Predictive evidential threshold scaling (PETS): does the actual evidence meet confirmatory standards?
Michael Branson, Novartis Pharma  *Beat Neuenschwander, Novartis Pharma AG  Satrajit Roychoudhury, BDM Oncology, Novartis Pharmaceuticals 

Keywords: evidence synthesis, heterogeneity, hierarchical model, meta-analysis, prediction

Making better use of evidence in drug development is one of the tenets of the 21st Century Cures Act. Here, we consider the case where it is not possible to collect evidence according to a confirmatory standard (e.g., two statistically significant trials), yet a formal comparison of all available evidence with such a standard is required. We propose predictive evidential threshold scaling (PETS), which compares the actual evidence with a confirmatory threshold on the predictive scale. Under PETS, the actual evidence fulfills the confirmatory standard if the probability of a favorable predicted treatment effect is higher for the actual data than for the hypothetical confirmatory data. PETS builds on hierarchical models with sensible between-trial heterogeneity assumptions for the predicted, confirmatory, and actual trials. After introducing the methodology, we will illustrate the evidential strength of recently published data.