Role of Detection Bias
© 2000, Lauren A. Colby. Version 2.5
Chapter 8: Smoking Myths and the
Role of Detection Bias
A common myth about smoking assert that the lungs of smokers become brown or even black from years of accumulation of tars and goo. Not true, according to Wray Kephart. Mr. Kephart presently works as an engineer but he previously worked in a hospital, performing autopsies, most of which were paid for by insurance companies, seeking to determine whether the deceased committed suicide, or died from "natural causes". Kephart tells me that he's done approximately 1560 autopsies, and he's seen some strange things, such as the lungs of auto painters, which were "effectively sealed with catalyzed lacquers".
Kephart insists, however, that it is normally impossible to tell, from autopsy, whether the deceased was or was not a smoker. Upon resection, the lungs are always clear, unless the deceased lived in a large city where there was significant industrial pollution. In that event, carbon deposits may be found, but these are unrelated to smoking. So the "brown lungs" myth is exactly that: a myth.
Recently, I posed a question to Ed Uthman, M.D., a pathologist practicing in Dallas, TX. The question was whether a surgeon, at autopsy, could determine from an examination of the deceased's lungs, whether the deceased was or was not a smoker. Here is Dr. Uthman's response: I don't think one can tell if the deceased were a tobacco smoker or not by the appearance of the lungs. The absence of any black pigment suggests that the person was either a nonsmoker or a very light smoker. Heavy black pigmentation suggests that the person was either a heavy smoker, or lived in a city with heavy particulate air pollution, or was a coal miner, or some combination of the three. The black pigment in question is elemental carbon, which most investigators believe to be inert in its effects on the lungs (although in the extremely heavy doses that coal miners used to get, it may have had a partial role in coal-workers' lung disease).
When I point these things out to anti-smokers, they frequently say, "But I've seen photographs of smoker's lungs that were shown to me in grade school, and they looked simply horrible." I've seen these photographs also, but they are phonies. A popular Internet web site features side by side photographs of two lungs. One is labeled "Smoker's lung - dead at 50". The other is labeled "Non-smokers's lungs, alive at 70". The problem is simply that the photograph of the smoker's lung is a photograph of a lung ravaged by lung cancer; it is not a photograph of the lung of some smoker who died from some other disease. Therefore, even if the cancerous lung is from somebody who smoked, and the "healthy" lung is from somebody who did not, the photographs prove nothing except that cancerous lungs look different from non-cancerous lungs.
Of course, both photographs are photographs of dead people's lungs, because it's not possible to take a photograph of the lung of a living person. Also, rather obviously, the photographs show the outside surface of the lungs. The outside surfaces of lungs are not exposed to either air or smoke; therefore, it would be impossible for smoke to stain those surfaces.
Another myth, propagated by the anti-smoking crowd, is the notion that lung cancer was a rare disease in this country until some time in the 1930's, when it began to raise its ugly head as the result of smoking. Not long ago, George Will told a story on TV about a physician in the early part of the century who ran across a case of lung cancer and declared it to be such a rare disease that he assembled the medical students to witness the autopsy, believing it to be a rare opportunity.
The story may be true, but it proves nothing, because, in the early part of this century, the diagnosis of lung cancer was complicated by the "consumption factor". "Consumption" was a name applied to any disease characterized by emaciation, wasting away and coughing. It doubtless included the disease which we now know as "tuberculosis", but it also included other diseases, as well.
Funk and Wagnalls Encyclopedia, published in 1912, has an entry for "consumption". It says, "See: Pthisis". Under "Pthisis" we are told that "strictly speaking, the name includes a group of affections, but it is generally used to indicate pulmonary consumption, i.e., a more or less advancing process of lung destruction, associated with progressive emaciation and other characteristics and symptoms. This is a disease of grave importance, from its frequency and fatal tendency. It has been estimated that consumption is responsible for one-seventh of the mortality of Europe. "
"Tho pthisis was early recognized as a definite disease, and its clinical course fairly well studied, much obscurity has rested over its causation. Medical opinion was divided until 1882, when Koch announced the fact that he had discovered an organism, which he believed to be present in all cases of consumption proper. This organism, the bacillus tuberculosis, is a minute rod-like structure, capable of cultivation outside the human body, and easy of demonstration in the expectorants of consumptive patients...".
"Any condition that weakens the constitution favors the development of phthisis. Thus, malnutrition, syphilis, overcrowding, lack of fresh air, and defective hygiene, are all factors in the causation of phthisis. More especially is this true of occupations whose performance necessitates the inhalation of dust particles, e.g., stone masonry, knife grinding, metal polishing, wood carving, etc...."
"The early symptoms vary much. There may be nothing but a gradual loss of strength, it may be of flesh; there may be slight discharge of blood from the throat or chest; there may be a more or less persistent tickling cough; there may be breathlessness, with or without pain; or there may be little except a tendency to take cold easily...."
Clearly, the state of medical knowledge about "phthisis" was confused. The article implies that all cases of the disease were caused by the tuberculosis germ, discovered by the great Dr. Koch. But many of the symptoms described are applicable to lung cancer and, in 1912, most people were treated by family physicians who made house calls, and probably diagnosed most disease from the symptoms, rather than from any sort of laboratory analysis.
The Historical Statistics of the United States, published by the Government Printing Office, give cancer statistics from 1900 to 1970, but these statistics do not differentiate between different types of cancer. The following table, derived from the Historical Statistics, shows the number of deaths per 100,000 of the population, for tuberculosis, influenza and pneumonia, and malignant neoplasms (cancer), for the years from 1900 to 1970:
The government statistics contain no item for "consumption" or for "phthisis". However, as we have seen, "consumption" was still a recognized disease as late as 1912 (and probably later). No doubt, those early death certificates which listed the cause of death as "consumption" have been classified as "tuberculosis", in the later years. Note the nice, linear and inverse relationship between cancer deaths and deaths from "tuberculosis" ("consumption") over the time period covered by the chart. There is no doubt that some of the early deaths reported from "consumption" were really lung cancer. I've also thrown in the figures for influenza, because, in the early years, some terminal lung cancers may have been diagnosed as pneumonia, and also because it's simply interesting to note the devastating impact of influenza and pneumonia in the early years.
It is generally assume that today, doctors can easily recognize lung cancer when they see it. But can they? In 1959, in England, Heasman and Lipworth 25 surveyed reports from 75 hospitals of the National Health Service. Attending physicians diagnosed 338 cases of cancer of the lung, while pathologists discovered 417 cases, by post mortem autopsy. The attending physicians and the pathologists agreed, however, in only 227 instances. If the pathologists were correct, 111 (33%) of the diagnoses of the attending physicians were false positive, while 190 genuine cases of lung cancer (46%) were missed.
A similar result was obtained by Feinstein, in a study conducted at the Yale University School of Medicine, and published in September, 1986, in the Archives of Internal Medicine 26 . Researchers at Yale obtained records on 3,286 adults who had died between 1971 and 1982. 153 of these patients were found, upon autopsy, to have died of lung cancer. The researchers then went back and obtained the death certificates for these 153 patients and attempted to obtain information about their smoking habits. For 13 patients, adequate smoking information was not available, so they were thrown out of the survey. The researchers reported, however, that out of these 13 patients, seven had been correctly diagnosed as having lung cancer during life, but 6 had not.
Working with the remaining 140 cases, it turned out that there were 37 "surprise" cases of lung cancer, i.e., cases which had not been correctly diagnosed during life. 57% of these cases involved non-smokers; 30% involved moderate smokers; but only 16% involved heavy smokers. The researchers concluded that there was a detection bias; that doctors were very ready to diagnose lung cancer in a smoker; very reluctant to make the diagnosis in a non-smoker.
Before leaving this study, it is important to point out that, by reason of the methodology used, working from autopsies backwards to death certificates, the study could only expose false negatives, i.e., cases of lung cancer which had not been discovered during life. It is a pity that the researchers could not have conducted another study, working from death certificates forward to autopsies. That would have yielded a number for false positives, i.e., the number of cases diagnosed as having lung cancer which, upon autopsy, turned out not to be lung cancer.
At the beginning of this book, I said I would describe the work of a British medical researcher, who questioned the hypothesis that smoking causes disease. The researcher was the late Philip R. J. Burch, a professor of Medical Physics at the University of Leeds. He was a non-smoker, whose principal life work was an attempt to develop a unified theory of cancer.
In 1976, Doll and Peto issued a paper in which they reported that daily cigarette consumption by the British doctors who had been studied in connection with the 1964 SG's report had declined from 9.1 in 1951 to 3.6 in 1971. Doll and Peto claimed that, as a result there was a 38% reduction in lung cancer death rates amongst the doctors. In a paper 27 , however, Burch showed that Doll and Peto had compared the lung cancer death rates among the doctors with the lung cancer death rates for the entire British male population. Burch re-plotted the data to compare the doctors with themselves and showed that, on that basis, the risk for lung cancer amongst the doctors had actually increased by 31%.
Burch may have been on to something here, even beyond what he, himself, saw. His chart shows that during the time period 1955 to 1971, the risk of lung cancer amongst all men in England and Wales more than doubled, while the risk amongst the doctors increased only 31%. Remember our earlier discussion of socio-economic status? The doctors, of course, were, as a group, in a socio-economic class far higher than most other men. They worked indoors at a sedentary occupation, ate different food, and were not as susceptible to depression. Could these factors account for the difference between the doctors and ordinary men?
In the same paper, Burch plotted cigarette consumption for women and men in England and Wales against lung cancer death rates, during the period 1890 to 1971. He showed that the largest increases in LCDR's in both sexes came during the time periods 1916- 1920 and 1931-35, when at a time when cigarette consumption among women in England and Wales was very small. From this Burch concluded that the rise in lung cancer was due to improved diagnosis, not smoking. In England and Wales, there was, in fact, a 30 year gap between the time when males began smoking and females. So it is not surprising that the anti-smoking crowd in Britain made the argument that recent (in 1966) increases in lung cancer among women resulted from a "30 year incubation period". Burch effectively refuted that argument by plotting lung cancer rates for males in 1906 through 1926, against female rates for 1936 to 1966, and showing that while, if the incubation theory was correct, the two curves should have been synchronous, they were in fact completely dissimilar.
Burch also wrote, extensively, about the problem of "detection bias". Primary lung cancer can be simulated by pulmonary metastases from carcinoma of the pancreas, kidney, stomach, breast and thyroid, and by malignant melanoma. He suggested that many cases diagnosed as "primary lung cancer" are not, in fact, "primary lung cancer", but simply metastasized tumors, originating in some other site 28 .