Multistate Markov model for functional recovery in stroke: probability of state transition and prognosis prediction

Stroke rehabilitation involves complex transitions between different functional states. This study developed and validated a five-state Markov model to quantify state transition probabilities and predict functional outcomes in stroke patients, providing quantitative support for clinical decision-making. We analyzed 2000 functional status observations from 1000 stroke patients (baseline and follow-up) at Tianjin Medical University General Hospital. Using modified Rankin Scale scores, patients were classified into five functional states (mild/moderate/severe disability, recovery, death). A continuous-time Markov model estimated transition intensities and probabilities, with validation through observed-versus-predicted comparisons and comprehensive sensitivity analyses. The Markov model revealed dynamic transitions among functional states: the monthly transition intensity from severe to moderate disability was the highest (3.13%), while the annual cumulative probability of full recovery was highest among patients with moderate disability (15.99%). At the 12-month follow-up, 65.4%, 59.6%, and 49.0% of patients with mild, moderate, and severe disability, respectively, remained in their original state, while 14.3%, 16.0%, and 14.5% achieved full recovery. One-way and probabilistic sensitivity analyses indicated that the model was robust to parameter variations, with narrow 95% confidence intervals. The multi-stata Markov model provided insights into the dynamic process of functional recovery after stroke in a specialized setting,offering a methodological framework that may support clinical prognosis assessment with appropriate validation. Patients with moderate disability exhibited the highest recovery potential, while those with severe disability demonstrated the fastest improvement rate, suggesting that individualized rehabilitation strategies may be considered tailored to different functional states.