A bidimensional finite mixture model for longitudinal data subject to dropout.

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Longitudinal studies often face challenges with incomplete data due to subject dropout.
• Nonignorable dropout mechanisms require specific statistical approaches to avoid biased results.
• Existing models may not adequately account for the complex relationship between dropout and longitudinal outcomes.

Purpose of the Study:

• To propose a novel statistical model for analyzing longitudinal data with nonignorable dropout.
• To accurately capture the dependence between the longitudinal response and the dropout process.
• To provide a flexible framework that nests ignorable dropout models within a nonignorable structure.

Main Methods:

• Modeling dependence using discrete, outcome-specific latent effects.
• Utilizing a probability matrix to define the joint distribution of latent effects.
• Introducing an index for sensitivity analysis to assess the impact of dropout assumptions.

Main Results:

• The proposed model effectively handles heterogeneity in individual response profiles.
• It allows for separate modeling of within-outcome and between-outcome dependencies.
• Demonstrated the model's utility in analyzing cognitive functioning dynamics in elderly populations.

Conclusions:

• The developed statistical approach offers a robust method for longitudinal data with nonignorable dropout.
• This model improves the accuracy of parameter estimation by accounting for dropout mechanisms.
• The findings have implications for understanding cognitive decline and developing interventions.