SERdigital Spring 2017

A goal of epidemiology is to learn about causal relationships that affect population health. This pursuit of causal inference has led to the establishment of an armamentarium of analytical and methodological strategies, aimed at the control of confounding, to estimate the effect of causes. In particular, two approaches to causal inference have been advanced to handle confounding. First, the measurement of sufficient variables to achieve conditional exchangeability between the exposed and unexposed within levels of those variables. In practice, adjustment for those confounders takes place via stratification, matching, or modeling. Second, the identification of situations when exchangeability can be assumed because the exposure is assigned based on factors unrelated to the outcome. In practice, adjustment for all confounders is not necessary.

For this SERdigital, we aim to discuss three questions: When might we do best using the first approach approach? When might we do best using the second approach? And when might both approaches be insufficient?

We have two superbly qualified discussants for our first multidisciplinary debate. Dr. Sandra Decker, an economist with the Agency for Healthcare Research and Quality, will present an approach to causal inference in which adjustment for confounders is not needed. Dr. Miguel Hernán, an epidemiologist with the Harvard T.H. Chan School of Public Health, will present an approach to causal inference in which confounders need to be measured. Each discussant will discuss settings for which their approach is well suited and will provide some examples from their work. Dr. Jay Kaufman will help moderate a discussion about the uses of and assumptions for both approaches to causal inference.