Background: A promise of real-world data in the post-market setting is to fill knowledge gaps on the treatment effect of a newly approved product in conditions beyond those studied in clinical trials. For example, the ICH-E11 recognizes real-world data as a useful source to support extrapolation of treatment effects from adults to pediatrics. Similarly, several regulatory guidance documents issued in the past few years recognize real-world data as a useful source of information for characterizing treatment effects in underrepresented populations.
Objectives: Our main objective is to propose a practical approach that leverages clinical trial findings and real-world-data to extrapolate the treatment effect on a binary outcome using causal effects methods and statistical modeling of biological mechanisms, borrowing evidence when possible.
Methods: In practice, extrapolation from adults to pediatrics typically rely on information borrowing and outcome modeling, more frequently born out of pharmacodynamic or pharmacokinetic processes. Conversely, applications of causal inference methods in clinical development have focused on reproducing findings from existing clinical trials using similar designs in real-world data or emulating novel clinical trials rather than transporting the findings and borrowing evidence. We propose a new practical approach combining the best in both worlds. Our approach leverages clinical trial findings and real-world-data to extrapolate the treatment effect, borrowing evidence when possible. We also discuss relevant scenarios and metrics in this exercise, including the degree of treatment effect heterogeneity as well as the degree of similarity in prognostic and predictive factors across data sources. We apply our proposed extrapolation approach with simulated clinical trial and real-world data under several plausible scenarios.
Results: The proposed approach is applicable in many plausible scenarios and can provide insights to the feasibility of using a real-world-data source to fill a knowledge gap, and reliability of those results.
Conclusions: The use of methods such as causal inference and modeling can extrapolate clinical trial findings using real-world data. The approach proposed in this paper can inform feasibility discussions of whether real-world data is fit-for-purpose to extrapolate treatment effect in new or underrepresented populations and when feasible, support the reliability of extrapolation findings in clinical development
