An accessible overview of epidemiology

Interpreting epidemiologic evidence: connecting research to applications. 2nd edition. David A Savitz, Gregory A Wellenius. Oxford University Press, 2016 (240 pp, £32.99). ISBN 9780190243777

After years of cautious epidemiology, feeling forced to limit our inferences to associations, and always hedging them with caveats, the work on causal inference over the past 25 years has been a breakthrough. It is refreshing to finally admit that, as epidemiologists, we have always been interested in causality, and the development of directed acyclic graphs (DAGs) has been instrumental in renewing our confidence in making inferences about cause and effect. DAGs are a formalised way of drawing the relationships between an exposure, an outcome and other variables of interest, allowing us to tease out what may be directly or indirectly causal, and what may be confounding or biasing.

It is therefore heartening to see a textbook that has truly integrated DAGs throughout its discussion of causality, bias and confounding, rather than leaving it as an isolated chapter or appendix. Because of its roots in formal logic and artificial intelligence, the early work on DAGs has often been dense and impenetrable. By contrast, Interpreting epidemiologic evidence is a highly readable book, written in an engaging conversational tone, and just the right length to not be intimidating. As such, it is amenable to reading cover to cover by the interested clinician. It is easily the most accessible introduction to DAGs that I have come across and is the perfect introduction for beginners in this area.

The book also has excellent discussions of selection bias and measurement bias. Rather than providing dry and theoretical frameworks, the authors have distilled years of expertise into a kind of epidemiological common sense, a way of asking questions and approaching design that almost intuitively highlights potential errors. There is also a brief and lucid chapter on P values and confidence intervals, which should be required reading for all clinicians.

The authors convey all this information in a way that is virtually devoid of mathematical equations; for those who are maths phobic, there is nothing here to scare you off. There is an almost guilty pleasure in getting all the insights of an experienced epidemiologist without any of the hard maths to get there.