Improving the decision to pursue a phase 3 clinical trial by adjusting for patient-specific factors in evaluating phase 2 treatment efficacy data.

Phase 2 clinical trials are undertaken to provide evidence of treatment efficacy and safety. A test statistic that accounts for individual patient risk in the patient population is proposed and applied to a phase 2 clinical trial for castrate metastatic prostate cancer. The test statistic is computed to compare, for each patient, the observed 2-year survival outcome to the predicted 2-year survival probability. A logistic regression model, developed using historical data in the same patient population, is used to adjust for patient risk in predicting the 2-year survival probability. Goodness-of-fit procedures are performed to ensure that a proper model is fit to the data. The test result is compared to the score test, the binomial exact test, and Fisher's exact test, all of which use the average 2-year survival probability in the population as the parameter of interest. The results demonstrate the benefit of risk adjustment in determining treatment efficacy in a single-arm phase 2 trial. By adjusting for patient risk, this method can provide a more precise assessment of phase 2 treatment efficacy, thereby improving the decision whether to proceed to a phase 3 clinical trial.