EPA 600/R-94/062
My 1, 1994
REVALUATION OF INHALATION HEALTH RISKS
ASSOCIATED WITH METHYLCYCLOPENTADIENYL
MANGANESE TRICARBONYL (MMT) IN GASOLINE
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
-------
DISCLAIMER
This document has been reviewed by the Office of Research and Development.
Mention of trade names or commercial products does not constitute endorsement or
recommendation.
11
-------
TABLE OF CONTENTS
LIST OF TABLES iv
LIST OF FIGURES viii
AUTHORS xi
REVIEWERS AND CONTRIBUTORS xii
ACKNOWLEDGMENTS xiii
1. INTRODUCTION 1
2. HEALTH EFFECTS ASSESSMENT 2
2.1 Background 2
2.2 Earlier Assessments 3
2.3 1993 Revised RfC 4
2.4 Alternative Approaches to Deriving RfCs 9
2.4.1 Conventional NOAEL- or LOAEL-Based Approach ... 11
2.4.2 NOSTASOT Approach 11
2.4,3 Benchmark Analyses 13
2.4.4 Bayesian Analyses 16
2.4.5 Summary of RfC Estimates 17
3. EXPOSURE ASSESSMENT 18
3.1 Background 18
3.2 Additional Canadian Studies . 19
3.3 The FTEAM Study 20
3.4 Estimated Mn Exposure Levels Associated with MMT Usage ... 21
4. RISK CHARACTERIZATION 25
5. SUMMARY AND CONCLUSIONS . . 32
6. REFERENCES 34
APPENDIX A: DOSE-RESPONSE ASSESSMENT ANALYSES A-l
APPENDIX B: EXPOSURE ASSESSMENT FOR MANGANESE B-l
APPENDDCC: MANGANESE INHALATION REFERENCE
CONCENTRATION C-l
111
-------
LIST OF TABLES
Number Page
A-l Prevalence of Abnormal Responses in Workers Exposed to Manganese
Dust A-3
A-2 NOSTASOT Values for Individual LffiD Exposure Data of Roels
(1993) A-5
A-3 NOSTASOT Values for Individual CRD Exposure Data of Roels
(1993) A-5
A-4 NOSTASOT Values for Individual ACRD Exposure Data of Roels
(1993) A-6
A-5 Summary of the Fit to the Eye-Hand Data of Roels et al. (1992,
Figure 3) Using LTRD Grouped Exposure for Six Different Models . . A-l 1
A-6 BMD Values for Eye-Hand Coordination Data of Roels et al.
(1992, Figure 3) for IIRD by Model and Specified Effect Level .... A-15
A-7 BMDL Values for Eye-Hand Coordination Data of Roels et al.
(1992, Figure 3) for IIRD by Model and Specified Effect Level .... A-16
A-8 Summary of the Fit to the Eye-Hand Data of Roels (1993) Using
Individual LORD Exposure Estimates for Six Different Models A-17
A-9 BMD Values for Eye-Hand Coordination Data from Roels (1993)
Using Individual IIRD by Model and Specified Effect Level A-17
A-10 BMDL Values for Eye-Hand Coordination Data from Roels (1993)
Using Individual LffiD by Model and Specified Effect Level ...... A-l8
A-ll Summary of the Fit to the Eye-Hand Data of Roels (1993) Using
CRD Exposure Estimates for Six Different Models A-19
A-12 Summary of the Fit to the Eye-Hand Data of Roels (1993) Using
ACRD Exposure Estimates for Six Different Models A-21
A-13 Logistic Regression Coefficients for Eye-Hand Coordination
Data of Roels (1993) Using IIRD as an Estimate of Exposure A-23
A-14 Logistic Regression Coefficients for Eye-Hand Coordination
Data of Roels (1993) Using CRD as an Estimate of Exposure A-23
IV
-------
LIST OF TABLES (cont'd)
Number page
A-15 Logistic Regression Coefficients for Eye-Hand Coordination Data
of Roels (1993) Using ACRD as an Estimate of Exposure A-24
A-I6 BMD Values for Eye-Hand Coordination for CRD by Model and
Specified Effect Level A-24
A-17 BMDL Values for Eye-Hand Coordination for CRD by Model and
Specified Effect Level A-25
A-18 BMD Values for Eye-Hand Coordination for ACRD by Model and
Specified Effect Level A-25
A-19 BMDL Values for Eye-Hand Coordination for ACRD by Model and
Specified Effect Level A-25
A-20 BMD Values for Visual Reaction Time for Individual LJDRD by
Model and Specified Effect Level A-26
A-21 BMDL Values for Visual Reaction Time for Individual LIED by
Model and Specified Effect Level A-27
A-22 BMD Values for Hand Steadiness for Individual ORD by Model and
Specified Effect A-27
A-23 BMDL Values for Hand Steadiness for Individual LIRD by Model and
Specified Effect A-27
A-24 BMDL Values for LIRD Based on Individual Dichotoraous Response
Data from Roels et al. (1992) A-28
A-25 LJRD Mn Concentration for 10% Increase in Abnormal Eye-Hand
Coordination Values Based on Figure 3 of Roels et al. (1992) A-32
A-26 LIRD Mn Concentration for 10% Increase in Abnormal Hand
Steadiness Values from Figure 3 of Roels et al. (1992) A-33
A-27 LTRD Mn Concentration for 10% Increase in Visual Reaction
Time Values from Figure 3 of Roels et al. (1992) A-34
A-28 LIRD Mn Concentration for 10% Increase in Abnormal Eye-Hand
Coordination Values from Roels (1993) A-34
-------
IIST OF TABLES (cont'd)
Number Page
A-29 URD Mn Concentration for 5% Increase in Abnormal Eye-Hand
Coordination Values from Roels (1993) A-35
A-30 LIRD Mn Concentration for 1 % Increase in Abnormal Eye-Hand
Coordination Values from Roels (1993) A-35
A-31 CRD Mn Concentration for 10% Increase in Abnormal Eye-Hand
Coordination Values from Roels (1993) A-36
A-32 CRD Mn Concentration for 5% Increase in Abnormal Eye-Hand
Coordination Values from Roels (1993) A-37
A-33 CRD Mn Concentration for 1 % Increase in Abnormal Eye-Hand
Coordination Values from Roels (1993) A-37
A-34 ACRI> Mn Concentration for a 10% Increase in Abnormal Eye-Hand
Coordination Values from Roels (1993) A-37
A-35 ACRD Mn Concentration for a 5 % Increase in Abnormal Eye-Hand
Coordination Values from Roels (1993) A-38
A-36 ACRD Mn Concentration for a 1 % Increase in Abnormal Eye-Hand
Coordination Values from Roels (1993) A-38
A-37 Comparison of Lower Limit (and HEC) Estimates for 10% Increase
in Eye-Hand Coordination Dichotomous Response Derived by BMD
or Bayesian Analyses A-52
A-38 Concentration That Yields Indicated Results by Analyses of
Continuous Data A-53
A-39 RfC Estimates Derived by Different Approaches Using Data
from Roels et al. (1992) and Roels (1993) A-57
Attachment A-l Individual Exposure and Dichotomized Response Data of
Roels (1993) . A-66
Attachment A-2 Grouped CRD Exposure Data and Dichotomized Response
of Roels (1993) A-69
Attachment A-3 Grouped ACRD Data and Dichotomized Response of Roels
(1993) A-70
VI
-------
LIST OF TABLES (cont'd)
Number Page
Attachment A-4 Grouped URD Exposure and Dichotomized Response Data
of Reels (1993) A-71
B-l Summary of Annual Ambient PMi0 Mn in Toronto B-6
B-2 Summary of PTEAM Mn Measurements in jtg/m3 Mn B-21
B-3 Comparison of Average Mn Personal Exposure Levels from Canadian
and PTEAM Studies B-25
B-4 Data Derived from Lyons et al. (1993) Used To Estimate Automotive
Source Contribution to Different Size Fractions of Mn Paniculate
Matter B-31
B-5 PTEAM Weighted Mean Elemental Concentrations B-33
B-6 Relation of TSP Measurements in Riverside and Other Stations in
the Los Angeles Basin B-43
B-7 Summary of Mn Concentrations from 22 Locations in the
IP Network, 1980 B-44
Attachment B-l PTEAM Mn Data Collected by PM10 Personal Samplers
During Daytime Hours B-53
Attachment B-2 PTEAM Mn Data Collected by PM10 Personal Samplers
During Nighttime Hours B-59
Attachment B-3 PTEAM Mn Data Collected by PM10 Personal Samplers
During Daily Periods Covering Both Daytime and Nighttime
Hours B-64
Attachment B-7 Pilot PTEAM Mn Data Collected by PM10 Personal Samplers
for 24 h on Nonconseeutive Days B-83
Vll
-------
LIST OF FIGURES
Number Page
1 RfC estimates derived by several different approaches, as explained
in the accompanying text and Appendix A, for three measures of Mn
exposure obtained from Roels et al. (1992) and Roels (1993) . 12
2 Logarithmic-probability plot of projected estimates of long-term
personal exposure levels of PM4 Mn for the fall season in Riverside,
CA, if MMT were used in 100% of unleaded gasoline at
1/32 g Mn/gal 24
3 Summary of Mn RfC estimates and personal exposure levels of Mn . . 31
A-l Quanta! linear model fitted to Roels (1993) eye-hand coordination
data, with LTRD as the exposure variable A-12
A-2 Quanta! quadratic model fitted to Roels (1993) eye-hand coordination
data, with IIRD as the exposure variable A-12
A-3 Unrestricted Weibull model fitted to Roels (1993) eye-hand coordination
data, with IIRD as the exposure variable A-13
A-4 Restricted Weibull model fitted to Roels (1993) eye-hand coordination
data, with IIRD as the exposure variable A-13
A-5 Unrestricted log-logistic model fitted to Roels (1993) eye-hand
coordination data, with IIRD as the exposure variable A-14
A-6 Restricted log-logistic model fitted to Roels (1993) eye-hand data,
with IIRD as the exposure variable A-14
A-7 Roels (1993) eye-hand coordination data fitted to a quanta! linear
model, with CRD as the exposure variable A-20
A-8 Roels (1993) eye-band coordination data fitted to a restricted
log-logistic model, with CRD as the exposure variable A-20
A-9 Roels (1993) eye-hand coordination data fitted to a quanta! linear
model, with ACRD as the exposure variable A-22
A-10 Roels (1993) eye-hand coordination data fitted to a restricted
log-logistic model, with ACRD as the exposure variable A-22
Vlll
-------
UST OF FIGURES (cont'd)
Fjuml)er Page
A-l 1 Schematic of computing a posterior distribution from a likelihood
function and a prior distribution . A-3Q
A-12 Posterior distributions generated by each of six different models
that would produce a 10% increase over background in the rate of
abnormal eye-hand coordination responses A-32
A-l 3 Concentration to increase abnormal eye-hand coordination responses
by 10% A-40
A-14 Concentration to increase abnormal eye-hand coordination responses
by 5% A-41
A-15 Concentration to produce observed difference in means of control
versus exposed cohorts for eye-hand coordination A-41
A-16 Concentration to produce just statistically significant difference
in means of control versus exposed cohorts for
eye-hand coordination responses , A-42
A-17 Mn exposure times of individual exposed workers
(Roels, 1993) A-49
B-l Frequency distribution of ambient Mn for Montreal for years 1984
to 1987 showing the effect of increased average daily traffic B-4
B-2 Logarithmic-probability plot of daytime Mn PM10 personal exposure
in Riverside, CA, and concentrations measured inside and outside
subjects* homes B-12
B-3 Logarithmic-probability plot of nighttime Mn PM10 personal exposure
in Riverside, CA, and concentrations measured inside and outside
subjects' homes B-13
B-4 Logarithmic-probability plot of 12-h Mn PM2.s indoor and outdoor
concentrations for both daytime and nighttime B-14
B-5 Logarithmic-probability plot of daytime outdoor Mn concentrations . . . B-15
B-6 Logarithmic-probability plot of nighttime outdoor Mn
concentrations B-16
IX
-------
LIST OF FIGURES (cont'd)
Number Page
B-7 Logarithmic-probability plot of daytime indoor Mn concentrations .... B-17
B-8 Logarithmic-probability plot of nighttime indoor Mn concentrations ... B-18
B-9 Logarithmic-probability plot of indoor and outdoor Mn PM10
concentrations for both daytime and nighttime . B-19
B-10 Summary of personal Mn exposure data from various studies (discussed
in the text), showing local ambient Mn data during the period of each
study where available B-24
B-l 1 Logarithmic-probability plot of 24-h average PM10 Mn personal
exposures in Riverside, CA, in 1990 B-28
B-12 Size distributions of Mn measured at Pico Rivera and Upland, CA, in
the winter and summer of 1989 B-29
B-13 Estimated size distributions for Mn with the suspended dust component
subtracted B-30
B-14 Logarithmic-probability plot of projected distributions (by two different
approaches explained in text) of 24-h PM4 personal exposures for the
fall season in Riverside, CA, assuming 100% MMT usage
at 1/32 g Mn/gal B-35
B-15 Logarithmic-probability plot of projected estimates of long-term
personal exposure levels of PM4 Mn for the fall season in Riverside,
CA, if MMT were used in 100% of unleaded gasoline at
1/32 g Mn/gal B-39
B-16 Cumulative frequency distribution of cities in the EPA
Inhalable Paniculate Network, 1979 to 1983, with ambient
PM10 Mn concentrations less than the specified levels . B-4S
-------
AUTHORS
Dr. J, Michael Davis (Team Leader)
Environmental Criteria and Assessment
Office (MD-52)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Joseph V. Behar
Environmental Monitoring Systems
Laboratory
U.S. Environmental Protection Agency
P.O. Box 93478
Las Vegas, NV 89193-3478
Dr. James Braddock
Atmospheric Research and Exposure
Assessment Laboratory (MD-46)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Mr. Stanley Durkee
Office of Science, Planning, and
Regulatory Evaluation (8105)
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC 20460
Dr. Judith A. Graham
Environmental Criteria and Assessment
Office (MD-52)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Victor Hasselblad
(consultant under contract to EPA)
Center for Health Policy
Research and Education
Duke University
Durham, NC 27705
Ms. Annie Jarabek
Environmental Criteria and Assessment
Office (MD-52)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. David Mage
Atmospheric Research and Exposure
Assessment Laboratory (MD-56)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Wayne Ott
Atmospheric Research and Exposure
Assessment Laboratory (MD-56)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. David Svendsgaard
Health Effects Research Laboratory
(MD-55)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Lance Wallace
Environmental Monitoring Systems
Laboratory
U.S. Environmental Protection Agency
Building 166, Bicher Road
Wanenton, VA 22186
XI
-------
REVIEWERS AND CONTRIBUTORS
Mr. Tim Backstrom
Office of General Counsel (2344)
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC 20460
Mr. Frank Black
Atmospheric Research and Exposure
Assessment Laboratory (MD-46)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. William K. Boyes
Health Effects Research Laboratory
(MD-74B)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Robert Elias
Environmental Criteria and Assessment
Office (MD-52)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Lester D. Grant
Environmental Criteria and Assessment
Office (MD-52)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. William F. Farland
Office of Health and Environmental
Assessment (8601)
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC 20460
Dr. Kenneth Knapp
Atmospheric Research and Exposure
Assessment Laboratory (MD-46)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Mr. David Kortum
Office of Mobile Sources (6406J)
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC 20460
Dr. Kathryn Mahaffey
Environmental Criteria and Assessment
Office (MS-114)
U.S. Environmental Protection Agency
26 Martin Luther King Drive
Cincinnati, OH 45268
Dr. Allan Marcus
Environmental Criteria and Assessment
Office (MD-52)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Woodrow Setter
Health Effects Research Laboratory
(MD-55)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Chon Shoaf
Environmental Criteria and Assessment
Office (MD-52)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Jeannette Wiltse
Office of Health and Environmental
Assessment (8601)
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC 20460
Dr. Hal Zenick
Health Effects Research Laboratory
(MD-51)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Xll
-------
ACKNOWLEDGEMENTS
Several individuals contributed significantly to the production of this report. From
ManTech Environmental Technology, Inc., they include Marianne Barrier, John Barton,
Sheila Lassiter, Wendy Lloyd, Edie Smith, and Peter Winz. From Information Organizers,
Inc., they include Blythe Hatcher. From the U.S. Environmental Protection Agency, they
include Linda Bailey-Becht, Douglas Fennell, and Judy Theisen.
Xlll
-------
Revaluation of Inhalation Health Risks
Associated with Methylcyclopentadienyl
Manganese Tricarbonyl (MMT) in Gasoline
1, INTRODUCTION
In 1990, the Office of Research and Development (OKD) of the U.S. Environmental
Protection Agency (EPA) assessed the potential health risks associated with the use of
methylcyclopentadienyl manganese tricarbonyl (MMT) as an additive in unleaded gasoline
(U.S. Environmental Protection Agency, 1990), That assessment was provided in technical
support to the Office of Mobile Sources (OMS) as they considered a decision on Ethyl
Corporation's request for a waiver for MMT. Later, ORD (Preuss, 1991) reaffirmed its
assessment after considering a resubmitted waiver application for MMT from Ethyl
Corporation. Based primarily on data suggesting increased hydrocarbon emissions, EPA did
not approve that waiver request. After Ethyl Corporation submitted additional emissions
date, EPA reconsidered its decision to deny the waiver request. Subsequently, EPA and
Ethyl Corporation agreed to an extension of the deadline for a decision on the waiver
request, in order to permit EPA to consider any new information in a reevaluation of health
and exposure issues related to MMT in unleaded gasoline. As identified in earlier OKD
evaluations (U.S. Environmental Protection Agency, 1990; Preuss, 1991), a key issue is the
potential health risk associated with inhalation exposure to manganese tetroxide (MrijO^,
which is the primary by-product resulting from the combustion of MMT in gasoline. New
information on manganese (Mn) health effects and exposure is incorporated in this revised
risk assessment prepared for OMS.
This reevaluation has four components: (1) a health effects assessment, (2) an exposure
assessment, (3) a risk characterization relating the first two, and (4) a summary and
conclusions. This report summarizes earlier ORD assessments and incorporates information
from certain other major new reports and analyses. The reader is referred to the appendices
to this report for more detailed background information. Appendix A presents dose-response
analyses, Appendix B presents an exposure assessment, and Appendix C contains the current
verified Mn inhalation reference concentration (RfC) as it appears in the U.S. EPA
Integrated Risk Information System (IRIS, 1993).
1
-------
2. HEALTH EFFECTS ASSESSMENT
2.1 Background
The toxicity of Mn varies according to the route of exposure (see Appendix C).
By ingestion, Mn has relatively low toxicity at typical exposure levels due in part to a low
rate of absorption from the gastrointestinal tract and in part to efficient regulation by
homeostatic mechanisms. Manganese is considered a nutritionally essential trace element and
is required for certain enzymes important for normal functioning of the central nervous
system and other body organs. However, by inhalation, Mn has been known since the early
1800s to be toxic to workers. Neurobehavioral, respiratory, and reproductive effects are the
primary features of excessive occupational exposure to Mn. Manganism is characterized by
various psychiatric and movement disorders, with some general resemblance to Parkinson's
disease in terms of difficulties in the fine control of some movements, lack of facial
expression, and involvement of underlying neuroanatomical and neurochernical factors.
Neurobehavioral effects of Mn intoxication are generally more clinically prominent than
respiratory or reproductive effects. However, respiratory effects (e.g., pneumonitis) and
reproductive dysfunction (e.g., reduced Ebido) are also frequently reported features of
occupational Mn intoxication. The available evidence is inadequate to determine whether or
not Mn is carcinogenic; some reports suggest that it may even be protective against cancer.
Based on this mixed but insufficient evidence, EPA has placed Mn in a Group D weight-of-
evidence category, which signifies that it is not classifiable as to human carcinogenicity.
Given these features of Mn toxicity, the health assessment focuses on the potential for
chronic noncancer effects.
Various epidemiological studies of workers exposed to Mn at average levels below the
current American Conference of Governmental Industrial Hygienists Threshold Limit Value
^J
(TLV) (5 mg/m )l have shown neurobehavioral, reproductive, and respiratory effects, both
by objective testing methods and by workers' self-reported symptoms on questionnaires
(see Appendix C). Neurobehavioral effects generally have reflected disturbances in the
control of hand movements (e.g., tremor, reduced hand steadiness) and/or the speed of
movement (e.g., longer reaction time, slower finger-tapping speed). Reproductive effects
'"The American Conference of Governmental Industrial Hygienists (1992) has given notice of intent to lower the
TLV to 0.2 mg/m.
-------
have included a decrease in the number of children bom to Mn-exposed workers (compared
to matched controls) and various self-reported symptoms of sexual dysfunction. In recent
studies at low to moderate occupational exposure levels, respiratory effects have been
reflected primarily in self-reported symptoms of respiratory tract illnesses rather than in
differences between objective spirometric measurements in Mn-exposed and control workers.
However, the lack of studies using more sensitive investigation^ methods and the existence
of some limited evidence from an epidemiological study of school children raise a degree of
concern about pulmonary function effects in relation to lower level Mn exposure.
The precise mechanisms of Mn neurotoxicity are not well understood, but it appears
that Mn can affect several different aspects of central nervous system (CNS) function and
structure. Some experimental evidence suggests that the mechanisms of Mn toxicity may
depend on the oxidation state of Mn. However, both the trivalent form (Mn "*") and the
n_i_
divalent form (Mn ) have been demonstrated to be neurotoxic. Also, both forms of Mn
can cross the blood-brain barrier, although research suggests that Mn + is predominantly
transported bound to the protein traiisferrin (Aschner and Gannon, 1994), whereas Mn +
may enter the brain independently of such a transport mechanism (Murphy et al., 1991).
Unlike ingested Mn, inhaled Mn is transported directly from the lungs to the vicinity of the
brain before its first pass by the liver. Depending on the form of Mn inhaled, its conversion
to other oxidation states (e.g., oxidation of Mn2"1" to Mn3+ or reduction of Mn + to Mn +)»
and its ability to enter the brain (through a protein transport mechanism or otherwise), it is
quite possible that a significant fraction of even small amounts of inhaled Mn would be able
to reach target sites in the CNS. Thus, the apparently greater toxicity of inhaled versus
ingested Mn may reflect important pharmacodynamic and pharmacokinetic differences of Mn
that enters the body by different routes. A more definitive understanding of these issues will
require more empirical information.
2.2 Earlier Assessments
Earlier ORD health assessments have been based on the RfC, which is defined as an
estimate (with uncertainty spanning about an order of magnitude) of a continuous inhalation
exposure level for the human population (including sensitive subpopulations) that is likely to
be without appreciable risk of deleterious noncancer effects during a lifetime. The basic
-------
procedure for derivation of an IfC entails identifying a no-observed-adverse-effect level
(NOAEL) and a lowest-observed-adverse-effect level (LOAEL) from a "principal" study,
generally defined as the available study that best defines the highest NOAEL or lowest
LOAEL for the most sensitive endpoint affected by a chemical. When an investigation of
occupationally exposed humans is the principal study (as in the case of the Mn RfC), the
NOAEL or LOAEL is adjusted for differences in ventilation rates and exposure durations
3
between the occupational exposure scenario (10 m air breathed per 8-h workday,
5 days/week) and the "general public" scenario (20 m air breathed per 24-h day,
7 days/week). The adjusted NOAEL or LOAEL is then divided by uncertainty factors and a
modifying factor. In the case of the original RfC for Mn, uncertainty factors of 10 each
were used for extrapolating from a healthy worker population to the general population
(including sensitive subpopulations) and for extrapolating from a LOAEL to a NOAEL.
Also, an uncertainty factor of 3 (approximately one-half of 10 on a log scale) was used for
extrapolating from subchronic to chronic exposure. A modifying factor of 3 was used
because of statements by the authors of the principal study (Reels et al., 1987) that past
exposure levels of workers in the subject study were probably lower than those measured at
the time the study was conducted. The resulting RfC of 0.4 ftg Mn/m3 was used for the
earlier ORD risk assessment (U.S. Environmental Protection Agency, 1990) and was entered
on EPA's IMS computer database of human health risk and regulatory information in
December 1990.
2.3 1993 Revised RfC
The original RfC for Mn was revised, in part, because newer information supplied in
conjunction with the resubmittal of the MMT waiver application by Ethyl Corporation
indicated that the workers' exposure levels in the principal study had probably not increased
over time, and thus the modifying factor could be "eliminated" (i.e., set equal to 1).
Another reason for revising the original RfC was that more recent studies (Reels et al.,
1992; Mergler et al., 1994) provided additional evidence of health effects in workers at
relatively low airborne concentrations of Mn.
Independently of their earlier study of Mn-exposed workers (Roels et al., 1987),
Roels et al. (1992) conducted a cross-sectional study of neurobehavioral and other endpoints
-------
in another group of workers from a different factory—namely, 92 male alkaline-battery plant
workers exposed to manganese dioxide (MnO^ dust—who were compared to a matched
control group of 101 male workers without industrial Mn exposure. The geometric mean
occupational-lifetime integrated respirable dust concentration was 793 ftg Mn/m3 x years
(range: 40 to 4,433). The equivalent value for total dust was 3,505 pg Mn/m3 x years
(range; 191 to 27,465). The authors noted that the monitored concentrations were
representative of the usual exposures of the workers because work practices had not changed
during the last 15 years of the plant's operation. No data on particle size or chemical purity
were provided in the report by Roels et al, (1992), but based on information provided by
Roels et al. (1992) and Roels (1993), the median cut point for the respirable dust fraction
was 5 /xm aerodynamic diameter. The respirable fraction is more representative of the
lexicologically significant particles (i.e., the smaller particles that are inhaled and deposit
predominantly in the lower respiratory tract). Total dust measurements comprised the
respirable dust as well as larger particles that deposit predominantly in the nose and throat
region (via nasal breathing) and would be cleared more rapidly from the respiratory tract
than the smaller particles retained in the lower regions. Therefore, the respirable dust
measurements were considered to be a more accurate indicator of exposure in relation to the
observed health effects.
Manganese-exposed workers in the 1992 study by Roels et al. performed significantly
worse than matched controls on several measures of neurobehavioral function, particularly
visual reaction time, eye-hand coordination, and hand steadiness. Similar neurobehavioral
impairments were also found in the earlier study by Roels et al. (1987) of a different
occupational population exposed to mixed Mn oxides and salts at approximately the same
levels of total dust (respirable dust was not measured). In addition, a recent study in Canada
by Mergler et al. (1994) indicated that, among other effects, performance on tests of the
ability to make rapid alternating hand movements, to maintain hand steadiness, and to
perform other aspects of fine motor control was significantly worse, compared to matched
controls, in workers who were exposed to even lower concentrations of respirable dust
(35 fig Mn/m3 at the time of the study). If Mergler et al. had included information on
integrated past exposure levels (which they have since provided to OKD in a preliminary
form not yet submitted for publication), their study would have provided a fivefold lower
-------
LOAEL for the derivation of the RfC. In addition, reports of a Swedish study of
Mn-exposed steel workers (Iregren, 1990; Wennberg et al., 1991, 1992) provided compelling
evidence of comparable neurobehavioral impairments, including slower reaction time and
finger-tapping speed. The median total dust concentration in the Swedish study was
140 11% Mn/m3, with respirable dust reported as constituting 20 to 80% of individual
workers' total dust exposures. Thus, the LOAEL from this study would be somewhat lower
than that from Roels et al. (1992), but the less fully characterized exposure histories in the
Swedish study made it more appropriate as a supporting (rather than principal) study for
deriving the Mn IfC.
Taken together, the above epidemiological studies provide a consistent pattern of
evidence indicating that neurotoxicity is associated with low-level occupational Mn exposure.
The fact that speed and coordination of motor function are especially impaired is particularly
noteworthy, given its consistency with other epidemiological, clinical, and experimental
animal evidence of higher concentration Mn intoxication.
Differences among these studies in the duration of workers' exposure to Mn raise
another issue of relevance to this discussion. In the Roels et al. (1992) study, the mean
period of exposure was 5.3 years (range: 0.2 to 17.7 years). In the other studies, the mean
durations of exposure were longer: 7.1 years in Roels et al. (1987), 9.9 years in Iregren
(1990), and 16.7 years in Mergler et al. (1994). The indications of lower LOAELs in the
Canadian and Swedish studies suggest that neurobehavioral effects might occur at lower
concentrations of Mn if the exposure periods were longer. In addition, the age of the
workers may be an important factor in interpreting these findings. The oldest worker in the
Roels et al. (1992) study was less than 50 years old; also, the average age in that study was
only 31.3 years, versus 34.3 years in Roels et al. (1987), 43.4 years in Mergler et al.
(1994), and 46.4 years in Iregren (1990). These points suggest that longer exposure and/or
testing later in life might result in the detection of effects at lower concentrations than is
possible after shorter periods of exposure and/or in younger workers. On the other hand, it
is also evident from these studies that a much shorter period than a full lifetime of
occupational Mn exposure may be sufficient to induce Mn neurotoxicity.
As Roels et al. (1992) and other investigators have noted, a threshold for the neurotoxic
effects of Mn has not been reported in the epidemiological literature. Therefore, instead of a
-------
NOAEL, a LOAEL was obtained from the study by Roels et al. (1992) by dividing the
geometric mean integrated respirable dust concentration (793 fig Mn/m3 x years) by the
average period of worker exposure (5.3 years) to eliminate time (in years) from the
time-weighted average, thereby yielding a LOAEL of 150 pg Mn/m3. (The geometric mean
concentration was used to represent the average exposure because the workers' exposure
measurements were log-normally distributed, and the arithmetic mean exposure period was
used because it was the only value reported by Roels et al. [1992].) The workplace-based
LOAEL of 150 /xg Mn/m was then adjusted for nonoccupational lifetime exposure by
multiplying it by (1) the quotient of 10 m3/day divided by 20 m3/day (for worker versus
nonworker ventilation rates) and (2) the quotient of 5 days divided by 7 days (for work week
versus full week). The resulting adjusted LOAEL, labeled the human equivalent
concentration (HEC), was 50 fj.g Mn/m , which was then divided by a total uncertainty factor
of 1,000 to yield an RfC of 0.05 /*g/m3. The total uncertainty factor of 1,000 incorporated
the following factors: 10 to protect sensitive individuals; 10 for using a LOAEL in lieu of a
NOAEL; and a composite factor of 10 for database limitations reflecting the less-than-
chronic periods of exposure and the lack of reproductive and developmental toxicity data, as
well as potential but unqualified differences in the toxicity of different forms of Mn.
A modifying factor was not used (i.e., it was set equal to 1).
Each RfC is assigned an overall rating of low, medium, or high confidence level, based
on two subsidiary confidence ratings reflecting the quality of the evidence from the principal
studies and the quality of the overall database for the chemical in question, respectively.
The revised Mn RfC was assigned a medium level of confidence. The evidence for the
neurobehavioral effects of low-level Mn exposure by inhalation was compelling and
consistent across several well-conducted studies. However, the limited duration of exposure
and the lack of a NOAEL for neurotoxicity in any of the principal or supporting studies
prevented assigning a confidence level greater than medium. Also, the lack of definitive data
on the concentration-response relationship and on the potential reproductive and
developmental toxicity of inhaled Mn limited the degree of confidence in the database to a
medium rating. Virtually all of the human health evidence is based on healthy, adult male
workers. No known studies have investigated human female reproductive function, and even
though male worker reproductive function is known to be affected by Mn exposure, it has
-------
not received adequate investigation. The limited available information concerning the
developmental toxicity of inhaled Mn suggests the possibility that prenatal exposure of
laboratory rodents to MaO2 (via the air supplied to the pregnant mother) may depress
neurobehavioral activity in neonatal rate and that continued postnatal exposure of the pups
may intensify this depression. In addition, several studies have demonstrated alterations in
neurochemical (dopamine) levels in young mice and rats exposed during early postnatal
development to Mn via other routes. Thus, the potential for developmental toxicity due to
Mn exposure exists. The concentrations and durations of exposure sufficient to induce such
effects are not known. Although adequate epidemiological studies of children and the elderly
have not been conducted, it is known that certain populations, such as children, pregnant
women, elderly persons, iron- or calcium-deficient individuals, and individuals with liver
impairment, may have an increased potential for excessive Mn body burdens due to increased
absorption or altered clearance mechanisms.
Another concern raised by the lack of studies involving longer periods of exposure
and/or older subjects is that the compensatory or reserve capacity of certain neurological
mechanisms may be stressed by Mn exposure earlier in life, with manifestations of
impairments only becoming evident much later, perhaps at a geriatric stage. One reason for
the latter concern is that Parkinson's disease is typically a geriatric disease in which
symptoms are only seen when the loss of brain cells that produce dopamine (which is also
apparently involved in Mn toxicity) reaches 80% or more. Indeed, some neurologists think
that a long latency period of perhaps several decades may precede various parkinsonian
syndromes. These points lead to a concern that if Mn reduces the compensatory or reserve
capacity of the nervous system, parMnsonian-type effects might occur earlier in life than they
would otherwise. Thus, several questions remain to be answered before higher confidence in
the protectiveness of the RfC can be achieved.
The two studies of Roels et al. (1992, 1987) were considered coprincipal studies for the
derivation of the revised RfC, with supporting evidence in the reports of Mergler et al.
(1994), Iregren (1990), and Wennberg et al. (1991, 1992). Given the fact that these studies
involved exposure to various oxides and salts of Mn, the RfC is designated as applying to
Mn and Mn compounds (including Mn3O£, The previous RfC of 0.4 fig Mn/m3 applied to
Mn only, due to undifferentiated forms of Mn in the principal study. Given that different
8
-------
forms of metals may have different toxic properties (due to different oxidation states,
different solubilities, and possibly other factors), it is likely that different compounds of Mn
vary in toxicity. However, sufficient data on the comparative toxicity of various compounds
of Mn are not available to judge the relative toxicity of Mn3O4 specifically.
As noted above, Mn affects multiple organ systems, including the respiratory and
reproductive systems as well as the CNS. However, because die available evidence suggests
that the CNS is the most sensitive target for Mn toxicity, neurobehavioral endpoints were the
focus of the RfC derivation. Although other types of effects remain a concern, it is
presumed, based on the limited data now available, that protecting against neurotoxicity
provides protection against these other, apparently less sensitive endpoints.
In revising the RfC for Mn, a draft version was subjected to peer review by external
experts (from academic and non-EPA governmental institutions) as well as internal experts.
Following this peer review, a further-revised version was submitted to and verified by an
EPA-wide RfD/RfC work group in September 1993. The current RfC for Mn was made
available through ISIS in early November 1993 through two mechanisms. A special notice
beginning November 1 in the news section of EPA's internal DRIS2 database announced the
availability of a hard copy of the text to EPA requesters who contacted the Risk Information
Hotline; also, the text was obtainable through the National Library of Medicine's publicly
accessible on-line computer database, TOXNET, beginning November 10, 1993. It also
became available on line via the EPA IRIS database beginning December 1, 1993. The
complete text of the revised RfC as it exists on IRIS2 may be found in Appendix C,
2.4 Alternative Approaches to Deriving R/Cs
After the revised RfC for Mn became available to the public, 'Ethyl Corporation and
other interested parties submitted comments on the RfC and issues related to it. One of the
primary comments concerned the availability of various statistical techniques for deriving a
NOAEL from the study by Roels et al. (1992) and/or from supplementary date for that study
provided to ORD by Roels (1993). In response to Ethyl Corporation's request that EPA
consider alternative approaches to analyzing these data and deriving an RfC for Mn, further
analyses of the subject data were undertaken using a variety of statistical methods. These
approaches may be identified as (1) conventional NOAEL- or LQAEL-based analyses,
-------
(2) "no statistical significance of trend" (NOSTASOT) analyses of the type described by
Tukey et al. (1985), (3) benchmark dose analyses of the type described by Crump (1984),
and (4) Bayesian analyses of the type described by Jarabek and Hasselblad (1991). These
analyses and their results, which are presented and discussed in more detail in Appendix A,
yield several possible RfC estimates, so designated because the current and only verified RfC
for Mn is that which has been verified by the EPA-wide RfD/RfC work group and entered
on IRIS. It must be emphasized that the RfC estimates developed for the purpose of this risk
assessment do not represent a revision of the current verified RfC for Mn, Reexamination of
the current Mn RfC, and any decision to revise or reaffirm the current RfC, will be under
the purview of the EPA-wide RfD/RfC work group at some future date.
A fundamental issue pertaining to all of the approaches presented here is the selection
of a measure of exposure. Roels et al. (1992) described two measures of respirable dust, the
occupational lifetime respirable dust concentration (LERD), expressed as jtg/m x years, and
the current concentration of respirable dust (CRD), expressed as /*g/m3. The CRD
concentration was measured at the time the study was conducted by Roels et al. and refers to
a representative concentration measured for the type of job performed by a worker (e.g.,
electrician, maintenance worker). The LERD value for each worker was a cumulative
exposure measure derived by adding the CRD values over the worker's entire period of
employment. If a worker changed jobs within the plant during his period of employment,
the CRD for each job held was multiplied by the number of years the worker performed that
job. Thus, if more than one job classification was worked, the worker's URD was the sum
of the products of CRD multiplied by years of performance of the respective jobs. However,
if a worker held only one job classification, his LERD was simply equal to his CRD
multiplied by the number of years employed. Another measure of exposure may be derived
from LERD by dividing an individual worker's LERD value by his total number of years of
employment. The latter measure, designated as the average concentration of respirable dust
(ACRD), reflects a worker's time-weighted cumulative exposure level but removes years
•a
from the unit of measurement of EJRD and is expressed as uglm . Although Roels et al.
(1992) did not refer to ACRD, this value could be calculated for each individual and for the
entire cohort by using the unpublished data provided to ORD by Roels (1993). For reasons
10
-------
to be discussed later, ACRD offers advantages for certain analyses and, unless otherwise
noted, is the exposure measure used in the alternative RfC estimates discussed here,
2.4.1 Conventional NOAEL- or LOAEL-Based Approach
The conventional method, and only method used thus far by EPA, to derive an RfC has
been to identify a NOAEL or LOAEL from a study and divide that concentration by
uncertainly factors, as described above for the Mn RfC, In the case of the study by Roels
et al. (1992), the geometric mean URD concentration of the Mn-exposed workers was used
as a LOAEL. Roels et al. (1992) also performed an exposure-response analysis of their data
by grouping the exposed workers into three exposure categories and comparing the
prevalence of abnormal neurobehavioral scores for each of the three groups to those of
controls. As indicated in the summary sheet for the Mn RfC (see Appendix C), the results
of this exposure-response analysis were not used in deriving the revised Mn RfC because the
reported analysis did not correct for multiple comparisons. However, ORD's analyses of
additional data provided by Roels (1993) suggest a possible RfC estimate of 0.03
(versus the current RfC of 0.05 jtg/m3), if a one-tailed test of statistical significance is
accepted (see Appendix A). Because it was based on an exposure-response analysis, this
RfC estimate is labeled as such in Figure 1 .
2.4 .2 NOSTASOT Approach
Another approach to analyzing dose-response data makes use of a procedure known as
NOSTASOT, described by Tukey et al. (1985). In essence, the procedure applies a trend
test sequentially to determine the highest noneffective dose of a series of doses by eliminating
one dose at a time. In this manner, the dose level at which the response is not significantly
different from controls is determined to be the NOSTASOT dose, which could therefore be
considered a NOAEL. Applied to Roels' (1993) epidemiologic data by beginning with the
highest individual ACRD exposure and moving downward (i.e., a "top-down" approach), the
procedure yielded a NOSTASOT of 285 /tg/m3 for eye-hand coordination (see Table A-4,
Appendix A). This approach implies that once nonsignificance is reached, further application
of trend tests to lower dose groups would also yield nonsignificance. However, this was not
the case with Roels' (1993) epidemiologic data, and thus it was important to determine not
11
-------
io -
1 -
0.1 -
0.01-
I1'"
c" 10^-
io-5-
KT6-
*fl"T-
I v :
i
Not
Calculable
*
**
*
^ * * * «4» •»
A • •*> -U • • ••
• A • + A • •
^_ +
^ A • A •
V +
^
*
X
(111 II II II II II II I ^__ 'jULJtJC
f" E l10^0 5% 1%||10% 5% 1%jf^
$ | BMD Bayesian $
-------
simply the highest NOABL but the highest NOAEL below the lowest LOAEL. By this
"bottom-up" approach, the highest nonsignificant exposure below the lowest statistically
significant exposure was 21 ng/m , for visual reaction time. Using the latter value as a
NOAEL and a total uncertainty factor of 100 (the same as that used for the current Mn RfC,
except omitting a factor of 10 for extrapolating from a LQAEL to a NOAEL), one would
2
obtain a value of 0.07 jig/m for an RfC estimate (Figure 1). Disparities in the
NOSTASOTs obtained for various endpoints by the top-down and bottom-up approaches raise
questions about the suitability of this technique for deriving a NOAEL from the data of Roels
(1993).
2.4.3 Benchmark Analyses
Another approach to deriving an RfC estimate is the benchmark dose (BMD) approach,
which has been described by Cramp (1984) and others (e.g., Kimmel and Gaylor, 1988;
Faustman et al., 1994; Allen et al., 1994), A BMD is an estimate of the dose (the term dose
is used interchangeably here with concentration, although the latter is more appropriate for
inhalation exposure) that will produce a specified effect (e.g., a 10% increase in the
prevalence of abnormal scores on a neurobehavioral test in the case of the study by Roels et
al. [1992]). The BMD is calculated by fitting a mathematical model to the available data and
obtaining a maximum likelihood estimate of the dose associated with a specified increase in
response (typically 10, 5, or 1 %). A lower confidence limit is then calculated for the BMD
(usually the 95th pereentile), and the result is denoted as a benchmark dose level (BMDL),
which has been proposed as a substitute for a NOAEL in deriving RfDs or RfCs (Crump,
1984; Barnes et al., 1994). Subscripts designate the effect level (10, 5, or 1 %) for which
the BMDL has been calculated, as in BMDL10, EMDL5, or BMDLj.
A large number of mathematical models could be used for deriving BMDLs, but six
frequently used models have been selected for the present exercise (as discussed in
Appendix A). In applying these models to the dataset provided by Roels (1993), it appears
that the models fit the CRD and ACRD data better than the LLRD data. (As explained in
Appendix A, it made little difference whether the URD values were obtained from the group
data provided in the report by Roels et al. [1992] or from the individual exposure data
supplied by Roels [1993], so the latter IIRD data were used here.) In principle, LERD is
13
-------
superior to CRD as a measure of long-term or cumulative exposure. One reason for the
difference in goodness of fit between LIRD and either CRD or ACM) is that two workers
with low LIRD values had abnormal eye-hand coordination responses (exceeding the 95th
percentile of control scores). These two subjects appear to have had rather short exposure
durations (0.3 and 0.4 years) and moderately high CRD values (201 fig/m each). Thus,
these two data points suggest an LffiD exposure-response relationship that is better fit by a
supralinear curve with a power term < 1 (see Figures A-3 and A-5 in Appendix A) than by
the more nearly linear curve such as that produced by the quanta! linear or restricted Weibull
model (see Figures A-l and A-4 in Appendix A). If CRD or ACRD exposure data are used,
however, the two individuals tend to fall in line better with a linear model (see Figures A-7
to A-10 in Appendix A). Another factor contributing to difficulty in fitting the Roels (1993)
data with linear models is a tendency for the prevalence of abnormal eye-hand coordination
responses to decline slightly at the highest LIRD, CRD, and ACRD concentrations (evident
in Figures A-l to A-6).
The supralinear exposure-response curve for abnormal eye-hand coordination scores
suggests a possible corollary to the healthy worker phenomenon; namely, the existence of
newly employed, relatively sensitive workers vis-i-vis long-term, relatively nonsensitive
workers. It may be that the two above-noted workers happened to be rather susceptible to
Mn toxicity but had not been employed long enough for their greater sensitivity to become
otherwise evident. There could be a tendency for such workers to move to other types of
employment, leaving a greater proportion of relatively less sensitive individuals among the
older workers. Although irregularities of the type posed by the data for these two workers
create complexities for model fitting, it is important to recognize that statistical curve fitting
is secondary to the objective of selecting the most biologically appropriate exposure variable.
However, given that the final results obtained with LIRD, CRD, and ACRD are roughly
equivalent, ACRD has been selected for discussion here because it estimates an average
exposure over time and yet provides as good or better goodness of fit as CRD or LIRD for
most of the models considered.
Of the six models considered, the quantal linear model fits the data reasonably well and
is the least complex (see Appendix A). It also gives equivalent results to the restricted
Weibull model for BMDL calculations (although the two models differ slightly when used in
14
-------
the Bayesian analyses, to be described below). Although much more conservative results
would be obtained if the unrestricted Weibull or unrestricted log-logistic model were used,
the NOAEL/LOAEL surrogates obtained with the latter models are so small as to be
practicably incalculable and extend far below the range of actual measurements. Therefore,
the following discussion is focused on the results obtained with the quanta! linear model.
In addition to choosing a model, a specified rate of increase in the effect of concern
must be selected in using the BMD approach. This percentage increment is expressed in
terms of the effective concentration that would yield the stated increase. Increases of 10, 5,
and 1 % in the incidence of abnormal eye-hand coordination scores (as dichotomized by Roels
et aL, 1992) have been considered, with the concentrations associated with these levels called
"effective concentrations" and designated as ECt, EC5, and EC10, respectively. One guide
to the choice of an effect level is that the resulting BMD (before calculating the lower
confidence limit) is preferably near or within the range of observed exposure concentrations
(cf. Barnes et ai, 1994). Because the BMD for ECj falls outside this range of observed
concentrations, the primary focus in this discussion is devoted to the BMDI^ and the
BMDL10.
It should be kept in mind that the BMDL represents the lower 95th percent confidence
interval for the effective concentration in question, and therefore the BMDL probably
inherently reflects some degree of conservatism. However, the degree of conservatism
obviously varies with the effective concentration for different percentage effect levels and
with the nature of the effect (e.g., severe versus moderate impairment). For the purposes of
this assessment, if one treats the BMDL10 derived from the dichotomized (quanta!) data of
Roels as if it were a minimal (less severe) LOAEL and the BMDL5 as if it were a NOAEL,
uncertainty factors of 3 and 1, respectively, would be warranted. On this basis, as shown in
Figure 1 and in Table A-39 of Appendix A for the quanta! linear model using ACRD, an
RfC estimate of 0.09 /ig/m3 would be obtained by using the quanta! BMDLjQ as if it were a
LOAEL and a total uncertainty factor of 300 (10 for intraspecies sensitivity, 10 for database
limitations, and 3 for a minimal severity LOAEL). Similarly, the quanta! BMDL5 would
yield an RfC estimate of 0.1 /ug/m3, based on a total uncertainty factor of 100 (10 for
intraspecies sensitivity, 10 for database limitations, and 1 for a NOAEL). As applied here,
the benchmark approach yields candidate RfC estimates of 0.09 to 0.1
15
-------
2.4.4 Bayesian Analyses
Another approach to deriving a substitute for a conventional LOAEL or NOAEL,
which bears some resemblance to the BMD approach just described, is known as the
Bayesian approach (see Appendix A). In essence, the Bayesian approach yields a distribution
of concentrations (rather than a point estimate) associated with a specified effect. Some
features of the BMD approach are common to the Bayesian approach; a mathematical model
must be fit to the data, an effect level must be selected, and a confidence bound on the
estimated concentration associated with a given effect level must be calculated (although the
calculation procedures are different). If these choices are consistent with those for the BMD
approach, the results are quite similar. By the Bayesian analysis, for a 10% increase in
abnormal eye-hand coordination scores, the lower 90% credible set limit (roughly equivalent
to the quanta! BMDL 95% confidence limit [see Appendix A]) based on the estimated median
^
concentration obtained with the quanta! linear model is 73 ^g/m . Adjusting this value to a
human equivalent concentration (HEC) and treating the result (26 figlm ) as if it were a
LQAEL(HEC), one may divide by a total uncertainty factor of 300 (10 for intraspecies
sensitivity, 10 for database limitations, and 3 for a minimal severity LOAEL) and obtain an
RfC estimate of 0.09 /*g/m3. Similarly, the 5% effect level yields an RfC estimate of
0.1 (igfm , based on a total uncertainty factor of 100. Thus, as applied here, the Bayesian
approach yields candidate RfC estimates of 0.09 to 0.1 jig/m , essentially identical to the
results of the benchmark analysis (Figure 1).
One advantage of the Bayesian approach is that it lends itself well to using continuous
as well as dichotomous data. Although Roels et al. (1992) did not provide individual
continuous data (i.e., actual raw scores instead of designations of normal/abnormal) on the
performance of the workers in their study, they did report mean differences and standard
deviations. With this information, it is possible to estimate the concentration at which certain
effect levels would occur based on the Bayesian posterior distribution. For example, a 10%
increase in the proportion of subjects with abnormal scores would be associated with a
median concentration of 112 /*g/m3, which has a lower 90% credible set limit of 90
Adjusting the latter value as if it were a LOAEL(HEC) yields a concentration of 32
and an RfC estimate value of 0.1 fig/m , based on a total uncertainty factor of 300 (10 for
intraspecies sensitivity, 10 for database limitations, and 3 for a minimal severity LOAEL).
16
-------
Note that these calculations based on continuous data essentially approximate the quanta!
BMD and Bayesian calculations for a 10% effect level based on dichomotous data (see
Figure 1). Similar calculations for the actually observed difference (i.e., 13%) between the
Mn-exposed and control workers in the Roels et al. (1992) study yield an RfC estimate of
'S
0.2 (jLg/m , based on a total uncertainty factor of 300 (including a factor of 3 for a minimal
severity LOAEL). Calculating the concentration associated with the difference between the
exposed and control mean values that just achieves statistical significance (a 4% difference in
this case) also results in a candidate RfC value of 0.2 /tg/m3, based on a total uncertainty
factor of 100 (eliminating the minimal LOAEL factor of 3). Thus, as applied here, the
Bayesian analyses of continuous data yield candidate RfC estimates of 0.1 to 0.2 jig/m .
2.4.5 Summary of RfC Estimates
Figure 1 displays the current, verified RfC along with over 100 possible Mn RfC
estimates based on various exposure measures, models, effects measures, and uncertainty
factors. Not all of these RfC estimates are equally plausible or worthy of consideration in
assessing the potential health risks associated with Mn inhalation exposure due to MMT
usage. As discussed above, some combinations of the three exposure measures and six
mathematical models fit one another better than other combinations. Based primarily on
considerations of cumulative dose toxicity, statistical goodaess-of-fit, and parsimony, ACRD
and the quanta! linear model appear to achieve the best results in this respect. Given the
similarities of the benchmark and Bayesian analytic results using ACRD and the quanta!
linear model, little distinction can be made between the two analytic approaches in the
present application. As for the results obtained for different effect levels, using a severity
uncertainty factor of 3 with a 10% effect level (for either benchmark or Bayesian analyses) is
essentially equivalent to using a severity UF of 1 with a 5% effect level. Note that the terms
LOAEL and NOAEL do not actually correspond to the results for 10% and 5% effect levels,
and therefore neither is preferable to the other in the sense that a NOAEL is preferable to a
LOAEL in deriving an RfC. Therefore, benchmark and Bayesian results for 10% and 5%
effect levels (using ACRD with the quantal linear model) are regarded as equally worthy of
consideration here. These particular analyses yield Mn RfC estimates of 0.09 to 0.1 /*g/m .
17
-------
In general, continuous response data are preferred to dichotomized data, primarily
because they provide more information and avoid the basically arbitrary division of effect
measurements into categories (e.g., normal versus abnormal). The Bayesian analysis based
on mean differences between exposed and control groups offers some of the advantages of
using continuous data, in that the reported means and standard deviations (from Roels et al.,
1992) provide a basis for estimating the distribution of continuous response measures.
However, this use of continuous data is not immune to certain common problems, such as
the issue of statistical power associated with studies of limited size, for the approaches
calculating observed or just-statistically significant differences. Also, whereas the
dichotomous data analyses yield more precision in estimating the effective concentration
associated with a somewhat imprecise response variable, the continuous data analyses offer
the opposite trade off (i.e., more precision in the response variable but less in the exposure
estimate). Nevertheless, the continuous data analyses appear to merit consideration as well
as the analyses based on dichotomous data. By the Bayesian analyses of continuous data, Mn
RfC estimates of approximately 0.1 to 0.2 /ig/m are obtained.
Based on the available data and on decisions and assumptions involved in analyses of
these data, the leading candidate estimates for an alternative Mn RfC appear to fall in a range
of approximately 0.09 to 0.2 pg/m . (Ethyl Corporation [1994] has proposed an alternative
RfC estimate based on a BMDLj0 value of 87 feg/m3. Treating this value as essentially a
NOAEL [thereby eliminating an uncertainty factor for use of a LOAEL], Ethyl Corporation
divided the adjusted NOAELpHEC] by a single uncertainty factor of 10 for sensitive
subpopulations to derive a Mn RfC estimate of 3
3. EXPOSURE ASSESSMENT
3.1 Background
Very limited data have been available by which to estimate potential Mn personal
exposure levels likely to be associated with the use of MMT as an additive in unleaded
gasoline. For example, after the completion of ORD's 1990 exposure assessment for Mn
(U.S. Environmental Protection Agency, 1990), Ethyl Corporation provided EPA a brief
report of a personal monitoring study as part of Ethyl Corporation's resubmittal of a waiver
application for MMT. The study focused on 6 taxi drivers and 17 office workers in Toronto,
18
-------
ON, where the allowable MMT concentration in gasoline is 1/16 (0.062) g Mn/gal. (In the
Toronto study, the actual concentration was reported as 1/26 [0.039] g Mn/gal, which is only
slightly greater than the 1/32 [0.031] g Mn/gal concentration proposed for the United States,
As confirmed by Kirshenblatt [1993], MMT concentrations in Canadian gasoline average
well below the allowable limit there.) In comments on the Ethyl Corporation's resubmittal,
O1D considered the Toronto data in conjunction with results from independent field studies
of personal exposures to carbon monoxide to develop a revised Mn exposure assessment
(Preuss, 1991). A key element of the 1991 ORD assessment was the assumption that taxi
drivers (six of whom were monitored in Toronto within a 2-week period) were members of a
high-exposure cluster reflecting the upper 4% of the population in a model based on the
carbon monoxide field studies. The result of the 1991 assessment was an estimate that
4% of the general public might be exposed to Mn at approximately 0.09 jtg/m3, although this
estimate had an undetermined amount of uncertainty due to the inadequacies of the available
data.
3.2 Additional Canadian Studies
Since the 1991 ORD assessment, additional personal exposure studies have been
completed in Montreal and Toronto (described in Appendix B). As shown in Figure B-10 of
Appendix B, the average concentrations reported in the Canadian studies vary by as much as
an order of magnitude for small groups (5 to 19 persons each) of garage mechanics, taxi
drivers, and office workers. The highest average Mn personal exposure level was
0.25 pg/m for Montreal garage mechanics while at work; the other averages ranged from
0.002 to 0.035 Mg/m for various particle size fractions. Although it is impossible to
extrapolate the results of these studies to the distribution of Mn exposure levels for the
general population, it does appear that there is a general relationship between personal
exposure levels of Mn and proximity to vehicular emissions of combusted MMT. Thus,
populations living near high traffic-volume areas such as inner cities and expressways would
probably tend to experience higher Mn exposure levels in relation to MMT usage.
Some of the limitations of the Canadian studies with respect to development of a
quantitative exposure assessment are reviewed in detail in Appendix B and may be
summarized briefly as follows.
19
-------
* The studies did not have adequate simple sizes and did not sample according to a
probabilistic statistical design that would help ensure the representativeness of the
sampled individuals.
* The sampling periods were relatively short, 1 to 2 weeks at most. Meteorological
and other factors that would be expected to influence ambient measurements over
relatively short periods of time were not characterized.
* Because the studies did not use ambient monitors collocated with reference monitors
(such as the dichotomous samplers used by Canadian agencies), it is difficult to
relate date from the studies to larger databases from the government monitoring
networks.
* Because the studies did not use identical monitors to measure personal exposure
levels and outdoor ambient levels, it is difficult to distinguish between personal
exposures and ambient levels or to relate one to the other.
* Quality assurance and certain other important methodological details are not fully
provided in the available reports.
Because of the substantial limitations of the above exposure studies, no quantitative
assessment of personal exposures to Mn in a Canadian population is possible at present.
3.3 The PTEAM Study
The only published study that has used a probability-based representative sampling
design for evaluating exposure levels of Mn in a general population is the Particle Total
Exposure Assessment Methodology (PTEAM) study, which was conducted in Riverside, CA,
over a 7-week period in the fall of 1990 (Pellizzari et al., 1992). This study used personal
and stationary monitors to measure Mn concentrations indoors and outdoors. The personal
samplers collected PM10, and the stationary samplers collected PM2.s as well as PM10 (see
glossary of terms in Attachment B-6 to Appendix B). Of the 139,000 nonsmoking residents
age 10 years and older in Riverside, 178 individuals were selected through a stratified
sampling plan to represent the general population and were monitored over two 12-h periods
(daytime and nighttime). More than 2,750 particle samples were collected. Quality
assurance and other procedures are summarized in Appendix B and are described extensively
elsewhere (e.g., Pellizzari et al., 1992; Clayton et al., 1993; Thomas et al,, 1993). The
PTEAM study has been presented in various peer-reviewed publications and discussed in
several scientific forums (see Attachment B-5 to Appendix B). It represents the best
20
-------
available information on an actual distribution of general population exposures to Mn. It also
provides valuable information on potential Mn exposure associated with MMT use, because
MMT was used in leaded gasoline in California prior to and during the period of the
PTEAM study.
3.4 Estimated Mn Exposure Levels Associated with MMT Usage
As noted above, the substantial limitations of the available Canadian exposure studies
make them unsuitable for estimating population exposure levels of Mn in relation to MMT
usage. In addition, ambient monitoring data typically underestimate and may be uncorrelated
with personal exposure levels of automotive-source pollutants. Therefore, of the currently
existing published evidence pertaining to Mn exposure levels in relation to MMT usage, only
the PTEAM Riverside study (Pellizzari et al., 1992) provides a reasonable basis for
estimating potential future exposure levels in relation to a scenario where 100% of unleaded
gasoline contains 1/32 g Mn/gal as proposed by Ethyl Corporation.
In the FTEAM study, measurements of personal exposure levels of PM10 Mn indicated
that approximately half of the population in Riverside in the 1990 study period had 24-h
personal exposures to PM10 Mn above 0.035 jtg/m , with the highest 1 % of the population
having exposures above 0.223 jKg/m PMj0 Mn. However, given the use of PM5 Mn
exposure measurements in the study of Roels et al. (1992), it would be preferable to consider
a population distribution of personal exposure levels of VM^ Mn. Due to limitations in the
available data, the exposure assessment in Appendix B focuses on estimated personal
exposure levels for PM4 Mn, not PM5. Although the difference is probably small, PM4
levels are an underestimate of PM5 levels. The derivation of the projected exposure
estimates involved several steps, which may be summarized as follows.
The automotive and nonautomotive contributions to particulate Mn exposures in the
PTEAM study were estimated using data from Lyons et al. (1993), who reported particle
size distributions up to PM4 of selected trace metals, including Mn, at two locations near
Riverside in the winter and summer of 1989. They attributed most of the PM4 Mn to
automotive sources. Based on their findings and data from other sources, it is possible to
estimate that 69% of the PM2 5 fraction of PM* Mn they measured was derived from
automotive sources (namely the combustion of MMT in motor vehicle fuel, as then allowed
21
-------
in leaded gasoline in California) and that 31 % was derived from paved road dust (mostly
earth crustal material). Next, the PTEAM Mn measurements from stationary indoor
monitors (SIMs) were used to estimate personal exposure levels by adjusting the SIM PM2,5
Mn date to reflect the typically higher levels of all elements measured by personal exposure
monitors (PEMs). This adjustment was made in two ways, either by the PEMrSIM ratio
obtained for Mn or by the ratio obtained for lead (Pb), another element related to automotive
fuel usage. These two methods of adjusting the SIM data to PEM values resulted in two
projected distributions, as will be described below. The next step in the derivation procedure
involved adjusting the PM2-5 personal exposure estimates to reflect PM in the size range
from 2.5 to 4 pm (based again on data from Lyons et al., 1993).
With these estimates of PM4 Mn personal exposure levels due to automotive sources, it
was then possible to project from the situation in Riverside around the time of the PTEAM
study (when leaded-MMT gasoline constituted about 14% of the gasoline sold and contained
an average of 0.048 g Mn/gal) to a future scenario that assumes 100% of the unleaded
gasoline contains MMT at 1/32 (0.031) g Mn/gal. This aspect of the derivation is described
in detail in Attachment B-4 to Appendix B. In essence, a factor was calculated to reflect the
estimated increase in MMT usage between 1990 and 1995 (i.e., the first full year in the near
future). This projection factor assumed an increase of 1 % per year in gasoline usage and no
difference in the Mn emission rate (grams Mn emitted per gram Mn in fuel combusted) for
noncatalyst vehicles using leaded-MMT gasoline in 1990 versus catalyst vehicles using
unleaded-MMT gasoline in 1995.
Next, the nonautomotive contribution to PM4 Mn was estimated and added to the
estimated automotive contribution to obtain the projected personal exposure levels of total
PM4 Mn. Assuming the estimated PM4 Mn distribution has the same form as the PM10 Mn
distribution from PTEAM (approximately lognormal with equal geometric standard
deviations), the ratio of the PM4:PM10 arithmetic mean personal exposure levels yields a
scaling factor that can be applied to the PM10 distribution to obtain the PM4 distributions.
Because of the alternative bases for adjusting the SIM PM2 5 data for personal exposures
(as noted above), two different scaling factors were multiplied by the PM10 distribution,
thereby producing a higher and a lower estimate of the distribution of 24-h average PM4 Mn
personal exposure levels.
22
-------
In addition, because long-term exposures arc likely to have less variance than 24-h
exposures, it is appropriate to adjust the distributions of 24-h average exposure levels to
better reflect longer periods of exposure. Of various methods that may be used for this
purpose (Wallace et ai, 1994), two approaches were applied to adjust the geometric standard
deviations of the two projected exposure distributions, based on data from either the PTEAM
study or the smaller pilot study that preceded the PTEAM Riverside study. These alternative
methods were applied to the two estimated distributions of 24-h average exposures to yield
the distributions of long-term average PM4 Mn personal exposure levels depicted as lines 1
and 2 in Figure 2 (only the highest and lowest of the four resulting estimates are shown).
It must be emphasized that these two distributions do not represent upper and lower bounds,
because even higher or lower estimates could be produced by alternative assumptions and
adjustments of the data. Moreover, if data were available for another time of the year (e.g.,
spring in addition to fall), the estimates would not be season-specific and could possibly be
much higher or much lower. Nevertheless, given the limited available data, lines 1 and 2 in
Figure 2 represent two reasonable estimates of the projected long-term (autumnal) personal
exposure levels of PM4 Mn in relation to MMT usage at 1/32 g Mn/gal in 100% of unleaded
gasoline.
By examination of the logarithmic-probability plot of long-term personal exposure
levels of Mn, it is estimated that half of the population would be exposed to PM4 Mn levels
*s
of more than approximately 0.045 to 0.050 /tg/m . Also, based on the two projection
estimates, approximately 5 to 10% of the population would have personal exposure levels
around 0.1 jtg/m PM^ Mn or higher. The highest 1 % would be predicted to have PM4 Mn
exposure levels above 0.15 /*g/m . It should be noted that these projections refer specifically
to Riverside, CA, with a population of more than 139,000 persons. However, in many
significant respects (e.g., meterology and traffic volume), Riverside is reasonably
representative of the greater metropolitan area of Los Angeles, which has a total population
of over 14.5 million persons. The exposure projection estimates for Riverside imply the
possibility that hundreds of thousands of persons in the Los Angeles area alone could be
exposed to PM4 Mn levels exceeding 0.1 /ig/m . To the extent that any other U.S. cities
(e.g., in the Southwest) share some degree of resemblance in meteorology, vehicle miles
23
-------
I.UUU
05m
Oflno
0.700
n Kfm
U.OUU
0.5OO
ft 4AA
v>4UU
n inn
U.ZOO
0100.
U.UOU
wSl rt IY7A
CO U.U/U
^ n fioA
^JJ; v.UOU
0.050
3 n ndn
(0 ,U.U*IU
& 0030
UJ
ffi
SK A ftnfi
«» VJ.UcU
§
1
5 0010
H*f§^
OOO7
0006
OOO5
A ftf\A
U.IAHf
n nfw
nmo
0001
0
^
20
X
5 '
X
i :
1
X
>
^
2
5 1
k
0 2
I"X
10 3
X
0 A
X
10 !
X
>
}0!
V
X*
»7
y
X1
'0 G
^
^.
X
10 S
x
X
10 9
X
x
5 S
X
X
189
X
?"*"""
9 9<
I.UUU
o'ftnn
.700
ORIVt
.DUU
A CAA
O.OOO
A 4AA
U.4vv
fl W\
U^sUV
01 Of)
o nsn
On7n
finon
Oncn
.uau
nnAf\
0030
Onon
-UtU
fi fiin
U.UIU
0.009
n Ana
U.UUtf
OAA7
.007
OIVU5
•UUO
A A/\C
U.UV9
0.004
0.003
0.002
A Aft4
U.UU1
3.8
Population Cumulative Frequency (%)
Figure 2. Logarithmic-probability plot of projected estimates of long-term personal
exposure levels of PM4 Mn for the fall season in Riverside, CA, if MMT were
used in 100% of unleaded gasoline at 1/32 g Mn/gal. The two lines reflect
different approaches for estimating personal exposure levels from stationary
indoor monitoring data and different methods of adjusting 24-h averages to
long-term averages, as explained in the accompanying text and Appendix B.
24
-------
traveled (VMT), and possibly other characteristics of relevance to automotive Mn levels, the
estimated exposure levels for Riverside could be pertinent, at least qualitatively, to other
locales or portions of locales as well. Similarities and differences in point-source
contributions to Mn exposure would also figure into comparisons with other communities.
The presence of a major point source or sources of Mn in a community (which was not a
factor in Riverside) would add some increment to the level of Mn exposure experienced by
the persons in that community. Although these Riverside estimates cannot be applied
quantitatively to any other U.S. metropolitan areas, the total population of the U.S. counties
with VMT levels greater than that of Riverside (apart from the four counties Los Angeles
comprises) is approximately 15 million persons. Possibly, then, several hundreds of
thousands of persons could be exposed to PM4 Mn levels of approximately 0.1 /ig/m3 or
higher if MMT were used in 100% of the unleaded gasoline as proposed by Ethyl
Corporation in all of these areas. However, it must be emphasized that because of the
limited available data, a great deal of uncertainty surrounds such estimates. The actual
exposure levels could be much higher or lower.
4. RISK CHARACTERIZATION
To assess the public health risk associated with the use of MMT in gasoline in the
United States, the available qualitative and quantitative health effects information on Mn must
be related to the available exposure information. From the standpoint of a qualitative hazard
identification, the available evidence amply demonstrates that inhaled Mn is toxic to the
nervous system, the respiratory system, and the male reproductive system. The toxicity of
Mn by different routes of exposure has been demonstrated by numerous medical reports and
epidemiological and experimental studies. However, available data do not allow quantitative
estimation of the relative lexicological potency of different Mn compounds or permit
quantitative route-to-route extrapolations for predicting the effects of Mn3O4-
The focus of the above health assessment discussion has been on the RfC and the types
of risks associated with chronic Mn exposures because, for the most part, acute effect levels
appear to be considerably higher than the highest projected exposure levels. However, the
issue of less-than-chronic exposures does arise with respect to the potential for developmental
toxicity. It is widely recognized that the human CNS develops over a period of several
25
-------
years, prenatally and postnatally, and can be vulnerable to long-term or irreversible effects if
damage occurs during certain "critical stages" of development Recent evidence from
ongoing longitudinal studies of children indicates that lead (Pb) exposure (measured as blood
Pb level) around 2 years of age in particular is associated with reduced cognitive
performance at 4 to 10 years of age. Such evidence raises the concern that exposure to
another neurotoxic metal such as Mn during part of early development might also be capable
of inducing permanent or irreversible damage to the developing CNS. Moreover, the
ramifications of such damage might extend to other important functions, such as
reproduction.
Children may also be at higher risk in terms of exposure because of biomedical and
metabolic differences at a young age (greater uptake and retention) and/or because of the
longer duration of their exposure over a lifetime. Over time, small impairments in
neurobehavioral function may accumulate. For this reason, the elderly, whose
neurobehavioral function may already be compromised by normal aging processes and
possibly by other disease states (e.g., parkinsonism or preclinical parkinsonism), also
represent a special population of concern. The ability of the elderly or other subpopulations
to compensate for such declines in neurobehavioral function may be overwhelmed eventually
by additional, albeit possibly quite small, insults due to Mn. If so, the effect could be
manifested as a more severe or earlier onset of declining function in senescence, with
consequent implications for increased societal health-care costs.
Special subpopulations at increased risk may be defined not only in terms of their
biological susceptibility, as exemplified above by the young and the elderly, but also by their
increased risk of exposure. In this respect, inner city residents and others who live near high
traffic areas such as expressways (e.g., low-income and minority communities) would
possibly have a disproportionate likelihood of higher Mn exposure levels due to their closer
proximity to vehicular emissions.
The nature of the neurobehavioral effects observed in occupational studies such as
Roels et al. (1992) should be understood as effects that probably would not be treated by, or
even be readily evident to, a clinical physician. Nonetheless, they are significant from a
public health standpoint when considered in terms of population effects. This concept is
illustrated by the well documented findings on low-level Pb neurotoxicity in children, where
26
-------
changes of as little as 1 or 2 points in IQ have been repeatedly demonstrated in several
independent, prospective, epidemiological studies in recent years. Such changes could not be
reliably demonstrated either in a clinical setting or in earlier cross-sectional epidemiological
studies; yet they are now well established to be "real" and significant from a public health
standpoint. With regard to the reductions in neurobehavioral function observed in various
epidemiological studies of Mn-exposed workers, these studies independently converge on
findings of impaired motor function (e.g., reductions in eye-hand coordination, slower hand
or finger movements, and less control of fine movement). As recently expressed in a
document prepared by the Subcommittee for Risk Assessment of the Federal Coordinating
Council for Science, Engineering and Technology (Federal Register, 1993), an adverse effect
can include "both unwanted effects and any alteration from baseline that diminishes the
ability to survive, reproduce or adapt to the environment." Thus, it can be argued that these
effects in themselves warrant consideration as adverse health effects. They may also have
ramifications for health and safety of an even more serious nature, if a person's ability to
react quickly and accurately to a situation (e.g., traffic conditions) was impaired.
Another aspect of the findings from available occupational studies concerns the
temporal relationship between exposure and effect. As noted above, the geometric mean
average period of Mn exposure of the workers in the Roels et al. (1992) study was only
4 years, with the longest period of Mn exposure for any one individual being less than
18 years. Also, the oldest worker in the Roels et al. (1992) study was less than 50 years
old. This relatively limited period of exposure along with the absence of older subjects in
the Roels study raises the question of whether sufficient time had elapsed for the full
expression of the toxic effects of Mn. Some reports in the literature indicate that Mn toxicity
may not be clinically evident until some years after exposure occurred or terminated (e.g.,
Cotzias et al., 1968; Rodier, 1955), and other reports point to a greater sensitivity of elderly
persons, compared to middle-aged or young adults, for acute as well as chronic Mn toxicity
(e.g., Kawamura et al., 1941). An uncertainty factor for extrapolation from a subchronic
exposure to chronic exposure was included in deriving the RfC estimates shown in Figure 1.
However, a "half-factor" of 3 was used for this area of uncertainty. If the average period of
Mn exposure (geometric mean: 4 years) to the Roels et al. (1992) study is compared to an
assumed lifetime of 70 years, one could argue that a factor of 70/4 = 17.5, or at least 10,
27
-------
would be a more appropriate adjustment for subchronic to chronic exposures. Given the
limited available data, this area of uncertainty is difficult to express in a quantitative manner,
but in general practice, EPA has used an uncertainty factor of 3 for other chemicals with
comparable databases. Nevertheless, this area of concern suggests that the Mn RfC estimates
derived here with an uncertainty factor of 3 for subchronic to chronic exposure probably do
not tend to err in the direction of being too conservative.
Another qualitative concern that should be recognized in considering the potential health
effects of Mn is the possibility of certain types of effects that are suggested by clinical and
other evidence but are difficult to measure quantitatively or demonstrate with currently
available methods. Specifically, much of the clinical literature on manganism refers to a
psychiatric component of the illness, which often involves striking emotional or mood
changes that tend to appear before changes in motor function are evident. Such effects are
inherently difficult to measure in a quantitative manner. However, the possibility of such
effects at lower levels of exposure than those at which motor control is affected should not
be discounted out of hand. Some reports in the literature (e.g., Gottschalk et al., 1991)
suggest that aggressive behavior may be associated with Mn exposure (as reflected in the
concentration of Mn in hair of prison inmates). Such reports require substantiation by
further studies, and the validity and relevance of hair Mn levels to environmental Mn
exposure remains to be established, but the suggestion that an association might exist between
hair Mn and behavior cannot be totally dismissed.
Quantitative analyses of neurobehavioral data obtained from an occupational cohort
provide a range of possible RfC estimates in addition to the current, verified Mn RfC value
of 0.05 /ig/m . In ORD's judgment, the leading candidates for a possible alternative RfC
estimate are approximately 0.09 to 0.2 ^g/m , based on currently available information.
By definition, RfC analyses do not yield a precise concentration that defines a demarcation
between safety and hazard. Rather, interpretation of a Mn RfC estimate is best made in
relation to an assessment of population exposures to Mn, with the understanding that the RfC
is a protective level, not a predictive value.
The exposure assessment, based largely on data from the PTEAM study, provides some
reasonable but necessarily uncertain estimates of personal exposure levels of Mn that might
result from the use of MMT in gasoline. These estimates indicate that if MMT (at 1/32 g
28
-------
Mn/gal) were used in all unleaded gasoline in Riverside, CA (or the greater Los Angeles
metropolitan area), approximately 40 to 50% of the population could experience PM4
Mn exposures exceeding the current RfC of 0.05 /*g/m3 (derived from PM5 Mn health
effects data), and approximately 5 to 10% could experience PM4 Mn exposure levels around
3
0.1 /Ag/m or higher (see Figure 2). In terms of the Los Angeles area population of
14.5 million persons, even an estimate of 5% of the population implies over
700,000 persons.
Uncertainties are inherent in any risk assessment. In this case, on the health assessment
side, the numerical uncertainty factors used in the RfC analyses presented here have been
explicitly described and explained in Appendix A and summarized above. These factors are
intended to provide a reasonable degree of public health conservatism reflecting areas of
biological knowledge as well as areas of information deficit. In addition, an RfC estimate by
definition reflects uncertainty spanning perhaps an order of magnitude, and thus there is no
significant difference between the verified RfC of 0.05 j*g/m and alternative estimates of
3
0.09 to 0.2 jtg/m . Other qualitative uncertainties are also discussed above.
On the exposure assessment side, the primary uncertainties are related to projections
from the PTEAM data, rather than the PTEAM data per se. Inferences about the relative
contributions of crustal and automotive sources to PM4 Mn were drawn from studies
conducted in geographical and temporal proximity to the PTEAM study, but both the data
from these studies and the inferences based on them introduce uncertainties. Attempts to
adjust the PTEAM data in various ways, including any extrapolation from the 24-h average
distribution obtained in the fall of 1990 to a long-term average for other seasons, introduce
progressively greater uncertainties at each step. Some adjustments have not even been
attempted. For example, a weighting of the daytime and nighttime PTEAM exposure data to
reflect a higher average ventilation rate during daytime activities and a lower ventilation rate
during nighttime activities (e.g., sleeping) would have resulted in higher personal exposure
estimates. Thus, given other approaches or assumptions, different projection estimates are
possible. It must therefore be emphasized that the two projections of Mn exposure levels in
Figure 2 should not be interpreted as upper and lower bound estimates, for even the higher
projection could possibly underestimate, or the lower projection overestimate, the PM4 Mn
exposure levels associated with MMT.
29
-------
As for the relevance of the PTEAM Riverside personal exposure estimates to other
communities, the PTEAM study was, strictly speaking, only designed to statistically
represent Riverside, CA. In that respect, the design and conduct of the PTEAM study
provide a high degree of confidence that it does accurately represent 24-h average Mn
exposure concentrations for the Riverside population in the fall of 1990. In ORD's
judgment, the PTEAM study provides a reasonable representation of the Los Angeles Basin
as well, given the commonalities in geography, vehicle usage, and meteorology. However,
the relevance of the Riverside data to other U.S. communities depends upon their similarities
or differences in the most relevant characteristics or dimensions. For example, to the extent
that several other major U.S. metropolitan areas (or subcommunities in these areas) also have
a high level of vehicle usage, the Riverside projections may have greater relevance. To the
extent that these same areas do not share the meteorological conditions that contribute to the
Mn exposure levels measured in Riverside, the Riverside projections have lesser relevance.
It is also important to consider other sources of Mn exposure apart from automotive and
crustal sources. Although Riverside had no major point sources of Mn contributing to the
personal exposure levels measured in the PTEAM study, other communities may have such
sources, and thus the personal exposure levels of Mn from all sources might be higher in
other communities than in Riverside.
The exposure estimates shown in Figure 3 are in the range of or exceed some candidate
RfC estimates as well as the current RfC. Exceeding the RfC does not necessarily indicate
that a public health risk will occur. At present, it is impossible to state whether projected
exposures above the RfC would result in an adverse health effect for either an individual or
the general population. At a sufficiently high level of exposure, adverse effects would be
expected to occur, first in any sensitive subpopulations, then with greater prevalence in the
general population and extending to other types of effects (e.g., reproductive and/or
respiratory as well as neurobehavioral effects in the case of Mn). However, the relationship
between such "sufficiently high" levels and the population exposure levels estimated by the
projection methods employed here is unknown. Expressed differently, given the gap between
observed or modeled effect levels and the RfC values obtained by applying uncertainty
factors of orders of magnitude, it is impossible to state whether projected population
exposures would lie above or below a presumed threshold level on the actual
30
-------
* * *
0.1
0.1
1
i
0.01
0.01
0.001
I
Current
Verified
RfC
ORD
Alternative RfC
Estimates
0.001
Figure 3. Summary of Mn RfC estimates and personal exposure levels of Mn.
Figure 2 are overlaid, with both medians shown as horizontal marks.
PM4 Exposure
Estimates
Assuming 100%
MMT usage
The two PM4 exposure distributions from
-------
concentration-response curve for Mn neurotoxicity. This gap between projected exposure
levels and the lowest concentrations obtained by modeling the concentration-response
relationship (at least, by the quanta! linear model) makes it impossible to make any assertion
regarding the likelihood of a health risk at projected exposure levels. However, this
conclusion should not be interpreted to imply that, therefore, no health risk is expected to
exist at exposure levels exceeding the RfC.
5. SUMMARY AND CONCLUSIONS
From a hazard identification perspective, Mn clearly is toxic to the CNS, the
respiratory system, and the male reproductive system. Quantitative analyses of health effects
data obtained with healthy adult workers provide a range of possible Mn RfC estimates that
vary depending on various assumptions and choices among mathematical models, exposure
variables, and uncertainty factors. In OKD's judgment, the leading candidate alternative Mn
RfC estimates, based on currently available information, are approximately 0.09 to
0.2 jig/m . This estimate of a protective level is inherently imprecise, and reflects only the
data and analyses considered to date.
Projections of estimated personal exposure levels to PM4 Mn, based on data from the
PTEAM study and other sources, indicate the possibility for substantial numbers of persons
to be exposed to PM4 Mn concentrations above 0.1 jtg/m if MMT were widely used in
gasoline. Although it is impossible to state whether a health risk would definitely exist at the
projected exposure levels, neither can the possibility of such a risk be ruled out. If a health
risk exists, the young, the elderly, and persons with certain preexisting conditions, such as
Parkinson's disease, may be at relatively greater risk. Also, populations of persons who Ive
or work near major vehicle emissions sources would be likely to experience higher exposure
levels and hence be at potentially greater risk.
Given the information that is available at present and the uncertainties discussed here, a
reasonable basis exists for concern regarding potential public health risks, especially for
sensitive subpopulations, if MMT were to be used widely in unleaded gasoline. Although all
risk assessments have some degree of uncertainty, in some cases it is reasonable to conclude
that the risk of adverse health effects is either very great or very small because estimated
exposure levels are either far above or far below a potential health effect level. However,
32
-------
this is not the case with MMT. If a more definitive estimate of risk is required, it is
essential that more relevant information be obtained on the health effects of and exposure to
Mn3O4. The types of information and research needed to provide a more definitive estimate
of risk were previously identified by ORD (U.S. Environmental Protection Agency, 1991).
Until those research and information needs are met, a high degree of uncertainty will attend
any judgment of the potential risks to public health associated with the use of MMT.
33
-------
6. REFERENCES
Allen, B. C,; Kavlock, R. J.; Kimmel, C. A.; Faustman, E. M. (1994) Dose-response assessments for
developmental toxicity: II. Comparison of generic benchmark dose estimates with NOAELS. Fundam.
Appl. Toxicol.: in press.
American Conference of Governmental Industrial Hygienists. (1992) 1992-1993 threshold limit values for
chemical substances and physical agents and biological exposure indices. Cincinnati, OH: American
Conference of Governmental Industrial Hygienists, Technical Information Office; pp. 40-45.
Aschner, M.; Gannon, M. (1994) Manganese (Mn) transport across the rat blood-brain barrier: saturable and
transferrin-dependent transport mechanisms. Brain Res. Bull. 33: 345-349.
Barnes, D. G.; Daston, G. P.; Evans, J. S.; Jarabek, A. M.; Kavlock, R. J.; Kimmel, C. A.; Park, C.; Spitzer,
H. L. (1994) Benchmark dose workshop: criteria for use of a benchmark dose to estimate a reference
dose. Regul. Toxicol. Pharmacol.: submitted.
Clayton, C. A.; Perritt, R. L.; Pellizzari, E. D.; Thomas, K. W.; Whitmore, R. W.; Ozkaynak, H.; Spengler,
J. D.; Wallace, L. A. (1993) Particle total exposure assessment methodology (FTEAM) study:
distributions of aerosol and elemental concentrations in personal, indoor, and outdoor air samples in a
southern California community. J. Exposure Anal. Environ. Epidemiol. 3: 227-2%.
Cotaas, G. C,; Horiuchi, K.; Fuenzalida, S.; Mena, L (1968) Chronic manganese poisoning: clearance of tissue
manganese concentrations with persistence of the neurological picture. Neurology 18: 376-382.
Crump, K. S. (1984) A new method for determining allowable daily intakes. Fundam. Appl. Toxicol.
4: 854-871.
Ethyl Corporation. (1994) Comments of Ethyl Corporation in response to EPA's December 9, 1993 Federal
Register notice 58 Fed. Reg. 64761 (1993). Washington, DC: Ethyl Corporation, Office of the Vice
President for Government Relations. Available for inspection at: U.S. Environmental Protection Agency,
Central Docket Section, Washington, DC: docket no. A-93-26.
Faustman, E. M.; Allen, B.C.; Kavlock, R.J.; Kimmel, C.A, (1994) Dose-response assessment for
developmental toxicity: I. characterization of data base and determination of NOAELs. Fundam. Appl.
Toxicol.: in press.
Federal Register. (1993) Draft report: principles of neurotoxicology risk assessment. F. R. (August 4)
58: 41556-41599.
Gottschalk, L, A.; Rebello, T; Buchsbaum, M. S.; Tucker, H. G.; Hodges, E. L. (1991) Abnormalities in hair
trace elements as indicators of aberrant behavior. Compr. Psychiatry 32: 229-237.
Iregren, A. (1990) Psychological test performance in foundry workers exposed to low levels of manganese.
NeurotoxicoL TeratoL 12: 673-675.
IRIS, Integrated Risk Information System [data base]. (1993) [Printout of reference concentration (RfC) for
chronic manganese exposure as revised November, 1993]. Cincinnati, OH: U.S. Environmental
Protection Agency, Office of Health and Environmental Assessment, Environmental Criteria and
Assessment Office, Available online from: TOXNET, National Library of Medicine, Rockville, MD.
34
-------
Jarabek, A. M.; Hasselblad, V. (1991) Inhalation reference concentration methodology: impact of dosimetric
adjustments and future directions using the confidence profile method. Presented at: 84th annual meeting
and exhibition; June; Vancouver, British Columbia, Canada. Pittsburgh, PA: Air and Waste Management
Association; paper no. 91-173,3.
Kawamura, R.; Ikuta, H.; Fukuzami, S.; Yamada, R.; Tsubaki, S.; Kodama, T. (1941) Intoxication by
manganese in well water. Kitasato Arch. Exp. Med. 18: 145-169.
Kimmel, C. A.; Gaylor, D. W. (1988) Issues in qualitative and quantitative risk analysis for developmental
toxicology. Risk Anal. 8: 15-20.
Kirahenblatt, M. (1993) [Letter to Ms. M. T. Smith, U.S. EPA, regarding the August 30, 1993, rebuttal
comments from Ethyl Corporation in support of their waiver request for HTTEC 3000 fuel additive.]
Ottawa, Ontario, Canada: Environment Canada, Transportation Systems Division, Industrial Programs
Branch; September 23.
Lyons, J. M.; Venkataraman, C.; Main, H. H.; Friedlander, S. K. (1993) Size distributions of trace metals in
the Los Angeles atmosphere. Atmos. Environ. Part B 27B: 237-249.
Mergler, D.; Huel, G.; Bowler, R.; Iregren, A.; Belanger, S.; Baldwin, M.; Tardif, R.; Smargiassi, A.; Martin,
L. (1994) Nervous system dysfunction among workers with long-term exposure to manganese. Environ.
Res. 64: 151-180.
Murphy, V. A.; Wadhwani, K. C.; Smith, Q. R.; Rapoport, S. I. (1991) Saturable transport of manganese (II)
across the rat blood-brain barrier. J. Neurochem. 57: 948-954.
Pellizzari, E. D.; Thomas, K. W.; Clayton, C. A.; Whitmore, R. W.; Shores, R. C.; Zelon, H. S.; Perritt,
R. L. (1992) Particle total exposure assessment methodology (PTEAM): Riverside, California pilot study,
volume I [final report]. Research Triangle Park, NC: U.S. Environmental Protection Agency,
Atmospheric Research and Exposure Assessment Laboratory; EPA report no. EPA/600/R-93/050.
Available from: NTIS, Springfield, VA; PB93-166957/XAB.
Preuss, P. W. (1991) ORD's comments on Ethyl Corporation's July 12, 1991 resubmittal of a waiver application
for the use of methylcyclopentadienyl manganese tricarbonyl (MMT) in unleaded gasoline [memorandum
to Richard Wilson]. Washington, DC: U.S. Environmental Protection Agency, Office of Research and
Development; December 12.
Rodier, J. (1955) Manganese poisoning in Moroccan miners. Br. J. Ind. Med. 12: 21-35.
Rods, H. (1993) [Letter to Dr. M. Davis, U.S. EPA, on definitions of "respirable," "total," and "inhalable"
dusts]. Bruxelles, Belgium: Universite Catholique de Louvain, Unite de Toxicologie Industrielle et
Medecine du Travail; October 19.
Roels, H.; Lauwerys, R.; Genet, P.; Sarhan, M. J.; de Fays, M.; Hanotiau, L; Buchet, J.-P. (1987)
Relationship between external and internal parameters of exposure to manganese in workers from a
manganese oxide and salt producing plant. Am. J. Ind. Med. 11: 297-305.
Roels, H. A.; Ghyselen, P.; Buchet, J. P.; Ceulemans, E.; Lauwerys, R. R. (1992) Assessment of the
permissible exposure level to manganese in workers exposed to manganese dioxide dust. Br. J. Ind. Med.
49: 25-34.
35
-------
Thomas, K, W.; Pellizzari, E. D.; Clayton, C. A.; Whitaker, D. A.; Shores, E. C.; Spengler, J. D,; Ozkaynak,
H.; Wallace, L. A. (1993) Particle total exposure assessment methodology (PTEAM) study: method
performance and data quality for personal, indoor, and outdoor aerosol monitoring at 178 homes in
southern California. J. Exposure Anal. Environ. Epidemiol. 3: 203-226.
Tukey, J. W.; Ciminera, J. L.; Heyse, J. F. (1985) Testing the statistical certainty of a response to increasing
doses of a drug. Biometrics 14: 295-301.
U.S. Environmental Protection Agency. (1990) Comments on the use of methylcyclopentadienyl manganese
tricarbonyl in unleaded gasoline. Washington, DC: Office of Research and Development.
U.S. Environmental Protection Agency. (1991) Information needed to improve the risk characterization of
manganese tetraoxide (MnjO^ and methylcyclopentadienyl manganese tricarbonyl (MMT). Washington,
DC: Office of Research and Development; December 12.
Wallace, L. A.; Duan, N.; Ziegenfus, R. (1994) Can long-term exposure distributions be predicted from short-
term measurements? Risk Anal. 14: 75-85.
Wennberg, A.; Iregren, A.; Struwe, G.; Cizinsky, G.; Hagman, M.; Johansson, L. (1991) Manganese exposure
in steel smelters a health hazard to the nervous system. Scand. J. Work Environ. Health 17: 255-262.
Wennberg, A.; Hagman, M.; Johansson, L. (1992) Preclinica! neurophysiological signs of parkinsonism in
occupational manganese exposure. Neurotoxicology 13: 271-274.
36
-------
APPENDIX A
DOSE-RESPONSE ASSESSMENT ANALYSES
1. INTRODUCTION
This report describes certain statistical analyses of possible relevance to the derivation
of an inhalation reference concentration (RfC) for manganese (Ma). An RfC is defined as an
estimate (with uncertainty spanning about an order of magnitude) of a continuous exposure
level for the human population (including sensitive subpopulations) that is likely to be
without appreciable risk of deleterious noncancer effects during a lifetime.
The discussion herein focuses on the results of the Roels et al, (1992) study, which was
used as a principal study for the derivation of the current, verified inhaktion RfC for Mn
(IRIS, 1993). The Roels et al. (1992) study measured neurobehavioral function in 101
matched control "nonexposed" workers (actually, exposed to unmeasured but presumably
very low levels of Mn) and 92 alkaline battery plant workers who were exposed to levels
ranging from 40 to 4,433 jtg Mn/m3 x years. The specific health endpoints included
(1) visual simple reaction time, (2) hand steadiness, and (3) eye-hand coordination.
This appendix considers four types of statistical approaches to evaluating the
information from Roels et al. (1992): (1) a conventional no-observed-adverse-effect level
(NOAEL)- or lowest-observed-adverse-effect level (LQAEL)-based approach, (2) a "no
statistical significance of trend" (NOSTASOT) approach, (3) a benchmark dose approach,
and (4) a Bayesian approach. These approaches are described in the following sections,
followed by a discussion of several considerations involved in deriving RfC estimates by
alternative analyses.
A fundamental issue pertaining to all of the approaches presented here is the selection
of a measure of exposure. Roels et al. (1992) measured both total and respirable dust.
Subsequent information provided by Roels (1993) indicated that respirable dust was
equivalent to PM5 (i.e., sampler 50% cut point of 5 /tm; maximum size collected 7 jon) and
that total dust included much larger particles, perhaps as large as 35 jtm in diameter.
Because respirable dust is more relevant toxicologically, it was used in these dose-response
analyses. Roels et al. (1992) described two measures of respirable dust, the occupational
A-l
-------
lifetime integrated respirable dust concentration (JURD), expressed as pgltv? x years, and
the current concentration of respirable dust (CRD), expressed as /ig/m . The CRD
concentration was measured at the time the study was conducted by Roels et al. and refers to
a representative concentration measured for the type of job performed by a worker (e.g.,
electrician, maintenance worker). The LIED value for each worker was constructed from
the CRD value by summing over the worker's entire period of employment. If a worker
changed jobs within the plant during his period of employment, the CRD for each job held
was multiplied by the number of years the worker performed that job. Thus, if more than
one job classification was worked, the worker's ORD was the sum of the products of CRD
multiplied by years of performance of the respective jobs. However, if a worker held only
one job classification, his LERD was simply equal to his CRD multiplied by the total number
of years employed. Another measure of exposure was proposed by Ethyl Corporation (1994)
in a submission from Ethyl Corporation. This third measure may be derived from LERD by
dividing a worker's LERD value by the total number of years of his employment. The latter
measure, designated as the average concentration of respirable dust (ACRD), integrates over
a worker's various exposure levels but removes years from the units of measurement in
LERD and is expressed as ^g/m . The ACRD can be thought of as a tune-weighted average
concentration. Although Roels et al. (1992) did not refer to ACRD, it is possible to
calculate this value for each individual and for the entire cohort by using unpublished data
(see Attachment A-l) provided to EPA by Roels (1993).
2. NOAEL-LOAEL APPROACH
The most common approach to derivation of dose-response estimates for noncancer
toxicity is to determine a NOAEL and/or a LOAEL based on available experimental or
epidemiological data. These are determined for a specified adverse effect from the exposure
levels of a given individual study. A NOAEL is defined as the highest exposure level at
which neither a biologically nor statistically significant increase in the frequency or severity
of the specified adverse effect is produced and is therefore, by definition, intended to be a
subthreshold level. Similarly, a LOAEL is defined as the lowest exposure level that
produces a biologically or statistically significant increase in the frequency or severity of the
specified adverse effect. In the case of continuous exposure data (e.g., from epidemiologic
A-2
-------
studies) that are categorized into subgroupings, the best available estimate of a representative
exposure level for a subgroup may be used appropriately as a NOABL or LOAEL. The
entire data array of individual NOAEL and LOAEL values for a chemical is then evaluated
to determine a critical effect. This effect is often the one with the lowest NOAEL.
The NOAEL and LOAEL values for each of the tests used by Roels et al. (1992)
(eye-hand coordination, visual reaction time, and hand steadiness) were derived by
calculating the mean values of the exposure groupings (<600, 600 to 1,200, and
> 1,200 jtg/m x years) based on individual exposure and response date (see
Attachment A-l) supplied by Roels (1993). Given the significance of the increased
prevalence of abnormal scores for eye-hand coordination in the lowest exposure group in
Table A-l, 338 pglm x years is designated as a LOAEL, and eye-hand coordination is the
critical effect. A NOAEL for the data array is not evident by this analysis.
TABLE A-l. PREVALENCE OF ABNORMAL RESPONSES IN WORKERS
EXPOSED TO MANGANESE DUST3
Ecdpoiut
Simple reaction time
Hand steadiness
Eye-hand coordination
Control
(O)
1/101
4/101
5/101
Mn Concentration (fig/m x years)
<600 600-1,200
(3381*) (8421*)
1/30 2/26
3/30 3/26
6/30c 1/26*
> 1,200
(2,182b)
4/36e
7/36°
8/36
BBased on Figure 3 of Roels et al. (1992). Data are expressed as number of subjects showing
an adverse response in relation to the total subjects within a concentration grouping.
Mean of each exposure grouping calculated from data provided by Rods (1993).
°Lowest concentration producing a statistically significant (p = 0.05) difference from controls using & one-sided
Fisher's exact test and Duunett's correction for multiple comparisons.
dLowest concentration producing a statistically significant (p = 0.05) difference from controls using a two-sided
Fisher's exact test and Dunuett's correction for multiple comparisons.
As noted in the text for the Mn RfC, a correction for multiple comparisons was not
applied by Roels et al. (1992) in the published report of their study. However, post hoc
analyses of the data provided by Roels (1993) indicated that statistical significance by the
Dunnett test was achieved only for the one-sided test. The lack of significance by the
A-3
-------
two-sided test yields a statistical NOAEL, but it does not enhance one's confidence that the
difference between a LOAEL and a NOAEL in Reels' data, as determined by this approach,
is as distinct as perhaps implied by the terms "significant" and "nonsignificant", lather than
relying upon the choice between a one-sided or two-sided test of statistical significance to
determine whether a NOAEL exists in these data, other analytic approaches to estimating a
NOAEL or NOAEL-surrogate are considered below,
3. NOSTASOT APPROACH
Another approach is to use the NOSTASOT test as defined by Tukey et al. (1985).
It represents a statistically more accurate method of estimating a NOAEL when several dose
levels are available. The idea is to sequentially test for a linear trend until statistical
significance is no longer reached. As described by Tukey et al. (1985), the procedure is
applied to all of the data first and then entails sequentially deleting the highest dose groups in
succession downward (i.e.» "top-down"). Although apparently intended for use with
experimental studies with multiple dose groups, the NOSTASOT procedure may be applied
to epidemiological data by deleting exposure groups or, in the case of individual exposure
data, each subject (or group of subjects having identical exposure levels). The NOSTASOT
procedure includes the Dunnett test to correct for multiple comparisons between each
exposure level and control. However, because of the continuous nature of Reels' (1993)
exposure variable, Dunnett's test was not appropriate for the analyses in Tables A-2 to A-4.
Also, because of the dichotomous nature of the response variable, linear regression could not
be applied, and therefore the Cochran-Armitage test for linear trend in proportions was used
for the top-down analyses. According to Margolin (1988), this test has advantages for
testing for a monotonic trend over certain other tests, unless there is a substantial downturn
in response at high exposure levels.
An alternative to the top-down approach to NOSTASOT is to start at the lowest
noncontrol exposure or dose level and move upward (i.e., "bottom-up"). The objective is to
determine the highest level of nonsignificance before a significant difference is detected. The
highest nonsignificant dose is declared the NOSTASOT. The bottom-up NOSTASOT was
performed with and without a correction for multiple comparisons. For LERD, significance
was judged at the 0.05 level and at the 0.0006 level (0.05 divided by 82, because there were
A-4
-------
82 different levels of LIRD based on the individual data of Reels [1993], obtained by using a
Bonferroni correction for multiple comparisons). The calculations were repeated for CRD
using a Bonferroni correction of 0.002 (i.e., 0.05 divided by 21 rather than 82, because there
were 21 different CRD levels). The ACRD calculations were adjusted in the same manner
as for LIED. A summary of the results is shown in Tables A-2 through A-4 for LIRD,
CRD, and ACRD, respectively.
TABLE A-2. NOSTASOT VALUES FOR INDIVIDUAL LIRD Otg/nT x years)
EXPOSURE DATA OF RQELS (1993)
Exposure below the Lowest
First Nonsignificant Exposure Last Nonsignificant*1 Significant Exposure after
Deleting Data from the Exposure Adding Data from Adjustment for Multiple
Highest Exposure Down the Lowest Exposure Up Comparisons
Visual reaction time
Eye-hand
coordination
Hand steadiness
2,651
2,421
1,389
163
55
516
163
604
3,162
Rp < 0.05.
p < 0.0006 by Bonferroni correction.
TABLE A-3. NOSTASOT VALUES FOR INDIVIDUAL ORD Oig/m3) EXPOSURE
DATA OF ROELS (1993)
Exposure below the
First Nonsignificant Last Nonsignificant8 Lowest Significant
Exposure Deleting Data from Exposure Adding Data from Exposure after Adjustment
the Highest Exposure Down the Lowest Exposure Up for Multiple Comparisons
Visual reaction time
Eye-hand
coordination
Hand steadiness
319
104
634
46
21
21
46
104
46
p < 0.05.
p < 0.002 by Bonferroni correction.
The NOSTASOT method can be very sensitive to both sample size and dose (exposure)
spacing. For example, consider the following situation. If there are a few intermediate
A-5
-------
TABLE A-4. NOSTASOT VALUES FOR INDIVIDUAL ACRD Oig/m3) EXPOSURE
DATA OF ROELS (1993)
Exposure below the
First Nonsignificant Last Nonsignificant Lowest Significant
Exposure Deleting Data from Exposure Adding Data from Exposure after Adjustment
the Highest Exposure Down the Lowest Exposure Up for Multiple Comparisons
Visual reaction time 468 21 21
Eye-hand coordination 158 1S8 174
Hand steadiness 434 46 S24
ap < 0.05,
p < 0,0006 by Bonferroni correction.
exposures, but the response rate at the high exposures is much larger than the background
rate, the NOSTASOT method will not find statistical significance until it has reached a large
number of the observations in the high-exposure range. Thus, the NOSTASOT level will be
estimated as a level where there is a very high response rate. On the other hand, when using
benchmark or Bayesian approaches (see Sections 4 and 5 below), any model that is close to
linear will be influenced by this high exposure level so that when such a model is used to
estimate a concentration producing a small rate increase, estimates much below the
NOSTASOT level and having relatively narrow confidence bounds result. This can be
appreciated by examination of the resultant RfC estimates calculated by the NOSTASOT
procedure as opposed to the estimates obtained by the other procedures shown in Table A-39
at the end of this appendix.
The sensitivity to sample size and dose spacing of the NOSTASOT procedure is also
illustrated by the difference between the top-down versus the bottom-up estimates. In most
animal experiments, for which the procedure was developed, with exposures typically only at
two, three, or four levels, the NOSTASOT will be the same whether analyzed from the top
down or the bottom up. Application to epidemiologic data is more complicated. First, it
must be assumed that there are sufficient data to estimate the control response rate quite
accurately. If a few individuals with the very lowest nonzero exposures happen to be
responders, a large number of middle-exposure individuals are nonrcsponders, and high-
exposure individuals are nearly all responders, then the NOSTASOT procedure from the
bottom up will find statistical significance very early, but as subjects are added from the
middle exposure range, this statistical significance may disappear. When enough subjects
A-6
-------
from the high exposure level are added, the statistical significance will reappear. When
calculated from the top down, statistical significance would again be found among the
responders in the high-exposure category. This appears to be part of the explanation for the
different results obtained by the bottom-up and top-down approaches and may also explain
the differences in endpoint sensitivity for the NOSTASOT analysis.
4. BENCHMARK DOSE APPROACH
Ciump (1984) and others have proposed using a "benchmark dose" (BMD) approach to
health effects assessment to address certain problems in using NOAELs or LOAELs. These
problems include (1) the designation of the effect levels does not readily account for the
number of test subjects or sample size, (2) the slope of the dose-response curve is generally
ignored, and (3) the estimate is typically presented as a single number that does not explicitly
express the range of variability.
A BMD is calculated from a specified (adverse) health effect. For example, a 10%
increase in the rate of a specific lesion in rats could be the specified health effect. From this
value, an estimate of the dose that will produce the specified effect is calculated (usually
using maximum likelihood estimation [MLE| procedures) and this value is denoted as the
BMD. In addition, a lower confidence limit (usually the lower 95th) of this dose is
calculated and denoted as the BMD lower limit (BMDL). The BMDL has been proposed as
a substitute for the NOAEL in the derivation of an RfC (Barnes et aL, 1994).
Confidence intervals for nonlinear models are usually calculated from the likelihood
function, which is assumed to be approximately normal. The variances are calculated from
the second partial derivatives and the confidence intervals are calculated assuming normality.
However, the calculation of BMDs results in likelihood functions that are very nonnormal in
shape. Thus, confidence intervals calculated assuming normality can be incorrect by several
orders of magnitude. Crump (1984) recognized this problem and constructed the confidence
limits using the asymptotic (large sample) distribution of the likelihood ratio (see Cox and
Hinkley, 1992). This method descends down the likelihood surface until the ratio of the
likelihoods reaches a specified value. Nuisance parameters are allowed to move to maximize
the likelihood. This method also yields an approximate confidence interval, but the method
does not depend on the asymptotic normality (i.e., that which only approaches normality for
A-7
-------
large sample sizes) of the likelihood functions. However, there is no way of knowing the
accuracy of the confidence interval for any given data set.
Although the BMD approach eliminates many of the problems associated with the
NOAEL/LQAEL approach, it also has some limitations of its own: (1) the BMD carries
with it no estimate of uncertainty; (2) there is no obvious method for comparing or
combining BMDs calculated from similar studies using the same specified health effect; and
(3) because confidence intervals for certain parameters are difficult to compute and are based
on asymptotic (large sample) theory, there is no way of knowing if the sample sizes of a data
set are large enough to ensure the accuracy of the asymptotic results without performing
thousands of simulations.
One key point in the calculation of a BMD is the choice of a model. In the report
submitted by Ethyl Corporation (1994), they stated that "These calculations were made using
the Weibull model with no threshold." This model can be written as
P (response) = 7 + (1 - 7) [1 - exp(-/3jc")], (A-1)
where x is the exposure concentration and a, j8, and 7 are parameters to be estimated.
In particular, 7 is the nonzero background rate of response. Normally a and /3 are restricted
to be positive. Actually, Ethyl Corporation (1994) used a restricted Weibull model, where a
is restricted to be greater than or equal to 1 .
Several other models including special cases of the Weibull model may also be used
(Crump, 1984). One of these is the quanta! linear model, defined as
P(response) = 7 + (1 - 7) [1 - exp (-£*)], CA'2)
where 0 is a slope to be estimated. This is a special case of the Weibull model with
a restricted to 1.
A similar model is the quanta! quadratic model, defined as
A-8
-------
P(response) = 7
where jS is a slope to be estimated. This is also a special case of the Weibull model, but
with a restricted to 2.
Finally, a model used often by epidemiologists is the log-logistic model,
P(response) -
where $(x) is the cumulative logistic distribution and where a and £ are parameters to be
estimated. Normally 0 is restricted to be positive. As was done with the Weibull model,
Cramp (1984) restricted ft to be greater than one.
The restriction of a (or jS) s 1 for the Weibull (or log-logistic) models appears to be
somewhat arbitrary, and may have its origins in EPA's approach to cancer risk assessment.
This type of restricted structure was chosen for cancer assessment procedures because its
attributes were applicable to the available tumor bioassay data, were consistent with
prevailing views of pathogenesis that may now be inappropriate for certain chemicals given
recent mechanistic insights, and were protective of the public health. As more mechanistic
data, cell proliferation data, pharmacokinetic data, and insights on indirect mechanisms of
pathogenesis become available, these attributes of the linearized multistage model may be
inappropriate. Indeed, "thresholds" and different modeling structures are now being
considered for carcinogens. As an example, the Moolgavkar-Venzon-Knudson model
(Knudson et al., 1975; Moolgavkar and Venzon, 1979) is being used to better describe data
for certain types of chemicals based on mechanistic insights into the pathogenesis process.
Thus, a general mathematical convenience, particularly one that is based on historical
precedent that may not be applicable to the type of data at hand, should not take priority over
fitting the observed data. Supralinear models (those with a < 1) are very plausible to mimic
data that exhibit saturation phenomena or that may result from differences in affected
subpopulations.
Another, unstated reason for the use of restricted models is that it eliminates certain
computational difficulties. Approximate lower confidence bounds are difficult to compute
when the slope is very steep near the origin. This difficulty does not imply that the model is
A-9
-------
incorrect but rather that the statistical methods have difficulty. (One solution to this problem
is to use Bayesian methods [see below], which yield exact results [subject to numerical
accuracy],) Also, it should be noted that when a logarithmic scale is used for the x-axis
(which is common), the slope is finite everywhere for all a and ft, and the standard S-shaped
sigmoidal curve results. Thus, there is no reason a priori to rule out the use of the
unrestricted Wiebull and log-logistic models. Indeed, many more models in addition to the
six selected here could have been considered.
From the results shown in Table A-l, eye-hand coordination is the most sensitive
endpoint by the Fisher's exact test. The eye-hand coordination data were therefore used to
explore a number of different models for the benchmark analyses. Results for the other two
endpoints, visual simple reaction time and hand steadiness, are included at the end of this
section.
The models described earlier can be fitted to the eye-hand coordination data of Roels
et al. (1992, Fipre 3). The Weibull and log-logistic models can be fitted with and without
restrictions. The results of fitting these models using IIRD as the measure of exposure are
in Table A-5. Graphs of the fits are shown in Figures A-l through A-6.
The results in Table A-5 show that not all models give equivalent results. First, note
that the restricted Weibull and the quantal linear models give identical results. This occurs
when the optimal value of a (given that a ^ 1) is a = 1. Second, only the unrestricted
Weibull and log-logistic models provide good fits to the data. A good fit is one which
produces a small chi-squared value, which in turn leads to a p-value much greater than 0.05.
In fact, it is possible to test whether the a parameter of the Weibull model (or the
j8 parameter of the log-logistic model) is equal to 1, versus the alternative that a (or ft) are
less than 1. The tests are made using likelihood ratio statistics, the same statistics that
Crump (1984) used to calculate lower confidence bounds. For this problem, the likelihood
ratio test for the Weibull model was 5.242 (p = 0.0220) and for the log-logistic model was
4.645 (p = 0.0311). Thus, the hypothesis that the parameter of interest is greater than or
equal to 1 can be rejected for both models. This calls into question one of the basic
assumptions of the restricted models used by Ethyl Corporation.
A-10
-------
TABLE A-5. SUMMARY OF THE FIT TO THE EYE-HAND DATA OF
ROELS ET AL. (1992, FIGURE 3) USING LERD GROUPED EXPOSURE
FOR SIX DIFFERENT MODELS
Model
Quantal linear
Quanta! quadratic
Weibull
(unrestricted)
Weibull
(oil)
Log-logistic
(unrestricted)
Log-logistic
0*1)
Parameter
Estimates
0 = 0.000133
7 = 0.0702
|8 =0.0000000404
y = 0.1009
0 = 0.1383
a = 0.0600
7 = 0.0495
j8 = 0.000133
a = 1
7 = 0.0702
0 = 0.0671
a = -1.921
7 = 0.0495
0-1
a = -7.711
7 = 0.0656
Chi-Squared
Goodness of Fit
(p-value)
5.11
(0.0777)
9.602
(0.0082)
0.28
(0.5961)
5.11
(0.0220)
0.28
(0.5963)
4.40
(0.0359)
Log-Likelihood
-71.9274
-74.1074
-69.3064
-71.9274
-69.3062
-71.6289
Finally, it should be noted that the unrestricted Weibull and log-logistic models are very
similar for small concentrations. This is not surprising because
[a + /31og(x)] =
(for small x) e"*"*** = a'x" (where a'
• 1 - e-'*'.
(A-5)
Thus, there is a correspondence between (1) ea and /? (0.1465 and 0.0671) in the log-logistic
model and (2) j8 and a (0.1383 and 0.0600) in the Weibull model. The estimates for 7 were
A-ll
-------
1.00!
0.75
0.50
h^
f
0.25
600 1,200 1.600
URD Mn (jig/m8 x years)
2,400
Figure A-l. Quantal linear model fitted to Roels (1993) eye-hand coordination data,
with LIRD as the exposure variable.
1.00
0.7&
0.50
0.25
600 1,200 1.800
URD Mn (mj/m3 x years)
2,400
Figure A-2. Quantal quadratic model fitted to Roels (1993) eye-hand coordination data,
with LORD as the exposure variable.
A-12
-------
1.001
0.75
§0.50
0.26
0 600 1,200 1,800 2,400
LJRD Mn (|ig/m» x years)
Figure A-3. Unrestricted Weibull model fitted to Roels (1993) eye-hand coordination
data, with LIRD as the exposure variable.
1.00
0.76-
-5
g 1
o 0.50
025
800 1,200 1.«00
URD Mn (ng/m3 x years)
2,400
Figure A-4. Restricted Weibull model fitted to Roels (1993) eye-hand coordination data,
with LIRD as the exposure variable.
A-13
-------
1.00-1
0.75
0.50
0.25
o-l-
600 1,200 1^00
LIRD Mn (ng/m3 x years)
2.400
Figure A-5. Unrestricted log-logistic model fitted to Roels (1993) eye-hand coordination
data, with LIRD as the exposure variable.
1.00
0.75
3
i
i
0.50-
OJS
600 1,200 1.800
LJRD Mn (|ig/m* x years)
2,400
Figure A-6. Restricted log-logistic model fitted to Roels (1993) eye-hand data, with
LIRD as the exposure variable.
A-14
-------
the same for both models, namely 0,0495. Clearly there is no way to distinguish these two
models on the basis of the observed data.
From these equations, it is possible to estimate the BMD directly. The BMD was
estimated for eye-hand coordination using the grouped results taken from Figure 3 of Roels
et al. (1992). Because no definitive guidelines exist for the definition of a BMD, three
different specified increase rates (i.e., MLE values at 0.10, 0.05, and 0.01) were calculated.
The estimates of the BMD for different models and at the different specified effect levels are
in Table A-6, and the estimates of the BMDL are in Table A-7. Note that these results show
no consistency. The values vary by several orders of magnitude depending on the model.
TABLE A-6. BMD VALUES FOR EYE-HAND COORDINATION DATA OF ROELS
ET AL. (1992, FIGURE 3) FOR LIRD (pg/m3 x yeare)
BY MODEL
Model
Quanta! linear
Quanta! quadratic
Weibull (unrestricted)
Weibull (a > 1)
Log-logistic (unrestricted)
Log-logistic 08 2 1)
AND SPECIFIED
Specified
0.10
793
1,615
0.01
793
0.02
674
EFFECT LEVEL
Increase in Rate
0.05
386
1,127
< 0.0001
386
<0.0001
319
of Abnormals
0.01
76
499
<0.0001
76
<0.0001
61
The exposure and response data for individual Mn-exposed workers in the study by
Roels et al. (1992) are listed in Attachment A-l. Actual individual test score results were
not provided by Dr. Roels, but an individual whose test score was abnormal, as specified by
Roels et al. (1992), is indicated by a "1". Because various parameters measured by each of
the three tests (eye-hand coordination, visual reaction time, and hand-steadiness) were not
strictly independent, only one representative parameter per test was selected. Roels et al.
(1992) designated eye-hand coordination abnormal when the percent precision value was
lower than the 5th percentile of the control group. Hand steadiness was considered abnormal
when the score for the 3.5-mm hole exceeded the 95th percentile value for the control group,
A-15
-------
TABLE A-7. BMDL VALUES FOR EYE-HAND COORDINATION DATA
OF ROELS ET AL. (1992, FIGURE 3) FOR LffiD Otg/m3 x years)
BY MODEL AND SPECIFIED EFFECT LEVEL
Model
Quanta! linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a ^ 1)
Log-logistic (unrestricted)
Log-logistic (0 Si 1)
Specified Increase in Rate
0.10 0.05
462 225
1,112 776
(*) (*)
462 225
< 0.0001 (*)
366 173
of Abnonnals
0.01
44
343
(*)
44
(*)
33
""Computational difficulties resulted from exponentiating values with large absolute magnitudes. The actual
results are much less than 1 and correspond to very low values for the same unrestricted cases in Table A-6.
and visual reaction time was considered abnormal when the subject's simple reaction time
measurements were increased for each 2-min examination period. The selection of these
cut-off values in order to dichotomize the continuous measurements may be viewed as
somewhat arbitrary.
For comparison to the benchmark analyses using grouped LERD exposures, all six
models from the eye-hand coordination data were rerun using individual URD exposure data.
These results are in Table A-8. The likelihoods for each model using the individual URD
data were similar but slightly better and the relative rankings of all models remained the
same. None of the conclusions about choice of model were affected. Summary cbi-squared
goodness of fit tests cannot be calculated, however, because the subjects are no longer
grouped into exposure intervals.
The BMD and BMDL values based on Table A-8 are shown in Tables A-9 and A-10.
These values are reasonably similar to those in Tables A-6 and A-7 except that the
unrestricted models give slightly larger estimates in Tables A-9 and A-10.
All of the above calculations were based on using the ORD exposure measure.
Because Roels also gave the CRD concentration for each subject, it is of interest to see the
impact of this variable on BMD and BMDL. Although the ORD variable may better reflect
A-16
-------
TABLE A-8. SUMMARY OF THE ITT TO THE EYE-HAND DATA OF
ROELS (1993) USING INDIVIDUAL LJRD Qtgfm3 x years) EXPOSURE
ESTIMATES FOR SIX DIFFERENT MODELS
Model
Quanta! linear
Quanta! quadratic
WeibuE
(unrestricted)
WeibuE
(<*S» 1)
Log-logistic
(unrestricted)
Log-logistic
Parameter
Estimates
jg = 0.000133
7 = 0.0716
ft = 0.0000000373
7 = 0.1011
0 = 0.0761
a - 0.1411
7 = 0.0496
ft = 0.000133
a = I
7 = 0.0716
ft - 0.1614
a = -2.550
7 = 0.0496
& i
p — i
a = -8.723
7 = 0.0680
Log-Likelihood
-71.2001
-73.0517
-69.1675
-71.2001
-69.1790
-71.0014
TABLE A-9. BMD VALUES FOR EYE-HAND COORDINATION DATA FROM
ROELS (1993) USING INDIVIDUAL URD (jtg/m3 x years)
BY MODEL AND SPECIFIED EFFECT LEVEL
Model
Quantal linear
Quanta! quadratic
Weibull (unrestricted)
Weibul (a S 1)
Log-logistic (unrestricted)
Log-logistic OS ^ 1)
Specified
0.10
793
1,681
9
793
9
683
Increase in Rate
0.05
386
1,172
0.07
386
0.09
324
of Abnormals
0.01
76
519
< 0.0001
76
< 0.0001
62
A-17
-------
TABLE A-10. BMDL VALUES FOR EYE-HAND COORDINATION DATA
FROM ROELS (1993) USING INDIVIDUAL LIRD Qig/m3 x years)
BY MODEL AND SPECIFIED EFFECT LEVEL
Model
Quantal linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a S 1)
Log-logistic (unrestricted)
Log-logistic (j8 5: 1)
Specified
0.10
465
1,167
(*)
465
(*)
370
Increase in Rate of Abnormals
0.05
226
814
(*)
226
(*)
175
0.01
44
361
(*)
44
(*)
34
''Computational difficulties resulted from exponentiating values with large absolute magnitudes. Hie actual
results are much less than 1 and correspond to very low values for the same unrestricted cases in Table A-9.
the history of previous exposure, it should be noted that the LIRD exposure values were
constructed using current exposure measurements for various job classifications and
constructing a cumulative exposure based on the time spent in each type of job. The results
of fitting the same six models to the CRD exposure values are in Table A-ll.
In order to compare goodness of fits, the CRD exposure levels were grouped (as shown
in Attachment A-2) in order to calculate chi-squared values and to show the predicted curves
compared with the actual data. The data were grouped into intervals that would minimize
the exposure variation within the group. The parameters were estimated from the original
data, however. All models (except for the quanta! quadratic, which has shown a consistently
poor fit thus far) provide a similar fit, so only the log-logistic and quanta! linear fits are
shown in Figures A-7 and A-8.
Goodness-of-fit statistics were also calculated for comparison using ACRD as the
exposure measure, which (as explained above) is a time-weighted average exposure. The
data were again grouped into intervals that would minimize the exposure variation within the
group, but the parameters were estimated from the original data (as shown in
Attachment A-3). The goodness-of-fit statistics are provided in Table A-12. All models,
with the exception of the quanta! quadratic (as before, using LIRD and CRD), provide a
similar fit. Figures A-9 and A-10 graphically illustrate the fit for the quantal linear and
restricted log-logistic models, respectively, using ACRD as the exposure variable.
A-18
-------
TABLE A-ll. SUMMARY OF THE FIT TO THE EYE-HAND DATA OF
ROELS (1993) USING CRD (jig/m3) EXPOSURE ESTIMATES FOR SIX
DIFFERENT MODELS
Model
Quantal linear
Quantal quadratic
Weibull
(unrestricted)
Weibull
Log-logistic
(unrestricted)
Log-logistic
Parameter
Estimates
ft = 0.000714
7 = 0.0530
ft = 0.000000909
7 = 0.0819
ft = 0.001838
a = 0.8452
7 = 0.0503
ft = 0.000714
a = 1
7 = 0.0530
0 = 0.9952
a = -7.0182
7 = 0.0508
a = -7.0468
7 = 0.0509
Chi-Squared
Goodness of Fit
(p-value)
4.54
(0.6040)
9.35
(0.1548)
4,12
(0.5323)
4.54
(0.4745)
4.32
(0.5043)
4.32
(0.5043)
Log-Likelihood
-66.4129
-69.9783
-66.3305
-66.4129
-66.3349
-66.3349
There are several key results from this analysis. First, all models fit better (as tested
by the statistical significance of the difference in log-likelihoods) using CRD as the measure
of exposure than do the same models using the LERD variable. Models using ACRD provide
a better (but not statistically significantly better) fit than using CRD as the exposure variable.
Second, the Weibul and log-logistic models give almost identical results to each other,
whether restricted or unrestricted. Third, the fit (as measured by the log-likelihood) is very
similar for all models except the quantal quadratic model. This is logical because both the
Weibull and log-logistic models have parameters for the exponent of dose near 1.
Because Roels provided the individual LJRD and CRD exposure values along with the
variables of age (AGE) and length of exposure (TY1MN) (see Attachment A-l), it is
A-19
-------
1.00
0.75
73
g
§ 0.50
I
0.25'
300 000
CRD
900
1,200
Figure A-7. Roels (1993) eye-hand coordination data fitted to a quanta! linear model,
with CRD as the exposure variable.
1.00
076.
OSO
0.2S
300
600
900
1,200
Figure A-8. Roels (1993) eye-hand coordination data fitted to a restricted log-logistic
model, with CRD as the exposure variable.
A-20
-------
TABLE A-12. SUMMARY OF THE FIT TO THE EYE-HAND DATA OF
ROELS (1993) USING ACRD Qtg/ni3) EXPOSURE ESTIMATES FOR SIX
DIFFERENT MODELS
Model
Quanta! linear
Quantal quadratic
Weibull
(unrestricted)
WeibuU
(«s 1)
Log-logistic
(unrestricted)
Log-logistic
<0i 1)
Parameter
Estimates
ft = 0.00954
7 = 0.0473
ft = 0.00000299
y = 0.0564
18 = 0.000014
a = 1.737
y = 0.0533
jS = 0.000014
a = 1.737
•y = 0.0533
ft = 2.1031
a = -13.075
7 = 0.0545
ft = 2.1031
a = -13.075
y = 0.0545
Chi-Squared
Goodness of Fit
Rvalue)
5.92
(0.8219)
6.28
(0.7912)
6.07
(0.7329)
6.07
(0.7329)
6.30
(0.7096)
6.30
(0.7096)
Log-Likelihood
-65.7264
-65.1447
-65.0726
-65.0726
-65.1030
-65.1030
possible to fit a model that includes all these variables. For example, a multiple logistic
model with a nonzero background could be fitted:
P(response) - y + (1 - 7)* [a + fttlog(CKD) + jS2log(TYEMN) + 03AGE], (A-6)
where $(x) is the cumulative logistic function. If the estimated coefficients for ft1 and j82
turn out to be similar, then the data are consistent with an exposure based on concentration
multiplied by time (CRD x TYRMN). The estimated coefficients when URD is used as the
concentration variable are shown in Table A-13. Note that the estimate of exposure
duration, ft2, is actually negative, suggesting that a shorter exposure at the same
A-21
-------
1.00
0.75
0.50
S
Q.
0.25
Figure A-9. Roels (1993) eye-hand coordination data fitted to a quantal linear model,
with ACRD as the exposure variable.
1.00
0 ' 200 400 600
ACRD (|ig/m*)
Figure A-10. Roels (1993) eye-hand coordination data fitted to a restricted log-logistic
model, with ACRD as the exposure variable.
A-22
-------
TABLE A-13. LOGISTIC REGRESSION COEFFICIENTS FOR EYE-HAND
COORDINATION DATA OF ROELS (1993) USING LIRD Otg/m3 x years) AS AN
ESTIMATE OF EXPOSURE
Coefficient
a (Intercept)
0, (Log(LffiD))
02 (LogCTYRMN))
03 (AGE)
7 (Background)
Parameter
Estimate
-13.7652
2.3290
-2.8391
0.0019
0.0544
Standard Error
of Estimate
5.1407
0.8438
1.0994
0.0678
0.0219
concentration is worse than a longer one, which is contrary to expectations. Further,
because neither the time nor the age coefficient are close to significance, they can be dropped
from the model without any loss of predictive value. The same analysis was performed
using CRD as the concentration variable in the multiple logistic model (Equation A-6). The
results using CRD are shown in Table A-14. The results for the analysis using ACRD as the
concentration variable are presented in Table A-15. Although the coefficients are different,
the conclusions are the same, namely that neither age nor exposure duration add significantly
to the model. However, it should be recalled that the duration of exposure (geometric mean
of 4 years) was relatively brief and the age range was limited (a range of less than 30 years).
TABLE A-14. LOGISTIC REGRESSION COEFFICIENTS FOR EYE-HAND
COORDINATION DATA OF ROELS (1993) USING CRD (jtg/m3) AS AN
ESTIMATE OF EXPOSURE
Coefficient
a (Intercept)
0! (Log(CRD))
02 (LogCTYRMN))
03 (AGE)
•y (Background)
Parameter
Estimate
-7.3618
1.1152
-0.4604
0.0067
0.0526
Standard Error
of Estimate
3.2171
0.4770
0.4216
0.0588
0.0223
A-23
-------
TABLE A-15. LOGISTIC REGRESSION COEFFICIENTS FOR EYE-HAND
COORDINATION DATA OF ROELS (1993) USING ACRD (jig/m5) AS AN
ESTIMATE OF EXPOSURE
Coefficient
a (Intercept)
£, (Log (ACRD))
fa (Log (TYRMN))
/33 (AGE)
j (Background)
Parameter
Estimate
-13.9579
2.3569
-0.5215
0.0033
0.0546
Standard Error
of Estimate
5.1804
0.8477
0.4622
0.0682
0.0219
Because these results using the CRD or ACRD exposure variables are quite different
from the results using the LERD variable, the BMDs for eye-hand coordination were also
recalculated. The resulting BMD values using CRD are shown in Table A-16 and the
BMDL values are in Table A-17. Now the results of these models are more consistent
(except for the quantal quadratic model), although the two unrestricted models do give lower
estimates. This kind of consistency is to be expected when similarly fitting models are used
to estimate a concentration. The BMD and BMDL estimates calculated using ACRD as the
exposure variable are provided in Tables A-18 and A-19, respectively. The consistency
between models is improved slightly when ACRD is used.
TABLE A-16. BMD VALUES FOR EYE-HAND COORDINATION FOR
CRD (jeg/m3) BY MODEL AND SPECDTED EFFECT LEVEL
Model
Quantal linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a > 1)
Log-logistic (unrestricted)
Log-logistic 03 & 1)
Specified Increase in Rate of Abnormals
0.10 0.05 0.01
147 72 14
340 237 105
120 51 7
147 72 14
127 60 11
128 60 12
A-24
-------
TABLE A-17. BMDL VALUES FOR EYE-HAND COORDINATION
FOR CRD (ftg/m3) BY MODEL AND SPECIFIED EFFECT LEVEL
Model
Quantal linear
Quanta! quadratic
Weibull (unrestricted)
Weibull (a Si 1)
Log-logistic (unrestricted)
Log-logistic 0? S 1)
Specified
0.10
95
256
13
95
15
76
Increase in Rate
0.05
46
178
1.0
46
1.4
36
of Abnormals
0.01
9
79
0.003
9
0.007
7
TABLE A-18. BMD VALUES FOR EYE-HAND COORDINATION FOR
ACRD (Mg/m3) BY MODEL AND SPECDTED EFFECT LEVEL
Specified Increase in Rate of Abnormals
Model
Quanta! linear
Quanta! quadratic
Weibull (unrestricted)
Weibull (« 3s 1)
Log-logistic (unrestricted)
Log-logistic (0 S 1)
0.10
110
188
170
110
176
104
0.05
54
131
112
54
123
49
0.01
11
58
44
11
56
9
TABLE A-19. BMDL VALUES FOR EYE-HAND COORDINATION
FOR ACRD Qtg/m3) BY MODEL AND SPECD7PSD EFFECT LEVEL
Specified Increase in Rate of Abnormals
Model 0.10 Q.Q5 0.01
Quanta! linear 73 36 7~
Quintal quadratic 150 105 46
Weibull (unrestricted) 69 26 3
Weibull (a * 1) 73 36 7
Log-logistic (unrestricted) 74 30 6
Log-logistic 08 S> 1) 64 30 6
A-25
-------
Ethyl Corporation (1994, Table 8) also gives results based on continuous response data.
Their approach assumes that the outcome measures are normally distributed, with the mean
defined in such a way that the Weibull model (Equation A-l) will apply. This is unusual in
that it forces continuous data to follow a diehotomous model. A more logical assumption
would be that the mean of the normal distribution is described by a standard dose-response
model. Apparently Ethyl Corporation (1994) actually used the quanta! linear model because
there were only two exposure levels for the continuous data. The Roels et al. (1992) paper
only gives the means and standard deviations for the unexposed and pooled exposed groups.
For completeness, BMD and BMDL estimates were calculated for the endpoints of
visual reaction time and hand steadiness. These values were calculated using the individual
LERD values for specified increases of 10, 5, and 1% in the rate of abnormals. The results
are in Tables A-20 through A-23. Note that the BMD results for both endpoints are very
consistent for the different models if the quanta! quadratic model is ignored.
TABLE A-20. BMD VALUES FOR VISUAL REACTION TIME TOR INDIVIDUAL
LIRD Otg/m3 x years) BY MODEL AND SPECIFIED EFFECT LEVEL
Specified Increase in Rate of Abnormals
Model
Quanta! linear
Quanta! quadratic
Weibull (unrestricted)
Weibull (a > 1)
Log-logistic (unrestricted)
Log-logistic (ft 5: 1)
0.10
1,841
2,143
1,869
1,869
1,860
1,860
0.05
896
1,495
1,009
1,009
962
962
0.01
176
662
250
250
224
224
An attempt was made to verify the calculations in Table 6 of Ethyl Corporation (1994).
The comparison is shown in Table A-24 with Ethyl Corporation's values given in
parentheses. The results are somewhat close for the individual data, but quite different for
the grouped data. It appears that differences in the grouping of data may be responsible for
the differing results.
A-26
-------
TABLE A-21. BMDL VALUES FOR VISUAL REACTION TIME FOR INDIVIDUAL
LIRD Qtg/m3 x years) BY MODEL AND SPECIFIED EFFECT LEVEL
Specified Increase in Rate of Abnormals
Model • 0.10 0.05 0.01
Quantal linear 958 466 91
Quantal quadratic 1,464 1,021 452
WeibuU (unrestricted) 706 (*) (*)
Weibull (a a 1) 958 466 91
Log-logistic (unrestricted) 682 (*) (*)
Log-logistic (ft fe 1) 915 434 83
Computational difficulties resulted from exponentiating values with large absolute magnitudes. The actual
results are much less than 1.
TABLE A-22. BMD VALUES FOR HAND STEADINESS FOR INDIVIDUAL
LIRD Otg/ni3 x years) BY MODEL AND SEECDHED EFFECT
Specified Increase in Rate of Abnormals
Model
Quantal linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a S> 1)
Log-logistic (unrestricted)
Log-logistic O S* 1)
0.10
1,172
1,754
1,139
1,172
1,121
1,121
0.05
571
1,224
529
571
531
531
0,01
112
542
93
112
102
102
TABLE A-23. BMDL VALUES FOR HAND STEADINESS FOR INDIVIDUAL
LIRD Otg/m3 X years) BY MODEL AND SPECIFIED EFFECT
Model
Quanta! linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a S» 1)
Log-logistic (unrestricted)
Log-logistic 0? *> 1)
Specified Increase in
0.10 0.05
662 322
1,243 867
256 8
662 322
259 9
595 282
Rate of Abnormals
0.01
63
384
(*)
63
(*)
54
*Computational difficulties resulted from exponentiating values with large absolute magnitudes. The actual
results are much less than 1.
A-27
-------
TABLE A-24. BMDL VALUES FOR ORD 0*g/m3 x years) BASED ON INDIVIDUAL
DICHOTOMOUS RESPONSE DATA FROM ROELS ET AL. (1992)*
Eye-Hand Coordination
Additional
Risk
0.10
0.05
Grouped
Datab
308
(457)
235
(223)
Individual
Data*
465
(492)
226
(239)
Hand Steadiness
Grouped
Data
507
(688)
247
(335)
Individual
Data
662
(682)
322
(332)
Visual Reaction Time
Grouped
Data
738
(965)
359
(420)
Individual
Data
958
(966)
466
(470)
"Values in parentheses reported by Ethyl Corporation (1994, Table 6) for BMD analysis using quanta! linear
model.
'Groups*
(1993).
model.
Grouped exposure data from Figure 3 of Roels et al. (1992); individual exposure data provided by Roels
As an alternative to forcing the continuous responses into a dichotomous framework, it
is logical to model the continuous outcomes directly. This is done in the following section
on the Bayesian approach.
5. BAYESIAN APPROACH
Because classical estimation methods rely on approximations to derive BMDLs, it is
logical to compare the results with Bayesian methods. Once the specified health effect has
been established, the concentration Otg/m3) required to obtain this effect is estimated as a
posterior distribution (i.e., the specified health effect of a particular study can be denoted as
9), In general, 9 will be a function of the concentration, x:
9 = f(x). (A-7)
The general Bayesian approach has been published under the title of the Confidence Profile
Method (Eddy et al., 1992). It produces posterior distributions for the parameters of
interest. The posterior distribution is a continuous plot describing the belief about the
location of the parameter of interest (e.g., concentration associated with a specified health
effect). The basic formula of Bayesian statistics is
p'(0) = U(0 | data) p(9), (A-8)
A-28
-------
where 9 is the parameter of interest, p*(0) is the posterior distribution for 6, L(0 | data) is the
likelihood for 6 given the new data, and p(fi) is the prior distribution for 8. Because p'(0)
will become the prior for the next experiment, it is often denoted by the same letter.
For example, assume that the health effects are expressed as a continuous measure, and
a linear regression model is assumed for the health effect:
(A-9)
where 6 is the adverse health effect, 0 is a linear regression coefficient, *e is the exposure
concentration of the exposed group, and xc is the exposure concentration of the control
group. If 5 is fixed at a value, 60, corresponding to a predetermined specified health effect,
then the concentration, XQ, that will produce the effect, 50, is estimated by an inverse
function (f " ):
*0 = f -» = a0 / ft + xc. (A-io)
Because the estimation of XQ requires the inverse of the estimate of |3, the problem is
often referred to as an "inverse estimation problem". Based on the evidence in the study and
the prior distribution for |3, a posterior distribution for /S can be derived. After specifying
§Q, the posterior distribution for XQ can be derived.
Because XQ is not defined for j8 < 0, it is reasonable to choose the prior for 0 as
pO) = 1 for $ > 0; pO) = 0 otherwise. This is consistent with a belief that exposure to a
toxic agent is not beneficial. This prior distribution is the horizontal dashed line in
Figure A- 11. Assume that an experiment to determine information about 0 was conducted,
resulting in the likelihood, L(/3), shown as a dotted line in Figure A-ll. Note that the
likelihood is positive for values of & less than 0. The posterior distribution, p'(/J), is the
product of the two distributions (properly normalized to be a probability distribution) and is
shown as a solid line in Figure A-ll. Note that this distribution has the same general shape
as the likelihood function, except that it has no mass below zero. This kind of distribution is
often referred to as a truncated distribution. The posterior distribution of XQ can be
calculated from the posterior distribution of /8.
A-29
-------
Figure A-ll. Schematic of computing a posterior distribution [p' (/?)] from a likelihood
function [L (ft \ data)] and a prior distribution [p 03)].
The calculation is similar for a dichotomous measure. Assume that the increase over
background of the adverse health effect is given by the log-logistic model:
8 =
(A-ll)
If 5 is fixed at a value 50 (the predetermined specified health effect typically equal to 0.10),
then the value of x that will produce the effect 60 is estimated by
- log[50/(l -
(A-12)
A-30
-------
The posterior for XQ is then a function of the parameters a and /3.
One advantage of the Bayesian approach is that it allows synthesis of multiple studies.
Qualitative synthesis of data routinely occurs when deriving an RfC. The Bayesian approach
provides for statistical combination of the data with the variability (e.g., due to sample size)
of the estimates incoiporated. The Bayesian approach has been applied to many areas of
medicine and has recently been applied to combining evidence of the effects of nitrogen
dioxide on respiratory disease in children (Hasselblad et al., 1992). Combination of the data
from the Roels et al. (1987) study with those of the Roels et al. (1992) study could be
considered because some of the same health measures were used in each study.
The posterior distributions calculated in this report are for the concentration associated
with a specified adverse effect. It has been previously proposed that the lower 5th percentile
(or lower 90% credible set limit) of such a distribution can be used as a surrogate to a
NOAEL for derivation of an RfC (Jarabek and Hasselblad, 1991).
For the models described earlier, the natural parameters are those used in Equations A-l
through A-4. The concentration, x, that produces the specified health effect is a functional
parameter. A uniform prior over the interval (0,1) is used for y. The natural prior
distribution for (a, j?) would be the uniform (improper) prior defined over the entire plane.
However, any value of ft less than or equal to zero is inconsistent with the assumption that a
particular compound has an adverse health effect. Thus, the logical prior is the uniform
prior over the half-plane where ft is positive. Finally, a posterior distribution for x can be
calculated by reparameterizing the likelihood function and integrating out the remaining
nuisance parameter.
S.I Bayesian Analyses of Dichotomous Data
Using the same models described earlier, posterior distributions for the
Mn concentration that would produce a 10% additional increase in abnormal eye-hand
coordination values were calculated using grouped LIRD values as in Table A-6. The
posterior mode values and 90% credible set limits are in Table A-25. The 90% credible set
limits were chosen so that the lower limits would correspond to the BMDL, which is a lower
5% confidence limit. Graphs of the six posterior distributions are shown in Figure A-12.
These graphs show that the unrestricted log-logistic and Weibull models produce similar
A-31
-------
TABLE A-25. LIRD Mn CONCENTRATION FOR 10% INCREASE IN ABNORMAL
EYE-HAND COORDINATION VALUES BASED ON FIGURE 3 OF
ROELS ET AL. (1992)
Estimated Median Concentratio
Model 0*g/ni3 x years)
Quanta! linear
Quanta! quadratic
Weibull (unrestricted)
Weibull (a £ 1)
Log-logistic (unrestricted)
Log-logistic (ft ^ I)
111
1,539
16
864
331
1,322
90% Credible Set Limits
n
Lower
457
1,083
0.06
508
1
601
Upper
1,850
2,960
760
1,957
2,956
3,898
-logistic - unrestricted
eibull - unrestricted
antal linear
1,250
URD (ng/m3 x years)
2.500
Figure A-12. Posterior distributions generated by each of six different models that
would produce a 10% increase over background in the rate of abnormal
eye-hand coordination responses.
A-32
-------
results with no real lower bound estimate. The quantal linear and restricted Weibull models
are similar, with clear lower limits. The restricted log-logistic has a similar lower limit but a
much larger upper limit. Finally, the quantal quadratic lies to the right (gives higher
estimates) than the other five curves.
These same calculations were made for the other two endpoints in the Roels et al.
(1992) study, and the results are in Tables A-26 and A-27. In general, the estimates of the
Mn concentration to produce a 10% increase in abnormal values for hand steadiness and
visual reaction time are higher than they are for eye-hand coordination. Many of the same
statistical problems remain, although the estimates tend to be higher than they are for eye-
hand coordination.
TABLE A-26. URD Mn CONCENTRATION FOR 10% INCREASE IN ABNORMAL
HAND STEADINESS VALUES FROM FIGURE 3 OF ROELS ET AL. (1992}
Model
Quantal linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a S> 1)
Log-logistic (unrestricted)
Log-logistic 0 S 1)
Estimated Median
Concentration —
0*g/m X years)
1,187
1,599
789
1,265
1,980
2,086
90% Credible Set Limits
Lower
656
1,158
5
723
631
1,125
Upper
3,163
2,691
10,804
3,089
20,924
12,273
The above results were all based on the original data in the paper of Roels et al.
(1992). The program used to calculate the Bayesian posteriors is similar to the benchmark
dose program in that it cannot handle too many concentration levels. In order to use the
Roels (1993) data with actual exposure levels, the Office of Research and Development
(ORD) had to create new exposure groupings. Exposure levels that were similar were
grouped together, producing the data set in Attachment A-4. The results for eye-hand
coordination using individual URD data for the exposure variable are shown in Table A-28.
A-33
-------
TABLE A-27. LffiD Mn CONCENTRATION FOR 10% INCREASE IN VISUAL
REACTION TIME VALUES FROM FIGURE 3 OF ROELS ET AL. (1992)
Model
Quantal linear
Quanta! quadratic
Weibull (unrestricted)
Weibull (a ^ 1)
Log-logistic (unrestricted)
Log-logistic (/3 s 1)
Estimated Median
Concentration —
Oig/m3 x years)
1,397
1,743
3,023
1,817
2,970
2,606
90% Credible Set Limits
Lower
950
1,359
96
1,010
1,402
1,519
Upper
5,101
3,578
27,515
4,574
71,987
13,762
TABLE A-28. LffiD Mn CONCENTRATION FOR 10% INCREASE IN ABNORMAL
EYE-HAND COORDINATION VALUES FROM ROELS (1993)
Model
Quantal linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a S 1)
Log-logistic (unrestricted)
Log-logistic (ft s 1)
Estimated Median
L-onceniraiion
0*g/m X years)
774
1,586
8
611
345
1,254
90% Credible Set Limits
Lower
458
1,128
0.0004
413
0.2
589
Upper
1,747
2,772
579
961
2,696
3,203
Note that the value of 458 /ttg/m3 x years for the lower credible set limit on a 10%
incidence in abnormal response for the quantal-Bnear model (Table A-28) is very close to the
2
BMDL (462 pg/m X years) in Table A-7, which is as it should be. This demonstrates the
comparability of the two methods. The Bayesian estimates are completely dependent on the
model, as they were with the BMD approach. The unrestricted Weibull or the unrestricted
log-logistic models produce very small values, whereas the other models produce much
larger values. The data cannot provide a definitive estimate of a lower limit. The results
were recalculated using a 5 % increase over background in the rate of abnormal eye-hand
A-34
-------
coordination as the specified health effect (Table A-29). Using the quanta! linear model,
a median posterior value of 377 /ig/m3 x years was obtained, with 90% credible set limits
of 223 to 850 /tg/m3 x years. Using the restricted Weibull model, the median was
312 jtg/m3 x years, with 90% credible set limits of 207 to 502 jig/m3 x years. The results
calculated using a 1 % increase over background rate as a specified health effect are shown in
Table A-30.
TABLE A-29. USD Mn CONCENTRATION FOR 5% INCREASE IN ABNORMAL
EYE-HAND COORDINATION VALUES FROM ROELS (1993)
Model
Quanta! linear
Quanta! quadratic
Weibull (unrestricted)
Weibull (a & 1)
Log-logistic (unrestricted)
Log-logistic 08 > 1)
Estimated Median
Concentration
3
(jig/m X years)
377
1,106
(*)
312
11
747
90% Credible
Lower
223
787
(*)
207
0.0001
301
Set Limits
Upper
850
1,931
(*)
502
75
2,304
^Computational difficulties resulted from exponentiating values with large absolute magnitudes. Hie actual
results are much less than 1.
TABLE A-30. LffiD Mn CONCENTRATION FOR 1% INCREASE IN ABNORMAL
EYE-HAND COORDINATION VALUES FROM ROELS (1993)
Model
Quanta! linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a S 1)
Log-logistic (unrestricted)
Log-logistic $ & 1)
Estimated Median
i^onceiuntuon
(ftg/m x years)
74
490
(*)
67
(*)
222
90% Credible Set Limits
Lower
44
348
(*)
43
(*)
64
Upper
167
855
(*)
121
(*)
1,047
'"Computational difficulties resulted from exponentiating values with large absolute magnitudes. The actual
results are much less than 1.
A-35
-------
The analyses were rerun using the CRD exposure data as was done in the benchmark
dose section. As expected, this variable produced sightly more consistent results between
the different models than using LERD. These estimates are shown in Table A-31. The
results were also calculated using a 5 or 1 % increase over background in the rate of
abnormal eye-hand coordination as the specified health effect (Tables A-32 and A-33,
respectively). Using the quanta! linear model with a 5% increase as the specified health
effect, a median posterior value of 71 /*g/m was obtained, with 90% credible set limits of
46 to 128 pg/m3. Using the restricted Weibull model the median was 75 ^g/m3, with 90%
credible set limits of 49 to 123 jtg/m . The estimates for the same analyses using a 10%
increase as the specified health effect level, performed using ACRD, are shown in
Table A-34. Estimates using a specified health effect level of 5 or 1 % rather than 10% are
shown in Tables A-35 and 36, respectively. Using ACRD as the exposure variable produces
slightly more consistent estimates of the median (within a twofold factor for 10% and
threefold for 5%) by the various models than using CRD.
TABLE A-31. CBD Mn CONCENTRATION FOR 10% INCREASE IN
ABNORMAL EYE-HAND COORDINATION VALUES FROM ROELS (1993)
Model
Quanta! linear
Quanta! quadratic
Weibull (unrestricted)
Weibull (a ^ 1)
Log-logistic (unrestricted)
Log-logistic 08 Ss 1)
Estimated Median
Concentration
145
335
29
145
170
213
90% Credible Set Limits
Lower
94
251
0.4
97
135
118
Upper
263
509
175
229
394
407
5.2 Bayesian Analyses of Continuous Data
Bayesian methods can also handle continuous outcome measures. The only continuous
data reported by Roels et al. (1992) were the means and standard deviations of the
101 control workers and the 92 exposed workers. The exposed workers had a geometric
A-36
-------
TABLE A-32. CRD Mn CONCENTRATION FOR 5% INCREASE IN ABNORMAL
EYE-HAND COORDINATION VALUES FROM ROELS (1993)
Set
.oauiiiaicu ivicuidii
Model Concentration (Mg/m3)
Quantal linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a S: 1)
Log-logistic
(unrestricted)
Log-logistic OS 2s 1)
71
234
4
75
78
119
Lower
46
175
0.02
49
4
58
Upper
128
355
54
123
245
256
TABLE A-33. CRD Mn CONCENTRATION FOR 1% INCREASE IN ABNORMAL
EYE-HAND COORDINATION VALUES FROM ROELS (1993)
Model
Quantal linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a S 1)
Log-logistic
(unrestricted)
Log-logistic O S 1)
Concentration Otg/m3)
14
105
0.08
18
16
39
90% Credible
Lower
9
79
<0.01
11
0.07
12
Set Limits
Upper
25
162
6
36
94
131
TABLE A-34. ACRD Mn CONCENTRATION FOR A 10% INCREASE IN
ABNORMAL EYE-HAND COORDINATION VALUES FROM ROELS (1993)
Estimated Median 90% Credible Set Limits
Model
Quantal linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a S: 1)
Log-logistic (unrestricted)
Log-logistic (ft > 1)
uoncenrranon
Gtg/rn3)
109
186
47
129
212
215
Lower
73
149
4
85
113
134
Upper
186
249
167
206
322
311
A-37
-------
TABLE A-35. ACRD Mn CONCENTRATION FOR A 5% INCREASE IN
ABNORMAL EYE-HAND COORDINATION VALUES FROM ROELS (1993)
Model
Quantal linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a £ 1)
Log-logistic (unrestricted)
Log-logistic (0 S 1)
Estimated Median
Concentration
0*g/m3)
53
130
17
71
151
155
90% Credible Set Limits
Lower
35
104
1.2
44
53
74
Upper
91
173
82
121
258
259
TABLE A-36. ACRD Mm CONCENTRATION FOR A 1% INCREASE IN
ABNORMAL EYE-HAND COORDINATION VALUES FROM ROELS (1993)
Estimated Median 90% Credible Set Limits
Model
Quantal linear
Quantal quadratic
Weibull (unrestricted)
Weibull (a S» 1)
Log-logistic (unrestricted)
Log-logistic OS 3: 1)
lAJiiuciiuauuii
10
57
1
18
75
78
Lower
7
46
0.04
10
11
21
Upper
18
77
17
40
169
171
mean URD exposure of 793 uglm x years. There are three different continuous
approaches that were used to analyze these data.
Each approach assumes that the mean score (e.g., for the exposed versus control
cohort) is linearly related to Mn exposure. This assumption is considered appropriate for
effects at the low range of concentrations. The posterior for the slope then can be calculated
assuming normality of the measurements. The difference in the approaches is the cutoffs that
are used to define an abnormal outcome.
The first approach is to calculate the concentration of Mn that will raise the fraction of
abnormals from 0.050 to 0.145 (a 10% increase). This approach uses essentially the same
A-38
-------
definition of the specified health effect that was chosen for the dichotomized data and results
in the posterior distribution shown in Figure A-13. It has a median of 451 jtg/m3 x years
(1JRD) and 90% credible set limits of 360 and 601 j*g/m3. These values are well within the
range of those seen in Table A-25. If CRD were used as the exposure measure, the median
of the posterior would be 122 jig/m3 with 90% credible set limits of 98 and 163 /tg/m3. The
median of the posterior distribution using ACID as the exposure measure would be
112 pg/m with 90% credible set limits of 90 and 150 fig/m . This approach can also be
used to calculate the Mn concentration that will result in a 5% increase in the fraction of
abnormals. The posterior distribution for this effect size is shown in Figure A-14. Using
LIRD as the exposure measure, the median of this posterior distribution would be 268 /ig/m
X years with 90% credible set limits of 214 and 357 jig/m3 X years. Using CRD as the
exposure measure and the 5 % increase as the specified health effect, the median would be
73 jig/m3 with 90% credible set limits of 58 and 98 jtg/m3. If ACRD is used, the median
would be 67 /*g/m3 with 90% credible set limits of 53 and 89 /ig/m3. Thus, the continuous
endpoint results for eye-hand coordination using this measure are comparable to the
dichotomous results.
As a second approach, it is possible to ask what concentration of Mn (LIRD) will
produce the difference seen between the means for the control versus exposed cohorts in the
Reels et al. (1992) data. Starting with a uniform prior distribution for the difference and the
variance, the posterior distribution is calculated for the LIRD that produces this difference.
This posterior is shown in Figure A-15. The posterior distribution has a median of
793 ptg/m3 x years (the average concentration of the exposed group) and 90% credible set
limits of 635 and 1,057 ng/m3 x years (LIRD). The same model was used with CRD
0*g/m3) as the index of exposure. This results in an identically shaped posterior distribution
with a median CRD Mn concentration of 215 jtg/ra3 and 90% credible set limits of 172 and
287 jig/m3. The model using ACRD again produces an identically shaped posterior
distribution with a median ACRD Mn concentration of 198 jtg/m3 and 90% credible set
limits of 158 and 264 jtg/m3.
The third approach estimates the Mn concentration that would be just statistically
significant, assuming again that the mean response was linearly related to Mn exposure. The
A-39
-------
URDMn(|tfl/ta»x years) 0
CRD Mn ((ig/m3)
ACRD Mn (i»tfm8)
122
112
Figure A-13. Concentration to increase abnormal eye-hand coordination responses by
10%. Median is indicated for each distribution.
posterior corresponding to this specified health effect was calculated using LTJRD, CRD, and
ACRD as exposure measures and is shown in Figure A- 16. For LIRD, the median
concentration is 236 jtg/m3 x years with 90% credible set limits of 189 and 315 /»g/m3 x
years. For CRD, the median concentration is 64 /ttg/m with 90% credible set limits of
51 and 85 pg/m . For ACRD, the median concentration is 59
limits of 47 and 78 jig/m3.
with 90% credible set
6. CONSIDERATIONS FOR DERIVATION OF ALTERNATIVE RfC
ESTIMATES
As shown above, various statistical approaches may be used to arrive at a surrogate for
the NOAEL for the derivation of an RfC. Otherwise, however, essentially the same
methodology is used for deriving an RfC as usual (i.e., dosimetric adjustment of the NOAEL
or its surrogate to a human equivalent concentration [HEC] and application of uncertainty
factors).
A-40
-------
URD Mn (|ig/m* x years) 0
CRD Mn (tig/m3)
ACRD Mn (ng/m')
268
73
67
Figure A-14. Concentration to increase abnormal eye-hand coordination responses by
5%. Median is indicated for each distribution.
URDMn(|iflAn»xyw«) 0
CRD Mn (|itfm»)
ACRD Mn (ng/m»)
215
198
Figure A-15. Concentration to produce observed difference in means of control versus
exposed cohorts for eye-hand coordination. Median is indicated for each
distribution.
A-41
-------
UHD Mn (ntfm1 x years) 0
CRDMn(ng/m»)
ACRD Mn (tig/m«)
236
64
59
Figure A-16. Concentration to produce just statistically significant difference in means
of control versus exposed cohorts for eye-hand coordination responses.
Median is indicated for each distribution.
As in previous derivations of the original and the current verified Mn RfC, in using a
BMD or Bayesian approach to calculate a surrogate NOAEL from the Roels et al. (1992)
study, one of three measurements (eye-hand coordination, hand steadiness, or visual reaction
time) must be selected as the critical effect. Although Roels et al. (1992) give results for the
health endpoints of eye-hand coordination, visual reaction time, and hand steadiness, the
endpoint of eye-hand coordination consistently showed effects at lower Mn levels than did the
other two endpoints for any of the three exposure measures (LJRD, CRD, or ACRD) when
using either the BMD or Bayesian approaches. With the exception of some applications of
the NOSTASOT procedure (see Tables A-2 through A-4), the various approaches discussed
above show that the eye-hand coordination measurement is the most sensitive.
The data from Roels et al. (1992) and Roels (1993) under consideration here require
specific decisions in order to derive an RfC. These include (1) the choice of model for the
analysis; (2) whether to use the data dichotomized into abnormal/normal responses; and
(3) which exposure measure, LIRD, CRD, or ACRD, should be used as the exposure
estimate for the "dose-response" (exposure-response) calculation. The choice of an
A-42
-------
appropriate exposure estimate has always been a critical choice, when using the
LOAEL/NOAEL approach previously as well, but some statistical approaches bring
additional considerations to bear on the choice, such as the goodness of fit of different
mathematical models.
6.1 Choice of Model
The use of the BMD and Bayesian approaches requires the choice of a model to fit the
data. Six different plausible models were explored as part of the analyses presented above.
Chi-square and log-likelihood estimates were calculated as goodness-of-fit statistics and are
summarized in Tables A-5, A-8, A-ll, and A-12. Tables A-5 and A-8 provide the
goodness-of-fit statistics for grouped and individual LERD data, respectively, and
Tables A-ll and A-12 provide the same statistics for CRD and ACID data, respectively.
The difference between the chi-square and log-likelihood statistics is that the chi-square is
calculated when the data are grouped into categories. In both cases for LJRD, the
goodness-of-fit statistics indicate that the unrestricted Weibull and unrestricted log-logistic
models have a significantly better fit than the restricted models. As discussed, there is no
reason a priori to discount the use of unrestricted models. Indeed, the reason that both the
unrestricted Weibull and log-logistic models fit the LJRD data better is that there are two
individuals with quite short durations of exposure and moderate CRD exposure
concentrations that have abnormal responses. Recall that the LERD exposure measure is a
cumulative sum exposure. In these two cases, the short exposure durations were each only at
one concentration (i.e., at one job category). That these two individuals were abnormal
responders after such a short duration of exposure at moderate levels suggests the possibility
of a sensitive subpopulation. Because the individual continuous response data are not
available, there is no way to determine the severity of the two abnormals responses.
However, there is no reason to consider these two responses less or more abnormal than the
others. These two abnormal responses are the reason that the unrestricted models provide a
better fit versus the restricted models, which underestimate the incidence of abnormal
responses at these low ORD exposures.
Given these considerations, a number of alternatives were considered for choosing a
model for the Roels (1993) data. The first was to acknowledge that these unrestricted
A-43
-------
models can give extremely low (virtually zero) estimates but nevertheless describe the
observed supralinear response. Another alternative considered was to construct an entirely
new nonparametric model structure with additional terms that might describe the time for
onset of abnormal responses. Because these types of models have not been discussed for
benchmark application and because they would require additional parameter estimation from
a limited data set, particularly in view of the alternative described below, ORD decided not
to pursue this alternative. The third alternative was to use the restricted models in order to
ensure stable and consistent estimates and take a discerning look at the differences in
exposure measures (see the next section).
For the restricted models, the quanta! linear and restricted Weibull have the same fit
and are the next best fit to that of the unrestricted models mentioned. One guide to selecting
one of several competing models with similar fits was articulated by William of Occam
(ca. 1290 to 1349), who in a discussion of scientific logic, reportedly said, "It is vain to do
with more what can be done with fewer" (Russell, 1945). Using the principle of "Occam's
Razor", selecting the model with the fewest number of parameters that adequately explains
the data, the quanta! linear model was selected for the analyses. (Differences in the range of
estimates due to different model structures will be discussed in Section 6.6.) A similar
pattern holds for model fit using CRD data, and it should be noted that the CRD data provide
statistically significantly better model fits than do the URD data. Using ACRD improves the
model fit slightly again over that using CRD and is considered to be a more appropriate
measurement of overall exposure (see discussion below).
6.2 Choice of Exposure Estimate
Given that the RfC is intended as a dose-response estimate for a lifetime exposure, the
use of LffiD is appealing because it incorporates an attempt to account for the previous
exposure history of the individuals in the exposed cohort. The URD was constructed using
the same exposure measurements as CRD, but it attempts to account for different exposure
histories by adjusting for time spent in different job categories. For example, if an
individual worked in two other exposure categories 1 and 2 previously for 2 and 3 years,
respectively, then the ORD was constructed as a cumulative sum using 2 years of exposure
at the level of category 1 plus 3 years of exposure at the level of category 2, This assumes
A-44
-------
that exposures in each of the categories measured as CRD (i.e., at the time of the study) are
the same as those that occurred in previous years (i.e., that exposure concentrations were
constant for each category over the entire exposure period). Although Roels et al. (1992)
indicated that the production practices had not changed significantly in the past IS years,
there may nevertheless be some variability introduced by the LJRD construct and hence
uncertainty in the quantitative characterization of previous exposure history.
As discussed in the previous section, if the UKD exposure variable is accepted as the
most appropriate, then two choices remain. The first is to accept the data as being a
representative sample and base conclusions accordingly. The two parameter unrestricted
models (Weibull and log-logistic) are driven by a few data points indicating abnormal
responses at very low exposure levels. Any curve fitted to these data will yield a very low
BMDL (near or below 1 /ig/m3 x years). The other choice is to assume that these points
were an aberration and fit a restricted linear model to the data, even though these models
provide a poorer fit. These models produce BMDL estimates near 500 jtg/m3 X years.
Another choice is to accept the results of the analyses that strongly suggest that CRD
provides a statistically significantly better fit of the data. This eliminates the problem of
model selection because all but the quanta! quadratic model fit equally well. The biological
motivation for the use of this exposure metric is not compelling, however. The CRD is a
measure only of current exposures and does not account for previous exposures that may
have occurred at different exposure levels.
The calculation of ACRD, derived from UKD by dividing a worker's URD value by
the total number of years of his employment, is another alternative that is more appealing
from the perspective of biological appropriateness of the dose-response model. The ACRD
can be thought of as a time-weighted average concentration that takes into account exposures
in previous work categories (because it is derived from USD). The use of ACRD as the
exposure measure also provides the best fit when using the restricted models and reduces the
disparity between model estimates.
As discussed above, a multiple logistic model with a nonzero background prevalence
that included the variables age, length of exposure, and concentration was fit to the data.
The results show that both age and length of exposure can be dropped from the model
without any loss of predictive value. These results do not argue that duration of exposure is
A-45
-------
not an important biologic consideration, but rather that these particular data of Roels et al.
(1992), with relatively short exposure durations (arithmetic and geometric mean exposures of
5.3 and 4.0 years, respectively), are insufficient to ascertain adequately whether duration is
an important explanatory variable for the outcome. Because the pharmacokinetics and
potential for accumulation of Mn and/or its damage under chronic inhalation exposure
conditions have not been adequately characterized, an argument can be made for the use of
LIED or ACRD as the better exposure estimates. The ACRD provides a statistically
significantly better model fit than does LORD, even when using the unrestricted structures to
model LORD. Therefore, ACRD is chosen as the best exposure measure candidate, based on
both model fit considerations as well as considerations of biological motivation for the
metric. All three exposure measurements are used to calculate HEC values and resultant
RfC calculations in the interest of completeness,
6.3 Use of Dichotomized Data or Continuous Measurements
The choice of how to designate an effect is different for dichotomous data versus
continuous measurements when using the statistical approaches applied here. For
dichotomous data, the BMDL has been proposed as a surrogate for the NOAEL (Barnes
et al., 1994). The use of the confidence limit ensures that variability due to sample size is
incorporated in the estimate. The incidence for which to calculate the BMD has been
proposed in the range of 10 to 1 %. The only commonly accepted procedural guidance
offered is to make sure that the BMD is close to the range of the observed data. Therefore,
calculations are presented for the BMD at 10, 5, and 1 %.
Roels et al. (1992) selected a representative parameter for each of the three tests
administered (eye-hand coordination, visual reaction time, and hand steadiness) and
dichotomized these continuous measures into normal or abnormal responses according to the
following criteria: (1) eye-hand coordination was considered abnormal when the percent
precision value was lower than the 5th percentile of the control group, (2) band steadiness
was considered abnormal when the score for the 3.5-mm test hole exceeded the 95th
percentile for the control group, and (3) visual reaction time was considered abnormal when
the subject's simple reaction time measurements increased for each 2-min interval within the
8-min testing period. The selection of these cutoff values in order to transform the
A-46
-------
continuous measurements to dichotomous data can be viewed as somewhat arbitrary.
Another concern about transforming these data is that it misrepresents the biological nature
of the normal population, in that most individuals are not simply normal or abnormal, but
exist in a distribution. Even though individuals throughout the distribution may experience
some degree of impairment or reduced function, the impairment of those in the middle or
upper half of the distribution would likely not be detected because a very severe reduction in
function would be necessary to result in their being classified as abnormal. Due to these
concerns, the use of continuous measurements was also explored. Because the individual
raw scores for each test are not available, the modeling of continuous measurements is
limited here to the use of the means and standard deviations of the control versus exposed
cohorts.
Hie choice of how to designate an effect for the cohort mean of a continuous measure
is not as straightforward. Four possible approaches to the continuous data have been
explored. The first approach is that of Ethyl Corporation (1994), whose results are presented
in Table 8 on page 71 of their report. A dose-linear mean model was used, and the BMDL
was calculated assuming a fraction of unexposed persons (controls) who are assumed to show
abnormal responses at a rate of either 1 or 5 %. Note that the 1 and 5 % values refer to
background rates (p0) of abnormal responses in controls for eye-hand coordination and hand
steadiness (5% each) and visual reaction time (1 %); they are not the percentages of increased
risk.
The second approach is a Bayesian analysis, which assumes that the effect on eye-hand
coordination is linear with respect to exposure and assumes a normal distribution of the
measurement. The concentration that raises the fraction of abnormals in the exposed cohort
from 0.05 to 0.145 (a 10% increase) was calculated. This approach to the continuous
measurements closely matches the transformation to dichotomous data described above.
The third and fourth approaches are also Bayesian and designate the effect on eye-hand
coordination according to the differences in the means of the control and Mn-exposed
cohorts. In one case, the observed difference in the means of the control versus exposed
cohorts, as reported by Roels et al. (1992), is taken at face value and used as the designated
adverse effect. In this case, the effect is designated as a percent decrease in the mean
percent precision measurement for the exposed versus control cohort, which rums out to be
A-47
-------
13% (Table 4 of Reels et al., 1992). This mean percent decrease was statistically significant
by t-test with a p value of 0.0005. The fourth approach is to designate an effect as the
concentration at which the difference in the means between the control versus exposed cohort
just achieves statistical significance.
The continuous approaches have the advantage over the dichotomous in that they treat
eye-hand coordination performance data appropriately as continuous measurements. There
is, however, no general consensus on how best to approach continuous data and that is why
four different approaches have been explored. The dichotomous approaches present risk
estimates as population incidence values, which can be easier to interpret in applications to
risk characterization.
Although the general BMD concept has been advocated for use with dichotomous data
(Barnes et al., 1994), implementation has largely centered on using it to address laboratory
animal bioassay data. Epidemiological data pose some additional considerations, as will be
discussed below. In general, because the exposure measures are really estimates of
workplace exposures, regression models such as those used for the BMD generally produce
biased estimates of the slope. If the error in the exposure estimate is unbiased, then the
overall effect is to bias the estimate of slope toward zero and the relative risk for a
predetermined effect level will be underestimated. If sensitive subjects self-selected to work
in lower exposures, then that also would serve to underestimate the risk. The results of
these approaches and considerations of exposure measure and of assigning severity to the
designated effects will be discussed below with respect to their impact on the derivation of
RfC estimates. These approaches and their respective results are described below.
6.4 Dosimetric and Duration Adjustment to Human Equivalent
Concentration
Regardless of the mathematical dose-response model used to calculate the BMDL
estimate, the doshnetric and duration adjustments to the HEC are applied in the same
fashion. These adjustments differ for LIRD versus CRD and ACRD as described below.
A-48
-------
6.4.1 Using LIRD as the Exposure Estimate
If LERD is used as the exposure estimate, then the BMDL is expressed in units of
2
fj.g/m x years and must be converted to an average concentration by dividing by an estimate
of the average length of exposure in years. In order to translate the units to LIRD
cumulative exposure (/ig/m x years), an estimate of the average length of exposure (in
years) was required. A histogram of these data, along with a fitted lognormal curve, is
shown in Figure A-17. Hie geometric mean time and the median time are both 4.0 years
(arithmetic mean was 5.3 years). Because the length of exposure is distributed lognormally
(Figure A-17), the geometric mean (4.0 years) for exposure duration is used for this
calculation. Use of the arithmetic mean (5.3 years) does not substantially alter this
calculation.
Length of Exposure (years)
Figure A-17. Manganese exposure times of individual exposed workers (Roels, 1993).
Adjustment to an HEC must also account for differences in dosimetry due to
occupational versus ambient exposure scenarios. The adjustment reflects differences in
A-49
-------
ventilation rates during exertion (occupational exposure) versus predominantly resting
(ambient exposure) activities and in the work week (5 days) versus the full week (7 days).
The default ventilation rates for occupational (10 m3/day) and ambient (20 m3/day) scenarios
are used in the calculation of HEC values from LXRD exposure data as follows:
[IIRD BMDL 0*g/m3 x year) / 4.0 years] x (10 m3/day)/(20 m3/day) x 5 days/7 days =
LffiD(HEC)0*g/m3).
6.4.2 Using CRD as the Exposure Estimate
If CRD is used as the exposure estimate, the calculation of an HEC is achieved by
adjusting for differences in dosimetry and exposure duration from the occupational to
ambient exposure scenarios as follows:
CRD BMDL 0*g/m3) x (10 m3/day)/(20 m3/day) x 5 days/7 days = CRD(HBC) (^g/m3).
6.4.3 Using ACKD as the Exposure Estimate
If ACRD is used as the exposure estimate, the calculation of an HEC is achieved by
adjusting for differences in dosimetry and exposure duration from the occupational to
ambient exposure scenarios as follows:
ACRD BMDL (jig/m3) x (10 m3/day)/(20 m3/day) x 5 days/7 days =
ACRD(HEC)
6.4.4 Discussion of HEC Calculations
Because the HEC adjustments are the same between the different model estimates for
particular exposure measures (URD, CRD, and ACRD), differences in the HEC values
reflect differences in model fit. The BMDL values used to calculate HEC values using the
BMD approach are taken from Tables A-10, A-17, and A-19 for LffiD, CRD, and ACRD
exposures, respectively. The lower 90% credible set limit estimates from the Bayesian
A-50
-------
approach used to calculate the HEC values are taken from Tables A-28 through A-30, A-31
through A-33, and A-34 through A-36 for USD, CRD, and ACRD exposures, respectively.
The URD(HEC), CRD(HEC), and ACRD(HEC) values are calculated in the same
fashion as above using LERD, CRD, or ACRD NOSTASOT values from Tables A-2 through
A-4, respectively. The lower 90% credible set limit estimates were used to calculate the
HEC values for the continuous Bayesian approaches.
The results of the calculations of HECs are shown in Tables A-37 and A-38 for
dichotomous versus continuous approaches, respectively. As shown in Table A-37, for the
specified health effect of a 10% increase in abnormal eye-hand coordination responses, there
is essentially no difference between the BMD approach and the Bayesian approach for
calculation of an HEC when using the quanta! linear model.
Likewise, there is little difference between the BMD and Bayesian approaches using the
restricted Weibull model with the dichotomous response data. Disparity between the two
approaches does exist for the unrestricted models, in part due to differences in lower limit
computations. Similar results between the BMD and Bayesian approaches are also obtained
when using 5 or 1 % increases in the abnormal eye-hand coordination response.
When restricted models are used with dichotomous data, the resultant HEC values for
either the BMD or Bayesian approaches are essentially the same regardless of whether ORD,
CRD, or ACRD is used as the exposure estimate (maximum disparity is a factor of 1.6).
As discussed previously, use of ORD in the unrestricted models, however, results in
extremely low estimates.
The Bayesian continuous approach that estimates a 10% increase in the mean of the
eye-hand coordination response measurement of the exposed versus control cohorts shows
good agreement with the dichotomous approach that specifies 10% increased abnormal
response as the effect.
6.5 Application of Uncertainty Factors
As discussed above, the use of the mathematical approaches is aimed at arriving at a
surrogate "NOAEL" (i.e., a more statistically robust NOAEL [or LOAEL, depending on the
associated severity]) to use in the derivation of an RfC. As such, the mathematical
approaches do not obviate the requirement to apply factors that extrapolate the observed data
A-51
-------
Ui
JO
TABLE A-37. COMPARISON OF LOWER LIMIT (AND HEC) ESTIMATES FOR 10%
INCREASE IN EYE-HAND COORDINATION DICHOTOMOUS RESPONSE
DERIVED BY BENCHMARK DOSE OR BAYESIAN ANALYSES
LIRD Qiglm X year)8
Model"
QL
QQ
W-U
W-R
LL-U
LL-R
BMDL
462
•1,112
462
<4X10"4
1,366
Group
(HEC)
(41.2)
(99.2)
(NC)
(41.2)
(<4xlO"4)
(121.9)
Individual
Bayes
457
1,083
0.06
508
1
601
(HEC)
(40.7)
(96.6)
9,005)
(45.3)
(0.089)
(53.6)
BMDL
465
1,167
NC6
465
NCe
370
(HEC)
(41-5)
(104.1)
(NC)
(41.5)
(NC)
(33.0)
Bayes
458
1,128
3X10"5
413
0.2
589
(HEC)
(41.0)
(100.7)
(3X10"S)
(36.9)
(0.018)
(52.6)
CRD (pg/m3)
Individual
BMDL
95
256
13
95
15
76
(HEC)
03.9)
(91.4)
(4,81)
03.9)
(5.36)
(27.1)
Bayes
91
236
1
104
127
127
(HEC)
02.5)
(g4.3)
(0,357)
(37.1)
(45.3)
(45.4)
BMDL
73
150
69
73
74
64
ACRDC
W5/«3)
Individual
(HEC)
(26.1)
(53.6)
(24.6)
(26-4)
(22.8)
Bayes
73
149
4
85
113
134
(HEC)
(26.1)
(53.2)
(1.43)
00.3)
(40.3)
(47.8)
°Use of LIRD in deriving RfC requires division of result by 4 years (geometric mean duration of Mn workers' exposure), as described in Section 6.4.1.
Dose-reponse model; QL = quanta! linear, QQ <
and LL-R = log-logistic restricted.
CNC = Not calculable.
quanta! quadratic, W-U = Weibull unrestricted, W-R = Wcibull restricted, LL-U = log-logistic unrestricted,
-------
TABLE A-38. CONCENTRATION THAT YIELDS INDICATED RESULTS8 BY
ANALYSES OF CONTINUOUS DATA
Type of Analysis <
Ethyl Corporation 10% additional
Assuming 1 % unexposed affected
Assuming 5% unexposed affected
Bayesian Approaches
10% Increase
5% Increase
Observed mean
difference
Minimum statistical
significance
URD(HEC)
[/tg/m3 x years)
NCb
NCb
360(32.1)
214(19.1)
635 (56.7)
189 (16.9)
CRD(HEC)
NCb
NCb
98 (35.0)
58 (20.7)
172 (61.4)
51 (18.2)
ACRD (HEC)
Otg/m3)
190 (67.9)
74 (26.4)
90(32.1)
53 (18.9)
158 (56.4)
47 (16.8)
^Change in abnormal responses, based on continuous data.
Not calculable.
to the intended scenario (continuous ambient exposures to the general population, which
includes sensitive subgroups). In the case of the uncertainty factor (UF) applied to account
for severity of the endpoint being extrapolated (previously LOAEL to NOAEL), this must be
judged on a case-by-case basis; the choices presented below pertain specifically to the Roels
(1993) and Roels et al. (1992) data and are not necessarily applicable to future data for Mn
or other chemicals. Each area of extrapolation uncertainty will be discussed below.
6.5.1 Intraspecies UF
Because the mathematical dose-response models only provide a way to manipulate the
observed data and do not inherently account for biological considerations pertaining to
susceptible subpopulations, the rationale for application of this UF remains the same. In the
case of Mn, the elderly and children in particular are considered potential subpopulations of
increased susceptibility (see Appendix C). The Roels et al. (1992) data were derived from a
group of young, healthy, adult male workers. Therefore, susceptible subpopulations have not
been adequately investigated, and a full factor of 10, as applied previously for this
extrapolation, is still warranted.
A-53
-------
6.5.2 Database Limitations
As with the intraspecies UF, the use of mathematical models for dose-response analysis
has not obviated the need for application of a factor for database limitations. A composite
factor was applied in this case, essentially a 3-fold factor for subchronic to chronic
extrapolation and another 3 to account for other database deficiencies (specifically the lack of
developmental/reproductive and chemical-speciation data). The threefold factors are
approximations of 3.33 (i.e., 10 ' ) and therefore, taken together, equal a composite factor
of 10.
The first aspect of uncertainty subsumed by this composite factor is the lack of chronic
exposure in the Roels et al. (1992) study. As discussed above, the duration of exposure
(arithmetic and geometric mean of 5.3 and 4.0 years, respectively) was limited in the Roels
et al. (1992) study. In addition, other data considered, particularly those of Mergler et al.
(1994) and Iregren (1990), suggest that longer exposures may result in lower effect levels.
The other database deficiencies addressed with the other threefold factor include the
lack of adequate characterization of developmental and reproductive toxicity as well as
uncertainties regarding chemical speciation. Although a full UF of 10 may be applied when
both reproductive and developmental data are deficient, a partial UF of 3 was applied in
recognition that (1) although the developing organism may be partially addressed by the UF
of 10 for intraspecies sensitivity, reproductive toxicity is not adequately addressed; and
(2) the potentially different potencies of the chemical species of Mn remain uncharacterized.
6.5.3 Severity UF
The UF for severity replaces the UF to extrapolate from a LOAEL to NOAEL. The
numerical elements of the discussion to follow are specific to this Mn assessment and are not
necessarily transferable to different pollutants with different databases. The UF for severity
is the only UF that may vary in magnitude depending on the nature of the alternative
mathematical approach applied. A different UF may be applied depending on the level of the
dichotomous response used (e.g., 10, 5, or 1 %) or on the magnitude of the difference
between the continuous response measurements. These considerations are discussed in
greater detail below.
A-54
-------
Data for eye-hand coordination used to develop the Mn dose response reflect a
subclinical neurological deficit that would likely not be apparent by a clinical neurological
examination. However, such subtle impairments may still significantly impact a person's life
in a variety of ways, such as impaired performance of jobs and leisure activities involving
fine motor skills. In addition, these measurements could be indicators for other neurological
impairments not tested in this study and could also reflect early stages in the progression to
more severe chronic neurological impairments. Furthermore, small changes induced by Mn
may be additive to other age- or disease-related deficits in motor control, such that together
the changes may be more clinically significant. These considerations argue that it is
appropriate to consider a UP for severity in this case and that the nature of the effects, albeit
subtle and subclinical, warrant at least a threefold factor.
Recent analyses (Faustman et al., 1994; Allen et al., 1994) have compared estimates
from the BMD approach using both quanta! and continuous models applied to oral dose
developmental data against quanta! or continous NOAEL and LOAELs derived using the
NOSTASOT procedure (Tukey et al., 1985). For the quantal approach, the BMDLi0 values
were closer, on average, to the NOAELSs. For the continuous approach, the BMDLj0
values were closer to LOAELs and the BMDL5 values were, on average, closer to NOAELs.
Given the uncertainty in dichotomizing the continuous neurobehavioral response data of Roels
(1993) into quantal data, and the differences between route of administration (oral versus
inhalation) and type of toxicity (developmental versus neurotoxicity), the relevance of these
analyses to the present assessment is limited. Another recent analysis of 39 studies (Ekuta
et al., 1994), which used both models for dichotomous and continuous data and which
included both oral dosing and inhalation exposures as well as general toxicity endpoints,
suggests that the BMDL5 approximates a NOAEL as determined by expert judgement.
Definitive application of such analyses, however, awaits a systematic investigation of data for
various target endpoints in relation to inhalation exposures only.
A factor of 3 has been applied to the BMDL10, as if it were a minimal LOAEL for
these calculations. The BMDL5 estimates were considered as if they were NOAELs, and
thus no UF for severity (i.e., a UF of 1) was applied. The BMDLj also could be considered
a NOAEL, requiring no additional factor for severity.
A-55
-------
Modeling the data as a continuous outcome variable in which the critical effect is the
observed difference in means between the exposed and control cohorts has the advantage that
it treats eye-hand coordination performance appropriately as a continuous variable.
In addition, similar relative impairment in any individual in the population can be considered
as an adverse outcome. If the critical effect is defined as the observed difference between
the means of the control and exposed groups, then the Bayesian analysis has a disadvantage
in that the definition of the critical effect is highly dependent upon the exposure levels
present in the particular situation. In this case, a mean response difference of 13% was
observed. As discussed, an alternative approach is to calculate the exposure at which the
difference in means would just achieve statistical significance. This analysis has both the
advantage of treating the variables as continuous and of being less dependent on the exposure
levels of the particular study. In this case, a mean response difference of 4% just achieves
statistical significance. Given these considerations, a threefold factor for severity may be
applied to the HEC derived from the continuous approach using the observed differences in
means (with an overall UP of 300), and no factor would be applied when calculating the just
significant difference (with an overall UF of 100). The resultant RFC estimates are 0.1 and
•>
0.2 pg/m respectively. These estimates are in agreement with those derived from the
dichotomous approaches.
6.6 Discussion of Resultant RfC Calculations
Table A-39 provides the results of the calculations for the various approaches, at
various effect levels, and with the UFs applied. As discussed previously, the NOSTASOT
procedure gives different results when the analysis is approached either from the bottom up
or the top down. The bottom-up analysis gives the lower estimates of the two, and estimates
for visual reaction time as the critical effect fall within the range of the estimates for
eye-hand coordination obtained by the parameterized models. These estimates are, however,
sensitive to dose spacing and sample size as discussed.
Similar results are obtained by either the BMD or Bayesian approaches using the
dichotomous data, with the exception of the unrestricted Weibull and log-logistic models.
As discussed in Sections 6.1 and 6,2, this disparity among models is accentuated when the
UED exposure estimate is used with these unrestricted models because of the presence of
A-56
-------
TABLE A-39. RfC ESTIMATES DERIVED BY DIFFERENT APPROACHES USING DATA FROM
ROILS ET AL. (1992) AND ROELS (1993)a
1>
Ux
NOAEL (HEC)
Method
Dichotomous Approaches
Current RfC
Conventional Exposure-
Response
One-sided significance
Two-sided significance
NOSTASOT
Bottom Up
VRT
EHC
HST
Top Down
VRT
EHC
HST
Benchmark Approaches
Quanta! Linear
BMDL|0
BMDLs
BMDLj
Quanta! Quadratic
BMDL10
BMDLj
BMDLt
Weibull Unrestricted
BMDLj0
BMDLj
BMDLj
URD
30
15
54
282
237
216
124
20
4
73
32
NC
NC
CRD
NA
16
37
16
114
37
226
16
3
64
28
0.4
0,001
ACRD
NA
7
62
187
167
56
155
13
2
37
16
9
1
LOAEL (HEC) Uncertainty Factors
URD CRD ACRD 10H x
10DB
50 NA NA 100
30 NA NA 100
100
100
100
100
100
100
100
42 34 26 100
100
100
104 91 25 100
100
100
NC 5 25 100
100
100
Severity
10
10
1
1
1
1
1
1
1
3
1
1
3
1
1
3
1
1
Total
Uncertainty
Factor
1,000
1,000
100
100
100
100
100
100
100
300
100
100
300
100
100
300
100
100
LIRD
O.OS
0.03
0.3
0.1
0.5
3
2
2
1
0,1
0,2
0.04
0.3
0.7
0,3
NC
NC
NC
RfC
CRD
NA
NA
NA
0.2
0.4
0.2
1
0.4
2
0.1
0.2
0.03
0.3
0.6
0.3
0.02
0.004
1E-5
ACRD
NA
NA
NA
0.07
0.6
2
2
0.6
2
0.09
0.1
0.02
0.08
0.4
0.2
0.08
0.09
0.01
-------
TABLE A-39 (cont'd). RfC ESTIMATES DERIVED BY DIFFERENT APPROACHES USING DATA FROM
ROELS ET AL. (1
-------
TABLE A-39 (cont'd). RfC ESTIMATES DERIVED BY DIFFERENT APPROACHES USING DATA FROM
ROELS ET AL. (1992) AND ROELS (1993)a
NOAEL (HEC)
Method
Weibull Restricted
Bayesian10
Bayesians
Bayesianj
Log-Logistic Unrestricted
BayesianjQ
Bayesian5
Bayesianj
Log-Logistic Restricted
Bayesian10
Bayesian5
v,. Bayesianj
Ul
Continuous Bayesian
Approaches
Observed Difference (13%)
10% Increase
5% Increase
Minimum Statistical
Significance (4%)
LIRD
18
4
9E-6
NC
27
6
19
17
CRD
17
4
1.4
0.02
21
4.3
21
18
ACRD
16
4
19
4
26
7
19
17
LOAEL (HEC) Uncertainty Factors
URD CRD ACRD 10H X 10DB
37 35 30 100
100
100
0.02 12 40 100
100
100
53 42 48 100
100
100
57 61 56 100
32 35 32 100
100
100
Severity
3
1
1
3
1
1
3
1
I
3
3
1
1
Total
Uncertainty
Factor
300
100
100
300
100
100
300
100
100
300
300
100
100
LIRD
0.1
0.2
0.04
7E-5
9E-8
NC
0.2
0.3
0.06
0.2
0.1
0.2
0.2
RfC
CRD
0.1
0.2
0.04
0.04
0.01
2E-4
0.1
0.2
0.04
0.2
0.1
0.2
0.2
ACRD
0.1
0.2
0.04
0.1
0.2
0.04
0.2
0.3
0.07
0.2
0.1
0.2
0.2
-------
TRfC
NOAEL
HEC
LOAEL
LIRD
CRD
ACRD
10H X 10DB
Current RfC
NA
NOSTASOT
VRT
EHC
HST
BMDL
NC
Bayesian
Bayesian continuous
= Inhalation reference concentration.
= No-observed-effect level.
= Human equivalent concentration.
= Lowest- observed-effect level.
= Lifetime integrated respirable dust.
= Current concentration of respirable dust.
= Average concentration of respirable dust.
= 10-fold factors applied each for human intraspecies extrapolation and database deficiencies.
= Mn RfC on-line IRIS November 1993.
= Not Applicable.
= No statistical significance of trend approach.
= Visual reaction time measurement.
= Eye-hand coordination measurement.
= Hand steadiness test measurement.
= Lower confidence limit of BMD for 10, 5, or 1% incidence using benchmark dose approach.
= Not calculable.
= Lower confidence limit of posterior distribution for 10, 5, or 1 % incidence using Bayesian approach.
= Approaches as described in text.
OS
o
-------
two abnormal individuals at very low USD values. Among the restricted models, the
quanta! linear model provides the best fit to the various exposure measures and uses the least
number of parameters. The ACRD measure provides the best fit when using this model and
in addition represents a time-weighted average exposure that takes into account previous
work history at other exposure levels. Because there exists a potential for cumulative Mn
exposure and/or toxicity, an exposure measure that accounts for the overall work history of
the individuals in the exposed cohort is considered superior to one that addresses only the
current exposure category (such as CRD). Use of ACKD provides a superior model fit using
the quantal linear model to that using IIRD with either the unrestricted or restricted models.
Thus, the use of ACRD with the quantal linear model is based both on biological motivation
for its choice as the exposure measure as well as consideration of fit for the dose-response
model. Also note that with the use of the quantal linear model, there is essentially no
difference between the BMD and Bayesian results and little difference (maximum disparity is
a factor of 1.6) using different exposure estimates (IIRD, CRD, or ACRD). Although
estimates based on 1 % effect levels are shown in Table A-39, analysis of the BMDLs
showed that the BMDs were outside the range of the data and were highly variable across the
different models.
Table A-39 presents over 100 possible Mn RfC estimates based on various exposure
measures, models, effects measures, and uncertainty factors. Not all of these RfC estimates
are equally plausible or worthy of consideration in assessing the potential health risks
associated with Mn inhalation exposure due to MMT usage. As discussed above, some
combinations of the three exposure measures and six mathematical models fit one another
better than other combinations. Based primarily on considerations of cumulative dose
toxicity, statistical goodness-of-fit, and parsimony, ACRD and the quantal linear model
appear to achieve the best results in this respect. Given the similarities of the benchmark and
Bayesian analytic results using ACRD and the quantal linear model, little distinction can be
made between the two analytic approaches in the present application. As for the results
obtained for different effect levels, using a severity uncertainty factor of 3 with a 10% effect
level (for either benchmark or Bayesian analyses) is essentially equivalent to using a severity
UF of 1 with a 5% effect level. Note that the terms LOAEL and NOAEL do not actually
correspond to the results for 10% and 5% effect levels, and therefore neither is preferable to
A-61
-------
the other in the sense that a NOAH- is preferable to a LOAEL in deriving an RfC.
Therefore, benchmark and Bayesian results for 10% and 5% effect levels (using ACRD with
the quanta! linear model) are regarded as equally worthy of consideration here. These
^j
particular analyses yield Mn RfC estimates of 0,09 to 0.1 ^g/m ,
In general, continuous response data are preferred to dichotomized data, primarily
because they provide more information and avoid the basically arbitrary division of effect
measurements into categories (e.g., normal versus abnormal). The Bayesian analysis based
on mean differences between exposed and control groups offers some of the advantages of
using continuous data, in that the reported means and standard deviations (from Roels et al.,
1992) provide a basis for estimating the distribution of continuous response measures.
However, this use of continuous data is not immune to certain common problems, such as
(be issue of statistical power associated with studies of limited size, for the approaches
calculating observed or just-statistically significant differences. Also, whereas the
dichotomous data analyses yield more precision in estimating the effective concentration
associated with a somewhat imprecise response variable, the continuous data analyses offer
the opposite trade off (i.e., more precision in the response variable but less in the exposure
estimate). Nevertheless, the continuous data analyses appear to merit consideration as well
as the analyses based on dichotomous data. By the Bayesian analyses of continuous data, Mn
RfC estimates of approximately 0.1 to 0.2 /teg/m3 are obtained.
All of these selected approaches reflect certain basic assumptions and decisions about
how to view the database in question. For example, the selection of ACRD over LIRD and
the quanta! linear model over the unrestricted Weibull or log-logistic models reflects a
judgment about the data for two workers in the Roels dataset. If the two workers in
question, who had abnormal neurobehavior test scores but short work histories (and
consequently low LERD levels) are considered aberrations of little significance, then they can
be disregarded for the purpose of achieving the best model fit. If, however, the two workers
represent relatively rare but possibly quite real "sensitives" in a worker population, the focus
on ACRD and the quantal model could be at the expense of identifying a more biologically
appropriate model fit. Clearly, with only two cases, the evidence is not sufficient to support
derivation of an RfC estimate. However, such evidence should not be dismissed out of hand
in judging the most appropriate estimate of an RfC for Mn.
A-62
-------
Taking these and other considerations into account, the leading candidate alternative
RfC estimates appear to fall in a range of approximately 0.09 to 0.2 jtg/m3, based on
currently available data. These values are not verified RfCs, and unless or until an
alternative RfC is verified by the EPA RfC/RfD Woikgroup, the current verified Mn RfC of
0.05 /tg/m3 is the only official value.
A-63
-------
REFERENCES
Allen, B. C.; Kavlock, R. J.; Kimmel, C. A.; Faustman, E. M. (1994) Dose-response assessments for
developmental toxicity: n. Comparison of generic benchmark dose estimates with NOAELS. Fundam.
Appl. Toxicol.: in press.
Barnes, D. G.; Daston, G. P.; Evans, J. S.; Jarabek, A. M.; Kavlock, R. J.; Kimmel, C. A.; Park, C.; Spitzer,
H. L. (1994) Benchmark dose workshop: criteria for use of a benchmark dose to estimate a reference
dose. Regul. Toxicol. Phannacol.: submitted.
Cox, D. R.; Hinkley, D. V. (1992) Theoretical statistics. London, United Kingdom: Chapman and Hall;
pp. 207-249.
Crump, K. S. (19S4) A new method for determining allowable daily intakes. Fundam. Appl. Toxicol.
4: 854-871.
Eddy, D. M.; Hasselblad, V.; Shachter, R. (1992) Meta-analysis by the confidence profile method: the statistical
synthesis of evidence. New York, NY: Academic Press, Inc.
Ekuta, J. E.; Martens, M. A.; Stevens, M. W.; Nair, R. A. (1994) Comparison of benchmark doses (BMD)
with no-observed-adverse-effect levels (NOAELs) and low-observed-adverse-effect levels (LOAELs) for
selected subchronic toxicity studies conducted by Monsanto Company [abstract], lexicologist 14: 401.
Ethyl Corporation. (1994) Comments of Ethyl Corporation in response to EPA's December 9, 1993 Federal
Register notice 58 Fed. Reg. 64761 (1993). Washington, DC: Ethyl Corporation, Office of the Vice
President for Government Relations. Available for inspection at: U.S. Environmental Protection Agency,
Central Docket Section, Washington, DC: docket no, A-93-26.
Faustman, E, M.; Allen, B.C.; Kavlock, R.J.; Kimmel, C.A. (1994) Dose-response assessment for
developmental toxicity: L characterization of data base and determination of NOAELs. Fundam. Appl.
Toxicol.: in press.
Hasselblad, V.; Eddy, D. M.; Kotchmar, D. J. (1992) Synthesis of environmental evidence: nitrogen dioxide
epidemiology studies. J. Air Waste Manage. Assoc. 42: 662-671.
Iregren, A. (1990) Psychological test performance in foundry workers exposed to low levels of manganese.
Neurotoxicol. Teratol. 12: 673-675.
IRIS, Integrated Risk Information System [data base]. (1993) [Printout of reference concentration (RfC) for
chronic manganese exposure as revised November, 1993]. Cincinnati, OH: U.S. Environmental
Protection Agency, Office of Health and Environmental Assessment, Environmental Criteria and
Assessment Office. Available online from: TQXNET, National Library of Medicine, Rockville, MD.
Jarabek, A. M.; Hasselblad, ¥. (1991) Inhalation reference concentration methodology: impact of dosimetric
adjustments and future directions using the confidence profile method. Presented at: 84th annual meeting
and exhibition; June; Vancouver, British Columbia, Canada. Pittsburgh, PA: Air and Waste Management
Association; paper no. 91-173.3.
Knudson, A. G., Jr. Hethcote, H. W.; Brown, B. W. (1975) Mutation and childhood cancer; a probabilistic
model for the incidence of retinoblastoma. Proc. Natl. Acad. Sci. U.S.A. 72: 5116-5120.
Mergler, D.; Huel, G.; Bowler, R.; Iregren, A.; Belanger, S.; Baldwin, M.; Tardif, R.; Smargiassi, A,;
Martin, L. (1994) Nervous system dysfunction among workers with long-term exposure to manganese.
Environ. Res. 64: 151-180.
A-64
-------
Moolgavkar, S.; Venzon, D. (1979) Two-event models for carcinogenesis: incidence curves for childhood and
adult tumors. Math. Biosci. 47: 55-77.
Roels, H. (1993) [Letter to Dr. M. Davis, U.S. EPA, on definitions of "nsspirable," "total," and "inhalable"
dusts]. Bruxelles, Belgium: Universite Catholique de Louvain, Unite de Toxicologie Industrielle et
Medecine du Travail; October 19.
Roels, H.; Lauwerys, R.; Buchet, J.-P.; Genet, P.; Sarhan, M. J,; Hanotiau, I.; de Fays, M.; Bernard, A.;
Stanescu, D. (1987) Epidemiological survey among workers exposed to manganese: effects on lung,
central nervous system, and some biological indices. Am. J. Ind. Med. 11: 307-327.
Roels, H. A.; Gfayselen, P.; Buchet, J. P.; Ceulemans, E.; Lauwerys, R. R. (1992) Assessment of the
permissible exposure level to manganese in workers exposed to manganese dioxide dust. Br. J. Ind. Med.
49: 25-34.
Russell, B. (1945) A history of Western philosophy. New York, NY: Simon and Schuster; pp. 468-475.
Tukey, J. W.; Ciminera, J. L.; Heyse, J. F. (1985) Testing the statistical certainty of a response to increasing
doses of a drug. Biometrics 14: 295-301.
A-65
-------
ATTACHMENT A-l. INDIVIDUAL EXPOSURE AND DICHOTOMIZED RESPONSE
DATA OF ROELS (1993)a
VRT
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
EHC
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
1
1
1
0
0
0
1
0
0
1
1
0
1
0
0
0
1
1
0
0
1
1
0
1
HST
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
0
1
0
0
0
1
0
0
0
1
0
0
CRD
179
163
179
104
201
163
201
201
468
163
104
201
468
201
21
104
179
163
179
201
179
201
163
468
201
179
104
104
179
468
468
265
244
179
314
530
179
104
243
179
179
201
907
179
499
IJRD
1,665
2,421
1,790
516
1,389
163
362
262
1,123
700
516
40
2,380
302
63
1,053
627
718
1,307
40
859
60
575
2,668
403
1,951
390
605
1,844
1,872
2,480
677
477
1,647
618
941
1,074
569
3,519
2,506
1,844
80
3,715
806
578
TYRMN
9,3
12,1
10.0
3.3
6.9
1.0
1.8
1.3
2.4
4.0
3.3
.2
5.0
1.5
3.0
6.3
3.5
4.1
7.3
0.2
4.8
0.3
3.3
5.7
2,0
10,9
2.6
3.8
10.3
4.0
5.3
3.3
2.3
9.2
2.7
3.3
6.0
3.6
8.0
14.0
10.3
0.4
7.0
4.5
2.0
AGE
34.7
48.2
32.2
32.0
31.1
33.3
23.1
39.0
25.3
33.4
29.1
22.5
34.8
36.1
39.3
26.3
23.5
29.5
29.0
22.2
26.7
22.1
22.6
30.8
23.4
33.4
26.4
29.3
49.0
24.6
30.8
23.6
23.5
49.4
22.9
30.8
31.5
24,2
31.0
49.6
36.3
23.9
32.0
27.6
28.6
A-66
-------
ATTACHMENT A-l (cont'd). INDIVIDUAL EXPOSURE AND DICHOTOMIZED
RESPONSE DATA OF ROELS (1993)*
VRT
0
1
0
0
0
1
0
0
0
1
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
1
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
EHC
0
0
0
0
0
0
1
0
0
1
0
0
1
1
0
0
0
0
1
0
0
0
0
0
0
0
0
1
0
0
0
0
0
1
0
1
0
0
0
0
0
0
0
0
HST
0
0
0
0
0
0
1
0
0
1
0
0
1
0
0
1
i
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
1
0
0
0
CRD
201
65
179
179
179
179
46
65
1,317
1317
179
907
907
498
49
179
104
179
499
179
21
319
179
244
613
201
243
907
179
634
907
21
179
796
499
179
163
243
46
314
65
530
244
1,201
LIRD
2,012
256
1450
752
430
752
520
1,178
1,603
1,729
627
3,668
1,562
750
460
2,685
498
1,396
768
1,307
55
569
1,110
477
2,061
1,469
2,911
3,574
3,162
1,117
4,433
65
1,862
1,082
929
609
1,309
97
262
565
1,390
977
477
2,651
TYRMN
10.0
8.0
8,1
4.2
2.4
4.2
11.3
7.6
2.6
3.3
3.5
6.9
2.4
2.4
10.0
15.0
3.2
7.8
2.5
7.3
2.6
2.4
6.2
2.3
7.8
7.3
6.7
6.7
17.7
3.7
9.4
3.1
10.4
2.6
3.4
3.4
7.4
0.4
5.7
2.4
8.4
3.5
2.3
9.1
AGE
35.6
32.7
27.4
36.3
22.0
31.7
40.0
32.8
23.8
25.6
23.8
35.7
36.6
33.8
37.5
39.5
24.0
35.7
26.9
29.7
27.1
22.7
25.8
23.0
29.3
36.7
29.6
38.3
49.3
23.8
28.9
31.2
41.2
24.9
31.5
26.0
46.4
23.5
39.6
22.9
44.0
49.4
24.0
30.0
A-67
-------
ATTACHMENT A-l (cont'd). INDIVIDUAL EXPOSURE AND DICHOTOMIZED
RESPONSE DATA OF ROELS (1993)a
VRT
0
0
0
Abbreviations:
EHC
0
0
0
0
1 —
VRT
EHC =
msr
CRD =
LUD
TYRMN =
HST CRD
0 201
0 201
0 201
Normal response.
Abnormal response.
Visual reaction time.
Eye-band coordination.
Hand steadiness test.
IIRD
624
604
1,328
TYRMN AGE
3.1 34.0
3.0 39.9
6.6 28.8
Concentration of respirable dust.
Lifetime integrated respirable dust.
Total years of manganese exposure.
A-68
-------
ATTACHMENT A-2. GROUPED CRD EXPOSURE DATA AND DICHOTOMIZED
RESPONSE OF ROELS (1993)a
CRD
0
44
131
179
201
271
475
1,032
Abbreviations:
VRT-pos
1
1
0
1
0
0
2
3
CRD
VRT-pos =
VRT-neg =
EHC-pos =
EHC-neg =
HST-pos =
HST-neg =
VRT-neg EHC-pos EHC-neg HST-pos
100 S 96 4
5152
13 4 9 1
23 2 22 1
14 2 12 0
11 3 8 3
11 5 8 2
5443
Current concentration of respirable dust.
Visual reaction time, number of positive (abnormal) responses.
Visual reaction time, number of negative (normal) responses.
Eye-hand coordination, number of positive (abnormal) responses.
Eye-hand coordination, number of negative (normal) responses.
Hand steadiness test, number of positive (abnormal) responses.
Hand steadiness test, number of negative (normal) responses.
HST-neg
97
4
12
23
14
8
11
5
A-69
-------
ATTACHMENT A-3. GROUPED ACRD DATA AND DICHOTOMIZED RESPONSE
OF ROELS (1993f
ACRD
0.00
23.75
46.00
159.00
179.00
202.00
236.00
289.00
420.00
470.00
530.00
634.00
EHC-pos
5
0
1
2
4
3
0
3
2
2
3
1
EHC-neg
%
4
2
8
24
16
4
6
1
4
1
1
aAbbreviations: ACRD = Average concentration of respirable dust.
EHC-pos = Eye-bind coordination, number of positive (abnormal) responses.
EHC-neg = Eye-band coordination, number of negative (normal) responses.
A-70
-------
ATTACHMENT A-4. GROUPED L1RD EXPOSURE AND DICHOTOMIZED
RESPONSE DATA OF ROELS (1993f
LIRD
0.0
62.5
236.0
427.0
587.0
763.0
1,009.0
1,132.0
1,372.0
1,656.0
1,921.0
2,541.0
3,037.0
3,619.0
4,433.0
Abbreviations:
VRT-pos
1
0
1
0
0
2
0
0
0
1
0
1
0
1
1
URD
VRT-pos =
VRT-neg =
EHC-pos =
EHC-neg =
HST-pos =
HST-neg =
VRT-neg
too
8
3
10
16
6
6
4
9
5
7
6
2
3
0
EHC-pos
5
2
0
0
7
3
1
0
0
3
0
2
0
3
0
Lifetime integrated respirable
Visual reaction time
Visual reaction time
, number
, number
EHC-neg
96
6
4
10
9
5
5
4
9
3
7
5
2
1
1
dust concentration.
of positive (abnormal)
HST-pos
4
0
0
1
2
0
2
0
1
2
0
2
0
2
0
responses.
HST-neg
97
8
4
9
14
8
4
4
8
4
7
5
2
2
1
of negative (normal) responses.
Eye-hand coordination, number of positive (abnormal) responses.
Eye-hand coordination, number of negative (normal)
Hand steadiness test
Hand steadiness test
, number
, number
of positive (abnormal)
response.
responses.
of negative (normal) responses.
A-71
-------
APPENDIX B
EXPOSURE ASSESSMENT FOR MANGANESE
This exposure analysis for manganese (Mn) is divided into four parts: (1) a review of
several small studies of Mn exposure levels in certain nonindustrial occupational groups and
of ambient air Mn levels in three Canadian cities, (2) a description of a major study of Mn
exposures in a representative sample of the population of a U.S. community,
(3) a comparison of the Canadian and U.S. exposure studies, and (4) estimates of the
projected distribution of personal exposure levels of Mn if methylcyclopentadienyl manganese
tricarbonyl (MMT) is used in gasoline in the United States as proposed by Ethyl
Corporation. A glossary of selected terms may be found in Attachment B-6.
1. SURVEY OF CANADIAN STUDIES OF Mn EXPOSURE LEVELS
In this section, several studies of Mn exposure levels in small cohorts of subjects in
Canada, where MMT has been in use for several years, are described and reviewed. These
studies focused on personal exposures to Mn in selected taxi drivers, garage mechanics, and
office workers, as well as ambient concentrations of Mn that might be correlated to exposure
distributions.
1.1 Montreal Area
In June 1992, Zayed et al. (1994) monitored 10 taxi drivers and 10 garage mechanics
(working at the same garage) for Mn exposure by means of personal samplers. The cut point
of the samplers was not stated in the published report by Zayed et al. (1994) but has been
since characterized as total suspended paniculate (TSP) matter by Ethyl Corporation (1994b).
The sampling periods were a 40-h workweek and two nonwork periods of 16 h per
participant to represent their weekday off-work exposures. Because these subjects do not
constitute a probability sample from all Montreal taxi drivers and garage mechanics, these
results represent only the exposures of the cohort sampled and cannot be used to make
statements about the exposures of the larger population of Montreal,
B-l
-------
Exposures of the on-duty taxi drivers ranged between 0.01 and 0.07 jtg/m3 Mn, and
off-duty exposures ranged between 0.01 and 0.03 ftg/m Mn. Mechanics working at the
same garage had on-duty exposures ranging from 0.01 to 2.07 /*g/m3, and their off-duty
exposures ranged from 0.01 to 0.03 /tg/m3. Unfortunately, the authors neither measured the
ambient Mn in the Montreal air for comparison during their study period nor collocated one
or more of their sampling pumps and filters at a nearby Montreal Urban Commission (MUC)
air monitoring station where Mn was being measured. Had they done so, it would have
enabled them to obtain a site-specific relationship between their TSP Mn instrument and the
MUC TSP instrument, a high-volume sampler with a 27-jtm cut point.
The maximum 24-h ambient average TSP (PM27) Mn measured in the MUC network in
June of 1992 was 0.33 jttg/m at Station 68, an urban high-traffic location (Gagnon, 1994a),
whereas the maximum 16-h off-work value for the Zayed et al. (1994) study was
0.027 /tg/m3, an order of magnitude lower. It is not clear why Zayed et al. (1994) cited a
1990 ambient Mn value for comparison, when MUC had June 1992 Mn data available.
In 1990, the Beauharnois ferromanganese plant 20 km from Montreal was in operation and
presumably contributed to the ambient Mn in Montreal when the wind was from the
southwest (cf. Loranger et al., 1994).
The MUC city-wide average ambient PM10 Mn for the month of June 1992, based on
samples taken every sixth day, was 0,027 /*g/m (Gagnon, 1994a), as compared to the
downtown value of 0.058 jfg/m3 in 1990 cited by Zayed et al. (1994). The city-wide
average PM10 Mn value of 0.027 uglm coincidently equals the highest off-work 16-h
average (inferred PM3 5) Mn value measured by Zayed et al. (1994) in June 1992.
The June 1992 monthly mean TSP (PM27) Mn at Station 68 was 0.12 /tg/m3 (Gagnon,
1994a), indicating a potential for high Mn exposures for persons living in areas of high
density traffic, given the fact that automotive Mn is in the fine particle mode and effectively
penetrates into indoor microenvironments with little, if any, interference. In June of 1992,
there was 76 mm of rainfall on 13 days (Veinot, 1994) in Montreal. This factor would need
to be considered in more detail to determine if the weather in the sampling period was
atypical, because rainfall is known to wash out Mn (World Health Organization, 1987). This
study is of limited utility for Mn personal exposure assessment purposes because of its
nonprobability sampling design and the fragmentary nature of the data.
B-2
-------
Zayed and colleagues performed another study of Mn exposures in a nonprobability
sample of taxi drivers and office workers in January 1994 in Montreal, as described in an
unpublished report submitted by Ethyl Corporation (1994a). The report discusses these data
in relation to the previous Montreal data (Zayed et ah, 1994) and to the results of a parallel
study conducted in Toronto by Ethyl Corporation, described in the same report (see
discussion below). Although the Ethyl Corporation (1994a) report describes the data as
PM3 5 and TSP, a later communication (Ethyl Corporation, 1994b) indicates that the data
were for PM5 and TSP.
The tabulated data contain several typographical errors, and total Mn is less than PM5
Mn in several entries, suggesting a problem with quality assurance for these data. In their
1992 study, Zayed et al. (1994) had problems with negative mass (i.e., the weight of
collected paniculate Mn was less than the weight of Mn on blank filters according to Zayed
[1993]); this may still be a problem in the 1994 study, as the reported exposures are almost
an order of magnitude lower than the 1992 exposures of the taxi drivers (0.024 j*g/m3 in
•a
1992 and 0.006 jig/m in 1994). Also, the office workers have a much lower exposure
H ^
(0.0023 jig/m ) compared to the off-work exposures of the taxi drivers (0.007 pg/m ) and
mechanics (0.011 pg/m ) measured in 1992.
As in the 1992 study, Zayed and coworkers (Ethyl Corporation, 1994a) did not
collocate one of their monitors with a MUG monitor to determine the interrelationship with
MUC and to allow later comparison with MUC ambient Mn data. It is noted that the total
ambient PM10 measured by MUC decreased from 0.059 to 0.032 /ig/m3 between June 1992
and January 1994 (Gagnon, 1994b), which may also explain a portion of the decrease in Mn
measured by Zayed between June 1992 and January 1994 (Ethyl Corporation, 1994a). Also,
in January 1994, Dorval International Airport in Montreal recorded an above-normal 70 cm
of snow and 20 mm of rain. In a discussion of the relative effectiveness of rain and snow in
removing particles from the atmosphere, Garland (1978) states, "Diffusion and interception
may be of greater significance in snow because of the larger surface area of precipitation
elements." There was also a ground cover of 14 to 26 cm of snow throughout the month
(Veinot, 1994), which could cause severe reduction of Mn by reducing traffic volume and
reducing resuspension of Mn by traffic from road dust. Because of the fragmentary nature
of these nonrandomly sampled data in time and number, they cannot be interpreted as
B-3
-------
representing anything other than a measure of the exposure of the subjects who carried the
monitors during those exact days.
Dann (1990) reported a frequency distribution of PM10 Mn for two stations in Montreal
covering the period 1984 to 1987. One station (n = 56) was listed as having > 30,000
vehicles in average daily traffic (ADT), and the other (n = 40) > 100,000 ADT. Figure B-l
shows the frequency distribution of Mn at both of these stations, indicating the effect of the
increased traffic volume.
1.0 T
to.
O
1
8
§0.01
O
> 30,000 ADT
Min. 10 30 50 70
Cumulative Frequency (%)
90 Max.
Figure B-l. Frequency distribution of ambient Mn for Montreal for years 1984 to 1987
showing the effect of increased average daily traffic (ADT). Data are for
PM10 Mn from chemical analyses of combined sets of fine and coarse
fractions from dichotomous samplers. Individual size fraction distributions
were not given.
Source: Dann (1990).
The mean ]PM10 and PM2-5 Mn was 0.083 /ig/m and 0.047 ^g/m, respectively, at the
high-ADT station, and 0.053 and 0.033 /tg/m3, respectively, at the low-ADT station. All
B-4
-------
high-ADT samples were collected on the same days as the low-ADT samples, on a l-in-6
day sampling schedule, but 16 additional low-ADT days were collected. Although these data
were presumably influenced by the Beauharnois ferromanganese plant emissions, the
contribution to both distributions at a 20-km distance should be similar with the reduction in
peak-to-mean of the plume at such a distance. Assuming that the background Mn is similar
between these two stations, these data suggest that, given the usage of MMT in Canadian
gasoline from 1984 to 1987, the increase in traffic produced an increase in PM10 Mn on the
order of 0.03 i*,g/m3 (mean of high-ADT samples minus mean of low-ADT samples).
Loranger et al. (1994) reported results of TSP PM27 Mn measurements in Montreal
during the year 1990, when the Beauharnois ferromanganese plant was in full operation
20 km to the southwest of Montreal. However, the authors state (italics added) that "no
major source of contamination by Mn can be found within a 20-km radius of Montreal," and
the very first dendrogram cluster in their analysis "includes the dominant winds (west,
southwest)" that follow the St. Lawrence River Valley from Beauharnois to Montreal. Yet
two of the same authors state in a different paper (Loranger and Zayed, 1994) with regard to
the "substantial decrease in Mn after 1990" in Montreal that it "may be [related to] the
shutdown in 1991 of a ferromanganese plant located 20 km southwest of Montreal (24 April
1991 [Paquet, 1994; Baldwin et al., 1992]).... Hence, any Mn in Montreal air due to MMT
would have been hidden by the more substantial variations associated with this industrial
activity." Perhaps Loranger and Zayed discovered the presence of the Beauharnois plant
after the Loranger et al. (1994) and Zayed et al. (1994) papers went to press, as this factor is
not mentioned in either of them. In the Loranger et al. (1994) paper, they cite the annual
average TSP (PM27) Mn as 0.03 ng/m3 for 1990 at four stations with low traffic
(< 15,000 ADT) and 0.053 /*g/m3 at two stations with high traffic (> 15,000 ADT).
If Beauharnois and crustal materials influenced both stations equally, then the difference of
0.023 jig/m3 would presumably be from the increased traffic.
1.2 Toronto Area
The annual average PM10 Mn concentrations at the Breadalbane Street monitoring
station of the Inhalable Particulate Monitoring (TPM) Network (Dann, 1990), based on
sampling every sixth day, are shown hi Table B-l. Based on post hoc analyses, these data
B-5
-------
TABLE B-l. SUMMARY OF ANNUAL AMBIENT PM10 Mn IN TORONTO fcg/m3)
Year Mean Standard Deviation
__ 0.025 0.012
1987 0.030 0.017
1988 0.023 0.013
1989 0.027 0.018
Average 0.0261 0.015
Source: Dann (1990).
are statistically indistinguishable from a set of random samples from a stationary distribution
with a mean of 0.0261 ^ig/m and a standard deviation of 0.015 /tg/m .
Lynam et al. (1994) measured exposure levels of 6 taxi drivers and 24 office workers
in Toronto from February 4 through 19, 1991, using a personal particle sampler. The
authors did not state the flow rate and cut point to which the sampler pump was calibrated,
nor is information on quality assurance procedures given in the report. However, the size
fraction approximated that of TSP (PM27), based on information provided by Ethyl
Corporation (1994b). Also, the authors do not state how the filters were analyzed for Mn
(e.g., X-ray fluorescence or neutron activation) or whether filter blanks were kept and mass
change corrections were made, as done by Zayed et al. (1994).
Lynam et al. (1994) do not mention how their subjects were recruited, but apparently it
was not through a probability sampling of all taxi drivers and office workers in Toronto, as
the taxi drivers all worked for the same company. Consequently, these results can be used
only to represent the exposures of the subjects themselves during that specific sampling time.
Lynam et al. (1994) did not collocate their monitor at an Environment Canada (EC) or
Ontario Ministry of the Environment (MOE) station, so no direct comparison can be made
between the sampling and analytical techniques. However, during the periods of the study,
Ontario MOE (Radell, 1994) measured Mn with high-volume TSP PM27 samplers at three
stations; Grosvenor, Mosley/Leslie, and Junction Triangle. Mean Mn TSP concentrations
were 0.083 jwg/m3 at Grosvenor (n = 12), 0.040 jttg/m3 at Mosley/Leslie (n = 14), and
0.023 Mg/m3 at Junction Triangle (n = 3).
B-6
-------
The ten 1-week average Mn exposures of six taxi drivers ranged from 0.015 to
0.049 pg/m and 33 measurements of weekly exposures of 24 "office workers" ranged from
0.002 to 0.048 jig/m3 in 1991 (Lynam et al., 1994). The almost identical maxima for an
office worker who commuted by subway and the highest taxi driver suggest one or more of
the following: an unanticipated nontraffic source of Mn is present; a quality assurance
problem exists with the sampling, weighing, and/or chemical analyses; or a chance sampling
error (e.g., the commuter is unusually high or the taxi driver is unusually low). Without
quality assurance information for the study by Lynam et al. (1994) on duplicate and replicate
analyses, Mn increase in filter blanks, and analysis of Mn standards by their unspecified
analytical technique, sampling error cannot be isolated from analytical error or
instrumentation differences as possible reasons for this discrepancy.
Ethyl Corporation (1994a) repeated their 1991 study by Lynam et al. (1994) on
unspecified days in January 1994. Unfortunately, the Ontario MOE and EC Mn data for
Toronto in January 1994 are not available for comparisons at this time. The average of taxi
driver exposures decreased from 0.035 jtg/m3 for 1991 to 0.014 j&g/m3 PM5 for 1994, and
office worker average exposures decreased from 0.013 ^g/m3 for 1991 to 0.010 /ig/rn3 PM5
for 1994. This difference may have contributed to the finding of higher Mn in Toronto than
in Montreal. As discussed regarding the 1994 Mn study in Montreal (Ethyl Corporation,
1994a), the decrease from 1991 to 1994 may be due to quality assurance problems (reflected
in the fact that the respirable fraction is reported higher than total PM for some pairs and
some blanks weighed more than collected samples). Other possible factors may be
differences in the analytical techniques used in 1991 (not stated) and 1994 (neutron activation
at University of Montreal).
Because the 1994 samples were shipped from Toronto to Montreal for analysis, there is
a strong likelihood of loss of Mn from the total filters, as larger particles containing Mn do
not "stick" tightly to the filter and may be lost in shipment (Appel et al., 1980, p. 91) or in
the neutron activation itself. Such losses may help explain why respirable Mn is more than
total Mn in several samples.
It should also be noted that Pearson International Airport in Toronto reported for
January 1994 a total of 30 mm rainfall and 40 cm snowfall, with snow cover every day
except for the first 2 days of January (Veinot, 1994). Precipitation occurred on 19 of
B-7
-------
31 days. In view of the fact that the dates on which the samples were collected by Ethyl
Corporation (1994a) are not specified, it is not possible to determine if the weather may have
influenced these data in an unusual manner.
In summary, the report of the January 1994 study submitted by Ethyl Corporation
(1994a) provides a limited account of a study of a nonrepresentative cohort. In addition, it
seems possible that severe weather with high winds and particle scavenging from rain and
snow storm activity could have created an atypical reduction in all atmospheric pollution,
including Mn.
1.3 Windsor Area
Lusis et al. (1992) reported that in June and July of 1991 the Ontario MOE conducted a
pilot study of personal exposures to organics and of indoor/outdoor metals in Windsor and
j-f
Toronto (Toronto was not sampled for metals). They reported a median of 0.010 /ig/m Mn
in 14 homes and a median of 0,030 /*g/m Mn outdoors, sampled with a high-flow-rate
sampler of 25 L/m that was estimated to have a cut point at 15 [im. These data were not
corrected for pressure and temperature (Bell, 1994). However, the outdoor values are
consistent with other data on trace metal concentrations (e.g., the average PM10 Mn
concentration measured by the EC IPM Network was 0.033 /*g/m3 in Windsor from 1984
through 1987). Individual data for indoor measurements are not given, so the variance is
unknown. Nontraffic sources of Mn in the Windsor/Detroit area are not identified. This
study provides no usable data on personal exposures to Mn.
2, POPULATION DISTRIBUTION OF EXPOSURES TO Mn
The only published probability-based representative study of the exposures of a
community population to Mn is the Particle Total Exposure Assessment Methodology
(PTEAM) field study conducted in liverside, CA (Pellizzari et al., 1992; Clayton et al.,
1993; Thomas et al., 1993). This study, conducted in fall of 1990 (September 22 through
November 9), used personal exposure monitors (PEMs) to measure the paniculate exposures
of 178 participants who were selected according to a statistical design to represent the people
living in the overall community. The PTEAM study also used stationary monitors to obtain
concurrent readings of indoor Mn concentrations in the homes and outdoor Mn
B-8
-------
concentrations in the backyards of each participant. The paniculate concentrations were
measured in two size fractions, PM]0 and PM2 5. The PTEAM study results have been
presented in several peer-review publications and other scientific forums (see
Attachment B-5).
2.1 Methodology of PTEAM
Stitdy Design
Probability-based representative sampling methods were used to select persons to
participate in the PTEAM study to allow inferences to be drawn from the sample to the
larger community population (Pellizzari et al., 1992). A three-stage probability sampling
procedure was used. Thirty-six areas within Riverside were selected for study following
socioeconomic stratification. Several homes from each area were sent letters explaining the
study. Interviewers then collected information about each household and invited eEgible
residents to participate. The 178 participants represented 139,000 ± 16,000 (S.E.)
nonsmoking residents of age 10 years and above.
Smokers were excluded from participating, but nonsmoking members of their families
were not. Employed persons were slightly oversampled because employment was thought to
be a possible risk factor for exposure to particles. (Occupational Mn exposures related to
motor vehicle use are reflected to some extent in the PTEAM data, but the study was not
designed to obtain representative occupational samples.) Each respondent wore a PEM
measuring PM10 for two consecutive 12-h periods (nominally 7 p.m. to 7 a.m. and 7 a.m. to
7 p.m.). Concurrent PM10 and PM2
-------
residents contacted, 443 (70%) completed the screening interview. Of these, 257 were asked
to participate and 178 (69%) agreed. More than 2,750 particle samples were collected (about
96% of those attempted). More details about the study design can be found in Pellizzari
et al. (1992).
Site Selection
The technical and logistical criteria considered in selecting Riverside as the PTEAM
study site included the following.
* Ambient aerosol levels should be somewhat heterogeneous and variable over the
study area to make it possible to discover covariates of the ambient levels,
* Ambient air pollution should not be dominated by a particular stationary source
(e.g., steel mills, smelters).
* Size-fractioned particle data (e.g., PM2.5, PM10, PM]5, or TSP) that could
provide a basis for determining relative source contributions must be easily
available.
• Housing stock (apartment, multifamily, single-family housing) in the communities
should be heterogeneous.
* The site must be readily accessible by personnel conducting the study.
Standard Metropolitan Statistical Areas (SMSAs) that would yield communities that
might meet many of these selection criteria included Riverside, Los Angeles, Philadelphia,
Baltimore, St. Louis, and Steubenville. Historic PM2,5, PM10, PM15, and TSP data were
analyzed, and the top 20 communities/SMSAs were ranked according to the highest levels of
PM2.s and PM10.
Based on an analysis of the U.S. Environmental Protection Agency inhalable particle
network and on recent air quality data from the California Air Resources Board, the
Riverside area was judged the most suitable for conducting the study. Riverside also met the
criteria listed above, and appeared to provide the greatest potential for observing a wide
range of ambient concentrations of inhalable particles. A wide range of outdoor levels in the
sample makes it easier to investigate a relationship between outdoor measurements and
indoor or personal measurements.
B-10
-------
The city of Riverside was operationally defined by selecting sample areas in terms of
the 1980 Census geographic units. The current (1990) city limits were approximated by
including 1980 Census blocks that had been annexed since 1980. The area around March
Air Force Base south of Allessandro Boulevard was excluded from the target population
because it was expected to have different inhalable particle concentrations from the rest of
Riverside (Pellizzari et al., 1992).
Measurement Methods
The PEM was designed to collect PMjQ using a sharp-cut impactor with a circular set
of holes 1.9 mm in diameter (Pellizzari et al., 1992, 1993; Clayton et al., 1993; Thomas
et al., 1993). Particles were collected at a flow rate of 4 L/min on a 37-mm Teflon® filter
mounted below a greased impactor plate. The PEM has been described in detail by
Pellizzari et al. (1992). A small quiet monitor for concurrent indoor and outdoor sampling
also was created. The stationary monitor to collect PM10 employed sampling heads and flow
rates that were identical to the PEM, but it operated off the home power line instead of
batteries. The sampling head was replaced with one having holes 1.4 mm in diameter to
collect fine particles (PM2.s). Laboratory studies indicated that the PEM and the PMjg
monitor had a sharp cut point at about 11 pm, whereas the PM2 5 monitor had a cut point at
2.5
2.2 Results
Interpretation of Graphs
Logarithmic probability plots provide a convenient way to represent the exposures of
large populations. Because the inference population consists of 139,000 people in Riverside,
the "Cumulative Frequency" on the abscissa of the graphs in Figures B-2 to B-9 describes
the proportion of the 139,000 people exposed at or below a particular concentration. The
ordinate of the graphs gives the exposure concentration in /ug/m . A logarithmic scale is
used because it helps "compress" the wide range of exposures, allowing them to fit on one
graph. A probability scale (the integral form of the normal distribution) is used on the
abscissa because graphs of environmental pollutant concentrations often appear as straight
B-ll
-------
1.000
0.700
0.600
0.500
0.400
0.300
0.200
0.100
£"•* ulbao
E 0.070
,3 0.050
0) 0.040
M 0.030
o
Q.
0.020
i
0)
g> 0.010
e 0.009
m 0.008
S 0.007
0.008
0.005
0.004
0.003
0.002
0.001
Personal Exposure Monitors (PEM
*«UI
XV
10 20 30 40 50 60 70 80 90 95
Cumulative Frequency (%)
98 99
1.000
0.700
0.600
0.500
0.400
0.300
0.200
0.100
0.090
0.080
0.070
0.060
0.050
0.040
0.030
0.020
0.010
0.009
0.008
0.007
0.006
0.005
0.004
0.003
0.002
0,001
99.8
figure B-2. Logarithmic-probability plot of daytime Mn PM10 personal exposure in
Riverside, CA, and concentrations measured inside and outside subjects'
homes.
Source: Pellizzari et al. (1992, Table 10-22, page 10-49).
B-12
-------
§
I
-------
n
I
kA.
ffi
.
NighttimeX,
0.010
0.009
0.008
0.007
0.006
0.005
0.010
0.
0.
0.007
0.006
0.005
Indoors
Daytime
Nighttime
0.001
99.8
Figure B-4. Logarithmic-probability plot of 12-h Mn FM2 5 indoor and outdoor
concentrations for both daytime and nighttime. The lowest data point for
nighttime indoors (not shown) is 0.0005.
Source: Pellizzari et al. (1992, Table 10-22, page 10-49).
B-14
-------
1.000
0.900
0.800
0.700
0.600
0.500
0.400 h
0.300 •
0.200
«- 0.100
c 0.090
-£. 0.080
5 0.070
— ' 0.060
§ 0.050
§ 0.040
0.030
0.020
O>
0.010
0.007
0.006
0.005
0.004
0.003
0.002
0.001
1.000
0.700
0.600
0.500
0.400
0.300
0.200
0.100
0.090
0.080
0.070
0.060
0.050
0.040
- 0.030
- 0.020
0.010
0.009
0.008
0.007
0.006
0.005
0.004
0.003
0.002
0.001
2 5 10 20 30 40 50 60 70 80 90 95 98 99 99.8
Cumulative Frequency (%)
Figure B-5. Logarithmic-probability plot of daytime outdoor Mn concentrations.
Source: Pellizzari et al. (1992, Table 10-22, page 10-49).
B-15
-------
0.001
10 20 30 40 50 60 70 80 SO 95 98 99
Cumulative Frequency (%)
0.001
99.8
Figure B-6. Logarithmic-probability plot of nighttime outdoor Mn concentrations.
Source: Pellizzari et al. (1992, Table 10-22, page 10-49).
B-16
-------
1.000
0.900
0.800
1.000
•O.S
0.001
0.700
0.600
0.500
0.400
0.300
0.:
0.100
0.090
0.080
0.070
0.060
0.050
0.040
0.009
0.008
0.007
0.008
0.005
0.004
•H 0.003
• 0.002
0.001
2 5 10 20 30 40 50 60 70 80 90 95 98 99
Cumulative Frequency (%)
99.8
Figure B-7. Logarithmic-probability plot of daytime indoor Mn concentrations.
Source: Pellizzari et al. (1992, Table 10-22, page 10-49).
B-17
-------
1.000
0.900
0.800
0.700
0.600
0.500
0.400
0.300 -
Q)
1.000
• 0!800
0.700
0.600
0.500
0.4
0.100
0.090
0.080
0.070
0.060
0.050 h
0.040 -
0.030 -
8 0.020 -
CO
09
S 0.010
0.009
0.008
0.007
0.006
0.005
0.004
0.003
0.002 -
0.001
2 5 10 20 30 40 50 60 70 80 90 95 98 99 99.8
Cumulative Frequency (%)
Figure B-8. Logarithmic-probability plot of nighttime indoor Mn concentrations.
Source: Pellizzari et al. (1992, Table 10-22, page 10-49).
B-18
-------
o.'soo
0.700
0.600
0.500
0.400
0.300
0.200
•s- 0-100
lil
O
to
O>
(0
0.060
0.050
0.040
0.030
0.020
0.010
o!oo7
0.006
0.005
0.004
0.003
0.002
0.001
Indoors
-Daytime
-Nighttime
1.000
0.700
0.600
0.500
0.400
0.300
0.200
0.100
0.090
0.080
0.070
0.060
0.050
0.040
0.030
0.020
0.010
0.009
0.008
0.007
0.006
0.005
0.004
0.003
0.002
0.001
2 5 10 20 30 40 50 60 70 80 90 95 98 99 99.8
Cumulative Frequency (%)
Figure B-9. Logarithmic-probability plot of indoor and outdoor Mn PM10
concentrations for both daytime and nighttime.
Source: Pellizzari et al. (1992, Table 10-22, page 10-49).
B-19
-------
lines when plotted on logarithmic-probability graphs. The reader can determine the
percentage of the Riverside population at or below a particular concentration by reading the
values of the two axes for each point on the line. For example, the curve for the daytime
PEM data in Figure B-2 passes through the 50% value (the median) at an exposure level of
0.049 jig/m3, and this result implies that 50% of the population (69,500 persons) had
daytime exposures to PM10 Mn up to 0.049 /ug/m3. Similarly, the 95th-percentile value is
0.170 pglm , so 5 % (6,950 persons) were exposed to daytime Mn concentrations above
0.170 jtg/m3 in Riverside in the fall of 1990. The date on which Figures B-2 to B-9 are
based were taken from Table 10-22 of Pellizzari et al. (1992), which is reproduced here in
Table B-2.
PEM and Indoor/Outdoor Concentrations
Figure B-2 shows that the daytime PM10 PEM values are higher than the indoor PM10
concentrations, presumably because people engage in activities that elevate their exposures
(driving a car, vacuuming, etc.), whereas nighttime PM10 PEM exposures (Figure B-3) are
nearly the same as those found indoors. Outdoor PM10 Mn concentrations tend to be higher
than indoor concentrations during the day and especially during the evening because the
larger particles settle out on walls and floors after entering the home.
J*M2 5 Concentrations
The absorptive characteristic of the home is less evident for the PM2.s particles
(Figure B-4). The frequency distribution of daytime outdoor PM2 5 Mn concentrations is
similar to the distribution for daytime indoor PM2 5 Mn concentrations, particularly at the
higher end (cumulative frequencies of 90 to 95%). Likewise, the distribution of nighttime
indoor PM2 5 Mn concentrations is similar to the distribution of nighttime indoor PM2j Mn
concentrations at the higher levels. This similarity between indoor and outdoor
concentrations occurs partly because the particles are so small that they do not settle out on
walls and floors as fast as larger particles. The frequency distributions of daytime and
nighttime PM2(s concentrations also are similar in shape. None of the distributions exceeds
0.05 figlm3 below the 99th percentile. Thus, less than 1 % of the Riverside population was
exposed to a PM2.s Mn concentration above 0.05
B-20
-------
TABLE B-2. SUMMARY OF PTEAM Mn MEASUREMENTS IN jtg/m3 Mn
Daytime
Percentile
10th
25th
50th (median)
75th
90th
95th
99th
PM2
SAM"
0.0042
0.0065
0.011
0.016
0.020
0.022
0.038
.5
SIM0
0.0010
0.0050
0.0082
0.012
0.019
0.024
0.045
SAM"
0.026
0.032
0.046
0.061
0.079
0.096
0.230
PM,0
SIMC
0.010
0.019
0.030
0.047
0.069
0.087
0.190
Nighttime*
PM2.s PMjQ
PEMa
0.019
0.028
0.049
0.093
0.140
0.170
0.390
SAM"
0.0022
0.0051
0.0086
0.014
0.019
0.023
0.032
SIMC
0.0005
0.0034
0.0063
0.010
0.017
0.020
0.026
SAM0
0.017
0.023
0.030
0.048
0.065
0.074
0.100
SIMC
0.0066
0.010
0.020
0.028
0.038
0.047
0.074
PEMfl
0.0084
0.013
0.020
0.031
0.049
0.058
0.072
Daytime samples were nominally 7 a.m. to 7 p.m. and nighttime samples were nominally 7 p.m. to 7 a.m.
SAM = Stationary ambient monitor placed outside subject's home.
SIM = Stationary indoor monitor placed inside subject's home.
PEM = Personal exposure monitor (identical to SIM and SAM monitors) carried by the subject during the identical period of operation of the SIM and SAM
monitors at the home.
Source: Pellizzari et al. (1992).
-------
Outdoor PM10 and PM2S Concentrations
The outdoor PM10 Mn concentrations are much higher than the outdoor PM2.5 Mn
concentrations, both in the daytime (Figure B-5) and at nighttime (Figure B-6). For
example, the median daytime PM10 Mn concentration outdoors is 0.046 jig/m3, compared
with 0.011 ftg/m3 for PM2 5. Over much of the range plotted, the daytime PM10 and PM2 5
distributions have similar slopes, but the PM10 Mn concentrations are about four times higher
than the PM2 5 Mn concentrations. A similar pattern occurs for nighttime outdoor
concentrations. The median of the nighttime PM10 concentrations is 0.030 /tg/m3, whereas
the median of the nighttime PM2 5 concentrations is 0.009 /ttg/m3, and the PM10
concentrations are about three times higher than the PM2 5 concentrations over most of the
distribution.
Indoor PMJO and PM2S Concentrations
When indoor PM10 and PM2 5 Mn concentrations are compared, PM10 concentrations
are much higher than PM2 5 concentrations, both in the daytime (Figure B-7) and nighttime
(Figure B-8). For example, the median for daytime indoor PM10 Mn concentrations is
31 3
0.030 ftg/m , compared with 0.008 jig/m for PM2 5. Similarly, the median for nighttime
indoor PM10 Mn concentrations is 0.020 /ig/m , compared with 0.006 ng/m for PM2>5.
Daytime and Nighttime PMIO Concentrations
A final analysis compares the daytime and nighttime PMio concentrations at the
residences (Figure B-9). The differences between daytime and nighttime Mn concentrations
are greater for PM10 than for PM2 5 (see Figure B-4). The outdoor PM10 Mn concentrations
in the day are higher than those at night. For example, the median outdoor PM10
concentration was 0.046 f*g/m in the day and 0.030 ftg/m at night. The 95% cumulative
frequencies on the two curves indicate that 5 % of the daytime PM10 concentrations are above
0.096 ftg/m3 and 5% of the nighttime concentrations are above 0.
The indoor daytime and nighttime Mn concentrations also differed. The indoor median
PM10 Mn concentration was 0.030 £tg/m in the day and 0.020 /*g/m at night. The
differences were, greater at the higher concentrations: 1 % of the homes were above
3 ' 3
0.190 jtig/m in the day; at night, 1 % of the homes were above 0.074 jig/m .
B-22
-------
3. COMPARISON OF CANADIAN DATA WITH PTEAM EXPOSURE
DATA
The Canadian and PTEAM Mn personal exposure data sets described to the previous
sections are summarized here in Figure B-10 and Table B-3. The data are shown in
Figure B-10 as the arithmetic means along with the range between minima and maxima of
the data sets. In the PTEAM study, the weighted average PM10 Mn exposure level for the
population of Riverside (Pelizzari et al., 1992, Table 10-18) was 0.069 jtg/m3 in the daytime
sample and 0.024 jtg/m in the nighttime sample, corresponding to a weighted 24-h average
of 0.046 jttg/m . This value is higher than the values generally reported from the Canadian
studies, with the notable exception of the mechanics' on-duty exposure in Montreal in 1992
of 0.250 j*g/m3 (Zayed et al., 1994). The differences between the Riverside and Canadian
PM5 measurements may be due in part to the fact that a larger size fraction (PM10) was
measured in the PTEAM study. In addition, the Mn data from the Canadian winter period of
January 1994, with high wind speeds and extensive precipitation, may well represent an
underestimate of the annual Mn exposure levels for the Canadian study areas. It is also
important to note differences in the scope of the Canadian and PTEAM studies, as the
PTEAM study greatly exceeded the Canadian studies in both time (7 weeks vs. 1 to 2 weeks)
and numbers of subjects sampled (178 vs. 5 to 24). Other differences in quality assurance
and data treatment have been noted in more detail above.
4. ESTIMATED Mn EXPOSURE LEVELS ASSOCIATED
WITH MMT USAGE
For risk assessment purposes, exposure and health effects assessments must be
compared and related. In deriving the current inhalation reference concentration (RfC) for
Mn from a study of health effects in Mn-exposed workers (Roels et al., 1992), measurements
of paniculate Mn collected by a sampler with a cut point of 5 j«n were used (Roels, 1993).
Although the RfC is expressed as a concentration independent of particle size, the most direct
and preferable comparison of population Mn exposure levels and the principal health effects
data from which the current Mn RfC was derived would be based on PMS Mn. Because of
its probabilistic sampling design and other features described above, the PTEAM study
(Pellizzari et al., 1992) provides the best available data, for estimating population
B-23
-------
Is)
0.30-
0.26-
£ 0.20-
e
0.16-
0.10-
O.OS-
Riverside, CA
Pelizzarietal.
(1992)
24-havg. PM10
Fall 1990 1°
Montreal, PQ Toronto, ON
Zayedetal.
11994)
8-hon, 16-hofl
June 1 99
^
T .1
2pM*
^2.07
_
^
MUG
(1994a,b)
24-r
June*
Amfatent Taxi Driven (offJ Mechanic* (off)
-------
TABLE B-3. COMPARISON OF AVERAGE Mn PERSONAL EXPOSURE LEVELS
FROM CANADIAN AND PTEAM STUDIES
Location Date
Riverside, CA Fall 1990
(Pellizzari et al.,
1992)
Montreal, PQ June 1992
(Zayedetal,, 1994)
Montreal, PQ January 1994
(Ethyl Corporation,
1994a,b)
Toronto, ON February 1991
(Lynam et ah,
1994)
Toronto, ON January 1994
(Ethyl Corporation,
1994a)
Sampled Population
General population
Taxi drivers, on duty
Taxi drivers, off duty
Mechanics, on duty
Mechanics, off duty
Taxi drivers
Office workers
Taxi drivers
Office workers
Taxi drivers
Office workers
Particle Size Measured
PM10
PM*
PM*
PM*
PM5
PM5
PM*
PM*
PM5
PM5
Mn (/ig/m )
0,046
0.024
0.00?
0.250
0,011
0.006
0.002
0.035
0,013
0.013
0.010
Size not specified in published report; characterized by Ethyl Corporation (1994b) as
Sampler ran for 7 days (40 h on-work, 128 h off-work).
exposure levels of Mn; it is also pertinent to an assessment of MMT health and exposure
risks in view of the fact that MMT was used in leaded gasoline in the Riverside area prior to
and during the period of the PTEAM study. However, the PTEAM study measured
personal exposure to paniculate Mn as PM10, not PM5. Moreover, the Mn measurements
from PTEAM reflected total Mn from all sources, not just automotive. Given that the focus
of the Office of Research and Development (ORD) exposure assessment is Mn in relation to
a projected use of MMT in unleaded gasoline, it is important to distinguish the contribution
of automotive sources to total Mn personal exposure in the PTEAM study.
In this section, data from the PTEAM study in Riverside, CA, and from other
independent studies in the vicinity of Riverside are analyzed to derive estimated personal
exposure levels of Mn that would be predicted to occur if MMT were used in unleaded
gasoline as proposed by Ethyl Corporation. Several steps in deduction and calculation are
B-25
-------
involved in arriving at these estimated personal exposure levels. As explained below, the
major features of this derivation process involve (1) estimating the automotive and
nonautomotive contributions to paniculate Mn based in part on a study by Lyons et al.
(1993), who measured PM size fractions up to PM^, (2) estimating personal exposure levels
of PM2 5 from PTEAM stationary indoor monitoring data for PM2 5 Mn; (3) adjusting the
PM2 5 personal exposure estimates to reflect PM in the size range from 2,5 to 4 /am, based
on the study by Lyons et al. (1993); and (4) projecting future PM4 Mn exposure levels in
Riverside in relation to increased usage of MMT in unleaded gasoline and increased gasoline
usage since 1990. Note that this approach estimates personal exposure levels of PM4 Mn
based on PM2 s data from indoor monitoring data rather than PM10 data from personal
samplers. Although direct measurements of personal exposure levels are preferable, the
automotive source contribution to PEM PM4 Mn can be more accurately estimated from SIM
PM2 5 data than from PEM PM10 Mn data, due to the much greater proportion of crustal Mn
in the PMjg fraction. Note also that the estimates of Mn personal exposure levels are limited
to PM4 (although PMs would be preferable) due to a lack of specific data for Mn PM in the
size range from 4 to 5 /*m in the Los Angeles-Riverside area; however, the mass of vehicular
source Mn in the range from 4 to 5 pm is thought to be somewhat less than the PRD and
crustal Mn in this size range, based on an extrapolation from the data of Lyons et al. (1993).
In the PTEAM study, personal Mn data were collected for approximately 12 h as a
daytime sample (nominally 7 a.m. to 7 p.m.) and 12 h as a nighttime sample (nominally
7 p.m. to 7 a.m.) and were separately tabulated as weighted averages and weighted
frequency distributions for the daytime and nighttime periods. In order to create a
distribution of daily (approximately 24-h) averages, the original data were retrieved and
matched by subject ID to develop a ranked list of daily exposure values by subject. In the
PTEAM study, there were 171 subjects with valid daytime values (see Attachment B-l),
168 subjects with valid nighttime values (see Attachment B-2), and 161 subjects with both
valid daytime and nighttime values (see Attachment B-3). Each subject was assigned a
sampling weight that was derived from the statistical survey design, as described by
Pellizzari et al. (1992). The mean daytime and mean nighttime Mn concentrations have been
recomputed for this assessment to verify that our procedures were identical to those used by
Pellizzari et al. (1992) to compute the values reproduced in Attachments B-l and B-2. Using
B-26
-------
this same procedure, the daytime and nighttime Mn concentrations for the 161 subjects were
combined to create the distribution of 161 24-h Mn PM10 levels shown in Attachment B-3.
Figure B-ll is a plot of the observed frequency distribution of daily personal exposures to
PM10 Mn for the population of Riverside, CA, as provided by the weighting functions. The
arithmetic mean PM10 Mn exposure of this distribution is 0.0456 /xg/m , with a standard
deviation of 0.0356 jtg/m3 and a 95% confidence interval of ±0.0054 jtg/m3.
4.1 Automotive Contribution to Mn PM in the Vicinity of Riverside, CA
Lyons et al. (1993) reported size distributions of selected trace metals, including Mn
and lead (Pb), sampled in the winter and summer of 1989 at two sites in the vicinity of
Riverside, CA, Upland and Pico Rivera. Samples were collected by an eight-stage low
pressure impactor, which had 50% efficiency cutoffs in aerodynamic diameters of 4.0, 2.0,
1.0, 0.5, 0.26, 0.12, 0.075, and 0.05 j*m. Figure B-12 shows the distributions of the total
mass of Mn collected in these size fractions.
Based on data on the ratios of iron (Fe) to Mn and Fe to Pb in paved road dust (PRD)
in the Los Angeles area (Cooper et al., 1987) and data on the contribution of crustal Mn to
PRD in Upland and other areas around Los Angeles (Gray et al., 1988 [also see Gray et al.,
1989]), Lyons et al. (1993) were able to distinpish the automotive versus PRD
(or suspended dust, in Lyons' terminology) contributions to the PM4 Mn they collected in
Upland and Pico Rivera. Although PRD included some settled participate emissions of Mn
from fuel combustion, tire particles, brake lining dust, and other particles from automotive
sources, PRD was primarily a function of soil dust and, as such, represented the earth crustal
material in a source apportionment analysis (Gray et al., 1989). Thus, by subtracting the
contribution of PRD (suspended dust), Lyons et al. (1993) were able to determine the Mn
mass attributable to fresh emissions from automotive sources, as shown in Figure B-13.
Table B-4 shows the total PM4 (column A) and automotive PM4 (column B) depicted in
Figures B-12 and B-13. From these data, the relative contribution of automotive sources to
PM4 Mn can be calculated (dividing column B by column A) as an average of approximately
62%.
B-27
-------
oisoo
0.700
0.600
0.500
0.400
0.300
0.200
0.100
S~* 0.090
E 0.080
"o> 0.070
3.
£
g 0.040
O
Q.
&
0.030 ••
-------
IS
£0
s
JS
OB
oo
Winter Upland
Concentration - 25 ng/m1
Mn ~f
]
T
/'-
1+
r-
0.01 0.10 1.00
Aerodynamic diameter dp(t»m)
2.0
0.0
Summer Upland
Concentration - 24 ng/m*
Mn
OA1
0.10 1.00
Aerodynamic dlamaler dp(|im)
10.00
25
8,0
§
1*
0X1
Winter Pico
Concentration - 27 ng/m*
Mn
fi H-
0.01 aio t.oo
Aerodynamic diameter dp(nm)
10.00
2S
2.0
0.6
Summor Pkso
Conoentration • 22 ng/m»*
i
Mn
0.0 •—
0.01
0.10 1.00
Aerodynamic diameter dp(nm)
10.00
*An Mn concentration of 28 ng/m for Pico Rivera in summer is reported in Figure 2 of
Lyons et al. (1993), but a concentration of 22 ng/m is reported for the same place and time
in Figure 4 of their report. According to Lyons (1994), the latter value (22 ng/m ) is
correct.
Figure B-12. Size distributions of Mn measured at Pico Rivera (Pico) and Upland, CA,
in the winter and summer of 1989.
Source: Lyons et al. (1993).
By computing the area under the curve (AUC) in Figures B-12 and B-13, it is
possible to estimate the mass of Mn in the PM2<5 size fraction. Because Lyons et al. (1993)
did not measure the PM2.5 fraction as such, a proportional amount of the AUC for the 2 to
4 /*m fraction was added to the AUC for the 0.05 to 2 /am fractions to estimate the PM2 5
B-29
-------
2.0
[1JO
as
Mn
Winter Upland Corrected for
Suspended Dust
Concentration -15 ng/m'
0.10 1.00
Aerodynamic diameter dpftim)
10.00
Summer Upland Corrected for
Suspended Dust
Concentration -13 ng/m*
1.0
Mn
Aerodynamic diameter dp(iun)
2.0
3
J'-O
OJO
Winter Rco Corrected for
Suspended Dust
Concentration -15 ng/m*
Mn
am 1.00
Aerodynamic diameter dp(|im}
s
Summer Pico Corrected for
Suspended Dust
Concentration -17 ng/m*
Mn
0.10 1.00
Aerodynamic diameter dp(»m)
Figure B-13. Estimated size distributions for Mn with the suspended dust component
subtracted.
Source: Lyons et al. (1993).
fraction. Based on these AUC calculations, Table B-4 shows the total PM2 5 (column D) and
automotive PM2 5 (column E), from which the relative contribution of automotive sources to
total PM2 5 can be calculated (dividing column E by column D) as an average of
approximately 75%. Subtracting PM2 5 from PHj yields the total (column G) and
automotive (column H) Mn mass in the PM size range from 2.5 to 4 pm. The average
relative contribution of automotive sources to Mn PM in the 2.5 to 4 /*m size range (dividing
column H by column G) is therefore 14%. (The decreasing percentage automotive
contribution with increasing particle size is evidence that the PEM PM10 Mn data from
PTEAM may be strongly influenced by nonautomotive sources and would not, as noted
B-30
-------
TABLE B-4. DATA DERIVED FROM LYONS ET AL. (1993) USED TO ESTIMATE AUTOMOTIVE
SOURCE CONTRIBUTION TO DIFFERENT SIZE FRACTIONS OF Mn PARTICULATE MATTER (ng/m3)
Location
Upland Winter
Pico Rivera Winter
Upland Summer
Pico Rivera Summer
Average Fraction
Auto + PRD
A
25
27
24
22a
PM4
Auto Only
B
15
15
13
17
Auto Fraction
(B/A)
C
0.600
0,555
0.542
0.773
0.618
PM2.5
PM4 - PM2 5
Auto Fraction
Auto + PRD
D
20.8
22.0
15.5
17.7
Auto Only
E
15
15
10.6
15.8
(E/D)
F
0.721
0.682
0.684
0.893
0.745°
Total
(A-D)
G
4.2
5
8.5
4.3
Auto
(B-E)
H
0
0
2.4
1.2
Auto Fraction
(H/G)
I
0
0
0.28
0.27
0.14
W
aAn Mn concentration of 28 ng/m for Pico Rivera in summer is reported in Figure 2 of Lyons et al. (1993), but a concentration of 22 ng/m is reported for
the same place and time in Figure 4 of their report; according to Lyons (1994), the latter value (22 ng/m ) is correct.
bLyonsetal. (1993) used a Fe:Mn ratio of 50:1 for PRD in the Los Angeles area; according to Gray et al. (1989) the Fe:Mn ratio for PRD in Riverside is 41:1.
Use of the latter ratio changes the auto fraction from 0.745 to 0.69,
-------
above, be as useful as PM2.5 da** ^or estimating personal exposure levels of Mn originating
from automotive sources.)
Note that the winter values for Upland and Pico Rivera were shown as zeros because of
the calculation procedure used by Lyons et al. (1993) to subtract PRD. Their procedure used
the Fe:Mn ratio (50:1) in PRD in the Los Angeles area, which in these two cases resulted in
negative values that were set to zero. The negative values indicate that their procedure
apparently overcorrected for the PRD contribution. Thus, the percentage contribution of
automotive sources to PM4 Mn derived from the report of Lyons et al. (1993) could be
somewhat higher than that presented here. However, data on the Fe:Mn ratio for PRD for
Riverside (Gray et al., 1989), which is more pertinent than a ratio for the general Los
Angeles area, indicates that a ratio of 41:1 is more appropriate than 50:1. If the data from
Lyons et al. (1993) are adjusted on the basis of a 41:1 Fe:Mn ratio for PRD, the percentage
contribution of automotive sources declines from 75 to 69%.
4.2 Personal Exposure Levels of Automotive Source PM2 5
In the PTEAM study, PM2 5 Mn measurements were obtained in both the subjects'
homes by a stationary indoor monitor (SIM) and immediately outside the homes by a
stationary ambient monitor (SAM) during two 12-h periods. These SIM PM2 5 Mn
3 3
measurements averaged 9.8 ng/m in the day and 7.5 ng/m at night (Table B-5). As is
evident in Table B-5, the PEM levels of PM10 Mn were higher than the SIM levels of PM10,
particularly during the day (see also Figures B-2 and B-3). This pattern of higher PEM than
SIM levels was clearly evident for all of the measured elements during the daytime in the
PTEAM study and for almost all of the nighttime measurements. (The reason for this
pattern is poorly understood but may be related to activities such as driving a car hi heavy
traffic, parking in an indoor parking garage, or working around motor vehicles.) Therefore,
a PEM adjustment factor was necessary to estimate personal exposure levels based on PM2 5
Mn data obtained from stationary indoor monitors. Assuming that the same ratio of PEM to
SIM for PMjQ Mn would apply to PM2.5 Mn, an estimated PEM level of PM2 5 Mn was
calculated by multiplying, respectively, the day and night PEM:SIM ratios for PM10 Mn by
the daytime and nighttime average PM2.s Mn values shown in Table B-5 (i.e., daytime:
B-32
-------
TABLE B-5. PTEAM WEIGHTED MEAN ELEMENTAL
CONCENTRATIONS (ng/m3)8
Element
Mn
Pb
PM
SAM
12
20
2.5
SIM
9,8
17
Daytime
SAM
51
30
PM10
SIM
38
27
Nighttime
PM2.5
PEM
69
40
SAM
9.9
23
SIM
7.5
20
SAM
37
32
PM,0
SIM
22
27
PEM
24
26
aData are weighted to reflect the target population of person-days (PEM) or household-days (SIM and SAM),
Source: Pellizzari et al. (1992).
9.8 x (69/38) = 17.8 ng/m3; nighttime: 7.5 x (24/22) = 8.2 ng/m3). Averaging these
two values yields an estimate of a PEM measure of PM2,s Mn (i.e., 13 ng/m3).
To estimate the personal Mn exposure level in Riverside in 1990 due to automotive
sources, the above value of 13 ng/m3 is multiplied by the percentage of PM2 5 attributable to
automotive sources (i.e., 69%) to yield 9 ng/m ; by subtraction, 4 ng/m of PM25 Mn may
be attributed to nonautomotive sources (i.e., PRD or suspended dust).
To deduce the personal exposure levels for PM4 Mn, an adjustment of the PM2 j levels
was based on the ratio of the automotive PM2 54 (column H, Table B-4) to automotive
PM2 5 (column E, Table B-4), which averages 0.08. Therefore, increasing the automotive
PM2 5 level of 9 ng/m3 by a factor of 0.08, the estimated PM4 Mn personal exposure level
in Riverside, 1990, would be 9.7 ng/m3.
4.3 Projections of PTEAM Exposure Levels with Increased MMT Usage
As explained more fully in Attachment B-4, approximately 13,6% of the gasoline sold
in California around the period of the PTEAM study contained MMT at a production-volume
weighted average of 0.048 g Mn/gal (at refineries in the Los Angeles area). If 100% of the
gasoline sold were unleaded gasoline and contained 1/32 g Mn/gal, Mn emissions would
increase by an estimated factor of 4.8, assuming the same relative amount of gasoline
consumption and the same average emission rate (grams Mn emitted per gram Mn
combusted) as that which existed around the time of the PTEAM study in Riverside.
However, overall unleaded gasoline consumption is expected to increase due to an increase in
vehicle miles traveled [VMT]) by an average of approximately 1 % per year between 1990
B-33
-------
and 1995 (the first full year in the near-term future). Therefore, MMT usage and, hence,
Mn emissions would also increase. Adjusting the factor of 4.8 for a 5% increase in gasoline
usage between 1990 and 1995 yields an MMT-increased-usage factor of 5. Multiplying this
factor of 5 by the estimated automotive PEM level of PML* Mn (9.7 ng/m ), the arithmetic
mean PM4 Mn level due to automotive sources is 49 ng/m3, or 0.049
To determine the projected total PM4 Mn exposure level, the contribution of
nonautomotive sources must be added to the automotive source projection. The
nonautomotive contribution to PM4 Mn was derived by adjusting the above estimate of
4 ng/m for nonautomotive PM2 5 Mn in a manner similar to that used for the automotive
portion of PM2 5. Subtracting the automotive portion from the total yields the nonautomotive
portions for PM2.5ut (column G minus column H, Table B-4) and for PM2 5 (column D
minus column E, Table B-4). The ratios of these values average 1.08, which when
multiplied by the estimate of the PM2 5 nonautomotive fraction (4 ng/m ) yields 4,3 ng/m
for the PM size range of 2.5 to 4 ^tm. Adding 4.3 and 4 yields 8,3 ng/m3 for the
nonautomotive PM4 Mn exposure level. Combining the nonautomotive and projected
automotive exposure levels, the arithmetic mean personal exposure level of PM4 Mn (with
1/32 g Mn/gal MMT in 100% of the unleaded gasoline) would be 57 ng/m3, or
0.057 jtg/m3.
To estimate the distribution of PM4 Mn exposure levels, the geometric standard
deviation (ag) of the PM4 Mn distribution is assumed to be identical to that of the PM10 Mn
distribution in Figure B-ll (as suggested by the similarity of the slopes for PM25 and PM10
in Figures B-7 and B-8). Given that these data are approximately lognormally distributed,
the mean is related to the median for either size fraction as mean = median x
exp([ln 0g] 12), Therefore, a scaling factor may be derived by dividing the PM4 arithmetic
mean by the PM10 arithmetic mean (i.e., 57 / 46.5 = 1.23). Multiplying the values plotted
in Figure B-ll by the factor of 1.23 yields a projected distribution of PM4 Mn exposure
levels if MMT were used as proposed by Ethyl Corporation. This projected distribution is
shown as line 1 in Figure B-14.
A second projected distribution is shown in Figure B-14 as line 2. The second estimate
differs from the first in the approach used to estimate PEM PM2 5 from SIM PM2 §5 Mn.
(A PEM adjustment factor was used to correct for the consistently higher values of PEM
B-34
-------
1.000
0.900
0.800
0.700
0.600
0.500 -
0.300 -
0.200 •
0.100
«T 0.090
£ 0.080
"o> 0.070
^ 0.060
-------
than of SIM in the FTEAM data for day and night PM10 Mn measurements.) The first PEM
adjustment factor, which was used in deriving the estimates represented by line 1, was based
on the ratio of PEM:SIM for PM10 Mn. The first approach is justified in part because of the
consistency between the SAM:SIM ratios (approximately 1.2:1) for both Pb and Mn in the
FTEAM data (see Table B-5). As discussed by Lyons et al. (1993), such consistency
suggests that both Mn and Pb in the PM2 5 fraction come from the same automotive source,
which might be expected because Mn was present in only leaded gasoline around the time of
the studies under consideration. However, based on data from Cooper et al. (1987), there is
reason to suppose that there may be more crustal Mn than Pb in the Riverside PM10, and
therefore, an adjustment factor based on the PM10 Mn data could possibly overstate the
automotive contribution in the estimation of PEM PM2>5 from the PTEAM Riverside data.
Because Pb was also a gasoline additive during the period of the PTEAM study, it provides
another means of determining a PEM adjustment factor. Using the daytime and nighttime
ratios of PEM:SIM for PM10 Pb (see Table B-5), the calculations to derive a projected
arithmetic mean exposure level of PM4 Mn may be summarized as follows:
PEM PM2 5 » average of adjusted daytime and nighttime SIM PM2 5
_ T9.8 X (40/27)1 + 17.5 X (26/27YI
2
~ 10.9 ng/m3.
Automotive PM2 5 = 10.9 x 0.69 (from column F, Table B-4, adjusted for PRD Fe:Mn
ratio of 41:1)
= 7.5 ng/m3.
PEM PM4 = 7.5 x 1.08 (the PM4:PM2 5 ratio for automotive Mn)
= 8.1 ng/m3.
Projected PEM PM4 = 8.1 x 5 (the MMT increased usage factor)
= 40.5 ng/m3.
B-36
-------
Nonautomotive PM2.5 = 10.9 - 7.5
= 3.4 ng/m3.
Nonautomotive PM2 5^ Mn = 3,4 X 1.08 (i.e., PRD PM2>5.4:PM2 5)
= 37 nm3
3.7 ng/m3
Nonautomotive PM4 = PRD PM2 5 + PRD
= 3.4 + 3.7
= 7.1 ng/m3.
Total PM4 ME exposure = automotive VM^ + nonautomotive PM4
= 40.5 + 7.1
= 47.6 ng/m3 or 0.048 /*g/m3.
By the previous assumption of equal geometric standard deviations for the PM4 and PM10
distributions, a scaling factor may be derived by the ratio of the arithmetic means for PM4
and PM10 (i.e., 48/46.5 = 1.03), which is a reasonably close approximation to the first
estimate. The resulting distribution of PM4 is represented by line 2 in Figure B-14.
A distribution of individual 24-h average exposure levels, such as that of the PTEAM
Riverside study, is likely to overestimate the upper tail of the distribution of longer term
average exposure levels, which would be of more relevance to an RfC or health assessment
of MMT than 24-h exposures. The most desirable way to estimate long-term personal Mn
exposures is to have data on long-term personal Mn exposures (by sampling the same
individuals over the period in question). In the absence of such data, different approaches to
adjusting the variance of the 24-h average distribution are possible (Wallace et al., 1994).
One approach to this problem is to use the two consecutive 12-h personal exposures
actually measured for each person in the PTEAM study as an estimate of the temporal
variation in personal exposure levels (Wallace et al., 1994). A related alternative, based on
the work of Wallace et al. (1994), uses the observed geometric means of the distributions
instead of the observed geometric standard deviations. By either approach, the natural
logarithm of the
-------
Another approach to estimating the variance in personal exposure levels over a longer
term period relies upon 24-h PEM PM2.5 Mn exposure data (see Attachment B-7) collected
on the same individuals for nonconsecutive days in the pilot study that preceded (in March
1989) the PTEAM Riverside study. From a set of 16 pairs of these data, separated by one
day, it is possible to calculate the intrapersonal variance of the logarithms of these 24-h
exposures and thereby estimate the natural logarithm of
-------
l.UUU
OQflft
A ffcnn
0.700
U.oOO
0.500
n Ann
n son
A 3Aft
U.cUU
0100
n u.uyu
c n nun
C U.UOU
O) 0.070
HI ft ftfifi
^~ u.uou
^ 0.050
— ' n i"Ldn
CO U.U*W
O
ft n oao
lU
M n nin
/langanei
0 C
2 g
9 C
ones
0008
0007
0006
0005
n (VLi
0003
0009
0001
0.
X"
20.
X
5 1
X
£
1
X
>
<*
2
s ^
^>
0 2
fX
0 3
/
0 4
X
^
0 £
X
ioe
x
X
107
.x
rf"
0 fl
>•;
X,
jjX»"'
0 9
1,,1J-i«!
^
0 g
X
r
5 9
X
X
39
9 a
0900
0800
OTfift
.700
ORnn
OKflfl
0400
0300
n win
0100
n nan
OOBO
On?A
.U(U
Ons/i
OfWI
004O
00%)
ft flQfl
n nin
n nno
0.007
n fVus
O.UUD
n nnc
u.uuo
0.004
0.003
0.002
ft ftA4
U.UUl
3.8
Population Cumulative Frequency (%)
Figure B-1S. Logarithmic-probability plot of projected estimates of long-term personal
exposure levels of PM4 Mn for the fall season in Riverside, CA, if MMT
were used in 100% of unleaded gasoline at 1/32 g Mn/gaL The two lines
reflect different approaches for estimating personal exposure levels from
stationary indoor monitoring data and different methods of adjusting 24-h
averages to long-term averages, as explained in the accompanying text and
Appendix B.
B-39
-------
example, an adjustment were made for the presence of some automotive-source Mn in PRD.
It is also possible that the adjustment of PM2-s data based on the PEM:SIM ratio for PM10
Mn could be biased upward by the presence of more crustal Mn in the PM10 fraction,
although examination of size fraction data for Pb and Mn from the Caldecott Tunnel (Lyons
et al., 1993) indicates more Pb in the 2 to 4 pm PM fraction than in the 1 to 2 /*m fraction
and the opposite pattern for Mn. Given the coexistence of Pb and Mn in leaded gasoline to
California, the PEM:SIM ratio for PM10 Pb provides an alternative means of adjusting the
PM2.5 SIM data to estimate automotive-source PM2 j PEM levels. To the extent that the
results obtained by the two approaches (one based on the Pb PEM:SIM ratio, the other based
on the Mn PEM:SIM ratio) tend to converge, confidence in the validity of this method is
enhanced. (Note that various assumptions are involved in comparing the two approaches.
For example, although both Pb and Mn were present in leaded gasoline in California, the
ratio of the two metals varied from refinery to refinery. Also, the size fractions emitted by
vehicles may have differed for Pb and Mn.)
The comparability of line 2 to line 1 in Figure B-14 is an indication that the approach
used to derive personal exposure levels from stationary indoor monitoring data is reasonable
and appropriate. To make no such adjustment of the PTEAM PM2 5 data in this respect
would not create a reasonable lower bound; it would simply be in error. However, the two
resulting estimates are simply alternatives and should not be denoted as bounds to a range of
values. Nevertheless, two reasonable estimates of projected distributions of long-term
average personal exposure levels of PM4 Mn for the fall season in Riverside are represented
by lines 1 and 2 in Figure B-14..
The MMT increased usage factor used in the derivation of exposure projections
involves multiple sources of data, deductive steps, and assumptions, as described in more
detail in Attachment B-4. One matter of potential significance is the assumption of
equivalent Mn emission rates from leaded gasoline-fueled and unleaded gasoline-fueled
vehicles. As noted in Attachment B-4, the percentage of the Mn in the fuel actually emitted
by a vehicle varies depending on many factors, including: operating conditions; the age,
mileage, and maintenance of the vehicle; and the type and condition of its emissions control
and exhaust systems. The limited data available on Mn emissions are not adequate to
indicate that there would be any substantial difference in emission rates between
B-40
-------
catalyst-equipped vehicles using unleaded gasoline and noncatalyst vehicles using leaded
gasoline in 1990. Therefore, for puiposes of this analysis, the Mn emission rates for these
two vehicle groups were considered to be equivalent. Nevertheless, it should be recognized
that the exposure projection estimates are sensitive to the MMT increased usage factor and
hence to the assumption of equal Mn emission rates. The projections of automotive PM4
exposure levels with 100% MMT usage would be lowered in direct proportion to the
assumed relationship between Mn emission rates from leaded and unleaded gasoline-fueled
vehicles. For example, if the emission rates for noncatalyst cars and catalyst cars were
assumed to differ by 25%, the MMT usage factor would be reduced by 25%. However,
because the nonautomotive Mn would be independent of the automotive Mn, the net exposure
levels would be lowered by roughly only 22%. These calculations are presented only to
illustrate the sensitivity of the derivation to a difference in emission rates; they do not
represent a quantitative lower bound to OlD's best estimates presented in Figure B-15.
Other uncertainties due to inadequate data must also be acknowledged. The estimate of
the Mn fraction due to automotive sources is based largely on the report of Lyons et al.
(1993). Uncertainties related to the most appropriate apportionment factor for automotive
versus PRO sources (based on the Fe:Mn ratio) were previously noted above. Other
uncertainties in the study by Lyons et al. (1993) pertain to the accuracy and precision of the
sampler instrument and the method used to analyze the Mn content.
The limitation of the PTEAM Riverside study to measurements made during 24-h
periods over a 7-week span in early fall presents other uncertainties. As noted above, a
better estimate of long-term personal exposures requires that sampled individuals be
measured at least twice at sufficiently separated times to account for seasonal variation and
the autocorrelation in personal exposures. Attempts to adjust 24-h average exposures to
longer term periods or to other seasons introduce progressively greater uncertainties that
reduce confidence in the ultimate estimates.
Other variables that could influence these projections have been judged either negligible
or too uncertain to include. The contributions of point sources to PM4 Mn in the Riverside
area are assumed to be negligible with the prior closing of the Fontana steel mill. The
accuracy of the data on projected total fuel usage affects the accuracy of the estimate of
future Mn exposure due to MMT. Also, the projections described above do not consider the
B-41
-------
potential impacts, if any, of electrical, natural gas, or other altemative-ftiel vehicles, or of
diesel fuel vehicles. Discrepancies or uncertainties in the cut points of samplers used to
collect the data considered here are assumed to be negligible.
Because the health data of focal interest in the health effects assessment were measured
in relation to PM5 Mn (Roels, 1993), the limitation of exposure projections to PM4 Mn
implies the possibility of underestimating population exposure levels of ?!% Even if no
increase in the automotive contribution to particulate Mn in the size fraction between 4 and
5 ^m were predicted, total PM5 exposure levels (i.e., including nonautomotive source
contributions) would presumably be higher than PM4 levels, if only slightly higher,
4.5 The Relationship of Riverside to Other U.S. Communities
Riverside was chosen for the PTEAM study because it served to represent the Los
Angeles (L. A.) area and a situation where there was significant PMjg air pollution.
Meteorology plays an important role in creating the distribution of PM10 in the L.A. Basin,
where the predominant weather pattern is the diurnal onshore sea breeze that brings the air
mass over downtown L.A. and on up through the Basin eastward toward Riverside. Primary
pollutants, such as carbon monoxide, maximize over the downtown area, and secondary
pollutants, such as the fine fraction of PM10 nitrates, increase with time as the air mass flows
eastward. The PM10 maximizes at, or about, Rubidoux, which is some 15 km from
Riverside (Cohanim et al., 1991). As shown in Table B-6, the TSP (PM27) is higher in
Rubidoux than in Riverside, which in turn is Mgher than in downtown L.A. The PM10 data
are not available for Riverside in the report of Cohanim et al. (1991), but the same ordinal
relationship between Rubidoux and L.A. holds for PM10. According to Hoggan (1994), the
arithmetic mean ambient PM10 Mn was 0.066 /*g/m3 in Rubidoux in 1985 to 1986. The
mean ambient PM10 Mn in Riverside in 1990 at a central PTEAM site was 0.036 /tg/m
(Pellizzari et al., 1992, Table 10-17).
A detailed meteorological analysis of the PMJO data at Rubidoux (highest of the three
L.A. Basin sites) shows "that on the average, about half of the PM10 mass consists of road
dust and wind blown dust, most of it of local origin" (Hoggan et al., 1993). In Rubidoux in
1985 to 1992, the PM10 in the summer (May through October) averaged some 40% higher
B-42
-------
TABLE B-6. RELATION OF TSP (PM27) MEASUREMENTS 0*g/m3) IN RIVERSIDE
AND OTHER STATIONS IN THE LOS ANGELES BASIN
Year
1987
1988
1989
1990
Average
Los Angeles
Oig/n?)
0.091
0.107
0.114
0.105
0.104
Riverside
(jig/m3)
0.107
0.133
0.126
0.106
0.118
Rubidoux
0*g/m3)
0.117
0.144
0.147
0.127
0.134
Source: Cohanim et al. (1991).
than to the winter (November through April), with values on the order of 100 and 70
respectively (Hoggan et al., 1993). Although air-Pb data for L.A. are not available for
1990, Cohanim et al. (1991) report that for 1987 to 1989, when virtually all detected lead
was from motor vehicle emissions, Pb was maximized in the downtown L.A. areas of highest
traffic levels (Lynnwood and Pico Rivera), as expected for a primary pollutant. Lead, and
presumably Mn from MMT, followed the same pattern of maximization to December
through February and minimization in May through August. In summary, different
meteorological conditions to the L.A. basin to summer and winter lead to the maxima for
PM10 to the summer (when high summer winds and dry conditions increase the crystal
material to the PM10) and maxima for Pb fa the winter (when weaker winds and stronger
inversions increase the buildup of automotive Pb and Mn from traffic). Thus, the fall time
period and location of the PTEAM study appear to be intermediate to terms of paniculate
pollution levels to the L.A. area.
As shown to Table B-7 (from Davis et al., 1984), Riverside also appears to be
intermediate among 22 U.S. sites, particularly for fine mode Mn. Within the United States,
TSP (PM27) Mn data for 1990 have a wide range of mean values, from 9.6 j*g/m3 near a
battery plant to Covington, TN, and 0.67 pg/m3 to Granite City, EL (where steel is
manufactured), to values approaching minimum detectable levels at rural locations, such as
0.003 pgfm3 to Columbia, SC (AIRS, 1994). Figure B-16 shows data for PM10 Mn from
the EPA IP Network (1984) for the years 1979 to 1983. Rubidoux, the location to the
network closest to Riverside, had a mean ambient PM10 Mn level of 0.038 ftg/m3, which
placed it at the 74th percentile of the national distribution. This is consistent with data from
B-43
-------
TABLE B-7. SUMMARY OF Mn CONCENTRATIONS (/tg/m3) FROM
22 LOCATIONS IN THE IP NETWORK, 1980a
Location
Akron, OH
Portland, OR
Ingleoook, AL
Buffalo, NY
Tarrant, AL
Honolulu, HI
Dallas, TX
Cincinnati, OH
Hartford, CT
Riverside, CA
Kansas City, KS
Kansas City, MO
San Jose, CA
Five Points, CA
Minneapolis, MN
El Paso, TX
Seattle, WA
Research Triangle Park, NC
Winnemucca, NV
St. Louis, MO
Braidwood, IL
Boston, MA
Fine (PM2.5) Mn
0.085
0.052
0.037
0.033
0.032
0.020
0.015
0.011
0.007
0.007b
0.006
0.006
0,006
0.005
0.005
0.004
0.004
0.003
0.003
0.002
0.001
0.001
Coarse (PM2.5_i0) Mn
0.044
0.056
0.041
0.078
0.035
0.014
0.020
0.020
0.021
0.042b
0.020
0.025
0.021
0.050
0.026
0.023
0.018
0.003
0.041
0.017
0.017
0.029
Total PM10 Mn
0.129
0.108
0.088
0.111
0.067
0.034
0.035
0.031
0.028
0.049b
0.026
0.031
0.027
0.055
0.031
0.027
0.022
0.006
0.044
0.019
0.018
0.030
Arranged in descending order of fine mode Mn concentration.
^Riverside rank: 9-10th for fine (tied with Hartford); 5th for coarse; 7th for total PM10.
Source: Davis et al. (1984).
Davis et al. (1984), as shown in Table B-7. Although many (but certainly not all) of the
locations with higher Mn levels had major industrial sources, the ambient PMi0 Mn level in
Riverside in 1990 (0.0465 /tg/m3) was comparable to that of the Rubidoux average for 1979
to 1983 (0.038 Mi/m3).
Given that no MMT was used in California gasoline during the years 1979 to 1983
(Davis et al., 1988), a rough estimate of the increment in ambient Mn levels due to MMT
usage can be derived from the increase in PM10 Mn in the Rubidoux area between the period
of 1979 to 1983 (0.038 /*g/m3) and 1990 (0.0465
), assuming that nonautomotive
sources remained constant and that any increment in PM10 was due to an increase in PM4
from automotive sources. The difference (0.0465 - 0.038 = 0.0085 /*g/m3) is quite
B-44
-------
w
I
1.000
0.900
0.800
0.700
0.600
0.500
0.400
0.300,
0.200
0.100
0.090
0.080
0.070
0.060
0.050
0.040
0.030
0.020
'•»
«
0
••
@©c
^
©•
Cle
f Rubidoux, CA
Rank 103 of 139
O-^Phili
< Ptttsburf
Beloit, Wl
/eland, OH
idalphla, I
h,PA
1.0CO
0.900
0.800
0.700
0.600
0.500
0.400
0.300
»A
0.200
0.100
0.090
0.080
0.070
0.060
0.050
0.040
0.030
0.020
n mn
0.01
I
0.1 0.5 1 2
5 10 20 30 40 50 60 70 80 90 95 98 99
Cumulative percentage
99.9 99.99
Figure B-I6. Cumulative frequency distribution of cities in the EPA Inhalable Particulate Network, 1979 to 1983, with ambient
PM10 Mn concentrations less than the specified levels.
Source: IP Network (1984).
-------
comparable to the estimated personal exposure PM^ Mn level of 0.0097 jtg/m due to
automotive sources in Riverside in the fall of 1990 (see above).
It is interesting to note that Davis et al. (1988) independently derived a value of
0.009 /tig/m3 for automotive-source Mn in the ambient PM2 5 fraction for the period July to
December 1985 in Riverside. The authors also reported that, for the same half-year period,
the ambient PM10 Mn level in Rubidoux was lower (0.045 jig/m3 Mn from July to December
1985) than in the following half-year (0.058 j*g/m from January to June 1986). The annual
average of PM10 Mn (0.051 /*g/m3) during this period (July 1985 through June 1986) in
Rubidoux was 0.013 j*g/m higher than the 0.038 jtg/m measured at the Rubidoux
JJP Network (1984) station prior to the introduction of MMT. (Hoggan [1994] reported an
average PM10 Mn level of 0.066 ftg/m3 for the full 2-year period of 1985 and 1986 at
Rubidoux.) If the increase of 0.013 fig/m from the IP Network average value of
0.038 /ig/m3 is all in the PM4 fraction, multiplying 0.013 jttg/m by a factor of 1.16 for the
increase in MMT usage from 1985/86 to 1990 (see Attachment B-8) yields 0.015 ftglm3
ambient PM4 Mn, which is even higher than the above estimated mean of 0.0097 /tg/m
PEM PM4 Mn for Riverside in the fall of 1990. Also note that the annual average
automotive-source Mn in Rubidoux in 1985 to 1986 was 0.016 /ig/m , which is comparable
%
to an average of 0.013 ^g/m for all southern California sites during the same period (Davis
et al., 1988). The similarity of these annual averages suggests that the Rubidoux-Riverside
area was not atypical in relation to southern California. However, the automotive Mn in the
period January to June 1986 was 0.021 jtg/m , whereas the July to December 1985 Mn
(which included a fall season, as sampled by PTEAM in 1990) was much lower,
0.011 jtg/m35 which suggests that seasonal differences could be significant.
Underlying the similarity of PMjQ pollution levels in Riverside to other areas of the
L. A. Basin are the relatively high levels of VMT in the greater L. A. metropolitan area. The
four counties making up the L.A. Basin rank in the 99th percentile of U.S. counties for
VMT (U.S. Environmental Protection Agency, 1993). Taken together, these counties
contain a population of over 14.5 million persons. This implies that several hundred
thousand persons could be exposed to PM4 Mn levels exceeding 0.1 jtg/m , by the above
projections.
B-46
-------
Both meteorological conditions and high VMT are important factors in the L.A. Basin
PM10 pollution levels. To the extent that other metropolitan areas of the United States have
high VMT and/or comparable meteorological conditions, the projected exposure estimates for
the L.A. area may be relevant. In addition, the presence of point sources of Mn in other
locales would contribute to higher overall Mn PM exposure levels. Although no quantitative
comparison of the L.A. estimates to other areas of the United States is possible with the data
available, approximately 19 U.S. counties (excluding the other three counties in the L.A.
Basin) have higher VMT levels than Riverside, which has approximately 9 billion miles per
year (U.S. Environmental Protection Agency, 1993). In the southwest United States,
Phoenix has average wind speed and calm conditions similar to the Los Angeles area, and
Maricopa County, AZ (of which Phoenix is the county seat), has VMT levels approximately
double those of Riverside. Cook County (Chicago), EL, has a VMT of approximately
36 billion miles per year; Harris County (Houston), TX, has approximately 27 billion miles
per year. The combined population of the 19 counties with greater VMT levels than
Riverside (apart from the Los Angeles area) is roughly 15 million. Based on the ORD
projection estimates for Riverside, possibly several hundred thousands of persons altogether
3
could be exposed to PM4 Mn levels exceeding 0.1 /*g/m if MMT were used in unleaded
gasoline as proposed by Ethyl Corporation.
5. SUMMARY AND CONCLUSIONS
Of the available data pertaining to personal exposure levels of Mn in relation to MMT
usage in gasoline, the PTEAM data provide the most appropriate basis for estimating the
distribution of Mn exposure levels in a general population. Uncertainties and assumptions
are involved in any attempt to project future Mn exposure levels from past data, and thus -
higher or lower estimates are possible. Therefore, the distributions in Figure B-15 should
not be interpreted as upper and lower bound estimates. However, based on the limited data
available, two reasonable estimates of the projected distribution of PM4 Mn personal
exposure levels that might result from the use of MMT in unleaded gasoline as proposed by
Ethyl Corporation are depicted as lines 1 and 2 in Figure B-15. This attempt to estimate
B-47
-------
possible Mn exposure levels based on limited data should not in any way be considered an
adequate substitute for empirical data from properly designed and conducted personal
monitoring studies, as previously discussed (U.S. Environmental Protection Agency, 1991).
B-48
-------
REFERENCES
MRS, Aerometric Information Retrieval System [database]. (1994) [TSP and PM,n 1990 data on manganese from
various sites in the U.S. as of 3/22/94]. Research Triangle Park, NC: U.S. Environmental Protection
Agency, Office of Air Quality Planning and Standards.
Appel, B. R.; Wesolowski, J. J.; Aleocer, A.; Wall, S.; Twiss, S.; Giauque, R,; Ragaini, R.; Ralston, R.
(1980) Quality assurance for the chemistry of the aerosol characterization experiment. In: Hidy, G. M.;
Mueller, P. K.; Grosjean, D.; Appel, B. R., Wesolowski, J. J., eds. The character and origins of smog
aerosols: a digest of results from the California Aerosol Characterization Experiment (ACHEX). New
York, NY: John Wiley & Sons, Inc.; pp. 69-105. (Advances in environmental science and technology:
v. 9).
Baldwin, M.; Belanger, S.; Gignac, N.; Mergler, D. (1992) Estimation du niveau d'exposition au manganese de
travailleurs d'une usine de production d'alliages de manganese [Determining the level of exposure to
manganese in workers in a manganese alloy factory]. In: Proceedings of the 14th congress of the
Industrial Hygiene Association at Quebec; Montreal, Quebec, Canada; pp. 7-25.
Bell, R. (1994) [Telephone conversation with Dr. David Mage, U.S. EPA, concerning cut point of sampler and
P&T corrections for Windsor, Ontario data in paper by Lusis et al., 1992]. Toronto, Ontario, Canada:
Ontario Ministry of the Environment, Air Resources Branch; February 28.
Clayton, C. A.; Perritt, R. L.; Pellizzari, E. D.; Thomas, K. W.; Whitmore, R. W.; Ozkaynak, H.; Spengler,
J. D.; Wallace, L. A. (1993) Particle toted exposure assessment methodology (PTEAM) study;
distributions of aerosol and elemental concentrations in personal, indoor, and outdoor air samples in a
southern California community. J. Exposure Anal. Environ. Epidemiol. 3: 227-2SO.
Cohanim, S.; Hoggan, M.; Sin, R,; Jones, L. W.; Hsu, M.; Tom, S. (1991) Summary of air quality in
California's south coast and southeast desert air basins 1987—1990. Diamond Bar, CA: South Coast Air
Quality Management District.
Cooper, J. A.; Redline, D. C.; Sherman, J. R.; Valdovinos, L. M.; Pollard, W. L.; Scavone, L. C.; Badgett-
West, C. (1987) PMjQ source composition library for the South Coast Air Basin; volume I, source
profile development documentation final report. El Monte, CA: South Coast Air Quality Management
District; July 15.
Dann, T. (1990) Results from inhalable particulate matter (PM10) sampling network (1984—1987). Ottawa,
Ontario, Canada: Environment Canada, Pollution Measurement Division; PMD no. 90-3.
Davis, B. L.; Johnson, L. R.; Stevens, R. K.; Courtney, W. J.; Safiiet, D. W. (1984) The quartz content and
elemental composition of aerosols from selected sites of the EPA inhalable particulate network. Atmos.
Environ. 18: 771-782.
Davis, D. W.; Hsiao, K.; Ingels, R.; SMkiya, J. (1988) Origins of manganese in air particulates in California.
JAPCA 38: 1152-1157.
Ethyl Corporation. (1994a) Comments of Ethyl Corporation in response to EPA's December 9, 1993 Federal
Register notice 58 Fed. Reg. 64761 (1993). Washington, DC: Ethyl Corporation, Office of the Vice
President for Government Relations. Available for inspection at: U.S. Environmental Protection Agency,
Central Docket Section, Washington, DC: docket no. A-93-26.
B-49
-------
Ithyl Corporation. (1994b) Re: Etbyl Corporation ("Ethyl") waiver application under §211(f) (4) of the Clean Air
Act. Washington, DC: Ethyl Corporation, Office of the Vice President for Government Relations.
Available for inspection at: U.S. Environmental Protection Agency, Central Docket Section, Washington,
DC; docket no. A-93-26.
Gagnon, C, (1994a) Manganese data from MVC sampling network June '92 [fax communication to Dr. David
Mage, U.S. EPA]. Montreal, Quebec, Canada: Communaute urbaine de Montreal, Service de
1'environment; February 25.
Gagnon, C. (1994b) [Pax communication to Dr. David Mage, U.S. EPA, with participate concentrations for June
1992 and January 1994]. Montreal, Quebec, Canada: Communaute urbaine de Montreal, Service de
1'environment; February 25.
Garland, J. A. (1978) Dry and wet removal of sulphur from the atmosphere. Atmos. Environ. 12: 349-362.
Gray, H. A.; Landry, B.; Liu, C. S.; Henry, R. C.; Cooper, J. A.; Sherman, J. R. (1988) Receptor modeling
for PM|0 source apportionment in die South Coast Air Basin of California. In: Mathai, C, V,;
Stonefield, D. H., eds. PM-10: implementation of standards. Pittsburgh, PA: Air Pollution Control
Association; pp. 399-418.
Gray, H. A.; Landry, B.; Liu, C, S.; Henry, R. C.; Cooper, J. A.; Sherman, J. R. (1989) Receptor modeling
for PMjQ source apportionment in the South Coast Air Basin of California: final air quality management
plan, 1989 revision. El Monte, CA: South Coast Air Quality Management District; final appendix V-K.
Hoggan, M. C. (1994) [Letter to Dr. David Mage, U.S. EPA, with summary statistics for manganese for
Rubidoux from August 1985—December 1986], Diamond Bar, CA: South Coast Air Quality Management
District; March 22.
Hoggan, M. C.; Banayan, S. C.; Sin, R. W.; Davidson, A. (1993) An analysis of PM,0 air quality in
California's South Coast Air Basin, 1985-1992. In: Criteria air pollutants: volume 3B, papers from the
86th annual meeting & exhibition of the Air & Waste Management Association; June ; Denver, CO.
Pittsburgh, PA: Air & Waste Management Association; paper no. 93-WA-75.04. (A&WMA publication
no. AM93-3).
IP Network, Inhalable Particulate Network [data base]. (1984) [Printout of manganese data for 1979-1983].
Research Triangle Park, NC: U.S. Environmental Protection Agency, Environmental Monitoring Systems
Laboratory.
Loranger, S.; Zayed, J. (1994) Manganese and lead concentrations in ambient air and emission rates from
unleaded and leaded gasoline between 1981 and 1992 in Canada: a comparative study. Atmos. Environ.:
in press.
Loranger, S.; Zayed, J.; Forget, E. (1994) Manganese contamination in Montreal in relation with traffic density.
Water Air Soil PolluL 74: 1-12.
Lnsis, M. A,; Bell, R. W.; Chapman, R. E.; Kruschel, B. D.; Spencer, M. J.; Smith, K. V. (1992) Personal
exposure pilot studies in Toronto and Windsor, Ontario. Presented at: 9th world congress on atmospheric
chemistry and environment; August; Montreal, Quebec, Canada; paper no. IU-21A.02.
Lynam, D. R.; Pfeifer, G. D.; Ter Haar, G. L.; Fort, B. F. (1994) Exposure to manganese-in-air in an urban
environment. Presented at: International conference of the Society for Environmental Goechemistry and
Health; July 1993; New Orleans, LA. Proc. Annu. Meet. Soc. Environ. Geochem. Health: in press.
B-50
-------
Lyons, J. M. (1994) [Fax coiranunication to Dr. David Mage, U.S. EPA, concerning error in reporting Mn
concentration in Lyons, et al., 1993, Atmos. Environ, 27B, 237-249]. Sacramento, CA: Sierra Research;
April 18.
Lyons, J. M.; Venkataraman, C,; Main, H. H.; Friedlander, S. K. (1993) Size distributions of trace metals in
the Los Angeles atmosphere. Atmos. Environ. Part B 27: 237-249.
Paquet, M. (1994) [Fax communication to Dr. David Mage, U.S. EPA, with information on the ferromanganese
plant at Beauhamois]. Sainte-Foy, Quebec, Canada: Ministers de L'Environment; February 28.
Pellizzari, E. D.; Spengler, J. D.; Ozkaynak, H. (1990) Nine home particle TEAM study. Research Triangle
Park, NC: Research Triangle Institute; RTI report no. R1T/166/00-01F; EPA contract no. 68-02-4544.
Pellizzari, E. D.; Thomas, K. W.; Clayton, C. A.; Whitmore, R. W.; Shores, R. C.; Zelon, H. S.; Perritt,
R. L. (1992) Particle total exposure assessment methodology (PTEAM): Riverside, California pilot study,
volume I [final report]. Research Triangle Park, NC: U.S. Environmental Protection Agency,
Atmospheric Research and Exposure Assessment Laboratory; EPA report no. EPA/600/R-93/050.
Available from: NTIS, Springfield, VA; PB93-166957/XAB.
Pellizzari, E. D.; Thomas, K. W.; Clayton, C. A.; Whitmore, R. C.; Shores, H.; Zelon, S.; Peritt, R. L.
(1993) Particle total exposure assessment methodology (PTEAM): Riverside, California pilot
study—volume I [project summary]. Research Triangle Park, NC: U.S. Environmental Protection
Agency, Atmospheric Research and Exposure Assessment Laboratory; EPA report
no. EPA/600/SR-93/050.
Radell, R. (1994) 1991 Mn & TSP data—Toronto [fax communication to Dr. David Mage, U.S. EPA].
Etobicoke, Ontario, Canada: Ontario Ministry of Environment and Energy, Environmental Monitoring
and Reporting Branch, Air Quality and Meteorology Section; March 4.
Reels, H. (1993) [Letter to Dr. M. Davis, U.S. EPA, on definitions of "respirable," "total," and "inhalable"
dusts]. Bruxelles, Belgium: Universite Catfaolique de Lou vain, Unite de Toxicologie Industrielle et
Medecine du Travail; October 19.
Roels, H. A.; Ghyselen, P.; Buchet, J. P.; Ceulemans, E.; Lauwerys, R. R. (1992) Assessment of the
permissible exposure level to manganese in workers exposed to manganese dioxide dust. Br. J. Ind. Med.
49: 25-34.
Thomas, K. W.; Pellizzari, E. D.; Clayton, C. A.; Whitaker, D. A.; Shores, R. C.; Spengler, J. D.; Ozkaynak,
H.; Wallace, L. A. (1993) Particle total exposure assessment methodology (PTEAM) study: method
performance and data quality for personal, indoor, and outdoor aerosol monitoring at 178 homes in
southern California. J. Exposure Anal. Environ. Epidemiol. 3: 203-226. ,
U.S. Environmental Protection Agency. (1991) Information needed to improve the risk characterization of
manganese tetraoxide (Mn3O4) and methylcyclopentadienyl manganese tricarbonyl (MMT). Washington,
DC: Office of Research and Development; December 12.
U.S. Environmental Protection Agency. (1993) Regional interim emission inventories (1987-1991),
v. I: development methodologies. Research Triangle Park, NC; Office of Air Quality Planning and
Standards; EPA report no. EPA-454/R-93-021a. Available from: NTIS, Springfield, VA;
PB93-236107/XAB.
B-51
-------
Veinot, E. (1994) [Fax communication to Dr. David Mage, U.S. EPA, with monthly meteorological summaries
for January 1994 from Toronto's Pearson and Montreal's Dorval international airports]. Downsview,
Ontario, Canada: Environment Canada, Atmospheric Environment Service, Canadian Climate Center;
March 4.
Wallace, L. A.; Duan, N.; Ziegenfus, R, (1994) Can long-term exposure distributions be predicted from short-
term measurements? Risk Anal. 14: 7S-8S.
World Health Organization. (1987) Manganese. In: Air quality guidelines for Europe. Copenhagen, Denmark:
Regional Office for Europe; pp. 262-271. (WHO regional publications, European series no. 23).
Zayed, J. (1993) [Personal communication with S. Durkee, U.S. EPA, concerning raw data of personal sampling
results from 1992 Montreal study]. Montreal, Quebec, Canada: Universite de Montreal, Department de
medecine du travail et d'hygiene du milieu; October 15.
Zayed, J.; Germ, M.; Loranger, S.; Sierra, P.; Begin, D.; Kennedy, G. (1994) Occupational and environmental
exposure of garage workers and taxi drivers to airborne manganese arising from the use of
methyleyclopentadienyl manganese tricarbonyl in unleaded gasoline. Am. tad, Hyg. Assoc. J. 55: 53-58.
B-52
-------
ATTACHMENT B-l. PTEAM Mn DATA COLLECTED BY PM10
PERSONAL SAMPLERS DURING DAYTIME HOURS
Data are weighted according to sampling statistics.
Rank
0.715
1.466
2.023
3.728
4.662
5.858
7.563
8.274
8.958
9.950
10.987
11.787
12.599
13.337
14.272
15.153
15.989
16.937
17,948
19.654
20.483
21.258
22.195
22.910
23.639
24.731
25.530
26.568
27.505
28.702
29.478
30.321
ID
102A
155A
024A
166A
179A
002A
095A
025A
050A
092A
152A
078A
139A
034A
149 A
100A
098A
I ISA
004A
051 A
014A
016A
032A
183 A
109A
131A
089A
046A
135A
028A
041A
146A
Mn (ng/m )
4.200
4.950
6.150
9.900
11.000
12.000
12.700
14.200
14.534
14.900
16.400
16.900
17.100
17.500
18.100
18.300
18.700
19.200
19.250
19.263
19.300
19.700
19.900
20.400
20.600
21.400
22,000
22.100
22.200
22.400
22.800
23.300
23.400
24.000
Percentile
0.00416
0,00853
0.01176
0.02167
0.02710
0,03406
0.04397
0.04811
0,05
0.05208
0,05785
0.06388
0.06853
0.07325
0.07754
0.08298
0.08810
0.092%
0.09847
0.10
0.10435
0.11426
0.11909
0,12359
0.12904
0,13320
0.13744
0.14378
0.14843
0.15447
0.15991
0.16687
0.17138
0.17629
B-53
-------
ATTACHMENT B-l (cont'd). PTEAM Mn DATA COLLECTED BY PM10
PERSONAL SAMPLERS DURING DAYTIME HOURS
Data are weighted according to sampling statistics.
Rank
31.036
31.880
32.986
33.989
34.605
35.399
36.137
37.245
38.836
39.552
40.364
41.599
42.247
43.773
44.553
45.382
47.087
47.813
48.370
48.930
49.768
50.606
51.166
52.109
52.908
53.745
54.582
55.757
57.069
57.784
58.628
59.460
ID
065A
111A
003A
053A
119A
005A
162A
031A
156 A
091A
172A
096A
027A
062A
103A
013A
144 A
090A
132A
153A
043A
015A
113A
110A
047A
108A
033A
018A
143A
159A
181A
134A
Mn (ng/m )
24.100
24.500
25.700
25.800
25.867
25.900
26.000
26.500
26.700
27.300
27.500
27.700
27.800
27.900
27.900
29.100
29.300
29.600
29.900
30.000
30.500
30.700
30.900
30.900
31.269
31.700
33.000
33.000
33.200
33.300
33.400
33.800
34.200
35.000
Percentile
0.18044
0.18535
0.19178
0.19761
0.20
0.20119
0.20581
0.21010
0.21654
0.22579
0.22995
0.23467
0.24185
0.25449
0.25903
0.26385
0.27376
0.27798
0.28122
0.28448
0.28935
0.29747
0.30
0.30296
0.31247
0.31734
0.32417
0.33180
0.33596
0.34086
0.34570
B-54
-------
ATTACHMENT B-l (cont'd). PTEAM Mn DATA COLLECTED BY PM10
PERSONAL SAMPLERS DURING DAYTIME HOURS
Data are weighted according to sampling statistics.
Rank
60.489
61.466
62.177
63.542
64.110
64.875
65.982
66.770
68.230
68.974
70.434
71.411
72.555
73.342
74.106
76.185
77.023
77.958
78.961
79.956
80.616
81.411
82.087
83.125
83.869
84.653
85.309
86.881
87.697
88.423
89.149
89.887
ID
136A
060A
117A
055A
123 A
176A
079A
115A
039A
045A
019A
(MSA
141A
167 A
142 A
061 A
012A
097A
104A
107A
122A
038A
085A
023A
006A
007A
064A
099A
168A
082A
058A
151A
3
Mn (ng/m )
36.400
37.000
37.000
37.700
37.900
38.400
38.600
38.600
38.900
39.130
39.200
39.400
39.500
39.500
41.350
41.400
42.800
44.200
44.200
44.400
44.800
44.900
45.800
46.000
46.200
46.350
47.000
47.400
47.776
48.700
48.700
49.800
50.600
51.900
Pereentile
0.35168
0.36149
0.36943
0.37273
0.37718
0.38820
0.39669
0.40
0.40101
0.40950
0.41518
0.42183
0.42641
0.43085
0.44294
0.44781
0.45324
0.45908
0.46486
0.46870
0.47332
0.47725
0.48329
0.48761
0.49217
0.49598
0.50
0.50986
0.51409
0.51831
0.52260
B-55
-------
ATTACHMENT B-l (cont'd). PTEAM Mn DATA COLLECTED BY PM,0
PERSONAL SAMPLERS DURING DAYTIME HOURS
Data are weighted according to sampling statistics.
Rank
90.531
93.232
93.990
94.822
95.644
96.952
97.704
98.364
100.942
101,897
102.912
104.991
105.828
106.572
107.310
108.347
109.532
110.669
11L278
112.446
113.231
114.841
115.879
116.658
117.453
118.290
119.328
119.903
120.779
122.615
123.254
124.050
ID
116A
164A
037A
077A
075A
035A
093A
145A
084A
008A
127A
160A
120A
044A
125A
163A
036A
106A
137A
069A
052A
126A
101A
083A
138A
001A
022A
112A
054A
154A
129A
105A
Mn (ng/m )
53.000
53.800
54.700
57.200
57.400
58.400
59,100
59.400
59.700
60.400
60.900
61.163
62.800
64.000
64.500
65,000
65.200
65.800
66.000
66.700
68.500
69.800
71.400
71.800
72.400
74.100
76.000
76.900
78.100
78.192
78.600
78.600
79.100
79.700
Percentile
0.52634
0.54216
0.54645
0.55129
0.55607
0.56367
0.56805
0.57189
0.58687
0.59243
0.59833
0.60
0.61041
0.61528
0.61960
0.62390
0.62992
0.63681
0.64343
0.64697
0.65376
0.65832
0.66768
0.67372
0.67825
0.68286
0.68773
0.69377
0.69711
0.70
0.71288
0.71659
0.72122
B-56
-------
ATTACHMENT B-l (cont'd). PTEAM Mn DATA COLLECTED BY PM10
PERSONAL SAMPLERS DURING DAYTIME HOURS
Data are weighted according to sampling statistics.
Rank
125.755
126.410
127.008
127.772
128.428
129.003
129.954
130.897
132.234
133.378
134.191
135.098
135.910
136.748
137.885
139.054
139.798
140.626
141.746
142.681
144.210
145.353
146.151
146.938
149.181
150.025
152.161
153.199
154.732
156.624
157.362
158.377
ID
059A
086A
180A
178A
020A
124A
067A
121A
161A
074A
171A
021A
133A
Oil A
066A
128A
080A
026A
040A
174A
175 A
076A
170A
130A
068A
169A
030A
184 A
158A
148A
057A
182 A
3
Mn (ng/m )
83.000
84.200
85.100
87.400
93.300
96.100
97.600
97.600
98.300
98.600
99.050
100.400
101.800
104.300
104.525
104.600
109.800
110.400
113.300
114,200
114.300
115.200
116.500
118.000
120.300
128.600
129.800
130.800
133.900
138.000
138.021
138.600
139.800
141.150
Percentile
0.73114
0.73494
0.73842
0.74286
0.74667
0.75002
0.76103
0.76880
0.77546
0.78018
0.78545
0.79018
0.79505
0.80
0.80166
0.80845
0.81278
0.81759
0.82410
0.82954
0.83843
0.84507
0.84972
0.85429
0.86733
0.87224
0.88466
0.89069
0.89961
0.90
0.91060
0.91490
0.92080
B-57
-------
ATTACHMENT B-l (cont'd). PTEAM Mn DATA COLLECTED BY FM10
PERSONAL SAMPLERS DURING DAYTIME HOURS
Data are weighted according to sampling statistics.
Rank
160.005
160.769
161.782
163J14
165.546
166.419
167.224
167.880
168.680
169.547
171.000
N
Mean
Standard deviation
Geometric mean
ID
073A
049A
071 A
140A
094A
087A
I57A
088A
042A
017A
114A
= 171
= 69.0887
= 64.8165
= 50.2227
Mn (ng/m )
141.500
143.300
163.600
171.471
173.000
193.900
220.600
221.700
275.600
310.541
316.700
393.400
421.377
448.850
Percentile
0.93026
0.93470
0.94059
0.95
0.95183
0.96248
0.96755
0.97223
0.97605
0.98
0.98070
0.98574
0.99
0.99419
Geometric standard deviation = 2.2185
B-58
-------
ATTACHMENT B-2. PTEAM Mn DATA COLLECTED BY PM10
PERSONAL SAMPLERS DURING NIGHTTIME HOURS
Data are weighted according to sampling statistics.
Rank
1.511
3.102
3.905
4.886
5.607
6.633
7.794
8.579
9.323
11.008
11.932
12.958
13.881
15.567
16.369
17.590
18.140
19.166
19.895
20.886
22.571
23.920
24.5%
25.386
26.386
27.213
27.763
28.946
30.388
30.942
31.697
32.484
33.667
34.592
36,165
37.314
38.339
39.124
41.179
42.864
43.513
ID
175A
126A
172A
092A
109A
023A
018A
005A
093A
059A
097A
101A
179A
166A
17IA
096A
132A
022A
151A
053A
095A
055A
050A
047A
004A
033A
024A
002A
019A
153A
176A
105A
028A
149A
156A
056A
163A
038A
160A
051A
088A
Mn
2.700
3.000
3.600
3.750
3.900
4.050
4.200
4.264
4.350
4.350
4.500
4.500
4.500
4.950
5.100
5.100
5.190
5.400
5.400
5.600
8.500
9.500
9.600
9.900
10.100
10.500
10.700
10.700
10.900
11.400
11.500
11.700
11.700
11.800
11.850
11.853
11.900
11,900
12.000
12.200
12.500
13.200
13.300
13.400
Percentile
0.00894
0.01836
0.02311
0.02891
0.03318
0.03925
0.04612
0.05
0.05516
0.07667
0.08214
0.09686
0.10
0.10734
0.11341
0.11772
0.12359
0.13356
0.14154
0.14554
0.15022
0.16103
0.16428
0.17128
0.17981
0.18756
0.19221
0.19922
0.20
0.21399
0.22079
0.22686
0.23151
0.24366
0.25364
0.25747
B-59
-------
ATTACHMENT B-2 (cont'd). PTEAM Mn
PERSONAL SAMPLERS DURING
Data are weighted according to
DATA COLLECTED BY PM10
NIGHTTIME HOURS
sampling statistics.
Rank
45.533
46.361
47.657
48.736
49.861
50.826
51.495
52.224
53.014
53.781
54.873
55.576
57.185
58.870
59.577
60.410
61.058
62.041
62.595
63.414
64.061
65.192
65.922
67.894
68.684
69.387
70.176
70.931
71.758
72.465
73.491
74.354
75.198
76.032
76,810
77.539
79.093
79.800
80.409
81.333
ID
009A
015A
143A
131A
106A
048A
085A
162A
078A
016A
003A
025A
073A
144A
065A
111A
020A
107A
113A
013A
086A
141A
125A
029A
042A
117A
170A
178A
098A
183 A
184A
087A
177 A
146A
167A
034A
099A
102A
119A
174A
Mn
13.600
14.000
14.200
14.300
14.900
15.161
15.200
15.200
15.200
15.300
15.500
15.600
15.600
15,600
15.600
16.100
16.100
16.WO
16.200
16.300
16.400
16.700
17.000
17.300
17.334
17.400
17.400
17.400
17.400
17.400
17.700
18.100
18.100
18.400
18.900
18.950
19.300
19.500
19.500
19.700
19.700
19.850
Percentile
0.26942
0,27432
0.28200
0,28838
0.29503
0.30
0.30902
0.31369
0.31823
0.34834
0.35746
0.36711
0.37038
0.37523
0.37906
0.38575
0.39007
0.40
0.41971
0.42460
0.43486
0.43996
0.44496
0.44989
0.45450
0.46801
0.47579
0.48126
B-60
-------
ATTACHMENT B-2 (cont'd). PTEAM Mn
PERSONAL SAMPLERS DURING
Data are weighted according to
DATA COLLECTED BY PM10
NIGHTTIME HOURS
sampling statistics.
Rank
82.104
82,895
83,624
84.331
85.097
85.659
86.465
87.636
88.372
89.397
90.528
91.130
93.677
95.120
95.916
98.605
99.390
101.205
101.796
102.774
103.492
104.312
105.082
105.901
106.926
108,362
109.137
109.854
110.858
111.728
112.434
113.950
114.762
115.498
116.498
117.490
ID
103A
089A
037A
091A
041A
123A
168A
036A
044A
046A
074A
137A
084A
039A
157A
164A
138A
154A
180A
173A
090A
014A
083A
026A
152A
114A
007A
082A
127A
100A
159A
158A
075A
045A
071A
104A
Mn
19.900
20.150
20.200
20.300
20.376
20.900
20.900
21.100
21.200
21.500
21,600
21.600
21.800
22.100
22.500
22.600
22.600
22.900
23.100
23.100
23.500
23.700
23.800
23.900
24.700
24.800
24.900
25.300
25.400
25,600
26.000
26.000
26.100
26,300
26.400
26.600
26.600
Percentile
0.48582
0.49050
0.49482
0.49900
0.50
0.50686
0.51163
0.51856
0.52291
0.52898
0.53567
0.53923
0.55430
0.56284
0.58346
0.58811
0.60
0.60234
0.60813
0.61238
0.61723
0.62179
0.62663
0.63270
0.64120
0.64578
0.65003
0.65596
0.66529
0.67426
0.67906
0.68342
0.69521
B-61
-------
ATTACHMENT B-2 (cont'd). FTEAM Mn DATA COLLECTED BY PM
PERSONAL SAMPLERS DURING
Data are weighted according to
NIGHTTIME HOURS
sampling statistics.
10
Rank
117.490
118.584
119.327
120.422
121,248
122.181
123.008
124.516
125.373
126.399
127.051
127.983
128.759
129.489
131.399
133.510
134.079
134.912
135.735
136.513
137.316
138.150
138.781
139.560
140.576
141.473
142.301
143.241
144.232
144.873
145.799
146.626
147.492
148.436
149.172
151.041
152.363
153.165
153.988
155.142
156.249
ID
104A
031 A
155A
079A
108A
110A
012A
062A
017A
070A
145A
121A
052A
057A
140A
030A
112A
1S1A
134A
130A
133A
169A
129 A
115A
136A
021A
043A
067A
010A
027A
135A
120A
054A
008A
080A
148A
161A
139A
077A
069A
040A
Mn
26.600
26.909
27.300
27.300
28.300
28.400
28.800
28.900
29.500
30.500
30.600
31.200
31.900
32.200
32.600
32.700
32.900
33.000
33.200
33.305
33.500
35,300
35.300
35.400
35.900
36.700
38.200
38.400
38.700
40.500
41.100
41.300
41.700
42.200
43.400
44.200
44.900
46.800
48.642
49.100
49.700
51.200
52.900
54.000
Percentile
0.69521
0.70
0.70608
0.71255
0.71745
0.722%
0.72786
0.73678
0.74185
0.74792
0.75178
0.75730
0.76189
0.76620
0.77751
0.79000
0.79337
0.79830
0.80
0.80317
0.81252
0.81745
0,82119
0.82580
0.83181
0.83712
0.84202
0.84758
0.85345
0.85724
0.86272
0.86761
0.87273
0.87832
0.88267
0.89374
0.90
0.90156
0.90630
0.91117
0.91800
0.92455
B-62
-------
ATTACHMENT B-2 (cont'd). PTEAM Mn DATA COLLECTED BY PM10
PERSONAL SAMPLERS DURING NIGHTTIME HOURS
Data are weighted according to sampling statistics.
N
Mean
Standard
Rank
157,004
157.641
158.209
158.944
160.237
162.454
163.609
165.419
166.422
167.347
168.000
deviation
Geometric mean
ID
142A
116A
124A
006A
035A
068A
128A
094A
182 A
032A
122A
= 168
= 24.2351
= 16.6313
= 19.1028
Mn
54.300
54.900
55.500
56.600
57.700
58.074
60.350
65.400
71.700
71.760
72.000
93,014
93.900
98.400
Percentile
0.92902
0.93278
0.93615
0.94050
0.94815
0.95
0.96127
0.96810
0.97881
0.98
0.98475
0.99
0.99022
0.99408
Geometric standard deviation = 2.0830
B-63
-------
ATTACHMENT B-3. PTEAM Mn DATA COLLECTED BY EM10 PERSONAL SAMPLERS DURING DAILY PERIODS
COVERING BOTH DAYTIME AND NIGHTTIME HOURS
Data are weighted according to sampling statistics.
w
4
Daytime
Rank
1.706
2.641
3.634
4.191
4.906
6.613
7.810
ID
001A
009A
010A
QUA
029A
049A
056A
OS8A
060A
061A
064A
066A
070A
076A
118A
173A
177 A
166A
179 A
092A
024A
102A
095A
002A
Mna
76,00
24.90
96.00
104.30
36.70
143.30
85.50
50.60
37.00
42.80
47.40
104.60
42.10
116.50
21.70
23.50
18.90
9.90
11.00
16.40
6.15
4.20
9.30
12.00
Duplicate 1
Duration
475
713
709
706
693
563
801
559
638
606
594
591
562
502
652
568
702
655
674
498
486
718
680
592
Flag"
1
5
5
1
5
1
5
1
I
1
1
1
5
1
1
1
1
1
1
1
1
1
1
1
Duplicate 2
Mn" Duration Flag* Mna
4 . « .
13.60
41.10
4.65
17.40
. . .
12.00
.
...
*
. . ,
. * .
30.60
17.10
16,8 652 1 19.40
...
*
5.10
4.95
3.75
10.90
19.70
16.1 680 1 9.60
11.40
Duplicate
Duration
,
530
655
656
575
.
610
,
.
.
,
712
774
560
«.
626
588
809
754
553
702
624
Nighttime
1
Hagb
1
I
5
1
,
I
.
,
4
1
5
5
1
1
1
1
1
1
1
Duplicate 2
Mna Duration Flag* Average* Percentile
» » * « «
» « . * .
» . .
.
wit . .
" * * * 1
. * . . 4
, » . .
* * * . «
. » .
* . . . »
... . .
... . *
« » * . .
... . ,
... . ,
7.554 0.01053
8.181 0.01630
8.570 0.02243
9.038 0.02587
10.944 0,03029
11.125 0.04082
11.692 0.04821
11.767 0.05
-------
ATTACHMENT B-3 (cont'd). PTEAM Mn DATA COLLECTED BY PM10 PERSONAL SAMPLERS DURING DAILY
PERIODS COVERING BOTH DAYTIME AND NIGHTTIME HOURS
Data are weighted according to sampling statistics.
Daytime
Duplicate 1
Rank
8.494
9.224
10.036
10.832
11.768
12.780
13.491
15.197
15.997
W 17.195
SS 18.199
18.975
19.726
20.902
22.138
22.695
23.787
25.380
26.217
26.955
27.671
28.231
29.337
ID
050A
109A
I72A
005A
149A
004A
025A
051A
078A
028A
053A
016A
155A
018A
096A
132A
131A
156A
098A
034A
183A
153A
003A
Mna
14
22
27
26
18
19
14
19
17
23
25
20
4.
33
27
30
22
27
19
18
21
30
.90
.00
.70
.00
.30
.30
.20
,70
.10
.30
.80
.40
95
.30
.80
.00
.10
.30
.20
.10
.40
.50
25.70
Duration
445
600
514
620
592
631
584
572
596
606
601
483
564
593
649
591
580
526
505
688
519
539
552
Flagb
1
1
1
3
1
1
1
1
1
1
3
1
3
1
1
3
6
Duplicate 2
Duplicate
Mntt Duration Flag Mn" Duration
10
3
3
4
11
10
15
13
15
12
9
15
27
4
5
5
14
11
17
19
17
11
15
.10
.90
.60
.35
.90
.70
.60
.30
.30
.00
.50
.50
.30
.20
.40
.40
.30
.90
.70
.50
.70
.70
.60
763
715
793
678
670
643
764
798
683
717
675
789
710
712
548
586
708
773
708
570
796
734
774
Nighttime
1 Duplicate 2
Flag0 Mn" Duration Flag™
1 ...
1 ...
1 ...
1 ...
1 ...
1 ...
I ...
1 ...
1 ...
1 H.7 717 1
1 ...
1 ...
1 ...
1 ...
1 ...
1 ...
1 ...
1 ...
I ...
1 ...
1 18.5 796 1
1 ...
1 ...
Average
11.868
12.159
13.078
14.691
14.902
14.959
14.993
15.972
16,139
16.301
17.095
17.177
17.361
17.406
17.423
17.545
17.752
17.812
18.136
18.324
18,734
19.402
19.660
19.805
Percentile
0.05243
0.05694
0.06195
0.06686
0.07264
0.07889
0.08328
0.09381
0.09875
0.10
0.10614
0.11234
0.11713
0.12177
0.12902
0.13665
0.14009
0.14683
0.15666
0.16183
0.16639
0.17081
0.17426
0.18109
-------
ATTACHMENT B-3 (eont'd). FTEAM Mn DATA COLLECTED BY PMIO PERSONAL SAMPLERS DURING DAILY
PERIODS COVERING BOTH DAYTIME AND NIGHTTIME HOURS
Data are weighted according to sampling statistics.
Rank
AM
PM
Duplicate 1
Duplicate 2
Duplicate 1
Duplicate 2
ID Mn Duratioa Flag
Mn Duration Flag Mn Duration Flag
Mn Duration Flag
Average Pereentile
30.076
31.114
31.958
37.648
38.686
39.525
40.406
41.236
42.274
43.118
43.894
44.454
45.767
46.383
47.099
48.465
49.245
50.706
51.471
52.407
53.145
54.122
54.849
162A
152 A
111A
013A
046A
015A
100A
014A
023A
146A
041A
113A
143A
119A
091A
055A
103A
019A
176 A
097A
151A
048A
090A
26.50
18.00
24.50
29.30
22.40
30.90
18.70
19,90
46.20
24.00
23.40
30.90
33.40
25.90
27.50
37,70
29.10
39.40
38.40
44.20
51.90
39,50
29.90
573
842
587
511
512
599
630
650
535
890
699
500
562
736
560
626
621
594
576
760
500
575
600
1
1
1
1
1
1
1
6
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
15.20
15.8 842 1 24.80
16.10
16.40
21.60
14.00
26.00
23.80
4.05
18.70
20.90
16.30
14.20
19.70
20.30
9.90
19.90
11.50
11.70
4.50
8.50
13.80
23.70
791
533
680
781
796
672
560
880
707
470
575
718
734
694
698
652
715
682
582
642
714
651
680
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
I
1
1
19.947
19.962
19.992
20.340
21.502
21.913
21.965
22.135
22.143
22.206
19.2 470 1 22.255
22.272
22.293
22.526
22.891
23.505
23.517
23.632
24.176
24.488
24.981
26.021
26.375
16.6 651 1 26.597
26.606
0.18565
0.19206
0.19727
0.20
0.23239
0.23880
0.24398
0.24942
0.25454
0.26095
0.26616
0.27095
0.27441
0.28251
0.28632
0.29073
0.29916
0.30
0.30398
0.31300
0.31772
0.32350
0.32806
0.33409
0.33857
3
-------
ATTACHMENT B-3 (cont'd). PTEAM Mn DATA COLLECTED BY PM,0 PERSONAL SAMPLERS DURING DAILY
PERIODS COVERING BOTH DAYTIME AND NIGHTTIME HOURS
Data are weighted according to sampling statistics.
AM
PM
Duplicate 1
Duplicate 2
Duplicate 1
Duplicate 2
Rank ID Mn Duration Flag"
Mn Duration Flag
Mn Duration Flag
Mn8 Duration Flag" Average Pereentile
55.561
56.706
57.814
63.203
64.664
65.341
66.178
67.122
67.875
69.448
70.556
71.301
72.145
73.149
73.982
75.020
77.100
77.939
78,751
79.689
80.827
81.665
82.449
117A
141A
031A
167A
039A
085A
108A
110A
093A
099A
079A
045A
181A
104A
134A
022A
I60A
043A
139A
135A
106A
012A
007A
37.00
39.50
26.70
42.80
38.90
46.00
33.00
31.70
59.10
48.70
38.60
39.20
34.20
44.40
35.00
76.90
62.80
30.70
17.50
22.70
66.00
44.20
47.00
651
564
521
601
598
608
509
717
650
539
565
708 (
690
491
622
554
521
655
631
505
507
502
572
I ... 17.40
1 ... 17.00
[ ... 27.30
[ 39.9 601 I 19.30
I ... 22.50
I ... 15.20
I ... 28.40
1 ... 28.80
I ... 4.35
[ ... 19.50
I ... 28.30
> ... 26.40
33.20
26.60
33.50
3.60
13.20
38.70
49.70
22.9 505 1 41.70
14.90
I ... 28.90
I ... 25.30
713 1
710 1
706 I
683 1
751 1
675 1
688 1
557 I
709 I
731 1
799 1
647 1
620 1
696 1
779 1
811 1
690 1
614 1
721 1
860 1
795 I
715 1
625 1
26,755
26.961
27.045
29.621
29.770
29.775
29.796
30.356
30.432
30.537
31.893
32.566
33.088
33.727
33.963
34.166
7.6 811 I 34.538
34.539
34.571
34,672
34.708
34.798
34.883
35.211
35.670
0.34297
0.35004
0.35688
0.39014
0.39916
0.40
0.40334
0.40851
0.41433
0.41898
0.42869
0.43553
0.44013
0.44534
0.45153
0.45668
0.46309
0.47593
0.48110
0.48612
0.49191
0.49893
0.50
0.5041 1
0.50895
-------
w
OS
oo
ATTACHMENT B-3 (cont'd). FTEAM Mn DATA COLLECTED BY PM10 PERSONAL SAMPLERS DURING DAILY
PERIODS COVERING BOTH DAYTIME AND NIGHTTIME HOURS
Data are weighted according to sampling statistics.
Rank
AM
PM
Duplicate 1
Duplicate 2
Duplicate 1
Duplicate 2
ID MiT Duration Flag
Mn Duration Flag Mn Duration Flag
Mn" Duration Flag" Average Percentile
83.098
83.825
85.436
88.159
88.975
90.013
92.081
92.869
93.691
94.430
95.174
97.754
98.492
100.198
101.384
102.399
103.060
103.669
104.466
105.245
107.083
107.739
108.504
109. 103
027A
082A
126A
164A
168A
163A
136A
USA
075A
037A
044A
084A
125 A
059A
036A
127A
145 A
137A
105A
083A
154 A
020A
142 A
180A
27.90
49.80
71.40
53.80
48.70
65.20
36.40
38.60
59.40
54.70
64.50
59,70
65.00
83.00
65.80
60.90
59.40
66.70
79.70
72.40
78.60
93.30
41.40
85.10
564
565
574
535
715
637
622
716
501
720
517
625
650
571
588
585
526
655
595
566
529
502
563
541
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
41
25
3.
22
21
12
38
36
55.4 501 1 26
20
21
.30
.40
00
.60
.10
.20
.20
.70
.30
.20
.50
22.10
17.30
4.
50
21.20
25.60
31.20
21.80
11.80
23.90
23.10
16.20
54.30
23.10
794
752
606
670
564
721
705
551
799
561
697
35
35
36
36
36
37
37
37
38
.735
.868
.273
,452
.529
.061
.356
.774
.285
39.591
39
40
657 1 ... 40
653
646
704
694
706
710
675
713
767
700
732
758
41
41
41
41
43
.812
.298
.431
.095
.331
.498
.746
.240
43.345
43
45
45
48
48
48
,611
.363
.754
.400
.692
.921
0.
0.
0.
0.
0.
0.
0.
0.
51295
51744
52738
54419
54923
55564
56840
57327
0.57834
0.
0.
58290
58750
0.60
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
60342
60798
61851
62583
63209
63617
63993
64485
64966
66101
66506
66978
67347
-------
w
ATTACHMENT B-3 (cont'd). PTEAM Mn DATA COLLECTED BY PM10 PERSONAL SAMPLERS DURING DAILY
PERIODS COVERING BOTH DAYTIME AND NIGHTTIME HOURS
Data are weighted according to sampling statistics.
AM
Duplicate
Rank
109.758
110,543
111.338
112.083
113.039
116.757
117.633
118.570
119.383
120.913
122.222
122.797
123.966
124.795
125.940
126.753
127.697
128.496
129.157
130.065
131.016
132.354
132.930
ID
086A
052A
138A
006A
008A
077A
054A
032A
171A
175A
035A
112A
069A
026A
074A
133A
121A
170A
122A
021A
067A
161A
124A
Mn" Duration
84.20
69.80
74.70
50.00
60.40
57.20
78.60
20.60
90,
20
115.20
58.
78.
68.
113.
98.
101.
97.
118.
44.
100.
97.
98.
96.
40
10
50
30
60
80
60
00
90
40
60
30
10
591
618
652
562
683
598
504
655
705
645
599
903
652
541
629
511
709
641
722
647
623
572
518
1
Flagb
1
3
3
3
1
3
1
1
1
1
1
1
1
1
1
1
1
1
1
6
1
3
1
Duplicate 2
PM
Duplicate 1 Duplicate 2
Mn* Duration Flagb Mn° Duration Flag" Mn* Duration Flag" Average"
16
32
73.5 652 3 22
42.7 562 3 56
44
51
42
93
107.9 705 1 5
2
57
33
52
24
21
35
31
17
98
38
40
49
55
.70
.20
.90
.60
.20
.20
.10
.90
.10
.70
.70
.00
.90
.70
.60
.30
.90
.40
.40
.40
.50
.10
.50
624 1 ... 49.533
709 1 ... 49.711
584 1 ... 49.908
692 1 ... 52.006
651 1 . . . 52.494
52.675
653 1 ... 54.068
797 1 44.7 797 1 57.036
659 1 ... 57.362
560 1 ... 57.459
668 1 ... 57.965
701 1 . . . 58.023
625
642
788
578
762
802
720
503
719
642
633
714
59.653
60.760
60.767
61.727
61.994
62.728
64.780
66.868
67.306
67.766
68.621
72.455
72.570
Percentile
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
.67752
.68237
.68727
.69187
.69777
.70
.72072
,72613
.73192
.73693
.74638
.75446
,75801
.76522
,77034
.77741
.78243
.78825
.79319
.79726
.80
.80287
.80874
.81700
.82055
-------
ATTACHMENT B-3 (eont'd), PTEAM Mn DATA COLLECTED BY PM10 PERSONAL SAMPLERS DURING DAILY
PERIODS COVERING BOTH DAYTIME AND NIGHTTIME HOURS
Data are weighted according to sampling statistics.
AM
Duplicate
Rank
133
134
135
136
137
138
140
142
142
144
146
147
148
150
151
153
154
155
157
157
158
,866
.611
.649
.493
.613
.401
.539
.073
.812
.441
.334
.502
.516
.760
.566
.500
.515
,389
.222
.878
.678
ID
174A
080A
184A
169A
040A
130A
030A
158A
057A
073A
148 A
128 A
071A
068A
157 A
1 40 A
182 A
087A
094A
OSSA
042A
Mntt Duration
114
110
.30
.40
133.90
129
114
120
130
138
139
141
138
.80
.20
.30
.80
.00
.80
.50
.60
109.80
163
122
221
173
138
220
193
275
316
.60
.80
.70
.00
.10
.60
.90
.60
.70
755
570
636
536
500
630
620
608
652
749
562
700
621
658
512
621
683
715
528
723
652
1
Flag1'
Duplicate 2
PM
Duplicate 1
Mn" Duration Flag" Mn* Duration Flag"
21
44
.30
.90
18.10
35
.40
54.00
1
1
1
1
1
1
1
1
1
3
1
3
1
1
3
1
35
32
26
32
15
46
.30
.90
.10
.60
.60
.80
65.40
26
134.4 658 1 58
22
.60
.10
,60
32.70
144.2 683 3 72.00
18.40
71.70
13.40
17.40
564
686
631
690
799
640
741
666 1
701 1
573 1
709 1
680 1
652 1
634 1
794 1
573 1
582
653
682
637
654
Duplicate 2
Mna Duration Flagb Average*
18.4 564 1 73.913
74.625
76.228
76.671
77.172
77.465
77.498
79.503
84.259
86.931
87.261
87.391
87.922
93.432
62.6 634 1 95.109
100.655
105,670
107.115
109.336
124.082
125.024
152.790
166.821
169.779
Percentile
0.82633
0.83093
0.83734
0.84255
0.84947
0.85433
0.86752
0.87700
0.88156
0.89161
0.90
0.90329
0.91051
0.91676
0.93062
0.93559
0.94753
0.95
0.95380
0,95919
0.97050
0.97456
0.97950
0.98
-------
ATTACHMENT B-3 (cont'd). PTEAM Mn DATA COLLECTED BY FM,0 PERSONAL SAMPLERS DURING DAILY
PERIODS COVERING BOTH DAYTIME AND NIGHTTIME HOURS
Data are weighted according to sampling statistics.
AM
Rank
160.133
161,000
ID Mn"
I14A 441.60
017A 393.40
Duplicate ]
Duration
588
703
I
Flagb
1
1
Duplicate 2
Mn"
456.1
Duration
588
r-i "
Flag
1
Mna
24.90
30.50
Duplicate
Duration
693
560
PM
1
Flagb
1
6
Duplicate 2
Mn" Duration Flagb Average"
219.500
223.207
232,494
Percentile
0.98847
0.99
0.99383
a , 3
ng/m
Flag designates validity of data: 1 = usable, no problems; 3 = questionable, comments noted; 5 - not usable, originally questionable; 6 = usable, originally
questionable
N =161
Mean = 45.6415 (±5,4, 95% confidence interval)
Standard deviation = 35.5933
i Geometric mean = 36.1667
**"*
""* Geometric standard deviation = 1,9568
-------
ATTACHMENT B-4. PROJECTED MMT USAGE FACTOR
To estimate the projected exposure levels of Mn associated with MMT usage in
gasoline, a base situation with a set of defined parameters is compared to another situation
with different values of the same parameters. In this case, the base situation refers to
Riverside, CA, in the fall of 1990, when the PTEAM personal exposure study (Pellizzari
et al., 1992) was conducted. Because MMT was used in leaded gasoline in California prior
to and during the period of the PTEAM study, it provides a basis for extrapolating to a
future scenario involving MMT usage in unleaded gasoline.
Leaded gasoline constituted approximately 13,6% of the total gasoline sold in
Los Angeles in 1989 and 1990 (Lundberg Survey, Inc., 1991). According to 1989 and 1990
production data for leaded gasoline for 11 Los Angeles area refineries reported to EPA
pursuant to lead phasedown regulations (confidential business information provided under
40 CFR 80.20) and Mn concentration data collected by the California Air Resources Board
(1994), leaded gasoline contained MMT at concentrations of 0.0003 to 0,0726 g Mn/gal, for
a production-weighted average concentration of 0.049 g Mn/gal in 1989 and 0.048 g Mn/gal
in 1990.
If 100% of the gasoline contained MMT, automotive Mn emissions would increase by
a factor of 7.4 (i.e., 100/13.6). However, because the leaded gasoline produced in the
Los Angeles area in 1989-1990 had an average MMT concentration of 0.048 g Mn/gal,
automotive Mn emissions would be proportionately lower if gasoline contained
0.03125 g Mn/gal (1/32 g Mn/gal), as proposed by Ethyl Corporation in a waiver petition to
EPA. In this respect, Mn emissions would be lower by a factor of 0.65 (i.e.,
0.03125/0.048). Based on these factors, automotive Mn emissions would be predicted to
increase by a factor of 4.8 (i.e., 7.4 X 0.65) from the situation that existed in the
Los Angeles area (including Riverside) in 1990 (i.e., 13.6% of the gasoline containing
0.048 g Mn/gal) to a scenario that assumes 100% of the unleaded gasoline contains MMT at
the concentration sought by Ethyl Corporation (0.03125 g Mn/gal).
Two additional factors come into consideration in attempting to estimate the increase
in Mn emissions related to MMT usage m comparing Riverside in 1990 to a future scenario
that assumes 100% of unleaded gasoline contains MMT at 0.03125 g Mn/gal. One factor
B-72
-------
concerns the level of gasoline usage; the other concerns the Mn emission rate (grams Mn
emitted per gram Mn in fuel combusted) for noncatalyst and catalyst vehicles.
According to projections from the U.S. Department of Energy (1993), gasoline usage
was expected to increase at an average annual rate of approximately 1 % per year between
1990 and 1995 (the nearest full year in the future). This estimate appears to be slightly
conservative, in view of actual supply and disposition data indicating an increase from
110.9 x 109 gallons for 1990 to 114.7 x 109 gallons for 1993 (U.S. Department of Energy,
1994), or an observed rate of increase of slightly more than 1 % per year. Adjusting the
factor of 4.8 for a 5% change in gasoline usage between 1990 and 1995 yields an
MMT-increased-usage factor estimate of 5.
The other factor involved in projecting between Riverside in 1990 to the 1995 scenario
in question concerns the Mn emission rates of catalyst-equipped and noncatalyst vehicles.
In 1990, only leaded gasoline in California was permitted to contain MMT. Therefore, only
noncatalyst vehicles would have been using leaded-MMT gasoline. In the 1995 scenario,
only catalyst vehicles using unleaded gasoline are assumed. The present extrapolation
assumes that the Mn emission rate for catalyst-equipped vehicles combusting unleaded
gasoline with MMT is the same as that for noncatalyst vehicles combusting leaded gasoline
with MMT. According to limited data obtained with noncatalyst and catalyst cars, Mn
emission rates vary widely for different vehicles and conditions. Data on inorganic Mn
emissions from noncatalyst cars combusting MMT-gasoline appear to be limited to three
studies of only a few vehicles (Ter Haar et al., 1974; Hum et al., 1974; Moran, 1975). The
tests were conducted over a limited mileage range using outmoded procedures and analytic
methods. Reported Mn emission rates (which cannot always be verified by inspection of the
reported data) ranged from 4 to 40%.
Data for Mn emissions from catalyst vehicles are also limited. Four recent studies
involving relatively small numbers of vehicles (a total of 24) indicate a wide range of Mn
emission rates when MMT is combusted under various conditions. Relevant conditions and
factors that may affect the emission rates of these tests include the type and condition of the
vehicle (e.g., newer, well-maintained versus older, poorly maintained cars), the operating
conditions (e.g., stop-and-go city driving with high acceleration rates versus highway driving
with relatively constant speeds; winter versus summer) for both the accumulation of mileage
B-73
-------
and the testing period, and the mileage of the vehicle in relation to MMT-gasoline use. The
last factor may be especially important if Mn particles deposit and accumulate in the engine
and exhaust system over time.
The number of conditions and variables is nearly as great as the number of tests in
these studies, thereby making it impossible to determine statistically or by casual inspection
what a representative rate of Mn paniculate emission would be for catalyst vehicles over
their expected lifetimes. However, of the studies conducted thus far, a Mn mass balance
study reported by Ethyl Corporation (1991) appears to have been conducted under conditions
more nearly representative of "real world" driving than any other available study. Although
more than two-thirds of the mileage was accumulated under highway conditions, which
would tend to produce lower emission rates, this study indicated an average emission rate of
approximately 27% for three new 1991 Chevy S-10 pickup trucks with 4.3-L V-6 engines
and 1991 emissions control technology.
Results obtained in other studies range from as low as approximately 4% to as high as
approximately 45%, However, a simple arithmetic average of these data does not provide a
meaningful representative emissions rate. Although a rate of approximately 30%, based in
large part on the results of the Ethyl Corporation mass balance study, may be a reasonable
estimate for catalyst vehicles operating under more-or-less real world conditions, the data are
inadequate to support this estimate with any confidence. Therefore, given the broad
comparability but basic inadequacy of the limited available emission rate data for noncatalyst
and catalyst vehicles, no specific values can be estimated as representative emission rates for
the two types of vehicles and the rates are assumed to be equal. No further adjustment of
the MMT-increased-usage factor of 5 is needed.
REFERENCES
California Air Resources Board. (1994) California Air Resources Board gasoline test data for lead and
manganese content of leaded gasoline for Los Angeles area refineries for 1989 & 1990. Sacramento,
CA: California Air Resources Board; March 28.
Code of Federal Regulations. (1991) Regulation of fuels and fuel additives; controls applicable to gasoline
refiners and importers. C. F. R. 40: §80.20.
B-74
-------
Ethyl Corporation, (1991) Re: Application for a fuel additive waiver under §211(F)(4) of the Clean Air Act:
appendices to the waiver application for the HiTec 3000 performance additive, v. 1, appendix 9.
Washington, DC: Ethyl Corporation.
Hum, R. W.; Allsup, J.R.; Cox, F. (1974) Effect of gasoline additives on gaseous emissions. Report prepared
for Office of Research and Development, U.S. Environmental Protection Agency, Washington, DC,
EPA report no. EPA-650/2-75-014. Available from: NTIS, Springfield, VA; PB-253782.
Lundberg Survey, Inc. (1991) U.S. refinery gasoline grade sales report by state, 1989-1990. North Hollywood,
CA: Lundberg Survey, Inc.
Moran, J. B. (1975) Hie environmental implications of manganese as an alternate antiknock. Presented at:
Automobile engineering meeting; October; Detroit, MI. Warrendale, PA: Society of Automotive
Engineers, Inc.; SAE technical paper no. 750926.
Pellizzari, E. D.; Thomas, K. W.; Clayton, C. A.; Whitmore, R. W.; Shores, R. C.; Zelon, H. S.; Perritt,
R. L. (1992) Particle total exposure assessment methodology (PTEAM): Riverside, California pilot
study, volume I [final report]. Research Triangle Park, NC: U.S. Environmental Protection Agency,
Atmospheric Research and Exposure Assessment Laboratory; EPA report no. EPA/600/R-93/050.
Available from: NTIS, Springfield, VA; PB93-166957/XAB.
Ter Haar, G. L.; Griffing, M. E.; Brandt, M.; Oberding, D. G.; Kapron, M. (1974) Methylcyclopentadienyl
manganese tricarbonyl as an antiknock: composition and fate of manganese exhaust products. Presented
at: 67th annual meeting of the Air Pollution Control Association; Denver, CO. Pittsburgh, PA: Air
Pollution Control Association; paper no. 74-199.
U.S. Department of Energy. (1993) Annual energy outlook 1993 with projections to 2010. Washington, DC:
Energy Information Administration; report no. DOE/ELA-0383(93). Available from: NTIS, Springfield,
VA; DE93-005983.
U.S. Department of Energy, (1994) Monthly energy review: May 1994. Washington, DC: Energy Information
Administration; p. 55; report no. DOE/EIA-0035(94/05),
B-75
-------
ATTACHMENT B-5. PUBLICATIONS FROM THE PTEAM STUDY
9-HOME PREPILOT STUDY
Peer-Reviewed Journal Articles
Clayton, C. A.; Pellizzari, E. D.; Wiener, R. W. (1991) Use of a pilot study for designing a
large-scale probability study of personal exposure to aerosols. J. Exposure Anal.
Environ. Epidemiol. 1: 407-422.
Project Reports
Pellizzari, E. D.; Spengler, J. D.; Ozkaynak, H. (1990) Nine-home particle TEAM study.
Research Triangle Park, NC: Research Triangle Institute; RTI report no.
RTI/166/00-01F; EPA contract no. 68-02-4544.
Conference Proceedings
Anderson, R.; Kamens, R.; Rodes, C. (1989) A collocation study of PM-10 and PM-2.5
inertia! impactors for indoor aerosol exposure assessment. In: Measurement of toxic
and related air pollutants: proceedings of the 1989 U.S. EPA/A&WMA international
symposium. Pittsburgh, PA: Air & Waste Management Association; pp. 464-469.
(A&WMA publication VIP Network-13).
Spengler, J. D.; Ozkaynak, H.; Lubwig, J.; Allen, G.; Pellizzari, E. D.; Wiener, R. (1989)
Personal exposures to paniculate matter: instruments and methodologies of PTEAM.
In: Measurement of toxic and related air pollutants: proceedings of the 1989 U.S.
EPA/A&WMA international symposium. Pittsburgh, PA: Air & Waste Management
Association; pp. 449-463. (A&WMA publication VIP Network-13).
Wiener, R. W. (1988) Measurement and evaluation of personal exposure to aerosols.
In: Measurement of toxic and related air pollutants: proceedings of the 1988
EPA/APCA international symposium; May; Research Triangle Park, NC. Pittsburgh,
PA: Air Pollution Control Association; pp. 84-88. (APCA publication no. VIP
Network-10).
Wiener, R. W. (1989) Particle total exposure assessment methodology—an overivew of
planning and accomplishments. In: Measurement of toxic and related air pollutants:
proceedings of the 1989 U.S. EPA/A&WMA international symposium. Pittsburgh,
PA: Air & Waste Management Association; pp. 442-448. (A&WMA publication VIP
Network-13).
B-76
-------
Wiener, R. W.; Wallace, L.; Pahl, D.; PeUizzari, E.; Whittaker, D.; Spengler, J.;
Ozkaynak, H. (1990) Review of the particle TEAM 9 home field study.
In; Measurement of toxic and related air pollutants: proceedings of the 1990
EPA/A&WMA international symposium, v. 2; May; Raleigh, NC, Pittsburgh, PA:
Air and Waste Management Association; pp. 452-460, (A&WMA publication VIP
Network-17).
178-HOME PILOT STUDY
Peer-Reviewed Journal Articles
Clayton, C. A,; Perritt, R. L.; Pellizzari, E. D.; Thomas, K. W,; Whitmore, R. W.;
Ozkaynak, H.; Spengler, J. D.; Wallace, L. A. (1993) Particle total exposure
assessment methodology (PTEAM) study: distributions of aerosol and elemental
concentrations in personal, indoor, and outdoor air samples in a southern California
community. J. Exposure Anal. Environ. Epidemiol. 3: 227-250.
Thomas, K. W.; Pellizzari, E. D.; Clayton, C. A.; Whitaker, D. A.; Shores, R. C.;
Spengler, J. D.; Ozkaynak, H.; Wallace, L. A. (1993) Particle total exposure
assessment methodology (PTEAM) study: method performance and data quality for
personal, indoor, and outdoor aerosol monitoring at 17S homes in southern California.
J. Exposure Anal. Environ. Epidemiol. 3: 203-226.
Project Reports
Mamane, Y. (1992) Characterization of PTEAM indoor aerosol samples (electron
microscopy analysis). Research Triangle Park, NC: U.S. Environmental Protection
Agency, Atmospheric Research and Exposure Assessment Laboratory.
Ozkaynak, H.; Xue, J.; Weker, R.; Butler, D.; Spengler, J. (1993) The particle TEAM
(PTEAM) study: analysis of the data, volume in [draft final report]. Research
Triangle Park, NC: U.S. Environmental Protection Agency; EPA contract
no. 68-02-4544.
Pellizzari, E. D.; Thomas, K. W.; Clayton, C. A.; Whitmore, R. W.; Shores, R. C.; Zelon,
H. S.; Perritt, R. L. (1992) Particle total exposure assessment methodology
(PTEAM): Riverside, California pilot study, volume I [final report]. Research
Triangle Park, NC: U.S. Environmental Protection Agency, Atmospheric Research
and Exposure Assessment Laboratory; EPA report no. EPA/600/R-93/050. Available
from: NTIS, Springfield, VA; PB93-166957/XAB.
B-77
-------
Pellizzari, E. D.; Thomas, K. W.; Clayton, C. A.; Whitmore, R, C.; Shores, H.; Zelon, S.;
Peritt, R. L. (1993) Particle total exposure assessment methodology (PTEAM):
Riverside, California pilot study—volume I [project summery]. Research Triangle
Park, NC: U.S. Environmental Protection Agency, Atmospheric Research and
Exposure Assessment Laboratory; EPA report no. EPA/600/SR-93/Q5Q.
Pellizzari, E. D.; Spengler, J. D. (1990) Particle total exposure assessment methodology
(PTEAM) pilot study: volume I, study design [work plan]. Research Triangle Park,
NC: U.S. Environmental Protection Agency; EPA contract no. 68-02-4544; EPA work
assignment 67.
Pellizzari, E. D.; Spengler, J. D. (1990) Particle total exposure assessment methodology
(PTEAM) pilot study: volume n, protocols for environmental sampling and analysis
[workplan]. Research Triangle Park, NC: U.S. Environmental Protection Agency;
EPA contract no. 68-02-4544; EPA work plan 67.
Pellizzari, E. D.; Spengler, J. D. (1990) Particle total exposure assessment methodology
(PTEAM) pilot study: volume HI, quality assurance project plan [work plan].
Research Triangle Park, NC: U.S. Environmental Protection Agency; EPA contract
no. 68-02-4544; EPA work assignment 67.
Sheldon, L.; Clayton, A.; Keever, J. Perritt, R. L.; WWtaker, D. (1992) PTEAM:
monitoring of phthalates and PAHs in indoor and outdoor air samples in Riverside,
California, volume n [final report]. Sacramento, CA: California State Air Resources
Board. Available from: NTIS, Springfield, VA; PB93-205649/XAB.
Conference Proceedings
Jenkins, P. L.; Hui, S. P.; Phillips, T. J.; Lum, S. B. (1992) Toxic air pollutants in
California residences. In: Edwards, L., ed. Current issues in air toxics: proceedings of
the third annual west coast regional Air and Waste Management Association
conference; November. Pittsburgh, PA: Air & Waste Management Association.
Ozkaynak, H.; Spengler, J. D.; Ludwig, J. F.; Butler, D. A.; Clayton, C. A.; Pellizzari,
E.; Wiener, R. W. (1990) Personal exposure to paniculate matter: findings from the
Particle Total Exposure Assessment Methodology (PTEAM) prepilot study. In: Indoor
air '90: precedings of the 5th international conference on indoor air quality and
climate, volume 2, characteristics of indoor air; July-August; Toronto, ON, Canada.
Ottawa, ON, Canada: International Conference on Indoor Air Quality and Climate,
Inc.; pp. 571-576.
B-78
-------
Ozkaynak, H.; Spengler, J. D.; Xue, J,; Koutrakis, P.; Pellizzari, E, D.; Wallace, L. (1993)
Sources and factors influencing personal and indoor exposures to particles, elements
and nicotine: findings from the particle TEAM pilot study. In: Jantunen, M.;
Kalliokoski, P.; Kukkonen, E.; Saarela, K.; Seppanen, O.; Vuorelma, H., eds. Indoor
air '93: proceedings of the 6th international conference on indoor air quality and
climate, v. 3, combustion products, risk assessment, policies; July; Helsinki, Finland.
Helsinki, Finland: Indoor Air '93; pp. 457-462.
Perritt, R. L.; Clayton, C. A.; Pellizzari, E. D.; Thomas, K. W.; Wallace, L. A.; Spengler,
J. D.; Ozkaynak, H. (1991) Particle Total Exposure Assessment Methodology
(PTEAM) Pilot Study: personal, indoor, and outdoor paniculate concentration
distributions for southern California fall 1990—preliminary results. In: Measurement
of toxic and related air pollutants: proceedings of the 1991 U.S. EPA/A&WMA
international symposium, v. 2; May; Durham, NC. Pittsburgh, PA: Air & Waste
Management Association; pp. 665-671. (A&WMA publication VBP Network-21).
Sheldon, L.; Whitaker, D.; Keever, J.; Clayton, A.; Perritt, R. (1993) Phthalates and PAHs
in indoor and outdoor air in a southern California community. In: Jantunen, M.;
Kalliokoski, P.; Kukkonen, E.; Saarela, K.; Seppanen, O.; Vuorelma, H., eds. Indoor
air '93: proceedings of the 6th international conference on indoor air quality and
climate, v. 3, combustion products, risk assessment, policies; July; Helsinki, Finland.
Helsinki, Finland: Indoor Air '93; pp. 109-114.
Wallace, L.; Clayton, A.; Whitmore, R.; Shores, R.; Thomas, K.; Whitaker, D.; Reading,
P.; Pellizzari, E.; Spengler, J.; Ozkaynak, H.; Froehlich, S.; Jenkins, P.; Ota, L.;
Westerdahl, D. (1991) Initial results from the PTEAM study: survey design,
population response rates, monitor performance and quality control. In: Measurement
of toxic and related air pollutants: proceedings of the 1991 U.S. EPA/A&WMA
international symposium, v. 2; May; Durham, NC. Pittsburgh, PA: Air & Waste
Management Association; pp. 659-664. (A&WMA publication VJJP Network-21).
Wallace, L. A.; Pellizzari, E.; Sheldon, L.; Whitmore, R.; Zelon, H.; Clayton, A.; Shores,
R.; Thomas, K.; Whitaker, D.; Reading, P.; Spengler, J.; Ozkaynak, H.; Froehlich,
S.; Jenkins, P.; Ota, L.; Westerdahl, D. (1991) The TEAM study of inhalable
particles (PM10): study design, sampler performance, and preliminary results.
Presented at: 84th annual meeting of the Air & Waste Management Association; June;
Vancouver, British Columbia, Canada. Pittsburgh, PA: Air & Waste Management
Association; paper no. 91-171.3.
Wallace, L. A.; Gzkaynak, H.; Spengler, J. D.; Pellizzari, E. D.; Jenkins, P. (1993)
Indoor, outdoor, and personal air exposures to particles, elements, and nicotine for
178 southern California residents. In: Jantunen, M.; Kalliokoski, P.; Kukkonen, E.;
Saarela, K.; Seppanen, O.; Vuorelma, H., eds. Indoor air '93: proceedings of the 6th
international conference on indoor air quality and climate, v. 3, combustion products,
risk assessment, poEcies; July; Helsinki, Finland. Helsinki, Finland: Indoor Air '93;
pp. 445-450.
B-79
-------
ATTACHMENT B-6. GLOSSARY OF SELECTED TERMS RELATING TO
Mn EXPOSURE ASSESSMENT
ADT. Average daily traffic.
Aerodynamic Diameter of a Particle. The diameter of a unit density sphere with the
identical drag coefficient (settling velocity) as the particle. All terms in this report and
glossary refer to the aerodynamic diameter of the particle.
AUC. Area under curve.
CARB. California Air Resources Board.
Coarse Mode Particles. These particles, defined by EPA as >2.5 ^m aerodynamic
diameter, are usually solid and of natural origin (e.g., dust devils, etc.). Wind-blown dust
has a "lognormal type" distribution, extending up to 100 /*m, usually with a mode in the
2.5 to 10 /*m range.
Cut Point. A filtration particle sampler has an efficiency that decreases from nearly 100%
for particles of the order of the pore size of the filter paper, depending on the air velocity
through the pores, to 0% for the largest particles with a settling velocity greater than the
vertical flow velocity into the inlet. The inlet cut point, defined as the size with a 50%
capture probability, is a function of the flow rate, and to a lesser extent, the air motion (and
direction for an asymmetrical inlet).
EC. Environment Canada,
line Mode Particles. These particles are usually of anthropogenic origin, either as emitted
from combustion sources (motor vehicles, etc.) or from chemical conversion of gaseous
species to aerosol droplets. The fine mode as defined by EPA is in the <2.5 pm range.
IP Network. Inhalable Particle Monitoring Network.
B-80
-------
MMT. Methylcyclopentadienyl manganese tricarbonyl.
MOB. Ontario Ministry of Environment.
MUC. Montreal Urban Commission.
HEM. Personal exposure monitor.
Penetration Curve. A plot of the capture efficiency (%) versus aerodynamic diameter (jim).
Although two different types of PM samplers may have identical cut points (e.g., 50% at
5 jim), they may have different penetration curves. This can be a complication in
interpretation and comparison of results in this report, because two such samplers will have a
bias between them that will vary with the particle size distribution,
PM|0. The mass fraction collected by a sampler designed with a penetration curve giving a
50% cut point for particles of 10 microns (jim) aerodynamic diameter. For example, if two
particles of 10 fan approach the sampler, one would be captured and one would escape.
. Same as for PM10 only with a 5 pm cut point.
PM2-5. Same as for PM10 only with a 2.5 pm cut point.
PM3 s. Same as for PM10 only with a 3.5 nm cut point.
PRD. Paved road dust, which may contain some automotive Mn.
PTEAM. Particle Total Exposure Assessment Methodology Study.
SAM. Stationary ambient monitor.
SCAQMD. South Coast Air Quality Management District,
B-81
-------
SIM. Stationary indoor monitor.
SMSA. Standard Metropolitan Statistical Area.
TSP. Total suspended paniculate matter, which includes both solid and liquid aerosol
particles. The high volume sampler for TSP has a cut point of 27 /tm at 50 cfm.
VMT. Vehicle miles traveled.
B-82
-------
ATTACHMENT B-7: PILOT PTEAM Mn DATA COLLECTED BY PM,0
PERSONAL SAMPLERS FOR 24 h ON NONCONSECUTTVE DAYS*
Person ID
11
12
21
22
31
32
41
42
51
52
62
72
81
82
91
92
Day
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
1
2
PM10
0.0258
0.0327
0.0281
0.0406
0.0284
0.0623
0.01935
0.02135
0.0291
0.0512
0.04505
0.0379
0.0355
0.0358
0.04475
0.0334
0.0191
0.03215
0.0397
0.027
0.03805
0.02445
0.0747
0.02925
0.04325
0.0206
0.0438
0.04445
0.03035
0.0887
0.03505
0.02145
Mean: 0.037
Variance of log PM10
0.028
0.068
0.308
0.005
0.159
0.015
0.000
0.043
0.136
0.074
0.098
0.440
0.275
0.000
0.575
0.121
0.1465
"These data were reproduced from a computer file, for which summary statistics are reported in Pellizzari,
E. D.; Spengler, J. D-; Ozkaynak, H. (1990) Nine home particle TEAM study. Research Triangle Park, NC:
Research Triangle Institute; report no. RTI/166/00-0IF; EPA contract no. 68-02-4544.
B-83
-------
ATTACHMENT B-8. MMT INCREASED USAGE FACTOR FOR 1985/86 TO 1990
To estimate the increase in MMT usage between 1985/86 (the period that was the
subject of the report of Davis et al, [1988]) and the last quarter of 1990 (the period of the
PTBAM Riverside study by Pellizzari et al. [1992]), the amounts of Mn in California leaded
gasoline during the two periods may be compared. According to Davis et al. (1988), the
average quarterly tonnage of Mn added to gasoline in all of California between July 1985 and
June 1986 was 17.5 tons. Based on confidential business information concerning leaded
gasoline production volumes for the Los Angeles area reported to EPA pursuant to lead
phasedown regulations (confidential business information provided under 40 CFR 80,20) and
information on the Mn concentrations in leaded-gasoline samples from 11 Los Angeles area
refineries in 1990 (CARB, 1994), the tonnage of Mn used in Los Angeles area leaded
gasoline in 1990 is estimated as 50 tons. Coupled with information from the Lundberg
Survey, Inc. (1991), the Mn tonnage for all of California in the last quarter of 1990 is then
estimated as 20.3 tons. Thus, the MMT increased usage factor for the period from 1985/86
to 1990 is estimated as 20.3/17.5 = 1.16.
REFERENCES
California Air Resources Board. (1994) California Air Resources Board gasoline test data for lead and
manganese content of leaded gasoline for Los Angeles area refineries for 1989 & 1990. Sacramento,
CA: California Air Resources Board; March 28.
Code of Federal Regulations. (1991) Regulation of fuels and fuel additives; controls applicable to gasoline
refiners and importers. C. F. R. 40: §80.20.
Davis, D. W.; Hsiao, K.; Ingels, R.; Shikiya, J. (1988) Origins of manganese in air particulates in California.
JAPCA38: 1152-1157.
Lundberg Survey, Inc. (1991) U.S. refinery gasoline grade sales report by state, 1989-1990. North
Hollywoood, CA: Lundberg Survey, Inc.
Pellizzari, E. D.; Thomas, K. W.; Clayton, C. A.; Whitmore, R. W.; Shores, R. C.; Zelon, H. S.; Perritt,
R. L. (1992) Particle total exposure assessment methodology (PTEAM): Riverside, California pilot
study, volume I [final report]. Research Triangle Park, NC: U.S. Environmental Protection Agency,
Atmospheric Research and Exposure Assessment Laboratory; EPA report no. EPA/600/R-93/050.
Available from: NTIS, Springfield, VA; PB93-166957/XAB.
B-84
-------
APPENDIX C
MANGANESE INHALATION REFERENCE
CONCENTRATION
c-i
-------
Manganese RfC-1
REFERENCE CONCENTRATION FOR CHRONIC INHMATION EXPOSURE (RfC) =
Substance Name:
CASRN:
Manganese
7439-96-5
The inhalation Reference Concentration (RfC) is analogous to the oral RfD and
is likewise based on the assumption that thresholds exist for certain toxic
effects such as cellular necrosis, but may not exist for other toxic effects
such as carcinogenicity. The inhalation RfC considers toxic effects for both
the respiratory system (portal-of-entry) and for effects peripheral to the
respiratory system (extrarespiratory effects). It is appropriately expressed
In units of mg/cu.m. In general, the RfC is an estimate (with uncertainty
spanning perhaps an order of magnitude) of a daily inhalation exposure of the
human population (including sensitive subgroups) that is likely to be without
an appreciable risk of deleterious effects during a lifetime. Inhalation RfCs
are derived according to the Interim Methods for Development of Inhalation
Reference Doses (EPA/600/8-88/066F August 1989) developed by U.S. EPA
scientists and peer-reviewed. For more information on the interim nature of
these methods and future plans see the IRIS News field. RfCs can also be
derived for the noncarcinogenic health effects of compounds which are
carcinogens. Therefore, it is essential to refer to other sources of
information concerning the carcinogenicity of this substance. If the U.S. EPA
has evaluated this substance for potential human carcinogenicity, a summary of
that evaluation will be contained in the Carcinogenicity Assessment Section of
this file when a review of that evaluation is completed.
RfC ASSESSMENT SUMMARY TABLE
Crit. Doses
Ws 1000 MFs
0.05 mg/cu.m
1 RfC:
[Study 1 LOAEI.(HEC)]
51-5 mg/cu.m Confidence: Medium
Crit Effects
Reported
ADJ
EEC
Scenario
Study Type
Reference
(1) Impairment of neurobehavioral function
(2) impairment of neurobehavioral function
-NOAEL (Study l)-r-LOAEL
None mg/cu.m ppm
None mg/cu.m
None mg/cu.m
particle f extrarespiratory
Occupational exposure to
manganese dioxide
Roels et al., 1992
(Study 1)-
0.15 mg/cu.m ppm
0.05 mg/cu.m
0.05 mg/cu.m
particle, extrarespiratory
Occupational exposure to
manganese dioxide
Roels et al., 1992
Other Refs: (2) Roels et al., 1987
1) Roels et al., 1992
Occupational exposure to manganese dioxide
Critical Effect:
Impairment of neurobehavioral function
-------
Manganese
RfC-2
REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
Defined Concentrations;
NOAEL=
NOAEL(ADJ}=
NOAEL(HEC)=
LOAEL=
LOAEL(ADJ)-
LOAEL(HEC)=
Scenario:
None
None
None
0.15 mg/cu.m
0.05 mg/cu.m
0.05 mg/cu.m
Conversion Factors:
The LOAEL refers to an extrarespiratory effect of
particulate exposure and is based on an 8-hour TWA
occupational exposure. MVho = 10 cu.m/day, MVh =
20 cu.m/day. LOAEL(HEC) =0.15 mg/cu.m x
(MVho/MVh) x 5 days/7 days = 0.05 mg/cu.m.
The LOAEL is derived from an occupational-lifetime
integrated respirable dust (IRD) concentration of
manganese dioxide (Mn02) (based on 8-hour TWA
occupational exposure multiplied by individual work
histories in years) expressed as mg manganese
(Mn)/cu.m x years. The IRD concentrations ranged
from 0.040 to 4.433 mg Mn/cu.m x years, with a
geometric mean of 0.793 mg Mn/cu.m x years and a
geometric standard deviation of 2.907. The
geometric mean concentration (0.793 mg/cu.m x
years) was divided by the average duration of Mn02
exposure (5.3 years) to obtain a LOAEL TWA of 0.15
mg/cu.m.
2) Roels et al., 1987
Occupational exposure to manganese oxides and salts
Critical Effects
Defined Concentrations:
Impairment of neurobehavioral function
NOAEL=
NOAEL(ADJ)=
NOAEL(HEC)=
LOAEL=
LOAEL (ADJ)-
LOAEL(HEC)=
Scenario:
None
None
None
0.97 mg/cu.m
0.34 mg/cu.m
0.34 mg/cu.m
Conversion Factors:
This is an extrarespiratory effect of a particulate
exposure. MVho = 10 cu.m/day, MVh = 20 cu.m/day.
LOAEL(HEC) =0.97 mg/cu.m x (MVho/MVh) x 5 days/7
days - 0.34 mg/cu.m.
The LOAEL is based on an 8-hour TWA occupational
exposure. The TWA of total airborne manganese dust
ranged from 0.07 to 8.61 mg/cu.m, and the median
was 0.97 mg/cu.m.
-------
Manganese RfC-3
——= REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC) =
DISCUSSION OF PRINCIPAL AND SUPPORTING STUDIES
Roelsf H.f R. Lauwerys, J.-P. BUG he t et al. 1987. Epideraiological survey
among workers exposed to manganese; Effects on lung, central nervous system,
.and some biological indices. Am. J. Ind. Med. 11: 307-327.
Jtoels, H.A., P. Ghyselen, J.p. BUG he t, E. Ceulemans, and R.R. Lauwerys. 1992.
^Assessment of the permissible exposure level to manganese in workers exposed to
zmanganese dioxide dust. Br. J. Ind. Med. 49s 25-34.
jtoels et al. (1992) conducted a cross-sectional study of 92 male workers
exposed to manganese dioxide (Mn02) dust in a Belgian alkaline battery plant. A
control group of 101 male workers was matched for age, height, weight, work
schedule, coffee and alcohol consumption, and smoking; educational level was
-slightly higher in the control group (p = 0.046 by chi square test).
3Hie manganese (Mn)-exposed group had been exposed to Mn02 for an average of 5.3
•years (range: 0.2-17.7 years). The geometric means of the workers' TWA
.airborne Mn concentrations, as determined by personal sampler monitoring at the
breathing zone, were 0.215 mg Mn/cu.m for respirable dust and 0.948 mg Mn/cu.m
if or total dust. No data on particle size or purity were presented, but the
aoedian cut point for the respirable dust fraction was 5 urn according to
.information provided by Roels et al. (1992) and Roels (1993). Total and
^respirable dust concentrations were highly correlated (r = 6.90, p < 0.001),
Tfith the Mn content of the respirable fraction representing on average 25% of
-the Mn content in the total dust. The authors noted that the personal
ononitoring data were representative of the usual exposure of the workers
because work practices had not changed during the last 15 years of the
reparation of the plant.
-Occupational-lifetime integrated exposure to Mn was estimated for each worker
±>y multiplying the current airborne Hn concentration for the worker's job
classification by the number of years for which that classification was held
.and adding the resulting (arithmetic) products for each job position a worker
diad held. The geometric mean occupational-lifetime integrated respirable dust
(IRD) concentration was 0.793 mg Mn/cu.m x years (range: 0.040-4.433 mg Mn/cu.m
3C years), with a geometric standard deviation of 2.907 mg Mn/cu.m x years,
Jaased on information provided by Roels (1993). The geometric mean
roccupational-lifetime integrated total dust (ITD) concentration was 3.505 mg
-Mn/cu.m x years (ranges 0.191-27.465 mg Mn/cu.m x years). Geometric mean
Concentrations of blood Mn (MnB) (0.81 ug/dL) and urinary Mn (MnU) (0.84 ug/g
^creatinine) were significantly higher in the Mn-exposed group than in the
control group, but on an individual basis no significant correlation was found
:±>etween either MnB or MnU and various external exposure parameters. Current
^respirable and total Mn dust concentrations correlated significantly with
geometric mean MnU on a group basis (Spearman r = 0.83, p < 0.05).
.A self-administered questionnaire focused on occupational and medical history,
^neurological complaints, and respiratory symptoms. Lung function was evaluated
•joy standard spirographic measures. Neurobehavioral function was evaluated by
rtests of audio-verbal short-term memory, visual simple reaction time, hand
steadiness, and eye-hand coordination. Blood samples were assayed for several
^hematological parameters (erythrocyte count, leukocyte count, hemoglobin,
-iiematocrit, mean corpuscular volume, mean corpuscular hemoglobin, platelets,
-------
Manganese RfC-4
=— REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
and differential leukocyte count); Mnj lead; zinc protoporphyrin; and serum
levels of calcium, iron, follicle stimulating hormone (FSH), luteinlzing
hormone (IiH), and prolactin. Urinary Mn, cadmium/ and mercury concentrations
were also determined.
Jtesponses to the questionnaire indicated no significant differences between
^groups in either respiratory or neurological symptoms, nor were spirometric,
Jiornonal, or calcium metabolism measurements significantly different for the
-two groups. In addition, a separate report (Gennart et al., 1992) indicated no
significant difference in the fertility of 70 of these workers, in contrast to
•earlier findings in 85 workers exposed not only to MnO2 but also to other Mn
Dxides and salts at higher concentrations (Lauwerys et al., 1985).
:Erythropoietic parameters and serum iron concentrations were consistently and
significantly lower in the Mn-exposed workers, albeit within the normal range
of values.
Of particular note, Mn workers performed worse than controls on several
^measures of neurobehavioral function. Visual reaction time was consistently
2nd significantly slower in the Mn-exposed workers measured in four 2-minute
periods, with more pronounced slowing over the total 8-minute period and
significantly greater variability in reaction times for the Mn-exposed group.
.Abnormal values for mean reaction times (defined as greater than or equal to
:the 95th percentile of the control group) also were significantly more
prevalent in the Mn-exposed group during three of four 2-minute intervals of
-the 8-minute testing period.
live measures of eye-hand coordination (precision, percent precision,
-imprecision, percent imprecision, and uncertainty) reflected more erratic
zontrol of fine hand-forearm movement in the Mn-exposed group than in the
controls, with mean scores on all five measures being highly significantly
.different for the two groups. There was also a significantly greater
^prevalence of abnormal values for these five measures in the Mn-exposed group.
ZPhe hole tremormeter test of hand steadiness indicated a consistently greater
amount of tremor in the Mn-exposed workers, with performance for two of the
dEive hole sizes showing statistically significant impairment.
Jtoels et al. (1992) performed an exposure-response analysis by classifying IRD
-values into three groups (<0.6, 0.6-1.2, and >1.2 mg Mn/cu.m x years) and
comparing the prevalence of abnormal scores for visual reaction time, hand
steadiness, and eye-hand coordination with controls. This analysis indicated
•±hat the prevalence of abnormal eye-hand coordination values was significantly
greater in workers whose IRD levels were less than 0.6 mg Mn/cu.m x years.
however, the relationship between exposure and response was not linear across
groups. Visual reaction time and hand steadiness showed linear
exposure-related trends but did not achieve statistical significance except at
-Levels of >1.2 mg Mn/cu.m x years. As noted by the authors, "analysis of the
jdata on a group basis ... does not permit us to identify a threshold effect
-level for airborne Mn." Although suggestive of a LOAEL of <0.6 mg Mn/cu.m x
•years, the exposure-response analysis by Roels et al. (1992) possibly could
^reflect the small disparity in educational level between exposed and control
-workers that was noted above with regard to the matching criteria for this
study. If educational level were in fact a covariate of exposure as well as
aaeurobehavioral performance, it could confound the exposure-response analysis.
.Although it is not clear that such was the case, the possibility of confounding
suggests that the LOAEL should not be based on the results of the
-------
Manganese RfC-5
—— REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
exposure-response analysis until these results can be confirmed by other
studies. Also, statistical correction for multiple comparisons should be
included in the exposure response analysis.
A LOAEL may be derived from the Roels et al. (1992) study by using the IRD
concentration of Mn02, expressed as mg Mn/cu.m x years (based on 8-hour TWA
occupational exposures for various job classifications, multiplied by
.individual work histories in years). Dividing the geometric mean IRD
concentration (0.793 mg/cu.m x years) by the average duration of the workers'
exposure to Mn02 (5.3 years) yields a LOAEL of 0.15 mg/cu.m. The LOAEL(HEC) is
0.05 mg/cu.m.
Roels et al. (198?) conducted a cross-sectional Roels et al. (1987) conducted a
cross-sectional study in 141 male workers
exposed to Mn02, manganese tetroxide (Mn304), and various Hn salts (sulfate,
carbonate, and nitrate). A matched group of 104 male workers was selected as a
control group. The two groups were matched for socioeconoraic status and
-background environmental factors; in addition, both groups had comparable
^work-load and work-shift characteristics.
The TWA of total airborne Mn dust ranged from 0.07 to 8.61 mg/cu.m, with an
overall arithmetic mean of 1.33 mg/cu.m, a median of 0.97 mg/cu.m, and a
geometric mean of 0.94 mg/cu.m. The duration of employment ranged from 1 to 19
years, with a mean of 7.1 years. The particle size and purity of the dust were
not reported. Neurological examination, neurobehavioral function tests (simple
reaction time, short-term memory, eye-hand coordination, and hand tremor),
spirographic measurements, blood and urine tests, and a self-administered
questionnaire were used to assess possible toxic effects of Mn exposure. The
questionnaire was designed to detect CNS and respiratory symptoms.
Significant differences in mean scores between Mn-exposed and reference
subjects were found for objective measures of visual reaction time, eye-hand
coordination, hand steadiness, and audio-verbal short-term memory. The
prevalence of abnormal scores on eye-hand coordination and hand steadiness
tests showed a dose-response relationship with blood Mn levels; short-term
memory scores were related to years of Hn exposure but not to blood Mn levels.
The prevalence of subjective symptoms was greater in the exposed group than in
controls for 20 of 25 items on the questionnaire, with four items being
statistically significant: fatigue, tinnitus, trembling of fingers, and
irritability.
A significantly greater prevalence of coughs during the cold season, dyspnea
during exercise, and recent episodes of acute bronchitis was self-reported in
the exposed group. Lung function parameters were only slightly (<10%) lower in
the Mn-exposed workers, with the only significant alterations evident in
-Mn-exposed smokers. These mild changes in Mn-exposed workers (apart from the
effects of smoking) and the absence of respiratory effects in the more recent
study by Roels et al. (1992) suggest that the nervous system is a more
sensitive target for Mn toxicity.
.Based upon the findings of impaired neurobehavioral function in workers whose
average Hn exposure was estimated by the geometric mean TWA of total airborne
_Mn dust at the time of the study, a LOAEL of 0.97 mg/cu.m was identified, with
.a LOAEL(HEC) of 0.34 mg/cu.m. Note that this LOAEL(HEC) is based on total Mn
:dust of mixed forms, whereas the LOAEL(HEC) from the more recent Roels et al.
(1992) study is based on the measured respirable dust fraction of Mn02 only.
-------
Manganese RfC-6
n REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
However, the geometric mean total dust concentrations in the 1987 and 1992
studies by Roels et al. were approximately the same (0.94 and 0.95 mg/cu.m,
respectively).
The findings of Roels et al. (1987, 1992) are supported by other recent reports
that provide comparable and consistent indications of neurobehavioral
dysfunction in Mn-exposed workers (Mergler et al., 1993; Iregren, 1990;
Wennberg et al., 1991, 1992).
Mergler et al. (1993) conducted a cross-sectional study of 115 male
ferromanganese and silicomanganese alloy workers in southwest Quebec. A
matched-pair design was employed because of presumptively high environmental
pollutant levels; 74 pairs of workers and referents were matched on age,
educational level, smoking status, number of children, and length of residency
in the region.
Air concentrations of respirable and total dust were sampled by stationary
monitors during silicomanganese production. The geometric mean of a series of
8-hour TWAs was 0.035 mg Mn/cu.ra (range: 0.001-1.273 mg Mn/cu.m) for respirable
dust and 0.225 mg Mn/cu.m (range: 0.014-11.480 mg Mn/cu.m) for total dust. The
authors noted that past dust levels at certain job sites had been considerably
higher. The mean duration of the workers' Mn exposure was 16.7 years and
included Mn fumes as well as mixed oxides and salts of Mn. Geometric mean MnB
was significantly higher in the Mn alloy workers, but MnU did not differ
significantly between exposed workers and controls.
The number of discordant pairs, in which workers reported undesirable symptoms
on a self-administered questionnaire but their matched pairs did not, was
statistically significant for 33 of 46 items, including the following: fatigue;
emotional state; memory, attention, and concentration difficulties; nightmares;
sweating in the absence of physical exertion; sexual dysfunction; lower back
pain; joint pain; and tinnitus. Workers did not report symptoms typical of
advanced Mn poisoning (e.g., hand tremor, changes in handwriting, loss of
balance when turning, difficulty in reaching a fixed point) significantly more
than referents, which suggests that the other reported symptoms were probably
not due to bias on the part of the workers.
The greatest differences in neurobehavioral function were evident in tests of
motor function, especially tests requiring coordinated alternating and/or rapid
movements. Workers performed significantly worse on the motor scale of a
neuropsychological test battery both in overall score and in eight subscales of
rapid sequential or alternating movements. Worker performance also was
significantly worse on tests of hand steadiness, parallel-line drawing
performance, and ability to rapidly identify and mark specified alphabetic
characters within strings of letters. Performance on a variety of other tests
of psychomotor function, including simple reaction time, was worse in
Mh-exposed workers but marginally significant (0.05
-------
Manganese RfC-7
——-— REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
differences in neurobehavioral performance between workers and referents. This
possibility is supported by the fact that the finger-tapping speed of both
workers and referents on a computerized test was slower than that of Mn-exposed
workers assessed on the same test by Iregren (1990) in Sweden. In the absence
of a NQAEL, the LOAEL from the study of Mergler et al. (1993) is based on the
geometric mean respirable dust level (0.035 mg Mn/cu.m), with a LOAEL(HEC) of
approximately 0.01 mg/cu.m, which is about five-fold lower than the LOAEL(HEC)
Identified in the study by Roels et al. (1992).
Workers exposed to Mn in two Swedish foundries (15 from each plant) were
evaluated in a study first reported by Iregren (1990). The exposure to Mn
varied from 0.02 to 1.40 mg/cu.m (mean - 0.25 mg/cu.m; median =0.14 mg/cu.m)
for 1-35 years (mean = 9*9 years). Earlier monitoring measurements made in
both factories suggested that essentially no changes in exposure had occurred
in either factory for the preceding 18 years. Each exposed worker was matched
for age, geographical area, and type of work to two workers not exposed to Mn
in other industries. Neurobehavioral function was assessed by eight
computerized tests and two manual dexterity tests. There were significant
differences between exposed and control groups for simple reaction time, the
standard deviation of reaction time, and finger-tapping speed of the dominant
hand. In addition, digit-span short-term memory, speed of mental addition, and
verbal (vocabulary) understanding differed significantly between exposed and
control groups. The difference in verbal understanding suggested that the two
groups were not well matched for general cognitive abilities. With verbal
performance used as an additional matching criterion, differences between the
groups in simple reaction time, the standard deviation of reaction time, and
finger-tapping speed remained statistically significant, despite a decrease in
statistical power due to reducing the size of the reference group to 30
workers. Further analyses using verbal test scores as a covariate also
indicated that these same three measures of neurobehavioral function were
statistically different in exposed and control workers. No significant
correlation was found within the exposed group to establish a
concentration-response relationship.
Additional reports of neurobehavioral and electrophysiological evaluations of
-these same workers have been published by Wennberg et al. (1991, 1992).
Although none of the latter results achieved statistical significance at p =
0.05, increased frequency of self-reported health symptoms, increased
prevalence of abnormal electroencephalograms, slower brain-stem auditory-evoked
potential latencies, and slower diadochokinesometric performance were found in
•the exposed workers. Diadochokinesis refers to the ability to perform rapidly
Alternating movements such as supination and pronation of the forearm, and is
mo. indicator of extrapyramidal function (see Additonal Comments /Studies).
Also, an increase in P-300 latency, as suggested by these results, has been
observed in patients with parkinsonism according to Wennberg et al. (1991), who
viewed these results in Mn-exposed workers as early (preclinical) signs of
disturbances similar to parkinsonism. Based on the impairments in reaction
time and finger-tapping speed reported in the Swedish studies (Iregren, 1990;
Wennberg et al., 1991, 1992), the LOAEL(HEC) is calculated to be 0.05 mg/cu.m.
Although numerically the same value as that derived from Roels et al. (1992),
the Swedish study measured total dust. However, Wennberg et al. (1991) stated
tthat the respirable dust level was 20-80% of total dust, which implies that the
lOAEL(HEC) from the Swedish studies could be somewhat lower than that from
Heels et al. (1992).
-------
Manganese RfC-8
y..t... a. REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC) —
All of the above studies taken together provide a consistent pattern of
evidence indicating that neurotoxicity is associated with low-level
occupational Mn exposure. The fact that the speed and coordination of motor
function are especially impaired is consistent with other epidemiological,
clinical, and experimental animal evidence of Mn intoxication (see Additional
Comments/Studies). Moreover, the LOAEL(HEC)s obtained from these studies are
not appreciably different. Nevertheless, some differences between the studies
jure evident in the durations of exposure and forms of Mn to which workers were
exposed. In the Roe Is et al. (1992) study, the mean period of exposure was 5.3
years (range: 0.2-17.7 years), and exposure was limited to Mn02. In the other
studies, mixed forms of Mn were present, and the mean durations of exposure
were longert 7.1 years in RoeIs et al. (1987), 9.9 years in Iregren (1990),
and 16.7 years in Mergler et al. (1993). The findings of Mergler et al. (1993)
suggest that the LOAEL(HEC) could be at least as low as approximately 0.01
mg/cu.m. However, the variable concentrations and mixed compounds of Mn to
which workers were exposed make it difficult to rely primarily upon the
findings of Mergler et al. (1993) in deriving the RfC. Nevertheless, their
results provide support for the findings of Roels et al. (1992) and suggest
that the longer period of exposure (16.7 years in Mergler et al. (1993) vs. 5.3
years in Roels et al., 1992) may have contributed to the apparent differences
in sensitivity. Although analyses by Roels et al. (1987, 1992) and Iregren
(1990) generally did not indicate that duration of exposure correlated
significantly with neurobehavioral outcomes, none of these studies involved
average exposures as long as those in the Mergler et al. (1993) study. Also,
the oldest worker in the Roels et al. (1992) study was less than 50 years old,
and the average age in that study was only 31.3 years vs. 34.3 years in Roels
et al. (1987), 43.4 years in Mergler et al. (1993), and 46.4 in Iregren (1990).
These points suggest that chronic exposure to Mn and/or interactions with
aging could result in effects at lower concentrations than would be detected
after shorter periods of exposure and/or in younger workers.
Based on the findings of neurobehavioral impairment by Roels et al. (1987,
1992), with supporting evidence from Mergler et al. (1993) and the Swedish
reports (Iregren, 1990; Wennberg et al., 1991, 1992), the IiQAEL for derivation
of the RfC is 0.15 mg/cu.m, and the LOAEL(HEC) is 0.05 mg/cu.m.
UNCERTAINTY FACTORS:
UNCERTAINTY AND MODIFYING FACTORS
An uncertainty factor of 1000 reflects 10 to protect sensitive individuals, 10
for use of a LOAEL, and 10 for database limitations reflecting both the
less-than-chronic periods of exposure and the lack of developmental data, as
well as potential but unguantified differences in the toxicity of different
forms of Mn.
MODIFYING FACTORS:
None
ADDITIONAL COMMENTS / STUDIES
-Manganese toxicity varies depending upon the route of exposure. When ingested,
Jto is considered to be among the least toxic of the trace elements. In the
normal adult, between 3 and 10% of dietary Mn is absorbed. Total body stores
-------
Manganese RfC-9
——— REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
normally are controlled by a complex homeostatic mechanism regulating
absorption and excretion. As detailed in the Uncertainty Factor Text and the
Additonal Comments /Studies for the oral RfD, toxicity from ingested Bin is
rarely observed. However, deficiencies of calcium and iron have been noted to
increase Mn absorption (Mena et al., 1969? Murphy et al., 1991). Also, Mena et
al. (1969) found that anemic subjects absorbed 7.5% of ingested Mn, whereas
normal subjects absorbed 3%. Interestingly, manganism patients absorbed 4%, and
apparently healthy Mn miners absorbed only 3%. These differences suggest that
certain subpopulations, such as children, pregnant women, elderly persons,
iron- or calcium-deficient individuals, and individuals with liver impairment,
may have an increased potential for excessive Mn body burdens due to increased
absorption or altered clearance mechanisms, which may be of particular
importance for those exposed to Mn by multiple routes.
As a route of Mn exposure, the respiratory tract is the most important portal
of entry. The inhalation toxicity of Mn is in part a function of particle
dosimetry and subsequent phannacokinetic events, particle size determines the
site of deposition in the respiratory tract. Generally, in humans, fine mode
particles (<2.5 urn) preferentially deposit in the pulmonary region, and coarse
mode particles (>2.5 urn) deposit in the tracheobronchial and extrathoracic
regions. Those particles depositing in the extrathoracic and tracheobronchial
regions are predominantly cleared by the mucociliary escalator into the
gastrointestinal tract where absorption is quite low (about 3%). Particles
deposited in the pulmonary region are cleared predominantly to the systemic
compartment by absorption into the blood and lymph circulation. Disregarding
the possibility of counteracting mechanisms, 100% absorption of particles
deposited in the pulmonary region is assumed. Another possible route of
exposure may exist. Studies such as those of Perl and Good (1987) and Evans
and Hastings (1992) have indicated that neurotoxic metals such as aluminum and
cadmium can be directly transported to the brain olfactory bulbs via nasal
olfactory pathways (i.e., from extrathoracic deposition). The alteration in
olfactory perception that Mergler et al. (1993) found in Mn-exposed workers
lends support to the speculation that this pathway may also operate for Mn,
which would further complicate understanding of target-site dosimetry.
The human health effects data base on Mn does not include quantitative
inhalation pharmacokinetics information on the major oxides of Mn. Two of the
studies described in the Principal and Support Studies (Roels et al., 1992;
Mergler et al., 1993) measured respirable as well as total Mn dust, and one
study (Roels et al., 1992) dealt with workers exposed to only one form of Mn,
namely MnO2. However, this information does not allow quantitative
determinations of the dose delivered to the respiratory tract or estimates of
target-site doses. After absorption via the respiratory tract, Mn is
transported through the blood stream directly to the brain, bypassing the liver
and the opportunity for first-pass hepatic clearance. This direct path from
the respiratory tract to the brain is the primary reason for the differential
toxicity of inhaled and ingested Mn. Phannacokinetic analyses based on
inhalation of manganese chloride (MnC12) by macaque monkeys (Newland et al.,
1987) indicated that clearance from the brain was slower than from the
respiratory tract and that the rate of clearance depended on the route of
exposure. Brain half-times were 223-267 days after inhalation vs. 53 days
following subcutaneous administration (Newland et al., 1987) or 54 days in
humans given Mn intravenously (Cotzias et al., 1968). These long half-times
were thought to reflect both slower clearance of brain stores and replenishment
from other organs, particularly the respiratory tract. In rats, Drown et al.
-------
Manganese RfC-10
REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC) —
(1986) also observed slower clearance of labeled Mn from brain than from the
respiratory tract. Several occupational physicians have reported large
individual differences in workers' susceptibility to Mn intoxication, which
Rodier (1955) speculated might be due in part to differences in the ability to
clear particulate Mn from the lung.
The bioavailability of different forms of Mn is another matter for
consideration. Roels et al. (1992) noted that geometric mean blood and urinary
Jto levels of workers exposed only to Mn02 in their 1992 report were lower (MnBs
0.81 ug/dLj MnU: 0.84 ug/g creatinine) than those of workers exposed to mixed
oxides and salts in their 1987 report (MnBs 1.22 ug/dL| MnUs 1.59 ug/g
creatine), even though airborne total dust levels were approximately the same
(geometric means of 0.94 and 0.95 mg/cu.m, respectively). Mena et al. (1969)
observed no difference between the absorption of 1 urn particles of MnC12 and
manganese sesguioxide (Mn203) in healthy adults. Drown et al. (1986) found
that following intratracheal instillation of MnC12 and Mn304 in rats, the
soluble chloride cleared four times faster than the insoluble oxide from the
respiratory tract. However, despite this initial difference, after 2 weeks the
amounts of labeled Mn in the respiratory tract were similar for the two
compounds. Recent work by Koraura and Sakamoto (1993) comparing different forms
of Mn in mouse diet suggested that less soluble forms such as Mn02 were taken
up to a significantly greater degree in cerebral cortex than the more soluble
forms of MnC12 and manganese acetate [Mn(CH3COO)2]; however, the corpus
striatal binding characteristics of the +4 valence state of Mn in MnO2 were not
substantially different from those of the divalent forms in MnC12, Mn(CH3COO)2,
and manganese carbonate. Different oxidation states of certain metals (e.g.,
chromium, nickel, mercury) are known to have different toxicities, and some
researchers have suggested that endogenous Mn can have quite different roles in
Hn neurotoxicity depending on its oxidation state (e.g., Archibald and Tyree,
1987; Donaldson et al., 1982). There have been unsubstantiated claims that the
.higher valence states of Mn (Mn+3, Mn+4) and the higher oxides in ores (Mn203
and Mn304) are more toxic (Oberdoerster and Cherian, 1988). At present,
Jiowever, insufficient information exists by which to determine the relative
toxicities of different forms of Mn, and thus, for the purpose of deriving an
RfC for Mn, no distinction is made among various compounds of Mn.
.Because Mn is an essential element and is commonly ingested in diet, total Mn
exposure is an issue. It would be desirable to know the effective target-site
doses and apportion the dose to both the inhalation and oral routes of
exposure. However, given the lack of data regarding oral and inhalation
phanaacokinetics under environmental conditions, such quantitative
apportionment is not possible at present.
Among the primary effects associated with Mn toxicity from inhalation exposure
In humans are signs and symptoms of CNS toxicity. The first medical
description of chronic Mn neurotoxicity (manganism) in workers is generally
credited to Couper in the 1830s (HAS, 1973). Although the course and degree of
Jto intoxication can vary greatly among individuals, manganism is generally
considered to consist of two or three phases (Rodier, 1955). The first is the
psychiatric aspect, which includes disturbances such as excessive weeping and
laughing, sleep disturbance, irritability, apathy, and anorexia. These
symptoms can occur independently of the second phase, neurological signs. The
-latter may include gait disturbances, dysarthria, clumsiness, muscle cramps,
tremor, and mask-like facial expression. In addition, there may be a final
stage of Mn intoxication involving symptoms of irreversible dystonia and
-------
Manganese RfC-11
«« REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
hyperflexion of muscles that may not appear until many years after the onset of
exposure (Cotzias et al., 1968). Cotzias et al. {1976) noted a parallel
between these stages of symptoms and the biphasic pattern of dopamlne levels
over time in the Mn-exposed individuals noted above. Indeed, various specific
features of Mn toxicity show biphasic patterns in which there is generally
first an increase then a decrease in performance (e.g., a notable increase in
libido followed by impotence, or excitement followed by somnolence) (Rodier,
1955).
In addition to studies described in the Principal and Supporting studies,
numerous investigators have reported CNS effects in workers exposed to Mn dust
or fumes (Sjoegren et al,, 1990; Huang et al., 1989; Wang et al., 1989; Badawy
arid Shakour, 1984; Siegl and Bergert, 1982; Chandra et al., 1981; Saric et al.,
1977; Cook et al., 1974; Smyth et al., 1973; Emara et al., 1971; Tanaka and
iieben, 1969; Schuler et al., 1957; Rodier, 1955; Flinn et al., 1941).
Limitations in these studies generally preclude describing a quantitative
concentration-response relationship. Exposure information is often quite
limited, with inadequate information on the historical pattern of Mn
concentrations or on the chemical composition and particle size distribution of
the dust. In addition, exposure to other chemicals in the workplace is often
not adequately characterized. Despite these limitations, such studies
collectively point to neurobehavioral dysfunction as a primary endpoint for Mn
toxicity.
The neuropathological bases for manganism have been investigated by many
researchers and have indicated the involvement of the corpus striatum and the
extrapyramidal motor system (e.g., Archibald and Tyres, 1987; Donaldson and
Barbeau, 1985; Donaldson et al., 1982; Eriksson et al., 1987, 1992).
Neuropathological lesions have generally been associated with the basal
ganglia, specifically involving neuronal degeneration in the putamen and globus
pallidus (e.g., Newland et al., 1987). Brain imaging studies (e.g., Wolters et
al., 1989; Nelson et al., 1993) more recently have begun to provide additional
insight into the brain structures involved in Mn toxicity.
In terms of the neurochemistry of Mn toxicity, several studies have shown that
dopamine levels are affected by Mn exposure in humans, monkeys, and rodents,
with various indications of an initial increase in dopamine followed by a
longer term decrease (e.g., Cotzias et al., 1976; Bird et al., 1984; Barbeau,
1984; Brouillet et al., 1993). Some theories of Mn neurotoxicity have focused
on the role of excessive Mn in the oxidation of dopamine resulting in free
radicals and cytotoxicity (e.g., Donaldson et al., 1982; Barbeau, 1984). In
addition, the fundamental role of mitochondrial energy metabolism in Mn
toxicity has been indicated by the studies of Aschner and Aschner (1991), Gavin
et al. (1992), and others. Brouillet et al. (1993) have suggested that the
mitochondrial dysfunctional effects of Mn result in various oxidative stresses
to cellular defense mechanisms (e.g., glutathione) and, secondarily, free
radical damage to mitochondrial DNA. In view of the slow release of Mn from
mitochondria (Gavin et al., 1992), such an indirect effect would help account
for a progressive loss of function in the absence of ongoing Mn exposure
(Brouillet et al., 1993), as Mn toxicity has been known to continue to progress
in humans despite the termination of exposure (Cotzias et al., 1968; Rodier,
1955).
JBecause of the involvement of the dopaminergic system and extrapyramidal motor
system in both Parkinson's disease and manganism, symptoms of the two diseases
-------
Manganese RfC-12
REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
are somewhat similar, and several writers have suggested the possibility of a
common etiology; however, many neurological specialists make a clear
distinction in the etiologies and clinical features of Parkinson's disease and
manganism (Barbeau, 1984? Langston et al., 1987),
Another primary endpoint of Kn toxicity has been male reproductive dysfunction,
often manifesting in symptoms such as loss of libido, impotence, and similar
complaints (e.g., Rodier, 1955? Cook et al.f 1974). Some neuropathological
evidence suggests that the hypothalamus is a site of Mn accumulation (Donaldson
et al., 1973); thus, disturbance of the hypothalamic-pituitary-gonadal axis
hormones might be expected (Deskin et al., 1981) and has been examined in a few
occupational studies. Lauwerys et al. (1985) reported the results of a
fertility questionnaire administered to male factory workers (n = 85) exposed
to Mn dust. This study involved the same population of workers for which Roels
et al. (1987) reported neurobehavioral disturbances. The range of Mn levels in
the breathing zone was 0.07-8.61 mg/cu.m, with a median concentration of 0.97
mg/cu.m. Average length of exposure was 7.9 years (range of 1-19 years). A
group of workers (n = 81) with a similar workload was used as a control group.
The number of births expected during different age intervals of the workers
(16-25, 26-35, and 36-45 years) was calculated on the basis of the reproductive
experience of the control employees during the same period. A decrease in the
number of children born to workers exposed to Mn dust during the ages of 16-25
and 26-35 was observed. No difference in the sex ratio of the children was
found. The same apparent LOAEL(HEC) (0.34 mg/cu.m) that was identified in
Roels et al. (1987) for neurobehavioral effects is identified in this study for
human reproductive effects.
However, a more recent report from the same group of investigators (Gennart et
al., 1992), based on 70 of the alkaline battery plant workers evaluated by
Roels et al. (1992), indicated that the probability of live birth was not
different between the Mn-exposed and control workers. Also, in the study by
Roels et al. (1992), serum levels of certain hormones related to reproductive
function (FSH, LH, and prolactin) were not significantly different for the full
group of 92 Mn workers vs. 102 controls. The latter results are partially
supported by a preliminary report by Alessio et al. (1989), who found that
serum FSH and LH levels were not significantly different in 14 workers
generally exposed to <1 mg Mn/cu.m compared to controls, although prolactin and
cortisol levels were significantly higher in the Mn-exposed workers. It is
possible that differences in the forms of Mn to which workers were exposed in
these studies may have contributed to the similarities and differences in the
results, but insufficient information exists to substantiate this speculation.
Average concentrations of airborne Mn differed slightly in the reports of
Gennart et al. (1992) and Roels et al. (1992), evidently because only a subset
of Mn workers, presumably with different job functions, was used in the Gennart
et al. (1992) analysis. The median respirable dust concentration was 0.18
mg/cu.m, and the median total dust concentration (comparable to Roels et al.,
1987, and Lauwerys et al., 1985) was 0.71 mg/cu.m. Thus, if 0.34 mg/cu.m is
identified as a LOAEL(HEC) based on the reports of Lauwerys et al. (1985) and
Roels et al. (1987), 0.25 mg/cu.m total dust is the NOAEL(HEC) for reproductive
effects based on the report of negative findings by Gennart et al. (1992).
The respiratory system is another primary target for Mn toxicityj numerous
reports of Mn pneumonitis and other effects on the respiratory system have
appeared in the literature, dating back to 1921 (NAS, 1973). In their
-------
Manganese RfC-13
• ' ' REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
cross-sectional study of workers exposed to mixed Mn oxides and salts
(described in the Principal and Supporting Studies}f Roels et al. (1987) found
that significantly greater prevalences of coughs during the cold season,
dyspnea during exercise, and recent episodes of acute bronchitis were reported
an the exposed group on a self-administered questionnaire. However,
objectively measured lung function parameters were only slightly altered and
tmly in Mn-exposed smokers (also see Saric and Lucic-Palaic, 1977, regarding a
^possible synergism between Mn and smoking in producing respiratory symptoms).
In their more recent study, Roels et al. (1992) found no significant
differences between Mn02-exposed and control workers in responses to a
questionnaire regarding respiratory symptoms. Nor were objective spirometric
measurements significantly different for the two groups. The LOAEL(HEC) for
respiratory effects is 0.34 mg/cu.m total dust, based on the Roels et al.
(1987) study, and the NOAEL(HEC) is 0.05 mg/cu.m respirable dust, based on the
Roels et al. (1992) study. In view of the near equivalence of the geometric
mean total dust concentrations in the 1987 and 1992 studies by Roels et al.
(0.94 and 0.95 mg/cu.m, respectively), there in fact may be little difference
.between the LOAEL(HEC) and the NOAEL(HEC) in terms of air concentrations;
however, differences in the forms of Mn (Mn02 vs. mixed Mn oxides and salts) to
which the workers in the two studies were exposed make it difficult to compare
these values quantitatively.
Nogawa et al. (1973) investigated an association between atmospheric Mn levels
and respiratory endpoints in junior high school students. A questionnaire
focusing on eye, nose, and throat symptoms and pulmonary function tests were
given to students attending junior high schools that were 100 m (enrollment =
1258) and 7 km (enrollment = 648) from a ferromanganese plant. Approximately
97-99% of the students participated. Based on measurements obtained at another
time by a government agency, the 5-day average atmospheric Mn level 300 m from
•the plant was reported to be 0.0067 mg/cu.m.
Significant increases in past history of pneumonia, eye problems, clogged nose,
nose colds, throat swelling and soreness, and other symptoms were noted among
the students in the school 100 m from the plant. Those living closest to the
plant reported more throat symptoms and past history of pneumonia than did
students living farther away. Pulmonary function tests revealed statistically
.significant decreases in maximum expiratory flow, forced vital capacity (FVC),
iorced expiratory volume at 1 second (FEV-1), and the FVCsFEV-1 ratio in the
students attending the school closer to the plant, with some measures
suggesting a relationship between performance and distance of residence from
-the plant.
.Although the results from the study of Nogawa et al. (1973) suggest an
association between ambient Mn exposure and respiratory problems, limitations
in the study make it difficult to interpret. No direct measurements were made
of atmospheric Mn levels either in the schools or homes, and exposure levels
were inferred from the distance from the plant and other indirect measures of
Jtn in the environment. Also, the authors did not note whether socioeconoraic
"variables were controlled, and this factor could well be confounded with both
distance from the plant and health problems. A follow-up study by Kagamimori
*t al. (1973) suggested that, following reductions in Mn emissions (with
.apparently no reduction in sulfur dioxide or total dust) from the
ferromanganese plant, students nearest the plant showed improvements in
.subjective symptoms and pulmonary function tests. As before, however, exposure
-levels were not adequately characterized to allow clear-cut conclusions.
-------
Manganese RfC-14
REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
Lloyd-Davies (1946) reported an increased incidence of pneumonia in men
employed at a potassium permanganate manufacturing facility over an 8-year
period. During that period, the number of workers in the facility varied from
40 to 124. Dust measurements were well described in terms of collection
conditions and particle size and composition, but actual exposure levels were
not evaluated. Air concentrations ranged from 9.6 to 83.4 mg/cu.m as Mn02,
which constituted 41-66% of the dust. The incidence of pneumonia in the
workers was 26 per 1000, compared to an average of 0.73 per 1000 in a reference
group of over 5000 workers. Workers also complained of bronchitis and nasal
irritation. In a continuation of this study, Lloyd-Davies and Harding (1949)
reported the results of sputum and nasopharynx cultures for four men diagnosed
as having lobar- or bronchopneumonia. With the exception of one of these
cases, they concluded that Mn dust, without the presence of bacterial infection
or other factors, caused the observed pneumonitis.
Evidence from several laboratory animal studies supports findings in Mn-exposed
humans. For example, inhaled Mn has been shown to produce significant
alterations in dopamine levels in the caudate and globus pallidus of Rhesus
monkeys (Bird et al.f 1984) and behavioral changes in mice (Morganti et al.,
1985). However, species differences may complicate interpretation of certain
neurobehavioral findings in laboratory animals. Unlike primates, rodents do
not have pigmented substantia nigra, which is a brain region of relatively high
Mn uptake and consequent involvement in neurobehavioral dysfunction.
Nevertheless, rodent and primate studies show various neurochemical,
neuropathological, and neurobehavioral effects resulting from Mn exposure.
However, because most laboratory animal studies of Mn neurotoxicity involve
exposure by routes other than inhalation, they are not described here.
Other endpoints of Mn toxicity also have been investigated with laboratory
animal models of inhalation exposure. Experimental animal data qualitatively
support human study findings of respiratory effects in that Mn exposure results
in increased incidence of pneumonia in rats exposed to 68-219 mg/cu.m MnO2 for
2 weeks (Shiotsuka, 1984), pulmonary emphysema in monkeys exposed to 0.7-3.0
mg/cu.m Mn02 for 10 months (Suzuki et al., 1978), and bronchiolar lesions in
rats and hamsters exposed to 0.117 mg/cu.m Mn304 for 56 days (Moore et al.,
1975). Also, Lloyd-Davies and Harding (1949) induced bronchiolar epithelium
inflammation, widespread pneumonia, and granulomatous reactions in rats
administered 10 mg Mn02 by intratracheal injection, and pulmonary edema in rats
administered 5-50 mg MnCl2 in the same fashion. However, no significant
pulmonary effects were detected in other studies of rats and monkeys exposed to
as much as 1.15 mg Mn/cu.m as Mn304 for 9 months (Ulrich et al., 1979a,b,c) or
rabbits exposed to as much as 3.9 mg Mn/cu.m as MnC12 for 4-6 weeks (Camner et
al., 1985).
Laboratory animal studies also have shown that inhaled Mn may increase
susceptibility to infectious agents such as Streptococcus pyogenes in mice
(Adkins et al., 1980), Enterobacter cloacae in guinea pigs (Bergstrom, 1977),
Klebsiella pneumoniae in mice (Maigetter et al., 1976), and Streptococcus
hemolyticus in mice (Lloyd-Davies, 1946). In general, Mn concentrations were
relatively high (>10 mg/cu.m) in these studies. However, Adkins et al. (1980)
concluded that, based on the regression line of the relationship between
concentration and mortality in Mn-exposed mice, exposure to <0.62 mg/cu.m would
result in a mortality rate that differed from controls by at least 10%.
-------
Manganese RfC-15
«======» REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC) —=-
The developmental effects of Mn have been investigated primarily from the
viewpoint of the nutritional role of this element and therefore have generally
involved oral exposure. Some studies indicate that neonates of various species
have a greater body burden of Mn than mature individuals have, possibly because
neonates do not develop the ability to eliminate Mn—and thereby maintain Mn
iiomeostasis—until some time after birth (Miller et al., 1975; Cotzias et al.,
J.976; Wilson et al., 1991). Moreover, some evidence suggests that the
meonate's inability to maintain Mn homeostasis is due to a limitation in the
elimination of Mn rather than in its gastrointestinal absorption (Bell et al.,
1989), which would suggest a potentially greater vulnerability of young
individuals to excessive Mn exposure regardless of the route of exposure.
Several studies have demonstrated neurochemical alterations in young rats and
aaice exposed postnatally to Mn by routes other than inhalation (e.g., Kontur
and Fechter, 1988; Seth and Chandra, 1984; Deskin et al., 1981; Cotzias et al.,
1976). The only inhalation study of the developmental toxicity of Mn appears
-to be that of Lown et al. (1984). Female mice were exposed to Mn02 7
Jiours/day, 5 days/week for 16 weeks prior to conception and for 17 days
following conception (i.e., gestational days 1-18). For the first 12 weeks,
the air concentration was 49.1 mg Mn/cu.m; all later exposures were at 85.3 mg
.Mn/cu.m. To separate prenatal and postnatal exposure effects, a
cross-fostering design was used. Although mothers exposed to MnO2 prior to
conception produced significantly worse pups per litter, prenatally exposed
offspring showed reduced scores on various activity measures (open field,
xoto-rod, and exploration) and retarded growth that persisted into adulthood.
A decrease in roto-rod performance was also observed in the offspring of
nonexposed mice that were fostered to Mn-exposed females during lactation.
Thus, balance and coordination were affected by either gestational or
postpartum exposure to Mn02.
CONFIDENCE IN THE RfC
Study: Medium Data Base: Medium RfC: Medium
Confidence in the principal studies (Roels et al., 1987, 1992) is medium.
.neither of the principal studies identified a NOAEL for neurobehavioral
effects, nor did either study directly measure particle size or provide
information on the particle size distribution. The 1992 study by Roels et al.
jiid provide respirable and total dust measurements, but the 1987 study measured
t»nly total dust. These limitations of the studies are mitigated by the fact
±hat the principal studies found similar indications of neurobehavioral
dysfunction, and these findings were consistent with the results of other human
studies (Mergler et al., 1993; Iregren, 1990; Wennberg et al., 1991, 1992; as
well as various clinical studies). In addition, the exposure history of the
•workers in the 1992 study by Roels et al. was well characterized and
-essentially had not changed over the preceding 15 years, thereby allowing
calculation of integrated exposure levels for individual workers. However,
individual integrated exposures were not established in the 1987 study of Roels
«t al.
Confidence in the data base is medium. The duration of exposure was relatively
JLimited in all of the principal and supporting studies, ranging from means of
3.3 and 7.1 years in the co-principal studies by Roels et al. (1992 and 1987,
^respectively) to a maximum of 16.7 years in the study by Mergler et al. (1993),
Jforeover, the workers were relatively young, ranging from means of 31.3 and
-------
Manganese RfC-16
=== REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
34.3 years in the co-principal studies (Roels et al., 1992 and 1987,
respectively) to a maximum of 46.4 years (Iregren, 1990). These temporal
limitations raise concerns that longer durations of exposure and/or
interactions with aging might result in the detection of effects at lower
concentrations, as suggested by results from studies involving longer exposure
durations and lower concentrations (Mergler et al., 1993? Iregren, 1990). In
addition, except for the 1992 study by Roels et al., in which Mn exposure was
limited to Mn02, the other principal and supporting studies did not specify the
species of Mn and the proportions of the different compounds of Mn to which
workers were exposed. It is not clear whether certain compounds or oxidation
states of Mn are more toxic than others. Thus, it is not possible to
distinguish the relative toxicity of different Mn compounds in these studies,
despite some indications in the literature regarding the differential toxicity
of various oxidation states of Mn. Although the primary neurotoxicological
effects of exposure to airborne Mn have been qualitatively well characterized
by the general consistency of effects across studies, the exposure-effect
relationship remains to be well quantified, and a no-effect level for
neurotoxicity has not been identified in any of these studies thus far.
Finally, the effects of Mn on development and reproduction have not been
studied adequately. Insufficient information on the developmental toxicity of
Mn by inhalation exposure exists, and the same is true for information on
female reproductive function. The study of the reproductive toxicity of
inhaled Mn in males also needs to be characterized more fully.
Reflecting medium confidence in the principal studies and medium confidence in
the data base, confidence in the inhalation RfC is medium.
EPA DOCUMENTATION AND REVIEW
Source Document: This assessment is not presented in any existing U.S. EPA
document.
Other EPA Documention: U.S. EPA, 1984
Agency Work Group Review: 08/23/90, 09/19/90, 09/23/93
Verification Date: 09/23/93
> EPA CONTACTS —
J. Michael Davis / OHEA — (919)541-4162
Annie M. Jarabek / OHEA — (919)541-4847
• BIBLIOGRAPHY
Adkins, B.f Jr., G.H. Luginbuhl, F.J. Miller, and D.E. Gardner. 1980.
Increased pulmonary susceptibility to streptococcal infection following
inhalation of manganese oxide. Environ. Res. 23: 110-120.
Alessio, L., P. Apostoli, A. Perioli, and S. Lombardi. 1989. Interference of
manganese on neuroendocrinal system in exposed workers. Preliminary report.
Mol. Trace Element Res. 21: 249-253.
Jlrchibald, F.S. and C. Tyree. 1987. Manganese poisoning and the attack of
-------
Manganese RfC-17
—= REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
trivalent manganese upon catecholamines. Arch. Biochero. Biophys. 256:
638—650.
Aschner, M. and J.L. Aschner. 1991. Manganese neurotoxicity: Cellular effects
and blood-brain barrier transport. Neurosci. Biobehav. Rev. 15: 333-340.
Badawy, A.B.N. and A.A. Shakour. 1984. Chronic manganese intoxication
(neurological manifestations). Ins Proceedings of 5th International Synqposium:
Trace Elements in Man and Animals, p. 261-263.
Barbeau, A. 1984. Manganese and extrapyramidal disorders. Neurotoxicology.
5s 13-35.
Bell, J.G., C.L. Keen, and B. Loennerdal. 1989. Higher retention of manganese
in suckling than in adult rats is not due to maturational differences in
manganese uptake by rat small intestine. J. Toxicol. Environ. Health. 26:
387-398.
Bergstrom, R. 1977. Acute pulmonary toxicity of manganese dioxide. Scand. J.
Work Environ. Health. 3: 1-40.
Bird, E.D., A.H. Anton, and B. Bullock. 1984. The effect of manganese
inhalation on basal ganglia dopamine concentrations in rhesus monkey.
Neurotoxicology. 5(l)s 59-65.
Brouillet, E.F., L. Shinobu, U. McGarvey, F. Hochberg, and M.F. Beal. 1993.
Manganese injection into the rat striatum produces excitotoxic lesions by
impairing energy metabolism. Exp. Neurol. 120: 89-94.
Camner, P., T. Curstedt, C. Jarstrand, A. Johannsson, B. Robertson, and A.
Wiernik. 1985. Rabbit lung after inhalation of manganese chloride: A
comparison with the effects of chlorides of nickel, cadmium, cobalt, and
copper. Environ. Res. 38s 301-309.
Chandra, S.V., 6.S. Shukla, R.S. Strivastava, H. Singh, and V.P. Gupta. 1981.
An exploratory study of manganese exposure to welders. Clin. Toxicol. 18(4):
407-416.
Cook, D.6., S. Fahn, and K.A. Brait. 1974. Chronic manganese intoxication.
Arch. Neurol. 30: 59-64.
Cotzias, G.C., K. Horiuchi, S. Fuenzalida, and I. Mena. 1968. Chronic
manganese poisoning: Clearance of tissue manganese concentrations with
persistence of the neurological picture. Neurology. 18: 376-382.
Cotzias, G.C., S.T. Miller, P.S. Papavasiliou, and L.C. Tang. 1976.
Interactions between manganese and brain dopamine. Med. Clin. North Am.
60(4): 729-738.
Deskin, R., S.J. Bursian, and F.W. Edens. 1981. The effect of chronic
manganese administration on some neurochemical and physiological variables in
neonatal rats. Gen. Pharmacol. 12: 279-280.
Donaldson, J., T. St. Pierre, J.L. Minnich, and A. Barbeau. 1973.
Determination of Na+, KH-, Mg2+, Cu2H-, Zn2, and Mn2 in rat brain regions. Can.
-------
Manganese RfC-18
* REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
J. Biochem. 51s 87-92.
Donaldson, 3., D. McGregor, and F. LaBella. 1982. Manganese neurotoxicitys A
rmodel for free radical mediated neurodegeneration? Can. J. Physiol. Pharmacol.
60; 1398-1405.
-Donaldson, J. and A. Barbeau. 1985. Manganese neurotoxicity» Possible clues
±o the etiology of human brain disorders. Ins Metal Ions in Neurology and
Psychiatry, S. Gabay, J. Harris, and B.T. Ho, eds. Alan R. Liss, Inc., New
York, NY. p. 259-285.
Drown, D.B., S.G. Oberg, and R.P. Shanna. 1986. Pulmonary clearance of
soluble and insoluble forms of manganese. J. Toxicol. Environ. Health. 17:
201-212.
JEmara, A.M., S.H. El-Ghawabi, O.I. Madkour, and G.H. El-Samra. 1971. Chronic
manganese poisoning in the dry battery industry. Br. J. Ind. Med. 28s 78-82.
JEriksson, H., K. Magiste, L.-O. Plantin et al. 1987. Effects of manganese
oxide on monkeys as revealed by a combined neurochemical, histological and
neurophysiological evaluation. Arch. Toxicol. 61s 46-52.
Eriksson, H., J. Tedroff, K.A. Thuomas et al. 1992. Manganese induced brain
lesions in Macca fascicularis as revealed by positron emissions tomography and
magnetic resonance imaging. Arch. Toxicol. 66s 403-407.
Evans, J. and L. Hastings. 1992. Accumulation of Cd(Il) in the CNS depending
on the route of administration: Intraperitoneal, intratracheal, or intranasal.
Fund. Appl. Toxicol. 19? 275-278.
71inn, R.H., P.A. Neal, and W.B. Fulton. 1941. Industrial manganese
•poisoning. J. Ind. Hyg. Toxicol. 23(8): 374-387.
Savin, C.E., K.K. Gunter, and T.I. Gunter. 1992. Mn2+ sequestration by
.mitochondria and inhibition of oxidative phosphorylation. Toxicol. Appl.
Pharmacol. 115s 1-5.
Sennart, J.-P., J.-P. Buchet, H. Roels, P. Ghyselen, 1. Ceulemans, and R.
Jiauwerys. 1992. Fertility of male workers exposed to cadmium, lead, or
manganese. Am. J. Epidemiol. 135(11): 1208-1219.
-fluang, C.-C., N.-S. Chu, C.-S. Lu et al. 1989. Chronic manganese
Jjitoxication. Arch. Neurol. 46(10): 1104-1106.
-Iregren, A. 1990. Psychological test performance in foundry workers exposed
±o low levels of manganese. Neurotoxicol. Teratol. 12: 673-675.
JKagamimori, S., T. Makino, Y. Hiramaru et al. 1973. [Studies concerning the
Bffects on the respiratory organs of atmospheric pollution caused by dusts
consisting mainly of manganese. (Report No. 2). On the changes in the effects
on the human organism after the environment has been improved]. Nippon Koshu
.Eisei lasshi. 20: 413-421.
JKomura, J. and M. Sakamoto. 1993. Subcellular and gel chromatographic
distribution of manganese in the mouse brain: Relation to the chemical form of
-------
Manganese RfC-19
——= REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
chronically-ingested manganese. Toxicol. Lett. 66: 287-294.
Kontur, P.J. and L.D. Fechter. 1988. Brain regional manganese levels and
monoamine metabolism in manganese-treated neonatal rats. Neurotoxicol.
Teratol. 10s 295-303.
Langs ton, J.W., I. Irwin, and G.A. Ricaurte. 1987. Neurotoxins, parkinsonism,
and Parkinson's disease. Pharmacol. Ther. 32: 19-49.
Lauwerys, R., H. Roels, P. Genet, G. Toussaint, A. Bouckaert, and S. de Cooman.
1985. Fertility of male workers exposed to mercury vapor or to manganese
dust: A questionnaire study. Am. J. Ind. Med. 7: 171-176.
Lloyd-Davies, T.A. 1946. Manganese pneumonitis. Br. J. Ind. Med. 3:
111-135.
Lloyd-Davies, T.A. and H.S. Harding. 1949. Manganese pneumonitis; Further
clinical and experimental observations. Br. J. Ind. Med. 6: 82-90.
Lown, B.A., J.B. Morganti, R. D'Agostino, C.H. Stineman, and E.J. Massaro.
1984. Effects on the postnatal development of the mouse of preconception,
postconception and/or suckling exposure to manganese via maternal inhalation
exposure to MnO2 dust. Neurotoxicology. 5(1): 119-131.
Maigetter, R.Z., R. Ehrlich, J.D. Fenters, and D.E. Gardner. 1976.
Potentiating effects of manganese dioxide on experimental respiratory
infections. Environ. Res. 11s 386-391.
Mena, I., K. Horiuchi, K. Burke, and G.C. Cotzias. 1969. Chronic manganese
poisoning: Individual susceptibility and absorption of iron. Neurology. 19:
1000-1006.
Mergler, D., G. Huel, R. Bowler et al. 1993. Nervous system dysfunction among
workers with long-term exposure to manganese. Environ. Res. (In press)
Miller, S.T., G.C. Cotzias, and H.A. Evert. 1975. Control of tissue
manganese: Initial absence and sudden emergence of excretion in the neonatal
mouse. Am. J. Physiol. 229(4): 1080-1084.
Moore, W. , D. Hysell, R. Miller et al. 1975. Exposure of laboratory animals
to atmospheric manganese from automotive emissions. Environ. Res. 9: 274-284.
Morganti, J.B., B.A. Lown, C.H. Stineman, R.B. D'Agostino, and E.J. Massaro.
1985. Uptake, distribution and behavioral effects of inhalation exposure to
manganese (Mn02) in the adult mouse. Neurotoxicology. 6(1): 1-16.
Murphy, V.A., J.M. Rosenberg, Q.R. Smith, and S.I. Rapoport. 1991. Elevation
of brain manganese in calcium-deficient rats. Neurotoxicology. 12: 255-264.
MAS (National Academy of Sciences). 1973. Manganese. National Academy of
Sciences Printing and Publishing Office, Washington, DC.
Nelson, K., J. Golnick, T. Korn, and C. Angle. 1993. Manganese
encephalopathy: Utility of early magnetic resonance imaging. Br. J. Ind. Med.
-------
RfC-20
ManganeBeItEr.ERgHcr CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC)
50: 510-513.
Newland, M.C., C. Cox, R. Hamada, G. Oberdoerster, and B. Weiss. 1987. The
clearance of manganese chloride in the primate. Fund. Appl. Toxxcol. 9s
314-328.
Nogawa, K., E. Kobayashi, M. Sakamoto et al. 1973, [Studies of the effects on
the respiratory organs of air pollution consisting of dusts composed mainly of
manganese. (First report). Effects on the respiratory organs of junior high
school students], Nippon Koshu Eisei Zasshi. 20(6)$ 315-325.
Oberdoerster, G. and G. Cherian. 1988. Manganese. In5 Biological monitoring
of toxic metals, T.W. Clarkson, L. Friberg, G.F. Nordberg, and P.R. Sager, eds.
Plenum Press, New York, NY. p. 283-301.
Perl, D.P. and P.F. Good. 1987. Uptake of aluminum into central nervous
system along nasal-olfactory pathways. Lancet. (8540): 1028.
Rodier, J. 1955. Manganese poisoning in Moroccan miners. Br. J. Ind. Med.
12s 21-35.
Roels, H. 1993. Correspondence from H. Roels, Faculte de Medecine, Unite de
foxicologie et Medecine du Travail, Catholigue Universite de Louvain, Clos
Chapelle-aux-Champs 30, BTE 3054, 1200 Bruxelles, Belgium, to J. Michael Davis,
Environmental Criteria and Assessment Office (MD-52), U.S. EPA, Research
Triangle Park, NC 27711, October 19.
Roels, H., R. Lauwerys, J.-P. Buchet et al. 1987. Epidemiological survey
among workers exposed to manganese! Effects on lung, central nervous system,
and some biological indices. Am. J. Ind. Med. 11: 307-327.
Roels, H.A., P. Ghyselen, J.p. Buchet, E. Ceulemans, and R.R. Lauwerys. 1992.
Assessment of the permissible exposure level to manganese in workers exposed to
manganese dioxide dust. Br. J. Ind. Med. 49s 25-34.
Saric, M. and S. Lucic-Palaic. 1977. Possible synergism of exposure to
airborne manganese and smoking habit in occurrence of respiratory symptoms. In:
Inhaled Particles, IV, W.H. Walton, ed. Pergamon Press, New York, NY. p.
773-779.
Saric, M., A. Markicevic, and 0. Hrustic. 1977. Occupational exposure to
manganese. Br. J. Ind. Med. 34: 114-118.
Schuler, P., H. Oyanguren, V. Maturana et al. 1957. Manganese poisoning:
Environmental and medical study at a Chilean mine. Ind. Med. Surg. 26:
167-173.
Seth, P.K. and S.V. Chandra. 1984. Neurotransmitters and neurotransmitter
receptors in developing and adult rats during manganese poisoning.
Neurotoxicology. 5(1): 67-76.
Shiotsuka, R.N. 1984. Inhalation toxicity of manganese dioxide and a
rmagnesiuin oxide-manganese dioxide mixture. Inhalation Toxicology Facility,
Brookhaven National Laboratory, Upton, NY. BNL 35334.
-------
Manganese RfC-21
*-=^= REFERENCE CONCENTRATION FOR CHRONIC INHALATION EXPOSURE (RfC) "
Siegl, von P. and K.-D. Bergert. 1982. Eine fruhdiagnostische
Uberwachsungsmethode bei Manganexposition. Z. Ges. Hyg. 28: 524-526.
Sjoegren, B., P. Gustavsson, and C. Hogstedt. 1990. Neuropsychiatric symptoms
among welders exposed to neurotoxic metals. Br. J. Ind. Med. 47s 704-707.
.Smyth, L.T., R.C. Ruhf, N.E. Whitman, and T. Dugan. 1973. Clinical manganism
jand exposure to manganese in the production and processing of ferromanganese
Alloy. J. Occup. Med. 15: 101-109.
Suzuki, Y., N. Fujii, H. Yano, T. Ohkita, A. Ichikawa, and K. Nishiyama. 1978.
Effects of the inhalation of manganese dioxide dust on monkey lungs.
Tokushima J. Exp. Med. 25: 119-125.
Tanaka, S. and J. Lieben. 1969. Manganese poisoning and exposure in
Pennsylvania. Arch. Environ. Health. 19s 674-684.
Dlrich, C.E., W. Rinehart, and W, Busey. 1979a. Evaluation of the chronic
Inhalation toxicity of a manganese oxide aerosol. I—Introduction,
experimental design, and aerosol generation methods. Am. Ind. Hyg. Assoc. J.
40: 238-244.
Dlrich, C.E., W. Rinehart, W. Busey, and M.A. Dorato. 1979b. Evaluation of
the chronic inhalation toxicity of a manganese oxide aerosol. II—Clinical
observations, hematology, clinical chemistry and histopathology. Am. Ind. Hyg.
Assoc. J. 40: 322-329.
Dlrich, C.E., W. Rinehart, and M. Brandt. 1979c. Evaluation of the chronic
inhalation toxicity of a manganese oxide aerosol. Ill—Pulmonary function,
electromyograms, limb tremor, and tissue manganese data. Am. Ind. Hyg. Assoc.
J. 40s 349-353.
U.S. EPA. 1984. Health Assessment Document for Manganese. Final Report.
Office of Health and Environmental Assessment, Environmental Criteria and
Assessment Office, Cincinnati, OH. EPA/600/8-83/013F.
Wang, J.-D., C.-C. Huang, Y.-H. Hwang, J.-R. Chiang, J.-M. Lin, and J.-S. Chen.
1989. Manganese induced parkinsonismi An outbreak due to an unrepaired
ventilation control system in a ferromanganese smelter. Br. J. Ind. Med. 46:
856-859.
Wennberg, A., A. Iregren, G. Struwe, 6. Cizinsky, M. Hagman, and L. Johansson.
1991. Manganese exposure in steel smelters a health hazard to the nervous
system. Scand. J. Work Environ. Health. 17: 255-262.
Wennberg, A., M. Hagman, and L. Johansson. 1992. Preclinical
neurophyBiological signs of parkinsonism in occupational manganese exposure.
Heurotoxicology. 13s 271-274.
Wilson, D.C., T.R.J. Tubman, H.L. Halliday, and D. McMaster. 1991. Plasma
manganese levels in the very low birth weight infant are high in early life.
Biol. Neonate. 61: 42-46.
Wolters, E.Ch., C.-C. Huang, C. Clark et al. 1989. Positron emission
tomography in manganese intoxication. Ann. Neurol. 26(5): 647-650.
------- |