A “Divide-and-Conquer” AECM Algorithm for Large non-Gaussian Longitudinal Data with Irregular Follow-Ups

Presented by Reuben Retnam, Takeda Pharmaceuticals

Wednesday, September 21
4:00 p.m. ET
Virtual

Features of non-Gaussianity, manifested via skewness and heavy tails, are ubiquitous in databases generated from large scale observational studies. Yet they continue to be routinely analyzed via linear/non-linear mixed effects models under standard Gaussian assumptions for the random terms. In periodontal disease data, these issues are applicable to the modeling of clinical attachment level and pocket depth. These problems are maintained, if not exacerbated, in the longitudinal data framework.

In this research, we define and elucidate an extension of the skew-t linear mixed model suitable for a big data setting. This extensibility is achieved via the implementation of divide-and-conquer techniques that utilize the distributed expectation-maximization algorithm. Specifically, the E-steps of the AECM algorithm are run in parallel on multiple worker processes, while manager processes perform the M-steps with a updated fraction of the results from the local expectation steps. We prove convergence properties of this algorithm and show examples of its performance compared to traditional modelling methods on real and simulated data.

Speaker Bio:

Dr. Reuben Retnam is a recent graduate of Virginia Commonwealth University’s Department of Biostatistics. His research interests include longitudinal data,extensions of the EM algorithm, and extrapolation-based model acceleration. He joined Takeda Pharmaceuticals in 2022, focusing on modeling complex pre-clinical outcomes.