Joint Modeling of Longitudinal Data and Potentially Informative Visiting Process to Dynamically Predict Future Visit...

Area of Science:

• Biostatistics
• Epidemiology
• Health Services Research

Background:

• Frequent patient visits are crucial for monitoring disease progression and treatment efficacy in cohort studies, particularly for individuals with HIV.
• Extended intervals between visits in HIV cohorts are linked to negative outcomes, such as interruptions in antiretroviral therapy.
• Predicting future visit times can proactively identify patients at risk of disengaging from care.

Purpose of the Study:

• To propose a novel joint model for analyzing longitudinal continuous markers and the patient visiting process.
• To develop a dynamic prediction tool for estimating the time to the next patient visit, considering individual patient history.
• To address the gap in dynamic prediction methodologies for visit processes within joint models of longitudinal and survival data.

Main Methods:

• Employed a gap time mechanism for the visiting process and a linear mixed model for the longitudinal marker.
• Linked the longitudinal marker and visiting process submodels using correlated, normally distributed random effects.
• Utilized a joint model framework enabling uni-dimensional integration for marginal likelihood calculation, regardless of random effect dimensions.

Main Results:

• The proposed model successfully derives dynamic predictions for the time to the next visit, incorporating previous gap times and marker history.
• Simulations demonstrated the robust performance of the developed methodology.
• The model was effectively applied to real-world data from the Athens Multicenter AIDS Cohort Study.

Conclusions:

• The developed joint model provides a powerful tool for dynamic prediction of patient visit times in longitudinal studies.
• This methodology can aid in identifying patients at risk of disengagement, facilitating timely interventions.
• The approach offers advancements in analyzing complex longitudinal data with informative visiting processes, particularly relevant for HIV research.