Document Ed81G1J8x6xq0xMwjpZpd9Nxn

Scarid J IVork Enwbon Heullh 1989;15:360-363 Association of childhood cancer with residential traffic density by David A Savitz, PhD,' Lisa Feingold, MSPH,' S A V I T Z D A , FEINGOLD I-.Associatioil o f chiltlhood cancer witti rciidciilial ti-attic density. Scrriicl J U'ork E/7\~ir.oH~rol//7 1989;15:360-363. Daln f r o i n a tecentlv coiiipleted caie-I-efei-en1\tucly o f cliilclhood cancer were used to e\plore a po\iible i o l e o t e111iroiiiiienial c\posiii-es I'roiii trafl'ic e\liiiiist. Tlie street addresses of 328 cancer patients and 262 I~optilatioii-hasedIcf'ereiit\ w r e used to uiiigii traft'ic den5 i i y (vehicles per day) as a niarker o f p o t e n r i a l e\poitirc to iiioior \,eliicle e ~ l i a t i s l ,,411 odd\ ralio 0 1 I .7 [95 u/o confidence inrerval (95 O h CI) I.0--2.R] \\mfound 1.01. the iota1 n u m h e r 0 1 childhood cancets iind 2.1 (95 71)C I I,1-4.0) for l e u h e m i a ~in a contrast of high and l o w traffic density addreises ( 2 500 YCI~LIS < 500 vehicles per day). Stronger associations w e i r found \villi a ti-affie denily cutoff \core 01. 2 10 000 vehicles per day, with imprecise odds ratios of 3.1 (95 CI 1.2--8.0)arid 4.7 (95 Y o CI I .h-l3.5) tor l l i c lola1 nuiiibri 01 c a i i c c r ~and i e u h c i i i i a . i c s p e c i i v c l ~ .AdjLi\riiicnI 101 s u ~ i x c t e d[ i s h lac lor^ l o r ciiilcIhood cancer did not subsiantially change these rre5uIts. Though tlic results are inconclu\ive, tlic ideniilicd associalioii warranis fui-[her evaluatioii. Kev /erri~s:ail- pollution, b e n m i e . brain cancer, leukemia. nioior ieliicle exhaust. Environmental influences on the development of childhood cancer have been of interest for some time (I- 3), but epidemiologists have had very little success in identifying exogenous etiologic agents for these diseases (4).Given that the etiology of childhood cancer is largely unexplained a n d the study of environmental agents has been limited, thei-e is a need for further exploratory evaluations of exposures not previously considered. A n examination of risk factors for the most prevalent childhood cancers (leukemia, brain cancer, and lymphomas) in adults would also provide few etiologic clues. O n e of the few established causes of adult leuke- mia is benzene (3,whereas other solvents are suspected of causing leukemia (6) and possibly brain cancer (7). Evidence that benzene in gasoline produces detectable concentrations in the ambient ail- as high as 10-20 p g / d in urban areas (8, 9 ) makes this potential consequence of traffic density worthy o f evaluation, along with the contribution of motor vehicles to enviroiiniental lead and hydrocarbon levels. Data from a study designed to examine electimnayxtic fields and childhood cancer were used to asses5 \vhcther a n y of iliese esposui-es might be related to childhood cancer' incidence in Denver. Colorado. Materials and methods The details of the study methods have been provided elsewhere ( I O ) and are therefore outlined only briefly in this report. The eligible cases consisted of all incident childhood cancers (age of child 0-14 years) diagnosed between 1 January 1976 a n d 31 December 1983 among residents of the 1970 Standard Meti-opolitan Statistical Area (Adanis, Arapahoe, Boulder, Denver, and Jefferson counties) of Denver, Colorado. The Colorado Central Cancer Register was complete tor the period 1979-1982, and area hospitals provided x - cess to additional cases, which allowed comprehensi\,e ascertainment throughout the study period. Of the cases, 95.2 (5'0 were histologically confirmed, and an additional 3. I T o were confirmed by direct visualiiation o r radiography ( 1 1 ) . The referents were selected by random dizit dialinp, matched to cases by the age (plus 01- minus 3 year-5). sex, and telephone exchange area of the patient. A though referents \\ere sou_ght lor each case, the i i i c l t i sioii of a given case or referent in the analysis \vas iiot affected by our abilit!, to produce a marched pair. The intent \\as to produce a gi-oiiii halaiiced oil ilie marchin@criteria. and the data \\ere analhred rhiniigli stratified I-atliei-than marched anal!,iii ( I ? ) , In older to ensure that tlie i-elet-eni5 occupied t l i c i i homes i n tlie patient'j telephone e\change area at [lie time of tlie c a v ' i dia$iioii\. onl!. purentia1 r'etew!;t\ \\ho had btayed iii tlieii home I'roni rhe time 01 t h e ca>e'\ cliayno\i\ I O tlie t i r i i c ' o f t h e i r i t r t v ie\\ i\\l i e n !lit\ \ \ e r e detcr'inined I O !-e.>iclei t 1 [fir C ~ I . I C C Itelephone c\change nre:~)\\e~-e eligililc. :\ltliot~gli it \\otild II:I\L heen pret'erahle I O include porelitin1 i.elerc:~i, \\ IIO pi-e\.iouslh reiiclsd i n [ h e area biir \ah~ci;ui.ri~ln\i o \ zd. t t ~ \e\ a s !IO mtci1;ll1lbt11 I'W iciptltlt! 1112 m i L i ~ ~ l ~ i n ? i i i c l i r:iniilie< iii their tien l o i n r ! o n < . T l i u ~iret':: c'iils \!ere re5trictecl to h a t e been o ~ c t i p : i i ~ Ot ~I r l i e i ! JIOI~IP. ai the time the case \\'a<diagnosxl (\\heyea\ ;;I,+ \ \ e r e not s o i-ejtinctedj; theretoi-e more residen: ~ c i l l : ~siable referent\ than case5 \\,ere produced ( I O ) . 111 (hi\ analysis, the residence occupied at the time 0 1 rhc case's diagnosis (01 at the I-efei-ent'sage when I I I C n ~ a t c h e dcase was diagnosed) was characteri7ed \\it11 respect to traffic density. The location of each reported a d d r e s was identified on a detailed street map nt tlie Denver Standard Metropolitan Statistical Area. ( J I ~t l i c basis of the precise location, tlie a p p r o x i m a t e Ioiation of' the homes was identified o n traffic densiI! maps provided by t h e Denver Division of Highway Planning and Research, and the specific segment of ilic' street to which the density figures applied was carei'1111y noted. On the basis of the m a p s a n d lists of the stieets and 11-afficvolumes, the home was assigned a ti-affic densir), score of < 500 vehicles per day o r the recorded number of vehicles per day (ranging from 500 t o over 100 000 pel- day). Data on potential confounders wei-e collected principally through an interview with the parent (mother preferred) regarding numerous potential risk factors >L!cIi as family demography, cancer history, exposure ti) N rays and medications during pregnancy, parents' occupational history, and mothci-'s and child's illn e w s . Wire configuration codes were obtained as a mal-ker of long-term magnetic field exposure (IO). On the basis of associations with childhood cancer (either ri5k factoi-s or ai-tifacts of the study design), the following variables were examined in detail: sex, age, yeai01' diagnosis, single-family versus multiple occupancy residence, residence in 01- out of Denver at birth, r e d e n r i a l mobility from birth to diagnosis, mother's age. fathei-'s education, per capita income, mother's rnioking during pregnancy, and wire code at the time 01' diagnosis. (Note that the mal-kers of residential stahilit), would address the possible bias from the res: i-ictioii of refei-ents based on long-term occupancy of their homes.) Odds ratios a n d 95 "10 confidence intervals weie calculated with test-based methods (13). Stratified anal y w produced Mantel-Haensrel adjusted odds ratios aiel confidence intervals (14). Finally, unconditional logistic regression analysis was used to control for multiple potential confounder-s (IS). Results 4 iota1 o f 356, eligible cases (101 leukemias, 83 acute I! iiiphoc!'tic leukemias, 67 brain cancerq. 3 5 Iyniphoi i i i i ~ .32 soft-tissue tumors, and I19 others) and 278 12iiIen t i a1 I-efer ent s were id en t i fiecl t h roug 11 rand oin iligit dialing. For the eligible cases. traffic densities \ \ e r e obtained for 328 (92 To), \\it11 a siinilai- response ac! O\I<diagnostic groups. The response for referents \\:I\ 94 (hon :l!c basi5 of 262 completed residencej, \\ l l i i h iq comt?:i.:d \\,ith the estimated 79 Y o response i n telephone screening ( 10) t o produce a n overall 75 O7o response. Nonrespondent cases were more likely to have been diagnosed earlier and be Hispanic or Black; no information was available concerning nonrespondent referentr. The crude results relating traffic density to childhood cancer incidence are provided in table 1 . The association for total childhood cancers [odds ratio ( O R ) 1.71 was slightly greater for leukemia (OR 2. I ) though not acute lymphocytic leukemia (OR 1.6j. Similar effect estimates (OR 1.6-1.8) were obtained foi-brain cancer and other cancers, with a diminished association !'orsoft-tissue tumors and the absence o f a n association for lymphomas. Because the referent selection procedures failed to provide a referent for each case, the possibility of bias due to incomplete coverage was examined in a matched analysis. Though the precision was greatly reduced, the matched odds ratio for all childhood cancers was 2.0 (based on 36 discordant pairs), similar to the results of the unmatched analysis. T h e m o r e highly exposed g r o u p ( 2500 v e h i c l e d d ) was subdivided for a n examination of the possible exposure-response gradients (table 2), although small numbers limited this analysis to the total n u m b e r of cancers a n d leukemias, A cutoff score of 5000 vehicles/d yielded oddr ratios of 1.8 [95 070 confidence interval (95 Vo CI) 0.9-3.31 for the total number of cancers a n d 2.7 (95 To C I 1.3- 5.9) for leukemias in the highest exposure group. An even more pronounced gradient was found with a higher cutoff score of I O 000 vehicles/d, with o d d s ratios of 3.1 (95 O7o CI 1.2-8.0) for total cancers and 4.7 (95 Vo CI 1.6-13.5) for leukemias in the highest exposure group. As noted earliei-, data on potential confounders were obtained in an interview with the subjects' parents, with adjusted results restricted to the subset of interviewed cases and referents (71 %' of eligible cases a n d 80 Vo of identified referents). This restriction alone (in the absence of an!' confounding) raised the o d d s ratios slightly, especially for leukemia (OR 2.3) and brain cancer ( O R 2.6). Adjustn1cnt To)-5 e ~a,ge, y e a of diag- Table 1. Odds ratios (OR) and 95 O/O confidence intervals (95 % CI) for traffic density and childhood cancer Groi;p Reference Case Leukemias Acute lymphocytic leukemia Lymphoma Brain Soft tissue Other Total Veh tclesiday 1500 2500 OR 238 24 95 % CI 81 17 68 11 29 2 52 9 28 4 90 16 280 48 2 1 11- 40 1 6 06-34 07 02- 30 1 7 08- 39 1 4 05- 44 1 8 09- 35 1 7 10-28 36 I Table 2. Dose-response gradient for traffic density and total childhood cancer and leukemias odds ratios (OR) and 95 ' 0 confi dence intervals (95 % CI) Vehicles per day < 500 500-4999 2 5000 500-9999 2 10 000 Referents (N) 238 10 14 19 5 Total cases N OR 95 O h CI 280 1 0 19 1 6 0 7-3 5 29 i a 0 9-3 3 30 1 3 0 7-2 4 18 3 1 12-80 Leukemias N OR 95 ' 3 CI ai 1 0 4 1 2 0 1-3 9 13 2 7 13- 59 9 1 4 0 6-3 2 a 4 7 1 6-13 5 Table 3. Sex and age-specific odds ratios (OR) and 95 % confidence intervals (95 % CI) for traffic density and total childhood cancer leukemias, and brain cancers for selected stratification factors ______ Referents Total cancers Leuhernias Brain tumors u(neexxppoosseedd/) uEnxepxoDsoesdeld OR 95 `10 CI uEnxepxooosseedd/ OR 95 `I Exposed/ unexoosed OR 95 "' I` Sex Male Female Age (years) 0-4 5-14 Residential stability Stable Moved Year of diagnosis Before 1980 1980 or later 101115 4/79 5/84 91108 2/77 11/108 6/94 8/100 15/127 1.4 0.6-3.1 13/86 3.0 1.0-9.2 19/91 3.5 1.3-9.4 9/120 0.9 0.3-2.4 12/71 6.5 1.7-25.4 15/127 1.2 0.5-2.6 11/91 1.9 0.7-5.3 171122 1.7 0.7-4.2 7/36 2.2 0.8-6.2 3/24 2.5 0.5-11.4 9/27 5.6 1.9-16.7 1/33 0.4 0.1-2.8 3/18 6.4 1.2-34.0 6/36 1.6 0.6-4.7 4/25 2.5 0.7-9.3 6/35 2.1 0.7-6.5 3/22 1.6 0 4-6 1 4/15 5.3 1 3-20 9 4/13 5.2 1.4-19.6 3/24 1.5 0.4-5.9 4/14 11 0 2.4-49.8 3/21 1.4 0.4-5.5 2/18 1.7 0.3-9.2 5/19 3.3 1.0-10.6 nosis, type of residence (single family, other), location at birth (Denver, other), mother's age, father's education, per capita income, and wire configuration code a t diagnosis had little effect except for a tendency of residence type to diminish the odds ratios and mother's age a n d mother's smoking to elevate the odds ratios. T h e logistic regression analysis for t h r total number of cancers, leukemias, and brain tumors with these three variables in the model (plus wire code for leukemia) produced adjusted odds ratios of 1.7 for the total number of cancers (95 Vo CI 0.8-3.6), 1.9 for leukemias (95 Vo CI 0.7-5.1), and 2.0 for brain tumors (95 Vo CI 0.7-6. I ) . Overall, these results provide evidence against substantial confounding by the measured potential risk factors. T h e sex- and age-specific results for the total number of cancers, leukemias, a n d brain t u m o r s are presented in table 3. T h e imprecision within jtrata is apparent, but there is a suggestion of enhanced associations for the females (especially for brain tumors) a n d a consistently stronger effect for the 0- to +yearold than I`or the 5 - to 14-year-old children. T h o u g h only crude results are reported in table 3. logistic iregression with adjustment [or residence t!.pe, mother's age. and mother'<\rnokiiig did not materially change the effect esrirnates, ;ilthoush the precision \\as t'tii-ther [reduced. 362 Discussion These results indicate an association between traffic density near the home occupied at the time of diagnosis and childhood cancer which was not accounted for by potential confounders. The odds ratios wei-e o n the order of I .6--2.0 for all cancers except lymphomas a n d soft-tissue tumors. Stronger but less prccise associations were observed for females, younger cliildren (age 0-4 years), and residentially stable children. Evidence of increasing risk with increasing traffic density was found for the total number of cancel-s and leukemias. Potential sources of bias include nonresponsc. differential mobility of cases and referents, unmeasured confounders, and nondifferential exposure mi\classification. T h e precision of many of the odd\ iratio estimates was less than would be desirable, and I a i i dom variation must be included among the posvble wtirces of e r r o r . In order for nonr-esponse io ! i a \ c biased the odds I-atios, there would have to have heen ;idifferential loss of high and lo\\ traffic densit! L L i w s and refet-ents (16). In spite o f the diffei-entia1 re\pcin\e for cases a n d referents, such a pattern is unlihel!. The w n e reasoning is applicable to [lie conccrii :`or the differential mobility of the cases and relerent\. ihe cases were unrestricted by patterns o t ino\'ement, b u t rsicr-ent, \\,ho rcninined in theii- home from thc riIne 0: tile i.a\c'< diagnosis to the time of intervie\\ \\ere ~ ~ ~ c l t ~ cTl chcel .I-elationship of residential mobiliI! tc) tl-af'i'ic densit! is u n k n o w n , although o n e might 5pt.cuIare that inor-e stable residents live in more desirablc. IO\SCI traffic density homes. Referents cannot be e i aluateil directl!, ior theii- representativeness of the t . \ l ~ ~ \ t i r ct.li5tribution in the study base that generated thc caws. hut aftei-adjustmeiir for a number of potential wurces o l selection bias, they should provide valid rmiilt 5 . Unmeasured confounders are always a possibility, especially when so little is known about the determinant\ o f childhood cancer. For example, a variety 01' environmental a n d social factors is associated with li\,iiif in rural versus urban settings. Any component of living i n a n u r b a n area that is predictive of childhood cancer would confound the results for traffic density. Limited data on childhood leukemia d o not support strong gradients by rural/urban status, with approvimately a 10 Vo increase in childhood leukemia mortality in urban areas found in one study (17) and a 10 '30 decrease in childhood leukemia mortality reported in another ( I 8). These data argue against substantial confounding by some unmeasured correlate of urban residence. Finally, t h e inherent limitations in t h e marker of traffic density should be noted. Whether this is a surrogate for ambient benzene or some other component or a correlate of traffic exhaust, it contains substantial misclassification when compared with some more direct determinant of childhood cancer. Since only the street address was coded, nearby high-traffic streets were not included. T h e resulting misclassification is almost certain to be nondifferential with respect to the case-referent status and would thus bias the odds ratios towards the null (19). Because of these limitations a n d the novelty of this observation, the study results must be interpreted very cautioujly. Specifically, these data d o not strongly impliixtf- tI-nffi(.-r-elatedair pollution in general (or benzene in particular) in the etiology of childhood cancer. Nonetheless, the data d o suggest that an association is pre\ent between high traffic density a n d childhood cancer. There is no obvious methodological flaw ~ , h i c hindicates that these results a r e spurious. Given the limited knowledge of risk factors for childhood cancel-, further research is warranted towards identifying an environmental factor for which traffic densit!. niighr serve a s a marker. I f the observed results a1-eactually reflective of a n underlying causative relation. then m o r e precise measures of hypothesized etiologic agenis (eg, benzene) should produce much sii'onyer associations. Acknowledgments This stud\ Mas pait of a research program to deteim i n c rile possible adverse health effects resulting from exposure to the electric a n d magnetic fields of over- head high-voltas? trailmission lines. The program was administered b! the State o f New York Department of Health and Health Research, Inc. The author5 thank Ms EM John and Dr RC Kleck- ner foi theii- programming assistance a n d Dr T M o r - ris foi- his ci-itical review of the manuscript. References I . MacMahon 8. Hutchison GU. Prenatal x-ray and childhood cancer: a review. Acta Unio I n t Contra Cancrum 1964;20:1172-4. 2. Neutel CI, Buck C. Effect o f smoking during pregnancy on the risk of cancer in children. J Natl Cancer lnst 197I ;4?:59-63. 3 . Infante PF, Epstein SS, Newton W A J r . Blood dyscrasias and childhood tumors and exposure to chlordane and heptachlor. Scand J Work Environ Health 1978; 4: 137-SO. 4. Greenberg BS, Schuster JL Jr. Epidemiology of cancer in children. Epidemiol Rev l985;?:22-48. 5. Rinsky RA, Smith AB, Hornung R, et al. Benzene and leukemia: an epidemiologic risk assessment. N Engl J Med 198?;316:1044--50. 6. Checkoway H , Wilcosky T, Wolf P, Tyroler B. An evaluation of the associations of leukemia and rubber industry solvent exposures. Am J Ind Med 1984;s: 239-49. 7. Thomas TL, Waxweiler RJ. Brain tumors and occupational risk factors: a review. Scand J Work Environ Health 1986;12:1-lS . 8. Stenberg U , Alsberg T , Westerholm R. Emission of carcinogenic components with automobile exhausts. Enviran Health Perspect 1983;47:53-63. 9. Holmberg B, Lundberg P. Benzene: standards, occurrence, and exposure. Am J Ind Med 1985;7:375-83. 0. Savitz D, Wachtel H , Barnes FA, John EM, Tvrdik JG. Case-control study of childhood cancer and exposure to 60-Hertz magnetic fields. Am J Epidemiol 1988;28: 21-38. I. Savitz DA, Zuckerman DL. Childhood cancer in the Denver merropolitan area, 1976-1983. Cancer 1987; 59:1539-42. 2 . Brookmeyer R, Liang K Y , Linet M . Matched casecontrol designs and overmatched analyses. Am J Epidemiol 1986;124:693-701. 3 . Miettinen OS. Estiinability and estimation in casereferent \rudies. Am J Epiderniol 1976: 103:226-35. 4. Mantel, Haenszel W. Statistical aspects of the analysis of data from retrospective studies of disease. J Natl Cancer lnst 1959;22:719-48. 5 . Kleinbaum DG, Kupper LL, Morgenstern B. Epidemiologic research: principles and quantitative methods. Belniont, CA: Lifetime Learning Publications, 1982. 6. Greenland S, Criqui MH. Are case-control studies more vulnerable to responsc bias? Am J Epidemiol 1981; I 14:175-7. 7. Blair A . Fi-aumeni J F J r , Mason TJ. Geographic patterns of leukemia in the United States. J Chronic Dis 198O;33:251-60, 8. Selvin S. Levin LI, Merrill DW, Winkelstein W J r . Selecred epidemiologic observations of cell-specific leukemia mortality in the United States, 1969-1977. Am I Epidemiol 1983;117:140--52. 9. Copeland K T , Checkoway H , McMichael A J , Holbrook R H . Bias due io rnisclassification i n the estimation of relative risk. Am J Epiderniol 1977;105:488-95. ReceiLed for publication: 22 h4ay 1989 I 363