Estimating the incidence of dementia from two-phase sampling with non-ignorable missing data

Area of Science:

• Psychiatry
• Epidemiology
• Biostatistics

Background:

• Two-phase sampling is common in psychiatry for rare diseases like dementia.
• Longitudinal dementia studies face challenges with missing data due to participant death, which is often non-ignorable.
• Accurate estimation of disease incidence requires adjusting for both death-related missingness and complex sampling designs.

Purpose of the Study:

• To develop and apply a statistical approach for estimating dementia incidence in longitudinal studies.
• To address the issue of non-ignorable missing data caused by participant mortality.
• To account for two-phase sampling designs in incidence estimation.

Main Methods:

• A selection model approach was used to handle data missing due to death.
• A likelihood approach was employed for deriving incidence estimates.
• A modified Expectation-Maximization (EM) algorithm was utilized for sampling selection data.
• Non-parametric jack-knife variance estimation was applied for model and incidence parameters.

Main Results:

• The proposed methods were successfully applied to the Indianapolis-Ibadan Dementia Study data.
• The study provides a framework for more accurate incidence estimation in dementia research.
• The statistical approach effectively handles missing data and complex sampling.

Conclusions:

• The developed statistical methods provide a robust way to estimate dementia incidence.
• Adjusting for missing data due to death is critical for reliable longitudinal dementia studies.
• The approach is valuable for epidemiological research on rare diseases with mortality.