Modeling of correlated cognitive function and functional disability outcomes with bounded and missing data in a...

Area of Science:

• Gerontology
• Biostatistics
• Epidemiology

Background:

• Longitudinal studies on aging often face missing data due to mortality or health decline.
• Cognitive assessments like the Mini-Mental State Examination (MMSE) have floor and ceiling effects.
• Accurate analysis requires models that address data missingness, outcome correlation, and bounded score limitations.

Purpose of the Study:

• To propose a shared parameter model for analyzing correlated cognitive and disability outcomes in older adults.
• To effectively handle missing data (MAR/MNAR) and the floor/ceiling effects of MMSE scores.
• To assess the impact of model misspecification on risk factor estimations.

Main Methods:

• Developed a shared parameter model incorporating shared random effects and Tobit distribution.
• Applied the model to data from the Chinese Longitudinal Healthy Longevity Survey (CLHLS).
• Conducted a simulation study to evaluate model performance and bias.

Main Results:

• Ignoring MMSE floor/ceiling effects in CLHLS data led to non-significant systolic blood pressure association and underestimated age association with cognitive function.
• Simulation showed a 43-fold increase in bias for the female gender-cognitive function association when floor/ceiling effects were ignored.
• Results differed based on the missing data mechanism (MAR vs. MNAR), highlighting its influence.

Conclusions:

• Proper model specification is essential for accurate longitudinal analysis of correlated, bounded outcomes with missing data.
• The proposed shared parameter model provides a robust approach for analyzing complex aging data.
• Findings emphasize the need to account for MMSE score limitations and missing data mechanisms in gerontological research.