Predicting Probability of Success for Phase III Trials via Propensity-Score-Based External Data Borrowing

Given the rising costs and time length of confirmatory phase III trials, drug developers have become increasingly reliant on quantitative methods to support critical decisions such as whether drug should continue development after completing a phase II study. One such method that is commonly used is to estimate the probability of success (PoS) of a phase III trial. PoS is computed by averaging the traditional power function over a prior distribution for the unknown treatment effect, which is often estimated using observed phase II data. However, phase II trials are often small due to budgetary, logistical, or ethical considerations, which can increase the variability of phase II results and provide misleading PoS calculations. In this paper, we develop a new PoS framework that leverages external data sources to increase the understanding of the phase II study data and hence the accuracy of PoS calculations. To mitigate the risk of bias associated with external data borrowing, we augment the control arm of the phase II study using the propensity-score-based MAP (PS-MAP) prior, which allow to objectively incorporate patient-level information. We demonstrate via simulation and an example application in non-small cell lung cancer that incorporation of external data into the traditional PoS framework can enable more robust decision-making in clinical development by engendering larger values, on average, when the treatment under study is truly effective and smaller values, on average, when the treatment under study is truly ineffective.