Novel Scalar-on-matrix Regression for Unbalanced Feature Matrices

Area of Science:

• Nephrology
• Biomedical Imaging
• Statistical Modeling

Background:

• Kidney tubule image features from biopsies can serve as novel biomarkers for disease prognosis.
• Existing scalar-on-matrix regression methods struggle with the variable number of tubules across subjects.
• Accurate analysis of these features is crucial for understanding kidney disease progression.

Purpose of the Study:

• To introduce CLUstering Structured laSSO (CLUSSO), a novel scalar-on-matrix regression technique designed to predict scalar outcomes from kidney biopsy image feature matrices.
• To address the challenge of unbalanced numbers of tubules across subjects in kidney biopsy data analysis.
• To develop a robust method for identifying image features that are predictive of kidney disease outcomes.

Main Methods:

• Proposed the CLUstering Structured laSSO (CLUSSO) technique, a novel scalar-on-matrix regression method.
• Employed a clustering approach to classify tubules into distinct groups, enabling within-subject and within-cluster averaging and weighting of feature values.
• Developed theoretical properties for error bounds of feature coefficient estimates in large tubule samples.

Main Results:

• Simulation studies showed CLUSSO achieved lower false positive rates and higher true positive rates compared to a naive averaging method.
• CLUSSO demonstrated reduced bias and competitive predictive accuracy for kidney function outcomes.
• The method was successfully applied to kidney biopsy data from the Nephrotic Syndrome Study Network (NEPTUNE) and validated using the Cure Glomerulonephropathy (CureGN) study.

Conclusions:

• CLUSSO is an effective method for analyzing kidney biopsy image features, particularly when dealing with varying numbers of tubules.
• The technique offers improved accuracy and reliability in identifying prognostic biomarkers for kidney disease.
• CLUSSO provides a valuable tool for predicting kidney function and advancing research in nephrology.