Functional Linear Regression Models for Nonignorable Missing Scalar Responses.

Area of Science:

• Biostatistics
• Medical Imaging Analysis
• Health Informatics

Background:

• Medical imaging data are crucial for modern healthcare, serving as densely sampled functional data for diagnosis and prognosis.
• Predicting clinical outcomes from imaging data is vital but complicated by missing outcome information.

Purpose of the Study:

• To propose a functional linear regression model for predicting clinical outcomes using functional (imaging) predictors.
• To address the challenge of non-ignorable missing clinical outcomes in statistical modeling.

Main Methods:

• An exponential tilting semiparametric model was introduced to handle non-ignorable missing data mechanisms.
• Estimating equations and computational methods were developed for parameter estimation and tuning parameter selection.
• A bootstrap resampling procedure was proposed for statistical inference.

Main Results:

• Asymptotic properties, including consistency and convergence rate, were established for the proposed estimates.
• Simulation studies and a real data analysis demonstrated the finite sample performance of the methods.

Conclusions:

• The proposed functional linear regression model effectively predicts clinical outcomes from medical imaging data, even with missing outcomes.
• The developed methods provide a robust framework for statistical inference and analysis in medical imaging research.