The research that led Professor Bradford Hill and me to conclude that “cigarette smoking is a factor, and an important factor, in the production of carcinoma of the lung”(Doll, R., B. Hill, Smoking and carcinoma of the lung: a preliminary report, British Medical Journal, 1950; 2: 739–748) had been designed to find an explanation for the extraordinary increase in the mortality attributed to the disease over the previous 30 years. I have described elsewhere the background to the research (Statistical Methods in Medical Research 1998; 7: 87–117) and describe here only our attitude to that work and why we followed it with a cohort study of British doctors, which led to the demonstration of so many other harmful effects of smoking. Our initial work was planned to be a case-control study of patients suspected of having cancer of the lung, stomach, or large bowel who had been admitted to 20 large London hospitals. Patients with stomach or large bowel cancer were included to enable us to distinguish between findings that might relate to cancer of the lung and those (if there were any) that related to cancer in general. Patients for whom the diagnosis was only suspected were included in order to give us enough time to learn about their admission and arrange for them to be interviewed before they were discharged, which, in those days, would probably have been two or three weeks. Diagnoses were in fact often changed; subsequently, I had to visit each hospital to review the patients’ records and determine the discharge diagnoses and the strength of the evidence on which they had been based, noting for the lung cancer patients the presenting symptoms, the site of the lesion, whether the diagnosis had been confirmed histologically and, if it had been, the histological type.

Case-control studies, which have subsequently become one of the central planks of epidemiology, were not then common practice and there was no standard way of carrying them out. Our first important decision had, therefore, to be the choice of controls, and this was described in our first publication. The second was the design of the questionnaire. It was laudably short, filling only three and a half pages, but even so it included all the questions that we could think of that might be relevant to any of the three selected types of cancer. To guide us in its design we had only the knowledge that lung cancer had become progressively more common since the end of the First World War, was much more common in men than in women, and was somewhat more common in large towns than in the countryside, and that cancer could be caused in humans by ionizing radiation and the tar produced by the combustion of coal.