A sequential threshold cure model for genetic analysis of time-to-event data

In analysis of time-to-event data, classical survival models ignore the presence of potential nonsusceptible (cured) individuals, which, if present, will invalidate the inference procedures. Existence of nonsusceptible individuals is particularly relevant under challenge testing with specific pathogens, which is a common procedure in aquaculture breeding schemes. A cure model is a survival model accounting for a fraction of nonsusceptible individuals in the population. This study proposes a mixed cure model for time-to-event data, measured as sequential binary records. In a simulation study survival data were generated through 2 underlying traits: susceptibility and endurance (risk of dying per time-unit), associated with 2 sets of underlying liabilities. Despite considerable phenotypic confounding, the proposed model was largely able to distinguish the 2 traits. Furthermore, if selection is for improved susceptibility rather than endurance, the error of applying a classical survival model was nonnegligible. The difference was most pronounced for scenarios with substantial underlying genetic variation in endurance and when the 2 underlying traits were lowly genetically correlated. In the presence of nonsusceptible individuals, the method provides a novel and more accurate tool for utilization of time-to-event data, and has also been proven successful when applied to zero-inflated longitudinal binary data.