Transporting trial results to synthetic real-world populations in order to estimate real-world effectiveness of newly marketed medicines.

INTRODUCTION

Real-world effectiveness of a new treatment is relevant information for patients, healthcare professionals and payers, especially when patients encountered in routine clinical care differ significantly from those recruited in the randomised controlled trials (RCTs) that led to approval. However, obtaining effect estimates can be challenging when a new drug has only recently been marketed and real-world data (RWD) are not yet available. For new breast cancer (BC) therapies, we illustrate how RCT inferences can be transported to a target population and how a synthetic population can be generated to mimic a target population for which no RWD is yet available.

METHODS AND ANALYSIS

In our framework, we defined the data-generating process for the RCT population and the real-world (target) population with confounders, effect-modulating covariates and survival times as outcomes. First, we conducted generalisability and transportability (G&T) analyses to transport the RCT results to the simulated target population, applying the inverse probability of sampling weighting and outcome model-based estimator approach. We then used Synthea to generate a synthetic target population based on German BC survival rates and combined both approaches into a coherent strategy.

RESULTS

Effect estimates (HRs with 95% CIs) transported from the RCT to our defined target population closely matched the expected real-world effect (RCT: 0.68 (0.65; 0.71); real-world: 0.75 (0.71; 0.79); transported from RCT: 0.76 (0.71; 0.81)). BC survival rates were very similar between observed and synthetic data (prediction error in absolute survival rates: 1.62%).

CONCLUSION

Combining G&T with synthetic data may inform decision-making in situations where RWD are not (yet) available.