Protocol Complexity and Trial Failure: Predictive Modeling for Early Feasibility Assessment in Drug and Device Development

Background: Unplanned discontinuation of clinical trials can delay medical product development, waste resources, and reduce confidence in evidence generation. Many terminations result from design or feasibility issues that could be addressed earlier in the planning process. Purpose: To determine whether structured protocol features available at trial registration can predict early trial discontinuation across drug and device studies. Methods: We analyzed 40,677 interventional trials registered on ClinicalTrials.gov between 2015 and 2025. Using structured protocol metadata, we applied Random Survival Forests and penalized Cox regression to model time to discontinuation. Concordance indices ($C$-index) were used to assess model performance, with subgroup analyses by sponsor type, intervention class, and design features. Results: The Random Survival Forest model outperformed other approaches, identifying clear predictors of early discontinuation. Trials with longer eligibility criteria, more exclusion conditions, and higher site counts showed increased risk of failure. Results were consistent across both drug and device categories. Conclusion: Structured protocol data can support early feasibility screening by identifying trials at greater risk of early termination. This approach may assist sponsors and regulatory stakeholders in improving trial planning and reducing preventable failures.