Tobacco Documents Online

• by lacob Sullum 0 LAST FF_BRL'ARY THE ADMINISTRATOR OF THE Envinonmental Protection Agency told a House subcommitcee that Congress should ban smok- ing in places of business. Testifying in favor of the Smoke-Free Environment Act, which would forbid smoking in buildings open to the public, Carol Browner relied heavily on the EPA report that declared environmental tobacco smoke (ETS) to be "a known human lung car- cinogen." Since it was released in Januarv 1993, this 510- page document has become a favorite prop of the anti-smoking movement. It has helped jus- tify smoking bans in government agencies-in- ciuding the Departrnent of Defense-in cities such as Los Angeles and San Francisco, and in states such as Maryland and Washington. Be- cause the EPA's preliminary conclusions about ETS were rirst publicized in 1990, the report had an impact even before it appeared in its fi- nal form. "Hundreds of local ordinances have been passed or introduced in virtually every atea of the country since 1991; Browner testified. "In the year since publication of the EPA re- port...we have seen a rapid acceleration of inea- sures to protect non-smokers in a variety of settings:" And in March, the U S. Occupational Safety and Health Administration (OSHA) pro- posed a ban on smoking in indoor workplaces, induding bars and restaurants. .. .--- .- . ,- - In light of the legislation and policy changes it has generated, the EPA's Respimtory Health Ef- fects of Passive Smoking: Lung Cancer and Other Disonders may be the most influential report ever issued by the agenry. As one might expect, it has received extensive coverage from major newspapers. Between May 1990 and February 1994, The New York Times, Los Angeles Times, The Wall StreetJoumal and The Washington Post ran more than 100 news stories about ETS, of MEDIACRITIC 41

• • which about 45 focused on the EPA report Yet almost without exception, the coverage has been one-sided, credulous and superficial. Even before the EPA released its report. journalists were quick to accept the claim that second- hand smoke kills. And despite serious questions about the report's assertion chat ETS causes lung cancer and the process by which the EPA reached that conclusion, leading U.S. news- papers have treated this assertion as scientinc fact. In so doing, not only have thev exagger- ated what is known about the effects of ETS, but they have missed an important story about the corruption of science by the political cru- sade against smoking. To uncover the facts would not have required a lot of di "9ino, . They were repeatedly outlined by representatives of the tobacco industry for anyone who would listen. Indeed. that was a big part of the problem. "The tobacco industn• has established a reputation for disseminating mis- information:' sa}S Michael Fumento. one of the few journalists who took a critical look at the science behind the EPA's report. "At the very least. [the industry] has been known to put a twist on material that isn't «arranted. In a sense. it was the boy who didn't cry «•olf-the guv who year after year saw a wolf uId claimed there was no wolf there. When he savs. 'Look, there's no wolf there.' the media are not going to be quick to believe that." In fact. most reporters were so disinclined to believe the tobacco industry that they simply as- sumed there was a woif, %-ithout attempting to verify its e.,istence. On )anuar}• 6, 1993, Los An- geles Times writer Rudy Abramson reported: "The most bitter resistance to. the EPA's move to link secondarv smoke and lung cancer has been waged by Philip Morris Co., a leading cigarette manufacturer, and by the Tobacco In- stitute, the industry's chieflobbying organization. Some 30 years after the landmark surgeon gen- eral's report on smoking and health, the indus- try continues to argue that there is no scientific proof of a link between cancer and smoking." The message of this juxtaposition is dear. Since the tobacco industry has teiused to acirnowledge that smoking causes lung cancer, people should not give credence to their claims about ETS and lung cancer. This argiunent, which showed up 42 -MED[ACRITIC repeatedly ia coveiage-a. ~ie EPA izport, seeks to simultaneously discredit criticism of the agen- cy's position and bolster the case against ETS. It implies not only that the tobacco industry is lying, but that the evidence of a link between ETS and lung cancer is just as strong as the ev- idence of a link between smoking and lung can- cer. This analogy is very misleading. James Enstrom, a professor of epidemiology at UCLA, notes that thousands of studies have examined the link between smoking and lung cancer. Virtually all of them have found pos- itive associations, statistically significant in the vast majority of cases. This is an important point. In any study that tries to measure the association between a sus- pected risk factor and disease rates, there is al- ways the possibility that an observed difference between the exposed group and the control group occurred simply by chance and had nothing to do with the risk fa.ctor. Researchers do statistical tests to account for this possibil- itv By convention, epidemiologists call a result significant if the possibility that it occurred by chance is five percent or less. The associations between smoking and lung cancer are sizable as well as statistically significanr. Recent stud- ies indicate that the average male smoker is 20 times more likely to develop lung cancer than a male non-smoker, while the risk ratio for women is about 10 to one. The figures are even higher for heavy smokers. y contrast, the EPA report was based on D 30 epidemiological studies that looked for a link between ETS and lung can- cer, mainly by comparing disease rates among non-smoking women living with smokers to disease rates among women living with non- smokers. Most of the studies found positive as- sociations, but they were statistically significant in only six studies. (Nine found that living with a smoker was associated with a reduced risk of lung cancer, but these results were not statistically significant.) And aIl of the positive associations were weak by epidemiological standards, typically yielding risk ratios of less than three to one. The EPA estimated that a woman who lives with a smoker is 1.19 times as likely to develop lung cancer as a woman

• • who lives with a non-smoker. "Comparing that to a 10 to one ratio, you can see it's minute," Enstrom savs. "It's at least one order of magnitude different from the active smok- ing data: " With risk ratios this small, it's difficult to rule out confounding variables, such as diet and oth- er sources of pollucion, that might account for an observed association. "At least 20 con- founding factors have been identified as im- portant to the development of lung cancer," wrote Gary L. Huber, a professor of medicine at the University of Texas Health Science Cen- ter, and two colleagues in the July 1991 issue of Cvnsumers' Researrll. "No re- ported study comes anywhere dose to controlling, or even mentioning, half of these." Enstrom is not optimistic that future research will clarifv the issue. "You're talking about ra- tios that are so close to 1.0 that it's really beyond the realm of epidemiology;" he says. "You could do more studies, and you could probably arrive at more precise tatios. but as to whether from its usiial'la~assent *procedure" combining the results from these 11 studies in a"meta-analysis:' This technique is appropri- ate only when the underlying studies are com- parable in method and structure. Enstrom says using meta-analysis for studies such as those examined by the EPA "is not a particularly meaningful exercise," since the studies are apt to differ in the way they define smokers, the types oflung cancer they include, the confounding variables they take into ac- count and so on. "It's just fraught with dan-~ gers:" In any event, the result of the EPA's meta-analysis is significant only under the weak definition adopted especially for these data. By the con- ventional standard, the meta- analysis does not support the claim that ETS causes lung cancer. Furthermore, had the EPA included in its meta-anal- ysis a large US. study pub- lished in 1992, the result might not have been significant even by the revised standard. The contrivances employed by the EPA, which a July 31, The EPA finessed some key points to conclude that passive smoke causes lung canceY those ratios would mean anything, I doubt it....You're basicallv down in a noise-level sit- uation. and whether you can really see a true signal above the noise is doubtfiul:' Faced with evidence that was weak, inconsis- tent and ambiguous, the EPA finessed some im- portant points and gave the data a vigorous massage to arrive at the conclusion that ETS caus- es lung cancer. To begin with, the EPA used an uncom-entidnal definition of statisdcal sigrificance. In pre%rious risk assessments the EPA had always used the traditional standard. But in the case of ETS, the agency abandoned the usual definition of statistical significance and called a result sig- nificant if the probability that it occurred by chance was 10 percent or less-- a change that in effect doubles the odds of being wrong. Even according to the broader definition, only one of the 11 U.S. studies that the report an- alyzes found a statistically significant link be- tween ETS and lung cancer. And according to the usual definition, none of them did. In or- 1992 Science article described as "fancy statis- tical footwork:' indicate that the agency was de- termined to reach the conclusion that ETS kills non-smokers. That impression is supported by the fact that the EPA put together a "policy guide" for reducing workplace exposure to ETS well before it had officially decided that ETS was a hazard. The first draft of the guide was released in June 1990, three-and-a-half years before the EPA released the final version of its risk assessment. William Reilly, then ad- ministrator of the EPA, told The Wall Street Journal in January 1993 that he delayed release of the policy guide in its final form because he didn't want it to "look like we're trying to torque the science." Reilly had reason to be concerned about that perception. In March 1992, an expert panel that he convened had issued a report called Safeguarding the Future: Credible Science, Cred- ible Decisions. Among other things, the panel conduded that "EPA science is of uneven qual- der to bolster the evidence, the EPA departed ity, and the agency's policies and regulations are MEDIACRITIC 43

• • frequently perceived as ricking a strong sci- entific foundation." It cautioned that "science should never be adjusted to fit policy, either consciously or unconsciously." D espite these and other warning signs, the coverage by the major newspa- pers was generally unskeptical of the agency's conclusions and dismissive of the tobacco industry's criticism. The typical sto- ry opened with the government's claims, elaborated on them for several paragraphs, quoted anti-smoking activists who agreed with the EPA and described the tobacco in- dustry's response in a para- graph or two. The tobacco industry's comments usually amounted to little more than a denial, and no independent sources were provided to back them up. News consumers were left with the impression that, aside from industry rep- resentatives, no one had doubts about the EPA's posi- tion on the health effects of ETS. similar questions about the report were raised that summer in congressional hearings and in a tobacco industry lawsuit challenging the EPA's findings. During the year after Fumento's piece appeared. only one story in a major newspaper dealt with the issues he raised in a less than per- functory way. In a July 28, 1993 article about the tobacco industry's lawsuit, Wall Street Jour- nal reporter Jerry E. Bishop made it clear that questions about statistical significance and con- founding variables are legitimate and not eas- ily dismissed. Although he did not quote any critics of the report who were not affiliated with the tobacco industry, he at least showed that statisticians disagree about the quality of the EPA's work. By contrast, a June 23, 1993 story byJournal reporter Eben Shapiro unfairly and erro- neously attacked one of the in- dustry's major claims: that the EPA excluded from its meta- analysis a large US. study, pub- lished in the November 1992 issue of the American Journal of Public Health, that would have changed the report's conclu- Reporters don't trust the tobacco companies. But they do trust the EPA. But as Michael Fumento showed in his Jan- uary 28, 1993 story for Investor's Business Dai- ly, this was cleariy not true. "Some scientists and policy analysts who say they couldn't care less about tobacco company profits or even the rights of smokers are worrying aloud that the EPA report is paving the way for justif%ring new health-based government regulations and pro- grams without any real science behind them," he wrote. The story quoted a series of credi- ble sources, including epidemiologists and statis- cicians, who questioned the quality of the evidence linking ETS to lung cancer and took the EPA to task for manipulating the data to make its case. Fumento cited a 1992 article from Toxicologic Pathology in which Alvan Feinstein, an epidemiologist at Yale University, reported a comment by a leading public-health re- searcher: "Yes, it's rotten science, but it's in a wordrv cause. It will help us get rid of cigarettes and become a smoke-free societv" It's difficult to understand why virtually no one followed Fumento's lead, especially since sions. Shapiro wrote that the study, which was included in a tobacco industry press package about the lawsuit, "actually appears to support the EPA's decision. The report...condudes that there is 'a small but consistent elevation in the risk of lung cancer in non-smokers due to pas- sive smoking: " Thus Shapiro implied that the results of the study supported the claim that ETS causes lung cancer. But the sentence from which he quoted actually says that "our study and odcers conducted during the past decade suggest a small but consistent elevation in the risk of lung cancer." (Emphasis added.) In fact, the study itself did not find a statistically significant association between ETS and lung cancer. That is why the tobacco companies argued that it would have undermined the EPA's case. Shapiro also smugly quoted the researchers' opinion that "the proliferation of federal, state and local reg- ulations that restrict smoking in public places and work sites is well-founded:" This editori- al comment does not change the data. Many other stories raised false doubts about 44 MEDIACRITIC

th f th EPA' _ yIn thcJ'ii~""`i~ a Geaeral's ieport ia 1986" e ar uments o t a g e s ca . t ......;-~c .a ~.~.. 22. 1993 New York Times, for example, Philip ~ie wrote, . .ing-:an early version of the risk J. Hilts reported that Representatives Thomas assessment "The newer studies confirm [the re- J. Bliley, Jr. (R-VA) and Alex McMillan (R- suits] in the first 13 studies:' The reader is not NC) "suggested that the EPA's study of several likely to guess from this summary that the vast studies. or 'meta-analvsis,' used a lower standard majority of these studies failed to find a sig- of statistical proof than normally used in as- nificant link between ETS and lung cancer. sessing danger scientifically." Despite the im- Altman is not alone in failing to discuss sta- plication of the word suggested, this is not an tistical significance. Conrider Jane E. Brody's Jan- arguable point. although the report's detractors uarv 8, 1992 New York Tunes story about a stud~• and supporters disagree about its importance. directed by Elizabeth Fontham of Louisiana State Hilts also stated that "about 30 studies were re- University Medical Center. The headline read: viewed. of which 24 showed that second-hand "New Study Strongly Links Passive Smoking 'smoke was a risk" just the opposite was true. and Cancer." Brody reported: "The study, the And he had the congressmen conceding the largest ofits kind, found a 30 percent higher risk • very point they were disputing: "The biggest of lung cancer if the women's husbands smoked, study, the two lawmakers noted, found statis- a risk that rose with the number of cigarettes and tical proof that second-hand smoke caused can- years of exposure:' Brody failed to note that this cer with certainty only in those people overall association was not statistically significant subjected to the most smoke:' No study has (that is, the probability that the result occurred ever found "statistical proof that second-hand purely by chance was greater than five percent). smoke caused cancer with certainty." (In fact, Although Fontham et aL reported statistically sig it is impossible for an epidemiological study to nifieant associations for a few subgroups, the risk provide such proof.) ratios were all under 2.5, so it is wrong to say These errors in stories about the EPA report that the study "strongly links passive smoking reflect a general tendency in coverage of the and cancer." ETS controveny to exaggerate evidence and minimize criticism. An example is a May 29, Another common error involves con- 1990 The \'eIV York rmes story by Lawrence fnsing correlation with causation. In K. :~ltmzn. Under the headline, "The Evidence 1991, for example, the U.S. Centers Mounts on Passive Smoking," Altman described for Disease Control (CDC) did a survey that, agro~ring scientific consensus that ETS is a among other things, asked parents to assess health hazard. He quoted one scientist who said their children's health. The CDC reported • "the links between passive smoking and health that 4.1 percent of the children who lived problems are now as solid as any finding in epi- in households with smokers were said to be in demioloK;' and another who claimed "there's "fair" or "poor" health, compared to 2.4 no question" that ETS causes heart disease. percent of the children who lived in house- Both assertions are conuovetsial, to say the least, holds without smokers. From this information but Altman did not offer specific rebuttals ftrom it is impossible to conclude anything about amrone. In the +3-paragraph article, he devoted the effects of ETS, since the study did not onlv three paragraphs to skeptics, both identified control for variables that might account for with the tobacco industry. the difference in reported health. Poverty is And Altman himself exaggerated what the ev- the most obvious example. Research shows idence tells us. In the second paragraph, he as- that people with lower incomes are both serted that "the studies show" ETS "causes more likely to smoke and more likely to be in death not only by lung cancer, but even more poor health. bv heart attack:' Thus, he declared at the out- Yet on June 19, 1991, The New York Tirnes, set of the story that the case was dosed on ETS. The Wall Street Journal, Los Angeles Times and "The EPA reviewed 24 epidemiological stud- The Washington Post all ran stories under head- ies of passive smoking and lung cancer, 11 more lines asserting that the study had found that MEDIACAITIC 45

smolangin tlie lioine harms cukhrii.: Only The '.,---.'.~remeMMM.s m~"'ed could do more New. York Times and The Washington Post not- ' for the public health than any other bill in ed that the study did not control for income, years." So, just as the tobacco companies have and only the Post made the importance of this an interest in minimizing the dangers of ETS, fact clear. the government and the anti-smoking groups The errors that appear in these and other sto- have an interest in maximizing them. ries about ETS are not random, of course. They When reporters choose sides on the basis of consistently weigh in favor of the view that ETS trust, they fail to make independent assessments is a serious health hazard. Reporters are re- of the arguments of both sides. So readers of ceptive to that view for a number of reasons. stories about ETS might wish to keep in mind Even if they don't personally disapprove of the following points: smoking, they are well aware of its dangers. If 0 The Impuiance of SUdsWW SWWmnm When re- • • a lot'of tobacco smoke hurts smokers, it seems plausible that a little would hurt non-smokers, though not as much. Since most journalisrs do not have backgrounds in statistics or epidemiology, they rely on oth- er people to assess the issue. The most conspicuous sources for stories about ETS «ork for the tobacco industry, the gov- ernment and anti-smoking groups. Reporters don't trust the tobacco companies. But in contrast to the skepticism they bring to the pronouncements searchers do not come up with statistically sig- nificant results, thev tend to underplay this fact, for obvious reasons. Stories should be examined to see whether they disclose, as a good report should, whether a Irsult is statistically significant. Epidemiological studies include "confldence intervals" that in- dicate there is a 95 percent probability that the true risk ra- tio lies between two numbers. If the lower number is 1.0 or less, the result is not significant, even if the authors of the paper When someone cites a `pattern" or "trend in the data," it's time to look more closely. of other government agencies and special-in- tetest groups, they do tend to trust public-health authorities such as the EPA and anti-smoking organizations such as the American Cancer So- ciety. The governing assumption seems to be that the tobacco companies are m-ing to main- tain profits, while the government and anti- smoking groups are interested in promoting public health and getting out the facts. B ut sometimes these tno missions con- flict. Public health officials may be in- dined to shade the truth a bit if it helps to discourage smoking by making it less ac- ceptable. In her testimony last February, EPA Administrator Browner said the main benefit of the Smoke-Free Environment Act would be its impact on smokers. "The reduction in smoker mortalitv due to smokers that quit, cut back or do not start is estimated to range ftom about 33,000 to 99,000 lives per year," she said. And six former surgeons general, The New York Iimes reported, "echoed the 46 MEDIACRITIC play it up in the abstract. When researchers don't get a significant re- sult overall, they sometimes slice up the data into subgroups, seeing if they can find a sig- niflcant association at cermin levels of exposure, for certain kinds of cancer and so on. But the more such comparisons they do, the less like- ly it is that any association they find will be meaningful, since there is a five-percent chance of being wrong each time. Furthermore, the subgroup data for ETS and lung cancer are of- ten contradictory: One study will find a sig- nificant result for adenocarcinoma lung cancer but not for other types of cancer, or for spousal smoking but not childhood exposure, while an- other study will find the opposite. p The pitfails of Contla~ versus Causaiion. Even a statisticaIly significant association between A and B does not prove that A causes B. A and B could both be associated with another factor or set of factors. An article in the July 28, 1993 Journal of the American Medical Association re- ported that, allowing for differences in smok-

• • ' link lietween"'E~~'iiancer or fieaif dig- ease. In this case, the reporter misunderstool his source, and a phone call to the AHA would have cleared up the matter. In other cases, it might be necessary to consult an independent authority familiar with the research. Reporters will soon have an opportunity to do better. In testimony last February, EPA Ad- ministrator Carol Browner predicted that the Smoke-Free Environment Act would save the lives of 5,000 to 9,000 non-smokers each year.+ Dave Mudarri of the EPA's Indoor Air Divi- sion says fewer than 2,200 of these represent lung-cancer cases; the rest are heart-disease deaths. Yet the evidence of a link between ETS and heart disease is even weaker than the ev- idence of a link between ETS and lung can- cer, and the EPA has never done a risk assessment in this atra. The agency's fuIl report on the impact of the Smoke-Fiee Environment Act was scheduled to be released in the spring. If reporters want to get at the truth, they can- not continue to act as if only one side in this debate has an ax to grind. They need to be just as skeptical about the EPA and the Coalition on Smoking or Health as they are about Philip Morris. "I treat sources like lawyers, like ad- vocates in a court of law," Fumento says. In a court of law the jurors take for granted that each side has an agenda, but that does not stop them from weighing the arguments. Similar- ly, reporters should not dismiss a statement sim- ply because it comes fmm the Tobacco Institute. Writing in Toxicologic Pathology, Yale epi- demiologist Alvan Feinstein cautioned his fel- low scientists against automatically believing everything the "good guys" say and rejecting everything the "bad guys" say. His message ap- plies to journalists as well as scientists: "If pub- lic health and epidemiology want to avoid becoming a branch of politics rather than sci- ence, the key issues are methods and process, not the 'goodness' of the goals or investigators. In science even more than law, the 'bad guy'...should always have the right to state his case, and a well-stated case has the right to be heard, regardless of who pays for it:'* on the effects of secondary smoke." In fact, Jacob Sullum is managing editor of Reason ing rates, restaurant workers are 50 percent more likely to get lung cancer than people in other occupations. The study controlled for smoking but not for a wide range of other fac- tors that could affect lung cancer rates. Yet cov- erage in The Washington Post, The New York Times and Los Angeles Times supported the au- thor's conclusion that the higher incidence of lung cancer should be blamed on higher lev- els of tobacco smoke in restaurants. The concern about confounding variables is especially important when risk ratios are small. Epidemiologists generally consider an associ- ation "weak" when the ratios are between 1.0 and 3.0. In the restaurant study, the risk ratio emphasized by the author was about 1.5. "Any- thing with a risk ratio of less than 3.0, 1 don't trust," Fumento says. "It's like measuring the width of a hair with a standard 12-inch ruler. You can't do it. The little markings are too big. So it is with epidemiology. It's a blunt tool:' 0 Weasel Words. Readers should be alert to qual- ifiers and hedging; so should reporters. In the restaurant study, for example, the author wrote: "The epidemiologic evidence suggested that there may be a 50 percent increase in lung cancer risk among food-service workers that is in part at- tributable to tobacco smoke exposure in the workplaca' (Emphasis added.) The 1991 report on the CDC survey of childtrn's health said the results "show an apparent pattern suggesting that, for most children, fair or poor health appean to be associated with various exposures to cigarette smoke:' (Emphasis added.) When someone cites a"pattern" or a "trend in the data," it's time to look more closely. In rigorous science, close doesn't count. 0 Disonepancia. When two versions of a verifi- able fact diverge sharply, readers should reserve judgment. For instance, an Associated Press sto- ry that appeared in 71fe 1Vev Ioik Temes on June 11, 1992, quoted a physician who appeared at an American Heart Association (AHA) press conference as saying that "thousands of stud- ies have shown that secondary smoke increas- es the risk of heart and lung disease:" The Tobacco Institute, on the other hand, "insist- ed that fewer than 100 studies had been done about a dozen studies had found a significant magazina MEDIACRITIC 47