Unraveling Causality: Innovations in Epidemiologic Methods.

Abstract

For several decades, the counterfactual model and the sufficient cause model have shaped our understanding of causation in biomedical science and, more recently, the link between these two models has enabled us to obtain a deeper understanding of causality. In this article, I provide a brief overview of these fundamental causal models using a simple example. The counterfactual model focuses on one particular cause or intervention and gives an account of the various effects of that cause. By contrast, the sufficient cause model considers sets of actions, events, or states of nature which together inevitably bring about the outcome under consideration. In other words, the counterfactual framework addresses the question "what if?" while the sufficient cause framework addresses the question "why does it happen?" Although these two models are distinct and address different causal questions, they are closely related and used to elucidate the same cause-effect relationships. Importantly, the sufficient cause model makes clear that causation is a multifactorial phenomenon, and it is a "finer" model than the counterfactual model; an individual is of one and only one response type in the counterfactual framework, whereas an individual may be at risk of none, one, or several sufficient causes. Understanding the link between the two causal models can provide greater insight into causality and can facilitate the use of each model in appropriate contexts, highlighting their respective strengths. I will briefly present three topics of interest from our research: the relationship between the concepts of confounding and of covariate balance; distinctions between attributable fractions and etiologic fractions; and the identification of operating mediation and mechanism. It is important to scrutinize observed associations in a complementary manner, using both the counterfactual model and the sufficient cause model, employing both inductive and deductive reasoning. This holistic approach will better help us to unravel causality.