Tobacco Documents Online

OSHA POSTHEARING SUBMISSION t ~ Submitted by: Helen B. Hubert, Ph.D. Senior Research Scientist Department of Medicine and Health Research & Policy Stanford University Medical Center Stanford, California ~ August 29, 1995 ~ ~ I I I ~ ~ ~ I ~ 2~9 I p.i'

OSHA POSTHEARING SUBMISSION Helen B. Hubert. Ph.D. Senior Research Scientist Departments of Medicine and Health Research & Policy Stanford Universiry Medical Center STanford, California I have obtained and reviewed a series of five deliverables prepared by Kenneth G. Brown. Ph.D.. Incorporated as a subcontractor to Meridian Research Inc. (Task Order No. 3, contract No. J-9-F- 1-0065), entitled "OSHA's Critical Evaluation of Epidemiological Studies on Cardiovascular Disease Risk in Nonsmokers Exposed to Passive Tobacco Smoke." These deliverables were submitted by Dr. E.own over a period of I yeir (November 1993- July 1994) to Ms. Debra Janes at OSHA and include the following: First Deliverable: (1) Updated risk analysis on passive tobacco smoke and lung cancer in the form of revised material from the EPA document, including a revised version of Chapter 5, additions and revisions to Appendix A, and additions to the bibliography. (2) Review and analysis of epidemiologic data on the association between exposure to passive tobacco smoke in the workplace and the risk of lung cancer. Second Deliverable: A report on eptdamiologic studies on passive tobacco smoke and heart disease prepared by Dr. Brown as a subcontractor to Meridian Research Inc. Third Deliverable: Study results on factors that may affect risk of heart disease. Fourth Deliverable: Tier classification scheme for epidemiologic studies on heart disease and passive tobacco smoke. Fifth Deliverable: Tables and figures on heart disease. Although I recognize these deliverables to be drafts of Dr. Brown's analyses on OSHA's behalf, I have some major concerns regarding Dr. Brown's assessments of the epidemioiogic data on exposure to "pagsive" tobacco smoke (PTS, also referred to in this analysis as environmental tobacco smoke, ETS) and lung cancer and cardiovascular disease risks. In particular, I am concerned about errors and discrepancies in the report, inappropriate conclusions drawn from the ~ data, and improper use of statistics. ~ Cdl ~ 1 (~ ~ GO Io

2057837307 ww m

I I I I I I exposures to other occupational factors that may be associated with lung cancer or heart disease risks. Furthetmore. because of widely recognized variability in dietary, lifestyle, socioeconomic conditions, genetics, and other potentially disease-related factors, the relative'irtfluence of confounders may vary in different countries or regions of the world. Another weakness of the report on workplace exposure to environmental tobacco smoke and lung cancer is that none of the published studies adjusted for spousal or other sources of environmental tobacco smoke to derive the independent contribution of workplace environmental tobacco smoke on risk. On page 1 I of the first deliverable, Dr. Brown acknowledges that "the small sample sizes, the low power to detect an effect, the difficulty of assessing exposure in the workplace, the mix of surrogates used for workplace exposure, and the probable correlation of exposure to PTS at (work) with exposure in other environments, leave the evidence only suggestive." The data are obviously of poor quality and Dr. Brown equivocates in his conclusions on page I 1 by stating, "If PTS is a lung carcinogen, as concluded by EPA, then exposure to PTS in the workplace adds to the toxic burden from exposure in other environments to increase the risk of lung cancer." Such a statement is not, in effect, based on any hard, factual data. In comparison to Dr. Brown's assessment of the workplace lung cancer data, LeVois and Layard (1994) performed a meta-analysis of the results of 12 epidemiologio studies that reported risks of lung cancer and workplace exposure to environmental tobacco smoke. Sixteen relative risks were provided in the papers identified; 9 of these 16 were above 1.0 and 7 were less than or equal to 1.0. LeVois and Layard stated that: "Of the 16 reported relative risks, we combined 15 by computing a weighted average of their logarithms, me weights being the inverses of the variances of the log relative risks (the Butler relative risk of 0.0 for males could not be included in the meta-analysis). The summary relative risk for 12 of the 14 worldwide studies was 1.01, with 95% confidence interval (0.92, 1.11). The summary relative risk for 7 of the 9 U.S. studies was 0.98 with 95% confidence interval (0.89, 1.09). Although we could not include the Brownson et al. (1992) and Stockwell et al. (1992) studigs in the meta-analysis, since they did not report relative risk estimates for workplace exposure, the authors' comments ... indicate that including them would not change the conclusion that there is no epidemiologic evidence of an association between workplace ETS exposure and lung cancer." (p. 312) Whereas one could argue that any approach to combining data from these studies may be flawed, use of LeVois and Layard's technique does not present any evidence for an effect of workplace exposure to ETS on lung cancer risk. _ Inappropriate Conclusions Drawn from the Dose-Response Data The issue of an alleged dose-response demonstrated by studies of spousal or workplace exposure ~ to environmental tobacco smoke and lung cancer or cardiovascular disease is one that has ~

I I I I I I I I I I I I t I I I bias and confounding inherent in that design; that is. such a[meta-analysis] will likely be significant [at the conventional 5% level, irrespective of whether a one-sided or two-sided test is used] simply due to artifact" (p. 3 12). Statistical Power of Studies Most of the studies of lung cancer and cardiovascular disease examined by Dr. Brown have insufficient power to detect relative risks of 1.5 or less as shown in Tables 5-9 (first deliverable on spousal exposure to environmental tobacco smoke and lung cancer risk) and Table 6 (second deliverable on spousal exposure to environmental tobacco smoke and cardiovascular disease risk). Rigorous epidemiologic studies should be designed with sample sizes sufficient to achieve 80-90% power to detect the. risk in question, given the expected disease rate or exposure in the population (for cohort or case-control studies) (Breslow and Day 1987). Since most of the data c~ted by Dr. Brov.n were derived from s:udies that were not specifically designed to examine the role of environmental tobacco smoke on these particular health endpoints, power has been seriously diminished. The major drawback is that without sufficient power and with such low risks described, chance cannot be ruled out as an explanation for many of the elevated risks found. Furthermore, power may actually be more seriously compromised than is reported in the tables if, in fact, risks for both lung cancer and cardiovascular disease are lower than 1.5 as the majority of the data suggest. In Table 5-9 of the first deliverable on spousal exposure to environmental tobacco smoke and lung cancer, only 3 out of 12 U.S. studies of lung cancer (Brownson et al. 1992, Fontham et al. 1991, and Garflnkel et al. 1985) for which power could be calculated show sufficient (>80%) power to detect a risk of 1.5. These are considered by Brown to be Tier 1 or Tier 2 studies. Only one (Fontham et al. 1991) shows significantly elevated risk (OR=1.28), but only using a 90% (instead of 95%) confidence interval (1.0.3, 1.50). Brownson et al. 1992 shows an OR=0.91 (CI= 0.75, 1.05) and Garfinkel et al. 1985 shows an RR=1.16 (CI=0.89, 1.52). Of the remaining non- U.S. studies, only Wu-Williams and Samet (1990) has sufficient power, reporting an OR=0.78 (CI=0.63, 0.96). Greater power is also indicative of greater stability of estimates and reliability, if the study methods and analyses are appropriate and have ruled out bias and confounding. With regard to power, on page 5 of the second deliverable on spousal exposure to environmental tobacco smoke and cardiovascular disease risk, Dr. Brown states "If power is small, the lack of significance may be due to small sample size instead of due to ,io effect; if the power is large, so that failure to detect a real effect is unlikely, then lack of significance is more informative." However, Dr. Brown often fails to acknowledge the importance of nonsignificant findings in instances where power is large. " 12

3. Important biases addressed' (al Selection bias 1 1 r • or YES Population-based (geographically defined) study with minimal nonresponse, and/or drop-out rate. O or NO Evidence of differential response or drop-out rate in cases versus controls/exposed versus unexposed or inadequate control group selected. :1 or BLANK Insufficient information to determine bias. (b) Information bias • or YES No evidence of bias when validation of smoking and ETS exposure undertaken in cases and controls, blinding of interviewers to case/control status, and proxy response rate low and nondifferential in cases and controls. O or NO Proxy response greater in cases than controls or no validation of smoking and ETS exposure or unblinded interviewers. a or BLANK Insufficient information to determine bias. 4. Are the data internally consistent? . • or YES Evidence of crude or significant dose-response, subgroup results that go in the same direction (e.g., males v. females, blacks v. whites, home v. work), and exposure preceding outcome. . _ O or NO No evidence of dose-response, subgroup results that go in a different direction, or unclear temporal relationship. O or BLANK Insufficient_data to determine dos.e-response, subgroup results, or temporality. The relative quantities of open circles and blank spaces in Table I suggest that the studies on cardiovascular disease are not sufficiently reliable to make conclusions with regard to the impact of environmental tobacco smoke exposure on cardiovascular disease risk. Of greatest significance is the fact that studies that ruled out chance as an explanation for elevated risks were, without exception, unable to rule out systematic study bias or confounding as explanations for the fmdings. My evaluation as to whether the data collected in each study are reliable to assess the risk of cardiovascular disease associated with exposure to environmental tobacco smoke exposure is described below.

Additionally, as an expert in the epidemiology of cardiovascular disease, I have performed my own assessment of all of the available epidemiologic studies on environmental tobacco smoke exposure and cardiovascular disease, and it is my belief that these studies do not support a conclusion that the reported associations are real. Specifically, as my analysis shows, the 13 published studies, 2 abstracts, and 1 letter that were available to Dr. Brown at the time of his report are not reliable because they do not adequately exclude important sources of bias and confounding nor rule out chance as an explanation for the fmdings. Whereas the copy of the report I reviewed did not contain Dr. Brown's conclusions on the potential effect of environmental tobacco smoke exposure on cardiovascular disease risk (my copy of the deliverable ends on page 7 and the remainder of the document was not available), Dr. Brown's own critique of the epidemiologic studies similarly points out that the data are inconclusive to establish an effect and rule out bias and confounding. Based upon the rndings of my reassessment )f the cardiovascular disease studies and a recognition that the methodological criticisms apply as well to $TS studies involving other disease endpoints, I am confident that a similar unbiased reassessment of all the available lung cancer epidemiology studies would raise similar questions about Dr. Brown's conclusions. My opinions of Dr. Brown's assessments of lung cancer and cardiovascular disease risks, and the results of my independent review of the cardiovascular data form the basis for this posthearing submission to OSHA. KENNETH BROWN'S APPROACH TO ANALYZING EPIDEMIOLOGIC DATA The series of deliverables details Dr. Brown's assessment of the potential association between spousal and workplace exposure to passive tobacco smoke and lung cancer and cardiovascular disease risks. In general, I have several concerns regarding his approach including reliance on studies of limited epidemiologic quality, inappropriate conclusions drawn from his analysis of the available epidemiologic data and of dose-response, limited evaluation of temporality, limited evaluation of confounding and misclassification, limited evaluation of study heterogeneity, improper use. of statistics, and use of subjective tier classifications. In addition, I was able to identify several errors, discrepancies, and misinterpretations in the deliverables that could compromise the validity of the reports' conclusions. Reliance on Studies of Limited Epidemiologic Quality Dr. Brown readily admits to the limited quality of the epidemiologic studies that evaluated spousal and workplace exposure to environmental tobacco smoke and lung cancer and heart disease risks. His comments can be found in the Appendices accompanying each deliverable. For example, among the lung cancer studies reviewed by Dr. Brown to assess workplace. exposure (first deliverable), he comments that the Akiba study "precludes drawing any firm conclusion regarding workplace PTS and lung cancer" (App-1); the Buffler study has "little

t ' l I I I In Table 2 of the first deliverable, relative risks presented are not comparable since they do not incorporate uniform definitions of exposure or study population. For example, in the analysis of workplace exposure and lung cancer, the crude RRs for the Kabat and Wynder and Lee studies are calculated for both genders combined rather than separately for males and females, whereas odds ratios, by gender, are in fact presented in the table footnotes and/or in the appendices. In the Kabat study, these odds ratios by gender show a statistically significant increased risk for neversmoking males exposed to tobacco smoke at work (2.57) but not for females (0.68). In the Lee study, the crude ORs for nonsmoking males and females separately are 1.6 and 0.63, respectively, neither statistically significant. As another example, in the Akiba study, exposure is defined as outside the home vs. housewife, whereas other studies define exposure differently. On p. 4 of the analysis of workplace exposure and lung cancer, Dr. Brown states that "Most. (9 of 12) [crude] values are above 1.0, two are at or slightly below 1.0, and only a single value falls well below the null." I believe that this overstates the strength of the data. According to his Table 2, crude RRs are reported for 13 studies, 4 of the 13 values fail below 1.0, including CHAN 0.77, GARF 0.93, LEE 0.99, and BUTL 0.6. Six are below 1.5, including AKIB 1.08, FONT 1.12, KABA 1.11, KOO 1.36, SVEN 1.26, and WU WI 1.22. Only LAMW is statistically significant. 4. Table 3 of the analysis of workplace exposure and lung cancer, the adjusted risk reported for BUTL in Table 2, 1.06 does not match that reported for BUTL in Table 3, 1.08. 5. On p. 6 of the analysis of workplace exposure and lung cancer, second paragraph, Brown does not report whether these odds ratios are crude or adjusted. This is an important omission since Dr. Brown notes on p. 5-23 of the first deliverable, a revision of Chapter 5, that'`an adjusted RR is considered preferable to a crude RR unless the study review .... indicates a problem with the adjustment procedure.... our choice of RR is the smaller of the crude and adjusted values in ... studies providing both estimates." 6. On p. 8 of the analysis of workplace exposure and lung cancer, fourth line, Dr. Brown omits BROW from the list of elevated ORs. Again, Dr. Brown does not specify in the text which of these ORs are crude and which are adjusted. 7. Despite the observation that 90% confidence limits were calculated for the analysis of lung cancer and passive tobacco smoke exposure, on p. 6 of the second deliverable, passive tobacco smoke and heart disease, and in the accompanying figures, 95% confidence limits were reported instead. 8. On p. 6 of the second deliverable, passive tobacco smoke and heart disease, despite Dr. Brown's observation of the contrast of estimates of RR for males and females exposed to PTS in the home, he suggests no explanation for this observation. 14

employment status was used as a sutroeate for workplace exposure) are consistent with the hypothesis that workplace exposure to environmental tobacco smoke increases lung cancer risk. On page 6, of the first deliverable on workplace exposure and lung cancer risk. Dr. Brown inappropriately deletes a quite accurate statement, that "workplace exposure to other carcinogenic substances could arguably give rise to the same observation." Employment status is not a convincing measure of environmental tobacco smoke exposure since other risk factors may also be associated with working outside the home. 1 I I I Inappropriate Conclusiops I)rawn Erom the Analyses I have identified several examples of instances in which Dr. Brown's conclusions are not reflective of the data he presents. For example, in his analysis of the spousal lung cancer risk by tier and country, he states on page 5-71, "It is concluded that the association of ETS and lung cancer observed rrom the analysis of 33 epidemiologic studies in eight different countries is not due to chance alone and is not attributable to bias or confounding." However, even if one is prepared to accept the tier ranking approach and discount the Chinese studies, Dr. Brown's statements are inappropriate since he fails to acknowledge that pooling of first and second best ranked studies produces no statistically significant effect of environmental tobacco smoke on lung cancer in the United States or Europe. Only in Greece, Hong Kong, and Japan are results significant and consistent over tier pooling. The fact that these are not U.S. studies renders them of limited utility, given different cultural differences and smoking patterns. In the first deliverable on workplace exposure to environmental tobacco smoke and lung cancei risk, four studies (Brownson et al. 1987, Chan and Fung 1982, Svendsen et al. 1989, and Lam 1985) collected data on workplace exposure but presented only an index of exposure to all sources. Therefore; it is incorrect for Dr. Brown to deduce on page 6, "The observation of elevated ORs in 3 out of 4 studies using a measure of total PTS that incorporates workplace exposure is consistent with the hypothesis that workplace PTS contributes to risk of lung cancer ... Lacking analyses of workplace exposure :.. the observed association is implicit rather than explicit " There is really no way of determining anything about the relationship of workplace exposure to lung cancer from these data. Workplace exposure could have shown no elevated risk and the same results could have been achieved given the exposure classifications in _. these studies (i.e., 4 or more hours per day spent in the presence of a smoker; exposure at home or at work). In addition, observing that three of four studies reported elevated risks, only one of which was statistically significant, also does not rule out chance in establishing an association. This faulty reasoning is apparent throughout the documents on spousal and workplace exposures to environmental tobacco smoke and lung cancer and cardiovascular disease. On page 10 of the first deliverable on workplace exposure to environmental tobacco smoke an lung cancer risk, Dr. Brown points out the inconsistencies of the U.S. data on workplace exposure compared to Asia and Europe and states: "there is no reason to think that if workplace exposure is a risk factbr for lung cancer, it would be so only in European and Asian populations: " Consideration should also be given to regional variations in working conditions and potential 4 I

(1994) appropriately concluded that each smoking study is confounded by uncontrolled influences and produces a biased estimate of ETS effect suggesting that weak spousal smoking- lung cancer risk elevations in the individual studies may well be the result of artifact alone. I I I I I I I I I Use of One-tailed Statistical Tests Dr. Brown's continued reliance on one-tailed tests to assess the relationship between environmental tobacco smoke exposure and disease is improper. For example, on page 5-2 of the f rst deliverable on spousal exposure to environmental tobacco smoke and lung cancer risk, the revision to Chapter 5 states that: "Throughout this chapter, one-tailed tests of significance (p=0.05) are used, which increases the statistical ability (power)'to detect an effect. The 90% confidence intervals used for the analyses performed are consistent with the use of the one-tailed test. The justification for this usage is based on the a priori hypothesis (from the plausibility of a lung cancer effect documented in Chaptets 3 and 4) that a positive association exists between exposure to ETS and lung cancer." However, the data contained in Chapters 3 and 4 of the EPA report concern levels of exposure to environmental tobacco smoke (Chapter 3) and discussions of lung cancer in active smoking and laboratory animals (Chapter 4), and, therefore, cannot be relied on as the basis of a decision to utilize a one-tailed test. Second, Dr. Brown's rationale for the use of the one-tailed test (90% confidence limit) for an overall estimate in a meta-analysis, rather than the conventional 95% confidence limit, is not supported by statistical or epidemiological theory or methodology in this instance. Statistical theory grew out of experimental science, which measures variables or responses. The typical test of significance for measured responses is a t-test, which tests differences in sample means. Two-sided (two-tailed) tests of significance are generally employed in the analysis of simple contrasts in experimental studies. However, one-sided (one-tailed) tests of significance (in which the investigator only considers deviations from the null hypothesis in one direction, ignoring deviations in the other direction) may be appropriate when the investigator either knows in advance that a treatment will affect outcome in only one direction or if the investigator has no interest in the outcome of a treatment unless it is superior to an existing treatment (such as in testing the efficacy of a new drug). In contrast to experimental science, epidemiolo;ical and observational studies are generally concerned with categorical or discrete data. Standard practice in these studies is to calculate odds ratios or relative risks. Statistical significance in epidemiologic studies is normally assessed using Chi-square tests, or equivalent procedures, which generate a confidence interval around the odds ratio or relative risk. Although a Chi-square test uses only one-tail of the Chi- square distribution, the test is in fact two sided, since a significant Chi-square can be generated by deviations from expected values in either direction. 10