Estimating Disability Prevalence and Incidence in Italy Using a Multistate Model: Merging Information from Different Sources

Enrico Roma, University of Bologna, Department of Statistical Science "Paolo Fortunati"
Rossella Miglio , Department of Statistical Sciences "Paolo Fortunati", University of Bologna

Disability drives social exclusion, poor health, and premature mortality. Monitoring its dynamics and not just prevalence, is crucial in ageing societies. In Italy, two complementary data sources are available: the AVQ survey, conducted annually from 2013–2023, provides regionally representative prevalence estimates without longitudinal follow-up; the SHARE study, collected biennially between 2004–2022, follows individuals and enables estimation of transition dynamics. SHARE is interval-censored: transitions are known only to occur between interviews, except for death. We integrate these datasets to estimate age-specific disability prevalence and transitions. We model health as a continuous-time Markov process with three states (non-disabled, disabled, death), allowing recovery. Disability is measured with the Global Activity Limitation Indicator (GALI). Transition intensities are smooth, age-dependent functions (penalized splines) with sex and education entering under a proportional-hazards specification. Regional heterogeneity can be accommodated through constrained shifts on the log hazard scale. Intensities imply transition probabilities and model-based prevalences via the forward Kolmogorov equations. Band-averaged prevalences serve as parameters in a binomial model for AVQ. Parameters are estimated by maximizing a joint penalized likelihood, propagating uncertainty across sources. Results indicate transitions from non-disabled to disabled and from disabled to death rise sharply with age, while recovery declines toward zero at advanced ages. Patterns differ by gender and education: women experience disability transitions more often, men face higher mortality. Lower education is associated with greater disability risk, reduced recovery, and higher mortality. The joint model improves precision relative to single-source analyses and enables identification of regional components.

See extended abstract

 Presented in Session 51. Health, Morbidity and Wellbeing