Tobacco Documents Online

occupation, except for the lower estimate of the attributable risk for cancer. For this estimate the OTA Draft made a crude adjustment for lung cancer. 2. The Healthy Worker Effect on Estimates of Morbidity and Mortality. Individuals entering the labor force are in reasonably health. As a consequence, morbidity and mortality rates of good particular occupational groups are generally lower than those of the general population because the general population includes many individuals who do not seek employment or cannot find employment because of poor health. This effect was first recognized by William Ogle in 1885 and has received increasing discussion since 1976 (See for instance, Fox, 1976; McMichael, 1976; Ott, 1976; Vinni, 1980; Wang, 1982; Sterling, 1985 see APPENDIX 4). The healthy worker effect has two important consequences for an evaluation of the data on which the OTA Draft is based. 1. The mortality experience of nonsmokers in the American Cancer Society study of a million men and women (referred to henceforth as the ACS study) is the main source used by the. OTA Draft to estimate attributable risks. This volunteer population contained very few, if any, individuals without stable employment and a much higher percentage of professional, manageriall and technical workers than found in the national population. Unemployed and transient people 9 TIMN 308174

were largely excluded by the method for selecting volunteers. Consequently, any estimates of disease frequencies based on the ACS population would be biased. They are based on the disease frequencies of a population that is not only healthy to begin with (i.e. the healthy worker effect), but also usually unexposed to occupational hazards, dusts and fumes, so that the healthy worker effect would be maintained throughout their life. Similar biases are found for all volunteer cohorts, particularly for groups such as Mormons and Seventh Day Adventists who differ from the general population in respect to many health risk factors such as diet and occupation. (See APPENDIX 1 for a more detailed analysis.) 2. The healthy worker effect indirectly contributes to underestimating occupational health effects as studies of occupationally exposed cohorts frequently use the national population as a source of the expected rates of mortality or morbidity. (Expressed by the Standard Mortality Ratio (SMR) or Standard morbidity Ratio (SmR)). Underestimating occupational effects contributes to an overestimate of . smoking effects, as unfavorable health outcomes found for individuals who both smoke and work in hazardous jobs are invariably attributed to smoking. Goldsmith (1975) and Enterline (1975) suggested that mortality rates for blue-collar populations should be compared to .8 to .9 of the mortality rate for the general population. 10 TIMN 308175

This would crudely adjust for the healthy worker effect. Hammond (1980) determined SMR's for blue-collar workers in the ACS population, and found that they were over 1.35 in urban areas. However, reducing the expected mortality rates to .9 in order to account for the healthy worker effect would indicate that there was a 50% increase in expected mortality for blue-collar workers in the ACS population. Certainly this is not a trivial increase and the dismissal of these findings by Hammond and Garfinkel (1980) and Doll and Peto (1981) as not being substantive are not justified. SECTION II: DETERMINING ATTRIBUTABLE RISKS An estimate of the public health costs of smoking is dependent upon the proportion of various diseases that are assumed to be directly caused by smoking. The OTA Draft cites 16 references in OTA Table 1 which have apparently estimated the proportio.n of cancer, nonmalignant respiratory disease, heart disease, or several minor categories attributable to smoking. The Draft also produces it's own estimates of attributable risk (AR) due to smoking for cancer"(OTA Tables 3 and 4), chronic obstructive lung disease (OTA Tables 5 and 6) and for heart disease (OTA Tables 8-14). It is important to examine the~ methodology used in the OTA Draft to estimate AR's in order to determine how accurate these AR's are. Most of the-following discussion will be limited to the AR's for cancer, though the 11 TIM-S 308176

same methods are used to determine the AR's for other diseases. The 16 studies cited in OTA Table 1 suggest that there has been a widespread attempt to estimate the attributable risks due to smoking, and that, presumably, the OTA Draft draws on a large amount of available evidence. This is an erroneous impression. Four of the cited references are unpublished and therefore presently unavailable (From OTA Table 1: Rice and Hodgson, 1983; Minn Health Dept., 1984; Lewit, 1984; ACS, 1985) and one study was not listed in the references (From OTA Table 1: Whyte, 1976). Seven references did not discuss how their estimates for AR's were determined. Three of these studies cited other sources (Luce and Scweitzer, 1977; Richter and Gori, 1980; Kristein, 1977), which in turn were either incorrectly cited as they did not estimate AR's, or cited other unpublished sources. None of the four cited Surgeon General's Reports discussed how their estimates for the number of smoking related deaths were determined, though the 1982 Report's estimate of 129,000 cancer deaths due to smoking is probably.based on an adjustment of Doll and Peto's (1981) rates derived from ACS data. Only four of the studies in the OTA Table 1 were original published sources for estimates of AR ( Doll and Peto, 1981; Ravenholt, 1984; Hammond and Seidman, 1980; Enstrom, 1979a). A11 four of the studies used data from the 25 state ACS study conducted between 1959-1972 (Hammond, 1966; 1964; Garfinkel, 1980). The study by Ravenholt (1984) also used data from the U.S. veterans study (Rogot and Murray, 1980). In addition, the 12 TIMN 308177

AR's for cancer, chronic obstructive lung disease, and heart disease, provided in Tables 3-7 of the OTA Draft, were based on the ACS data. The available evidence for AR's is therefore based almost entirely upon the ACS 25 state study with minor use made of the U.S. veterans study. The methodology for estimating the AR's from these studies was essentially the same. Doll and Peto and the OTA Draft used the age-specific death rates for ACS nonsmokers, for specific cancer sites thought to be related to smoking, to determine the expected cancer incidence for the entire U.S. population if no one smoked. All of the deaths in excess of the predicted number of deaths based on the mortality rates for ACS nonsmokers were attributed to smoking. For example, there were 71,006 lung cancer deaths among U.S. males in 1978. 6,439 deaths were estimated to occur if the U.S. lung cancer mortality rates were the same as the ACS nonsmoker lung cancer mortality rates..The difference between the observed rate of 71;006 deaths and the predicted rate of 6,439 deaths was attributed to smoking (Doll and Peto, 1981). Hammond and Seidman, instead of using the difference between expected and observed deaths for the U.S. population, determined the difference between the death rates for ACS nonsmokers and the entire ACS population, adjusted to the standard 1970 U.S. population distribution by age and sex. The estimated AR from Enstrom, cited in OTA Table 1, was in fact made by Doll and Peto (1981). Enstrom had computed the age 13 TIMN 3~81~5

adjusted mortality rate for all cancers combined for the ACS population of nonsmokers and for the U.S. population of whites. The mortality rate for all cancer sites among ACS nonsmokers was only 62% of the U.S. mortality rate. Doll and Peto attributed the 38% difference in cancer mortality for U.S. whites to smoking. This is an incorrect comparison (and one that they did not make in their own estimate of AR), as it assumes that all cancers may be caused by smoking, when in fact there is only evidence to associate cancer of the lung, esophagus, bladder, larynx and buccal cavity to smoking. Enstrom did not use the ACS data to determine the AR for cancer from smoking because the ACS cancer mortality rates for both nonsmokers and smokers were significantly below the rate for the U.S. population as a result of the 'healthy volunteer' effect. Most of the 38% reduction,in cancer mortality for the ACS nonsmokers would have been due to this effect. Enstrom actually used a representative sample of U.S. whites who never smoked to estimate that 24% of cancers were attributable to smoking. Ravenholt determined an upper estimate for AR from the ACS study and a lower estimate from the study of U.S. veterans (Rogot and Murray, 1980) and took the average of the two as an estimate of AR. The AR's based on the veterans study are of doubtful validity, as one analysis of the study found 50,000 misclassification errors for smoking status (Sterling, 1979b, see APPENDIX 5). Ravenholt's upper estimate used relative risks for ACS male heavy smokers and an estimate of the proportion of 14 TIMN 308179

the U.S. population that ever smoked regularly to estimate the AR due to smoking. This.estimate assumed that all smokers are heavy•smokers and that all women smokers have the same increased risks as heavy male smokers. Consequently, little confidence can be placed on Ravenholt's 'average', and for this reason the following discussion will be limited to the estimates for AR derived from the ACS data. Problems in Using the ACS Data to Determine Attributable Risks The use of mortality rates for ACS nonsmokers to estimate the expected number of cancers or other diseases in the U.S. if no one ever smoked requires three major assumptions: '1. The ACS population of nonsmokers does not differ from the entire U.S. population in respect to other risk factors for cancer, heart disease, or nonneoplastic respiratory disease. These risk factors include age, sex, employment characteristics, occupation, alcohol consumption, race, socio-economic status, diet, urban or rural residence, etc., 2. Nonsmokers do not differ from smokers in respect to the above risk factors, and 3. Death rates for nonsmokers have not changed since the last follow-up of the ACS study populaton in 1972. Whether or not death rates for nonsmokers have changed in the last 15 years is controversial, with some evidence indicating that there has been an increase (Enstrom, 1979b) and 15 TIMN 308180

others disputing this evidence (Doll and Peto, 1981; Garfinkel, 1981). However, given the available evidence, the first two assumptions are the most problematic. 1. ACS Population Versus the U.S. Population The ACS population is known to differ from the U.S. population on many risk factors. The ACS population included only 3% nonwhites, compared to 12% in the U.S. population, a larger percentage of individuals from a high socio-economic status as indicated by the number of high school and college graduates, more rural residents, more native born whites, hardly any people without a stable long-term residence, and fewer people occupationally exposed to dusts and fumes (Hammond, 1980; Sterling, 1975; U.S. Stats Abs, 1985). An analysis by Hammond , (1980) on the effect of occupational exposure to dusts and fumes among the ACS population was limited to only 1500 out of a- population of over half a million male subjects who reported occupational exposures. This emphasizes the bias in the ACS study towards professional, technical and managerial workers. All of these differences between the ACS population and the U.S. population act to reduce the mortality rate in the ACS population both for diseases associated with smoking and for all other diseases. For example, the cancer mortality rate for the ACS male population, consisting of 78.2% ever smokers and 21.8% never smokers, was only 76% of the cancer mortality of a comparative age distribution of U.S. white males (Enstrom, 16 TIMN 308181

1979a; Hammond, 1966). The cancer mortality rate for the entire ACS cohort was 79% of a comparative age and sex distribution of U.S. whites. The difference between the ACS mortality rate and the U.S. rate including blacks would be even greater. Therefore, the AR estimates by Doll and Peto (1981) and the OTA Draft must substantially overestimate the true AR's. The estimates by Hammond and Seidman (1980) and Enstrom (1979a) are not affected by the 'volunteer effect', as Enstrom did not use data from a volunteer cohort, and Hammond and Seidman conducted an internal comparison that did not assume that the ACS population was representative of the U.S. population. 2. Nonsmokers Versus Smokers Nonsmokers differ in.several important ways from smokers. Nonsmokers are less likely to be occupationally exposed to toxic materials and are known to drink less alcohol, a major risk factor for oral and esophageal cancers. There is evidence to suggest that nonsmokers exercise more than smokers (Sterling, , 1971), eat more foods containing vitamin A or retinoids (Bjelke, 1975), which may have a protective effect against lung cancer (Graham, 1983), and drink less coffee than heavy smokers (Wynder, 1974). . Estimates of the AR due to smoking must adjust for the difference between the distribution of known and suspected risk factors among ACS nonsmokers versus the U.S. population. All of the studies discussed above and the OTA Draft adjusted for age, 17 TIMN 308182

but other possible risk factors were only partially adjusted in some of the studies, or not at all. This is especially true for occupational exposures to toxic chemicals, dusts and fumes. Table 2 summarizes the treatment of five major confounding factors by the studies which estimated the AR for cancer. The lack of adequate adjustment for risk factors which occur less frequently among nonsmokers compared to smokers means that all of the studies listed in Table 2, including Enstrom (1979b) and Hammond and Seidman (1980), overestimated the AR for cancer as a result of smoking. Comparing Estimates to Directly Calculated AR's Perhaps•the best evidence to•indicate that the AR's determined by the OTA Draft are unrealistic comes from case-control studies on specific cancer sites. In many case-control studies sufficient data is available to determine the AR for smoking, using the method discussed by Cole and McMahon (1971). Relevant examples were selected from a literature search of studies with information on both occupation and smoking. This does not, of course, provide a complete set of all relevant studies but there is no reason to suspect that the selection process would bias the results. None of the case-control studies selected were representative of the entire U.S. population of cancer cases, but the AR's determined from them should roughly indicate the extent of overestimate in the 18 TIMN 308183