Childhood asthma acute primary care visits, traffic, and traffic-related pollutants

Abstract

Previous studies have found associations between traffic-related air pollution and asthma exacerbation in children, where exacerbations were measured according to emergency department visits and hospital admissions. Fewer studies have been undertaken that look at asthma exacerbations in a less severe primary care setting. Therefore, the authors sought to examine the associations between childhood asthma exacerbations, measured as acute visits to a primary care setting, and vehicular-traffic measures in a population of children aged 18 and under in the metropolitan Atlanta area. Statistical tests for differences of mean monthly visits for members with traffic measures above the median compared with below the median and for the upper quartile compared with the lower quartile were conducted. We also compared the odds of having one or more visits in a month for those who lived closer to a major roadway were compared with those who lived farther (greater than 300 m) from a major roadway. Poisson general linear modeling was used to determine associations between daily levels of acute visits for childhood asthma and traffic-related pollutants (zinc, EC [elemental carbon], and PM₁₀ and PM_2.5 [particulate matter with an aerodynamic diameter of ≤10 and ≤2.5 μm, respectively]) for different levels of traffic and distance measures. This analysis found that both larger traffic volumes and smaller distances to the nearest major roadway were positively and significantly associated with larger numbers of childhood asthma visits, when compared with less traffic and larger distances. Our findings point to motor vehicle traffic as an important contributor to childhood asthma exacerbations.

Implications: 

Previous studies have found associations between traffic-related air pollution and asthma exacerbation in children. However, these studies were mainly conducted in emergency department or hospital admission settings; little is known regarding less acute health effects. This analysis of the association between vehicular traffic measures and childhood asthma in a primary care setting suggests that motor vehicle traffic is a contributor to less acute asthma episodes in children. The present analysis of traffic-related air pollutants and childhood asthma were less conclusive, likely due to methods limitations outlined in the paper. The implication is that further evidence of adverse respiratory health effects in children due to motor vehicle traffic can be found in a primary care setting and similar studies should be considered.

Introduction

Previous studies have found associations between traffic-related air pollution and asthma exacerbation in children (Health Effects Institute HEI Panel on the Health Effects of Traffic-Related Air Pollution [HEI Panel], 2010), where exacerbations were measured according to emergency department visits and hospital admissions. Fewer studies have been undertaken that look at asthma exacerbations in a less severe primary care setting. Therefore, we sought to examine the associations between childhood asthma exacerbations, measured as acute visits to a primary care setting, and vehicular traffic measures in a population of children aged 18 and under in the metropolitan Atlanta area. Although Atlanta’s traffic congestion ranks among the worst congestion in U.S. cities (U.S. Department of Transportation, 2005), we are aware of only one epidemiological study that specifically examined childhood asthma and traffic measures in Atlanta (Friedman et al., 2001). In that study, the impact of reduced traffic in downtown Atlanta due to traffic restrictions in place during the 1996 Olympic Games was associated with a 28% drop in daily peak ozone levels and a concurrent 20% drop in asthma hospitalizations.

The first objective of our study was to explore the relationship between childhood asthma exacerbations and two vehicular traffic measures—traffic density within a defined perimeter of household residence and distance from household to nearest major roadway. The measure of childhood asthma exacerbations was the number of monthly visits to a health plan’s non-emergency-room medical centers for acute, nonroutine asthma episodes. The second objective was to examine whether associations between acute childhood asthma exacerbations and traffic-related pollutants—zinc, elemental carbon (EC), and PM₁₀ and PM_2.5 (particulate matter with an aerodynamic diameter of ≤10 and ≤2.5 μm, respectively)—varied by the different levels of vehicular traffic measures analyzed for the first objective. Findings from our previous work as well as other studies that indicated associations between childhood asthma and particulate matter zinc (Sinclair and Tolsma, 2004; Hirshon et al., 2008; Clairborne et al., 2002) and EC (Sinclair and Tolsma, 2004; Delfino et al., 2003) prompted this secondary research objective. Zinc and EC are each produced in part by vehicular traffic. A study found that the quantity of Zn released by tire wear was similar in magnitude to that released by waste incineration, and that urban tire-wear release of Zn is 10 times greater than in rural settings (Councell et al., 2004). PM_2.5 EC is a component of diesel exhaust and is often used as a marker for this emissions source; in addition, EC has been the subject of recent concern in occupational settings such as mining (National Institute for Occupational Safety and Health [NIOSH], 2010).

Methods

The study population consisted of members of a health plan with home addresses in 20 counties in the metropolitan Atlanta, Georgia area who had asthma and who were under 19 years of age between January 1, 1998, and December 31, 2002. If a member turned age 19 during the study period, he/she was no longer included in the asthma cohort or in the visit counts from that point forward. Members were identified as having asthma if they were listed in the health plan’s asthma registry. There were 11,049 members in the study comprising 13,690 unique addresses. Member home addresses are updated in the health plan’s electronic administrative membership database on a monthly basis. The addresses were geocoded to latitude and longitude coordinates using Mapmarker software (Pitney Bowes Business Insight, Troy, NY). Over 90% of members mainly see a primary care physician based at one of the health plan’s medical care facilities, which are located around the metro Atlanta area. The health plan membership is racially diverse (54% of members being white, 35% black, 11% other) and reflective of the racial distribution of the Atlanta area.

For the study cohort, visits with a diagnosis of asthma to the health plan’s health care facilities were collected from electronic administrative databases. We wanted to capture only acute asthma episodes in the outcome measure; therefore, only acute, nonroutine visits were included in the study. An appointment type code that indicated whether visits were routine or acute identified these visits. The acute visit codes included urgent, walk-in, or after-hours care.

Traffic and distance measures

Two possible sources of spatial data sets of roads with associated traffic data were available: modeled week day traffic for calendar year 2000 from Atlanta Regional Council (ARC), and 2001 and 2002 Average Annual Daily Traffic (AADT) estimated from counts from Georgia Department of Transportation (GA DOT). Although the ARC data were developed to support technical and policy decisions required for compliance with federal legislation, including the 1990 Clean Air Act Amendments, they may not be suitable for household level exposure assessments because of their regional scale. The GA DOT data are the most complete and spatially accurate electronic statewide road data and align well with aerial photographs. In some locations, the ARC roads are 100–200 m displaced from the GA DOT roads. In order to evaluate which data would be best for a health study, we conducted a pilot study using a random selection of 1000 subjects’ addresses. The pilot determined that the 2001 data resulted in more members’ home addresses being within 300 m of a road with traffic data. In addition to their greater spatial accuracy, the 2001 GA DOT roads data had less missing data for medium-sized roads away from the city center. The pilot study also evaluated the difference between including traffic within 150 and 300 m. In previous work using case-control data from a different metropolitan area, we evaluated traffic at 100, 200, and 300 m and found that selecting distances within this range did not make much difference in overall results (Tonne et al., 2007). We concluded that using a 300-m distance with the 2001 data resulted in fewer missing estimates and sufficient variability in the AADT times road length measure for use in this study. Using the 2001 data, we found that study subjects lived in urban, suburban, and rural neighborhoods with varying road density as well as near roads with different traffic volumes.

The GA DOT traffic data were used to calculate traffic values for the full study data set. Average annual daily traffic values (AADT) for 2001 and 2002 and 2007 were obtained from GA DOT. The AADT values are based on traffic monitor counts for major roads and estimates for nonmajor roads. For roads missing 2001 or 2002 values with available data in other years, missing values were calculated by multiplying available data by correction factors for each county and for three functional classes to account for change in traffic over time. In 1991, the federal Intermodal Surface Transportation Efficiency Act (ISTEA) required states to classify streets according to functional classes, which are based on the character of traffic service the streets are intended to provide. We used available summary data by functional class to impute missing values and estimate changes in traffic over time. AADT values for 1998–2000 were not available electronically and were calculated in a similar fashion from 2002 AADT values using yearly adjustment factors by county and six functional classes based on Georgia’s Roadway Mileage and Characteristics Reports (400 Series). These reports provide vehicle miles traveled by county and function class for individual years. Within a county, roads with the same function class are assumed to have the same trends over time. Function classes distinguish between rural and urban. To obtain monthly traffic measures, each of the five annual estimates, 1998–2002, was multiplied by the monthly adjustment factors used by GA DOT with 2007 data (http://www.dot.state.ga.us/statistics/Documetns/MonthlyFactors.pdf; accessed December 4, 2008). Monthly factors are statewide factors, but include factor groups that distinguish between urban and rural, separate factor groups for Atlanta, and factors for a few individual highways (I-285, I-75, and I-85).

Using ArcGIS 9.2 (ESRI, Redland, CA) a 300-m circle around each member’s home address was intersected with roadway traffic count data. For roads segments within 300 m, the sum of the length of roads in meters was multiplied by the estimated average daily traffic to create a composite measure of traffic and roadway length. Finally, for each home address, the distance to nearest major road, defined as a road with greater than 10,000 vehicles per day in 2002, and the traffic counts on the nearest major road were derived from the 2002 AADT data.

Air pollution data

PM coarse Zn, PM_2.5 Zn, EC, PM₁₀, and PM_2.5 Federal Reference Method (FRM) monitor-based measures were collected as part of the Aerosol Research and Inhalation Epidemiology Study (ARIES) (Van Loy et al., 2000; Hansen et al., 2006). ARIES pollutants were measured daily at the Southeastern Aerosol Research and Characterization (SEARCH) Jefferson Street site, which is approximately centrally located in the Atlanta metropolitan area. Twenty-four-hour averages of the pollutants were calculated for this analysis. The ARIES aggregation rules require 21 valid hours of data to calculate a 24-hr average. ARIES spatial distribution analyses conducted by researchers at the Georgia Institute of Technology indicate that the Jefferson Street pollutant levels represent a reasonable average ambient air level. In addition to the pollutant data, ARIES also collected data on daily mean temperature and dew point temperature from the National Climatic Data Center at Hartsfield-Atlanta International Airport.

Analysis methods

Objective 1: Effects of spatial variation of exposure to traffic emissions

Statistical tests for differences of mean monthly visits for members with traffic measures above the median compared with below the median and for the upper quartile compared with the lower quartile were conducted. SAS’s Logistic procedure (SAS Institute, Cary, NC) was used to compare the odds of having one or more visits in a month for those who lived closer (less than 100, 100–200, or 200–300 m) to a major roadway with those who lived farther (greater than 300 m) from a major roadway. The odds of one or more visits compared with zero visits in a month was selected for comparison because most members have either zero or one visit in a given month.

Objective 2: Effects of traffic-related air pollution by variation of exposure to traffic

Poisson general linear modeling was used to determine whether associations between daily levels of acute visits for childhood asthma and traffic-related air pollutants varied by the different levels of traffic and distance measures analyzed in Objective 1. The models included control variables for temporal trends and meteorological variables: indicators for season, day-of-week, federal holidays, and study month (1 through 53); cubic splines of time (study day) with knots at the 21st day of each month; and cubic splines of average temperature and dew point. The models also included a control variable for possible visit trends that were not measurable in this study, defined as the count of all acute, nonaccident visits that did not have a diagnosis code of asthma, respiratory disease, cardiovascular disease, or a missing code. We excluded all respiratory and cardiovascular diseases in this definition so that the control variable would not be capturing air-pollution-related trends. Finally, a Poisson offset term of the log-transformed value of the total monthly membership of the specific subgroup based on the traffic or distance measure was included in the model, allowing estimation of population effects (risk ratios). The lag time between pollutant measurements and acute visits was the average of the 3-day period of 3, 4, and 5 days prior to the visit day. This lag structure was selected based on our previous finding that lag (3–5) was more significant for associations between childhood asthma and acute outpatient visits (Sinclair and Tolsma, 2004). Pollutant measurements were standardized by their standard deviations to simplify interpretation and comparison of results for the five pollution measurements (PM coarse Zn, PM_2.5 Zn, EC, PM₁₀, and PM_2.5).

Results

Descriptive statistics

Table 1 shows descriptive statistics for the study cohort. The overall average number of visits per member per month was 0.0551 (SD = 0.0259) (Table 1).

Table 1. Descriptive statistics for acute visit data, 1998–2002

Variable	Mean	SD	Min.	Max.
Age at visit (years)	7.05	4.69	0	18
Age (all cohort)	6.16	5.02	0	18
Monthly visit count	439.83	159.75	131	808
Monthly visits per member	0.06	0.03	0.02	0.12

Objective 1: Effects of spatial variation of exposure to traffic emissions

Looking at visits by traffic measure, members with a monthly traffic measure in the upper quartile of all monthly traffic measures had a statistically significant greater number of mean monthly visits compared with those in the lower quartile (Table 2). Similarly, members with a monthly traffic measure above the median of all monthly traffic measures had a greater number of mean monthly visits compared with those with values below the median.

Table 2. Mean monthly acute asthma visits per member by traffic measure category

Traffic Measure	Mean Visits (95% CI)	B Minus A	t (P value)
A. Lower quartile	0.053 (0.0514, 0.0546)	0.007	5.71 (<0.001)
B. Upper quartile	0.060 (0.0581, 0.0615)
A. Below median	0.055 (0.0539, 0.0562)	0.0019	2.35 (0.019)
B. Above median	0.057 (0.0558, 0.0581)

For the logistic regression models, we found that when compared with those who lived farther from a major roadway (greater than 300 m), there was a statistically significant greater odds of having one or more visits in a month for those with a home address within 100 m of a major road (odds ratio [OR] = 1.080), between 100 and 200 m (OR = 1.069), or between 200 and 300 m (OR = 1.066) (Table 3).

Table 3. Odds ratios for having one or more acute asthma visits in a month by resident distance to nearest major road

Group 1	Group 2	Group 1 vs. Group 2 Odds Ratio (95% CI)
≤100 m	>300 m	1.080 (1.030–1.133)
>100 to ≤200 m	>300 m	1.069 (1.018–1.122)
>200 to ≤300 m	>300 m	1.066 (1.017–1.117)

Objective 2: Effects of traffic-related air pollution by variation of traffic exposure

Our findings from general linear model (GLM) modeling of childhood asthma and pollutant measures are more problematic and difficult to interpret. Results of modeling EC in the subgroup of children with asthma exposed to traffic levels below the median were not significant, whereas results for those above the median were both positive and significant, with a relative risk of 1.047 (Table 4). That is, the risk of an acute asthma visit in this group increased 4.7% for each increase of 1 standard deviation in the pollutant. Results for zinc, however, were less clear. Zinc has been linked to traffic through tire wear. PM_2.5 Zn and PM₁₀, like EC, were not significant for traffic below the median but were positive and significant for traffic above the median, with relative risks of 1.056 and 1.050, respectively. On the other hand, PM coarse Zn was counterintuitively positive and significant for traffic below the median, while positive but not significant for traffic above the median. PM_2.5 was positive but not significant for above or below median models.

Table 4. GLM model results by monthly traffic measures and air pollutant

	RR (95% CI)
Pollutant	Below the Median	Above the Median
EC	1.022 (0.983–1.064)	1.047 (1.007–1.090)
PM2.5 Zn	1.047 (0.998–1.099)	1.056 (1.005–1.110)
PM coarse Zn	1.122 (1.056–1.193)	1.061 (0.995–1.130)
PM10	1.029 (0.984–1.076)	1.050 (1.004–1.099)
PM2.5	1.017 (0.974–1.061)	1.017 (0.973–1.063)
	Lower Quartile	Upper Quartile
EC	1.028 (0.972–1.086)	1.016 (0.963–1.071)
PM2.5 Zn	1.058 (0.987–1.133)	1.036 (0.969–1.108)
PM coarse Zn	1.124 (1.029–1.228)	0.972 (0.892–1.059)
PM10	1.034 (0.971–1.102)	1.014 (0.955–1.077)
PM2.5	1.038 (0.976–1.103)	1.004 (0.947–1.065)

When we limited these comparisons to the groups in the lowest quartile of traffic volume and in the highest quartile, only PM coarse Zn was significant, again for the lower quartile. EC, PM_2.5 Zn, and PM₁₀ were again positive in the upper quartile, but in this analysis were not significant. This loss of statistical significance may be related to the smaller sample size of persons in a quartile of children arrayed by traffic volume.

Findings from models that included distance from roadway were inconsistent with findings of traffic volume models. None of the pollutants was significant for the subgroup living within 150 m of a roadway with a traffic measure, and all except PM_2.5 were positive and significant in the subgroup living greater than 150 m of a roadway (Table 5). The same pattern was observed for persons living less than 300 m and those living greater than 300 m of a roadway.

Table 5. GLM model results by distance of home address to nearest major roadway and air pollutant

	RR (95% CI)
Pollutant	≤150 m of a Major Roadway	>150 m from a Major Roadway
EC	1.017 (0.949–1.090)	1.034 (1.003–1.067)
PM2.5 Zn	1.073 (0.987–1.167)	1.041 (1.002–1.081)
PM coarse Zn	0.996 (0.893–1.110)	1.101 (1.049–1.156)
PM10	1.033 (0.957–1.116)	1.039 (1.004–1.076)
PM2.5	1.021 (0.945–1.102)	1.014 (0.980–1.049)
	≤300 m of a Major Roadway	>300 m from a Major Roadway
EC	1.018 (0.969–1.070)	1.039 (1.004–1.074)
PM2.5 Zn	1.041 (0.979–1.107)	1.048 (1.005–1.093)
PM coarse Zn	1.025 (0.948–1.109)	1.113 (1.055–1.174)
PM10	1.010 (0.955–1.068)	1.051 (1.011–1.092)
PM2.5	1.002 (0.949–1.058)	1.021 (0.984–1.060)

Discussion

We found that both larger traffic volumes and smaller distances to the nearest major roadway were positively and significantly associated with larger numbers of childhood asthma visits, when compared with less traffic and larger distances. Results from some prior studies are consistent with these findings. A case-control study conducted in San Diego found that higher traffic flows (defined by quintiles) at the nearest street within a 550-m buffer were related to increased medical visits for children with asthma (English et al., 1999). Similarly, another study conducted in Southern California found that the risk of asthma-related outcomes were increased for those with decreasing distance from a freeway (Gauderman et al., 2005). On the other hand, this second study differed from ours in that they found no association between traffic volume and childhood asthma, although their measure of traffic volume was calculated differently from ours (i.e., estimated hourly traffic volumes were used to calculate the daily average number of vehicles traveling within 150 m of each residence, weighted by inverse distance from the home to each road. This local traffic density was expressed as traffic volume per square meter.) This finding seems to be an exception, as other studies have found associations between asthma outcomes and traffic volumes (Meng et al., 2006; Oyana and Lwebuga-Mukasa, 2004). As previously noted, comparisons between these studies are limited due to differences in traffic and distance estimation methods, whether or not air pollution data were included, and differences in the asthma outcome measure utilized. There does not appear to be an accepted or common methodology for conducting studies that relate traffic to asthma outcomes. This is in addition to the usual differences between studies, such as populations and settings.

Our results from models to assess associations between asthma and two traffic-related pollutants, Zn and EC, were less conclusive than our traffic and distance analyses. Our methods were limited by the lack of consistency in time resolution between the traffic measures and the pollutant measures. Traffic measures were estimated on a monthly basis, whereas pollutant measures were available on a daily basis. We are unaware of other studies that have attempted to incorporate measures of pollutant exposure from a central monitoring site, traffic and distance, and asthma exacerbation measures in modeling analyses. The lack of variation in pollutant measurements on a spatial scale makes it difficult to assess differences in model results when stratifying analyses by traffic and distance measures. A calculation of AADT times the length of road within 300 m for the central monitor found it to be above the mean for the study locations but within the middle 50% of the distribution. The central monitor was 370 m away from the nearest major road, which puts it in the lowest distance to major road exposure category. This implies that although the Jefferson Street site is believed to be generalizable for ambient air pollution in the metropolitan area, it does not represent the “near heavily trafficked roadways” experience. Model stratification applies to the outcome measure because different populations based on traffic or distance measures will have different numbers of asthma visits; however, stratification in this case does not allow for variations between the subgroups on daily pollutant exposure levels. Because the fixed monitor site was located 370 m away from the nearest major road, the pollution level obtained at the monitor may likely better capture the exposure by the subcohort who lived >300 m of a major road as compared with the rest of the cohort. Therefore, the stronger associations found among the former subcohort may reflect reduced errors in measurement assessment. On the other hand, it may also reflect the different distribution of socioeconomic status and lifestyles between the two subcohorts.

Besides the lack of spatial variation in daily pollutant measures, another reason for unexpected results for the GLM analyses may be due to our initial assumptions of the relationship between pollutants and traffic measures. For example, a recent study from the Los Angeles and Long Beach areas of California found that, although not significant, surface streets contributed more to EC exposure than major freeways (55% vs. 45%, respectively) (Wu et al., 2009). Further, when looking at PM_2.5 overall, the study also found that light-duty trucks contributed significantly more to PM_2.5 and only slightly less to EC than heavy-duty trucks.

Some other studies that have included a pollutant measure have estimated pollutant exposures based on traffic-related and meteorological measures. For instance, one study found that increased asthma was associated with estimated NO₂ from freeway roads (Gauderman et al., 2005), whereas another study found that nonfreeway traffic–modeled exposure was related to prevalent asthma (McConnell et al., 2006). Because including the pollutant measures was not the primary aim of our study, we did not extend the methods to include this approach.

Limitations

In addition to the methodological limitations in the GLM analysis discussed above, this study may be subject to other limitations to consider when interpreting results. All visit data included in the analyses were derived from visits to the health plan providers at one of the health plan’s medical centers. The integrated care delivery system of the health plan in this study should present fewer variations than other types of health care settings because of uniformity in patient forms and coding for the health care facilities. We are confident that the findings are valid for the population studied. However, generalizing to the entire population has limits inherent in the nation’s health care system. For example, noninsured people may exhibit different patterns for doctors’ office visits than insured people. Also, variations exist across delivery systems in access to care, patient coinsurance rates, and patient characteristics. Finally, although our visit data are drawn from clinical records and should be generally sound, it could of course include some coding errors due to variations in data entry, missing data, and physician coding of diagnoses.

Another potential limitation related to the visit data is that exposure is based on resident address distance from the roadway. We did not have data on school locations for members, which would have allowed us to consider exposure at the school address relative to the roadway. Basing exposure only on home address could result in exposure misclassification.

The mean monthly traffic measures used in this analysis rely on average annual traffic count data and estimates obtained from GA DOT. In creating monthly measures from annual data, it was assumed that monthly trends do not vary year to year and that traffic trends over multiple years are consistent within a county for a particular road class. Error in exposure classification due to these assumptions is likely to be small compared with the differences in exposure resulting from proximity to different classes of roads. Because of the large difference between traffic estimates for local roads (320–1900 vehicles per day) and traffic on major roads (10,000 vehicles per day in this study), we are confident that mean monthly traffic measures can be used to distinguish between subjects who live in completely residential neighborhoods and those who live near any major road or highway. Since traffic counts on some major roads are based on estimates rather than actual counts, subjects living near major roads other than highways have the biggest potential for exposure misclassification. Generally, this type of misclassification would bias results towards the null. Our approach does not take into account differences in exposure due to braking, accelerating, and idling of vehicles in areas of heavy congestion or at major intersections; such speed change phenomena have been shown in some previous work to be important with respect to health impacts (Riediker et al., 2004; Riediker, 2007). We also did not consider the effect of wind direction.

This study made use of available data on traffic counts and estimates for a detailed street network in the Atlanta region. It could be improved if more traffic count data were collected consistently throughout the region especially on major roads other than highways.

There are also limitations in the Zn and EC data. Pollutants vary in their amount of missing data, which should be considered when comparing findings of the different pollutants. Other limitations may include instrument measurement errors and the lack of data on the spatial variation of pollutants, as previously discussed. Although this study makes use of only a single monitoring site, it is important to note that the monitoring station is reasonably central to the health plan member population. Lastly, although there are concerns in using ambient air monitoring data to estimate and assign exposure in epidemiological analysis, at least one review study suggests that ambient air concentration measurements are adequate for acute effects time series studies (Sheppard, 2005).

Conclusion

We observed significant associations between childhood asthma acute visits to primary care doctors and traffic density around the home. Similarly, living closer to a major roadway was also associated with increased odds of childhood asthma visits. The consistency in these findings, along with those from other previous studies, suggests that motor vehicle traffic is an important contributor to childhood asthma exacerbations. Other research should consider methods that allow for consideration of both traffic measures and specific pollutants in modeling health outcomes. Future studies are also needed to encourage a “best practice” approach to research in this discipline.

Acknowledgment

The authors thank the ARIES coinvestigators and members of the ARIES Expert Advisory Committee for their valuable input on this study.

Funding

This study was supported by the Electric Power Research Institute, Inc.

Additional information

Notes on contributors

Amber H. Sinclair

Amber H. Sinclair is a research associate consultant, Dennis Tolsma is a senior research consultant, and Lauren Perkins was a researcher with Kaiser Permanente-Georgia, Atlanta, GA.

Steven Melly

John Spengler is the Akira Yamaguchi Professor of Environmental Health and Human Habitation and Steven Melly is a GIS specialist based in the Department of Environmental Health at Harvard University School of Public Health, Boston, MA.

Dennis Tolsma

Amber H. Sinclair is a research associate consultant, Dennis Tolsma is a senior research consultant, and Lauren Perkins was a researcher with Kaiser Permanente-Georgia, Atlanta, GA.

John Spengler

John Spengler is the Akira Yamaguchi Professor of Environmental Health and Human Habitation and Steven Melly is a GIS specialist based in the Department of Environmental Health at Harvard University School of Public Health, Boston, MA.

Lauren Perkins

Amber H. Sinclair is a research associate consultant, Dennis Tolsma is a senior research consultant, and Lauren Perkins was a researcher with Kaiser Permanente-Georgia, Atlanta, GA.

Annette Rohr

Annette Rohr is senior technical manager, Air Quality and Health, and Ronald Wyzga is senior technical executive and biostatistician at Electric Power Research Institute, Inc., Palo Alto, CA.

Ronald Wyzga

Annette Rohr is senior technical manager, Air Quality and Health, and Ronald Wyzga is senior technical executive and biostatistician at Electric Power Research Institute, Inc., Palo Alto, CA.