Tobacco Documents Online

{A'c ~~`and exposuxe to environmental tohacco ~ `~n s1i of lunicancer; = SHA ~~ thods. Data,from 14 studies.'_ _. 983; ;providing information on lung cancer ' `s °f smoke'at work zvere' examined Six ~ ` l~- quality cntena were developed for 996. de#ermming usable data. Ameta-analy-, - `sis_was' performed io cibtain a ooi'n-, ecfives:.T 'lus studq.ivas:under-~--'- a_,.;_., t .~s,... ken tq ietermtne whether exposure ~S to environinental tobaceo ~t wor „ ,.. ,,..•~ -~-snioke is`°associated_with an increased_ ~. ' ~p ~' m data that mefahe. !n.. edslr; foi those ialityiestriit•hons ~ ~ ~' ~ ~fs ` Fivestudies met the c '. _ , Their ciirnbined `relative risk was 1'`39 ,(95%o confidence uYtzrval [CV 1,1Y, 1.68) based on 835 lung , ~.._ - caficer-cases. In various meta-analvses ° prepared by :tobacco industry employ. , i ees or consultants,no zncrease,il,i,ris _ _- was found :-The main reason;fof=thi~ , fference is that the earlier analysts failed to-find errors in 2 underlying studies that resulted in overweighting of the odds_ ratios from those studies, both of which were less than unity. - ;-Conclusions: When annrobriate ~~nizauce is taken`of the quality of ` data inputs, the increase in lung cancer om wo lace~exposure to_ envi , ..;_ ,:,..,~,;.~ ,.. , : y . ; _ "ro.itmental_tobacco;smoke is abopY the . _ : saflie as_ thatfcom.household exiiosure:-. ~ y {11in, JPubliHealth. 1998;88:1t)25-- - c 1029) July 1998, Vol. 88, No. 7 Lung Cancer From Passive Smoking at Work A. Judson Wells, PhD Introduction The US Environmental Protection Agency (EPA)' has concluded that passive smoking is causally associated with lung cancer in adults and that environmental tobacco smoke is a group A (known human) carcinogen. This means that exposure to environmental tobacco smoke should cause lung cancer regardless of locale. However, 5 meta-analyses2-6 have become available recently showing no association between lung cancer and workplace exposure to envi- ronmental tobacco smoke. One of these meta-analyses was published in the Journal.6 The combined relative risks (RRs) ranged from 0.98 to 1.04, with an average of 1.01 (95% confidence interval [CI]=0.91, 1.11). All 5 of the meta-analyses were conducted by employees of, or consultants to, the tobacco industry and were included among their comments on the proposed rule by the Occupational Safety and Health Administra- tion (OSHA) to control environmental tobacco smoke in US workplaces.7 The pur- pose of this paper is to present an alternative meta-analysis of the workplace data on lung cancer and passive smoking that places more emphasis on the quality of the underlying studies and avoids several errors in the meta- analyses just mentioned. Methods Fourteen studies (which, together with their various updates, constitute 18 reference citations"~) contain potentially useful data on lung cancer and exposure to environmen- tal tobacco smoke at work. Criteria for deter- mining which studies to include in the meta- analysis were developed. The summary statistic for each study was the odds ratio (OR) (for case-control studies) or the rela- tive risk (for cohort studies), and both were assumed to be log-normally distributed. The meta-analysis was done in the usual way, by calculating a mean of the log odds ratios or relative risks of the various studies weighted (w) by the inverses of their respective vari- ances. The odds ratios were assumed to be reasonable approximations to the relative risks. The variances for the logarithms of the individual odds ratios and relative risks were determined from the 95% confidence inter- vals in the various studies via formula 16-8 in Rothman26: Var[ln(RR)] = {[ln(RlkpPe) - ln(RR1aw,er)]/(2 X 1.96)}2. When no confi- dence interval was given, it was necessary to calculate a weight from a confidence interval developed directly from the cell count using the procedure of Rothman and Boice.27 The 95% confidence interval for the combined relative risk (RR) was determined from the weighted point estimate and the total weight (W=2w) via the transpose of formula 16-8: RR~_uppe' = exp[ln(RR.) + 1.96/ W1 12] and RRc-iow,cr = exp[ln(RR) -1.96/ n]. The exposure estimates and exposure surrogates for many of the workplace studies of lung cancer and environmental tobacco smoke exposure are not of as good quality as those for spouse and household estimates. For example, in some of the studies there was no past history of workplace exposure; only current exposure status was provided. Current exposure might be meaningful in terms of spouse exposure, because marriages normally last a long time; turnover in jobs is high, however, so a subject might recently have moved from a work environment with high levels of environmental tobacco smoke to an environment with no exposure but still be classified as nonexposed. Also, some of the studies, because of the scarcity of never- Requests for reprints should be sent to A. Judson Wells, PhD, 5 Ingleton Circle, Kennett Square, PA 19348. This paper was accepted December 5, 1997. Editors Note. See related article by Breslow (p 1011) in this issue. American Journal of Public Health 1025

Welts smoking lung cancer patients, were forced to depend to a considerable extent on surrogate rather than direct responses. Surrogates, especially if they are family members, might be able to remember a good deal about a deceased subject's home exposure, but it would be very difficult for surrogates to remember workplace environmental tobacco smoke exposure over, say, a 40-year period. Also, the lung cancer effects of short or light exposure to environmental tobacco smoke at the workplace might be lost in the back- ground levels from other environmental tobacco smoke exposures, or the effects of even longer-temi or medium exposure might be lost if there is substantial non-environ- mental tobacco smoke exposure with strong lung cancer potential. Inclusion of ex-smok- ers in the cohort also would tend to dilute the passive smoking effect. For some studies, it is necessary to calculate weights from confi- dence intervals, an uncertain procedure unless there are cell counts to back up the confidence interval calculation. In the light of these considerations, the following 6 criteria were developed for inclusion of the workplace odds ratio or rela- tive risk from a particular study: (1) expo- sure history beyond current exposure only (with a minimum of 10 years preferred), (2) no more than 50% surrogate responses for cases, (3) exposure beyond "little" or "minimal," (4) no large non-environmental tobacco smoke exposure that might mask the weaker environmental tobacco smoke effect, (5) cohort limited to subjects reporting that they never smoked, and (6) availability of cell counts to, allow a reasonable check on the confidence intervals of any crude or adjusted odds ratios or relative risks. Results Of the 14 studies8-1a'i6-i9a2,z4,z5 with the most recent workplace data on lung cancer and environmental tobacco smoke exposure, only 510,13•il=a survived the 6 criteria used for inclusion. The first Kabat study8 involved data for current exposure only. The Garfmkel et al. study9 was excluded because of 88% surrogate case responses. Lee et al." was excluded because 9 of the 10 cases involved "little" exposure. Koo et al.1` was excluded because their history of workplace exposure averaged only 2.0 years. Butler's14 study was rejected because ex-smokers were included in the cohort. Wu-Williams et al."R was excluded because the authors believed that the effects of environmental tobacco smoke might be obscured by heavy exposure to coal-heating fumes in the northeast China locale. Brownson et al.19 was excluded because of 65% surrogate case responses. Schwartz et al.'S was excluded because of 83% surrogate case responses. Janerich et al.16 was excluded because there was no workplace cell count. The remaining 5 studies10•13•17=4 were accepted for the meta-analysis (see Table 1). Workplace exposure histories reflected life- time exposure except in the second Kabat study,Z4 in which the exposure history involved up to 4 jobs lasting a year or more (mean workplace exposures: 28.5 years for male cases and 22.1 years for female cases). The odds ratio and confidence interval from the Reynolds et al. version~2 of the study by Fontham et al.Zl were used because Reynolds et al. excluded nonemployed women from the reference category. The per- centages of surro~ate responses were 36% for Fontham et al. I and 0% for the other 4 studies. None of the 5 studies included ex- smokers. Workplace cell counts were included in 4 of the studies13,i7'na4 and were available for Wu et al." (A. H. Wu, written communication, June 23, 1986). The com- bined relative risk for the 5 studies was 1.39 (95% CI =1.15, 1.68). The relative risk for the 3 US studies was 1.43 (95% CI =1.15, 1.78). The mean weight for the 5 studies was 12.5 per 100 cases. It is evident that the meta-analysis was dominated by the large Fontham/Reynolds study,2°"`2 which constituted 63% of the cases and 56% of the statistical weight. This study was designed, from its initiation, to measure lung cancer effects from environ- mental tobacco smoke exposure at home, at the workplace, and in social situations. It was carried out in 5 US cities: Atlanta, Ga; New Orleans, La; Houston, Tex; Los Ange- les, Calif; and San Francisco, Calif. Lifetime smoking status was determined in a 3-tiered approach that weeded out essentially all ever smokers. A final urinary cotinine test was used to eliminate the remaining 0.6% of cases categorized as current smokers. Envi- ronmental tobacco smoke exposure history was obtained via in-person interviews. Years of exposure in occupational settings repre- sented the sum of years of employment in each job in which persons were reported to have smoked in proximity to the study sub- ject. The odds ratio was adjusted for age, race, study area, diet, family history of lung cancer, and employment in high-risk occupa- tions. Odds ratios were determined for 3 lev- els of workplace exposure. The point esti- mates for the 3 levels were as follows: 1-15 years, 1.46 (95% CI = 1.10, 1.94); 16-30 years, 1.55 (95% CI=1.13, 2.16); and more than 30 years, 2.08 (95% CI=1.35, 3.20) (P for trend: <.001). It is fortunate that the largest study of workplace environmental tobacco smoke exposure and lung cancer included here was of such high quality. The 5-study combined workplace rela- tive risk of 1.39 (95% CI=1.15, 1.68) com- pares with the combined spouse exposure relative risk for the same 5 studies (1.30, 95% CI =1.09, 1.55). Excluding Reynolds et al.,22 the other 4 studies had combined work- place relative risks of 1.21 (95% CI=0.91, 1.62) for men and women combined and 1.25 (95% CI=0.91, 1.72) for women only. Neither was statistically significant at the 95% level, but both were compatible with the combined relative risk of 1.19 (90% CI = 1.04, 1.35) for female never smokers shown in Table 5-17 of the EPA report.1 That value was based on relative risks from I 1 US household studies, 10 of which were of spousal exposure to environmental tobacco smoke. Discussion One might ask why the result indicated in Table 1 is so different from the results found in the earlier meta-analyses ?-4 The main reason is that there were errors in 3 of the underlying studies-Garfinkel et a1.,9 TABLE 1 -Meta-Analysis of Lung Cancer and Passive Smoking at Work No. of Odds Ratio or Relative Risk Study and Locale Sex Cases (95% Confidence Interval) Weight Wu et al.10: California Female 29 1.3 (0.5, 3.3) 4.3 Shimizu et al.13• Japan Female 90 1.2 (0.70, 2.04) 13.4 Kalandidi et al.": Greece Female 89 1.39 (0.76, 2.54) 10.5 Reynolds et al.22: United States Female 528 1.56 (1.21, 2.02) 58.5 Kabat et al 24: United States Male 41 1.02 (0.50, 2.09) 7.5 Female 58 1.15 (0.62, 2.13) 10.1 Total: all locations Both 835 1.39 (1.15, 1.68) 104.3 Total: United States only Both 656 1.43 (1.15, 1.78) 80.4 Note. The confidence interval for Shimizu et a113 was calculated from cell counts; the odds ratio and confidence interval for Kalandidi et al.'7 w ere calculated from cell counts for "little" plus "some" vs no exposure. J July 1998, Vol. 88, No. 7 1026 American Journal of Public Health

Tobacco Smoke, Work, and Lung Cancer ory ~M 're- : in ito ab- ge, tng pa 3fi- -15 •30 ore (P :he .tal -er Ia- m- .ire 30, ;et rk- )1, nd dy. '-he ith )% ers hat us of ;co :ed 11ts 'he of 1.,9 'a (D N N U Q~ OD(DQ') ' c7 q Nh~ ' tnt1JOLqNr LO tolnCOQ) J O ~ NGDQ} '- 'ct rNd C'7 r NrN '- ~ 0~ 10~ I~O - N~ O E 0 N O O O OILl7 QC7tnd' tAd'~r N ~~C9 C ~T V' O M t O a7 ' pettn F ~n(000 ~j.r h q N 0 ~ ~~^ rr ~,..~• (pNNN C.~'6 rNrr N NNNr 3 t~6 i pa - OM~ V: ° M.-r-d0'~ M~(D~W r•T t~CN0~0 C N U ~,n voo o000 °OOOOO oo~v y a~ 3 N(.50m .•. I~OOM NC m ~W O N O o Q~p N t0 tn 0>fqNQ NM O Or 0 ~m ~ CD Om vio0 r [Dr C rorr r r q o~ro U~CO o~ 0 ~ U C) ,~ °~ p O tnc+~t0 NtDf~ Orn N OLnpp O rrODO) ~•mrrn~no N O)Nr T N mnm V' U) N V ~ w •00 c=6~ Z U ~ L r N ~~ N ~ ~ a ~~ A =' O ) nU N M 0 d'~ hV~' Ch U ~ ~ _ N OC O O L ~ O ~ CV 1 00 r T N r O CO N CO CO I~ CO f- CV ~~ C_ s - c 6 r N tN 0 = v o~~ o~ o~o ~1 .C U c ~ r r N C6 T r N N 3. 3 Ca N ~^ ~U O M ~ N OOn N ~~ • :~r- : VO' ~•p C ~ ~ U ~ a ~°LO vOO o0 000 00 'o E ~;~ Cs U rn ~v ~p NCOO) NO ~p O) OM CO~ mC OOi omcc ~ tn y CM O O r N j r O O r O r O N ~ O ° % C ~ -.- `d ° °° O N N~T C~O ^O W d0 9'V' eY O (D 0 ZU r d CO T ~ 6 U ~~~ .n ~ ~ FD ~ WtO~t N M ~ O)Nnd; N' ' tDilIt o ~ iC> r,~ c0 , o > CO LO d r V T N d CO ~ I ~ Y~ •~- ~ ~ ' o'°a i L, 0 CO c S O ~~ O N 0 ' C I O M pM (OMLO O OOrO r (0 0 r ~ d(0 y rrr CO (OCVCVN r4 +- fVCV U O~ O ~ ~ 0) ~U omr ~ : mr~° vv ~rn 3 ~ 3 ~ =-J ~~ v00 O 0000 ' '00s 00 'O ~L Q ° U O ~v f~pp M NCDO ~ ' ~Q>O fOCO N r0~ O M Or 000 N O O ~ y~L ~ eioo C ? r r rorr ~ or*- .= d or r CO q~ ° c~ g °vas y ° ~ "u) c`~V n c\i °r W gi rn`c0~ ~ rri~l' n E$ c a i Z U Z0 0 a N ~ ~ Z O CO CO(O V) d7N1~tf) :f, tntfld. • r h ~ W .~ lV CO(6 4 rNC7C`! ONO~ ~ COi~ ~ ~ m ' > r C O -0 ~ c O $ U >, `D~r° O _ c~c~no go~^ 0 «s ~s ~ J ~^ rrr r CDNChN r+-N~ ,- NN r ~Z DN C ~U O '' ~ ~ ~- CO M r000~~ •O • ~ ~~ •~ Q[ U."'' :C ~ O °~ [ J r00 • S ' r tn D 0000 ? OQOOa 00 O . ~~ ~~ E ~ U~ NCOCO r- COCO ['~ N ~0~ C07.r [? OS O U•OC3 N G) O -J C00 O r 0 r r r O r r r O r N CO O a o~ °~ ' 0 CL ~ u i CO ~n cD N~I~ p~ N o tnro rrMC~ i.nM rrn ~n O O)Wr N ~ ~• Ln CO~I- M ~ x a zU r d v r ~~ a i ~ a) _r 3 N i O (D O O X ~ J~ c O O ~ O ' ~ ~ ,9 ~ m CO ~ N m fA RS ~ C 0 •cE .~a?E Qf E cn O O m ~@ a ~ M m ~ E 8°~ ' ~ ~ o ~ m.~ p ~ ~ C ~ V - N " ~ ~~ - • r ~(] T _J O • a) Fu 2 ~CII O . ' l'J N X > ~ No -O i 7 aj' @ O O m O E O O ~ L ~ O • 0 [C Y C, OW N S O ~ c ~ O U _7 j, a c O c3 Ec_ C cC a>~ N ~ 1=Z •- C, p V~- Ecc ~ > ~ ~ O ~ t tl O 3F' ;v mm • m c ~~ mMti~ ° . a o c ~i c~ ~~s a? o Z ~ la- U Y C7~i J YU)m ~Y1i~ Y U) Z Janerich et al.,16 and Wu-Williains et al.'8- that the earlier analysts failed to detect. To illustrate, I have chosen 3 of the earlier meta-analyses-Le Vois and Layard,3 Lee,4 and Chappell and Gratt6-as examples, although the other 2 analyses2•5 suffered from the same general problems. The details of the 3 meta-analyses are provided in Table 2, along with a column showing what the corrected combined results would have been if the 3 underlying errors had been detected and if, in a few other instances, a preferred underlying relative risk and confidence interval had been used. Le Vois and Layard3 chose to use 12 of the 14 studies worldwide, Lee4 chose 11, and Chappell and Grattb chose 8 of the US studies. The Schwartz et al. study,'5 shown in the fourth column of Table 2, appeared too late to be used in any of the earlier analyses. Also, the values for the second Kabat study24 in Table I and column 4 of Table 2 were derived from the final paper on that study, while other values shown in Table 2 were derived from an early version.23 The underlying error in Garfinkel et al 9 is detailed in Wells and Henley.28 Briefly, the confidence intervals in the paper do not match the cell counts. EPA' used the cell counts, which were con:ect, to calculate the weights, but the meta-analyses referred to eaxlier2-6 used the confidence intervals. The 95% confidence intervals in the paper were calculated with the observed odds ratio as the multiplier of the standard error rather than the correct critical value of 1.96. When the workplace odds ratio of 0.93 was used as the multiplier, the confidence interval calcu- lated in the paper was much too narrow, resulting in a weight approximately 5 times too large. In Wu-Williams et al.,'$ the workplace- adjusted odds ratio of 1.1 was not centered in the 95% confidence interval (0.9, 1.6). A. H. Wu, formerly Wu-Wiliiams (written com- munication, November 25, 1996) states that the confidence interval was correct, not the odds ratio. The correct odds ratio is 1.20 (95% CI = 0.93, 1.57). The third error in an underlying study-and the most important-involves Janerich et a1.16 It is evident that all 3 of the meta-analyses included in Table 2 were strongly dependent on this study, which had a weight of 223, 48% of the weight in Le Vois and Layard3 (but with only 12% of the cases), 45% of the weight in Lee4 (but with only 11% of the cases), and 43% of the weight in Chappell and Gratt (but with only 11% of the cases).6 Based on the mean weight-to-case ratio of 12.5 per 100 from Table 1, one would expect the weight for the 191 cases in Janerich et al.16 to be about 24. . 7 1 July 1998, Vol. 88, No. 7 American Journal of Public Health 1027

The only cell count in Janerich et ai. was that for total household exposure. An exposed/not exposed array developed from this cell count yielded a crude odds ratio of 0.86 (95% Cl = 0.53, 1.40) for the same 191 cases, and a weight of 16. The workplace result in Janerich et al.16 was for never smokers only and was taken from Table 19 of a dissertation by Varela.15 The disserta- tion included a matched-pairs analysis of environmental tobacco smoke exposure for 439 case-control pairs (218 women and 221 men, 197 never smokers and 242 ex-smok- ers) and was based on 33% surrogate inter- views. Matching included age, sex, county, never smoker vs exsmoker (for 10+ years), and surrogate vs direct interview. There were no cell counts in the dissertation. The data in Janerich et al.16 for household and spousal exposure appear to have been reworked, since "the results for these expo- sures were different from those included in the dissertation. The reworked odds ratios in Janerich et al.16 for matched analysis of spousal expo- sure (0.93, 95% CI = 0.55, 1.57, for direct interviews and 0.44, 95% CI = 0.19, 1.02, for surrogate interviews) involved a total weight of 19.4 for 188 pairs. The workplace odds ratio in Janerich et al.16 (also in Varela") was said to be "as a continuous variable for an equivalent differential of 150 person-years of exposure." In addition to matching, it was probably adjusted for religion, income, mari- tal status, and cigarettes smoked per day among former smokers. Unfortunately, such an odds ratio basically represents different units from the•relative risks of the other stud- ies (which were simply relative risks for "ever exposed" vs "never exposed"), and therefore it should not be used in a meta- analysis with them. Confidence intervals from such logistic regressions are very sensi- tive to the exposure differential chosen,26 resulting in calculated weights that also can vary substantially depending on the exposure differential. Varela's Table 17 presents unadjusted workplace odds ratios and 95% confidence intervals for all subjects for each of 8 levels of environmental tobacco smoke exposure. A weighted mean of the log odds ratios yielded a combined odds ratio of 0.91. How- ever, this odds ratio was for a mix of never smokers and ex-smokers. If one assumes that this is also a correct odds ratio for the 191 never-smoking cases in Janerich et al., and if one assumes a 53% workplace exposure level based on a workplace study with a sim- ilar locale,'9 one can construct a 2 X 2 table that results in a 95% confidence interval of 0.61, 1.35 and a weight of 24. A linear regression of the data included in Varela's Table 17, using the method in Rothman,`'6 yielded an odds ratio of 0.77 (95% CI = 0.58, 1.04) for a differential of 150 person-years. This resulted in a weight of 45.1 for 439 cases, which would be equivalent to a weight of 20 for 191 cases). In Varela's Table 18, the unadjusted odds ratio for all 439 subjects, for a differential of 150 person-years of expo- sure, was 0.9941 (95% CI = 0.9740, 1.0145), leading to the enormous weight of 9259. All of this indicates some kind of an error in Varela's methodology for calculating stan- dard errors and confidence intervals in his regression analyses. This error carries over into the workplace result in the Janerich et al. paper16 and casts doubt not only on the con fidence interval but on the odds ratio itself. If the odds ratio of 0.91 is used, which is not recommended, a weight of 16 to 24 should be used, as shown in column 4 of Table 2 (not the 223 that the other analysts used). Le Vois and Layard3 and Lee4 used the same odds ratios and confidence intervals for the various studies except for (1) Fontham et a120,2I (where Le Vois and Layard used the 1991 version20 and Lee used the 1994 ver- sionZl); (2) Kalandidi et a1.17 (where Lee used an odds ratio of 1.70 based on compar- ing "some" exposure and "minimal" expo- sure, whereas Le Vois and Layard preferred to use 1.39 [as I did] based on "some" plus "minimal" exposure vs nonexposed); and (3) Koo et a1.12 where Lee4 included nonsmok- ers exposed at home in his reference cate- gory, whereas Le Vois and Layard3 used only subjects with known exposure. Sears and SteichenZ used US results only and appear to have used the same odds ratios and confi- dence intervals used by Lee 4 Biggerstaff et al 5(their Table 4) used the same odds ratios and confidence intervals that Le Vois and Layard3 used, except for Koo et al.,12 where they based their odds ratio on the 2 cases involving only work exposure. Chappell and Gratt6 followed the other analysts in using the 223 weight for Janerich et al.t6 but went further with Garfinkel et al 9 Table 7 of the Garfinkel et al. paper provides 2 odds ratios for workplace exposure: (1) at any time during the previous 25 years (0.93, 95% CI = 0.73, 1.18) and (2) at any time dur- ing the previous 5 years (0.88, 95% CI = 0.66, 1.18). The other analysts used the odds ratio for the "last 25 years °" Chappell and Grattb also used the 25-year odds ratio in their Air and Waste Management Association presentarion30 but shifted to a "more appro- priate" odds ratio of 0.91 (95% CI = 0.76, 1.09) in their OSHA submission31 and inn their letter,6 apparently by adding to the 25-year odds ratio the data for those subjects exposed at work only during the previous 5 years. This raised the number of cases from 76 to 170 (there were only 134 total cases in the study) and increased the weight from 66.6 to 118.2. A careful reading of the Garfinkel et al. pape? discloses that the "any time in the last 5 years" data are a subset of the "any time in the last 25 years" data and that, there- fore, the 2 sets should not be added. Another large study that Chappell and Gra.ttb used was Brownson et al.,19 a study Lee4 had rejected because the only work- place odds ratio included in the paper, 1.2 (95% C1= 0.9, 1.7), was for the highest exposure level only. Chappell and Gratt used this value in their Air and Waste Manage- ment Association meta-analysis30 but shifted to 0.9 (95% CI = 0.7, 1.15, weight= 62.4) in their OSHA submission31 and their letter6 on the basis of an analysis of unpublished data from Brownson et al. in an OSHA submis- sion by another tobacco consultant, W. J. Butler.32 This value, taken from Butler's Table 2, was for never smokers combined with former smokers. A better choice, if But- ler's results are to be used, would be his never-smoker workplace odds ratio of 0.98 (95% CI = 0.74, 1.31, weight = 47. 1). Chappell and Grattb used odds ratios for Wu et aI.10 for never smokers, ex-smokers, and current smokers combined rather than the more appropriate odds ratio for never smok- ers only (see Table 1 and the Le Vois and Layard and Lee data in Table 2). They also used odds ratios for Butler14 that were incon- sistent in that the male odds ratio was for I to 10 years of exposure, while the female odds ratio was for 11 or more years of exposure. Table 2 (fourth column) shows that even if all 14 of the workplace studies are used, regardless of the criteria listed earlier (but with corrected odds ratios and weights), a combined relative risk of 1.19 (95% CI = 1.07, 1.34) is the result. This value is statistically significant and essentially the same as the EPA result' noted earlier. Conclusion The quality of the data concerning workplace exposure to environmental tobacco smoke varies considerably among the 14 available epidemiologic studies, and 3 of the studies contain important errors. When relative risks from the better studies are com- bined via meta-analysis, a statistically signifi- cant increase in lung cancer risk is found that is similar to the increased risk from house- hold studies. When all of the studies are included, regardless of quality, the increase in risk is still statistically significant. However, when incorrect, high weights are given to rel- ative risks less than unity in some of the stud- ies, no increase in risk is found. 11 1028 American Journal ofPublic Health July 1998, Vol. 88, No. 7

, and ;tudy loric- , 1.2 ;hest used tage- iifted 4) in t' on data )mis- N. J. ler's lined But- - his 0.98 ls for kers, n the mok- ; and also icon- rlto odds 'e. that s are arlier ~hts), 95% ue is y the ning -nta1 nong tnd 3 Nhen com- PfflY ause- s are ise in iever, o rel- stud- , Acknowledgments I wish to acknowledge statistical help from Allen Bostrom, Kenneth Rothman. and Robert Tarone. References 1. Respiratory Health Effects of Passive Smok- ing: Lung Cancer and Other Disorders. Wash- ington, DC: US Environmental Protection Agency; 1992. EPA publication 600/6- 90/006F. 2. Sears SB, Steichen TJ. Comments on Environ- mental Tobacco Smoke Preliminary Quantita- tive Risk Assessment. Washington, DC: Occu- pational Safety and Health Administration; 1994. OSHA docket 9-47533. 3. Le Vois ME, Layard MW. Inconsistency between workplace and spousal studies of environmental tobacco smoke and lung cancer. Regul Toxicol Pharmacol. 1994;19:309 316. 4. Lee PN. Environmental Tobacco Smoke, a Catnntentary on Some of the Evidence Pro- vided by OSHA in Support of Their Proposed Rules. Washington, DC: Occupational Safety and Health Administration; 1994. OSHA docket9-47513. 5. Biggerstaff BJ, Tweedie RL, Mengersen KL. Passive smoking in the workplace: classical and Bayesian meta-analyses. Int Arch Occup Environ Health. 1994;66:269-277. 6. Chappell WR, Gratt LB. A graphical method for pooling epidemiological studies [letter]. Am JPublic Health. 1996;86:748-750. 7. Occupational Safety and Health Administra- tion. Indoor air quality; proposed rule. Federal Register. Apri15, 1994:15968-16039. 8. Kabat GC, Wynder EL. Lung cancer in non- smokers. Cattcer. 1984;53:1214-1221. 9. Garfinkel L, Auerbach 0, Joubert L. Involun- tary smoking and lung cancer: a case-control study. J Natl Cancer Inst. 1985;75:463-469. 10. Wu AH, Henderson BE, Pike MC, Yu MC. Smoking and other risk factors for lung cancer in women. J Natl Cancer Inst. 1985;74: 747-751. July 1998, Vol. 88, No. 7 11. Lee PN, Chamberlain J, Alderson MR. Rela- tionship of passive smoking to risk of lung can- cer and other smoking-associated diseases. BrJ Cancer. 1986;54:97-105. 12. Koo LC, Ho JH-C, Saw D. Is passive smoking an added risk factor for lung cancer in Chinese women? Exp Clin Cancer Res. 1984;3:277-284. 13. Shimizu H, Morishita M, Mizuno K, et al. A case-control study of lung cancer in nonsmok- ing women. Tohoku J Exp Med. 1988;154: 389-397. 14. Butler TL. The Relationship of Passive Smok- ing to Various Health Outcomes among Sev- enth Day Adventists in California [disserta- tion]. Los Angeles, Calif: University of California; 1988:105-112. 15. Varela LR. Assessment of the Association between Passive Smoking and Lung Cancer [dissertation]. New Haven, Conn: Yale Univer- sity; 1987. 16. Janerich DT, Thompson WD, Varela LR, et al. Lung cancer and exposure to tobacco smoke in the household. N Engl J Med. 1990;323: 632-636. 17. Kalandidi A, Katsouyanni K, Voropoulou N, Bastas G, Saracci R, Trichopoulos D. Passive smoking and diet in the etiology of lung cancer among non-smokers. Cancer Causes Control. 1990;1:15 21. 18. Wu-Williams AH, Dai XD, Blot W, et al. Lung cancer among women in northeast China. Br J Cancer. 1990;62:982-987. 19. Brownson RC, Alavanja MCR, Hock ET, Loy TS. Passive smoking and lung cancer in non- smoking women. Am J Public Health. 1992; 82:1525-1530. 20. Fontham ETH, Correa P, Wu-Williams A, et al. Lung cancer in nonsmoking women: a multi- center case-control study. Cancer Epidemiol Biomarkers Prev. 199 1; 1:35-43. 21. Fontham ETH, Correa P, Reynolds P, et al. Environmental tobacco smoke and lung cancer in nonsmoking women: a multicenter study. JAMA. 1994;271:1752-1759. 22. Reynolds P, Fontham ETH, Wu A, Buflier PA, Greenberg RS. Occupational exposure to envi- Tobacco Smoke, Work, and Lung Cancer ronmental tobacco smoke [letter]. JAMA. 1996;275:441-442. 23. Kabat GC. Epidemiologic studies of the rela- tionship between passive smoking and lung cancer. In: 1990 Annual Winter Toxicology Forum. Washington, DC: Toxicology Forum; 1990:187-201. 24. Kabat GC, Stellman SD, Wynder EL. Relation between exposure to environmental tobacco smoke and lung cancer in lifetime nonsmokers. Am JEpidemio1.1995;142:141-148. 25. Schwartz AG, Yang P, Swanson GM. Familial risk of lung cancer among nonsmokers and their relatives. Am J Epidemiol. 1996;144: 554-562. 26. Rothman KJ. Modern Epidemiology. Boston, Mass: Little Brown & Co Inc; 1986:341-346. 27. Rothman KJ, Boice JD Jr. Epidemiologic Analysis with a Programmable Calculator. Boston, Mass: Epidemiology Resources; 1982:8-10. 28. Wells AJ, Henley SJ. Re: Involuntary smoking and lung cancer: a case-control study [letter]. J Natl Cancer Inst. 1997;89:821-822. 29. Muscat JE, Wynder EL. Exposure to environ- mental tobacco smoke and the risk of heart attack. Int J Epidemiol. 1995;24:715-719. 30. Gratt LB, Chappell WR Problems with the use of meta-analysis of epidemiological studies in risk assessment: worker environmental tobacco smoke lung cancer risk as an example. Pre- sented at the Air and Waste Management Asso- ciation 88th Annual Meeting and Exhibition, June 1995, San Antonio, Tex. 31. Chappell WR, Gratt LB. Summation of Com- ments and Recommendations on OSHA's Indoor Air Quality Proposed Regulation. Washington, DC: Occupational Safety and Health Administration; 1995. OSHA docket 496. 32. Butler WJ. Workplace Exposure to ETS and Lung Cancer: A More Detailed Presentation of the Data from a Negative Study, Brownson et al. (1992). Washington, DC: Occupational Safety and Health Administration; 1994. OSHA docket 9-47597a. American Journal of Public Health 1029