Biostatistics Seminar Series

Peter X.K. Song, University of Michigan

Thursday 10/4 3:30PM - 4:30PM
Public Health Auditorium (G23)
HASS: Hybrid Algorithm for Subgroup Search via ADMM and EM Algorithms

Identification of subgroups in a biomedical study with subjects sampled from a heterogeneous population has attracted considerable attention in recent years. Technically, subgroup group analysis may be formulated as a type of supervised clustering analysis with group labels being latent. The method of mixture model is the most widely used approach in which the Expectation-Maximization (EM) algorithm plays a central role in handling related optimization.  A well-known issue with the EM algorithm is its strong reliance on initial values and possible poor convergence, especially when mixture components are not very separable. We propose a fusion learning approach to overcoming this drawback, where a pairwise fusion penalty is utilized to automatically detect and identify homogeneous subgroups as initial features.  This fusion approach is implemented by an Alternating Direction Method of Multipliers (ADMM) algorithm, which, however, only provides subgroup centroids with no ability to reconstruct the underlying individual effect sizes.  Our proposed method, termed as Hybrid Algorithm for Subgroup Search (HASS), to blend computational speed and numerical stability with interpretability and reproducibility in supervised subgroup analysis.  We also establish key theoretic properties for the proposed HASS procedure. It is further illustrated by extensive simulation studies and analysis of a real example.


Last Updated On Friday, September 7, 2018 by Tang, Lu
Created On Friday, August 24, 2018