Design-phase prediction of potential cancer clinical trial accrual success using a research data mart

Background: Many cancer interventional clinical trials are not completed because the required number of eligible patients are not enrolled.

Objective: To assess the value of using a research data mart (RDM) during the design of cancer clinical trials as a predictor of potential patient accrual, so that less trials fail to meet enrollment requirements.

Materials and methods: The eligibility criteria for 90 interventional cancer trials were translated into i2b2 RDM queries and cohort sizes obtained for the 2 years prior to the trial initiation. These RDM cohort numbers were compared to the trial accrual requirements, generating predictions of accrual success. These predictions were then compared to the actual accrual performance to evaluate the ability of this methodology to predict the trials' likelihood of enrolling sufficient patients.

Results: Our methodology predicted successful accrual (specificity) with 0.969 (=31/32 trials) accuracy (95% CI 0.908 to 1) and predicted failed accrual (sensitivity) with 0.397 (=23/58 trials) accuracy (95% CI 0.271 to 0.522). The positive predictive value, or precision rate, is 0.958 (=23/24) (95% CI 0.878 to 1).

Discussion: A prediction of 'failed accrual' by this methodology is very reliable, whereas a prediction of accrual success is less so, as causes of accrual failure other than an insufficient eligible patient pool are not considered.

Conclusions: The application of this methodology to cancer clinical design would significantly improve cancer clinical research by reducing the costly efforts expended initiating trials that predictably will fail to meet accrual requirements.