Abstract
Latent class methods can be used to identify unobserved subgroups which differ in their observed data. Researchers are often interested in outcomes for the identified subgroups and in some disciplines time-to-event outcome measures are common, e.g., overall survival in oncology. In this study Monte Carlo simulation is used to evaluate the empirical properties of latent class effect estimates on a time-to-event distal outcome using one, two and three-step approaches. Both standard and inclusive bias-corrected three-step approaches are considered. One-step latent class effect estimates are shown to be superior to the evaluated alternatives. Both the two-step approach and a standard three-step approach, where subjects are partially assigned to latent classes, produced unbiased estimates with nominal confidence interval coverage when latent classes were well separated, but not otherwise.
Keywords: latent class analysis, time-to-event, two-step, joint modeling
Acknowledgments
We would like to thank two reviewers for their helpful comments on an earlier draft. We would also like to thank Professor John Neoptolemos for permission to use the ESPAC3v2 data set and Dr. Ian Smith for help and assistance with use of the high throughput Condor system at the University of Liverpool (http://condor.liv.ac.uk/).