Pattern Mixture Sensitivity Analyses via Multiple Imputations for Non-Ignorable Dropout in Joint Modeling of...

Area of Science:

• Biostatistics
• Epidemiology
• Neuroscience

Background:

• Longitudinal studies of aging often face missing data in memory assessments.
• Health-related dropout can lead to missing not at random (MNAR) longitudinal measurements.
• Joint modeling of longitudinal data and time-to-event data is crucial for understanding disease progression.

Purpose of the Study:

• To propose and implement a pattern-mixture sensitivity analysis for MNAR data within a joint modeling framework.
• To evaluate the impact of MNAR data on the association between memory decline and dementia risk.
• To assess the utility of multiple imputation for sensitivity analyses in joint modeling.

Main Methods:

• Utilized the Swedish Betula study data, combining longitudinal memory assessments and dementia onset.
• Employed a pattern-mixture sensitivity analysis using multiple imputation for MNAR longitudinal data.
• Developed imputation models allowing for accelerated memory decline post-dropout and fitted joint models to imputed datasets.

Main Results:

• Worse memory levels and steeper memory decline were consistently associated with an increased risk of dementia across all sensitivity scenarios.
• The proposed multiple imputation method proved pragmatic and accessible for evaluating MNAR data consequences.
• Pattern-mixture models effectively framed plausible MNAR assumptions for various missing data patterns.

Conclusions:

• Sensitivity analyses using multiple imputation are valuable for assessing the impact of MNAR data in joint modeling.
• The pattern-mixture approach offers flexibility in handling missing data in aging and dementia research.
• Joint modeling combined with sensitivity analysis provides robust inference for predicting dementia risk based on memory trajectories.