Regularized approach for data missing not at random.

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Missing data are prevalent in longitudinal studies due to various reasons like dropouts or missed visits.
• Handling data missing not at random (MNAR) is crucial for valid analysis but often faces identifiability and computational challenges.
• Existing models for MNAR data can struggle with estimation and convergence.

Purpose of the Study:

• To propose novel LASSO and ridge-regularized selection models to effectively handle MNAR data in longitudinal studies.
• To overcome the identifiability and computational issues associated with traditional MNAR models.
• To provide a robust framework for sensitivity analysis regarding missing data mechanisms.

Main Methods:

• Development of LASSO and ridge-regularized selection models to regularize the missing data mechanism model.
• Utilizing a cross-validation procedure for selecting the regularization parameter.
• Application of the proposed models to simulation studies and a real-world randomized clinical trial dataset.

Main Results:

• The proposed regularized models effectively handle MNAR data, improving estimation and computational convergence.
• Simulation studies demonstrated the performance and robustness of the developed methods.
• Analysis of a randomized clinical trial confirmed the practical utility of the models.

Conclusions:

• LASSO and ridge-regularized models offer a viable solution for analyzing longitudinal data with MNAR mechanisms.
• These models enhance the reliability of statistical inference in the presence of complex missing data patterns.
• The proposed approach facilitates robust sensitivity analyses for missing data assumptions.