Estimating treatment effects in observational studies

Neurology: October 17, 201

Randomized controlled trials are required to determine the effectiveness and safety of medical therapies for clinical use. However, observational studies are used frequently to evaluate treatment effects in a number of situations, such as postmarketing safety evaluations of medications or in situations where clinical trials are not feasible.1 Unlike randomized controlled trials, where randomization promotes an equal balance of known and unknown confounders between treatment groups, treatment groups in observational studies often differ in the prevalence of key prognostic variables associated with both treatment and outcome.1 While statistical approaches, such as multivariable adjustment, propensity-based matching, and propensity score risk adjustment, can be used to adjust for the influence of known confounders in observational studies, adjusting for unmeasured confounders is more obviously challenging. In the case of propensity-based approaches, the analyses specifically take account of the factors that predict whether a patient receives the treatment or not (e.g., age, disease severity). Read more