Biostatistics Non-Thesis Presentations
Biostatistics Non-Thesis Presentations
Where: Hybrid Event | 2001 ÎÛÎÛ²ÝÝ®ÊÓƵ College, Room 1140;
Ìý
Presentation #1
Title: Optimal Strategies for Interim Analysis Scheduling in Bayesian Adaptive Clinical Trials: A Simulation Study
Lily Chafetz is a second year Biostatistics Master’s student from Montreal. She received her Bachelor of Arts degree from ÎÛÎÛ²ÝÝ®ÊÓƵ University, where she majored in Economics and minored in Statistics. During and following her undergraduate career, Lily fulfilled roles in various professional fields, including finance and healthcare. Her recent research project, which was supervised by Dr. Shirin Golchi and Dr. Alexandra Schmidt, focuses on interim analyses in Bayesian adaptive clinical trials.
Abstract
A crucial component of Bayesian adaptive designs for clinical trials are interim analyses (IA’s), which allow for opportunities to stop a trial early if sufficient evidence exists to declare efficacy or futility. In this project, we evaluate two strategies for the timing of IA’s through the trial: (1) event-based, i.e., according to the expected number of events, and (2) sample size-based, i.e., based on predetermined interim sample size(s). Through simulations, the optimality of these two approaches is assessed by comparing the design operating characteristics (power, false positive rates) when the event/hazard rate in the control arm is misspecified. Our results confirm that for binary outcomes modelled as binomial experiments, the event-based strategy is more efficient since the number of events is in fact the sufficient statistic for the parameter of interest, i.e., the probability of event. Similar conclusions are drawn for time-to-event outcomes in some scenarios, but not overall, since other parameters such as the censoring/dropout rates should be considered in deciding the best strategy for scheduling the IA's.
Presentation #2
Title: Zero-state Markov Switching Count Models for Chikungunya Spread in Rio de Janeiro
Mingchi Xu is a master's student in biostatistics in the Department of Epidemiology, Biostatistics and Occupational Health at ÎÛÎÛ²ÝÝ®ÊÓƵ University. His research interests include Markov-Switching Models, Spatial Epidemiology and Bayesian Inference.
Abstract
In epidemiological studies, zero-inflated models and hurdle models are commonly used to handle excess zeros in reported infectious disease cases. However, they can not model the reemergence and persistence of a disease separately. Therefore, we propose a zero-state Markov switching Poisson hurdle model, based on the recently proposed zero-state Markov switching Poisson model, to accommodate this issue. To compare the model fits, we apply and compare the chikungunya cases in Rio de Janeiro by the Poisson, Poisson hurdle, zero-state Markov switching Poisson and our proposed model. We find a zero-state Markov switching Poisson model that fits the best by WAIC criteria. We also compare the 4-step-ahead prediction
Ìý