Shared parameter modeling of longitudinal data allowing for possibly informative visiting process and terminal event.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Survival Analysis

Background:

• Shared parameter models (SPMs) jointly analyze longitudinal and time-to-event data but often assume non-informative visiting processes.
• Informative visiting processes, where observation times depend on patient history, can bias SPM estimates.
• Current methods typically assume conditional independence between marker, visiting, and event processes.

Purpose of the Study:

• To develop a unified, flexible approach for jointly modeling longitudinal markers, informative visiting processes, and competing risks time-to-event data.
• To assess the impact of informative visiting processes on parameter estimation in joint models.
• To evaluate the robustness of different visiting process formulations (gap time vs. calendar time).

Main Methods:

• A novel joint model is proposed, conditioning on marker history, visit times, and random effects.
• The model accommodates both gap time and calendar time scales for the visiting process.
• Competing risks are incorporated into the time-to-event component.
• Extensive simulation studies are conducted to evaluate performance.

Main Results:

• Disregarding an informative visiting process leads to significantly biased marker estimates.
• Misspecification of the visiting process also introduces bias.
• The gap time formulation demonstrates greater robustness to misspecification than the intensity-based model.
• Including prior visit history in the visiting process model improves estimation accuracy.

Conclusions:

• Informative visiting processes are critical and should be explicitly modeled in joint analyses.
• The proposed unified approach provides a flexible and robust framework for joint modeling.
• Accurate modeling of the visiting process is essential for reliable inference in longitudinal and survival data analysis.
• The methodology is successfully applied to HIV longitudinal data with variable visit frequencies.