© 2000, Lauren A. Colby. Version 2.5
Chapter 7: The Surgeon
General's Reports

Chapter 7: The Surgeon General's Reports

At this point, the reader will likely ask, "But what about the Surgeon General's Reports? Don't they prove that smoking causes lung cancer?". Actually, they don't.

It's not easy to get copies of these Reports. When I started my research, I combed the local libraries without success, and called major libraries all over the country. Nobody had any copies. One reason the Reports may be so difficult to obtain is that they contain material which might be embarrassing to the anti-smoking lobby, e.g., the data on pipe and cigar smoking.

Ultimately, I found a small company in Alexandria, VA, which was able to supply copies of the reports from 1964 through 1982, on microfilm. During that time frame, there were a total of 15 Reports, issued sporadically between 1964 and 1982. The largest , most massive Report was issued in 1979, and dealt with programs to "educate" (force) people not to smoke. The last Report that I have was entitled the "Changing Cigarette", and dealt with such things as filters, tar content, etc. The basic "science", purporting to show that smoking causes lung cancer was set forth in the first Report, in 1964, and for that reason I will concentrate here on an analysis of the 1964 Report.

The 1964 Report was issued by a committee of ten "scientists", picked from a list of 150 scientists and physicians, heavily weighted towards government agencies and large organizations active in public relations, with a low representation from the scientific community. There were no statisticians on the panel, although statistical expertise was essential to a proper analysis of the epidemiological studies, which formed a large part of the "evidence" which was studied. In 1965, a prominent statistician, K.A. Brownlee, of the University of Chicago, wrote a scathing review of the Report, pointing to many discrepancies in the statistical data. I will refer to that later 17 .

Prior to the writing of the Report, numerous experiments had been conducted, attempting to induce lung cancer in laboratory animals by painting their lungs and trachea with cigarette tars, forcing the animals to inhale vast quantities of tobacco smoke, etc. All of these experiments failed, miserably! Consequently, at page 165 of its Report, the Committee was obliged to concede that "Bronchogenic carcinoma has not been produced by the application of tobacco extracts, smoke, or condensates to the lung o r the tracheobronchial tree of experimental animals with the possible exception of dogs".

The phrase "possible exception of dogs" related to a single experiment, of which the Committee wrote that "this work has not yet been confirmed". To this day, it remains unconfirmed and it remains true, to this day, that despite hundreds of experiments18, nobody has been able to induce a single cancer in lab animals by exposing them to ordinary tobacco products or smoke.

Other researchers attempted to induce lung cancer in lab animals by using nasty combinations of industrial strength carcinogens. They used mixtures of ozonized gasoline and mouse-adapted influence viruses; polycyclic aromatic hydrocarbons, directly applied to the lungs of rats; mixtures of benzo(a)pyrene and iron oxide dust; radioactive cerium; and beryllium oxide. Even with these noxious brews, the results were not entirely successful. For one thing, some of the experimenters reported "distant metastases", i.e., tumors occurring in sites far from the lungs (which makes me wonder whether the "treatments" had simply weakened the animals' immune systems to the point at which cancers were springing up spontaneously throughout their bodies). Moreover, no t all the animals got sick. For example, two out of ten rhesus monkeys injected with beryllium oxide developed cancers but 8 did not.

The animal experiments having failed, the Committee was left with retrospective studies and prospective studies. Retrospective studies are studies in which cancer patients are interviewed about their smoking habits and compared with another group of controls from the general population, whose smoking habits are likewise identified. In prospective studies, a population is sampled, their smoking habits are ascertained, and they are then followed for a number of years, to determine who develops the disease.

The Committee had a number of retrospective studies available, but wisely decided not to rely much upon them, because of well known problems with such studies. Instead, it chose to rely upon seven prospective studies, as follows:

(1) British doctors, a questionnaire having been sent to all members of the medical profession in the U.K. by Doll and Hill, in 1956.

(2) White American men in 9 states, enrolled by American Cancer Association volunteers, each of whom enlisted 10 white males between 50 and 60 years of age. Hammond and Horn, 1958.

(3) Policy holders of U.S. Government Life Insurance policies. Dorn, 1958.

(4) Men, 35-64 in nine occupations in California which were suspected of having a high occupational risk of lung cancer. Dunn, Linden and Breslow, 1960.

(5) California members of the American Legion and their wives. Dunn, Buell, and Breslow, 1961.

(6) Canadian War Veterans. Best, Josie and Walker, 1961.

(7) American men in ten states, enrolled by volunteers from the American Cancer society, each of whom was asked to enroll about ten families containing at least one person over 45. Hammond, 1963 19 .

Now, right off the bat, there were several sources of bias immediately apparent in the manner in which the surveys were conducted. It was obvious to everyone, including the participants and their doctors why these studies were being conducted, i.e., to prove that smoking causes lung cancer. Thus, an element of detection bias was introduced. I'll return to that point shortly.

There was also the matter of the selection of the survey participants. Not all the holders of U.S. Government Life Insurance policies participated; not all the British doctors participated, etc. Taking the five studies for which it had data on the non- response rate, the Committee concluded that the average non-response rate was about 32%. Then, at page 116 of its Report, the Committee made the following curious observation. Citing a paper by Berkson 20 , the Committee said, "The death rate in the complete population (3.000) was 42% higher than the respondent death rate. The non-smoker death rate was over 38 times as high among non-respondents as among respondents (60.1221/1.553), whereas among smokers it was only 1.8 times as high. [Berkson's] calculations referred to an early year of the study, in which the differential entry of ill persons among smokers and non-smokers are likely to be most marked. Further, as we interpret his writing, the example was intend ed as a warning against the type of subtle bias that can arise whenever a study has a high proportion of non respondents, rather than a claim that this numerical estimate of the bias actually applied to these studies".

Thus, the Committee was confronted with what should have been a red flag: a finding that the death rate amongst non responding non-smokers was 38 times as great as the rate amongst responding non-smokers, whereas the death rate among non-responding smokers was only 1.8 times as great as the death rate among corresponding respondents. It is apparent, even to a layman, that such a major discrepancy could greatly skew the results of the surveys. Yet, the Committee brushed the point aside, saying, in substance, that it didn't think that Berkson meant what he wrote!

There were troublesome discrepancies. The Committee found that the most potent carcinogen present in tobacco smoke is benz(a)pyrene (p. 27). According to the Committee, cigar smoke has 4 times as much benz(a)pyrene as cigarette smoke, and pipe smoke ten times as much as cigarette smoke (p. 58). Yet, the Committee found pipe and cigar smoke to be pretty much innocent of causing lung cancer, and even concluded that pipe smokers live longer than non-smokers (unless they quit - the Committee concluding that those pipe smokers who quit had done so because they were already ill).

Some would argue, of course, that cigar and pipe smokers inhale less than cigarette smokers (although, in my case, I inhale both pipes and cigars). If, however, inhalation is a factor in the development of disease, it should show up in relative inhalation rates for cigarette smokers. A study was, in fact, conducted by Hill and Doll, which sought to classify cigarette smokers as inhaling vs. non-inhaling. At page 188 of the Report, there is a reference to a "negative association" between inhaling and lung cancer, based on the "early" Hill and Dole studies.

In 1959, in fact, R.A. Fischer analyzed some of the Hill and Doll data and concluded that inhalers have a lower rate of lung cancer than non-inhalers 21 . Fischer's findings were incorporated into Table 8 of the 1982 Surgeon General's Report, but the Report did not deal with this apparent paradox.

The Committee did, to some extent, recognize the effect of socio-economic status on the various prospective studies which it analyzed. Table 26 at page 109 of the Report showed incidents of morbidity, derived from all seven prospective studies, for 25 different causes of death. In all but two categories (cancer of the rectum and intestines), smokers showed an increased risk of death, as opposed to non-smokers. Indeed, it was claimed that smokers have increased risks of dying from such diverse causes as accidents and suicide, cirrhosis of the liver and bladder cancer, as opposed to non smokers. This troubled Brownlee, because he failed to see the "specificity" of smoking to the disease which the Committee claimed to be "caused" by smoking, i.e., lung cancer. After all, common sense would seem to show no connection between smoking and prostate cancer, or smoking and cirrhosis of the liver. Perhaps, what the studies were really studying was social class. Cigarette smokers tend to come from lower socio-economic strata than cigar or pipe smokers, or non smokers. Perhaps it is socio-economic status that accounted for the paradoxical finding that pipe smokers lived longer than non smokers and that cigar smokers lived the same.

Studies published in recent years (and therefore not available to the 1964 Committee) bear out the relationship between socio-economic status (SES), smoking and morbidity. A 1990 study 22 showed the following relationships between smoking and levels of education:

Percentage who Smoke (U.S.)

Years of EducationMalesFemales
less than 13 4136

A 1973 Study 22a correlated morbidity with educational levels, as follows:

Ratio of Observed to Expected Deaths, U.S., ages 21-65

Years of EducationMalesFemales
less than 51.171.60

What these studies show is that low class people tend to smoke more than higher class people, and that low class people tend to die sooner than high class people: considerably sooner. There may be many reasons for the higher death rates in people with lower SES. They tend to work in hazardous occupations, exposed to hazardous fumes and chemicals. They eat a different diet, tend to become obese, tend to receive less medical care and lower quality care. Moreover, they tend to suffer more from mental depression 23 . So the Committee's concern that the study results might be biased by SES turns out to have been well founded. Subsequent studies confirm that, smoking aside, it is risky to belong to the lower socio-economic strata.

While the Committee did, in fact, acknowledge the possibility of bias due to SES, it appears to have overlooked entirely another important source of bias. That is detection bias. Remember, everybody enrolled in the studies knew what was being studied, and their doctors knew that, also. Thus, everybody was waiting with baited breath for the smokers to develop lung cancer. I will discuss the role of detection bias in more detail in the next chapter. It should be noted, however, that the methodology followed in the SG's studies was calculated to exaggerate the possibility of detection bias, because the researchers were concentrating heavily upon the hypothesis that smoking causes lung cancer.

In the British Doctor's study, for example, all deaths in which lung cancer was a contributing cause were classified as deaths from lung cancer, even though the direct cause of death may have been something else (Report, page 101). It is interesting, in that regard, that the British Doctor's study was the one which purported to show the highest risk for lung cancer, from smoking 24 .

There was, however, another indication of trouble, which has been heretofore overlooked. This troublesome indicator is best illustrated by a more detailed discussion of one of the 6 cancer society studies discussed in the 1964 Surgeon General's Report.

During the time period from October 1959 through February 1960, the American Cancer Society enrolled men in a smoker survey, described in the Report as the "Men in 25 States" study. Female volunteers were each asked to pick ten families among their acquaintances, each with at least one person over the age of 45, and study them to find out whether they would die during the survey period and, specifically, whether they would die from lung cancer.

There were 448,000 useable replies, representing 448,000 men between the ages of 35 and 89. We don't know how many replies were rejected as unusable because each volunteer was free to use her own criteria. We also don't know how many smokers were studied as opposed to non-smokers because the results, published in the 1964 Surgeon General's Report, don't furnish that information. We do know that during the approximately 22 months that the survey lasted, there were 11,612 deaths. As the Surgeon General acknowledged, this translates to a death rate for both smokers and non-smokers, considerably below the overall death rate for white males, meaning that the participants in the survey were considerably healthier than the average person. At least, that's what the Surgeon General thought that it meant. I have other ideas.

The observed mortality ratios for different types of smokers, as opposed to non-smokers, were as follows:

Cigarettes only 1.83 Cigarettes and other 1.54 Cigars only 0.97 Pipes only 0.86

Thus, once again, as with Doll's study, it appears that cigar and pipe smokers actually lived longer than non-smokers - something that modern anti smokers would vigorously dispute.

The SG's Report does not list the number of lung cancer deaths which were recorded by the Cancer Society volunteers. Instead, the results are lumped in with five other studies (some or all of which also seem to have been organized by the cancer societies ), and Doll's study of British doctors. Lumping all of the studies together, there were 26,223 smoker deaths and 11,168 non-smoker deaths. Of these, 1,833 deaths from cancer of the lung occurred in smokers while only 123 occurred in non-smokers, yielding a mortality ratio of 10.8 for death from lung cancer among smokers as opposed to non-smokers.

Table 15 of the Report shows that for all of the various studies, the age-adjusted death rates for the study subjects were much lower than the age adjusted death rate of 22.9 per 1000 man years for U.S. white males, in 1960. In the case of the 25 States study, the death rate for the non smokers was 12.8, for smokers of less than a pack a day, 18..5, and for smokers of 1 pack or more, 19.2. These results were similar to the 5 other cancer society studies, but the Men in 25 States results bear a footnote saying that "These results may be too low by about 1.7%, since the person-years used in the computation included some contribution by men who had not been full traced".

Table 2 at page 85 of the Report gives the mortality ratios for current smokers for various studies, including Men in 25 States. We are assured by the Surgeon General that the figures were age adjusted. Thus, we might expect that all figures given in the Report would be age adjusted and represent current smokers. It turns out, however, that this is not the case.

Table 19 at page 102 shows the number of deaths from each of 25 different causes (ranging from lung cancer down to cancer of the intestines). The figures given in Table 19 represent the sums of all of the deaths recorded in all of the seven studies. It bears a footnote, reading: "Current cigarettes only for four studies: all cigarettes (current and ex-) for the two California studies and Men in 25 States". That little word "ex" has tremendous significance. It means that for purposes of calculating the lung cancer death rate, the Cancer Society dropped the practice of classifying only current smokers as "smokers" and chose, instead, to treat anybody who had ever smoked a cigarette as a "smoker".

In 1961, 68% of the men in America smoked. Therefore, by the time a man reached the age where lung cancer becomes a problem (essentially 50+), the likelihood that he would have smoked, at some time in his life, surely approached or even exceeded 90%. It should not have been a surprise, therefore, nor did it prove anything, that 90% of the lung cancer deaths were in "smokers" since, if a smoker was defined as anybody who had ever smoked, 90% of the population susceptible to lung cancer was comprised of "smokers".

Lumping the seven studies together was also a statistical mistake. Each had different methodologies. Different age groups were studied and different populations (British doctors, U.S. Veterans, etc.). To really sort out what was going on, we need to see the numbers for each individual study but, at this late date, we probably never will.

The fact that there were so few deaths during the study period, compared to the deaths that would be observed in a cross section of ordinary white males, worried the Surgeon General. I gather that the footnote, suggesting that the figures from Men In 25 States might be a bit low, was part of an effort to explain the discrepancy but, if so, it was a misplaced effort since all of the figures from the other 5 cancer society studies were in the same ball park. Still, the Committee found it necessary to speculate, at some length, concerning the discrepancy, suggesting, among other things, that people who were already sick might not have been chosen as study participants, by the volunteers.

One thing that did not occur to the Committee or, if it did, was not mentioned, is that the reported data, itself, may have been wrong or incomplete. This seems to me to be the most logical possibility. According to the Surgeon General, the ladies who conducted the study were free to weed out any responses which, for any reason, they felt to be inappropriate. Also, according to the Surgeon General, the ladies were expected to get a death certificate when a death was reported. I have a feeling that the reason there were so few deaths, particularly among non-smokers, was simply that the ladies didn't report all the deaths. Getting a death certificate would have been as lot of trouble and, if somebody died from some cause which seemed totally unrelated to smoking, the ladies might well have concluded that it wasn't really relevant, and wasn't worth reporting.

In any event, in all of the cancer society studies, the overall death rates, for smokers and non smokers alike, but especially for non smokers, were much, much lower than the death rates for the general population. This should have been a red flag: it should have at least raised questions concerning the quality and/or completeness of the data. But to the Surgeon General's Committee, bent on proving that smoking causes lung cancer, it suggested only that the study subjects were, for some reason, exceptionally healthy.

In 1991, Doll did a forty year follow-up of the doctors in his study, which is available on line at:

There, we are told that most of the doctors who were smokers of cigarettes only (as opposed to cigars and pipes) at the time the study began had given up smoking by the time of the follow-up, so that only 6% continued to smoke. Indeed, most gave it up within a few short years after the study began. This, however, did not deter Doll, who continued to try to estimate the number of pack years smoked by the quitters, and to try to develop correlations to lung cancer. To do this, he had to go by the recollections of those interviewed by mail at infrequent follow-ups, as to how long they had smoked and when they gave it up. This seems to me to be a fallacious approach, since it introduces the very element of recall bias that the prospective studies were supposed to avoid. The same approach also required Doll to make numerous adjustments to take into account the effects of quitting - adjustments which, wittingly or not, allowed his biases to get in the way of objective analysis.

Perhaps, however, the most damaging element of the Doll study is an admission that he made when the study was finally terminated, in 2001. Writing in the December, 2001, issue of the British Medical Journal, Doll explained that the study was "devised by Sir Austin Bradford Hill to achieve maximum publicity for the critical link between smoking and lung cancer". In short it was never intended as a serious scientific study to test the hypothesis that smoking may cause lung cancer. From the beginning, it was just propaganda - well intended, perhaps, but propaganda none-the-less.

Table of Contents | Next Chapter (8): Smoking Myths and the Role of Detection Bias