United States
Environmental Protection
Agency
Environmental Monitoring Systems
Laboratory
Research Triangle Park NC 27711
Research and Development
EPA-600/S4-84-031 May 1984
Project Summary
Study of Carbon Monoxide
Exposure of Residents of
Washington, DC and
Denver, Colorado
T. D. Hartwell, C. A. Clayton, R. M. Michie, R. W. Whitmore, H. S. Zelon, S. M.
Jones, and D. A. Whitenurst
This report describes a study funded
by the EPA and conducted by the
Research Triangle Institute in 1982 and
1983 to evaluate methodology for
collecting representative personal ex-
posure monitoring (PEM) CO and cor-
responding activity data in an urbanized
area. This involved telephone screening
of households and sample selection of
respondents in the metropolitan areas
in and around Denver, Colorado and
Washington, D.C. Data on CO breath
levels were also collected in Washing-
ton, D.C. (PEDCo Environmental con-
ducted the field work in Denver.) The
target population in both cities consist-
ed of the non-institutionalized, non-
smoking adults (ages 18 to 70) of these
metropolitan areas. The data collected
in the field were edited and appropriate-
ly weighted to produce CO exposure
estimates for the target population.
Estimates of CO exposure for the
winter of 1982-83 in Washington, D.C.
were obtained using the data base
constructed from the raw CO levels by
activity data. This data was collected
over a 24-hour period when the respond-
ent carried a CO PEM and an activity
diary. The data consisted of hourly CO
values on 712 respondents, activity
patterns and corresponding CO levels
on 705 respondents, and CO breath
measurements corresponding to the
PEM CO data on 659 respondents. The
size of the target population was esti-
mated to be 1.22 million individuals.
The weighted average maximum hour-
ly PEM CO level in Washington, D.C.
was 6.74 ppm. The average maximum
8-hour CO level was 2.79 ppm. The
percentage of the population with maxi-
mum hourly CO values overthe 35 ppm
CO standard was estimated to be 1.28
percent while the percentage with an
8-hour maximum over the 9 ppm stand-
ard was 3.9 percent.
Estimates were also made for sub-
groups of the population. Persons in
high-exposure occupations (about 4.6%
of the total population) generally exhib-
ited higher CO exposure levels: it was
estimated that about 24% of this high-
exposure group had 1-hour CO expo-
sures above the 35 ppm standard and
that about 28% exceeded the 8-hour
standard. It was also shown that CO
levels were generally higher for com-
muters, especially for those with larger
amounts of travel.
By combining PEM data with data
from individuals' diaries, estimates of
both CO levels and time durations for
various activities and personal environ-
ments were made. For example, the
activities "in parking garage or parking
lot" and "travel, transit" had the high-
est average CO concentrations (6.93
ppm, and 4.51 ppm, respectively) while
"sleeping" had an estimated CO con-
centration of only .85 ppm.
Variation from duplicate hourly PEM
measurements under field condition
were also analyzed. An analysis of
variance of this data which considered
person-to-person, hour-to-hour, and
measurement variation indicated that
about 5 to 6 percent of total variation
-------
among the hourly duplicate readings
was due to deviation in measurements
made by two PEMs at the same hour for
the same person.
This Project Summary was developed
by EPA's Environmental Monitoring
Systems Laboratory, Research Triangle
Park. NC. to announce key findings of
the research project that is fully docu-
mented in a separate report of the same
title (see Project Report ordering infor-
mation at back).
Introduction
As the control of emission increases,
the burden of proof on EPA to show that a
particular level of emission control is
justified also increases. It has become
more and more important to show that a
given level of control is justified for each
air pollutant, with the relative risk of
public health approximately comparable
for each pollutant controlled.
A critical factor in determining the
degree of risk to the population is the
exposure of members of the population.
In the past, monitoring of airborne pollut-
ants has necessarily been based on the
assumption that fixed-site monitoring is
representative of concentrations sur-
rounding the site, since monitoring tech-
niques were generally not developed for
determining personal exposures. Then to
obtain estimates of population exposure,
techniques such as computer simulation
or overlaying isopleths of pollution con-
centrations measured at fixed sites on
population density maps have been used.
For some pollutants, these techniques
may be reasonable approximations; how-
ever, recent work has shown that many
pollutant concentrations are not homo-
geneous and that activity patterns play an
important role in an individual's actual
exposure. Therefore, data from ambient
fixed sites often differ significantly from
the concentrations with which people
actually come into contact.
Accordingly, RTI and EPA formulated a
study plan to develop and field test a
population exposure methodology using
CO while making sure that the method-
ology was broad enough to accommodate
other pollutants of concern. The specific
objectives of this study were the follow-
ing:
— To develop a methodology for meas-
uring the distribution of carbon
monoxide (CO) exposures of a repre-
sentative population of an urban
area for assessment of the risk to
the population.
— To test, evaluate, and validate this
methodology by employing it in the
execution of pilot field studies in
Denver, Colorado, and in Washing-
ton, D.C.
— To obtain an activity-pattern data
base related to CO exposures.
Carbon monoxide was selected for
primary emphasis in the current study
because:
— Accurate and portable field-tested
instruments now are available for
CO.
— Most of the CO to which the public
is exposed can be attributed to
motor vehicles.
— It appears that CO is a good "indi-
cator" (i.e., surrogate) pollutant for
estimating exposures to several
other motor vehicle pollutants of
interest.
— Because CO is a nonreactive air
pollutant, it is simpler to treat
analytically.
— The health effects of CO are reason-
ably well documented, and NAAQS
based on these effects have been
promulgated.
— Considerable data exist showing
that CO varies spatially and that
many locations in cities have con-
centrations that differ from those
reported at fixed air monitoring
stations.
The study was carried out in Washing-
ton, D.C. and Denver, Coloradoduring the
winter of 1982-83 (the period of the year
with maximum ambient CO concentra-
tions). The population exposure profile
was determined by direct measurement
of CO with personal exposure monitors
(PEMs) through the use of statistical
inference from the statistically drawn
sample. The study provided sufficient
data to determine exposure as a function
of concentrations within significant
microenvironments (home, in-transit,
work, and leisure) and individual activity
patterns.
The report describes in detail the
activities, results, and recommendations
evolving from the study. It is extremely
important to note that the study not only
developed and tested methodology for
measuring the distribution of CO in an
urban area but also produced direct
estimates of CO exposure that apply to
two large metropolitan areas. In addition,
a very important product of this work is a
unique and valuable data base on indi-
vidual exposures to CO and the corre-
sponding activities that led to these
exposures.
Summary of Study Design and
Procedures
The target population consisted of the
non-institutionalized, non-smoking adults
(ages 18 to 70) in the metropolitan areas
in and around Denver, Colorado and
Washington, D.C. A probability sample of
the target population was selected in both
cities. This sample was a stratified, three-
stage, probability-based design. Area
sample segments defined by Census
geographic variables were selected at the
first stage of sampling. Households were
selected at the second stage, and a house-
hold member was administered a short
screening interview covering all house-
hold members to identify individuals with
characteristics believed to be positively
correlated with CO exposure. Thus, house-
hold members with these characteristics
could be oversampled in the third stage.
Donnelley Market Corporation listings
were used to help select households for
the screening interview. The third stage
sample was a stratified sample of
screened eligible individuals (i.e., non-
smoking, aged 18 to 70). The individuals
in the third stage sample were admin-
istered a Computer Model Input Ques-
tionnaire and were asked to carry a
personal CO monitor and an Activity
Diary for 24 or 48 hours (for Washington
and Denver, respectively). A breath sam-
ple was also requested from these indi-
viduals and they were asked to fill out a
Household (Study) Questionnaire. The
third stage sample design also allocated
individuals to specific days within the
sampling period. A detailed discussion of
the sample design is given in the report.
To carry out the sample design, RTI
developed the data collection instru-
ments and worked with EPA in obtaining
OMB approval for the study. An initial
telephone screening was carried out in
both Denver and Washington, D.C. by
using RTI's Computer AssistedTelephone
Interviewing (CATI) system. This initial
screening was supplemented by limited
field screening in both sites. Specific
information collected during this inter-
view inclu'ded: time spent in regular
commuting and smoking status of each
household member, as well as presence
of gas appliances and attached garages in
their residences. After the initial screen-
ing and the initial selection of potential
participants, anothertelephone interview
was conducted. The purpose of this call
was to contact the selected individual to
further explain the study and attempt to
enroll him (her) into the study. If the
individual agreed to be part of the study.
-------
an appointment was established for a
field interview. In addition, during this
call, a Computer Model Input Question-
naire was administered which collected
additional data on commuting patterns,
demographics of household members,
and household characteristics.
Finally, participating individuals were
met at their home or other convenient
location and given all study materials.
These participants carried both a PEM (a
model COED-1, which utilized a data
acquisition package supplied by Magnus,
Inc.) for the 24 hours of their participation
and an Activity Diary in which to record a
description of their activities. Participants
were requested to push a button on their
PEM every time they changed activities
and to record descriptions of the new
activities in their diaries. In addition, for a
small sample of participants, a GE/HP
PEM (which utilized a Hewlett-Packard
HP-41CV programmable calculator) was
used which allowed the participant to
also enter an activity code into the
monitor. Participants were also asked to
complete a self-administered Household
Questionnaire which provided informa-
tion on themselves and on their home and
work environments. The telephone screen-
ing and sample selection of individuals
for both Denver and Washington were
carried out by RTI as was the field work in
Washington.
The results of the telephone screening
and field activities for the study are
described in detail in the report. Briefly,
8643 household screenings were at-
tempted by RTI in Washington, D.C. and
4987 were attempted in Denver, Colo-
rado. The successful screening rates
were 75.8 percent in Washington and
70.4 percent in Denver. From these
telephone and field screenings, 5418
eligible respondents were identified in
Washington and 2232 in Denver. From
this population of eligibles, 1987 indi-
viduals were selected for participation
(i.e., to carry a PEM) in Washington and
1139 in Denver. Of these selected indi-
viduals, 58 percent actually scheduled
appointments to carry a PEM in Washing-
ton. Finally, 35.8 percent of the indi-
viduals in Washington selected to partici-
pate contributed usable CO monitor data.
This represented 712 sample respond-
ents. Instrument failure was one of the
major reasons for the low response rate.
Specifically, CO data was not collected or
was unusable for analysis purposes for
232 respondents (22% loss rate) due to
monitor failure or malfunction. Usable
CO breath data corresponding to the
usable CO monitor data was collected on
659 sample respondents.
In order to successfully implement the
study in Washington, D.C., a field office/
laboratory was established in the offices
of the Metropolitan DC Council of Govern-
ments. This office was used for several
purposes including supervision of field
staff, storage of supplies, maintenance of
records, allocation of field assignments,
and maintenanee and repair of the PEMs.
This office was visited twice nightly by all
interviewers to receive PEMs and data
collection forms for that evening and for
return of completed study materials in-
cluding the PEMs used the previous 24
hours. All calibrations of the PEMs during
the study were carried out in this field
laboratory. In addition to the field super-
visorforthe interviewers, the field labora-
tory was staffed with two full-time tech-
nicians working seven days per week
throughout the study. A detailed descrip-
tion of the PEMs (COED-1 s and GE/HPs)
used in this study and the extensive daily
technical support that they required is
given in the report.
As mentioned above, breath samples
were collected from respondents during
the study. This required RTI to evaluate a
method for collecting and measuring
alveolar CO. The method essentially
required each respondent to blow into a
sample bag at the end of his 24-hour
sampling period. This sealed bag was
then returned to the field laboratory for
CO analysis.
Throughout the field work, a quality
control and assurance program was
maintainedfor the sampling and analysis
procedures employed. This included using
field standards to calibrate all the CO
monitors. The monitors were subject to
calibration (two-point, zero/span) before
they were put in the field and 24 hours
later when they were returned from the
field. The comparison of the two calibra-
tion curves was used to assign validity
codes to the PEM data. Other quality
control procedures employed were: a ten
percent check of data transcribed from
monitor memory to field data sheets;
monitoring control charts on each monitor
describing the course of differences be-
tween pre-sample and post-sample span,
zero, battery voltage, and flow rate
values; collecting duplicate colocated
samples for the purpose of characterizing
monitor precision; performing external
and internal QA and QC audits; perform-
ing multipoint calibrations to assess
monitor linearity during the study; and
obtaining duplicate breath samples from
respondents. The results of these exten-
sive quality control and quality assurance
procedures are given in the report.
After the field work was completed, the
data were returned to RTI where detailed
editing of the data was carried out by RTI
editors. The data were then entered into
computer files using RTI's mini-computer
data base entry system. All data were
keyed and then 100 percent key-verified.
Extensive machine editing was carried
out which resulted in identifying many
computer records which required further
manual editing. The process of editing the
computer files took extensive staff time.
In particular, checking the consistency of
the PEM data with the diary data was a
time consuming process.
Sampling weights were computed ac-
cording to prescribed formulas. This
involved extensive computations so that
the weights could be used to draw
inferences to the target populations. The
sampling weights were then put on a
computer file so that they could be
merged with the corresponding field data.
Detailed statistical analyses were car-
ried out using computer data files with
PEM COandactivitydiarydata. Estimates
computed during this analysis were
weighted estimates for the population of
inference—adult non-smokers in the
Washington, D.C. metropolitan area. Stand-
ard errors of estimates were produced by
using specially written software designed
for analysis of data from complex sample
surveys.
In particular, analyses were first pro-
duced for hourly CO exposure data. These
analyses included computing statistics
describing diurnal patterns, maximum
hourly CO concentrations, maximum 8-
hour CO concentrations, and mean hourly
CO concentrations. Statistics included
means, standard errors, and percentages
of the population exceeding certain speci-
fied CO levels. Estimates of these statis-
tics were computed for all days, week and
weekend days, and low and high CO days
(as indicated by fixed-site monitors). In
addition, CO hourly level comparisons
were also made for 3 occupational
groups; 6 commuter groups (i.e., non-
commuters; commuters who traveled up
to 5 hours/week; etc.); and 4 categories
describing the use of gas stoves.
Estimates were also produced for CO
exposure levels for various activities (e.g.,
in transit) and locations (e.g., indoors—at
residence). Statistics computed for each
activity and location included mean CO
level, the estimated standard error, and
estimates of the proportion of the popula-
-------
tion having CO levels above specified
levels. The distribution of times spent in
the various activities and locations were
also computed.
Breath measurements taken at the end
of each individuals' monitoring period
were used to produce estimates of the
distribution of CO breath levels in the
Washington, D.C. area. Finally, using the
duplicate CO monitor data, estimates
were computed to assess variation in
PEM measurements under field condi-
tions.
Summary of Study Results and
Conclusions
Based on the experience gained during
the Washington, D.C. and Denver PEM
CO studies, the methodology developed,
with some modifications (see the detailed
report), may be used effectively in other
areas of the country for collecting PEM
data. Experience gained during this
initial study will improve the execution of
such similar studies. Modifications that
are suggested include a different sam-
pling design using the classified tele-
phone directory listings, improvements in
the CO monitors, and additional refine-
ment of the method used to collect
activity data. These modifications should
make the methodology more cost effec-
tive, improve the response rate, and lead
to more accurate activity information.
Important new information was learned
for each of three sampling methodology
studies of the project: (1) It was found that
geographically classified telephone direc-
tory listings can be used in a cost-effective
manner in association with standard area
household sampling techniques for per-
sonal monitoring studies like the current
CO study. The sampling design for the
cost-effective use of these telephone
directory listings differs substantially
from the design used for the CO study,
however (details are given in the report).
(2)Sending lead letterstoindividualswho
were selected for personal monitoring
prior to calling to schedule an appoint-
ment was found to be an effective strate-
gy. (3) The need for person-day sampling
for studies that monitor personal expo-
sure to airborne pollutants is apparent.
The CO study gained valuable experience
with this technique. Further study, possi-
bly even another methodological study, is
needed to refine this technique.
Based on experience derived during
this project, two important conclusions
were reached concerning the use of the
COED-1 and GE/HP monitorsfor monitor-
ing personal CO exposure:
— The COED monitors exhibited a less
than desirable reliability during this
study producing a final successful
sample completion rate of only 78
percent. Since most of the lost
samples can be attributed to unreli-
ability of the monitor electronics,
the battery packs, or the sample
pump (169 of the 232 samples lost
due to monitor malfunction), these
monitors will probably become ac-
ceptable for future projects of this
type providing that the recom-
mendations discussed in the report
are successfully incorporated into
the monitor design. Excessive cali-
bration drift accounted for the re-
maining 63 of the 232 samples lost
due to monitor malfunction (approxi-
mately 6 percent of the samples
attempted. The monitors exhibited
high linearity (calibration r2 =
0.9997), acceptable stability (86
percent within ±10 percent of
initial response levels after 24
hours), and reasonable precision
(median standard deviation of dupli-
cate measurements s= 0.25 ppm)
during field monitoring.
— The GE/HP monitors will probably
be acceptable for such monitoring
following perfection of the design
and incorporation of the recom-
mendations suggested in the report.
The full user-programmability of
these monitors will add desirable
flexibility, not achievable with the
COED-1, to future monitoring proj-
ects. On-Board micro-processor
monitoring of, and compensation
for, parameterssuchascelltemper-
ature and battery voltage may in-
crease monitor stability and preci-
sion.
Concerning the monitoring of alveolar
carbon monoxide by the method utilized
during this project, the following con-
clusions were reached:
— The proposed method performed
well, producing a mean difference
between duplicate samples of 0.11
ppm ± 0.13 ppm at the 95 percent
confidence level and an estimated
accuracy of ± 0.3 ppm at 3.5 ppm
and ±1.0 ppm at 40 ppm. The
proposed modification to the pro-
cedure concerning use of humidi-
fied zero and calibration matrices
is, however, deemed necessary for
procedural stability. The method is
highly reliable (97.5 percent suc-
cessful sample completion rate).
Using the data collected in the Washing-
ton, D.C. and Denver metropolitan areas
with the Household Screening Question-
naire, weighted estimates were computed
of population characteristics. These esti-
mates were based on screening inter-
views in 4394 households in Washington
and 2128 households in Denver. In
particular, the population estimate for the
number of households in the two areas
was 953,714 for Washington and 345,163
for Denver. Population estimates of per-
centages of households with various
characteristics were as follows:
Washing-
ton Denver
Use Fireplace
Use Wood Stove
Use Gas Furnace
Use Gas Stove
Use Gas Hot Water
Have Attached
Garage or
Multi-Family
Garage
33%
4%
56%
64%
57%
22%
30%
6%
71%
25%
78%
35%
In addition to household characteristics,
several estimates were also obtained for
individuals' characteristics in the two
areas. For example,
Washing-
ton Denver
Male 48% 47%
Smokers (13 years 33% 38%
or older)
Work (13 years 70% 72%
or older)
Travel ^ 3 84% 82%
times/week
Regarding estimates of CO exposure
for the winter of 1982-83 in Washington,
D.C., a data base was constructed from
the raw CO levels by activity data which
consisted of hourly CO values on 712
respondents, activity patterns and cor-
responding CO levels on 705 respond-
ents, and CO breath measurements cor-
responding to the PEM CO data on 659
respondents. These data were used to
obtain estimates of CO exposure for the
population of inference—the adult (18 to
70 years old), non-smokers in the urban-
ized portion of the Washington, D.C.
SMSA. The size of this population was
estimated to be 1.22 million individuals.
The results presented beloware weighted
estimates which apply to this population.
-------
The weighted average maximum hourly
PEM CO level in Washington, D.C. was
6.74 ppm (this was computed as the
weighted average of the maximum hourly
CO value for each individual in the
sample). The average maximum 8-hour
CO level was 2.79 ppm. The percentage
of the population with maximum hourly
CO values over the 35 ppm CO standard
was estimated to be 1.28 percent while
the percentage with an 8-hour maximum
over the 9 ppm standard was 3.9 percent.
Estimates were also made for sub-
groupsof the population. Personsin high-
exposure occupations (about 4.6% of the
total population)generally exhibited high-
er CO exposure levels: it was estimated
that about 24% of this high-exposure
group had 1-hour CO exposures above
the 35-ppm standard and that about 28%
exceeded the 8-hour standard. It was also
shown that CO levels were generally
higher for commuters, especially for
those with larger amounts of travel. For
example, 8% of the commuters indicating
16 or more hours of travel per week were
estimated to have maximum 8-hour CO
concentrations over 9 ppm, whereas
less than 1% of the non-commuters were
estimated to have such levels.
Breath CO levels (taken at the end of
the sampling periods, usually in the
respondents' homes) for the adult non-
smoking population in Washington aver-
aged 5.12 ppm. Slightly higher levels
were observed for persons with high
occupational exposures and for persons
with large amounts of travel.
By combining PEM data with data from
individuals' diaries, estimates of both CO
levels and time durations for various
activities and personal environments
were made. In general, these results
were consistent with a priori expecta-
tions. For example, the activities "in
parking garage or parking lot" and "travel,
transit" had the highest average CO
concentrations (6.93 ppm and 4.51 ppm,
respectively) while "sleeping" had an
estimated CO concentration of only .85
ppm. Among the environments consid-
ered, the three with the highest average
CO concentrations were "indoor parking
garage," "outdoor parking area," and "in
transit-car." The average levels for these
environments were 10.36,4.67, and 5.05
ppm, respectively.
Variation from duplicate hourly PEM
measurements under field conditions
were also analyzed. An analysis of vari-
ance of this data which considered per-
son-to-person, hour-to-hour, and mea-
surement variation indicated that about 5
to 6 percent of total variation among the
hourly duplicate readings was due to
deviations in measurements made by two
PEMs at the same hour for the same
person.
T. D. Hartwell. C. A. Clayton, R. M. Mitchie. R. W. Whitmore, H. S. Zelon, S. M.
Jones, and D. A. Whitehurst are with Research Triangle Institute, Research
Triangle Park, NC 27709.
Gerald G. Akland is the EPA Project Officer (see below).
The complete report, entitled "Study of Carbon Monoxide Exposure of Residents
of Washington, DC and Denver, Colorado," (Order No. PB 84-183 516; Cost:
$20.50, subject to change) will be available only from:
National Technical Information Service
5285 Port Royal Road
Springfield, VA 22161
Telephone: 703-487-4650
The EPA Project Officer can be contacted at:
Environmental Monitoring Systems Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
------- |