Joint Modeling of Zero-Inflated Longitudinal Count Data and Survival Outcomes with Cure Fraction

Background Longitudinal data with excess zeros and survival data with a cured fraction are common in medical research, often exhibiting interdependence that complicates analysis. Joint modeling of these data types can address this interdependence, potentially improving estimation accuracy over separate models. This study investigates the performance of joint models compared to separate models under varying conditions of zero-inflation, cure rates, and sample sizes. Methods We analyzed 500 datasets with sample sizes of 500 and 800 as part of a simulation study that included 32 scenarios. Zero-Inflated Poisson (ZIP) and Zero-Inflated Negative Binomial (ZINB) distributions were used to generate longitudinal count data, with zero-inflation rates of 43–50% and 69–75%, respectively. A mixture cure model was used to model survival data, which included cure rates of 47–50% and 59–66%. Joint models, fitted using Bayesian techniques with JAGS in R, connected longitudinal and survival processes through shared random effects. By comparing the bias in parameter estimates between joint and separate models, the performance of the models was assessed. Results Joint models consistently outperformed separate models, showing lower bias across most parameters. ZINB-based joint models exhibited less bias than ZIP models in high zero-inflation settings due to better handling of overdispersion, while ZIP models performed better with lower zero-inflation (43–50%). Larger sample sizes (800) reduced bias compared to smaller samples (500). Lower cure rates generally improved estimation accuracy, especially in joint models. Conclusion Joint modeling of zero-inflated longitudinal count data and survival data with a cure fraction provides more accurate parameter estimates than separate models, particularly in complex scenarios with high zero-inflation and cure rates. These findings highlight the importance of joint modeling in medical research involving interdependent longitudinal and survival outcomes.