Protocol-Level Predictors of Clinical Trial Discontinuation: A Survival Analysis Using Structured Registry Metadata

Purpose: To evaluate whether structured protocol features can predict early trial discontinuation and support feasibility assessment during clinical planning. Methods: We analyzed 40,677 interventional trials registered on ClinicalTrials.gov (2015–2025) using survival models, including Kaplan–Meier, Cox-Ridge, Cox-Lasso, parametric AFT models, and Random Survival Forests (RSF). Protocol-level features were extracted from registry metadata, and model performance was evaluated using the concordance index (C-index) with stratified subgroup analysis. Results: RSF achieved the highest test set C-index (0.6882), outperforming Cox-Ridge (0.6335) and parametric models. The RSF risk scores highlighted trial complexity markers such as eligibility length, site count, and sponsor type as strong predictors of early termination. Subgroup evaluations showed stable RSF performance across design and regulatory strata, while Cox-Ridge exhibited reduced discrimination in FDA-regulated and crossover trials. Conclusion: Structured protocol data can be used to estimate trial termination risk before launch. RSF models offer accurate, non-parametric prediction for risk-based planning, while Cox-Ridge provides interpretable baselines. These tools may aid trial sponsors and planners in feasibility screening and early decision-making.