How to run trials better when patient outcomes stay hidden

Clinical trials often can't wait for all patient outcomes to appear—some drop out or survive beyond the study window. This paper solves that by developing an adaptive allocation policy that learns which patient groups to prioritize based on uncertainty in both event timing and dropout patterns. The authors derived the theoretically optimal allocation rule and built ASE, a framework that assigns patients sequentially while estimating survival effects. It works with any machine learning model for nuisance parameters and comes with formal asymptotic guarantees.