By incorporating eθk in the model, frailty models assume that the baseline hazard functions are proportional to one another (i.e., the hazard functions have similar shape). If instead the baseline hazard function for one HPV type increases with age while the baseline hazard function for another decreases with age this assumption would be violated. WLW, in contrast, is more flexible, in that the baseline hazard function for each infection can be entirely different from each other. In addition, as with other mixed effects models, frailty models: (i) explicitly model the correlation between multiple events and, therefore, can be more efficient than WLW (a marginal model) if the correlation is correctly specified; but (ii) tend to be more computationally intensive than marginal models, and obtaining model convergence can be problematic, especially with small datasets.
