The abstract is read at the publisher; the summary is JClub's.
TrialOutlook, using GPT-5 on standardized TrialDossier evidence, predicts whether 694 ongoing clinical trials meet prespecified primary endpoints with AUC 0.819 (95% CI 0.708–0.914).
- Why it matters: Trial outcome prediction depends on integrating biological and clinical evidence for a particular intervention, population, and study design, but it is unclear whether public evidence can support reliable forecasts.
- What they did: The team built TrialDossier and used GPT-5 to estimate endpoint-meeting probabilities for 694 trials, then evaluated forecasts against adjudicated outcomes disclosed by 20 October 2026.
- The result: The model achieved average precision 0.860, correctly flagged 27/29 failures (93.1%), and misclassified 17/40 successes as failures, enabling a public benchmark for archived evidence with fixed forecasts.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.