Predicting Success of Phase III Trials in Oncology

ImportanceWe developed a model predicting the probability of success (PoS) for single planned or ongoing PhIII trials based on information available at trial initiation. Such a model is highly relevant for study sponsors to capture risk and opportunity on a trial-to-trial basis through trial optimization, and for investors to select drugs whose trial design match their investment strategy. ObjectivesTo predict the outcome of planned or ongoing PhIII trials in oncology, given publicly available prior information Design, Setting, ParticipantsPredictive modeling using publicly available data for 360 completed PhIII and 1240 PhII studies initiated between 2003 and 2012. Success and failure of PhIII studies were modeled using Bayesian logistic regression model. Main Outcome MeasuresPredicted PoS of individual PhIII trials based on a Bayesian model calibrated on publicly available data translated into 16 composite scores. Those scores cover aspects such as trial design, indication, number of patients, phase II (PhII) study outcomes, experience of sponsor at time of trial initiation, and others. ResultsThe model allows to calculate the PoS distribution - including credible intervals - for a PhIII trial in oncology. The predictive performance was determined using an area under the receiver-operator curve (AUROC), resulting in an overall performance of 73%oPP (mean AUROC). We identified two key factors contributing to the predictive performance of the model: quality and strength of PhII data and experience of the sponsor at the time of study initiation. Conclusion and RelevanceWe describe the generation and application of a statistical model predicting the PoS for individual PhIII trials in oncological indications with unprecedented predictive performance. Compared to other approaches, this is the first study generating a fully transparent model resulting in trial-specific PoS distributions. Moreover, we have shown that qualitative concepts such as PhII knowledge or sponsor R&D strength can be captured in quantitative scores and that these scores have a high predictive power. Key PointsO_ST_ABSQuestionC_ST_ABSWhat is the probability of success (PoS) for single phase III (PhII)I trials in oncology? FindingsWe developed a model allowing to predict the PoS of single PhIII trials in oncology with a predictive performance of 73%PP and demonstrated that qualitative factors such as strength of PhII knowledge and sponsor R&D strength can be captured in quantitative scores that have significant predictive power. MeaningThe model can help study sponsors to analyze and amend planned clinical trials, and investors to choose where to invest best.