ESTIMATION OF CARBON MONOXIDE
EXPOSURES AND ASSOCIATED
CARBOXYHEMOGLOBIN LEVELS IN DENVER
RESIDENTS USING A
PROBABILISTIC VERSION OF NEM
by
Ted Johnson, Jim Capel, Roy Paul, and Luke Wijnberg
International Technology Air Quality Services
South Square Corporate Centre One
3710 University Drive, Suite 201
Durham, North Carolina 27707-6208
Contract No. 68-DO-0062
Work Assignment No. 1-4
JTN 830013-013-02
Thomas McCurdy, Project Officer
Richard B. Atherton, Project Manager
Prepared for
U.S. ENVIRONMENTAL PROTECTION AGENCY
OFFICE OF AIR QUALITY PLANNING AND STANDARDS
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27711
July 1992
-------
DISCLAIMER
The statements and conclusions presented in this report are those of the
contractor. They do not necessarily reflect policies of the U.S. Environmental
Protection Agency.
-------
CONTENTS
Figures v
Tables vi
Acknowledgment viii
1. Introduction 1
2. Overview of the Methodology 3
Define study area, population-of-interest,
subdivisions of study area, and exposure period 3
Divide the population-of-interest into an exhaustive
set of cohorts 6,
Develop an exposure event sequence for each cohort for
the exposure period 8
Estimate the pollutant concentration, ventilation rate, and
COHb level associated with each exposure event 14
Extrapolate the cohort exposures to the population-of-interest
and to individual sensitive groups 26
3. The Mass-Balance Model 32
Overview of the model 32
The air exchange rate algorithm 36
Air exchange rate.distributions 39
Probability of stove use 41
Gas stove emission rate 46
Residential volume 48
4. The Carboxyhemoglobin Algorithm 49
The physiological profile 50
The COHb parameters 53
5. Exposure Estimates for Denver Residents 56
-------
CONTENTS (continued)
6. Initial Efforts to Validate the Exposure Model 64
7. Discussion and Recommendations 71
References 74
Appendices
A. Denver Fixed-Site Monitors A-1
B. Assignment of Diary Activity Codes to Activity Classifications B-1
C. The Carboxyhemoglobin Module C-1
IV
-------
FIGURES
Number Page
1 Page from the activity diary used in the Cincinnati study 10
2 Page from the activity diary used in the Denver and Washington
studies 11
3 Cumulative distributions of 1-hour daily maximum
carbon monoxide exposures measured during the Denver
Personal Monitoring Study and simulated by pNEM/03
(persons residing in homes with gas stoves) 65
4 Cumulative distributions of 1-hour daily maximum
carbon monoxide exposures measured during the Denver
Personal Monitoring Study and simulated by pNEM/CO
(persons residing in homes without gas stoves) 66
5 Cumulative distributions of 8-hour daily maximum
carbon monoxide exposures measured during the Denver
Personal Monitoring Study and simulated by pNEM/CO
(persons residing in homes with gas stoves) 67
6 Cumulative distributions of 8-hour daily maximum
carbon monoxide exposures measured during the Denver
Personal Monitoring Study and simulated by pNEM/CO
(persons residing in homes without gas stoves) 68
-------
TABLES
Number Page
1 Summary Statistics for Hourly Average Carbon Monoxide Values
Reported by Five Denver Monitoring Sites During 1988 5
2 Exposure Districts Defined for the Denver pNEM/CO Analysis 6
3 Demographic Groups Defined for the Denver pNEM/CO Analysis
and Number of Associated Cohorts 7
4 Microenvironments Defined for the Denver pNEM/CO Analysis 13
5 Percent Distribution of Breathing Rate Categories Reported
by Cincinnati Subjects by Activity Classification 14
6 Initial Values for Previous Outdoor Carbon Monoxide
Concentration and Upper Bounds of Indexing Intervals 17
7 Parameter Values of Lognormal Distributions Used to Characterize
Equivalent Ventilation Rate in Denver pNEM/CO Analysis 25
8 Percentage of Persons Wwith Ischemic Heart Disease (IHD) by
Demographic Group 31
9 Distributions of Parameter Values Used in the pNEM/CO Mass
Balance Model 37
10 Percentage of Person-Days With Indicated Window Ratio by Air
Conditioning System and Temperature Range 39
11 Statistics on Gas Stove Use Obtained from a Survey
by Koontz et al. 42
12 Probablity of Gas Stove Use by Clock Hour and Assumed
Burner Operation Period 43
VI
-------
TABLES (continued)
Number Page
13 Proportion of PEM Values in Gas Stove Residences with Stove
in Operation by Clock Hour and Work Status 45
14 Coefficient Values for Carboxyhemoglobin Algorithm 52
15 Number of Person-Days in Which a Denver Adult with Ischemic
Heart Disease Experiences a 1-Hour Daily Maximum Carbon
Monoxide Exposure at or Above Specified Concentration
Under Each of Four Scenarios 57
16 Number of Person-Days in Which a Denver Adult with Ischemic
Heart Disease Experiences an 8-Hour Daily Maximum Carbon
Monoxide Exposure at or Above Specified Concentration Under
Four Scenarios 57
17 Number of Person-Hours in Which a Denver Adult with Ischemic
Heart Disease Experiences an End-of-Hour Caboxyhemoglobin
Level at or Above Specified Percentage Under Each of Four
Scenarios 58
18 Number of Person-Days in Which a Denver Adult with Ischemic
Heart Disease Experiences a 1-Hour Daily Maximum Carbon
Monoxide Exposure at or Above Specified Concentration for
Each of Three Model Runs Under Scenario 1 (Existing Conditions,
Sources On) 61
19 Number of Person-Days in Which a Denver Adult with Ischemic
Heart Disease Experiences an 8-Hour Daily Maximum Carbon
Monoxide Exposure at or Above Specified Concentration for
Each of Three Model Runs Under Scenario 1 (Existing Conditions,
Sources On) 61
20 Number of Person-Hours in Which a Denver Adult with Ischemic
Heart Disease Experiences an End-of-Hour Carboxyhemoglobin
Level at or Above Specified Percentage for Each of Three
Model Runds Under Scenario 1 (Existing Conditions, Sources On) 62
VII
-------
ACKNOWLEDGMENT
This report describes a probabilistic version of NEM applicable to carbon
monoxide (pNEM/CO). The model consists of three principal parts: a main program
which estimates carbon monoxide exposures within a defined population, a special
module which estimates the carboxyhemoglobin levels that result from these
exposures, and a program which tabulates the exposure and carboxyhemoglobin
estimates. Supplementary programs are used to process air quality and population
data for input into the main program.
The main pNEM/CO program, the tabulation program, and all supplementary
programs were developed by IT Air Quality Services (ITAQS) under the direction of the
Ambient Standards Branch of the U.S. Environmental Protection Agency (EPA). Mr.
Ted Johnson of ITAQS developed the general methodology for the model as
described in Section 2 of this report. He is also the principal author of this report. Mr.
Roy Paul served as project manager for the ITAQS effort and was responsible for
assembling the population data required by the main program. Mr. Paul also wrote
the tabulation program. Dr. Louis Wijnberg performed the statistical analyses used to
relate outdoor carbon monoxide concentrations to fixed-site monitor readings (Table
6). Jim Capel wrote the main pNEM/CO program, the carboxyhemoglobin module
program, and the majority of the supplementary programs. Dr. Eduardo Olaguer
performed a series of quality assurance checks on the pNEM/CO software. Mr. John
Carroll conducted a review of the scientific literature concerning air exchange rates,
residential volumes, and gas stove emission factors.
The carboxyhemoglobin module is based on algorithms developed by Dr.
William F. Biller under a separate contract with the Ambient Standards Branch. Dr.
Biller also developed the hourly average version of the mass balance model.
ITAQS work on this project was funded under EPA Contract No. 68-DO-0062.
Mr. Thomas McCurdy served as the EPA Work Assignment Manager and provided
guidance throughout the project. Mr. Richard Atherton was the EPA Project Officer.
Mr. Harvey Richmond of EPA assisted in developing the carboxyhemoglobin algorithm
and in characterizing the distributions of the variables contained in the algorithm. He
also developed the estimates of the prevalence of ischemic heart disease appearing in
Table 8.
The authors would like to express their appreciation to Mr. John Irwin of EPA's
Atmospheric Research and Exposure Assessment Laboratory and to Mr. Joseph
Somers of EPA's Office of Mobile Sources for their many helpful recommendations.
VIII
-------
SECTION 1
INTRODUCTION
Within the U.S. Environmental Protection Agency (EPA), the Office of Air Quality
Planning and Standards (OAQPS) has responsibility for establishing and revising
National Ambient Air Quality Standards (NAAQS). In evaluating alternative NAAQS
proposed for a particular pollutant, OAQPS assesses the risks to human health of air
quality meeting each of the standards under consideration.1 This assessment of risk
requires estimates of the number of persons exposed at various pollutant
concentrations for specified periods of time. The estimates may be specific to an
urbanized area such as Los Angeles or apply to the entire nation.
Several researchers2'3 have recommended that such estimates be obtained by
simulating the movements of people through zones of varying air quality so as to
approximate the actual exposure patterns of people living within a defined area.
OAQPS has implemented this approach through an evolving methodology referred to
as the NAAQS Exposure Model (NEM). An early overview of the NEM methodology is
provided in a paper by Biller et al.4 From 1979 to 1988, IT Air Quality Services
(formerly PEI Associates, Inc.) assisted OAQPS in developing and applying pollutant-
specific versions of NEM to ozone,5 particulate matter,6 and CO.7 These versions of
NEM are referred to as "deterministic" versions in that no attempt was made to model
random processes within the exposure simulation.
The deterministic versions of NEM were similar in that each was capable of
simulating the movements of selected segments of an urban population through a set
of environmental settings. Each environmental setting was defined by a geographic
area and a microenvironment. The size and distribution of the geographic areas were
determined according to the ambient characteristics of the pollutant. Ambient
-------
(outdoor) pollutant levels in each geographic area were estimated from either fixed-site
monitoring data or dispersion model estimates. To better utilize fixed-site monitoring
data, researchers developed special time series techniques to fill in missing values and
special roll-back techniques to adjust the monitoring data to simulate conditions under
attainment of a particular NAAQS.
The population of interest in each study area was divided into an exhaustive set
of cohorts, and an activity pattern was developed for each cohort. The activity pattern
assigned the cohort to a geographic location and a microenvironment for each time
interval of a defined exposure period. In early NEM analyses, the time interval was 1
hour; in later NEM analyses, the time interval was reduced to 10 minutes. Exposure
periods varied from three months to one year. The activity patterns were based on
reviews of time use surveys. Researchers estimated the number of persons
represented by each cohort by accessing census and commuting data at the census-
tract level.
The pollutant concentration in a particular microenvironment was estimated by a
linear model which included terms relating to the ambient pollutant level and to
emission sources within the microenvironment. Researchers developed both point
estimates and distributions for the parameter values of these linear models by
conducting comprehensive reviews of the scientific literature associated with each
pollutant.
Additional details concerning the evolution of the deterministic version of NEM
are provided by Paul et al.8 Critiques of deterministic MEM are included in surveys of
exposure models by Pandian9 and Ryan.10 Two staff papers11'12 prepared by EPA
discuss the use of NEM in evaluating proposed NAAQS for CO and ozone.
In 1988, work was begun to incorporate probabilistic elements into the NEM
methodology and to apply the resulting model (pNEM) to ozone and carbon
monoxide. This report describes a version of pNEM specific to CO and its application
to Denver, Colorado. A version of pNEM applicable to ozone has been described by
Johnson et al.13
-------
SECTION 2
OVERVIEW OF THE METHODOLOGY
The general NEM methodology consists of five steps.
1. Define a study area, a population-of-interest, appropriate subdivisions of
the study area, and an exposure period.
2. Divide the population-of-interest into an exhaustive set of cohorts.
3. Develop an exposure event sequence for each cohort for the exposure
period.
4. Estimate the pollutant concentration, ventilation rate, and physiological
indicator (if applicable) associated with each exposure event.
5. Extrapolate the cohort exposures to the population-of-interest and to
individual sensitive groups.
This approach has been followed in developing a probabilistic version of NEM
applicable to CO (pNEM/CO). To illustrate the concepts incorporated in pNEM/CO,
the application of the methodology to Denver, Colorado, is described in the remainder
of this section.
2.1 Define Study Area, Population-of-lnterest, Subdivisions of Study Area, and
Exposure Period
The pNEM/CO methodology provides estimates of the distribution of CO
exposures and associated carboxyhemoglobin (COHb) levels within a defined
population (the population-of-interest) for a specified exposure period. The exposure
period is usually a recent calendar year for which good data are available with respect
to ambient CO levels. The population-of interest is typically defined as all residents of
a defined residential zone. Persons living in the residential zone are assumed to work
-------
within a larger area, referred to as the commuting zone. The residential and
commuting zones are both defined as contiguous sets of census tracts. The
residential and commuting zones are subdivided into a set of districts such that the
ambient CO concentration in each district can be estimated by CO data provided by
one or more fixed-site monitors selected as representative of the district.
In the application of pNEM/CO to Denver, the residential zone was defined as
the union of Denver, Adams, Arapahoe, and Jefferson counties. The commuting zone
included all of the residential zone plus all additional census tracts within 50 km of a
"city center" point located within the central business district of Denver. In 1980, the
residential zone contained 1,138,141 residents and was divided into 340 census tracts.
The commuting zone contained 393 census tracts.
The exposure period was defined as calendar year 1988. Five fixed-site CO
monitors reported 1988 data meeting the minimum completeness criterion established
for NEM analyses (75% complete). These sites are listed in Table 1 where they are
identified by the codes A, B, C, L, and M. Site descriptions, addresses, and EPA
identification numbers are provided in Appendix A. Each of the sites reported at least
8190 (93 percent) of the possible 8784 hourly-average concentration values for 1988.
The residential and commuting zones were subdivided into a set of seven
districts labeled A, B, C, L, M, X, and Y. Six of the districts (A through X) were
common to both the residential and commuting zones; one district (Y) was located
only within the commuting zone. Table 2 provides descriptive information concerning
each district.
The districts labeled A through M were each associated with the fixed-site
monitor of the same name. These districts were constructed by assigning each
census tract within the residential zone to the nearest monitor, given that the distance
between census tract centroid and monitor site did not exceed 10 km. The census
tracts in the residential zone not assigned to a monitor were assigned to District X.
The census tracts in the commuting zone outside the residential zone were assigned
to District Y. This procedure assigned each census tract within 50 km of the city
center to one and only one district.
-------
TABLE 1. SUMMARY STATISTICS FOR HOURLY AVERAGE CARBON MONOXIDE VALUES REPORTED BY FIVE
DENVER MONITORING SITES DURING 1988
Map
code
A
B
C
L
M
Number
of 1-h
values
8668
8649
8448
8659
8190
Carbon monoxide concentration, ppm
Arithmetic
Mean
2.19
1.65
1.93
1.20
0.53
Std.
Dev.
2.80
1.82
1.82
1.52
0.56
Geometric
Mean
1.57
1.06
1.34
0.70
0.40
Std.
Dev.a
2.35
2.55
2.40
2.68
1.93
Percentiles
10
0.5
0.3
0.3
0.3
0.3
30
1.1
0.7
0.9
0.3
0.3
50
1.7
1.1
1.4
0.6
0.3
70
2.5
1.7
2.1
1.2
0.6
90
4.3
3.7
4.0
2.9
1.0
95
5.4
5.2
5.3
4.3
1.4
Maximum
50.5
18.0
22.9
17.9
9.8
"Dimensionless.
-------
TABLE 2. EXPOSURE DISTRICTS DEFINED FOR THE DENVER pNEM/CO
ANALYSIS
Exposure
district
A
B
C
L
M
X
Y
Home
district?
Yes
Yes
Yes
Yes
Yes
Yes
No
Work
district?
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Number of
census
tracts
37
50
83
43
31
96
53
Description of district
Census tracts within 10
km of Monitor A
Census tracts within 10
km of Monitor B
Census tracts within 10
km of Monitor C
Census tracts within 10
km of Monitor L
Census tracts within 10
km of Monitor M
Census tracts in
Residential Zone not
included in Districts A
through M
Census tracts in
Commuting Zone not
included in Districts A
through X
2.2 Divide the Population-of-lnterest Into an Exhaustive Set of Cohorts
In a pNEM analysis, the population-of-interest is divided into a set of cohorts
such that each person is assigned to one and only one cohort. Each cohort is
assumed to contain persons with identical exposures during the specified exposure
period. Cohort exposure is typically assumed to be a function of demographic group,
location of residence, and location of work place. Specifying the home and work
district of each cohort provides a means of linking cohort exposure to ambient CO
concentrations. Specifying the demographic group provides a means of linking cohort
exposure to activity patterns that vary with age, work status, and other demographic
variables. In some analyses, cohorts are further distinguished according to factors
-------
relating to proximity to emission sources or to time spent in particular
microenvironments.
In the Denver pNEM/CO analysis, each cohort was identified as a distinct
combination of 1) home district, 2) demographic group, 3) work district (if applicable),
and 4) residential cooking fuel. The home district and work district of each cohort
were identified according to the districts defined above. Table 3 lists 11 demographic
groups defined for the Denver pNEM/CO analysis. Four of the demographic groups
are identified as workers. Each cohort associated with one of these groups was
identified by both home and work district. The remaining cohorts were identified only
by home district.
TABLE 3. DEMOGRAPHIC GROUPS DEFINED FOR THE DENVER pNEM/CO
ANALYSIS AND NUMBER OF ASSOCIATED COHORTS
Demographic group
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
Children, 0 to 17
Males, 18 to 44, working
Males, 18 to 44, nonworking
Males, 45 to 64, working
Males, 45 to 64, nonworking
Males, 65+
Females, 18 to 44, working
Females, 18 to 44, nonworking
Females, 45 to 64, working
Females, 45 to 64, nonworking
Females, 65+
Includes commuting
cohorts?
No
Yes
No
Yes
No
No
Yes
No
Yes
No
No
Total
Number of cohorts
associated with
demographic group
12
84
12
84
12
12
84
12
84
12
12
420
The residential cooking fuel of each cohort was identified as either "natural gas"
or "other." This cohort index was used because a personal monitoring study14
-------
conducted in Denver suggested that proximity to operating natural gas stoves
contributed significantly to CO exposure.
Table 3 lists the number of cohorts associated with each demographic group.
Each of the seven nonworking demographic groups is associated with 12 cohorts, one
for each combination of home district and residential cooking fuel. Each of the four
working demographic groups is associated with 84 cohorts, one for each combination
of home district, work district, and residential cooking fuel. The total number of
cohorts is thus (7x12) + (4 x 84) or 420.
2.3 Develop an Exposure Event Sequence for Each Cohort for the Exposure
Period
In the pNEM/CO methodology, the exposure of each cohort is determined by
an exposure event sequence (EES) specific to the cohort. Each EES consists of a
series of events with durations from 1 to 60 minutes. To permit the analyst to
determine average exposures for specific clock hours, the exposure events are
defined such that no event falls within more than one clock hour. Each exposure
event assigns the cohort to a particular combination of geographic area and
microenvironment. In addition, each event specifies whether or not the cohort is in the
presence of smokers. Each event also provides an indication of respiration rate. In
typical applications, this indicator is a classification of slow, medium, or fast.
In typical pNEM applications, the EESs are determined by assembling activity
diary records relating to individual 24-hour periods into a year-long series of records.
Because each subject of a typical activity diary study provides data for only a few
days, the construction of a year-long EES requires either the repetition of data from
one subject or the use of data from multiple subjects. The latter approach is used in
pNEM analyses to better represent the variability of exposure that is expected to occur
among the persons included in each cohort.
In the Denver pNEM/CO analysis, activity diary data obtained from studies
conducted in Cincinnati,16 Denver,15 and Washington, D.C.16 were assembled into a
special three-city data base. The data base consisted of diary records organized by
8
-------
study subject and 24-hour time period. The time periods were defined as starting at
7:00 p.m. on one day and ending at 7:00 p.m. the next day. The 7:00 p.m. to 7:00
p.m. time period was selected to make maximum use of available diary data from the
three studies. All three studies used 7:00 p.m. as a nominal start time for filling out
activity diaries. Figure 1 presents a page from the Cincinnati diary. Figure 2 presents
a page from the diary used in the Denver and Washington studies.
The diary records for one subject for one 24-hour period were designated a
"person-day." The three-city data base contained 3568 person-days, each of which
was indexed by the following factors:
1. Demographic group
2. Season: summer or winter
3. Temperature classification: cool or warm
4. Day type: weekday or weekend.
The demographic group index was determined by the demographic group to which
the subject filling out the diary belonged. The season and day type indices were
based on the end calendar date of the 7:00 p.m. to 7:00 p.m. time period. The
temperature classification was based on the daily maximum temperature of the
associated city (Cincinnati, Denver, or Washington) on that date. The cool range was
defined as temperatures below 55 in winter and temperatures below 84 in summer.
The EES for each cohort was determined by a computerized sampling
algorithm. The algorithm was provided with the sequence of daily maximum
temperatures reported for Denver in 1988 and with the list of cohorts. The
temperature data were used to assign each calendar day in 1988 to one of the
temperature ranges used in classifying the activity diary data. To construct the EES
for a particular cohort, the algorithm selected a person-day from the three-city data
base for each 7:00 p.m. to 7:00 p.m. time period in 1988 according to the
demographic group of the cohort and the season, day type, and temperature
classification associated with the time period.
-------
TIME
PM
A. ACTIVITY (please specify)
B. LOCATION
In transit, car 01
In transit, other vehicle . . 02
Specify
Indoors, your residence ... 03
Indoors, other residence. . . 04
Indoors, office 05
Indoors, manufacturing
facility 06
Indoors, school 07
Indoors, store 08
Indoors, other 09
Specify
Outdoors, within 10 yards of
road or street 10
Outdoors, other 11
Specify
Uncertain 12
*Enter MIDN for midnight and NOON for noon. Otherwise enter four-digit
time (e.g., 0930 for 9:30 and 1217 for 12:17) and check a.m. or p.m.
C. BREATHING RATE
Slow (e.g., sitting) 13
Medium (e.g., brisk walk). . . 14
Fast (e.g., running) 15
Breathing problem 16
Specify
D. SMOKING
I am smoking 17
Others are smoking 18
No one 1s smoking 19
E. ONLY IF INDOORS
(1) Fireplace in use?
Yes 20
No 21
(2) Woodstove in use?
Yes 22
No 23
(3) Windows open?
Yes 24
No 25
Uncertain. 26
Figure 1. Page from the activity diary used in the Cincinnati study.
10
-------
TIME FROM MONITOR
A. ACTIVITY
B. LOCATION
In transit 1
Indoors, residence 2
Indoors, office ........ 3
Indoors, store 4
Indoors, restaurant 5
Other Indoor location 6
• Specify:
Outdoors, within 10 yards of road
or street 7
Other outdoor location .... 8
Specify:
Uncertain 9
C. ADDRESS (1f not In transit)
0. ONLY IF IN TRANSIT
(1) Start address _
(2) End address
(3) Mode of travel:
Walking . . .
Car ......
Bus ......
Truck
Train/subway
Other . . . .
Specify:
1
2
3
4
5
6
E. ONLY IF INDOORS
(1) Garage attached to building?
Yes 1
No 2
Uncertain ........ 3
(2) Gas stove 1n use?
Yes
No
1
2
Uncertain . . 3
ALL LOCATIONS
Smokers present?
Yes
No .......
Uncertain . . . .
1
2
3
Figure 2. Page from the activity diary used in the Denver
and Washington studies.
11
-------
Each exposure event within an EES was defined by 1) district, 2)
micro-environment, 3) breathing rate category, and 4) passive smoking status. The
district was either the home or the work district associated with the cohort. The
home/work determination was based on a decision rule which was applied to the
activity diary data associated with the exposure event.
Table 4 lists the 13 microenvironments used for event assignments. Each
microenvironment is identified as to a general location (e.g., outdoors) and a specific
location (e.g., near road). The list includes two indoor microenvironments related to
residences, five indoor microenvironments related to nonresidential buildings, three
outdoor microenvironments, and three vehicle microenvironments. The majority of
these microenvironments are aggregates of two or more location descriptions used in
the activity diary studies. Only location descriptions associated with similar average
CO exposures were combined in defining the aggregate microenvironments. Personal
CO monitoring data obtained during the Denver activity diary study14 were used to
determine an average CO exposure for each location description.
Four breathing rate categories were defined: slow - sleeping, slow - awake,
medium, and fast. Diary data for determining breathing rate category directly were
available for the Cincinnati study only. Consequently, a Monte Carlo procedure was
used to assign breathing rate categories to exposure events associated with the
Denver and Washington activity diaries.
Assignment probabilities varied according to an activity classification assigned
to each exposure event for this purpose. The activity classification was A, B, C, or D,
with A indicating the most strenuous class of activities and D indicating the least
strenuous class. Each of the codes used in the three activity diary studies to
characterize the diary entries under "Activity" was assigned to one of the four activity
classifications. Classification D was reserved for sleeping activities. Appendix B lists
these assignments.
12
-------
TABLE 4. MICROENVIRONMENTS DEFINED FOR THE DENVER pNEM/CO ANALYSIS
Microenvironment
Code
1
2
3
4
5
6
7
8
9
10
11
12
99
General
location
Indoors
Indoors
Indoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Outdoors
Vehicle
Vehicle
Outdoor
Vehicle
Specific
location
Residence
Nonresidence A
Nonresidence B
Nonresidence C
Nonresidence D
Nonresidence E
Residential
garage
Near road
Other locations.
Automobile
Other
Public parking
or fueling
facility
Airplane
Activity diary locations included in
microenvironment
Indoors-residence
Service station or auto repair
Other repair shop
Shopping mall
Restaurant
Other indoor location
Auditorium
Store
Office,
Other public building
Health care facility
School
Church
Manufacturing facility
Residential garage
Near road
Outdoor residential garage
Construction site
Residential grounds
School grounds
Sports arena
Park or golf course
Other outdoor location
Automobile
Bus
Truck
Bicycle
Motorcycle
Train/subway
Other vehicle
Indoor parking garage
Outdoor parking garage
Outdoor parking lot
Outdoor service station
Airplane
13
-------
An analysis was performed on the Cincinnati diary data to determine the
fraction of time that subjects reported breathing rates of slow, medium, or fast within
classifications A, B, and C (Table 5). These fractions were used to randomly assign a
breathing rate category to each nonsleepina exposure event derived from the Denver
and Washington studies. Sleeping events were always assigned a breathing rate
category of "slow - sleeping."
TABLE 5. PERCENT DISTRIBUTION OF BREATHING RATE CATEGORIES
REPORTED BY CINCINNATI SUBJECTS BY ACTIVITY CLASSIFICATION
Activity
classification
A
B
C
Percent of time spent in indicated breathing rate
category
Slow-awake
39.2
85.9
98.1
Medium
47.9
13.7
1.9
Fast
12.9
0.4
0
The effects of active smoking on CO exposure were not addressed in the
exposure analysis described here. Because of the coding conventions used in the
diary studies, passive smoking patterns could be determined for nonsmoking subjects
only. Consequently, the activity diaries sampled in constructing EESs were limited to
those of nonsmokers. The diary record associated with each exposure event provided
information on whether or not the subject was in the presence of smokers (Figures 1
and 2). This information was used to assign a passive smoking status to each event.
2.4 Estimate the Pollutant Concentration, Ventilation Rate, and COHb Level
Associated With Each Exposure Event
In the general pNEM methodology, the EES defined for each cohort is used to
determine a corresponding sequence of exposures, event by event. Each exposure is
defined by a pollutant concentration and a ventilation rate indicator. In some
applications, a biokinetics model is used to determine the status of a physiological
indicator at the end of each exposure event, based on the status of the indicator at the
14
-------
beginning of each event and the pollutant dose delivered during the event. The
delivered dose is a function of the pollutant concentration and ventilation rate values
assigned to the event and the demographic characteristics of the cohort.
2.4.1 Estimation of Pollutant Concentration
In the pNEM/CO analysis, each exposure event within a particular EES was
indexed according to district d, microenvironment m, person-day p, clock hour h, and
start time t. The exposure associated with a particular event, CEXP(d,m,p,h,t) was
estimated by the expression
CEXP(d,m,p,h, t) = CME(d,m,p,h) + SMOKE(m, t) . (1)
CME(d,m,p,h) is the CO concentration determined for microenvironment m in district d
for person-day p and hour h. SMOKE(m,t) is an assumed contribution from passive
smoking specific to microenvironment and event.
The SMOKE(m,t) term represented the short-term contribution of passive
smoking to CO exposure. The operation of this term was a function of two event
descriptors: passive smoking status and microenvironment. If the passive smoking
status for an indoor event indicated the presence of smokers, the value of
SMOKE(m.t) was set equal to 1.6 ppm for the duration of the event. Otherwise,
SMOKE(m,t) was set equal to zero. The 1.6 ppm value is based on the average
increase in CO exposure observed in subjects of the Denver study14 during indoor
periods when smokers were present.
The CME(d,m,p,h) term represented the component of exposure contributed by
ambient (outdoor) CO concentrations and by the operation of residential gas stoves.
An array of CME values was created for each cohort. Each array consisted of a set of
year-long sequences of hourly-average CME values, one for each combination of
microenvironment and district. The district was either the home or the work district
specified for the cohort. When an exposure event occurring during hour h assigned a
cohort to a particular combination of microenvironment and district, the cohort was
15
-------
assigned the CO concentration specified for hour h in the designated
microenvironment/district sequence.
Each year-long sequence of hourly average CME values for the indoor and
motor vehicle microenvironments was generated by the mass-balance algorithm
described in Section 3. Briefly, this algorithm estimated the hourly average indoor CO
concentrations during hour h as a function of the indoor CO concentration during the
preceding hour (i.e., hour h -1), the CO concentration outdoors during hour h, the air
exchange rate during hour h, and the indoor emissions of CO from gas stoves during
hour h (if applicable). Values for the air exchange rate and gas stove emission rate
were sampled from appropriate distributions on a daily basis. During each clock hour,
gas stoves were probabilistically determined as "on" for 30 minutes, "on" for 60
minutes, or "off" for the entire hour. The probability of being on varied with time of day
according to use patterns observed during the Denver activity diary study.
The outdoor CO concentration required by the mass-balance algorithm was
determined for each hour through a Monte Carlo process. According to this process,
the outdoor CO concentration estimated for a particular hour h was influenced by 1)
the microenvironment, 2) the outdoor CO concentration estimated for the preceding
hour for the specified microenvironment, and 3) the CO concentration estimated for
hour h at the fixed-site monitor representing district d.
For a particular combination of microenvironment and district, the process
consisted of stepping through the hours in the calendar year and selecting an outdoor
CO concentration for each hour. The selection for a particular hour was made by
randomly selecting a value from one of a set of empirical distributions established for
the microenvironment. Two methods were employed in making these selections. In
Method A, the selection procedure accounted for both the current fixed-site CO
concentration and the outdoor CO concentration determined for the preceding hour.
In Method B, the selection procedure accounted only for the current fixed-site CO
concentration. Method A was applied to the four microenvironments listed in Table 6.
Method B was applied to the remaining microenvironments.
16
-------
TABLE 6. INITIAL VALUES FOR PREVIOUS OUTDOOR CARBON MONOXIDE
CONCENTRATION AND UPPER BOUNDS OF INDEXING INTERVALS
Microenvironment
Code
1
4
5
10
General
Location
Indoors
Indoors
Indoors
Vehicle
Specific
location
Residence
Nonresidence C
Nonresidence D
Automobile
Applicable
districts
A-M
X and Y
All
All
All
Initial
value8
1.1
0.5
3.1
2.3
5.0
Carbon monoxide concentration, ppm
Upper bound of indicated interval
1
0
0
1.2
0.5
1.8
2
1.1
0.5
3.1
2.3
5.0
3
3.1
1.7
6.1
4.8
10.0
4
6.0
4.1
10.6
8.3
16.3
5
b
b
b
b
b
aFor outdoor concentration during previous hour.
bNot bounded.
-------
Twentyfive empirical distributions were established for each of the Method A
microenvironments. Each distribution represented the distribution of outdoor CO
concentrations expected to occur for a defined pair of events: 1) the fixed-site CO
concentration falls within interval i and 2) the outdoor CO concentration of the previous
hour falls within interval j.
The following five intervals were established for fixed-site CO concentrations:
CO concentration, ppm
Interval (i) Lower bound Upper bound
1 0 1
2 1 2
324
44 8
5 8 na
The first interval includes 0 and 1. Each of the remaining intervals excludes the
specified lower bound and includes the specified upper bound.
Five intervals were established for the previous outdoor CO concentration for
each of the four microenvironments. Table 6 lists the upper bounds of these intervals.
Each distribution was expressed as a cumulative probability distribution.
Values were selected randomly from a distribution by first selecting a random number
between zero and one and then determining the CO concentration associated with the
random number on the cumulative distribution. For example, the random number
0.87 would select the concentration value associated with the 87th percentile of the
cumulative distribution.
The following algorithm was used to simulate a year-long sequence of hourly-
average outdoor CO concentrations for each of the Method A microenvironments.
1. Identify microenvironment and district associated with simulation.
2. Go to first hour.
3. Set previous outdoor CO concentration equal to initial value listed in
Table 6 for microenvironment.
18
-------
4. See lookup table (Table 6) for microenvironment to determine interval
index (j) for previous outdoor concentration.
5. Determine fixed-site CO concentration associated with hour for specified
district and microenvironment.
6. See lookup table to determine interval index (i) for fixed-site CO
concentration.
7. Locate cumulative distribution associated with indices i,j in supplemental
array. If no cumulative distribution has been determined for (i,j), go back
to previous hour and redo Steps 8-12 (requires drawing a new random
number in Step 8).
8. Select random number between zero and one. Compare random
number (RN) with probabilities listed in cumulative distribution identified in
Step 7. RN will fall within one of the specified intervals, such that
LBCF < RN < = UBCF
where LBCF is the lower bound cumulative fraction and UBCF is the
upper bound cumulative fraction.
9. Associated with LBCF is a lower bound pollutant concentration (LBPC).
Associated with the UBCF is an upper bound pollutant concentration
(UBPC). Use the following interpolation equation to estimate the current
outdoor CO concentration value.
CO = LBPC + (RN - LBCF) (UBPC - LBPC) / (UBCF - LBCF) (2)
10. Go to the next hour.
11. The current outdoor CO concentration as determined by Step 9 becomes
the previous outdoor CO concentration.
12. Go to Step 4.
The empirical distributions used in Step 7 were developed through a statistical
analysis of data obtained from the 1982-83 Denver activity diary study14. During this
study, each of approximately 450 subjects carried a personal exposure monitor (PEM)
for two 24-hour periods. Each PEM measured CO concentration continuously. The
PEM readings were averaged by exposure event such that each event was associated
19
-------
with a single microenvironment and a single clock hour. Event durations ranged from
one minute to one hour. The microenvironment assigned to each PEM was
determined from entries made in the subject's activity diary.
The goal of the statistical analysis was to develop distributions representing the
outdoor CO concentrations associated with each microenvironment. Originally, these
distributions were to be developed using only outdoor PEM values. The Denver study
was found to contain relatively few outdoor PEM values, however, and these values
were difficult to associate with specific indoor microenvironments.
An alternative approach was subsequently implemented in which each indoor
PEM value was assumed to represent the outdoor CO concentration at the same
location one hour earlier, given that the indoor PEM value was not measured during a
period when a gas stove was in operation. Consistent with this assumption, an indoor
PEM value reported for hour h was indexed according to a fixed-site monitor
concentration reported for h -1. For outdoor and motor vehicle microenvironments,
each PEM value was assumed to represent the current outdoor concentration.
Consequently, each PEM value was indexed according to the fixed-site concentration
associated with the current hour.
In indexing the PEM values according to fixed-site concentration, a distinction
was made between indoor PEM values which occurred within 10 km of at least one
fixed-site monitor and those which were not within 10 km of any monitor. In the first
case, the index was determined by the fixed-site concentration reported by the nearest
monitor. In the second case, the index was determined by the average of the
concentration values reported by all monitors.
For the outdoor microenvironments, the i index of each PEM value was
determined according to the CO concentration at the nearest monitor, given that the
PEM value was obtained within 10 km of the monitor. The average of the fixed-site
concentration values was used to index each of the motor vehicle microenvironments.
The PEM values reported by a particular subject of the Denver study consisted
of a sequence of CO concentrations which could be indexed by microenvironment.
Whenever a particular PEM value was preceded by a PEM value reported for the same
20
-------
same microenvironment, the value of the preceding PEM value was used to determine
the preceding hour index 0')- No index could be determined in cases where a
particular PEM value was not preceded by a PEM value associated with the same
microenvironment.
The initial result of this approach was a file listing PEM values indexed
according to microenvironment, fixed-site CO interval (i), and previous PEM CO
interval (j)- This file was subsequently analyzed to determine the distribution of PEM
values for each combination of Method A microenvironment, i value, and j value. The
resulting distributions were accessed in Step 7 of the 12-step procedure described
above.
Method B was applied to the nine microenvironments not listed in Table 6. In
each of these cases, the Denver PEM data available for the microenvironment were
judged to be insufficient for the purpose of accounting for both the current fixed-site
CO concentration and the preceding outdoor CO concentration. An analysis of the
activity diary data for these microenvironments indicated that people tended not to
occupy these microenvironments for long time periods. Consequently, the decision
was made to ignore the effects of the preceding outdoor concentration in the selection
process.
Five empirical distributions were established for each of the Method B
microenvironments, one for each of the five ranges used in Method B for classifying
fixed-site concentrations. The outdoor value selected for each hour was randomly
selected from one of these five distributions according to the fixed-site concentration
associated with the hour. Apart from this use of five rather than 25 empirical
distributions, Method B is identical to Method A.
Methods A and B each require a complete (gapless) year of hourly average
fixed-site monitoring values for each district. These data sets were prepared by
applying a special interpolation program to the 1988 hourly average CO data reported
by each of the five Denver fixed-site monitors. The interpolation program provided an
estimate of each missing value. The resulting filled-in data sets were assumed to
represent existing conditions at each monitor.
21
-------
The interpolation program provides estimates of missing values through the use
of a time series model developed by Johnson and Wijnberg17. The time series model
is based on the assumption that hourly average air quality values can be represented
by a combination of cyclical, autoregressive, and noise processes. The parameter
values of these processes are determined by a statistical analysis of the reported data.
2.4.2 The Rollback Model
The procedure described in Section 2.4.1 was used to determine a sequence of
1-hour outdoor concentrations for each combination of microenvironment and district.
These sequences were assumed to represent existing ("as is") outdoor air quality
conditions in Denver. To represent outdoor air quality under a particular regulatory
scenario, the concentration values in each sequence were adjusted according to a roll-
back model. This model can be expressed as
COUT(m,d,t,s) =BG + P(s) x CDIF(m,dtt,e)
t
where
d,t,e) = COUT(m,d,t,e) - BG,
COUTtm.d.t.s) is the the outdoor value expected at time t in microenvironment m in
district d under scenario s, COUT(m,d,t(e) is the outdoor value expected at time t in
microenvironment m in district d under existing conditions, BG is the assumed
background concentration which is riot affected by the control scenario, and p(s) is the
rollback factor specific to scenario s.
The value of /o(s) was calculated by the expression
p(s) = [CMAX(S) - BG]/[CMAX(e) - BG] (5)
when CDIF(m,d,t,e) was greater than 0 and by the expression
22
-------
p(s) = 1 (6)
when CDIF(m,d,t,e) was less than or equal to 0. In Equation 5, CMAX(s) is the
highest CO concentration permitted under scenario s for a specified air quality indictor
(AQI) and CMAX(e) is the value of this AQI based on 1988 Denver monitoring data.
This rollback model was used by Johnson and Paul7 in a previous application of
NEM to CO. The model is based on the assumption that outdoor CO concentrations
can be partitioned into two parts: a constant component representing the background
CO concentration and a varying component which is proportional to the CO emissions
that are permitted under scenario s. The magnitude of the latter component is
assumed to be proportional to p(s).
Two scenarios were considered in the exposure analyses described below:
existing conditions and attainment of the current 8-hour NAAQS for CO. The "existing
conditions" scenario was implemented by setting COUT(m,d,t,s) equal to
COUT(m,d,t,e). The attainment scenario was implemented by determining an
appropriate AQI for the rollback process, specifying the value of CMAX(s) for this AQI,
estimating CMAX(e) for this AQI, determining an appropriate value for BG, calculating
p(s) by Equations 5 and Q, and then using Equation 3 to adjust the filled-in 1988 CO
data for each Denver monitor.
The current NAAQS for CO specifies that the second highest 8-hour CO
concentration shall not exceed 9 ppm. Attainment of the NAAQS within an urbanized
area can be determined by evaluating an AQI defined as the largest value reported by
any monitor for the second highest 8-hour CO concentration of the year. Based on
1988 Denver monitoring data, the value of this AQI was 16.2 ppm.
To simulate NAAQS attainment in Denver, the variables CMAX(s) and CMAX(e)
were assigned the values 9.0 ppm and 16.2 ppm, respectively. The value of BG was
set at 0.53 ppm, the smallest annual average CO concentration reported by a Denver
monitoring site for 1988. The value of p(s) determined by Equation 5 for these values
was 0.54.
23
-------
2.4.3 Equivalent Ventilation Rate
In addition to CO concentration, an equivalent ventilation rate (EVR) value was
estimated for each exposure event. EVR is defined as ventilation rate divided by body
surface area (BSA). Clinical research by EPA suggests that EVR exhibits less inter-
person variability than ventilation rate for a given level of exertion.18
The algorithm used to estimate EVR was employed previously in an application
of the pNEM methodology to ozone.13 This algorithm is based on an analysis of
activity diary data provided by Dr. Jack Hackney.19 The data were obtained from 36
subjects in Los Angeles who completed activity diaries identical to those used in the
Cincinnati study. The heart rate of each subject was monitored during the period
reported in the diary. Separate clinical trials were conducted to determine a
relationship between ventilation rate and heart rate for each subject. These
relationships and subject-specific BSA values were used to convert the one-minute
heart rate data associated with each diary activity to an average EVR value for the
activity. The resulting EVR estimates were then grouped by breathing rate category
(slow - sleeping, slow - awake, medium, fast). Statistical analysis indicated that a two-
parameter lognormal distribution provided a good fit to the EVR values in each group.
Table 7 lists the geometric mean and standard deviation of each fitted distribution.
The appropriate distribution was randomly sampled to provide an EVR value for
each exposure event in the pNEM/CO simulation. EVR values were not permitted to
exceed an upper limit which varied with demographic group and event duration. This
limit was based on research on maximum energy expenditure limits reported by Erb20
and by Astrand and Rodahl.21
24
-------
TABLE 7. PARAMETER VALUES OF LOGNORMAL DISTRIBUTIONS USED TO CHARACTERIZE
EQUIVALENT VENTILATION RATE IN DENVER pNEM/CO ANALYSIS
Age group
Children
Adults
Breathing rate
Slow-sleeping
Slow-awake
Medium
Fast
Slow-sleeping
Slow-awake
Medium
Fast
Parameter values of fitted lognormal
distribution
Geometric mean8
8.1
10.0
12.3
14.8
5.4
7.1
8.6
18.9
Geometric
standard
deviation
1.60
1.46
1.44
1.62
1.22
1.36
1.34
1.92
aLiters/min per m.
2.4.4 Carbaxyhemoglobin Level
An algorithm developed by Biller and Richmond22 was used by pNEM/CO to
estimate the COHb level at the end of each exposure event. The algorithm is based
on a differential equation proposed by Coburn, Forster, and Kane.23 Inputs to the
algorithm include:
Percent COHb at the start of the event
Average CO exposure concentration over the event, ppm
Time duration of event, min
Alveolar ventilation rate, ml/min
Haldane Constant
Atmospheric pressure at sea level, torr
Altitude above sea level, feet
Blood volume, ml
Total hemoglobin content of blood, gm/100 ml
Pulmonary CO diffusion rate, ml/min per torr
Endogenous CO production rate, ml/min.
Section 4 describes the method used to estimate each of these quantities. Appendix
C provides a detailed description of the COHb algorithm and a list of references.
25
-------
2.4.5 Hourly Average Exposure Sequences
Algorithms within pNEM/CO provided four estimates for each exposure event:
average CO concentration, average EVR, the product of average CO concentration
and EVR (CO x EVR), and COHb at the end of the event. These estimates were
processed to produce time-weighted estimates of CO concentration, EVR, and CO x
EVR for each clock hour, as well as end-of-hour estimates of COHb. The result was a
year-long sequence of hourly values for CO, EVR, CO x EVR, and COHb for each of
the 420 Denver cohorts. These sequences can be statistically analyzed to determine
the value of various multihour exposure indicators. Examples include the largest eight-
hour daily maximum CO concentration and the number of times the end-of-hour COHb
concentration exceeded a specified percentage value.
2.5 Extrapolate the Cohort Exposures to the Population-of-lnterest and to
Individual Sensitive Groups
2.5.1 General Population
The cohort-specific exposure estimates developed in Step 4 of the pNEM
methodology (Section 2.4) were extrapolated to the general Denver population by
estimating the population size of each cohort. Cohort populations were estimated in
three steps. The population of each demographic group within a particular home
district [Pop(d.h)] was first estimated from census data specific to that district. Each
of these groups was subdivided into a group residing in homes with gas stoves and a
group residing in homes with other cooking fuels. The population of each of these
groups was determined by the expression
Pop(d,h,f) =F(h,£) xp0p(d,h), (7)
where Pop(d,h,f) is the population of a group associated with demographic group d,
home district h, and cooking fuel f. F(h,f) is the fraction of homes in home district h
that use cooking fuel f. F(h,f) was determined from BOG census data specific to each
district.24
26
-------
The Pop(d,h,f) values provided an estimate of the population of each non-
commuting cohort residing within home district h. The populations of the commuting
cohorts (assumed to include all working cohorts) were determined by the expression
Com(d,h,f,w) = Pop(d,h,f) x Com(h,w)/Work(h). (8>
Com(d,h,f,w) is the number of persons in the commuting cohort associated with
demographic group d, home district h, cooking fuel f, and work district w; Com(h,w) is
the number of workers in all demographic groups that commute from home district h
to work district w; and Work(h) is the total number of workers in home district h.
Estimates of Work(h) were developed from census data specific to each district.
Estimates of Com(h,w) were obtained from an origin-destination (O-D) table produced
by a special commuting model.
The commuting model has been described in detail by Johnson et al.25 Briefly,
the commuting model uses a trip duration model to develop an O-D table. Trip
duration data are collected during each census year for all areas of the U.S. The data
are reported as the number of persons in each census unit with one-way commute
times that fall into each of the following eight commute duration ranges:
1. Less than 5 minutes
2. 5 to 9 minutes
3. 10 to 14 minutes
4. 15 to 19 minutes
5. 20 to 29 minutes
6. 30 to 44 minutes
7. 45 to 59 minutes
8. 60+ minutes.
The model assumes that each commute duration range can be converted into a
corresponding range of geodesic distances. Geodesic distance is defined here as the
shortest distance between two points on the globe, i.e., the distance "as the crow
files."
For example, a commuter in a large urbanized area may report that her
commute takes between 20 and 29 minutes. If the average geodesic commute speed
27
-------
in the area is 0.3 kilometers per minute (km/min), the commute duration range of 20
to 29 minutes is equivalent to a geodesic distance range (GDR) of 6 to 8.7 km.
In a similar manner, each of the eight commute distance ranges is converted to
a GDR indexed as i = 1, 2 8. The location of each census unit is represented by
its geographic centroid. If the census unit is an origin location for commuters (i.e., a
"home" district), one can delineate eight concentric rings centered on the centroid, one
for each GDR. The i-th GDR centered on home district h is identified as GDR(h.i).
Other useful terms are defined below. 4
COM(h): Number of commuters residing in home district h
COM(h,i): Number of commuters residing in home district h who commute to
a work district in GDR(h,i)
COM(h,i,w): Number of commuters residing in home district h who commute to
work district w where work district w is in GDR(h.i)
N(h,i): Number of work districts in GDR(h.i)
Tot(h,i,w): Total number of commuters who work in work district w in
GDR(h.i)
Tot(h.i): Total number of commuters that work in GDR(h.i) (includes
commuters from all home districts).
The following method is employed to develop an 0-D table.
1. Com(h,i) and N(h,i) are known. Make an initial estimate of Com (h,i,w)
using the expression
Com(h,i,v) = [Com(h,i)]/[W(h,i)]
2. Calculate Tot(h,i,w) by summing Com(h,i,w) values associated with
specified w value
3. Calculate Tot(h.i) by summing Tot(h,i,w) values for all work districts in
GDR(h.i)
28
-------
4. Make a new estimate of Com(h,i,w) using the expression
Com(h,i,w) = [Com(h,i) ] [Tot(h,i,w)) ]/[Tot(h,i) ] (10)
5. Go to Step 2.
In Step 1, commuters associated with a particular combination of home district and
GDR are evenly distributed across the work districts in the GDR. This step is Iteration
0. Steps 2 through 4 are repeated n times where n is determined by the user. In
each iteration of Steps 2 through 4, the commuters are redistributed across the work
districts in the GDR in proportion to the number of workers assigned to each work
district during the last iteration.
The commuting model was initially applied to Denver using individual census
tracts for the specified home and work districts. Based on the 340 census tracts
specified as home districts and the 393 census tracts specified as work districts, this
approach produced a high-resolution 0-D table containing 133,620 Com(h.w)
estimates. These estimates were then aggregated according to the six home districts
and seven work districts specified for the Denver exposure analysis. The resulting
aggregated 0-D table was the source of the Com(h,w) values used in Equation 8.
A special tabulation program in pNEM/CO combined the cohort-specific
estimates of exposure and population to produce histograms and cumulative
frequency tables for various population exposure indicators and averaging times.
Section 5 provides exposure estimates based on existing conditions in Denver and the
attainment of the current 8 hour NAAQS.
2.5.2 Persons with Ischemic Heart Disease
The cohort-specific exposure estimates developed in Step 4 were also
extrapolated to the sensitive population defined as persons with diagnosed and
undiagnosed ischemic heart disease (IHD). The extrapolation was performed using
the procedure described in Subsection 2.5.1 with a single variation: the following
equation was substituted for Equation 7.
29
-------
Pop(d,h,f) = IHD(d) x F(h,f) x Pop(d.h)
The term IHD(d) is the fraction of persons in demographic group d with IHD.
Estimates of the prevalence of IHD by demographic group were provided by H.
Richmond26. Table 8 lists these estimates as percentages. In each case a total
prevalence rate is provided which is the sum of a prevalence rate for diagnosed IHD
and a prevalence rate for undiagnosed IHD. Estimates of diagnosed IHD were
obtained from the National Health Interview Survey27, in which U.S. prevalence rates
were disaggregated by age and gender. Estimates of undiagnosed IHD were based
on two assumptions: 1) there are 3.5 million persons in the U.S. with undiagnosed
IHD and 2) persons with undiagnosed IHD are distributed within the population in the
same manner as persons with diagnosed IHD. The 3.5 million statistic was based on
an estimate by the American Heart Association28 that there are between three and four
million persons with undiagnosed IHD.
30
-------
TABLE 8. PERCENTAGE OF PERSONS WITH ISCHEMIC HEART DISEASE
(IHD) BY DEMOGRAPHIC GROUP
Demographic group
1. Children, 0 to 17
2. Males, 18 to 44, working
3. Males, 18 to 44, nonworking
4. Males, 45 to 64, working
5. Males, 45 to 64, nonworking
6. Males, 65+
7. Females, 18 to 44, working
8. Females, 18 to 44, nonworking
9. Females, 45 to 64, working
10. Females, 45 to 64, nonworking
11. Females, 65+
Percentage of persons with IHD
Diagnosed
0.33
0.33
8.6
8.6
16.9
0.18
0.18
2.6
2.6
11.3
Undiagnosed
0.17
0.17
4.3
4.3
8.5
0.09
0.09
1.3
1.3
5.7
Total
0.03
0.50
0.50
12.9
12.9
25.3
0.27
0.27
3.8
3.8
17.0
31
-------
SECTION 3
THE MASS-BALANCE MODEL
The application of pNEM/CO to Denver marks a milestone in the evolution of
NEM in that it represents the first time that a mass balance model has been
incorporated directly into the NEM methodology. This section provides an overview of
the pNEM/CO mass balance model together with descriptions of the algorithms used
in the model to estimate air exchange rates and emissions from gas stoves.
3.1 Overview of the Model
The pNEM/CO methodology includes a mass balance model which is used to
estimate CO concentrations when a cohort is assigned to an indoor or motor vehicle
microenvironment. The mass balance model is based on the generalized mass
balance model presented by Nagda, Rector, and Koontz.29 This model can be
expressed by the differential equation
B out in
dt B out cv in cV cv
- FB) "cout + — - ""Cin - — - in <1X>
O' OUt .—T r in —T r -*T r
where Cin = Indoor concentration (units: mass/volume)
FB = Fraction of outdoor concentration intercepted by the enclosure
(dimensionless fraction)
v = Air exchange rate (1/time)
Cout = Outdoor concentration (mass/volume)
S = Indoor generation rate (mass/time)
32
-------
cV = Effective indoor volume where c is a dimensionless fraction
(volume)
m = Mixing factor (dimensionless fraction)
\ = Decay rate (mass/time)
q = Flow rate through air cleaning device (volume/time)
F = Efficiency of the air cleaning device (dimensionless fraction).
As CO is a nonreactive pollutant, it is reasonable to assume 1) that the enclosure does
not intercept any of the CO as it moves indoors, 2) that the CO does not decay once
it enters the enclosure, and 3) that no CO is removed by air filtration devices. Under
these assumptions, the parameters FB, A, and F in Equation 11 would be set equal to
zero. If the additional assumptions are made that c and m are each equal to 1, the
resulting differential equation is
It can be shown that this equation has the following exact solution.
Cln(t) =^Cin(t-At) + Jc2Cout(t-At) + K3 (13)
where
jkt = e-"At (14)
fc (15)
33
-------
(16)
and At is a fixed time interval. Based on this relationship, the average indoor pollutant
concentration for hour h [Cm(h)] can be calculated by the expression
Cm(h) = a^Jh-l) + a2Cout(h) + a3 (17)
where Cm(h-1) is the indoor concentration at the end of the preceding hour, Cout (h) is
the average outdoor concentration during hour h,
ai = z(h), (18)
a, = 1 - z(h), (19)
a3 = _.[! - z(h)], (20)
z(h) = (1 - e-*)/j/. (21)
Equation 17 was used to construct a sequence of hourly average values for each
combination of microenvironment (indoor and motor vehicle) and exposure district.
In constructing each sequence, the value of Cout for a particular hour was set
equal to the value for outdoor concentration determined for that hour by the algorithm
described in Subsection 2.4.1. A value for air exchange rate (v) was selected from a
user-specified distribution each day at 7:00 p.m. and held constant until 7:00 p.m. of
the next day. This procedure was consistent with the procedure used to construct the
EES for each cohort. As discussed in Section 2, each EES consisted of a series of
person-days selected from an activity diary data base. Each person-day spanned a
24-hour period from 7:00 p.m. to 7:00 p.m.
The term S/V represents the contributions of indoor sources to indoor levels.
This term was included in the mass balance equation when the microenvironment was
34
-------
Normal: X = AM + ASD x Z (22)
Lognormal: X = CM x GSDZ (23)
In these expressions, AM is the arithmetic mean, ASD is the arithmetic standard
deviation, GM is the geometric mean, and GSD is the geometric standard deviation.
The distribution type (normal vs. lognormal) and the corresponding values for the
mean and standard deviation were determined by.fitting distributions to representative
data sets. A value for X was selected at random from a particular distribution by
randomly selecting a value for Z from the unit normal distribution [N(0,1)] and
substituting it into the appropriate equation. Table 9 lists the distribution type and
parameter values for each of the random variables used in the mass balance model.
3.2 The Air Exchange Rate Algorithm
An air exchange rate (AER) value was estimated for each indoor
microenvironment (IME) for each 24-hour period (7:00 p.m. to 7:00 p.m.) of the
exposure year. The estimation procedure consisted of randomly selecting AER values
from a distribution specific to each IME.
Two distinct AER distributions were established for the residential IME, one
representing windows open and one representing windows closed. The choice
between windows open and windows closed was conditioned on the air conditioning
system assigned to the residence and the outdoor temperature. For each 24-hour
period, an air conditioning (AC) algorithm was used to probabilistically specify the AC
system for the residence (central, window units, or none). A window status algorithm
was then used to probabilistically determine window status (closed or open). Based
on this determination, a value of AER for the 24-hour period was selected from either
the closed window distribution or the open window distribution.
36
-------
indoors - residence and the cohort was characterized as using natural gas for
cooking. The S parameter was assumed to represent emissions from a single gas
stove in the residence, and the V parameter was assumed to represent the total
volume of the residence.
Gas stove operation within the indoor - residence microenvironment was
simulated by assuming a constant emission rate attributable to pilot light operation and
a varying emission rate attributable to burner operation. Burner operation was
assumed to occur in discrete "burner operation periods" (BOP's) of 60 minutes
duration during normal, dinner hours and of 30 minutes duration at other times. No
more than one BOP was permitted to occur within a given clock hour, and each BOP
began and ended within the same clock hour. A monte carlo process was used to
randomly assign BOP's to clock hours throughout the year based on a table listing the
probability of a BOP occurring within each hour of a typical day. This table was
developed from an analysis of gas stove use patterns observed during the Denver
Personal Monitoring Study14.
An average burner emission rate was determined for each 7 p.m. to 7 p.m.
period by selecting annual gas use and emission factor values from user-specified
distributions. Pilot lights were assumed to operate continuously at a constant
emission rate. The simulated burner and pilot light emissions were summed for each
clock hour and presented to the mass balance model as an hourly average value for
S. The residential volume (V) receiving the CO emissions was determined for each 7
p.m. to 7 p.m. period by selecting values from a distribution representing the housing
stock in Denver.
The following subsections provide descriptions of the algorithms and data
bases used to determine the air exchange rates, burner operation probabilities, burner
emission rates, pilot light emission rates, and residential volumes used in the mass
balance model. Many of these algorithms require that values be selected at random
from normal or lognormal distributions. This selection was conducted by first defining
the distribution of interest by one of the following expressions.
35
-------
TABLE 9. DISTRIBUTIONS OF PARAMETER VALUES USED IN THE pNEM/CO
MASS BALANCE MODEL
Parameter
Air exchange
rate,
exchanges/h
Annual gas usage
by burners,
kilojoules
Annual gas usage
by pilot lights,
kilojoules
Emission factor,
mg/kilojoule
Residential
volume, cubic
meters
Category
Residence-
windows closed
Residence-
windows open
Nonresidential
locations
In vehicle
Distribution of parameter
Lognormal distribution
0 Geometric mean =0.53
0 Geometric standard deviation =
0 Lower bound = 0.063
0 Upper bound =4.47
1.704
Point estimate: 6.4
Lognormal distribution
0 Geometric mean = 1.285
0 Geometric standard deviation =
0 Lower bound = 0.19
0 Upper bound =8.69
1.891
Point estimate: 36.0
Lognormal distribution
e Geometric mean = 3.10 x 10
0 Geometric standard deviation =
0 Upper bound = 10.92 x 106
1.88
Point estimate: 1.82 x 106
Lognormal distribution
0 Geometric mean = 0.0294
0 Geometric standard deviation =
8 Upper bound = 0.400
Lognormal distribution
0 Geometric mean = 321
0 Geometric standard deviation =
0 Lower bound = 100
" Upper bound = 1200
2.77
1.642
The AC algorithm requires that the user specify the proportion of residences in
the study area that have central AC, window units, and no AC. According to the 1980
census, the breakdown for Denver is 18.4 percent central, 14.6 percent window, and
67.0 percent none. The algorithm as applied to Denver can be described as follows.
1. For each day, select a random number (RN) between zero and 1.
37
-------
2. If RN < = 0.184, then AC system is "central."
3. If 0.184 < RN < = 0.330, then AC system is "window units."
4. If 0.330 < RN, then AC system is "none."
The window status algorithm was originally developed for applications of
pNEM/O3 and has been described by Johnson et al.13 This algorithm determines
window status based on AC system and the daily average temperature according to
the probabilities listed in Table 10. The combination of window status algorithm and
AER algorithm can be described as follows.
1. For each day, determine daily average temperature from a
supplementary temperature file and AC system from AC system
algorithm. Select random number RN between zero and 1.
2. Assume Step 1 specifies 65 degrees and central AC. RN will be
evaluated against percentage values listed in Table 10 for central AC -
medium temperature range (i.e., 35.6, 29.4, and 34.6).
3. If RN < = 0.356, then windows are closed all day. AER value is selected
from the "windows closed" AER distribution.
4. If 0.356 < RN < = 0.650, then windows are open all day. AER value is
selected from the "windows open" AER distribution.
5. If 0.650 < RN, then windows are open for 58.2 percent of the day (see
last column). AER is determined by the expression
AER = (0.582 X open AER) + (0.418 X closed AER) , (24)
where open AER is selected from the open window AER distribution and
closed AER is selected from the closed window AER distribution.
38
-------
TABLE 10. PERCENTAGE OF PERSON-DAYS WITH INDICATED WINDOW RATIO BY
AIR CONDITIONING SYSTEM AND TEMPERATURE RANGE
Air conditioning
system
Central
Room units
No air
conditioning
Temperature
range*
Low
Medium
High
Low
Medium
High
Low
Medium
High
Percentage of person-days with indicated window
ratio"
Ratio = 0
86.0
35.6
62.1
73.2
12.0
17.1
80.0
4.7
1.4
Ratio = 1
0.8
29.4
12.9 .
2.0
44.2
34.3
1.0
59.1.
70.8
0 < ratio < 1
13.2
34.6
25.0
24.7
43.8
48.6
19.0
36.2
27.8
Mean of
ratios
not equal
to 0 or 1
0.260
0.582
0.503
0.316
0.618
0.521
0.276
0.716
0.774
'Low: 31' to 62'F
Medium: 63' to 75'F
High: 76'F and above.
"Ratio = (minutes windows open)/(minutes spent in residence).
3.3 Air Exchange Rate Distributions
A review of the scientific literature relating to air exchange rates identified 31
relevant references (list available on request). Of these, only a few were found to
contain sufficient data to construct a distribution of air exchange rates relating to a
particular building type such as residence or office. The two most useful studies were
conducted by Grimsrud et al.30 and by Turk et al.31
Residential Locations
Grimsrud et al. measured AER's in 312 residences. Reported AER values
ranged from 0.08 to 3.24. Researchers with IT Air Quality Services (ITAQS) analyzed
these data to determine which of two distributions (normal vs. lognormal) better
characterized the data. The lognormal distribution was found to yield a better fit, as
the data were highly skewed. The fitted lognormal parameters were geometric mean
= 0.53 and geometric standard deviation = 1.704. This distribution was used in
pNEM/CO to represent the distribution of AER's in residences with windows closed.
39
-------
Upper and lower limits of 4.47 and 0.063 air changes per hour were established to
prevent the selection of unusually extreme values of AER. These limits correspond to
the substitution of Z = 4 and Z = -4 in Equation 23 when GM = 0.53 and GSD =
1.704. The upper bound is 38 percent larger than the largest reported AER (3.24).
The lower bound is 21 percent smaller than the smallest reported AER (0.08).
No comparable data bases were identified which were considered
representative of residences where windows are open. Hayes has used 6.4 as the
AER value for open windows in applications of the PAQM model.32 This value was
used in the pNEM/CO analyses presented here. This value will be replaced by a
distribution when an appropriate data base becomes available.
Nonresidential Locations
Turk et al. measured AER's in 40 public buildings identified as schools (n = 7),
offices (n = 25), libraries (n = 3), and multipurpose buildings (n = 5). The minimum
reported AER was 0.3; the maximum was 4.1. Researchers with ITAQS fit normal and
lognormal distributions to the data for all 40 buildings and found that the lognormal
distribution produced a slightly better fit. The fitted lognormal parameters were
geometric mean = 1.285 and geometric standard deviation = 1.891.
The buildings can be grouped as offices (n = 25) and nonoffices (n = 15).
Lognormal fits to these data sets yielded geometric means and standard deviations of
1.30 and 1.93 for offices and 1.27 and 1.87 for nonoffices. ITAQS performed a two-
sample t test on the two data sets and found no significant difference in the means or
standard deviations of the data. Consequently, a single lognormal distribution
(geometric mean = 1.285, geometric standard deviation = 1.891) was used in
pNEM/CO for all public buildings.
To prevent the over-prediction of high AER values, an upper bound of 8.69 was
established. This value results when Z = 3 is substituted into Equation 23 with GM =
1.285 and GSD = 1.891. This value is over twice the largest AER value (4.1) reported
for the 40 buildings and corresponds to the 99.87 percentile of the specified lognormal
40
-------
distribution. A lower bound of 0.19 was also established. This value corresponds to a
Z value of -3 and represents the 0.13 percentile of the lognormal distribution.
In Vehicle Locations
A point estimate of 36 air changes per hour was used for in-vehicle locations.
This value was obtained from Hayes32 based on his analysis of data presented by
Peterson and Sabersky.33
3.4 Probability of Stove Use
The operation of gas stove burners in residences is simulated in the mass
balance model by specifying when the burners are on, the emission rate of the
burners during operation, and the volume of the residence where it is located.
As discussed above, burner operation was assumed to occur in discrete BOP's
such that use always began and ended within a single clock hour. BOP duration was
assigned a value of either 30 minutes or 60 minutes, depending on time of day.
These values were based on responses to a questionnaire administered by GEOMET
to 4312 survey participants. Each participant provided data on the type of cooking
facilities in the home, frequency of cooking, and average time spent in meal
preparation34.
Table 11 presents a summary of data from this survey by type of meal
(breakfast, lunch, and dinner). The values listed for average weekly time spent
cooking breakfast, lunch, and dinner are 65.8 minutes, 71.4 minutes, and 288.2
minutes, respectively. The total time for all three meals is 425.4 minutes per week.
The average daily cooking time based on this weekly value is 425.4/7 or 60.8 minutes.
In addition to duration, the data in Table 11 provide an indication as to the
frequency that a gas stove is used to prepare meals in the typical residence. In one
week, the stove will be used to prepare 2.5 breakfasts, 2.2 lunches, and 5.0 dinners --
41
-------
TABLE 11. STATISTICS ON GAS STOVE USE OBTAINED FROM
A SURVEY BY KOONTZ ET AL.34
Data item
Weekly duration of
gas stove use,
minutes
Weekly frequency of
gas stove use
Average duration of
use, minutes
Breakfast
65.8
2.5
26.3
Lunch
71.4
2.2
32.5
Dinner
288.2
5.0
57.6
Sum
425.4
9.7
a total of 9.7 meals per week. On an average day, the number of meals prepared on
a gas stove is 9.7/7 or 1.386.
Dividing the weekly cooking time associated with each meal type by the
average frequency of the meal yields average BOP's for breakfast, lunch, and dinner
of 26.3 minutes, 32.5 minutes, and 57.6 minutes, respectively. Based on these results,
pNEM/CO uses a value of 60 minutes for BOP's that occur during normal dinner
hours and 30 minutes for BOP's that occur at other times.
In pNEM/CO, stove operation is determined on an hourly basis by comparing a
randomly selected number between 0 and 1 with AP(h), the probability of a gas stove
being operated during the indicated clock hour h (h = 1, 2, ..., 24). If the random
number is less than AP(h), the stove is "on" for a duration of M(h) minutes, where M(h)
is either 30 or 60 minutes, depending on the value of h. If the random number is
greater than or equal to AP(h), the gas stove is "off" for the entire hour.
Table 12 lists the values of AP(h) and M(h) used in the pN EM/CO analysis by
clock hour. These values were developed to 1) reflect diurnal patterns in gas stove
usage specific to Denver, 2) yield an average daily duration for stove use of
approximately 60.8 minutes, and 3) yield an average daily frequency of stove use of
approximately 1.386.
Diurnal patterns in stove use were determined through an analysis of data from
the Denver Personal Monitoring Study14. In this analysis, the diary entries and
42
-------
TABLE 12. PROBABILITY OF GAS STOVE USE BY CLOCK HOUR
AND ASSUMED BURNER OPERATION PERIOD
Clock hour
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Total
AP(h): probability of
gas stove operation
0.025
0.023
0.023
0.023
0.023
0.026
0.049
0.058
0.081
0.073
0.062
0.075
0.085
0.071
0.067
0.064
0.107
0.130
0.091
0.058
0.052
0.047
0.040
0.035
1.386
M(h): assumed burner
operation period, minutes
30
30
30
30
30
30
30
30
30
30
30
30
30
30
60
60
60.
60
60
60
60
60
30
30
Product of AP(h)
and M(h), minutes
0.758
0.679
0.689
0.698
0.698
0.768
1.457
1.726
2.425
2.195
1.856
2.245
2.544
2.135
4.011
3.852
6.406
7.803
5.488
3.453
3.113
2.794
1.207
1.038
60.040
43
-------
background questionnaire provided by each study subject were used to determine 1)
when the subject was in a residence having a gas stove and 2) whether the stove was
on. As working subjects would not always be present when other family members
were operating a gas stove, it was assumed that workers would tend to
under-report stove use in their residences. It was further assumed that nonworkers
would use gas stoves more than the average person and that the diaries of
nonworkers would tend to over-represent typical gas stove use. Consequently, the
decision was made to average the worker and nonworker data and then adjust these
results so that the adjusted P(h) values would yield 1.386 hours of stove use "events"
per day, on average.
Table 13 presents the relevant data. For each clock hour, the table lists values
of P(h) for workers and nonworkers calculated as
P(h) = N(stove on, GSR)/N(GSR) , (25)
where N(stove on, GSR) is the number of diary entries indicating the subject was in a
gas stove residence when the stove was on and N(GSR) is the total number of diary
entries indicating the subject was in a gas stove residence. In calculating these
values, a stove was considered on during a particular clock hour if the subject's
activity diary indicated at least one minute of use during the hour.
The column labeled "average P(h)" lists the arithmetic mean of the worker and
nonworker P(h) values. These probabilities sum to 4.167 over 24 hours. It is
desirable that the probabilities sum to 1.386, as this will produce an average of 1.386
BOP's per day. The values labeled "adjusted P(h)" were calculated by multiplying the
average values by 0.333 (1.386/4.167). The adjusted values sum to 1.386.
The AP(h) values are listed also in Table 12. To the right of each AP(h) value is
the assumed value of M(h); that is, the number of minutes the stove will be assumed
to operate if the stove is determined to be "on" during the hour. The product of AP(h)
44
-------
TABLE 13. PROPORTION OF PEM VALUES IN GAS STOVE RESIDENCES
WITH STOVE IN OPERATION BY CLOCK HOUR AND WORK STATUS
Clock hour
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Nonworkers
n
63
59
5
58
58
62
67
84
119
87
80
76
72
62
51
64
96
103
151
148
102
82
82
75
P(h)
0.111
0.085
0.086
0.086
0.086
0.097
0.119
0.179
0.269
0.230
0.200
0.253
0.296
0.213
0.216
0.266
0.396
0.456
0.251
0.149
0.147
0.183
0.159
0.133
Workers
n
149
139
136
134
133
141
175
151
134
86
70
76
70
80
59
75
122
174
244
341
236
176
193
148
P(h)
0.041
0.051
0.052
0.053
0.053
0.057
0.173
0.167
0.216
0.209
0.171
0.197
0.214
0,215
0.186
0.120
0.246
0.326
0.299
0.196
0.165
0.097
0.083
0.075
Average
P(h)
0.076
0.068
0.069
0.070
0.070
0.077
0.146
0.173
0.243
0.220
0.186
0.225
0.255
0.214
0.201
0.193
0.321
0.391
0.275
0.173
0.156
0.140
0.121
0.104
4.167
AP(h):
adjusted
P(h)
0.025
0.023
0.023
0.023
0.023
0.026
0.049
0.058
0.081
0.073
0.062
0.075
0.085
0.071
0.067
0.064
0.107
0.130
0.091
0.058
0.052
0.047
0.040
0.035
1.386
45
-------
and M(h) is listed in the far right column. Summing these? values over all 24 hours
provides an estimate of the average number of minutes per day that a gas stove will
be operated according to the algorithm. The sum is 60.04 minutes, a value very close
to the desired value of 60.8 minutes.
3.5 Gas Stove Emission Rate
The gas stove algorithm in pNEM/CO is based on the assumption that the
mass of CO emitted by a gas stove during a particular hour (h) can be estimated by
the equation
MASSCO(h) = ERBURN x M(h)/60 + ERPILOT, (26>
where MASSCO(h) is expressed in mg, ERBURN is the hourly burner emission rate in
mg per hour, and ERPILOT is the hourly pilot light emission rate in mg per hour. M(h)
is the duration of burner use during hour h expressed in minutes.
M(h) is zero for each hour in which the algorithm assigns the stove a status of
"off". If the stove status is "on" for a particular hour, then M(h) is assigned a value of
30 or 60 minutes according to Table 12.
ERBURN is determined by the equation
ERBURN = (AUB/365.2) X EFBURN, <2?)
where AUB is the annual fuel usage of the burners in kilojoules, 365.2 is the number of
hours per year that the burners are operated assuming 60.04 minutes of use per day
(Table 12), and EFBURN is the burner emission factor in mg of CO per kilojoule. The
values of AUB and EFBURN are assumed to be constant within each 24-hour period.
Values of AUB are randomly selected from a normal distribution with mean =
3.42 million kilojoules and standard deviation = 1.14 million kilojoules. This distribution
is based on the distribution of annual gas stove fuel use measured in 46 homes in
which the stoves did not have pilot lights35. The value of AUB is not permitted to
exceed 6.84 million kilojoules. This value is based on the assumption that maximum
gas usage will not exceed a value which is double the mean value.
46
-------
Values of EFBURN are randomly selected from a lognormal distribution with
geometric mean = 0.0294 mg/kilojoule and geometric standard deviation = 2.77.
Values of EFBURN are not permitted to exceed 0.400 mg/kilojoule. These values are
based on the results of an ITAQS analysis of data reported by Davidson et al.36 and
represent a well-adjusted stove. As such, the assumed geometric mean is probably
low with respect to the overall population of gas stoves.
ERPILOT is determined by the equation
ERPILOT = (AUP/8160) x EFPILOT (28)
where AUP is the annual fuel usage by all pilot lights in kilojoules, 8760 is the number
of hours per year that the pilot lights are in operation, and EFPILOT is the pilot light
emission factor in mg of CO per kilojoule. The value of AUP is constant over all 24-
hour periods and is set equal to a point estimate of 1.82 million kilojoules. The value
of EFPILOT is assumed to be constant within each 24-hour period and is set equal to
the value determined for EFBURN in Equation 27.
The point estimate for AUP was developed as follows. A survey by the Public
Service Commission of Colorado found that the total annual gas use by a single stove
in Denver averaged 5000 cubic feet per year37. The Gas Research Institute estimates
that the average heating value of natural gas in Denver is 1048 kilojoules per cubic
feet38. Thus the average total fuel use by a gas stove in Denver (burners plus pilot
lights) is 5000 x 1048 or 5.24 million kilojoules per year. The mean annual usage of a
gas stove without pilot lights is estimated to be 3.42 million kilojoules per year.
Subtracting 3.42 million from 5.24 million yields an estimate of 1.82 million kilojoules for
the pilot light contribution to annual fuel use of a gas stove.
The value of MASSCO(h) determined by Equation 26 is used as the value of S
for hour h in Equation 20, regardless of the value of M(h). This approach permits the
use of the hourly average exact solution (Equation 17) to the mass balance equation
(Equation 12). The practical result of this simplification is a slight smoothing in the
simulated hour-to-hour variation in indoor CO concentrations with respect to the
pattern which would be simulated by a model with finer time resolution.
47
-------
3.6 Residential Volume
Data collected by Johnson et al.14 and by Grimsrud, et al.30, were analyzed to
determine typical distributions of residential volume. The Johnson data were collected
during the Denver Personal Monitoring Study, in which 409 residents of Denver
provided estimates of their residential living area in square feet. The percentiles of
these data are listed below.
Percentile Living area, square feet
10 750
25 1000
50 1425
75 2056
90 2600
These percentiles can be closely fit by a lognormal distribution with a geometric mean
of 1417 square feet (132 square meters) and a geometric standard deviation of 1.642.
Assuming a constant ceiling height of 8 feet (2.44 m), the corresponding lognormal
distribution for volume would have a geometric mean of 321 cubic meters. The
geometric standard deviation would remain the same (1.642).
A lognormal distribution was also fit to volume data for 312 single-family homes
located throughout North America provided by Grimsrud et al. The fitted lognormal
distribution had a geometric mean of 382 cubic meters and a geometric standard
deviation of 1.401.
The difference in geometric means (321 versus 382) between the Johnson and
Grimsrud studies may reflect the fact that the Denver sample included apartments
whereas the Grimsrud study was limited to single-family houses. As the pNEM/CO
analysis includes apartment dwellers in the exposed population, the Denver distribution
was considered to be more representative. Consequently, the Denver distribution
(geometric mean = 321 cubic meters, geometric standard deviation = 1.642) was
used in the analysis.
48
-------
SECTION 4
THE CARBOXYHEMOGLOBIN ALGORITHM
The EES for a particular cohort consists of a sequence of individual exposure
events. The main pNEM/CO program provides a start time, a duration, an average
CO concentration, and an average ventilation rate (expressed as EVR) for each of
these events. These data are combined with supplemental data in the
carboxyhemoglobin (COHb) algorithm to produce an estimate of the COHb level
(expressed as percent) at the end of each event.
The COHb algorithm is based on a differential equation developed by R. F.
Coburn, R. E. Forster, and R. B. Kane.23 William F. Biller and Harvey M. Richmond
developed a solution to the non-linear form of the Coburn-Forster-Kane (CFK)
equation which has been incorporated into the COHb algorithm.22 A paper describing
the research by Biller and Richmond is included in Appendix C.
The COHb algorithm is called by the main pNEM/CO program for each
exposure event. The algorithm produces an estimate of the COHb level at the end of
the event as a function of the following parameters.
Percent COHb at the start of the event
Average CO exposure concentration over the event, ppm
Time duration of event, min
Alveolar ventilation rate, ml/min
Haldane Constant
Atmospheric pressure at sea level, torr
Altitude above sea level, feet
Blood volume, ml
Total hemoglobin content of blood, gm/100 ml
Pulmonary CO diffusion rate, ml/min per torr
Endogenous CO production rate, ml/min.
49
-------
These parameters are discussed briefly in Subsection 4.2. The paper by Biller and
Richmond provides a more detailed discussion of each parameter together with a list
of references.
Values for several parameters were determined according to a physiological
profile which the COHb algorithm generated for each 24-hour time period in the EES.
Subsection 4.1 describes the method used to generate these profiles.
4.1 The Physiological Profile
Each 7:00 p.m. to 7:00 p.m. time period in an EES represents activity diary data
reported by one subject for one 24-hour time period (i.e., one "person-day"). The
COHb algorithm generates a physiological profile of this subject based on the
demographic group of the associated cohort. The physiological profile specifies
Age, years
Menstrual phase
Height
Weight
Body surface area
The profile is updated with each new person-day by selecting a value or category for
each parameter from appropriate distributions.
Age
Each cohort is associated with a particular demographic group. Three age
groups are specified in the demographic group descriptions: 18 to 44, 45 to 64, and
65+. The age assigned to the physiological profile generated for a particular person-
day was determined by randomly sampling a value from the age range of the
associated demographic group. For this purpose, age was assumed to be uniformly
distributed within each age range and the 65 + age group was assumed to include
ages 65 to 75.
50
-------
Menstrual Phase
The gender (male/female) of each cohort was specified by the demographic
group identifier. Whenever the gender was female, the physiological profile generated
for each person-day contained a menstrual phase assignment (either pre-menstrual or
post-menstrual) conditioned on age. If age was less than 65, the phase was assigned
randomly with equal probability of the two outcomes. If age was 65 or higher, then
the premenstrual phase was assigned.
Height and Weight
The value for height was selected from one of six normal distributions which
varied according to age and gender (Table 14). The value for weight was determined
from the equation
weight = A0 + A: x height + z x (S.E.) (29)
where weight is expressed in pounds, height is expressed in inches, A,, and A, are
regression coefficients, S.E. is the standard error of the regression, and z is a value
randomly sampled from the unit normal distribution [N(0,1)]. Table 14 lists values for
AQ, A,, and S.E. according to the gender and age group of the cohort.
Lower bounds were established for height and weight to prevent the selection
of unreasonably low values. The lower bound for height was 65 inches for men and
60 inches for women. The lower bound for weight was specified as the weight
corresponding to a z value in Equation 29 of -4. If a selected value was less than the
specified bound, a new value was selected.
51
-------
Body Surface Area
A formula proposed by Dubois and Dubois39 was used to estimated a BSA
value according to the height and weight values in the physiological profile. The
Dubois formula is
S = 10.092 x f/0-*25 x H°-125
(30)
where W is weight in pounds and H is height in inches. This formula has been used
by EPA researchers to estimate BSA.18
TABLE 14. COEFFICIENT VALUES FOR CARBOXYHEMOGLOBIN ALGORITHM
Parameter
Height, in
Weight, Ibs
Total
hemoglobin
content of
blood
Endogenous
CO
production
rate
Sampling source
Normal distribution
with arithmetic mean =
AM and arithmetic std.
dev. = ASD
Equation 27
Normal distribution
with arithmetic mean =
AM and arithmetic std.
dev. = ASD
Lognormal distribution
with geometric mean =
6M and geometric std.
dev. = GSO
Age group
18 to 44
45 to 64
65+
18 to 44
45 to 64
65+
18 to 44
45 to 64
65+
18 to 64
65+
Coefficient values
Males
AM = 69.5, ASD = 2.8
AM = 68.6, ASD = 2.6
AM = 67.3. ASD = 2.6
A. = -168.67, A, = 4.941
SE = 30.5
A. = -131.83, A, = 4.454
SE = 28.4
A. = -131.64, A, = 4.385
SE 26.0
AM = 15.3, ASD = 1.0
AM = 15.1, ASD = 1.2
AM = 14.8, ASD = 1.4
GM = 0.473, GSD = 1.316
GM = 0.473, GSD = 1.316
Females
AM = 64.2. ASO = 2.5
AM = 63.2. ASD = 2.4
AM = 62.3, ASO = 2.4
A. = -88.62, A, = 3.587
SE = 32.1
A, = -77.17, A, = 3.583
SE = 33.8
Ao = -76.38
A, = 3.583
SE = 29.0
AM = 13.3, ASD = 1.1
AM = 13.6, ASD = 1.2
AM = 13.7. ASD = 1.2
Premenstrual :
GM = 0.497, GSO = 1.459
Post menstrual :
GM = 0.311. GSD = 1.457
GM = 0.497. GSD = 1.459
aHourly average value.
52
-------
4.2 The COHb Parameters
Haldane Constant. Sea Level Pressure, and Altitude
The COHb algorithm treats the Haldane constant and atmospheric pressure at
sea level as constants over all calculations. These values were assumed to be 218
and 760 torr, respectively. Altitude is assumed to be constant for a particular city.
Denver's altitude was assumed to be 5183 feet.
Starting COHb Level for an Exposure Event
The starting COhb level for each exposure event was set equal to the ending
COHb level of the preceding exposure event. Consistent with this approach, the initial
COHb level for each person-day was set equal to the final COHb level for the
preceding person-day. In the special case of the first person-day of the year, the initial
COHb level was set equal to the final COHb level determined for an initializing person-
day. The initializing person-day was identical to the first person-day of the year. The
application of the COHb algorithm to the initializing person-day was performed by
setting the initial COHb level equal to zero.
Duration and Average CO Concentration
The duration and average CO concentration of each exposure event were
determined by another part of pNEM/CO as described in Section 2. No event
crossed a clock hour; consequently, no event lasted more than 60 minutes.
Alveolar Ventilation Rate
The pN EM/CO program calculated a value for the EVR associated with each
exposure event. As EVR is defined as alveolar ventilation rate divided by body surface
area (BSA), the EVR value determined for each exposure event was multiplied by an
appropriate BSA value to yield a ventilation rate value for the event. The BSA value
was specified by the physiological profile generated for the person-day.
53
-------
Blood Volume
A value for blood volume was estimated for each person-day according to the
following formulae.
Men: Vb = 20. W + 0.00683H3 - 30 (31)
Women: Vb = 14.6W + 0.00678H3 - 30 (32)
In these expressions, Vb is blood volume in ml, W is weight in pounds, and H is height
in inches. The values used for weight and height were those specified in the
physiological profile generated for the person-day.
Total Hemoglobin Content of the Blood
The value for total hemoglobin was assumed to be constant within each
person-day. The value was selected from one of six normal distributions according to
the gender of the cohort and the age specified by the physiological profile. Table 14
lists the parameter values of these distributions.
Pulmonary CO Diffusion Rate
The pulmonary CO diffusion rate (DL) was calculated by the following
equations.
Men: DL = 0.361H - 0.232Y + 16.3 (33)
Women: DL = 0.556H - 0.115JT - 5.97 <34>
54
-------
In these equations, DL is expressed in ml/min/torr, H is height in inches, and Y is age
in years. Height and age were determined from the physiological profile.
Endogenous CO Production Rate
A value for the hourly average endogenous CO production rate was selected
from one of six lognormal distributions according to the cohort's gender and the
menstrual phase specified by the physiological profile. Table 14 lists the parameter
values of these distributions. Minute rates were determined by dividing the hourly
rates by 60.
55
-------
SECTION 5
EXPOSURE ESTIMATES FOR DENVER RESIDENTS
The pNEM/CO methodology was used to develop estimates of CO exposure
and resulting COHb levels within the Denver study area population. This section
presents selected results for four scenarios:
1. Existing conditions - indoor sources "on"
2. Existing conditions - indoor sources "off"
3. Attains 8 hr NAAQS - indoor sources "on"
4. Attains 8 hr NAAQS - indoor sources "off."
In these descriptions, the qualifier "existing conditions" refers to the air quality
conditions reported by the Denver fixed-site monitors during 1988. The qualifier
"attains 8 hr NAAQS" indicates that the rollback procedure described in Subsection
2.4.2 was used to simulate the attainment of the 8-hour CO NAAQS in Denver. The
term "indoor sources" refers to gas stoves and passive smoking only. Indoor sources
"on" indicates pNEM/CO was run in the standard mode in which the CO contributions
from gas stoves and passive smoking are treated according to the procedures
described in Section 2. Indoor sources "off" indicates that pN EM/CO was run with the
CO contribution from these sources set equal to zero.
Tables 15, 16, and 17 present modeling results for a sensitive population
defined as all adults residing in the Denver study area with IHD. This population group
is estimated to contain 36,645 persons.
56
-------
TABLE 15. NUMBER OF PERSON-DAYS IN WHICH A DENVER ADULT
WITH ISCHEMIC HEART DISEASE EXPERIENCES A 1-HOUR DAILY MAXIMUM
CARBON MONOXIDE EXPOSURE AT OR ABOVE SPECIFIED CONCENTRATION
UNDER EACH OF FOUR SCENARIOS
CO
Concentrating
ppm
45
40
35
30
25
20
15
10
0
Number of person-days
Existing conditions
Sources on
28,885
49,528
71,672
105,243
170,540
350,673
1,328,538
3,655,352
13,412,070
Sources off
22,749
41,629
67,478
105,070
156,847
314,402
1,130,373
3,408,952
13,412,070
Attainment of 8-h NAAQS
Sources on
0
8,426
11,073
12,933
18,306
33,956
72,603
267,476
13,412,070
Sources off
0
1,380
6,562
7,397
15,137
33,465
76,582
221,000
13,412,070
TABLE 16. NUMBER OF PERSON-DAYS IN WHICH A DENVER ADULT
WITH ISCHEMIC HEART DISEASE EXPERIENCES AN 8-HOUR DAILY
MAXIMUM CARBON MONOXIDE EXPOSURE AT OR ABOVE SPECIFIED
CONCENTRATION UNDER EACH OF FOUR SCENARIOS
CO
Concentration
ppm
25
20
15
12
9
6
0
Number of person-days
Existing conditions
Sources on
3
2,007
27,840
118,268
624,715
1,840,121
13,412,070
Sources off
0
115
19,027
90,195
570,168
1,628,463
13,412,070
Attainment of 8-h NAAQS
Sources on
0
0
42
128
652
13,348
13,412,070
Sources off
0
0
0
1
503
12,240
13,412,070
57
-------
TABLE 17. NUMBER OF PERSON-HOURS IN WHICH A DENVER ADULT
WITH ISCHEMIC HEART DISEASE EXPERIENCES AN END-OF-HOUR
CARBOXYHEMOGLOBIN LEVEL AT OR ABOVE SPECIFIED PERCENTAGE
UNDER EACH OF FOUR SCENARIOS
COHb
level,
percent
6.0
5.0
4.0
3.0
2.5
2.4
2.3
2.2
2.1
2.0
1.5
1.0
0
Number of person-hours
Existing conditions
Sources on
0
0
145
3,794
31,816
49,673
73,915
116,188
183,126
307,373
2,978,224
13,640,275
321,889,680
Sources off
0
0
94
3,604
15,241
24,144
39,930
61,803
104,795
192,688
2,555,163
12,102,333
321,889,680
Attainment of 8-h NAAQS
Sources on
0
0
0
122
356
393
487
521
649
754
5,801
80,154
321,889,680
Sources off
0
0
0
0
0
0
4
9
13
36
3,778
79,885
321,889,680
58
-------
Table 15 presents estimates of the number of person-days in which members of
the sensitive population experienced a 1-hour daily maximum CO exposure at or
above each of the indicated CO concentrations. Table 16 is similar in format; it
presents estimates of the number of person-days in which members of the sensitive
population experienced an 8-hour daily maximum exposure at or above each indicated
CO concentration. The maximum possible value in each of these tables is
13,412,070 -- the product of the number of persons in the sensitive population
(36,645) and the number of days in the exposure period (366).
Table 17 presents estimates of the number of person-hours in which members
of the sensitive population experienced an end-of-hour COHb level at or above each of
the indicated levels. The maximum possible value in this table is 321,889,680 - the
product of the number of persons in the sensitive population (36,645) and the number
of hours in the exposure period (8,784).
Table 15 indicates that the sensitive population are expected to experience
71,672 person-days in which the 1-hour daily maximum CO exposure is at or above 35
ppm under Scenario 1 (existing conditions, sources on). This number is equivalent to
an average exposure rate of 1.96 days per person (i.e., 71,672 person-days divided by
36,645 persons). With sources off (Scenario 2), the number is reduced to 67,478.
The number is further reduced to 11,073 under attainment with sources on (Scenario
3) and to 6,562 under attainment with sources off (Scenario 4). The average exposure
rates associated with Scenarios 2, 3, and 4 are 1.84, 0.30, and 0.18 days per person,
respectively.
According to Table 16, the sensitive population is expected to experience
624,715 person-days in which the 8-hour daily maximum CO exposure is at or above 9
ppm under Scenario 1. This value corresponds to an average exposure rate of 17.0
days per person. The value is reduced to 570,168 (15.6 days per person) under
Scenario 2. There is a significant reduction in exposure under the two attainment
scenarios. Under Scenario 3, the sensitive population is expected to experience only
652 person-days at or above 9 ppm for an average exposure rate of 0.018 day per
59
-------
person. Under Scenario 4, the value is further reduced to 503 person-days (0.014 day
per person).
Carboxyhemoglobin levels of 2.1 percent and 3.0 percent are currently of
interest to EPA in evaluating health effects under proposed NAAQS. Table 17
indicates that the sensitive population will experience 183,126 person-hours in which
the end-of-hour COHb level is at or above 2.1 percent under Scenario 1 (an average
of 5.00 hours per person). The corresponding values for the other three scenarios are
104,795 person-hours for Scenario 2 (2.86 hours per person), 649 person-hours for
Scenario 3 (0.018 hour per person), and 13 person-hours for Scenario 4 (0.0004 hour
per person). The number of person-hours at or above 3.0 percent COHb is 3794 for
Scenario 1 (0.10 hour per person), 3604 for Scenario 2 (0.10 hour per person), 122 for
Scenario 3 (0.003 hour per person), and 0 for Scenario 4.
The overall results suggest that the attainment of the 8-hour NAAQS for CO
would significantly reduce the frequency that members of the sensitive population
experience high CO exposures and high COHb .levels. The results also suggest that
gas stoves and passive smoking make a significant contribution to CO exposure. In
general, CO exposures and resulting COHb levels are higher under Scenario 2
(existing conditions, sources off) than under Scenario 3 (attainment of 8-h NAAQS,
sources on).
The results presented for each scenario in Tables 15, 16, and 17 are based on
a single run of pNEM/CO. Because pNEM/CO contains a large number of stochastic
variables, estimates of CO exposure and COHb level will vary from run to run. This
variation is illustrated in Tables 18, 19, and 20 using the same exposure indicators
appearing in Tables 15, 16, and 17, respectively. Each table presents the results of
three runs based on Scenario 1 and the arithmetic mean of the three runs. Run B in
these tables is the run which produced the estimates presented in Tables 15 through
17.
Of particular interest to EPA is the variability of model estimates for 1-hour daily
maximum exposures above 35 ppm, for 8-hour daily maximum exposures above 9
ppm, and for end-of-hour COHb levels above 2.1 percent and 3.0 percent. In Table
60
-------
TABLE 18. NUMBER OF PERSON-DAYS IN WHICH A DENVER ADULT WITH ISCHEMIC
HEART DISEASE EXPERIENCES A 1-HOUR DAILY MAXIMUM CARBON MONOXIDE EXPOSURE
AT OR ABOVE SPECIFIED CONCENTRATION FOR EACH OF THREE MODEL RUNS UNDER
SCENARIO 1 (EXISTING CONDITIONS, SOURCES ON)
CO
concentration,
ppm
45
40
35
30
25
20
15
10
0
Number of
Run A
29,746
48,316
74,341
114,905
181,787
372,627
1,305,429
3,604,190
13,412,070
Run B
28,885
49,528
71,672
105,243
170,540
350,673
1,328,538
3,655,352
13,412,070
person-days
Run C
19,317
36,642
58,873
91,677
158,890
320,111
1,283,725
3,675,079
13,412,070
Mean
25,983
44,829
68,295
103,942
170,406
347,804
1,305,897
3,644,874
13,412,070
TABLE 19. NUMBER OF PERSON-DAYS IN WHICH A DENVER ADULT WITH ISCHEMIC
HEART DISEASE EXPERIENCES AN 8-HOUR DAILY MAXIMUM CARBON MONOXIDE EXPOSURE
AT OR ABOVE SPECIFIED CONCENTRATION FOR EACH OF THREE MODEL RUNS
UNDER SCENARIO 1 (EXISTING CONDITIONS, SOURCES ON)
CO
concentration,
ppm
25
20
15
12
9
6
0
Number of person-days .
Run A
43
1,145
22,478
112,740
632,714
1,789,710
13,412,070
Run B
3
2,007
27,840
118,268
624,715
1,840,121
13,412,070
Run C
779
2,612
23,846
110,687
651,208
1,878,934
13,412,070
Mean
275
1,771
24,721
113,898
636,212
1,836,255
13,412,070
61
-------
TABLE 20. NUMBER OF PERSON-HOURS IN WHICH A DENVER ADULT
WITH ISCHEMIC HEART DISEASE EXPERIENCES AN END-OF-HOUR
CARBOXYHEMOGLOBIN LEVEL AT OR ABOVE SPECIFIED PERCENTAGE
FOR EACH OF THREE MODEL RUNS UNDER SCENARIO 1
(EXISTING CONDITIONS, SOURCES ON)
COHb
level ,
percent
6.0
5.0
4.0
3.0
2.5
2.4
2.3
2.2
2.1
2.0
1.5
1.0
0
Number of person-days
Run A
0
0
289
2,388
21,901
33,586
54,692
106,100
177,821
323,055
2,984,354
13", 362, 834
321,889,680
Run B
0
0
145
3,794
31,816
49,673
73,915
116,188
183,126
307,373
2,978,224
13,640,275
321,889,680
Run C
26
52
196
3,571
22,649
36,025
57,461
98,614
166,409
269,342
3,069,737
13,772,509
321,889,680
Mean
9
17
210
3,251
25,455
39,761
62,023
106,967
175,785
299,923
3,010,773
13,591,873
321,889,680
62
-------
18, the number of persons experiencing 1-hour daily maximum CO exposures at or
above 35 ppm is 74,341 for Run A, 71,672 for Run B, and 58,873 for Run C. The
mean of these three runs is 68,295; the standard deviation is 8,268 (12.1 percent of
the mean). The variation is smaller with respect to the 8-hour indicator (Table 19).
The number of persons experiencing 8-hour daily maximum CO exposures at or above
9 ppm is 632,714 for Run A, 624,715 for Run B, and 651,208 for Run C. The mean is
636,212; the standard deviation is 13,584 (2.1 percent of the mean).
The variability of the COHb estimates depends greatly on the reference COHb
level. As indicated in Table 20, the number of person-hours with an end-of-hour
COHb level at or above 2.1 percent is 177,821 for Run A, 183,126 for Run B, and
166,409 for Run C. The mean of the three runs is 175,784, and the standard deviation
is 8,544 (4.9 percent of the mean). The number of person-hours at or above 3.0
percent COHb is 2388 for Run A, 3794 for Run B, and 3571 for Run C. These values
have a mean of 3251 and a standard deviation of 756 (23.3 percent of the mean).
63
-------
SECTION 6
INITIAL EFFORTS TO VALIDATE THE EXPOSURE MODEL
The validity of pNEM/CO was evaluated through the use of data collected
during the Denver Personal Monitoring Study.14 During this study, each of
approximately 450 subjects carried a PEM and an activity diary for a two-day period.
Each subject also completed a background questionnaire which provided information
on various demographic variables, including whether or not the subject lived in a home
with a gas stove. The study was conducted over a four-month period starting on
November 1, 1982 and ending on February 28, 1983.
Researchers analyzed the Denver data to determine the 1-hour daily maximum
and the 8-hour daily maximum CO exposures associated with each person-day.
These values were classified according to whether or not the subject's residence had
a gas stove. The solid lines plotted in Figures 3 and 4 (labeled "measured exposure")
indicate the distribution of the 1-hour daily maximum exposures for subjects with gas
stove residences and with non-gas stove residences, respectively. The solid lines
plotted in Figures 5 and 6 show the corresponding distributions of 8-hour daily
maximum exposures.
In an effort to validate pN EM/CO, researchers set up a special run of
pNEM/CO which attempted to simulate the conditions of the Denver Personal
Monitoring Study. In this application, the modeling approach described in Section 2
was followed with the following alterations. The exposure period was changed from
1988 to the four-month period of the Denver Personal Monitoring Study. Whenever
pN EM/CO algorithms required fixed-site CO concentrations, these values were
determined using fixed-site monitoring data for this four-month period. In a similar
manner, temperature and calendar data for this four-month period were used
64
-------
100
70
50
40
30
20
8 1°
-8
I 7
i
I 5
a
i i i
i i
Measured exposure
Simulated exposure
I I i
M 10 JO M M 70 fO fO 'H
Cumulative probability, percent
**.»
Figure 3. Cumulative distributions of 1-hour daily maximum carbon monoxide exposures
measured during the Denver Personal Monitoring Study and simulated by
pNEM/03 (persons residing in homes with gas stoves).
65
-------
100
70
50
40
30
§ 20
at
u
8 10
•8
i
3 4
1 I I I I I
1 I I I I I J
Measured exposure
Simulated exposure
i I
I
I i
i i
10 10 40 10 *0 rO M tO W
Cumulative probability, percent
f».t
t».t»
Figure 4. Cumulative distributions of 1-hour daily maximum carbon monoxide exposures
measured during the Denver Personal Monitoring Study and simulated by
pNEM/CO (persons residing in homes without gas stoves).
66
-------
&
Q.
C
at
o
c
o
u
X
o
o
•e
ID
CJ
30
25
20
15
10
~1 I I I I I I I
1 1 I I
l I
Measured exposure
Simulated exposure
l l I l i
l i I
I l l l l
jo « jo «o ro w to t5
Cumulative probability, percent
w tt
ft.W
Figure 5. Cumulative distributions of 8-hour daily maximum carbon monoxide exposures
measured during the Denver Personal Monitoring Study and simulated by
pNEM/CO (persons residing in homes with gas stoves).
67
-------
c
o
30
25
20
15
10
c '
8 '
8 5
I 4
o
IB
i i r
i i
Measured exposure
Simulated exposure
J
ill
I t ill
M JO 40 JO «0 70 10 »0 M
Cumulative probability, percent
M ft
f»J
w.w
Figure 6. Cumulative distributions of 8-hour daily maximum carbon monoxide exposures
measured during the Denver Personal Monitoring Study and simulated by
pNEM/CO (persons residing in homes without gas stoves).
68
-------
whenever these data were required by the model. For example, the exposure event
sequence for each cohort was generated using temperature and weekday/weekend
calendar data specific to the four-month period.
The exposure estimates from the special application were tabulated according
to the classification of each cohort with respect to cooking fuel (natural gas or other).
The results of these tabulations are indicated by the distributions represented by
dashed lines (labeled "simulated exposure") on Figures 3 through 6.
There is generally good agreement between the distributions of measured and
simulated exposures when the exposure indicator is the 1-hour daily maximum value
(Figures 3 and 4). The principal region of disagreement in each graph falls above the
99th percentile. As each of the PEM-derived points plotted in this range is based on
less than 10 person-days of data, it is difficult to determine whether the observed
discrepancy is real or the result of sampling error
The distributions of measured and simulated 8-hour daily maximum exposures
presented in Figures 5 and 6 do not agree as well as the 1-hour distributions. In each
case, the distribution obtained from pNEM/CO over-estimates the occurrence of small
CO exposures and under-estimates the occurrence of large CO exposures. The
simulated and measured distributions agree most closely in the range of CO
concentrations between 5 ppm and 12 ppm. This range brackets 9 ppm, the CO
concentration associated with the 8-hour CO NAAQS.
The PEM and activity diary data of Denver subjects reporting high 8-hour daily
maximum exposures were reviewed in an effort to identify the factor(s) associated with
these exposures that were not being adequately addressed by the pNEM/CO
methodology. A total of 52 person-days were identified as being associated with 8-
hour daily maximum exposures at or above 10 ppm. In most of these cases, the PEM
data indicated an extended period of CO exposures ranging between 10 ppm and 30
ppm had occurred rather than a short-term exposure at a much higher level.
In 18 cases, the subject spent most of the 8-hour period in the indoors-
residence microenvironment. A gas stove was in operation in the residence for an
extended period of time (more than two hours) in five of these cases. Smokers were
69
-------
present in the residence in two of these cases. Multiple potential sources of CO were
present in the residence in 11 cases.
In five of the fifty-two cases, the subject spent an extended period of time in a
motor vehicle. The subject stayed in a service station or public garage for an
extended time period in three cases. One or two of the 52 cases were associated with
each of the following exposure situations: 1) working in an office with or without
smokers, 2) riding in a truck with or without smokers, 3) visiting or working in
restaurant, 4) shopping, 5) working in a store, 6) working in a hospital, 7) riding a bus,
and 8) attending school.
In nine cases, the subject moved through a variety of microenvironments in
which high exposures were recorded. In four other cases, the situation associated with
a high 8-hour exposure could not be adequately characterized because of missing or
unreliable data.
A detailed review of the PEM data associated with these cases suggested that
high 8-hour daily maximum exposures often arise from an extended series of short-
term exposures which exhibit a high degree of autocorrelation over time. Although
many of the existing pNEM/CO algorithms incorporate some degree of
autocorrelation, it appears that additional autocorrelation is required if the model is to
adequately represent the upper end of the distribution of 8-hour daily maximum CO
exposures. In particular, autocorrelation should be increased in the algorithms that
determine the sequence of outdoor CO concentrations for each microenvironment and
that determine the sequence of on/off periods for gas stoves. In addition, it may be
necessary to provide pNEM/CO with special algorithms to simulate certain extended,
high exposure situations, such as working in a service station, parking garage, or
restaurant.
70
-------
SECTION 7
DISCUSSION AND RECOMMENDATIONS
N EM is a general approach to estimating population exposure to air pollution
which has been evolving since 1979. The first version of NEM to incorporate
probabilistic elements was the pNEM/03 model described by Johnson et al13. The
model described here, pNEM/CO, represents an advance over pNEM/O3 in at least
two respects. First, pNEM/CO employs a mass balance model to estimate pollutant
concentrations indoors and in motor vehicles. A series of regression equations were
used for this purpose by pNEM/03. Second, pNEM/CO employs a much larger
number of stochastic exposure factors (20) than does pNEM/O3. The term
"stochastic exposure factor" is here defined as any parameter within the model for
which values are determined over time by a process involving random elements.
The stochastic exposure factors in pNEM/CO include factors that were
determined on an event basis, on an hourly basis, and on a 24 hour basis. The
factors are listed below according to these classifications.
Event Basis
0 Breathing rate classification (except data obtained from Cincinnati activity
diary study)
0 Equivalent ventilation rate (EVR)
0 Alveolar ventilation rate
Hourly Basis
0 Estimates of missing values in fixed-site monitoring data .
0 Gas stove status (on/off)
71
-------
24-hour Basis
0 Activity pattern
0 Air exchange rate
0 Annual gas usage rate
- ° Gas stove emission factor
0 Residential volume
0 Air conditioning system (central, window units, none)
0 Window status (open/closed)
Age
0 Menstrual phase (pre or post)
Height
Weight
0 Body surface area
0 Blood volume
0 Total hemoglobin content of the blood
0 Endogenous CO production rate.
As demonstrated in Section 5, the use of stochastic variables in pNEM/CO can
produce a relatively large variation in exposure estimates when multiple runs of the
model are performed.
This variability can be characterized by first calculating the mean and standard
deviation for a particular exposure indicator based on the results of multiple runs and
then expressing the ratio of the standard deviation to the mean as a percentage value.
Based on three runs in which pNEM/CO was applied to Denver adults with IHD
(Section 5), the value of this variability index was 12.1 percent for the number of
72
-------
person-days with 1-hour daily maximum exposures at or above 35 ppm, 2.1 percent
for the number of person-days with 8-hour daily maximum exposures at or above 9
ppm, 4.9 percent for the number of person-hours with end-of-hour COHb levels at or
above 2.1 percent, and 23.3 percent for the number of person-hours with end-of-hour
COHb levels at or above 3.0 percent.
In situations where the analyst desires a "best estimate" of CO exposure and
resulting COHb levels under a given scenario, it is advisable to make multiple runs of
pNEM/CO and then calculate summary statistics (e.g., arithmetic mean and standard
deviation) for each exposure indicator of interest. These statistics can be used to
determine the central tendency of each indicator and to characterize the degree of
variability associated with each indicator.
Efforts were made to validate pNEM/CO by applying it to Denver for a four-
month period during the winter of 1982-83. Exposure estimates made by the model
were compared to measured personal exposures obtained in Denver during the same
time period. The results of this analysis suggest that pNEM/CO provides relatively
good estimates of 1-hour daily maximum CO exposures, but tends to underestimate
the occurrence of high 8-hour daily maximum CO exposures. This bias may be
corrected by increasing the level of autocorrelation within various pN EM/CO
algorithms and by providing special algorithms to represent specific long-term, high-
exposure situations.
73
-------
REFERENCES
1. Richmond, H. M. and T. McCurdy, 1988, "Use of Exposure Analysis and Risk
Assessment in the Ozone NAAQS Review Process," presented at the 81st
Annual Meeting of APCA, Dallas, Texas.
2. Ott, W. R., 1982, "Concepts of Human Exposure to Air Pollution," Environment
International, Vol. 7, page 179.
3. Duan, N., 1982, "Models for Human Exposure to Air Pollution," Environment
International, Vol. 8, page 305.
4. Biller, W. F., T. B. Feagans, and T. R. Johnson, June 1981, "A General Model
for Estimating Exposure Associated With Alternative NAAQS," presented at the
74th Annual Meeting of the Air Pollution Control Association, Dallas, Texas.
5. Paul, R. A. and T. McCurdy, June 1986, "Estimation of Population Exposure to
Ozone," presented at the 79th Annual Meeting of the Air Pollution Control
Association, Dallas, Texas.
6. Johnson, T. R. and R. A. Paul, 1981, 'The NAAQS Exposure Model (NEM)
Applied to Paniculate Matter," prepared by PEDCo Environmental, Inc. for the
Office of Air Quality Planning and Standards, U. S. Environmental Protection
Agency, Research Triangle Park, North Carolina.
7. Johnson, T. R. and R.A. Paul, 1983, The NAAQS Exposure Model (NEM)
Applied to Carbon Monoxide," prepared by PEDCo Environmental, Inc. for the
Office of Air Quality Planning and Standards, U. S. Environmental Protection
Agency.
8. Paul, R, A., T. R. Johnson, and T. McCurdy, 1988, "Advancements in Estimating
Urban Population Exposure," presented at the 81st Annual Meeting of the Air
Pollution Control Association, Dallas, Texas.
9. Pandian, M. D., 1987, "Evaluation of Existing Total Human Exposure Models,"
EPA-600/4-87-004, U. S. Environmental Protection Agency, Us Vegas, Nevada.
74
-------
10. Ryan, P. B., "An Overview of Human Exposure Modeling," submitted to
Toxicology and Industrial Health.
11. Richmond, H., P. Johnson, and T. McCurdy, 1984, "Review of the NAAQS for
Carbon Monoxide: Reassessment of Scientific and Technical Information," EPA-
450/5-84404, U. S. Environmental Protection Agency, Research Triangle Park,
North Carolina.
12.. McKee, D., P. Johnson, T. McCurdy, 1989, "Review of the National Ambient air
Quality Standard for Ozone: Assessment of Scientific and Technical
Information," U. S. Environmental Protection Agency, Research Triangle Park,
North Carolina.
13. Johnson, T. R., R. A. Paul, J. E. Capel, and T. McCurdy, 1990, "Estimation of
Ozone Exposure in Houston Using a Probabilistic Version of NEM," presented
at the 83rd Annual Meeting of the Air and Waste Management Association,
Pittsburgh, Pennsylvania.
14. Johnson, T. R., 1984, "A Study of Personal Exposure to Carbon Monoxide in
Denver, Colorado," EPA-600/54-84-014, U. S. Environmental Protection Agency,
Research Triangle Park, North Carolina.
15. Hartwell, T.D. et al., 1984, "Study of Carbon Monoxide Exposure of Residents of
Washington, D. C. and Denver, Colorado," EPA-600/54-84-031, U. S.
Environmental Protection Agency, Research Triangle Park, North Carolina.
16. Johnson, T. R., 1987, "A Study of Human Activity Patterns in Cincinnati, Ohio,"
Electric Power Research Institute, Palo Alto.
17. Johnson, T. R. and L. Wijnberg, 1981, 'Time Series Analysis of Hourly Average
Air Quality Data," presented at the 74th Annual Meeting of the Air Pollution
Control Association, Philadelphia, Pennsylvania.
18. McDonnell, W. F., D. H. Horstman, and M. J. Hazuch, 1983, "Pulmonary Effects
of Ozone Exposure During Exercise: Dose-Response Characteristics," Journal
of Applied Physiology, Vol. 54, page 1345.
19. Trim, S. C. Environmental Health Service, Rancho Los Amigos Medical Center,
Inc., Downey, California, February, 1990, Personal Communication.
20. Erb, B. D., 1981, "Applying Work Physiology to Occupational Medicine,"
Occupational Health Safety, Vol. 50, pp. 20-24.
75
-------
21. Astrand, P. 0. and K. Rodahl, 1977, Textbook of Work Physiology, 2nd ed.
McGraw-Hill, New York, New York.
22. Biller, W. F. and H. M. Richmond, September, 1991, "COHB Module for a
Probabilistic CO NEM."
23. Coburn, R. F., R. E. Forster, and R. B. Kane, 1963, "Considerations of the
Physiology and Variables That Determine Blood Carboxyhemoglobin
Concentration in Man," Journal of Clinical Investigations, Vol. 44, pp. 1899-
1910.
24. Bureau of Census, 1982, "Census Population and Housing, 1980: Summary
Tape File 3-A," Washington, D. C.
25. Johnson, T. R., J. E. Capel, and D. M. Byrne, 1991, The Estimation of
Commuting Patterns in Applications of the Hazardous Air Pollutant Exposure
Model (HAPEM)," presented at the 84th Annual Meeting of the Air and Waste
Management Association, Vancouver, Canada.
26. Richmond, H.R. "Sensitive Population Estimates for Use in Carbon Monoxide
Exposure Analysis" memorandum to Thomas McCurdy, March, 1991, U. S.
Environmental Protection Agency, Research Triangle Park, North Carolina. .
27. Vital and Health Statistics: Current Estimates from the National Health Interview
Survey, 1989, Series 10, No. 176, DHHS Publication No. PHS 90-1504, U. S.
Department of Health and Human Services, Hyattesville, Maryland.
28. 7990 Heart and Stroke Facts, American Heart Association, page 13, 1990.
29. Nagda, N. L, H. E. Rector, and M. D. Koontz, 1987, Guidelines for Monitoring
Indoor Air Quality, Hemisphere Publishing Corporation, Washington, D. C.
30. Grimsrud, D. T., M. H. Sherman, and R. C. Sondregger, 1982, "Calculating
Infiltration: Implications for a Construction Quality Standard," Proceedings of
the ASHRAE/DDE Conference: Thermal Performance of the Exterior Envelopes
of Buildings II, Las Vegas, Nevada.
31. Turk, B. H., D. T. Grimsrud, J. T. Brown, K. L Geisling-Sobotka, J. Harrison,
and R. J. Prill, 1989, "Commercial Building Ventilation Rates and Particle
Concentrations," ASHRAE Transactions, Vol. 95, Part 1.
32. Hayes, S. R., 1991, "Use of an Indoor Air Quality Model (IAQM) to Estimate
Indoor Ozone Levels." Journal of the Air and Waste Management Association,
Vol. 41, pp.161-170.
76
-------
33. Peterson, G.A. and R.H. Sabersky, 1975, "Measurements of Pollutants Inside an
Automobile," Journal of the Air Pollution Control Association. Volume 25, No.
10, pp: 1028 - 1032.
34. Koontz, M.S., M.A. Mehegan, and N. Nagda, "Distribution of Cooking
Appliances and Impact on Indoor Air Quality," Topical Report No. 5086-245-
1265, Gas Research Institute, Chicago, in preparation.
35. Data obtained from Bruce Smith, Quantum Consulting, Inc., 2030 Addison
Street, Berkeley, California, February 21, 1992.
36. Davidson, C.L, J.E. Borrazzo, C.T. Hendrickson, 1987, "Pollutant Emission
Factors for Gas Stoves," EPA/600/9-87-005, U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina.
37. Public Service Commission of Denver, "1989 Residential Energy Survey,"
Denver, Colorado, 1990.
38. Liss, W.E. and W.H. Thrasher, "Variability of Natural Gas Composition in Select
Major Metropolitan Areas of the United States," Interim Report Contract No.
5089-293-1848, Gas Research Institute, Chicago, 1991.
39. DuBois, D. and E.F. DuBois, 1916, "A Formula to Estimate the Approximate
Surface Area if Height and Weight be Known," Archives of International
Medicine, Vol. 17, pp. 863-871.
77
-------
APPENDIX A
DENVER FIXED-SITE MONITORS
A-1
-------
The following site descriptions were obtained from a report by Johnson14
describing the Denver personal monitoring study and from telephone conversations
with Frank Martelli on March 14, 1991.
CAMP (Site A): 2105 Broadway
The Continuous Air Monitoring Project (CAMP) station is the oldest monitoring
station in the Denver area. It is located in the central downtown region in flat terrain at
an altitude of 5220 feet MSL. This station is under a center-city commercial
designation exhibiting the major qualities of a heavy traffic oriented site. Major traffic
intersections and arteries exist within a few hundred meters in all directions. Carbon
monoxide values are significantly affected by local traffic. The station location and
traffic orientation is generally representative of the large downtown central business
district (CBD) with poor ventilation and high density traffic. To the east of the station
some residential housing starts. In the other directions, one to three story buildings
are present within one half block. The station itself is located on a triangular piece of
park land at the three-way intersection of Broadway, Champa, and 21st Streets.
Carriage (Site B): 2325 Irving
The Carriage site derives its name from its location in the middle of a city block
(i.e., along the alley at mid-block) on a plot of ground owned by Denver and known as
a "carriage lot." This site is at an elevation of 5300 feet MSL and is 2.5 miles west of
the downtown CBD. The terrain is relatively flat and residential housing surrounds the
site at distances of approximately 75 to 100 feet. The site is isolated enough that it is
unaffected by localized traffic patterns. Topographically, the site is located along a
ridge of the Platte River Valley that cuts through the CBD area. However, it is only 80
to 100 feet above the valley floor and well within the influence of major traffic influx into
and out of the CBD.
A-2
-------
National Jewish Hospital East (Site C): 14th and Albion
The National Jewish Hospital (NJH) East site is on the edge of a low influx
parking lot (i.e., filled and emptied once/day) surrounded by generally open space in
all directions for at least 100 feet. The station is three miles east of the downtown
CBD at an elevation of 5300 feet MSL. To the west one of Denver's major arteries,
Colorado Boulevard, is located at a distance of 200 feet from the station. Less major
streets are within 100 feet to the south and 600 feet to the north. This station would
fall into the worst-case neighborhood classifications and generally reflects levels
influenced by the high density traffic on Colorado Boulevard and Colfax Avenue to the
north. Generally flat terrain exists with a three story building being located 100 feet to
the north. The area is primarily residential with nor commercia sources in the vicinity.
Arvada (Site L): 5701 Garrison
The Arvada station is classified as a suburban residential site. This station is
located about 6 miles northwest of the central downtown area at an elevation of 5385
feet MSL This site is considered to be in relatively rugged terrain compared to the
other monitoring stations. The surrounding hills could influence local pollutant values
to a certain degree, depending upon existing meteorological conditions. The source
environment of both the near and far (up to 2 miles) vicinity shows no major industrial
sources that would significantly affect carbon monoxide. A number of indirect sources
do exist less than one mile from the station which appear to periodically influence the
carbon monoxide levels. These indirect sources are 1) major traffic intersections and
2) large shopping centers.
Highlands (Site M): 8100 South University Boulevard
The Highlands air monitoring station is located 12.5 miles south-southeast of
the downtown CBD, at the corner of County Line Road and University Boulevard. The
station itself is set back from both highways by a distance of over 500 feet and is at
the top of a rolling hill. The area surrounding the station is fenced, with the only traffic
being general maintenance personnel within the Water Department grounds. Terrain
A-3
-------
is generally rolling hill areas with the station elevation being 5700 feet above mean sea
level (MSL). Within one mile of the station in all directions are residential housing
areas. This site is the highest monitoring station in Denver. Because of the elevation
and location on top of a hill, it continually receives good ventilation and generally low
pollution readings.
A-4
-------
APPENDIX B
ASSIGNMENT OF DIARY ACTIVITY CODES
TO ACTIVITY CLASSIFICATIONS
B-1
-------
TABLE B-1. BREATHING RATE CATEGORIES OF ACTIVITIES IN THE
CINCINNATI STUDY
Activity
Code
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Description of Activity
All destination - oriented travel
(including walking)
Income • related work
Day - care
Kindergarten - 12th grade
College or trade school
Adult education and special training
Homework
Meal Preparation and cleanup
Laundry
Other indoor chores
Yard work and outdoor chores
Child care and child - centered activities
Errands and shopping
Personal care outside home (doctor,
hair dresser)
Eating
Sleeping
Other personal needs
Religious activities
Meetings of clubs, organizations,
committees, etc.
Other collective participation
Breathing
Rate
Category
B
B
C
C
C
C
C
C
B
B
A
C
C
C
C
D
C
C
C
C
B-2
(continued)
-------
(Table B-1 continued)
Activity
Code
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
Description of Activity
Spectator sports events
Movies, concerts, and other
entertainment events outside home
Cafe, bar, tea room
Museums and exhibitions
Parties and receptions
Visiting friends
Recess and physicial education
Active sports and games outside
school, including exercises and
aerobics
Hunting, fishing, hiking
Jogging or bicycling
Taking a walk
Artistic creations, music, and hobbies
Other active leisure
Reading
Television or radio
Conversation and correspondence
Relaxing, reflecting, thinking (no visible
activity)
Other passive leisure
Asthma attack
Other sudden illness or injury
Breathing
Rate
Category
B
C
C
B
B
C
A
A
B
A
A
C
A
C
C
C
C
C
C
C
B-3
(continued)
-------
(Table B-1 continued)
Activity
Code
43
44
45
Description of Activity
Interview
Wakeup
Baby crying
Breathing
Rate
Category
C
C
A
B-4
-------
TABLE B-2. BREATHING RATE CATEGORIES OF ACTIVITIES
IN THE DENVER STUDY
Activity
Code
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Description of Activity
All travel
Work (income - related) and
study
Cooking
Laundry
Other indoor chores and child
care
Yard work and other outdoor
activities
Errands and shopping
Eating
Sleeping
Other personal needs
Social, political, or religious
activities
Cafe or pub
Walking, bicycling, or jogging
(not in transit)
Other leisure activities
Breathing
Rate
Category
B
C
C
B
B< 30 min.
C=> 30 min.
A
C
C
D
C
C
C
A
C
B-5
-------
TABLE B-3. BREATHING RATE CATEGORIES OF ACTIVITIES
IN THE WASHINGTON STUDY
Activity
Code
1
2
3
4
5
6
7
8
9
11
12
13
14
15
16
17
Description of Activity
Transit, travel
Work, business meetings
Cooking
Laundry
Inside house - chores
Outside house - chores
Errands, shopping, etc.
Personal activities
Leisure activities
Sleeping
School, study
Eating, drinking
Sports and exercise
Church, political meetings, etc.
Inside house - miscellaneous
In parking garage or lot
Breathing
Rate
Category
B
B
C
B
B
A
C
C
C
D
C
C
A
C
C
B-6
(continued)
-------
(Table B-3, continued)
Activity
Code
18
19
21-36
89
Description of Activity
Outside, not otherwise specified
Doctor or dentist office
Same as Activities 1-16 including
suspected sleep
.Any other activity
Breathing
Rate
Category
B
C
See codes
1 -16
C
B-7
-------
APPENDIX C
THE CARBOXYHEMOGLOBIN MODULE
C-1
-------
COHB MODULE FOR A PROBABILISTIC CO NEM
W. F. Biller and H. M. Richmond
May 20, 1992
The original version of the carbon monoxide (CO) national ambient
air quality standard (NAAQS) exposure model (NEM), referred to as CO-
NEM, has been upgraded to include the probabilistic treatment of a
number of its inherently stochastic input variables. The new model,
pNEM/CO, just as the original version, includes a module for the
computation of carboxyhemoglobin (COHb) blood level distributions in
exposed populations. The upgraded module for the new version is also
probabilistic. This report describes the probabilistic COHb module
and discusses its basis.
I THE BASE PHYSIOLOGICAL MODEL FOR COMPUTING COHb LEVELS
The COHb module in the original CO-NEM (Johnson and Paul, 1983)
used as its basic model the differential equation derived by Coburn,
Forster, and Kane (1965) which described the dynamic relationship
between instantaneous blood levels of COHb, inspired CO, and other
physiological variables. This model, which will be referred to here
as the CFK model, continues to be the most widely used and is the
basic model for COHb computations in pNEM/CO. The CFK model is
described in Section II.
The CFK model describes the rate of change of COHb blood levels as
a function of the following quantities:
1. Inspired CO pressure
2. COHb level
3. Oxyhemoglobin (O2Hb) level
4. Hemoglobin (Hb) content of blood
5. Blood volume
6. Alveolar ventilation rate
7. Endogenous CO production rate
8. Mean pulmonary capillary oxygen pressure
9. Pulmonary diffusion rate of CO
10. Haldane coefficient (M)
11. Barometric pressure
12. Vapor pressure of water at body temperature (47 torr)
-------
If all of the listed quantities except COHb level are constant over
eome time interval, the CFK equation has a linear form over the
interval and is readily integrated. The solution to the linear form
gives reasonably accurate results for lower levels of COHb. However,
CO and oxygen compete for the available hemoglobin and are therefore
not independent of each other. If this dependency is taken into
account, the resulting differential equation is no longer linear.
Peterson and Stewart (1975) proposed a heuristic approach to dealing
with this dependency which assumed the linear form and then adjusted
the 02Hb level iteratively based on the assumption of a linear rela-
tionship between the 02Hb and COHb. This approach was used in the
COHb module of the original CO-NEM. Alternatively, once a relation-
ship between COHb and O2Hb is assumed it also possible to determine
COHb at any time by numerical integration of the nonlinear CFK equa-
tion (e.g. by use of the Runge-Kutta method). Muller and Barton
(1987) demonstrated that assuming a linear relationship between COHb
and O2Hb leads to a form of the CFK equation equivalent to the Michae-
lis-Menton kinetic model which is analytically integrable. However,
the analytical solution in this case cannot be solved explicitly for
COHb. Muller and Barton demonstrated a binary search method for
determining the COHb value.
The COHb module for pNEM/CO employs a linear relationship between
COHb and 02Hb which is consistent with the basic assumptions of the
CFK model but differs from the linear forms used by other modelers.
The Muller and Barton (1987) solution is employed. However, instead
of the simple binary search described in the Muller and Barton paper,
a combination of the binary search and Newton-Raphson root finding
methods was used to solve for COHb (Press et al., 1986). Using the
Muller and Barton solution increased computation time compared to the
Peterson-Stewart method but was shown to be faster than a fourth order
Runge-Kutta numerical integration.
-------
II THE CFK MODEL FOR ESTIMATION OF CARBOXYHEMOGLOBIN
Table I defines the variables which appear in the equations of this
section. Coburn, Forster, and Kane (1965) derived the following
differential equation governing COHb levels in the blood upon exposure
to CO.
d[COHb] = Vco + ^Jco _ ^"Wcot (1)
dt Vb BVb [02HJb]MBVb
where
B = -i- * Pg . Ptf2o) (2)
If the only quantity in this equation that can vary with time is
[COHb], the CFK equation is linear and can be readily integrated.
However, since oxygen (O2) and CO compete for the available Hb, [COHb]
^nd [O2Hb] must be related. Increasing [COHb] will result in decreas-
Png [O2Hb]. Thus the CFK equation is not linear and requires the
relationship between the two quantities to be known if it is to be
accurately integrated over a wide range of COHb levels.
Various linear relationships between [COHb] and [O2Hb] have been
used (See Marcus, 1980; McCartney, 1990; Muller and Barton, 1987; and
Tikuisis et al., 1987). A relationship not previously used follows
directly from the basic assumptions of the CFK model. The CFK model
employs the Haldane coefficient which is the equilibrium constant
associated with the following reaction depicting the replacement of O2
in 02Hb by CO:
CO + 02Hb = 02 + COHb (3)
At equilibrium the following equation applies approximately:
P- [COHb]
Pc(02Hb]
= M (4)
'•CO
Equation (4) is the Haldane relationship. The Haldane coefficient, M,
-------
TABLE I
DEFINITIONS OF CFK MODEL VARIABLES
t
[COHb]
[02Hb]
[RHb]
[COHb]0
[RHb]0
%[COHb]
%[02Hb]
%[COHb]
%[COHb]
Pc
HO
Lm
M
k
Vb
Hb
%MHb
Time from start of an exposure event, min
Concentration of carboxyhemoglobin (COHb) in blood at
time,t, ml CO per ml blood at STPD
Concentration of oxyhemoglobin (02Hb) in blood at time t,
ml 02 per ml blood at STPD
Concentration of reduced hemoglobin in blood as equivalent
ml CO per ml of blood at STPD
[COHb] at t = 0
[RHb] + [COHb] + [02Hb]
[COHb] expressed as percent of [RHb]0
[02Hb] expressed as percent of [RHb]0
%[COHb] at t = 0
%[COHb] at t = co
Pressure of inspired CO in air saturated with water vapor at
body temperature, torr .
Mean pulmonary capillary CO pressure, torr
Mean pulmonary capillary 02 pressure, torr
Barometric pressure, torr
Vapor pressure of water at body temperature, torr
(47 torr)
Alveolar ventilation rate, ml/min STPD
Endogenous CO production rate, ml/min STPD
Pulmonary CO. diffusion rate, ml/min/torr STPD
Haldane coefficient
Equilibrium constant for reaction O2 + RHb = O2Hb
Blood volume, ml
Total hemoglobin in blood, g/ 100ml
Methemoglobin as weight percent of Hb
-------
is the chemical equilibrium constant for reaction
(3)
The above reaction can also be viewed as the difference between two
competing chemical reactions:
CO + RHb = COHb (5)
O2 + RHb = 02Hb (6)
Subtracting (6) from (5) yields (3). If (3) is in equilibrium, then
(5) and (6) are in equilibrium. If k is the equilibrium constant for
(6) then:
[02Hb]
Pc [RHb]
°2
It is known that an individual breathing air free of CO for an
extended period will have about 97% of the reactive hemoglobin tied up
as 02Hb and the rest (3%) as RHb. It is also known that at one
atmosphere barometric pressure the mean pulmonary capillary oxygen
pressure is approximately 100 torr. Substituting into (7) yields 0.32
as the approximate value of k at body temperature. From mass balance
considerations:
[02Hb] + [COHb] + [RHb] = [RHb] 0 (8)
Eliminating [RHb] between (7) and (8) and solving for [02Hb] yields:
kPc
[02Hb] = ^— ([RHb] 0 - [COHb]) (9)
1 + kPc
This equation is the desired linear relationship. It has the same
form as a relationship given without explanation by McCartney (1990)
but replaces the constant in the McCartney equation by the term in (9)
involving the mean pulmonary capillary oxygen pressure and the equi-
librium constant k. Substituting (9) into (1) yields a CFK equation
free of [02Hb] and fully consistent with Coburn, Forster, and Kane's
original derivation.
-------
d[COHb] _ Vco + £co _ [COHb] c°* (10)
dt Vj, BVi, [£Hb]0- [COHb] kMBVb
In working with the CFK model it is convenient to express COHb as a
percent of [RHb]0. Multiplying (10) by 100 and dividing by [RHb]0:
d% [COHb] _ 100 VCQ + ££« _ % [COHb]
dt ~ [RHb]0[ Vb BVbj 100 - %[COHb] k[RHb]
Equation (11) can be written in the form suggested by Muller and
Barton (1987):
d% [COHb] _ c _ c % [COHb]
dt ~oi 100 _ % [COHb]
where
on ( ir P \
C = 10° V™ + ±£2 (13)
0 [RHb]0( Vh BVb\
_
1
k[RHb]aMBVb
Given values for the atmospheric pressure and the physiological
variables in equations (12) - (14) the value of %[COHb] at time t can
be found by numerical integration using such techniques as the fourth
order Runge-Kutta method (Press et al., 1986)
Muller and Barton (1987) pointed out that an equation of the form
of (12) is equivalent to a Michaelis-Menton kinetics model which is
integrable. Integration yields:
- (C0 + q) t + % [COHb] -% [COHb] 0
The equation for %[COHb]. is obtained by setting equation (12) equal
-------
to zero and solving for %[COHb] which is now equal to %[COHb]m:
<">
Equation (15) cannot be solved explicitly for %[COHbj. The Muller and
Barton paper suggests the binary search method as one way to find the
value of %[COHb]. Press and coauthors (1986) contend a combination of
the binary search and Newton-Raphson methods is faster on average.
-------
Ill APPLICATION OF THE BASIC COHb MODEL IN pNEM/CO
Description of pNEM/CO
pNEM/CO follows the daily activities over an extended period of a
finite set of cohorts residing within a given geographic area. The
period may be a single season or a calendar year. The cohorts are
defined such as to be representative of a group of individuals sensi-
tive to ambient levels of CO they may encounter as they go about their
daily activities. Individuals who smoke are not included in this
group. The exposure of each cohort over the period of interest is
followed by dividing the period into a contiguous set of exposure
events. An exposure event is a period in which the individual resides
in a single environment, at a single breathing rate class, within a
single clock hour. That is, passage from one exposure event to the
next is indicated by a change in microenvironment, breathing rate
class, or clock hour. Thus no exposure event can last more than one
hour. Nor can it span a clock hour. Given the district, the micro-
environment, and the time and duration of the exposure event the model
computes the average.CO exposure concentration for the event. Since
the exposure events for a cohort are contiguous the model can combine
these concentrations to output distributions of one-hour and running
eight-hour exposures for each cohort. Then, taking into account the
population represented by each cohort, the individual cohort exposure
distributions can be combined into exposure distributions for the
total sensitive population.
To treat the daily behavior of cohorts probabilisticly, cohorts are
identified according to their home district, work district, demograph-
ic group, and the use of cooking fuel in their residences. Currently
seven demographic groups are used 'to distinguish cohorts by sex and
age. A set of pools of 24-hour activity patterns is used based on
activity data drawn from studies performed in Cincinnati (Johnson,
1983), Washington D. C.(Hartwell et al., 1984), and Denver (Hartwell
et al., 1984). These patterns are twenty-four hour samplings of the
behavior of real people. The patterns in each pool represent the same
demographic cohort group, day type (weekday, weekend day), and ambient
temperature range. Each pattern consists of a 24-hour set of contigu-
ous exposure events defined by event start time, duration, microenvi-
ronment, and breathing rate class. For a given cohort a daily pattern
is randomly sampled from the appropriate pool each day during the
-------
period of the computation.
From the description in the preceding paragraph it is apparent that
while the cohort represents a demographic group, its 24-hour activity
pattern comes from a different member of that group each day. This
feature has an impact on the design of the COHb module.
The COHb Module
The COHb module of pNEM/CO employs the Muller and Barton (1987)
integration of the CFK model as represented by equations (12)-(14) to
compute the.COHb level of a cohort at the end of each exposure event.
To perform this computation it requires information on each of the
quantities listed in the section discussing the CFK model. In addi-
tion the COHb level at the beginning the exposure event must be known.
This latter quantity is usually the COHb level computed at the end of
the previous contiguous exposure event. To obtain the initial COHb at
the start of the exposure period the computation is started one day
before the beginning of the period. The effect of the initial COHb
value on the end value is negligible after about 15 hours. The
program stores the COHb levels at the end of each clock hour and
outputs distributions of COHb levels for the sensitive population.
Assignment of CFK Model Input Data for an Exposure Event
Section IV describes the equations and procedures used by the
pNEM/CO COHb module to obtain the values of the input variables to
equations (2) and (13) through (16). A brief overview is given here.
The actual inspired CO level can change significantly.
during an exposure event. The model supplies an average exposure
concentration for the event, which is used as the CO input. The time
constant for the change in COHb is sufficiently large that the use of
concentrations based on averaging times up to one hour can be used in
place of the instantaneous concentrations over the averaging time
period with little loss of accuracy in estimating the COHb level at
the end of the exposure event. Furthermore, applying the average
concentrations to a contiguous sequence of exposure events does not
cause an accumulation of error.
The model as presently used in pNEM/CO does not take into account
changing barometric pressure. It uses a constant barometric pressure
which is a function of the average elevation of an area above sea
-------
level. The pressure at sea level is taken to be 760 torr.
The remaining input variables to the CFK model are all physiologi-
cal. While the Haldane coefficient, the equilibrium constant k and
average pulmonary capillary oxygen pressure are treated as having the
same constant values for all cohorts, the remaining physiological
input variables will vary among individuals. As discussed in the
previous section a characteristic of pNEM/CO is that a new selection
is made each day of the pool member who will represent the cohort
group for that day. Therefore a new set of several of the physiologi-
cal parameters is selected each day. Because the variables may vary
with age, weight and height of an individual, the COHb module first
randomly selects an age, weight, and height for the individual that is
within the parameters of the cohort group. The basis for randon
selection and manner in which it is done are covered in Section IV.
Once the individual's age, weight, and height are determined the
physiological parameters can be determined using suitable correla-
tions and distributions. The procedures for each of the variables are
given along with the basis for the procedures.
10
-------
IV COMPUTATION OF INPUT DATA FOR THE COHB MODULE
The methods described in this section follow closely those used by
Biller and Richmond (1982) in a sensitivity analysis of the CFK equa-^
tion. The literature data in this study have been updated. The
updating has resulted in changes in the values of a number of the
equation parameters and in some cases the forms of the equations given
in this section.
pNEM/CO selects a cohort defined by: home district, work district,
demographic group, and residential use of cooking fuel. Of these only
the demographic group descriptors are used in the COHb module. The
demographic group is defined by: sex, age group, and work status. Of
these only sex and the age groups 18 to 44, 45 to 64 and 65+. are used
by the COHb module.
For each cohort, as defined above, pNEM/CO computes exposure for a
contiguous sequence of exposure events spanning the total time period
of the computation (the calendar year 1988). The sequence of exposure
events is determined by random sampling from a set of pools of 24-hour
activity patterns each exposure day. An individual 24-hour pattern in
one of these pools is referred to as a unit exposure sequence (UES).
Each pool consists of a collection of UESs which are specific to the
cohort demographic group, day type, and average daily temperature.
A UES is a contiguous set of exposure events spanning 24 hours.
Each event is characterized by start time, duration in minutes,
home/work status, microenvironment, and breathing rate. All exposure
events are constrained to occur entirely within a clock hour. The
UESs start at 7:00 p.m. and end at 7:00 p.m. the following day."
11
-------
The CFK model within the COHb module is called for each exposure
event. For each event it requires the following data:
Time duration of event, min
Inspired CO partial pressure averaged over the event, torr
Percent COHb at the start of the event
Alveolar ventilation rate, ml/min STPD
Average pulmonary capillary oxygen pressure, torr
Haldane Coefficient
Equilibrium constant for the reaction of 02 and RHb
Atmospheric pressure, torr
Blood volume, ml
Total potential reduced hemoglobin content of blood, ml CO/ml STPD
Pulmonary CO diffusion rate, ml/min/torr STPD
Endogenous CO production rate, ml/min STPD
Given these data as input the module computes the percent COHb at the
end of the exposure event. This value is used by the module as the
initial percent COHb for the next contiguous exposure event. The main
program retains only those COHb values at the end of each clock hour.
Some of the above data do not change during a pNEM/CO computer run
and therefore need to be supplied to the computer program only once at
the start. Some of the data vary with the cohort and therefore need
to be supplied at the beginning of each activity day. Other data tend
to change with the exposure event and therefore need to be supplied
for each new exposure event.
INPUT DATA SUPPLIED AT START OF THE PNEM/CO COMPUTATION
Barometric Pressure
A constant barometric pressure is assumed for the study area based
on the average height above sea level:
PB = 760*exp(-0.0000386*Altitude) (17)
where altitude is the average height (in feet) of the study area above
sea level (USEPA, 1978).
12
-------
Average Pulmonary CapillarY Oxygen Pressure
A substantial effort was expended on developing an equation for Pc
which accurately captures the effect of barometric pressure on this
parameter. Time constraints were such that the work could not be
completed in time to include such an equation in the COHb module.
Consequently, for the present, the equation employed is based on an
approximation used by Peterson and Stewart (1975) in which the 49 torr
is subtracted from the partial pressure of inspired oxygen. This
leads to the following approximate relationship:
Pc = 0.209(PB - 47) - 49 (18)
The constant 0.209 is the mole fraction of O2 in dry air. The constant
47 is the vapor pressure of water at body temperature. This expres-
sion was used in an investigation of the CFK equation by Tikuisis et
al. (1987). Modelers have tended to use the value 100 torr. Equation
(18) gives the value 100 torr for a barometric pressure of 760 torr.
Haldane Coefficient
The value of 218 will be used for the Haldarie coefficient. Mea-
sured values in the range 210 to 270 have been reported in the litera-
ture. Modelers have tended to use values in the range 210 to 240.
In the early 1980's The Clean Air Scientific Advisory Committee
expressed the opinion to EPA (Friedlander, 1982) that .the most careful
work done in this area was that by Rodkey (1969) who determined a
value of 218. This value was used in the COHb module of the earlier
CO NEM version. Other modelers using values in the range 218 to 220
are Peterson and Stewart, 1970; Marcus, 1980; Collier and Goldsmith,
1983; Muller and Barton, 1987. Since the value 218 continues to be
within the range used by modelers it has been decided to continue with
this value in pNEM/CO.
Equilibrium Constant for the Reaction of 07 and RHb
This quantity was estimated in Section II to have the value 0.32
based on the observation that %[RHb] is about 3% in individuals
breathing air which is free of CO and a value of 100 torr for P~ .
u
13
-------
INPUT DATA SUPPLIED AT START OF EACH 24-HOUR PERIOD
The last four quantities in the above list of input data are more
persistent physiological characteristics of the cohort. Since each
UES is based on the activities of an actual person, in effect the
cohort is representing a different person each day. It is reasonable
therefore to alter the four physiological characteristics of the
cohort each day and maintain them as constant for that day. To
determine these physiological parameters it is first necessary to
assign a weight and height and in one case age to the person repre-
senting the cohort for the 24-hour period.
Determination of Height
The distribution of heights is assumed to be given by a normal
distribution whose mean and standard deviation vary with sex and age
group as follows:
Distribution of Height in Inches
MEN --WOMEN
Mean Standard Mean Standard
Age Group Height Deviation Height Deviation
18 to 44 69.5 2.8 64.2 2.5
45 to 64 68.6 2.6 63.2 2.4
65+ 67.3 2.6 62.3 2.4
The data in this table were obtained as part of the DHEW National
Health and Nutrition Examination Survey (NHANES) for the period 1971-
1974 (USDHEW, 1976). These are the most recent data that could be
obtained in the time available. Subsequent NHANES studies did not
provide these data in a form that could be used here.
To use the table a random number is drawn from a standardized
normal distribution which will be referred to as N(0,l) where 0 is the
value of the mean and 1 is the value of the standard deviation. The
random variate being sampled is called the standardized normal vari-
ate, z, which can be thought of as representing the deviation from the
mean expressed as a multiple of the standard deviation. The height is
then obtained by multiplying the generated random number by the corre-
sponding standard deviation from the above table and algebraically
- 14
-------
adding the corresponding mean height. If z is the random number
generated from the distribution N(0,l), the height of the cohort is
given by:
height = (mean height) + z*(standard deviation) (19)
Determination of Body Weight
The average weight of individuals tends to vary with height, sex,
and age according to the following regression model:
weight = AO + A,*height + z*(S.E.) (20)
AO and A, are regression coefficients. S.E. is the standard error of
estimate for the regression. z is the standardized normal variate
randomly sampled from'the normal distribution N(0,l).
The height has already been determined as described above. A new z
must be obtained. The regression coefficients and standard error of
estimate are obtained from the following table:
Regression Expressions Relating Sex Height and Age Class to Weight
Regression Coefficients Standard Error
Sex and Age AO A! of Estimate
MEN:
18 - 44
45 - 64
65+
WOMEN :
18 - 44
45 - 64
65+
-168.67
-131.83
-131.64
-88.62
-77.17
-76.38
4.941
4.454
4.385
3.587
3.587
3.583
-
30.5
28.4
26.0
32.1
33.8
29.0
With the above coefficients height is in inches and the calculated
weight is in pounds. These data were obtained from the 1971-1974
NHANES study (USDHEW, 1977)
15
-------
Blood Volume
Blood volume for the cohort is calculated as follows:
Men: Vb = 20.4*weight + 0.00683*(height)3 - 30 (21)
Women: Vb = 14.6*weight + 0.00678*(height)3 - 30 (22)
where Vb is in milliliters, weight in pounds and height in inches.
These expressions were obtained by Allen et al. (1956) and were
modified to accept the units used for height and weight.
Total Reduced Hemoglobin in the Absence of O, and CO
The quantity [RHb]0 is expressed as equivalent milliliters of O2 or
CO at STPD per milliliter of blood. Total Hb blood levels are custom-
arily expressed as grams per deciliter of blood. The total Hb level
in the absence of COHb and O2Hb would consist principally of RHb which
can react with O2 or CO and MHb which cannot. Total Hb blood levels
also tend to be higher in people living at higher altitudes. There-
fore to relate [RHb]0 to Hb it is necessary to correct for the MHb
present, adjust for the effect of altitude, and convert to equivalent
milliliters of CO at STPD. The later conversion is based on the
observation that a gram of reduced Hb can react with a maximum of 1.39
ml of O2 or CO at STPD. The application of these" three factors yields
the equation:
[RHb]0 = 1.39*Hb(100 - %MHb)*(l + HbAlt*Altitude)/100 (23)
where HbAlt is a regression constant. Hb in equation (23) is a sea
level value. Hb level in a human population is normally distributed
with the mean Hb and standard deviation both dependent on sex and age
class according to the following table.
16
-------
Distribution of Blood levels of Hb by Sex and AGE
MEN WOMEN
Mean Standard Mean Standard
Age Group Hemoglobin Deviation Hemoglobin Deviation
18 to 44
45 to 64
65+
15.3
15.1
14.8
1.0
1.2
1.4
13.3
13 . 6
13.7
1.1
1.2
1.2
These data were obtained from the 1976-1980 NHANES study (USDHHS,
1982). Given the preceding table the hemoglobin content of the blood
is calculated in the same manner as the height of the cohort and
substituted into equation (23).
The weight percent MHb, %MHb, is taken to be 0.5% of the weight of
Hb (Muller and Barton, 1987).
The altitude correction factor, HbAlt, was developed by application
of simple regression analyses to Hb data obtained in 17 U.S. cities
(USEPA, 1973).
Men: 0.000161 S.E. = 0.000064
Women: 0.000115 S.E. = 0.000043
Two cities (Phoenix and Houston) were eliminated in the regression
analysis because the neasured hemoglobin levels were substantially
below that of the other cities. . The altitude factor is small. It
predicts about a 5% increase in Hb for residents of Denver over that
for people living at sea level.
Base Pulmonary Diffusion Rate of CO
A base lung diffusivity of CO for the cohort is calculated as
follows:
Men: DLm = 0.361*height - 0.232*age +16.3 (24)
Women: DL = 0.556*height - O.ll5*age - 5.97 (25)
17
-------
where height is in inches and age is in years. To obtain the cohort
age it is assumed the population is evenly distributed within each
weight class. For the 65+ group it is assumed the age is distributed
evenly between 65 and 75.
The regression equations were obtained from a paper by Salorinne
(1976) and modified to conform to the units used in the COHb module.
The Salorinne data were obtained for non-exercising individuals.
Tikuisis et al. (1992) working with eleven male subjects at various
exercise levels showed significant increase in lung diffusivity of CO
with increasing alveolar ventilation rate. Regression analyses on
data provided by Tikuisis for the individual subjects in the study
showed the relationship to be linear. From this relationship and the
heights and ages of the subjects in the Tikuisis et al. study it was
determined that the Salorinne equations (24) correspond to an alveolar
ventilation rate of 6.69 1/min STPD. In the absence of other data it
is assumed that this same value applies to equation (25) for women.
Thus for each twenty-four hour period equations (24) and (25) are used
to compute lung diffusion rates of CO for a base case alveolar venti-
lation rate of 6.69 1/min STPD. As will be seen this value is adjust-
ed to account for the actual ventilation rate experienced by the
cohort during each individual exposure event.
Endogenous Rate of CO Production
The endogenous CO production rates taken from a number of sources
show the rate to be distributed lognormally in the population (see
Appendix for data and sources). The distribution is different for men
and women. For a woman there is a further difference depending on
whether she is in her premenstrual or postmenstrual phase. The
parameters of the lognormal distribution are given in the following
table:
Parameters of the Lognormal Distribution of the Endogenous CO Produc-
tion Rate Among Adults.
Geometric Geometric
Mean Standard Deviation
(ml/hr) (ml/hr)
MEN 0.473 1.316
18
-------
WOMEN
Premenstrual 0.497 1.459
Postmenstrual 0.311 1.459
The calculation is as follows: If the cohort is female, a random
number is sampled from a uniform distribution of real numbers in the
range 0-1. If the number is less than 0.5, the cohort is considered
to be in the premenstrual half of the month. Otherwise she is consid-
ered to be in the postmenstrual period. If the woman is in the 65+
age category she is treated as premenstrual. The appropriate combina-
tion of geometric mean and geometric standard deviation are obtained
from the above table depending on the age class, sex, and menstrual
phase. As in the previous cases a random number, z, is sampled from
the standardized normal distribution, N(0,l). The appropriate endoge-
nous CO production rate is then obtained from:
= 0.01667* (geometric mean) * (geometric S.D.)Z (26)
The constant term converts ml/hr to ml/min.
19
-------
INPUT DATA SUPPLIED WITH EACH EXPOSURE EVENT
Duration of Exposure Event
The duration of the exposure event in minutes is supplied by the
main program to the COHb module.
Partial Pressure of Inspired Carbon Monoxide
The main program supplies the inspired CO concentration averaged
over the duration of the exposure expressed as ppm. This quantity is
converted to pressure via:
PI
-------
Adusted Pulmonary Diffusion Rate of CO
Given the alveolar ventillation rate for the exposure event the
associated adjusted pulmonary diffusion rate can be calculated from:
Z?Lco(Adjusted) = DLm(Ease) + 0.000845^ - 5.65 (30)
(See discussion of base pulmonary diffusion rate.)
21
-------
V REFERENCES
Biller, W. F. and Richmond, H. M. (1982) Sensitivity analysis on
Coburn model predictions of COHb levels associated with alternative CO
standards. Report to Strategies and Air Standards Division of the
Office of Air Quality Planning and Standards, U. S. Environmental
Protection Agency, Research Triangle Park, NC, November.
Brouillard, R. P., Conrad, M. E., and Bensinger, T. A. (1975) Blood
45:67-69.
Burke, P. D., Rodkey, F. L., Blaschke, T. F., Collison, H. A., and
Waggoner, J. G. (1974) Comparison of plasma bilirubin turnover and
carbon monoxide production in man. J. Lab. Clin. Med. 83:29-37.
Coburn, R. F., Forster R. E., and Kane, R. B. (1965) Considerations of
the physiology and variables that determine blood carboxyhemoglobin
concentration in man. J. Clin. Invest. 44:1899-1910.
Coburn, R. F., Blakemore, W. S., and Forster, R. E. (1963) Endogenous
carbon monoxide production in man. J. Clin. Invest. 42:1172-1178.
Coltman, C. A., and Dudley III, G. M. (1969) The relationship between
endogenous carbon monoxide and total heme mass in normal and abnormal
subjects. Am. J. Med. Sci. 258:374-385
Collier, C. and Goldsmith J. R., (1983) Interactions of carbon monox-
ide and hemoglobin at high altitude. Atmos. Environ. 17:723-728
Delivoria-Papadoppulos, M., Coburn, R. F., and Forster, R. E. (1974)
Cyclic variation of rate of carbon monoxide production in normal
women. J. Appl. Phvsiol. 36:49-71.
Dubois, D. and Dubois, E. F. (1916) A formula to approximate the
surface area if height and weight be known. Arch. Int. Med. 17:863-871
Friedlander, S. K. (1982) Letter from The Clean Air Scientific Adviso-
ry Committee to USEPA Administrator. August 31, 1982.
Hartwell, T. D., Clayton, C. A., Richie, R. M., Whitmore, R. W.,
Zelon, H. S., Jones, S. M., and Whitehurst, D. A. (1984) A study of
carbon monoxide exposure of residents of Washington, D. C. and Denver,
Colorado. EPA-600/S4-84-031, U. S. Environmental Protection Agency,
Research Triangle Park, NC.
22
-------
Johnson, T. and Paul, R. A. (1983) The NAAQS model (NEM) applied to
carbon monoxide. EPA 450/5-83-003, U. S. Environmental Protection
Agency, Research Triangle Park, NC.
Johnson, T. (1987) A study of human activity patterns in Cincinnati,
Ohio. Prepared by PEI Associates, Inc. for the Electric Power Re-
search Institute, Palo Alto, CA.
Luomanmaki, K. and Coburn, R. F. (1969) Effects of metabolism and
distribution of carbon monoxide on blood and body stores. Am. J.
Phvsiol. 217(2);354-362.
Lynch, S. R. and Moede, A. L. (1972) Variation in the rate of endoge-
nous carbon monoxide production in normal human beings. J. Lab, din.
Med. 79:85-95.
Marcus, A. H. (1980) Mathematical models for carboxyhemoglobin. Atmos,
Environ. 14:841-844.
McCartney, M. L. (1990) Sensitivity analysis applied to Coburn-
Forster-Kane Models of carboxyhemoglobin formation. Am. Ind. Hyg.
Assoc. J. 51(3):169-177
Merke, C. Cavallin-Stahl, E., and Lundh, B. (1975) Carbon monoxide
production and reticulocyte count in normal women. Effect of contra-
ceptive drugs and smoking. Acta Med. Scand. 198:155-160.
Muller, K. E. and Barton, C. N. (1987) A nonlinear version of the
Coburn, Forster and Kane Model of blood carboxyhemoglobin. Atmos.
Environ. 21:1963-1967.
Peterson, J. E. and Stewart, R. D. (1975) Predicting the carboxyhem-
oglobin levels resulting from carbon monoxide exposures. J. Appl.
Phvsiol. 39:633-638.
Press, W. H., Flannery, B. P., Teukolsky, S. A., and Vettering, W. T.
(1986) NUMERICAL RECIPES, Cambridge University Press.
Salorinne, Y. (1976) Single-breath pulmonary diffusing capacity.
Scand. J. Resp. Diseases Supplementum 96.
23
-------
Tikuisis, P., Madill H. D., Gill, B. J., Lewis, W. F., Cox, K. M.,
Kane, D.M. (1987) A critical analysis of the use of the CFK equation
in predicting COHb formation. Am. Ind. Hyg. Assoc. J. 48(3):208-213.
Tikuisis, P., Kane, D. M., McLellan, T. M., Buick, F., Fairburn, S.
M., (1992) Rate of formation of carboxyhemoglobin in exercising humans
exposed to carbon monoxide. J. Appl.Physiol. 72(4).
U.S. Department of Health and Human Services (1982) Hematological and
nutritional biochemistry reference data for persons 6 months-74 years
of age: United States, 1976-80. DHHS Publication No. (PHS)83-1682.
U.S. Department of Health, Education, and Welfare; Public Health
Service (1976) Height and weight of adults 18-74 years of age in the
United States. Advance Data, No. 3, November 19 1976.
U.S. Department of Health, Education, and Welfare; Public Health
Service (1977) Weight by Height and age of adults 18-74 years: United
States, 1971-74. Advance Data, No. 14, November 30, 1977.
U. S. Environmental Protection Agency (1978) Altitude as a factor in
air pollution. EPA-600/9-78-015, U. S. Environmental Protection
Agency, Research Triangle Park, NC
U. S. Environmental Protection Agency (1990) Air quality criteria for
carbon monoxide. EPA-600/8-90-045A, U. S. Environmental Protection
Agency, Research Triangle Park, NC
Werner, B. and Lindahl, J (1980) Endogenous carbon monoxide production
after bicycle exercise in healthy subjects and in patients with
hereditary spherocytosis. Scand. J. Lab. Invest. 40:319-324.
24
-------
APPENDIX
LITERATURE DATA USED TO DERIVE GEOMETRIC MEAN AND STANDARD DEVIATION
LOGNORMAL DISTRIBUTION OF ENDOGENOUS CO PRODUCTION RATE
TABLE A-l
ENDOGENOUS CO PRODUCTION RATE FOR MEN
Vco
(ml/hr)
0.35
0.35
0.40
0.39
0.43
0.35
0.51
0.42
0.57
0.45
0.40
0.81
0.26
0.65
0.51
0.62
0.44
REFERENCE
Coburn et al., 1963
n
u
n
n
n
n
n
u
n
Lynch and Moede, 1972
n
n
n
u
n
M
25
-------
TABLE A-l
ENDOGENOUS CO PRODUCTION RATE FOR MEN
Vco
(ml/hr)
0.43
0.58
0.52
0.59
0.80
0.72
0.54
0.46
0.26
0.60
0.45
0.39
0.40
0.81
0.57
0.33
0.70
REFERENCE
Berk et al., 1974
it
it
it
it
it
ii
Delivoria-Papadopoules et al., 1974
it
it *
M
ii
it
Brouillard et al., 1975
M
II
II
26
-------
TABLE A-l
ENDOGENOUS CO PRODUCTION RATE FOR MEN
Vco
(ml/hr)
0.58
0.38
0.51
0.55
0.37
0.49
0.45
0.50
0.33
0.45
0.36
0.54
0.76
0.48
0.31
0.70
0.36
0.65
0.38
0.42
0.41
0.54
0.38
REFERENCE
Coltman and Dudley, 1969
M
ii
n
ii
n
n
ii
M
n
n
Werner and Lindahl, 1980
n
n
n
ii
n
n
Luomanmaki and Coburn, 1969
n
n
n
M
27
-------
TABLE A-2
ENDOGENOUS CO PRODUCTION RATE FOR
WOMEN IN PREMENSTRUAL PERIOD
Vco
(ml/hr)
.72
.37
.23
.33
.42
.44
.29
.48
.57
. .54
.72
.99
.48
.53
.43
.64
.86
.35
.52
.80
.54
.68
.28
REFERENCE
Lynch and Moede, 1972
ti
it
ii
it
ii
H
H
Delivoria-Papadopoulos et al., 1974
ii
H
ti
ii
ii
n
Merke et al., 1975
n
n •
n
n
n
n
n
28
-------
TABLE A-3
ENDOGENOUS CO PRODUCTION RATE FOR
WOMEN IN POSTMENSTRUAL PERIOD
Vco
(ml/hr)
.48
.23
.25
.20
.22
.15
.21
.23
.51
.34
.41
.26
.16
.30
.40
.47
.23
.24
.55
.32
.43
.35
REFERENCE
Lynch and Moede, 1972
H
it
it
ii
H
H
Delivoria-Papadopoulos et al., 1974
ii
H
ti
ii
ti
M
Merke et al. , 1975
n
H
M
n
n
H
II
29
------- |