Multistate Joint Frailty Models for Interval Censored Data

Magdalena Muszynska-Spielauer , Vienna Institute of Demography, Austrian Academy of Sciences

Multistate survival models (MSMs) capture transitions through multiple health states over time (age). These models typically rely on the Markov assumption, which implies that transition intensities depend only on the current state and observed covariates. In practice, this assumption is frequently violated because transition hazards depends on unobserved heterogeneity, sojourn time, or unobserved transitions in the intervals between interviews or observations, i.e. interval censoring. This paper extends MSMs for interval-censored data by introducing a joint frailty model that captures unobserved heterogeneity across several transitions simultaneously. Whereas existing joint frailty models have primarily been used to link dependence between two transitions to a health event and death, the proposed joint frailty models generalize this concept to multiple transitions within a unified and computationally efficient structure. The model accommodates different baseline hazard specifications (e.g., Gompertz or Weibull) and alternative frailty distributions (commonly Gamma or Normal). By allowing transition-specific loadings of the same frailty term, the joint frailty model provides a flexible solution to represent teh effect of latent characteristics on multiple health transitions. A preliminary empirical application using data on German women from SHARE illustrates the feasibility and empirical relevance of the approach. The results confirm that including a frailty term improves model fit and mitigates bias caused by violations of the Markov assumption. The proposed models offer a theoretically coherent and computationally tractable extension of multistate models for interval-censored data.

See extended abstract

 Presented in Session 51. Health, Morbidity and Wellbeing