Routinely collected data may usefully supplement randomised controlled data on treatment effects for mortality

Hemkens and colleagues investigated the effects of treatment on mortality in routinely collected data and subsequent randomised trials.1 They found that observational studies with routinely collected data substantially overestimated the effects of treatment, despite the use of propensity scores to address confounding bias. This overestimation might, however, be partially due to the selection of studies. In randomised controlled trials, a pitfall is choosing a control that reflects the realities of medical care, and a major …