Abstract:
Survival analysis is a fundamental tool in medical science for time-to-event
data. However, its application to colony organisms like bees poses challenges
due to their social nature. Traditional survival models may not accurately
capture the interdependence among individuals within a colony. Frailty models, accounting for shared risks within groups, offer a promising alternative.
This study evaluates the performance of semi-parametric shared frailty models (gamma, inverse normal, and positive stable-in comparison to the traditional Cox model using bees’ survival data). We examined the effect of misspecification of the frailty distribution on regression and heterogeneity parameters using simulation and concluded that the heterogeneity parameter was
more sensitive to misspecification of the frailty distribution and choice of initial parameters (cluster size and true heterogeneity parameter) compared to
the regression parameter. From the data, parameter estimates for covariates
were close for the four models but slightly higher for the Cox model. The
shared gamma frailty model provided a better fit to the data in comparison
with the other models. Therefore, when focusing on regression parameters,
the gamma frailty model is recommended. This research underscores the
importance of tailored survival methodologies for accurately analyzing
time-to-event data in social organisms.
