Integrating genetic data in target trial emulations improves their design and informs the value of polygenic scores for prognostic and predictive enrichment

Randomized controlled trials (RCTs) are the gold standard for evaluating the efficacy and safety of medical interventions but ethical, practical, and financial limitations often necessitate decisions based on observational data. The increasing volume of such data has prompted regulatory bodies to rely more on real-world evidence, primarily obtained through trial emulations. This study explores how genetic data can improve the design of both emulated and traditional trials. We successfully emulated four major cardiometabolic RCTs within FinnGen (N=425 483) and showed how reduced differences in polygenic scores (PGS) between trial arms track improved study design and consequently reduced residual confounding. Complementing these results with simulations, we show that PGS cannot be directly used to adjust for residual or unmeasured confounding. Instead, we propose an approach that uses genetic instruments for confounding detection and apply this approach to identify likely confounders in Empareg trial emulation. Finally, our results suggest that trial emulations can inform the practical application of PGS in RCTs, potentially improving statistical power. Such prognostic enrichment strategies need to be assessed in a trial-relevant population, and we show that, for 2 out of 4 emulated trials, the association between PGS and trial outcomes in the general population was different from what observed in the population included in the trial.

In conclusion, our work shows that genetic information can improve the design of emulated trials. These results contribute to the establishment of a promising new era of genetically-informed clinical trials.