Title: Probability of success for regulatory decision-making in a pediatric indication
Authors: Aiesha Zia - Novartis Pharma (Switzerland) [presenting]
Simon Wandel - Novartis Pharma (Switzerland)
Nathalie Fretault - Novartis Pharma (France)
Shantha Rao - Novartis Pharma (United States)
Abstract: Two formulations of a marketed drug are investigated in a pediatric study. The primary objective is to evaluate treatment efficacy in the respective population of interest. Recruitment is particularly challenging, imposing a risk to meet regulatory requirements. An early analysis based on partially complete data was prepared for strategic discussion. This strategy involved probability of success (PoS) calculation to enrich the tools available for decision-making. The primary endpoint was a continuous longitudinal endpoint assessed at a specific time point. Several multivariate models, including a Bayesian approach with covariance matrix under different assumptions, to predict the yet unobserved data were used for PoS calculations. All models were fitted using SAS proc MI and proc MCMC. The PoS from general models properly reflects uncertainty due to missing values. Difficulties in fitting models were encountered due to sparse data, these challenges could be addressed by the models with more structural constraints. While PoS is a useful metric for decision-making, special considerations need to be given when data are sparse. The framework is particularly relevant to estimate PoS in the context of longitudinal partially complete data.