Comparing treatment effects between propensity scores and randomized controlled trials: improving conduct and reporting

This editorial refers to ‘Do observational studies using propensity score methods agree with randomized trials? A systematic comparison of studies on acute coronary syndrome’†, by I.J. Dahabreh et al. , on page 1893 Evaluating the effectiveness of a therapeutic intervention is ideally carried out in the setting of a randomized controlled trial (RCT). Patients are randomly allocated to the experimental and control groups ensuring that observed, pre-treatment key prognostic characteristics, but also unobserved patient characteristics, are balanced between the treatment groups, minimizing the variability in patient characteristics. Providing a sufficient number of patients have been randomized, this balance in observed and unobserved pre-treatment characteristics between the groups enables unbiased conclusions about the treatment effect to be drawn. There may, however, be instances where randomization is not possible due to, for example, ethical reasons (e.g. emergency surgery,1 transplantation2) or because it is impractical (e.g. rare events, financial reasons). When it is not possible to conduct an adequately powered RCT, observational studies are often carried out to examine and infer treatment effects. In addition, treatment effects observed in RCTs that involve highly selective populations are often examined in different patient populations and settings in observational studies. However, in observational studies, investigators have no control over treatment assignment, which is often part of a patient's routine medical care. In these instances, it is likely that potentially large systematic differences (typically confounding by indication) in observed patient characteristics could lead to large, biased, and ultimately misleading estimates of treatment effect. Propensity scores are increasingly being used to reduce the impact …