Simulation and Estimation of Continuous-Time Multi-State Markov Models for Panel Data

A higher-level interface to continuous-time Markov multi-state models for panel (interval-censored) data. Seven canonical clinical process structures are supplied with structurally valid generator matrices, so that transition matrices and starting values need not be constructed by hand. Panel data can be simulated from exact trajectories under regular or irregular observation schedules, with optional exactly observed absorption times and optional Weibull holding times for assessing the Markov assumption. A single fitting call validates the input against the assumed structure and returns the estimated generator with confidence intervals, mean sojourn times, transition probability matrices and observed transition counts, together with the optimiser's convergence code. A Monte Carlo driver reports Monte Carlo standard errors alongside bias, root mean squared error and interval coverage. Likelihood evaluation is delegated to 'msm' (Jackson, 2011, ); the panel-data likelihood is that of Kalbfleisch and Lawless (1985) .