Hill's Considerations for Causal Inference

Austin Bradford Hill's widely cited list of considerations, from his 1965 address to the Royal Society of Medicine, presents factors to consider before inferring causation from an observed association. This list is often erroneously referred to as the ‘Bradford Hill criteria’ or ‘causal criteria,’ although any list that lacks a basis for determining whether a condition is met, or for compiling such determinations to draw an overall conclusion, does not constitute a set of criteria, and Hill warns in the address that there are no ‘hardand-fast rules of evidence’ for causation (Hill, 1965, p. 299). This widespread misinterpretation is particularly unfortunate because it distracts from lessons in Hill's address that ‘offer ways to dramatically increase the contribution of health science’—for example, systematic error is often more important than random error; decisions must always be made using the epidemiologic evidence that exists; weighing costs and benefits matters more than statistical significance (Phillips & Goodman, 2004, p. 1).

Hill sought to challenge the mistake of treating tests of statistical association as sufficient evidence of causation, suggesting that we consider factors such as [Page 494](1) strength (of association), (2) consistency (across studied populations), (3) specificity (of cause and effect), (4) temporality (cause before effect), (5) biological gradient (dose-response), (6) plausibility, (7) coherence (with other evidence of various kinds), (8) experiment (intervention changes the outcome), and (9) analogy (with known causes of disease). Such lists of considerations do nothing to relieve the epistemic conundrum—well known since Hume—that causation cannot be directly observed. But pragmatically, conclusions about causation are needed to make decisions, and tools that improve these conclusions are useful.

Unlike many who cite him, Hill recognized that causal considerations (which he called ‘viewpoints’) are sources of insight, but do not provide a basis for objective conclusions. The simple observation that there is no well-defined basis for judging whether a ‘criterion’ has been fulfilled makes clear that the list can serve only as a guide to drawing subjective conclusions. In an address that covered many aspects of making decisions based on evidence, Hill cannot be faulted for not presenting a careful thesis on causal inference, or even the failure to define his viewpoints or provide guidance for those who wish to consider them (e.g., suggesting an epistemic hierarchy among them or methods for assessing whether they have been satisfied). He does, however, share blame for continued simplistic interpretation of an unelaborated list, having included his list in textbooks with no elaboration beyond what appeared in the original paper.