Improving the external validity of randomized clinical trials: The interesting place of weighting survey methods

Abstract Background: Randomized clinical trials aim to estimate the average treatment effect by randomly allocating the treatment to patients. However, their results are considered poorly applicable to real-world patients due, among other reasons, to the restriction of patient eligibility. We aimed to show that the reweighting of trial individuals to match the target population, a technique commonly used in surveys, allows unbiased estimates of the treatment effect on the target population to be provided. Methods: We first conducted a simulation study to assess whether such a weighting can provide some valid treatment effect estimation for the target population. We then used trial and registry or real-world data in COVID-19 and chronic lymphocytic leukemia patients as two illustrations. Results: The results of the simulations showed unbiased estimates of the treatment effect in the target population, regardless of the differences in the trial and target populations, the treatment effect, the potential interaction with patient characteristics, and the sample size. When applied to the trial settings, the estimated effect of treatment differed according to the severity of the target population condition, although calibrated estimates all fell into the confidence interval of the trial. Conclusions: This approach appears promising for extrapolating trial results to larger populations that are more representative of the real world. It could be of particular interest when very elderly or frail patients are not included in trials evaluating a new treatment, even though they are a significant part of the target population. Trial registration: COVIDICUS trial (NCT04344730);