The First Reports on Smoking and Lung Cancer-Why Are They Consistently Ignored?
Smith, George Davey, Egger, Matthias, Bulletin of the World Health Organization
Editor--In his commentary on early reports on smoking and lung cancer, Michael J. Thun ignores the actual first reports and concentrates on the series of case-control studies published in 1950, first in the United States and then in the United Kingdom (1). He also claims (as others have done) that in the pre-1950 era the most popular hypothesis was that "lung cancer mortality ... was more likely to have resulted from the widespread tarring of roads and exhaust from motor vehicles than from cigarette smoking". Could he please produce contemporary--i.e. pre-1950--references on this hypothesis that match the veritable mountain of publications from pre-1950 implicating cigarette smoking (2), in particular, formal analytical studies of road tarring and motor exhaust similar to the many analytical studies of smoking and lung cancer?
Of course, as is now well known (3, 4, 5-10), extensive research on smoking and lung cancer appeared before 1950, much--but not all --of it from Weimar and then Nazi Germany. In 1928 Schonherr from Chemnitz drew attention to the high rates of smoking in a series of lung cancer patients (11). He explicitly studied where the lung cancer cases lived and reported they were not closer to roads than expected (tar and exhaust already being an unsupported hypothesis) and suggested that lung cancer in non-smoking wives of smokers was caused by passive smoke inhalation. In his comprehensive review of smoking and lung cancer in 1929 (reviews were already being written 20 years before the 1950 "discovery") the influential actuary Frederich Hoffman wrote: "Unfortunately, German statistical discussions are invariably confusing and complicated by the omission of proper headings to statistical tables, which makes a full use of the elaborate investigation by Schonherr exceptionally difficult. Exceedingly rigorous reading is required to draw the best results from the mass of evidence presented. But every aspect of the problem is considered by this author, whose observations are deserving of being completely translated into English" (12). In a review published two years later, Hoffman considered that "smoking habits unquestionably increase the liability to cancer of the mouth, the throat, the oesophagus, the larynx and the lungs" (13).
In 1935, Fritz Lickint published an elegant synthetic review that considered time trends in lung cancer and cigarette smoking, ecological associations between smoking and lung cancer, autopsy series, experimental animal studies and clinical reports, which left him in no doubt that smoking was a cause of lung cancer (14). He had entitled a 1929 paper "Tobacco and tobacco smoke as etiological factors for carcinoma" (15); he simply did not think further studies were needed--what was required was to prevent smoking. His extraordinary 800-page work in 1939 (16) summarized all that was currently known from scientific research on the health consequences of smoking.
In 1939, a study of smoking habits of lung cancer cases compared with those of an ill-defined control group carried out by Franz Muller appeared in the Zeitschrift fur Krebsforschung (17), a leading international cancer journal of its time (a time when many of the major science journals were in German). Four years later a considerably more sophisticated study by Schairer & Schoniger appeared in the same journal (18); an English translation is now readily accessible (19). This study involved a population and hospital control group (findings were similar in both), analysed the potential effect of people with illnesses quitting smoking, and tabulated results of their study together with previous data to show consistency (a formal statistical meta-analysis was not, unsurprisingly, performed. The odds ratios that can now be calculated from these early studies are very similar to those of the later studies published in the 1950s, and for the Schairer & Schoniger study alone statistical evidence for an association can be shown to be strong (P < 0. …