Bayesian-based survival analysis: inferring time to death in host-pathogen interactions
Abstract
© 2019, Springer Science+Business Media, LLC, part of Springer Nature. The standard approach to modeling survival times, or more generally, time to event data, is often based on parametric assumptions that may not fit the data collected well. One of the goals of this article is to discuss and compare several commonly used parametric and non-parametric, as well as a Bayesian semi-parametric method for survival data. We do so in the context of the data from an experimental system where insect herbivores become infected when consuming a lethal virus along with the plant on which the virus resides. We used data collected on individual insects that were fed known doses of virus along with varying genotypes of a single plant species (soybean), to compare how the insect’s diet affects its time to death. Through hazard characterization and model selection, we found that the flexible semi-parametric analysis is better at describing the time-to-death data while maintaining a relatively parsimonious form. Unlike the standard parametric and non-parametric approaches, the Bayesian semi-parametric approach better captured the rapid decline in the hazard function after a window of time where the host was most vulnerable to the virus. For our study system, being able to accurately model time to death and quantify how plant genetics affects within-insect disease processes allows us to gain a better understanding of the host-pathogen interaction at an individual level. While we show the appropriateness of the Bayesian semi-parametric approach for infection data, this method readily applies to data sets concerned with characterizing a time until any event.