Estimating and predicting clinical trial outcomes in China.

Abstract

Around a quarter of drug development now has a Chinese origin, yet China-specific success benchmarks and forecasting tools remain scarce. We analyze all nongeneric drug trials registered on China's national drug clinical trial registry from 2013 to 2025, benchmark phase-transition probabilities of success (PoS)-where success denotes a program advancing to the next phase or to approval-and forecast trial outcomes with machine learning using structured registry features and Chinese text. China's overall phase I-to-approval PoS is below 30% across most therapeutic areas, while phase II-to-phase III transition rates run far above global benchmarks. In out-of-sample evaluations, random forests reach areas under the curves approaching 80% and logistic regression approaches 70%, with the sponsor's track record and text-derived signals as key drivers.