A Penalization Method for Improving the Parsimony of Network Meta-Analysis Models

Audrey Béliveau, PhD

Assistant Professor
Department of Statistics and Actuarial Science |
University of Waterloo

WHEN: Wednesday, September 18, 2024, from 3:30 to 4:30 p.m.
WHERE: Hybrid | 2001 McGill College Avenue, Room 1201; Zoom
NOTE: Audrey Béliveau will be presenting in-person

Abstract

Network meta-analysis (NMA) is a valuable statistical tool for combining evidence on the comparative efficacy and safety of medical treatments from multiple studies. In this work, we develop a penalization framework for NMA that penalizes all pairwise differences between treatments using a generalized fused lasso (GFL). This approach improves model parsimony, resulting in more precise estimates of treatment differences and increased statistical power. Practical advantages include: 1) no prior knowledge of the similarity of treatment effects is required, 2) treatments are only assigned separate ranks if there is sufficient evidence to suggest they are different, and 3) computing time is minimal. The novel GFL-NMA method is successfully applied to three separate real-world NMAs on diabetes, Parkinson’s disease, and depression, where the best-fitting GFL-NMA model outperformed the standard NMA model (ΔAICc > 6.5).

Speaker bio

Dr. Béliveau is an Associate Professor in the Statistics and Actuarial Science Department at the University of Waterloo. Her research interests include network meta-analysis, Bayesian modeling, capture-recapture methods, and survey sampling. Please visit: https://uwaterloo.ca/scholar/a2belive.