Document x11oRgGY70GzZZxr5QZm0Jqa0

aCHNICAL PAPER ISSN 1047-3289 1.Air & Waste Manage. Assoc. 50175-180 Copynght 2oM, Air &Waste ManagementAssociation Distance-Weighted Traffic Density in Proximity to a Home Is a Risk Factor for Leukemia and Other Childhood Cancers Robert L. Pearson Radian International LLC, Denver, Colorado Howard Wachtel University of Colorado, Boulder, Colorado Kristie L. Ebi Electric Power Research Institute, Falo Alto, California ABSTRACT occupational exposure to elevated concentrations of benzene is a known cause of leukemia in adults. Concentrations of benzene from motor vehicle exhaust could be elevated alonghighly trafficked streets. Severalstudieshave reportedsignificantassociationsbetweenproximity to N h l y traffickedstreets and the occurrence of childhood cancers and childhood leukemia. These associationsmay be due to chronic exposureto benzene or other carcinogeniccomponentsof vehicleexhaust fromthesenearby streetsor to some other factor (e.g., noise, increased light exposure, or some unaccounted-for socioeconomic variable). We used data forhomes studiedin an earlier childhood cancer study conducted in Denver, CO, in the 1980s.No air pollution measurements were made in the original study. We identified the highest trafficked street near each study home and obtained the traffic density in 1979 and 1990. Traffic density Was weighted for the distance from the street to the home using 3 different widths of Gaussian curves to approximate the decay of the emissions into the surrounding neighborhoods. The associations between the 750-ft-wide distanceweighted traffic density metria and all childhood cancers c IMPLICATIONS Motor vehicles are a significant source of air pollution emissions, includingbenzene,which may be carcinogenic. This study presents further evidence that children who live near heavily traveled streets or highways may be at elevated risk of developing cancer, including leukemia. Children living near such streets could be exposed via inhalation or exposure to soil where emissions deposit. Alternative hypotheses for these results could involve noise, increasedlight exposure, or some socioeconomic Variable. Future studies with actual exposure measurements are needed. \ and childhood leukemia are strongest in the highest traffic density category(120,OOOvehicles per day WD]). The odds ratio is 5.90 (95%confidence interval [CI] 1.69-20.56) for all cancers and 8.28 (95%CI 2.09-32.80)for leukemia. The results are suggestive of an association between proximal w h trafficstreetswith traffic counts120,OOOVPD and childhood cancer, including leukemia. INTRODUCTION Many occupational studies have linked chronic exposure to high levels of benzene to the occurrenceof leukemia in The association has been mainly with acute myeloid leukemia (Ah4L).1,2Although AML is rare in children, this raises the possibility that leukemia and other cancers in children could alsobe associatedwith chronic exposure to elevated levels of benzene, a major constituent of volatile organic compounds (VOCs) from motor vehicle^.^,^ It has been estimatedthat over 80%of the total benzene emissions in the United States originate from motor vehicles: Benzene has alsobeen found in tob;kco smoke? Elevated levels of motor vehicle exhaust are likely to be localized (from a few hundred up to one thousand feet) near heavily traveled roadways,'@'' which could produce localized high levels of VOCs containing benzene13J4that could also deposit in soil, but no measurements are available. Although traffic-related ambient air concentrations of benzene are well below occupational exposure levels normally considered leukernogeni~t,h~e~ U.S.Environmental Protection Agency (EPA) believes chronic exposures may pose a risk for leukemia or cancer in general.I3The causal agents for the most common childhood leukemia, acutelymphocytic leukemia (ALL), arelargely unknown, although severalstudies suggest an associationwith some measure of trafficden~ity.'~-T''raffic density may act as a surrogate exposure indicator for elevated levels of VOCs containing benzene. Journal of the Air & Waste Management Association 175 Pearson, Wachtel, and Ebi 'i 1 Savitz and Feingold16reported that traffic density is associated with all childhood cancers and with childhood leukemiain Denver, CO. In their study, traffic density was defined as the traffic count on the street of address of the home, usually located in frontof the home. They reported an odds ratio for children living on streets with traffic densities of 210,000 vehicles per day (VPD) of 3.1 (95% confidence interval [CI] 1.2-8.0)for total childhood cancers and 4.7 (95% CI 1.6-13.5) for childhood leukemias. These results suggest that children living on streets with trafficcounts of more than 10,000VPD are 3-5 times more evaluated many factors as potential confounders,including the child's age, gender, year of disease diagnosis, and residential stability.Savitz et al. also controlled for family electricity consumption, parents' ages, parents' ocmpations, parents' education levels, per capita income, traffic density, and family cancer history. Also evaluated as con. founders were the child's prenatal exposure to tobacco smoke and alcohol, birth order, birth weight, medications, illnesses, and X-rays. Savitz et al. found evidence of an elevated cancer risk associated with young maternal age, lower father'se&. 1 C a C V S I L V V likely to contract cancer than are children who live on streets with fewer than 500 VPD. cation, lower per capita income, maternal smoking dur- E ing pregnancy, and traffic density, all of which were taken t More recently, Feychtinget al."retrospectively mod- into account as potential confounders of the association 5 eled concentrations of nitrogen dioxide (NO,) from mo- between cancer and proximity to power lines. The od& t tor vehicle exhaust in Sweden. They reported that at NO, ratios for these factors were not large, and taking them < levels >80 ~ g / mout~side a child's home, the total child- into account did not change the point estimates, indicat- ..4 hood cancer odds ratio was 3.8 (95% CI 1.2-12.1). The ing that they were not confounders. Subsequent analyses t 3 1`-1s NO, concentrations were estimated using mobile source (Savitzand FeingoldI6)supported the suggestion that there I 5. air pollution models from the Swedish Environmental are few, if any, confounding factors in these data. I Protection Agency as the 99th percentile of the 1-hraver- i II": ages for one year. EXPERIMENTAL METHODS ! m I* Knox and Gilmanl8examined spatial clustering pat- We employed the case-controlchildhood cancer epidemio- 1 J.hFI ; 3m ` terns among 22,458 children who died from leukemia or other cancers in England, Scotland, or Wales from 1953 logic data used by Savitzand Feingold.16The data are from I a study conducted by Savitz et al.19 of the association be- to 1980.They found excess childhood leukemias in prox- tween magnetic fields from electric power lines and child- imity to motorways and industrial plants likely to be hood cancer in the Denver metropolitan area. Details of sources of airborne toxic substances such as VOCs. The the methods used to identify cases and controls have been largest excessesof cancer associatedwith trafficwere found provided e1~ewhere.Iln~ summary, eligiblecasesweIe chil- I 1 k m or less from motorways. Knox and Gilman con- dren from 0 to 14 years of age who were diagnosed with cluded, "Childhood cancers are geographically associated cancer from 1976 to 1983 and who lived in the greater with two main types of industrial atmospheric effluent, Denver metropolitan area. The 320 eligible cases consisted namely: (1)petroleum-derived volatiles and (2) kiln and of 97 leukemias (including 78 ALL), 59 brain cancers, 30 furnace smoke and gases, and effluent from internal com- lymphomas, 32 soft tissue cancers, and 102 other forms Of bustion engines" (p. 151). cancer. Data on these children were obtained from the This study is an extension of the work by Savitz and Colorado Central Cancer Registry; 259 controls were se- Feingold.16 We developed an exposure metric based on lected by random digit dialing with individual matching the highest trafficked street surrounding the house, not on gender, age f 3 years, and telephone exchange area. simply the street of address (usually located in front of The Colorado Central Cancer Registry of the Colorado the house), and took into account the decay of air pollu- Department of Health was the principal source of cases, tion concentration with distance from the street. These with complete coverage of the study area from 1979 to improvements are intended to reduce exposure 1982. To complete case ascertainment for 1976-1978 and misclassification for traffic air pollution. The results are 1983,recordswere reviewed of area hospitalsthat provided presented to allow for a direct Comparison of the results diagnosis and treatment for childhood malignancies.Over of this improved traffic density exposure metric with the 95% of the cases were microscopically confirmed, with an results reported by Savitz and Feingold.16 additional 3% confirmed by direct visualizationor radiog- Parents of case and control children were interviewed raphy. Diagnostic accuracy was also assured through le- in the original study to obtain data about their child's view by pediatric oncologist^.^^ past exposure to agents and activities potentially associ- Savitz et al. found elevated traffic density to be ass@ ated with cancer, including parents' cigarette smoking and ciated with childhood cancer, but no assessment of air the use of pesticides in and around the home. In order for pollution exposure was made in their study. To follow UP a factor to be a confounder, it must be associated with this observation, we used other publicly available data both the disease and exposure of interest. Savitz et al. that could act as markers for air pollution exposure for 176 Journal of the Air & Waste ManagementAssociation Volume50 FebruaryZ@ tw homes'of the case and control children. We obtained 1979 traffic density data from the Colorado Department of Transportation and 1990 traffic count data for Denver area streetsand highways from the Denver Regional Council of Governments.These years were chosen because they were the two years closest in time to the period of exposure of the cases studied by Savitz et al.19when coordinated traffic counts were made across the Denver metropolitan area. If a street did not have a traffic count, we assigned the average of the traffic counts for all streets with the same traffic-carrying capacity code from a geographic information system (GIS) digital mapzoof all of &e streets and highways in the Denver metropolitan area. Streets with low traffic-carrying capacity codes without traffic measurements (narrow residential and cul-de-sac streets)were assigned a default traffic count of 100VPD. We compared the traffic density measurements for these two years for all streets near the study homes and noted the similarities and trends between the years. Although the 1979trafficdensity data were collected closer in time to the other data collected in the Savitz et al. study,lgwe preferred to use the 1990 traffic density data because traffic counts for more streets were available.We compared the 1990 traffic density with the traffic density for 1979 (Figure 1)for those homes that had measured traffic density on the nearby highest trafficked street for both years (n = 110). Traffic density increased on average 22% from 1979 to 1990 and is highly correlated between these years. The R2 is 0.93. The straight line in Figure 1 is the least-squares linear regression. Therefore,we used the 1990 traffic data because there is a high correlation for traffic counts between these years 200000 # 150000 nE 0 100000 CI o\ b o\ 50000 R2 = .93 J . 0 0 50,000 100,000 150,000 20b,300 1990 Traffic Density FiQUre 1. A comparison of traffic density measurements in VPD for '979 and 1990for 110study homes inthe Savitz and Feingold study.14 Volume 50 February 2OOO Pearson, Wachtel, and Ebi with a fairly uniform trend of increasing traffic. A hightraffic street in 1979 also would be a high-traffic street in 1990with about the same rank ordering by traffic count, because few streets fall far from the least-squares regression line in Figure 1. Each of the study homes was geocoded for location from its street address. Using GIS methods, we identified the highest trafficked street within 1500ft in any direction from each home and calculatedthe perpendiculardistance from the center line of the streetto the center of the home. The concentration profile of airborne exhaust pollution from motor vehicles traveling on a street could approximate a Gaussian (normal) distribution11~12~sizm1~ilazr2 to that shown in Figure 2. This idealized profile may vary due to many factors such as prevailing winds, turbulence generated by passing vehicles, and the dispersion blocking effect of nearby buildings and hillside^.^^^^^ To bracket the spatial action that these and other factors may have on air pollution concentration profiles, 3 different widths (250, 500, and 750 ft) of Gaussian curves were used to distance-weight the traffic counts. The width of each Gaussian curve was determined at one-half of its maximum height (Figure 2). Distance-weightedtraffic density was computed by multiplying the traffic counts in VPD for the highest trafficked streetby the values derived from each of the 3 weighting curves at the street to home distance. For instance, a street with a traffic count of 10,000 VPD located at the distance from the home shown on the left in Figure 2 would have a distance-weightedtraffic density of 2500 VPD at the home. Odds ratios (OR) and 95Yo confidence intervals (CI) were computed between childhood cancer and childhood leukemia and the different distance-weighted traffic density metrics using standard statistical programs for stratified analysis.z5 RESULTS Savitz and Feingold16evaluated the association between childhood cancer and traffic counts up to 210,000 VPD for 328 case and 262 control homes, using the traffic counts on the street of address of each home. The associations for all childhood cancers and leukemias reached statistical significancein the highest category despite small numbers. Using their street of address data, we recomputed odds ratios for traffic counts up to 220,000 VPD (Table 1).The associations were not statistically significant or consistently elevated, except in the highest category, where the odds ratios abruptly increased. The associations were not precise because of the small number of controls, particularly in the higher-VPD categories. We determined 3 distance-weighted traffic density metrics for the highest trafficked street proximate to the home for 579 of the 590 homes analyzed by Savitz and Journal of the Air & Waste ManagementAssociation 177 Typical Concentration Profile of Traffic Air Pollution Corridok Width \ Figure 2. Generalizedprofileof concentrationof airborneexhaust pollutionfrom vehiculartraffic inthe absenceof a crosswind. In this example,the distance-weighting value for the home on the left will be 0.25. If the traffic density on the street is 10,000 VPD, then the distance-weighted traffic density at the home on the left would be 2500 VPD. Table 1. Odds ratios betweentraffic density on the street of addressand all childhood cancersand childhood leukemias using data reportedby Savitz and Feingold.16Thistable extends the earlier analysisto traffic strata above 10,000 VPD. Distance-Weighted Traffic Density (VPD) Controls N t500 500-4999 5000-9999 10,000-14,999 15,000-19,999 220,000 238 10 9 3 Oa 2 All Cancers N OR 95% CI 280 1.o Reference 19 1.62 0.74-3.54 11 1.04 0.42-2.55 6 1.70 0.42-6.87 4 8 3.40 0.72-16.17 Leukemias N OR 95% CI 81 1.o Reference 4 1.18 0.36-3.85 5 1.63 0.53-5.01 1 0.98 0.10-6.87 2 5 7.35 1.40-38.60 aOddsratios cannot be computed due to lack of controls in this category. N =number of children in each category, OR =odds ratio, and 95% CI = 95% confidence interval. Table 2. Odds ratios between 7504 distance-weightedtraffic density and all childhood cancers and childhood leukemias. Distance-Weighted Traffic Density (VPD) Controls N c500 500-4999 5000-9999 10,000-14,999 15 000-19,999 >20,000 118 86 32 14 6 3 All Cancers N OR 95% CI 121 1.o Reference 101 115 0.78-1.68 55 1.68 1.02-2.80 18 1.26 0.60-2.66 7 115 037-3 51 18 5 90 1 69-20 56 N 38 26 21 2 2 8 178 Journal of the As & WasteManagement Association Leukemias OR ~~ 10 0 94 2 04 0 48 1 04 8 28 95% CI Reference 0 53-1 66 1 05-3 95 0 10-2 21 0 20-5 34 2 09-32 80 - Volume50 February @ FeingoldI6i l l of the 590 homes did not have complete data). Table 2 shows the associations between all childhood cancers and childhood leukemias and the 750-ft distance-weighted traffic density metric. Note the shift of homes to the higher traffic density strata in Table 2 compared to Table 1when taking the highest trafficked street into account. The association between traffic density and all cancers and leukemia only becomes elevated and significant in the 220,000 VPD category. The results for the Soo-ft-widthweighting curve are similar with odds ratios for220,000 VPD of 9.45 (95% CI 1.19-74.79) for all cancers and 12.18 (95% C1 1.33-111.8)for leukemia. Both of these odds ratios are statistically significant. For the 250ft distance-weightedtraffic density metric, there were no controls with 15,000 or more VPD, so odds ratios could not be computed. If the children are groupedby exposure to greater than SOOVPD(750-ftdistance-wwted) versus lessthan 500VPD, the odds ratios become 1.70 (95% CI 1.01-2.86) for all cancersand 2.08 (95%CI 1.06-4.07)for leukemia.Even though these risks are small, they are statistically significant. CONCLUSIONAND DISCUSSION This study has several limitations. As in most studies of childhood cancer, the number of cases and controls in the high-exposure category is small, due to the relative rarity of childhood cancer. These small numbers affect the precision of the observedassociations. The confidence limits are wide because of the small number of cases and controls, but some of the point estimates are still statistically significant. To make the group of children as large as possible, Savitzet al.19made their diagnosisperiod fairly wide (1976-1983). The cumulativepre-diagnosis exposure period for these children may have stretched for more than 10years, but most of the children were less than 10 Years old when diagnosed. Savitz et al. did not directly measure traffic density for each child. Citywide coordinated traffic counts are only made in the Denver area about every 10 years. Retrospective estimates of exposure for the pre-diagnosis time period for each child could be extrapolated from these periodictraffic counts, assuming the child had lived in one residence for the whole period. However, data collected by Savitz et al. on the residential mobility showed that many children had moved. Therefore, we did not attempt to retrospectively estimate the traffic exposure during the pre-diagnosis exposure period for each child. To do so would require knowledge of each child's residential history during the relevant (and unknown) etiologic period. The results suggest that distance-weightedtraffic density is a risk factor for all childhood cancers and for childhood leukemias. This is consistentwith the earlier findings Volume50 Februaty 2000 Pearson, Wachtel, and Ebi of Savitz and Feingold,16Feychting et al.,''and Knox and Gilman." Elevated risks are found for 220,000 VPD for the 500- and 75043 metrics. There are cases but no controls with 220,000 VPD for the 2504%metric, so odds ratios cannot be computed. There is no evidence of elevated risk for distance-weighted trafficdensities less than 20,000 VPD. At 220,000 VPD, the risks of childhood cancers and leukemia increase with decreasing distance-weighting from the street, although the numbers are small and the confidence intervals overlap. These results are suggestive of a threshold effect for highly exposed children when distance-weighted traffic density near a home rises well above the average value for the area. This apparent threshold effect is to be expected where high densities of VPD increase the concentration of vehicle exhaust far above the ambient background in the surrounding neighborhoods. The gradient of traffic air pollution may be very steep (Figure 2) and diminish very rapidly with distance from the street. Thus, only those children who live very near high-traffic streets (within 750 ft) will have an exposure that is far above "ambient background" for the community. Noise, increased light exposure, or some socioeconomic factor may also help explain these results. For distance-weighted traffic densities 220,000 VPD, the associations we observe are stronger than those reported by Savitz and Feingold.16This may be due to possible misclassifications of exposure in the Savitz and Feingold study, which only analyzed traffic density on the street of address in front of a home. This would not account for high exposure coming from a busy street either behind or beside the home. Feychting et al. used a retrospective estimate of mobile source NO, concentration as their metric." The metrics used in our analyses are simpler to construct than estimating NO, concentrations and are also strongly associated with cancer. The confidence limits are wide because of the smallnumber of cases and controls, but some of the point estimates are still statistically significant. Together, these results are suggestivebut do not constitute conclusive evidence of an associationbetween living near a high-traffic street and increased incidence of childhood cancer, particularly leukemia. Further studies specificallydesigned to test this and other hypotheses are needed. Measurements of ambient or personal exposure to various pollutants are also needed. This studyused distance-weightedtraffic densityas an exposure surrogate for mobile source air pollution. This metric can be improved. Concentrations of motor vehicle VOCs containing carcinogens such as benzene can be retrospectively estimated for each home using mobile source models. The models could be used to estimate emissions from passing vehicles and the ambient concentrations Journal of the Air & WasteManagementAssociation 179 ... -, i "**P' 10 Iq 8 !; al 1 1 Lm ;: '; .b 1% %`I). Pearson, Wachtel, and Ebi resulting after dispersion into the neighborhood. Finally, measurements are highly desirable.If benzene and other emissions in motor vehicle exhaust are associated with childhood cancer, then estimates of these concentrations should be more stronglyassociatedwith childhood cancer than is distance-weighted traffic density. ACKNOWLEDGMENTS We would like to thank Dr. David Savitz for sharing the data from his epidemiologic study as well as his encouragement and helpful suggestions in conducting this research. The authors would like to thank David Simpson and SuzanneStrasserfor their statistical analyses presented in this article, and Jim Rowe and Jim Crawford for performing the GIS analysis of nearby streets. Finally, we would like to thank Dr. Leeka Kheifets and Dr. Ron Wyzga for their support and constructive suggestions. This work was sponsored by the Electric Power Research Institute under Contract No. WO2964-22. REFERENCES 1. Le Noire, M.; Claude, H. "Sur un cas de purpura attribue a l'intoxlcation par le benzene," Bull. Mem. SOC.Med. Hosp. Paris 1897, s", .er. 1631. 2. Delore, P.; Borgomano, J. "Leucemieaique au cours de intoxication benzenique. Sur originetoxique de certainesleucemiesaigues et leurs realtions avec le anemies graves,"J. Med. Lyon 1928, 9, 227. 3. Aksoy, M. Benzene Heterotoxicity. In Benzene Carcinogenicity; Aksoy, M.. Ed.: CRC Press: Boca Raton. FL. 1988: DD 59-112. 4. Goldstein,B.D. "Hematoxicityin humans,"). Toxicol.Environ. Health 1977, 2 (Suppl.), 69-105. 5. Klassen, C.; Amdur, M.; Doull, J. Casarette and Doull's Toxicolog, the Basic Science of Poisons. Klassen, C., Ed.; Macmillan Publishing Co.: New York, 1986. 6. Wallace, L. "Major sources of benzene exposure," Environ. Health Perspect. 1989, 82, 165-169. 7. Conner, T.; Lonneman, W.; Seila, R. "Transportation-relatedvolatile hydrocarbon source profiles measured in Atlanta," J. Air & Waste Manage. Assoc. 1995, 45, 383-394. 8 . Report to the Scientific Review Panel on Benzene; California Air Resources Board: Sacramento, CA, November 1984. 9. Wallace, L. "Cigarettes:Point source for benzene exposure,"Environ. Health Perspect. 1993, 101, 14-16. 10. Ott, W.R. "Developmentof criteria for siting air monitoring stations," J. Air Pollut. Control Assoc. 1977, 6 , 543-547. 11. Kono, H.; Ito, S . "Micro scale dispersion model for motor vehicle exhaust gas in urban areas-OMG volume-source model," Atmos. Environ. 1990, 24B,243-251. 12. Sivacoumar, R.; Thanasekaran, K. "Line source model for vehicular pollution prediction near roadways and model evaluation through statistical analysis,"Environ. Pollut. 1999, 104, 389-395. 3. Motor Vehicle-Related Air Toxics Study; EPA-420-R-93-005;U.S. hvi. I ronmental Protection Agency, Office of Mobile Sources: Ann Arbor, 1 MI, 1993. i 4. Kagawa,J. "Atmosphericpollution due to mobile sources and effects I on human health in Japan,"Environ. Health Perspect. 1994, 102 (Suppi, 4), 93-99. 5 . Aksoy, M. Benzene Heterotoxicity. In Benzene Carcinogenicity; A k q , M., Ed.; CRC Press: Boca Raton, FL, 1988, pp 113-151. 6. Savitz, D.A.; Feingold, L. "Associationof childhood cancer with resi. dential traffic density,"Scand. J. Work Environ. Health 1989, 15, 360. I 363. 7. Feychting, M.; Svensson, D.; Ahlbohm, A. "Exposure to motor ve. hicle exhaust and childhood cancer,"Scand. J. Work Environ. Health 1998.24. 8-11. 8. Knox; E.G.; %man, E.A. "Hazard proximities of childhood cancers in Great Britain from 1953-1980,"J. Epidemiol. Commun.Health 1997. 51, 151-159. 19. Savitz,D.A.; Wachtel, H.; Barnes, EA.; John, E.M.; Tvrdik, J.G. "Case- control study of childhood cancer and exposure to 60-Hz magnetic fields,"Am. J. Epidemiol. 1988, 128, 21-38. 20. EtakMap PREMI.0, Denver; Etak, Inc.: Menlo Park, CA, 1994. 21. Kono, H.; Ito, S . "A comparison of concentration estimates by the OMG volume-sourcedispersion model with three line source disper- sion models,"Atmos. Environ. 1990, 24B, 253-260. 22. Boubel, R.; Fox, D.; Turner, D.; Stern,A. Fundamentals ofAirPollution, 3rd edition; Academic Press: San Diego, CA, 1994. 23. Bensen, P.E. CALNE3-A Versatile Dispersion Model for Predicting Air Pollution Levels Near Highways and Arterial Streets; California Depart- ment of Transportation, Ofice of Transportation Laboratory: Sacra- mento, CA, 1979. 24. Bensen, P.E. CALNE4-A Dispersion Model for Predicting Air Pollution Concentrations NearRoadways; Stateof CaliforniaDepartment of Trans- portation: Sacramento, CA, 1989. 25. SAS/STAT Version 6; SAS Institute, Inc.: Cary, NC, 1990. About the Authors Robert L. Pearson is a project manager for Radian International, at 707 17th Street, Suite 3400, Denver, CO 80202; phone: (303) 292-0800; fax: (303) 292-5860. He is also a member of the adjunct faculty of the Universityof Colorado, Denver campus. Howard Wachtel is a professor of Electrical and Computer Engineering at the University of Colo- rado, Campus Box 425, Boulder, CO 80309, and is also a consultant for Radian International. Kristie L. Ebi is manager, epidemiology and toxicology, in the Environment Divi- sion of the Electric Power Research Institute,PO.Box 10492, Palo Alto, CA 94303. 180 Journal of the Air & Waste Management Association