Answers are grounded in this paper and cite their sources.
Medical cost data are often difficult to model because they are highly skewed, contain many zeros, and in longitudinal settings involve repeated measurements with potential censoring and informative dropout, introducing additional complexity due to within-subject correlation and time-varying covariate effects. Marginalized two-part joint models (MTJMs) offer a flexible framework for analyzing longitudinal medical cost data alongside survival outcomes by directly modeling the population mean, ensuring that regression effects correspond to the marginal mean. Existing MTJMs typically assume a log-normal (LN) distribution with constant variance, which may be restrictive in practice. This study extends the classical MTJM by incorporating a generalized gamma (GG) distribution, which encompasses the LN model as a special case and provides greater flexibility, while also broadening the applicability of existing GG-based marginalized two-part models to a longitudinal-survival joint modeling framework accommodating informative dropout. Simulation studies assess the finite-sample performance of the proposed model under correctly specified and misspecified medical cost distributions, comparing MTJM-LN, MTJM-W (Weibull), MTJM-G (gamma), and MTJM-GG using bias, standard error, and 95% coverage rate. The method is illustrated using electronic health records from Iran, where the GG-based MTJM outperforms the LN-based approach, and multimorbidity and proximity to death emerge as major cost drivers.
No takes yet. Share an insight, caveat, or question.
Shahrokhabadi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: