The scientific method is built on the principle that nothing can ever be proved as definitively true. Rather, once a hypothesis is proposed, evidence can be generated in favor of the hypothesis or in favor of an alternative hypothesis. When enough evidence is gathered in one direction or the other, the original hypothesis is either accepted or debunked in favor of an alternative. As scientific work is always in flux, any previously accepted theory can always be overturned by new evidence.1, 2 Many epidemiologists accept Popper’s thesis that causality can never be truly proven; although, once enough reliable evidence has been accumulated, a causal relationship can be inferred.2
The question of what constitutes a cause is a matter of ongoing inquiry among epidemiologists. Causality is extremely complex and has been described with a number of metaphors, images and guidelines, and has been summarized simply by Susser as “something that makes a difference”.2 A primary objective in epidemiology is to make inferences identifying a causal variable for the outcome of interest. However, these inferences can only be valid if the accumulation of evidence is done within a causal framework, rather than an associational one. A critical description of the difference between associational and causal concepts insists that while an associational relationship can be defined by the distribution of observed variables, a causal relationship cannot be.3 This paper will use the lens of Susser’s three characteristics of a cause, association, time order, and direction to argue that a causal relationship between exposure and outcome can only be inferred from experimental epidemiologic studies. The basis for this argument is an examination of the counterfactual model and an analysis of both observational and experimental study designs and their contributions to epidemiologists’ ability to infer causality.
As true causation in epidemiology can never be observed or definitively proven as in an ideal counterfactual scenario, the use of the best possible proxy of the counterfactual is essential to the validation of inferred causal relationships.4, 5 By using the counterfactual theory in experimental study design, the researcher seeks to mimic as closely as possible two parallel universes in which the same individuals are both exposed and not exposed within the same time dimensions.2, 4 ,5 The potency of the randomized controlled trial (RCT) is that participants can be selected with the concept of full comparability4 of the counterfactual model in mind before they are randomly separated into exposed and unexposed groups. These two components, full or close comparability and the random assignment of exposure, make the randomized controlled trial the closest to the counterfactual ideal and most capable of supporting a causal inference.
This susceptibility to bias makes observational study designs further from the ideal counterfactual scenario and threatens the validity of...