United States
Environmental Protection
Agency
Environmental Monitoring Systems
Laboratory
Research Triangle Park NC 27711
Research and Development
EPA/600/S4-85/046 Aug. 1985
Project Summary
Application of the
Microenvironment Monitoring
Approach to Assess Human
Exposure to Carbon Monoxide
Naihua Duan, Harold Sauls, and David Holland
Exposure estimates based on moni-
toring carbon monoxide in microenvi-
ronments are compared to exposure
estimates based on personal monitor-
ing with individual, portable monitors.
Methods of calculation are reviewed
and discussed, and results of calcula-
tions are presented. These data indicate
that population exposure estimates
based on data from the Washington
Microenvironment Study, combined
with people's activity data from the
Washington Urban Scale Study, are
about 40 percent higher than estimates
based on personal monitoring data from
the Urban Scale Study. The former set
of exposure estimates is found to be a
good predictor of the latter. Neverthe-
less, generalizations of these findings to
other data bases are not valid at this
time.
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
Due to high costs, equipment require-
ments, and people-related difficulties
associated with personal exposure moni-
toring, it is highly desirable to develop
methodology with which to estimate
population exposure to air pollution with-
out directly monitoring individuals sam-
pled from the population. Knowledge of
pollutant concentrations in microenviron-
ment types (METs) plus information
about the activities and mobility of a popu-
lation under study can be used to obtain
all the elements, presumably, needed to
produce a valid estimate of overall popu-
lation exposure. Variability of pollutants,
concentrations, and time within METs
are the principal limiting factors of reli-
ability, given that reported activities
match up well with the defined METs.
This study applies the microenviron-
ment monitoring (MEM) approach, called
the indirect approach, to estimate human
exposure to carbon monoxide (CO), using
activity time data from the Washington
Urban Scale Study and CO concentration
data from the CO Microenvironment
Study. The estimated exposures based on
the MEM approach are then compared
with estimated exposures based on the
personal monitoring (PM) approach,
called the direct approach.
For the specific data used in this study,
the MEM exposures are about 40 percent
higher than the PM exposures. However,
despite this discrepancy, the MEM expo-
sure is found to be a powerful predictor
forthe PM exposure. On the log scale, the
MEM exposure has the correct span rela-
tive to the PM exposure; the relationship
between the two sets of exposure esti-
mates is found to be a constant drift.
Several factors offer some explanation
of the observed difference between the
MEM and the PM exposures. The two
data collecting activities were not de-
signed primarily for comparative analy-
sis. Therefore, the microenvironments
are imperfect matches with the reported
-------
activities. The commuting routes of the
CO M icroenvironment Study were select-
ed as "heavily traveled" and sampled
only during the rush hour periods. The
PM study sampled travel in private cars at
any time it occurred. Also, Wallace, Thom-
as, and Mage noted that COHb levels
estimated from breath measurements
were higher than those estimated from
PM observations. It is believed that read-
ings decline as the monitor battery dis-
charges. Monitors were used for much
shorter periods with more frequent cali-
brationsduringtheCOMicroenvironment
Study than in the PM Study.
Exposure Assessment
Until recently, human exposure to air
pollution could be assessed only with
fixed-site ambient monitoring data. Typi-
cally, people residing in the same neigh-
borhood near a monitoring station were
treated as homogeneous receptors fixed
at the location of the monitoring station.
Recent field studies with personal expo-
sure monitors (PEMs) have found this
approach inadequate for such pollutants
as carbon monoxide, which are spatially
variable or have nonambient sources or
sinks. During the Washington M icroen-
vironment Study commuters were ex-
posed to 9-12 ppm CO averaged over the
entire commute route, while at the same
time of day fixed-site monitors in DC
logged an average of about 3 ppm CO. A
study by Nagda and Koontz observed CO
concentrations generally between the
MEM and PM values reported here for
comparable microenvironments. Obvious-
ly it is important to consider population
activities and mobility when assessing
exposure.
Incorporation of population mobility and
activities into the CO exposure assess-
ment process became a more practical
reality with the development of reliable,
continuous CO personal exposure moni-
tors (PEMs). There are two general
approaches to exposure assessment us-
ing PEMs. The first is the personal moni-
toring (PM) approach in which human
subjects are sampled from the target
population and are equipped with PEMs
for a certain time to measure their expo-
sures directly. This approach was taken in
the Washington Urban Scale Study. Ad-
vantages are simplicity of design and
freedom from modeling assumptions. The
main disadvantage is cost, too high for
large-scale investigations.
An alternative approach to assess expo-
sure is the microenvironment type (MET)
approach in which pollutant concentra-
tion data are combined with or enhanced
by activity time data. The MET approach
can be implemented either by the en-
hanced personal monitoring (EPM) meth-
od or by the microenvironment monitor-
ing method. The latter approach was
taken in the CO Microenvironment Study
in Washington, DC, during the Winter of
1983.
The MET method combines MET-speci-
f ic pollutant concentration data and activ-
ity time data to estimate exposures. This
approach incorporates information about
the mobility of the population under study.
A relatively inexpensive way to imple-
ment the MET approach is through micro-
environment (ME) monitoring. Instead of
MET concentration data from personal
monitoring, a number of MEs may be
sampled in each MET, with research staff
or trained technicians sent to the sampled
microenvironments to monitor those micro-
environments directly.
Methods for Estimating Expo-
sure
The MET concentration data and the
MET time data can be combined in sev-
eral ways to estimate exposure. If one is
interested only in average exposure, one
can use the average time-weighted sum-
mation formula and estimate average
exposure by
person-days), and the exposure for each
convoluted unit is estimated using a time-
weighted summation formula similar to
Equation (1).
E = I
k
Ck x T,
(D
where E is the average exposure, Ck is
the average MET concentration for the kth
MET, and Tk is the average MET time for
the kth MET. This method implicitly as-
sumes that the MET concentrations and
MET times are uncorrelated. The assump-
tion basically rules out responses to air
pollution episodes which might cause
people to stay away from high concentra-
tion METs during such days.
For most purposes the mere estimation
of average exposure is inadequate, and it
is necessary to estimate exposure distri-
bution or individual exposures. One way
of doing this is to use a simulation model
in which the concentration and activity
data are summarized by probabilistic dis-
tributions, human activity and concentra-
tion data are simulated from those prob-
abilistic distributions, and the simulated
data are used to estimate exposures. This
type of approach generally assumes that
the concentration and time are inde-
pendent. Another approach is the convo-
lution method. Units (e.g., persons) from
the activity data base are paired with
units (e.g., days) from the concentration
data base to form convoluted units (e.g..
Eim -
(2)
where E,m is the exposure combining the
ith unit in the activity data base and the mth
unit in the concentration data base, Cmk is
the MET concentration for the mth unit in
the concentration data base in the kth
MET, and T,k is the MET time for the ith
unit in the activity data base in the kth
MET.
To illustrate the application of Equation
2, consider a study that has 43 days of
MEM data, combined with a sample of
705 persons, each providing one day of
activity diary. If the ith person in the activ-
ity sample spent the day according to T,
and was exposed to concentrations Cm in
the METs encountered during that day,
he would receive exposure E,m. As inde-
pendence is assumed between the MET
concentrations and times, each of the 43
concentration vectors Cm is equally likely
for each of the 705 participants. With the
convolution method, the exposures Eim
are derived for each of the 43 x 705 =
30,315 pairings of persons and days in
the two data bases. Each such pairing
forms one convoluted person-day.
Another method can be viewed as a
hybridization between the average time-
weighted summation formula Eq. (1) and
the convolution method Eq. (2). With this
hybrid method, the average MET concen-
tration in each MET is used to estimate
the exposure for each unit (day or person-
day) from the activity data base by
E, = ICk x Tlk
k
(3)
This method ignores the variability in
exposures between microenvironments
of the same MET. If all microenvironments
belonging to the same MET have the
same concentration, this method is pref-
erable to the convolution method because
of its simplicity. If the microenvironments
belonging to the same MET vary substan-
tially, this approach is likely to underes-
timate the variability of the exposure
distribution.
Activity Time Data
A population-based study of CO expo-
sure was conducted during the winter of
1982-83 in the Washington, DC metro-
politan area. An area probability sample
of human subjects was enrolled for one
day for each in this study. The partici-
-------
pants filled out activity diaries giving the
activities they were engaged in during
each time period. The activities were
entered in the diaries as activity seg-
ments, where each activity segment was
defined to be the time period between two
reported changes in activities in the activ-
ity diary. The participants' exposures to
CO were measured using PEMs, which
recorded the average concentration over
each activity segment.
The participants in the Washington
Urban Scale Study were selected from a
probability sample. To extrapolate from
the sample to the target population, it is
necessary to weight the individual obser-
vations by the sampling weights based on
sampling probabilities. In preliminary
analysis, the summary statistics based on
the weighted and the unweighted proce-
dures were compared. The weighting did
not have a major effect on the results. For
example, the average time spent in car
commuting differs by about 2 percent
between the weighted and the unweight-
ed estimates. Because the primary goal of
the comparative study is to compare the
estimated exposures based on the MEM
and PM approaches for the observed
sample, the extrapolation to the target
population is not crucial. Therefore, to
simplify the analysis, it was decided not to
weight the individual observations.
In the Washington Urban Scale Study
each participant filled out activity diaries
for one day. During this sampling day,
whenever there was a new activity—e.g.,
the participant stopped reading a news-
paper in the living room (end of an old
activity) and went outside for a walk
(beginning of a new activity)—the partic-
ipant was required to record the start time
of the new activity and describe it. The
period between two entries in the activity
diary is referred to as an activity segment.
Each activity segment is regarded as one
microenvironment.
Based on information available, activity
segments are grouped into seven METs:
parking, public transportation, private car,
pedestrian, shops, offices, and other. The
rest of this section gives the heuristic
definitions of these METs.
The MET parking is restricted to indoor
parking because only indoor parking con-
centration data are available from the CO
Microenvironment Study. The MET public
transportation includes both bus and
metrorail. Because both buses and met-
rorails are monitored in the Microenvi-
ronment Study, it is possible to consider
them as distinct METs. However, in the
evaluation of MET classification schemes,
it was found unproductive to distinguish
between these two METs; therefore,
public transportation is considered as one
MET without further refinement.
The MET private car includes private
cars, trucks, motorcycles, and vans. It is
debatable whether this MET should be
restricted to the narrow definition includ-
ing private cars only. (Only private cars
were monitored in the Microenvironment
Study). The four modes of travel were
grouped into one MET for two reasons. (1)
The amount of time spent in trucks,
motorcycles, and vans is very small com-
pared with the amount of time spent in
private cars. The total amount of time
spent in the four modes of travel is 1.623
hours per person per day, out of which
only 0.106 hours belong to the three
modes other than private car, less than 7
percent of the total. (2) The MET concen-
trations based on PEM for those four
modes of travel are roughly similar. The
difference between car and truck is small
(about 1 ppm) and statistically insignifi-
cant. The difference between car and van
is not small (about 3 ppm) and is statisti-
cally significant, but only seven people
reported using a van in their travel.
The MET pedestrian includes walking,
biking, and jogging. It is again debatable
whether jogging and biking should be
grouped with walking into one MET. The
amount of time spent jogging and biking
is very small (less than 6 percent) com-
pared with time spent walking. The dif-
ference in concentrations between walk-
ing and jogging is very small (less than
0.1 ppm) and statistically insignificant (t =
0.09). The difference between walking
and biking is about 2 ppm and is statis-
tically significant(t = 2.09). However, only
five people reported biking during the
sampling period. Therefore, they are
combined into one MET.
The MET shops consists of the activity
segments reported as stores, shopping
malls, and theaters in malls. The amount
of time spent in the malls is small relative
to the time spent in stores (less than 5
percent). The difference in concentration
is very small (less than 0.5 ppm) and sta-
tistically insignificant (t = 0.65). There-
fore, they are combined into one MET.
The MET offices consists of activity
segments reported as offices. The MET
other is a residual category for activity
segments not considered above. The main
component of activity segments in this
MET is home. Because there are no
microenvironment monitoring data cor-
responding to these activity segments in
the Microenvironment Study, this MET
cannot be refined any further.
CO Concentration Data
The Washington CO Microenvironment
Study was conducted in the metropolitan
area during the winter of 1983. Primarily
the study focused on the measurement of
commuting microenvironments including
parking garages, driving an automobile,
riding a bus, riding a train, and walking.
For automobile commutes, the study
identified eight routes that "collectively
extend 160 miles, about 8.6% of the total
length (1,853 miles) of Washington's
arterials and freeways." (In 1980, the
Washington metropolitan area had 9,432
miles of streets and roads, including arte-
rials, freeways, and locals). The routes
selected were ones considered to be
heavily traveled and predicted to have
high CO exposures during rush hour
periods.
Although the routes were chosen to be
representative of the arterials and free-
ways, they might not be representative of
all routes traveled by the general popula-
tion. The empirical analysis found that for
the commuting METs, the MET concen-
trations from the CO Microenvironment
Study are substantially higher than cor-
responding MET concentrations based on
personal monitoring from the Urban Scale
Study.
A Commuter Study Links Data Base
was constructed from the commuting
part of the Microenvironment Study. Each
commuting route was divided into links
ranging from one-half to three miles,
each link being a physically distinct seg-
ment of the route and regarded as an
individual microenvironment. For quality
assurance, several commuting trips used
collocated monitors or insideX-outside
pairs. In the paired situation, this study
restricts attention to the primary monitor.
The ME study included monitoring on
some indoor microenvironments—shop-
ping centers and offices. Additional moni-
toring was conducted on walking micro-
environments. The pedestrian data are
combined with those from the commut-
ing part of the study and analyzed as
belonging to the same MET.
One major exclusion in ME coverage
was the home microenvironment. A
residual MET, referred to as the MET
other consists of all microenvironments
not covered in the Microenvironment
Study. For the exposure estimation, the
microenvironment monitoring data are
supplemented with personal monitoring
data from the Urban Scale Study for those
microenvironments not covered in the
Microenvironment Study.
-------
Observed MET Concentrations
Concentrations Based on MEM
For each MET except the MET other,
the measurements from the Microenvi-
ronment Study are aggregated into daily
averages, which are used as the MET
concentrations in further analysis. A total
of 43 days were measured during the
period from January 1 through March 18,
1983.
As expected, the concentrations in
parking garages are very high. The aver-
age concentration exceeds the one-hour
federal standard level of 35 ppm. The
concentration in private cars is also fairly
high. The average concentration exceeds
the eight-hour federal standard level of 9
ppm. Public transportation, walking, and
shops have moderate levels averaging
about 5 ppm. Offices have low levels,
averaging about 2 ppm.
Concentrations Based on PM
An alternative set of estimates of MET
concentrations can be derived from the
personal monitoring data in the Urban
Scale Study. For each activity segment
reported, the exposure for that activity
segment is computed as the product of
the duration of the activity segment and
its average CO concentration. For each
participant and for each MET, the expo-
sures from the activity segments belong-
ing to that MET are summed as the total
exposure for that MET. The total exposure
in the MET is divided by the total amount
of time(hours) in the MET to getthe aver-
age MET concentration.
For certain activity segments, the CO
concentrations are not available, possibly
because of monitor failure. Those activity
segments are not included in the calcula-
tion of the MET concentrations. To assess
the effect of those missing data, the
amount of time belonging to such activity
segments is calculated for each partici-
pant and for each MET. For three METs—
namely, shops, parking, and public trans-
portation—none of the participants had
any activity segments with missing CO
concentration data. For the other three
METs, some of the activity segments did
not have CO concentrations. However,
the amount of time for those activity seg-
ments is very small. For the MET/wvVafe
car, the average amount of time per par-
ticipant for which CO concentration is
missing is 0.004 hours. This is less than
one-half of 1 percent of the average time
of 1.623 hours spent in this MET. For the
MET office, the average amount of time
without CO concentration is 0.001 hours.
again very small compared with the aver-
age time of 0.269 hours in this MET.
Missing concentration data is, therefore,
of very little effect.
Comparison of MET
Concentrations
The MET concentrations based on PM
are substantially lower than the corres-
ponding MET concentrations based on
MEM, especially in the commuting METs.
The most dramatic difference of all is the
MET parking, in which there is a fourfold
difference between PM and MEM. The
average MET concentration for private
cars based on MEM is more than twice
the corresponding average concentration
based on personal monitoring. It is sus-
pected that incongruencies inherent in
the matchups of activities to MEs, ME
rush hour sampling, and monitor battery
rundown contributed considerably to
these differences.
Comparison of Exposure
Distribution Estimates
The comparison between the two sets
of summary statistics for the estimated
exposures indicates that the two distribu-
tions are substantially different. The
average MEM exposure is about 40 per-
cent higher than the average PM expo-
sure. The difference is highly significant (t
= 6.69 for the convolution method, t =
8.01 for the hybrid method). The compari-
son between the summary statistics for
the log estimated exposures also indi-
cates major differences between the
MEM and PM exposures. The average log
MEM exposure is significantly higher
than the average log PM exposure.
For certain situations such as qualify-
ing the health effects of air pollution, it is
only necessary that the estimated expo-
sure be a good predictor of actual expo-
sure. In such instances the appropriate
way to assess the validity of the estimated
exposure is to examine the regression
relationship between the actual and
estimated exposures. The slope coeffi-
cient in the regression relationship must
be significant, indicating that the esti-
mated exposure predicts the ranking of
actual exposures, even though the mag-
nitude might be off. Furthermore, the
slope coefficient should be close to one,
and the intercept coefficient close to zero,
implying that the estimated exposures
are approximately equal to the actual
exposures.
As usual the actual exposures are
unknown, therefore one cannot adequate-
ly define the relationship between the
estimated exposures and the unobserved
actual exposures. The PM exposure is
used as the benchmark and the regres-
sion relationship between the two esti-
mated exposures is tested, regressing the
PM exposure on the MEM exposure.
On the original scale, the regression
results show a very significant relation-
ship between the PM and the MEM expo-
sures. The convolution method gives a
more significant slope coefficient than
the hybrid method. This indicates that
even though the MET concentrations from
MEM and PM are substantially different,
the MEM exposures are still useful for
predicting the ranking of the PM expo-
sures. In other words, given that a certain
individual's MEM exposure is high, it is
reasonable to expect that his PM expo-
sure is also high.
The R2 statistic for the convolution
method is about 40 percent, indicating
that the MEM exposure is not only a sig-
nificant predictor for the PM exposure but
is also an informative predictor, explain-
ing an important fraction of the variability
in the PM exposure. The hybrid method
has a much smaller R2. With the convolu-
tion method, the slope coefficient in this
regression is about 0.5, substantially
smaller than one, and the intercept coef-
ficient is about 0.5 ppm, significantly
larger than zero. For simplicity the esti-
mated regression model may be approxi-
mated as follows:
PM exposure — 0.5 + 0.5
x MEM exposure.
At low levels (less than 1 ppm), the MEM
exposure underestimates the PM expo-
sures. For example, for an individual with
MEM exposure equal to zero, the regres-
sion model predicts that his actual expo-
sure is probably about 0.5 ppm. At higher
levels (more than 1 ppm), the MEM expo-
sure overestimates the PM exposure. For
example, for an individual with MEM
exposure equal to 10 ppm, the regression
model predicts that his PM exposure is
probably about 5.5 ppm, substantially
lower than the MEM exposure. Because
the average MEM exposure is about 2
ppm, for most people the MEM exposure
overestimates the PM exposure accord-
ing to the regression model.
On the log scale, too, the regression
results show a significant relationship
between the MEM exposure and the PM
exposure, indicating that the MEM expo-
sures successfully predict the ranking of
the PM exposures. The R2 statistic for the
convolution method is about 60 percent,
indicating that the log MEM exposure is
4
-------
fairly powerful in explaining an important
fraction of the variability of the log PM
exposure.
Conclusions
Methods for estimating population CO
exposures using microenvironment mon-
itoring (MEM) data, personal monitoring
(PM) data, and activity data have been
presented and results compared.
The MEM/activity data exposures aver-
aged about 40 percent higher than the
exposures estimated by the PM method.
The observed difference in the estimated
distributions is probably specific to this
data base and should not be generalized.
Given the imperfect matches of micro-
environments and problems associated
with personal monitoring, it is impressive
that the MEM exposure is such a success-
ful predictor of PM exposure, especially
on the log scale on which the MEM expo-
sure derived by the convolution method
has the correct span relative to the PM
exposure and the drift is constant over the
range.
The convolution method is preferable
to the hybrid method for this data set due
to the high variability within the MET
concentrations.
Naihua Duan is with Rand Corporation, Santa Monica, CA 90406; the EPA
authors Harold Sauls (also the EPA Project Officer, see below) and David
Holland are with the Environmental Monitoring Systems Laboratory, Research
Triangle Park. NC 27711.
The complete report, entitled "Application of the Microenvironment Monitoring
Approach to Assess Human Exposure to Carbon Monoxide," (Order No. PB
85-228 955/AS; Cost: $11.50, 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
U. S. GOVERNMENT PRINTING Of f ICE: 1985/559-111/20660
-------
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
^-r^x US.OFFiC.ALf.'JA;
• ''C:-->E»ALT-! i'S JJOViliP* '
- • " "i'^t. i OH
/ p«IVATf !
..;£ r.;o"' " r> n '? • •
' -; 5 C - :<
Official Business
Penalty for Private Use $300
EPA/600/S4-85/046
IL
------- |