Development and validation of a clinical trial accrual predictive regression model at a single NCI-designated comprehensive cancer center

Background: Clinical study sites often do not achieve anticipated accrual to clinical trials, wasting critical patient, material, and human resources. The expensive and extensive process to bring a drug to approval highlights the need to streamline clinical pipeline processes. We sought to create a predictive accrual model to be used when considering clinical trial activation at the level of the individual site. Materials and Methods: This retrospective cohort study used 7 years of registry data from treatment and supportive care interventional studies at a single academic cancer center to build a negative binomial regression model with local and protocol variables known prestudy. Actual, team-predicted, and model-predicted accrual and sensitivity/specificity were calculated. Results: To build the model, 207 trials were used. Investigational drug application, disease team, number of national sites, local Institutional Review Board use, total national accrual time, accrual completed, and national enrollment goal were independently and significantly associated with accrual. Predicted accrual was 94% of actual, maintaining predictive value at multiple cutoff values. Validation included 61 trials. The model correctly predicted whether a study would accrue at least 4 subjects 75% of the time. Correlation at the category level was 44.3%, and model sensitivity and specificity are 70% and 78%, respectively. Conclusions: We identified and validated national and local key factors associated with accrual at our site. This methodology has not been previously validated broadly with the intent of trial feasibility. Model validation shows it to be an accurate and quick metric in anticipating accrual success that can be used for resource allocation.