Hidden multistate models to study multimorbidity trajectories

TL;DR

Continuous-time hidden multistate models provide a robust alternative to traditional approaches, supporting individualized predictions and informing targeted interventions and secondary prevention strategies for multimorbidity in aging populations.

Abstract

Multimorbidity in older adults is common, heterogeneous, and highly dynamic, and it is strongly associated with disability and increased healthcare utilization. However, existing approaches to studying multimorbidity trajectories are largely descriptive or rely on discrete-time models, which struggle to handle irregular observation intervals and right-censoring. We developed a continuous-time hidden multistate modeling framework to capture transitions among latent multimorbidity patterns while accounting for interval censoring and misclassification. A simulation study compared alternative model specifications under varying sample sizes and follow-up schemes, and the best-performing specification was applied to longitudinal data from the Swedish National study on Aging and Care-Kungsholmen (SNAC-K), including 2,716 multimorbid participants followed for up to 18 years. Simulation results showed that hidden multistate models substantially reduced bias in transition hazard estimates compared to non-hidden models, with fully time-inhomogeneous models outperforming piecewise approximations. Application to SNAC-K confirmed the feasibility and practical utility of this framework, enabling identification of risk factors for accelerated progression toward complex multimorbidity and revealing a gradient of mortality risk across patterns. Continuous-time hidden multistate models provide a robust alternative to traditional approaches, supporting individualized predictions and informing targeted interventions and secondary prevention strategies for multimorbidity in aging populations.

Figures (4)

• Figure 1: Conceptual and analytical pipeline for modeling multimorbidity trajectories. Chronic disease diagnoses are first aggregated using Latent Class Analysis to identify multimorbidity patterns (Steps 1–2). Individuals are assigned pattern membership at each visit (Step 3), and transitions across latent states—including progression to death—are modeled using a continuous‑time hidden multistate model that accounts for misclassification and interval censoring (Step 4). LCA: Latent Class Analysis; MM: Multimorbidity. Created in BioRender.
• Figure 2: Overview of the data‑generating mechanism used in the simulation study. The process integrates population composition, chronic disease development, and observation schemes, with scenarios varying by sample size (3,000 vs. 10,000) and follow‑up structure (regular population‑based intervals vs. irregular visits). Transitions among multimorbidity states and death follow a Gompertz‑based continuous‑time process.
• Figure 3: Alluvial diagram of observed transitions across multimorbidity states in the SNAC‑K cohort. Each stream represents an individual’s assigned multimorbidity pattern over time (mild: light blue; complex: dark blue), with transitions to death shown in grey. Patterns are based on posterior membership probabilities from the latent class model and illustrate increasing movement toward complex multimorbidity with age.
• Figure :