Proof of Causality: Deduction from Epidemiological Observation

It is a particular pleasure to have the honor of giving a Fisher memorial lecture, for it gives me the opportunity to acknowledge the formative influence that Sir Ronald's book, Statistical Methods for Research Workers (1934) had on my career.

I had, as a boy, developed a love for mathematics and when, in October 1931, I started to study medicine, I sought ways in which I could apply numerical methods in my work. Nothing productive emerged until, a few years later, I discovered his book and realized how relevant it was to medical research. About that time, one of the teaching staff drew our attention to a report of the use of pituitary gonadotropic hormone in the treatment of undescended testes in adolescent boys, which the authors said had a smaller percentage of failure than had been achieved by reliance on spontaneous descent (Spence and Scowen 1935). There were, of course, many difficulties in making such a comparison; but suffice it to say that even if the series were properly comparable, no consideration had been paid to the possibility that the difference observed might have been due to chance. Fisher's book led me to Pearson's c ² test, which showed that with the numbers observed as big or bigger differences would have been expected to occur by chance about as many as six times out of 10—a finding that I published in the St. Thomas's Hospital Gazette and that is recorded as my first scientific publication (Doll 1936).

Another 20 years had to pass, however, before I met Sir Ronald personally, when we both took part in a debate organized by Cambridge University Medical students about the validity of Bradford Hill's and my claim that our case-control study of patients with and without lung cancer justified the statement that cigarette smoking was a cause, and an important cause, of the disease (Doll and Hill 1950, 1952).

Meaning of Causality

Whether our statement was justified depends on what is meant by "cause" and the distinction between a cause and the cause. Reference to the cause is appropriate when considering the mechanism by which a disease is produced, when a specific biological change may meet the philosophical requirement that it is both necessary and sufficient.The term is also used appropriately in reference to agents that are not sufficient but that satisfy Koch's postulates.These postulates, one of which requires that the agent should be found in all cases of the disease, were very helpful at the time that new discoveries were being made about bacteria and the origin of infectious disease, but were subsequently taught as essential requirements for a conclusion about causality, even though they were often quite inappropriate. I can remember, for example, being told, when a medical student, that smoking could not cause Buerger's disease because my teacher had seen a case in a non-smoker, and the same objection was occasionally raised some 15 years later to the idea that smoking might cause lung cancer. In both examples, the critics had confused the cause with a cause and had not appreciated what is now obvious: that many diseases have several causes, in the sense that we now give the word—that is, agents that are associated with a disease in such a way that increased exposure to them is followed by an increased risk of the disease and decreased exposure by a decreased risk. ¹ That asbestos is a cause of lung cancer in this practical sense is incontrovertible, but we can never say that asbestos was responsible for the production of the disease in a particular patient, as there are many other etiologically significant agents to which the individual may have been exposed, and we can speak only of the extent to which the risk of the disease was increased by the extent of his or her exposure.

Whether epidemiology alone can, in...