Quantile partially linear additive model for data with dropouts and an application to modeling cognitive decline.

Area of Science:

• Neurology
• Biostatistics
• Longitudinal Data Analysis

Background:

• Cognitive assessment is crucial for understanding Alzheimer's disease progression.
• Existing models may produce biased results due to non-ignorable dropouts in longitudinal studies.
• Modeling cognitive ability in low-performing patients requires specialized statistical approaches.

Purpose of the Study:

• To develop a robust statistical model for assessing cognitive ability in Alzheimer's patients, specifically addressing challenges posed by non-ignorable dropouts.
• To create a composite cognitive score from multiple tests for a more comprehensive measure.
• To utilize partially linear quantile regression for modeling complex relationships and non-central tendencies.

Main Methods:

• A composite cognitive score was created from ten tests within the National Alzheimer's Coordinating Center Uniform Data Set.
• A partially linear quantile regression model was employed to analyze longitudinal cognitive data.
• A weighted quantile regression estimator was developed to correct for non-ignorable dropouts, using inverse probability weighting.

Main Results:

• The proposed weighted estimator demonstrated consistency and efficiency in estimating both linear and nonlinear effects.
• The model effectively handles non-central tendencies in cognitive performance.
• The methodology provides a reliable approach for analyzing cognitive trajectories in the presence of dropouts.

Conclusions:

• The developed weighted partially linear quantile regression model offers a statistically sound method for analyzing cognitive changes in Alzheimer's disease research.
• This approach enhances the accuracy of cognitive ability modeling, particularly for underperforming patient groups.
• The findings contribute to a better understanding of Alzheimer's disease progression and the impact of missing data in clinical studies.