Early Prediction of Clinical Trial Outcomes Using Generative Data Augmentation: A Retrospective Simulation Study (The NAEC – Nantes Anticipation of Clinical Trials – Project)
Objectifs
To evaluate whether Avatar-based generative data augmentation method can reliably reproduce final treatment effect estimates in a survival clinical trial using partially recruited cohorts. To identify the minimum recruitment threshold associated with performance comparable to that observed at 90% recruitment.
Conclusion
In this retrospective simulation study, Avatar-based generative data augmentation progressively improved the agreement between interim and final treatment effect estimates as the proportion of observed data increased. Performance appeared to stabilize from approximately 50% patient enrollment onward, suggesting that reliable outcome estimation may be achievable well before full recruitment. These preliminary results suggest that synthetic data generation may help strengthen interim analyses, support earlier go/no-go decisions and trial decision-making before full enrollment is achieved. Although validation across additional datasets and therapeutic areas is ongoing, this approach offers promising new opportunities to accelerate clinical research while maximizing the value of partially recruited studies.