Predicting accrual success for better clinical trial resource allocation

Area of Science:

• Clinical trial management
• Biomedical data science
• Predictive analytics in healthcare

Background:

• Clinical trial success is critically dependent on patient accrual.
• Approximately 55% of terminated trials fail due to insufficient patient enrollment.
• Low accrual rates lead to significant financial and societal costs.

Purpose of the Study:

• To develop and validate predictive models for clinical trial failure caused by poor accrual.
• To enable early identification of trials at risk of not meeting enrollment goals.
• To inform resource allocation and improve overall clinical trial success rates.

Main Methods:

• Construction of a large dataset from ClinicalTrials.gov (57,846 trials).
• Feature engineering using literature review and natural language processing.
• Application of advanced supervised machine learning algorithms for prediction.
• Model validation using cross-validation and prospective testing.

Main Results:

• Developed models with robust predictive performance (cross-validation AUC = 0.744, prospective AUC = 0.737).
• Demonstrated stable performance over a 10-year period.
• Improved model calibration and explored the utility of the reject option for decision support.

Conclusions:

• The study presents the first models to predict clinical trial accrual failure using a comprehensive dataset and advanced machine learning.
• These models offer valuable decision support for optimizing clinical trial planning and resource allocation.
• The developed tools can enhance the efficiency and success rates of clinical research endeavors.