Establishing causality for population health and health policy


Over much of the last two decades, health research has been dominated by randomised controlled trials (RCTs). However, many interventions which target the social determinants of health are often difficult or even impossible to randomise, for political, ethical and pragmatic reasons. Critics have therefore expressed concern that privileging the RCT as the study design of choice may mean that the social determinants of health are side-lined within research. A diverse range of other approaches exist which attempt to establish causality in non-randomised studies. In contrast to the control researchers exert over the receipt of the exposure (e.g. treatment versus control), natural experiment studies try to gain causal understanding from variation in the exposure that has not been controlled by the researcher. Genetic variation across a population also provides an opportunity to understand the causal effect of exposures through Mendelian randomisation studies. The importance of clearly specifying the nature of the causal effect will be emphasised. Different approaches will be illustrated using examples drawn from evaluations of welfare policies in the UK and Brazil, a planned evaluation of minimum unit pricing of alcohol in Scotland and the use of Mendelian randomisation to understand whether alcohol really does reduce the risk of heart disease.