A Dynamic Bayesian Network Predictive Model with Stochastic Serial Structures in Detecting Safety Signals for Hematologic Responses in Cancer Patients

• Activity Number: 501
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Jagannath Ghosh*+ and D. Das Purkayastha
• Companies: Novartis and Novartis
• Keywords: Predictive model ; Bayesian network model ; Recursive conditional probability ; Stochastic serial structure

Abstract

In clinical trials, safety monitoring plays an important role in evaluating the disease conditions over time in a patient population detecting relevant safety signals. The primary objective of this paper is to develop a dynamic predictive model capturing interdependent behavior among hematologic abnormalities within a Bayesian network with a stochastic serial structure. Thus, recursive conditional probabilities of relevant safety signals with other predictors will be estimated based on a fitted stochastic model. In an empirical case study, using safety monitoring data of cancer patients, most frequently occurring hematologic safety indicators such as erythopenea, thrombocytopenia, and leukopenia with their stochastic variability, being affected by different pathophysiological conditions, will be considered. The predicted probabilistic pattern of different safety signals for pre- and post- treatment periods will be determined using different variations of the model.