CSDAresub.pdf (377.7 kB)
Faster Monte Carlo estimation of joint models for time-to-event and multivariate longitudinal data
journal contributionposted on 2020-06-11, 12:43 authored by P Philipson, GL Hickey, MJ Crowther, R Kolamunnage-Dona
Quasi-Monte Carlo (QMC) methods using quasi-random sequences, as opposed to pseudo-random samples, are proposed for use in the joint modelling of time-to-event and multivariate longitudinal data. The QMC integration framework extends the Monte Carlo Expectation Maximisation approaches that are commonly adopted, namely using ordinary and antithetic variates. The motivation of QMC integration is to increase the convergence speed by using nodes that are scattered more uniformly. Through simulation, estimates and computational times are compared and this is followed with an application to a clinical dataset. There is a distinct speed advantage in using QMC methods for small sample sizes and QMC is comparable to the antithetic MC method for moderate sample sizes. The new method is available in an updated version of the R package joineRML.
This research was funded by a Medical Research Council (MRC), United Kingdom grant (MR/M013227/1) awarded to RKD, and used to fund GLH’s position. MJC is part-funded by an MRC New Investigator Research, United Kingdom Grant (MR/P015433/1). The funder had no role in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript. Results for the simulation study and bootstrap standard errors for the application were performed on the Oswald HPC cluster at Northumbria University. We thank Jimmy Gibson (HPC Technology Specialist) for his assistance in using Oswald.
CitationComputational Statistics & Data Analysis Volume 151, November 2020, 107010
- AM (Accepted Manuscript)