BAYESIAN MODELING LONGITUDINAL DYADIC DATA WITH NONIGNORABLE DROPOUT, WITH APPLICATION TO A BREAST CANCER STUDY

Area of Science:

• Social and Behavioral Sciences
• Biostatistics
• Longitudinal Data Analysis

Background:

• Dyadic data, common in social sciences, feature correlated members due to interdependence.
• Analyzing longitudinal dyadic data is complicated by nonignorable dropouts (missing data).

Purpose of the Study:

• To propose a Bayesian selection-model-based approach for longitudinal dyadic data with nonignorable dropouts.
• To develop a statistical model that accounts for within-dyad correlations and complex missing data mechanisms.

Main Methods:

• A transition model was used for repeated measures, incorporating random effects for within-dyad correlations.
• A selection model was employed to address nonignorable missing data, where dropout probability depends on the missing outcome.
• A Gibbs sampler algorithm was developed for model fitting.

Main Results:

• Simulation studies demonstrated the proposed method's effectiveness in handling nonignorable dropouts.
• The methodology was illustrated using a real-world longitudinal breast cancer study.

Conclusions:

• The proposed Bayesian selection-model approach provides a robust framework for analyzing longitudinal dyadic data with nonignorable dropouts.
• This method enhances the accuracy of statistical inference in studies with complex interdependence and missing data patterns.