Statistical Modeling in Pre-Clinical Drug Proarrhythmic Assessment

• Activity Number: 351 - Statistical Modeling in Pre-Clinical Drug Proarrhythmic Assessment
• Type: Topic-Contributed
• Date/Time: Thursday, August 12, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Yu-yi Hsu* and Nan Xi*
• Companies: FDA/CDER and UCLA
• Keywords: Machine Learning; Ordinal random forest model; Bayesian additive regression tree; Ordinal logistics regression model; pro-arrhythmic ; TdP

Abstract

The aim of this panel is to discuss machine learning approaches in classifying pro?arrhythmic (torsade de pointes (TdP)) risks of drugs using pre-clinical data. Panelists will present ordinal logistic regression and ordinal random forest statistical models to predict drug-induced TdP risks using pre-clinical data. A Bayesian additive regression tree (BART) approach for correlated observations and decision making based on a posterior distance matric in TdP risk classification will also be presented. The panelists will discuss general principles to quantify models, metrics to be used to predict TdP risk, and when and how these novel approaches play a role in determining the proarrhythmic risk to inform clinical development.