Composite likelihood and two-stage estimation in family studies

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Genetics

Background:

• Register-based family studies motivate advanced statistical methods.
• Understanding familial aggregation of diseases requires robust modeling of multivariate failure time data.
• Existing methods may lack flexibility in capturing complex family structures.

Purpose of the Study:

• To develop and evaluate a two-stage estimation procedure using composite likelihood for multivariate failure time data.
• To provide a flexible modeling framework for family studies, accommodating various relationship structures (e.g., siblings, parent-child).
• To assess the efficiency and applicability of the proposed method in real-world family data.

Main Methods:

• A two-stage estimation procedure is linked with a composite likelihood approach.
• Asymptotic properties of estimators are derived for parametric and semi-parametric models.
• The method is applied to family structures with exchangeable members or hierarchical relationships.

Main Results:

• The proposed composite likelihood method demonstrates efficiency compared to maximum likelihood estimation.
• The method allows for highly flexible modeling, particularly in hierarchical family structures.
• Simulations confirm the method's efficiency and performance.

Conclusions:

• The composite likelihood approach offers an efficient and flexible tool for analyzing multivariate failure time data in family studies.
• The method effectively captures familial associations, as demonstrated in a deep venous thromboembolism study.
• A stronger age at onset association was observed between siblings compared to parent-offspring relationships.