Bayesian analysis of longitudinal quality of life measures with informative missing data using a selection model

Area of Science:

• Biostatistics
• Health Services Research
• Clinical Trials

Background:

• Health-related quality of life (HRQoL) is crucial for evaluating treatment efficacy in clinical studies.
• Longitudinal studies assessing HRQoL are prone to missing data, potentially biasing results.
• Non-ignorable missing data, where missingness relates to HRQoL itself, poses a significant analytical challenge.

Purpose of the Study:

• To propose a novel Bayesian approach for analyzing longitudinal, multivariate outcome data with non-ignorable missingness.
• To develop a method that accurately accounts for missing data mechanisms linked to health-related quality of life.
• To provide a robust analytical tool for clinical studies evaluating treatment effects on HRQoL.

Main Methods:

• A Bayesian selection model is employed for joint likelihood factorization to address non-ignorable missing data.
• Bayesian spike and slab variable selection is utilized within the missing data mechanism to identify informative factors among multiple outcomes.
• Linear mixed-effects models with a hierarchical structure capture correlations between multiple outcomes and covariates.

Main Results:

• Simulation studies demonstrate the proposed method's superior performance compared to the conventional last observation carried forward approach.
• The Bayesian method offers efficiency gains in estimating marginal associations.
• The approach successfully identifies associations between outcomes and the reasons for missing patient information.

Conclusions:

• The proposed Bayesian method provides a robust framework for analyzing longitudinal HRQoL data with non-ignorable missingness.
• This approach enhances the validity and efficiency of inferences in clinical studies, particularly in oncology.
• The method offers valuable insights into the relationship between patient outcomes and data attrition.