Enhancing knowledge of treatment effects by analysing observational data as a randomised trial.

Abstract

Randomised controlled trials are the gold standard design to obtain knowledge of causal effects of treatments. However, not all causal questions can be answered by randomised trials for practical or ethical reasons. Target trial emulation is a conceptual framework and set of analysis methods to obtain knowledge of causal effects of treatments using observational data, and is particularly valuable when randomised trials are not feasible. In this Perspective, we outline the concepts and application of target trial emulation, showing how it can be used to minimise bias due to confounding at baseline and over the course of sustained treatments, and how causal per-protocol analysis can be performed to provide informative yet unbiased estimates of treatment effects. We summarise a recent target trial emulation that examined the effects of intensity of early rehabilitation for hip fracture on performance in activities of daily living as a worked example. We encourage investigators to explore the target trial framework, and apply it to answer causal questions when using observational data.