Joint models for longitudinal and time-to-event data in a case-cohort design

Area of Science:

• Biostatistics
• Clinical Research Methodology
• Longitudinal Data Analysis

Background:

• Longitudinal studies with repeated measurements are crucial in clinical research for predicting time-to-event outcomes dynamically.
• Low event rates in studies necessitate efficient data utilization, often leading to the use of case-cohort designs.
• Standard case-cohort analysis excludes non-selected non-cases, potentially introducing bias due to case overrepresentation.

Purpose of the Study:

• To propose a novel method for analyzing case-cohort data that incorporates survival information from all patients.
• To address the challenge of missing longitudinal data in a subset of patients within a case-cohort design.
• To evaluate the performance of the proposed method compared to traditional case-cohort analysis and full cohort analysis.

Main Methods:

• A joint modeling approach is proposed to handle missing longitudinal data, treating it as a missing at random problem.
• Survival information from all patients in the cohort is included in the analysis.
• Simulations were conducted to compare the proposed method with classical case-cohort analysis and full cohort analysis.

Main Results:

• The proposed method, which includes all patient data and handles missingness appropriately, performs similarly to full cohort analysis in terms of parameter estimates and survival probability predictions.
• Classical case-cohort design analysis exhibits clear bias and poorer prediction performance.
• The method was successfully illustrated using data from the BIOMArCS biomarker study.

Conclusions:

• Including all patient survival data and appropriately handling missing longitudinal information in a case-cohort design leads to valid and robust results.
• The proposed joint modeling approach offers a superior alternative to traditional case-cohort analysis for predicting time-to-event outcomes.
• This methodology enhances the efficiency and accuracy of clinical research utilizing longitudinal data and case-cohort designs.