Classification of primary glomerulonephritis using machine learning models: a focus on IgA nephropathy prediction

Area of Science:

• Nephrology
• Medical Informatics
• Computational Biology

Background:

• IgA nephropathy (IgAN) is the most prevalent glomerulonephritis globally, characterized by immune complex deposition.
• Current diagnosis relies on invasive renal biopsy, posing risks like bleeding and infection.
• There is a critical need for non-invasive diagnostic methods for IgAN.

Purpose of the Study:

• To develop and validate a non-invasive diagnostic model for IgA nephropathy (IgAN).
• To leverage machine learning algorithms for improved IgAN diagnosis.
• To reduce reliance on invasive renal biopsies for IgAN detection.

Main Methods:

• Retrospective study of 292 IgAN patients and 310 controls.
• Utilized 82 clinical variables; random forest (RF) regression for missing values.
• Developed and compared diagnostic models (RF, SVM, ADB, doctor judgment) using 17 key features selected by RF.

Main Results:

• The RF model achieved the highest accuracy (82.3%) and AUC (0.89) on the test set.
• Key predictors for IgAN included high urinary protein, low serum albumin, and elevated IgG levels.
• RF model outperformed SVM (AUC 0.82) and ADB (AUC 0.88).

Conclusions:

• A non-invasive diagnostic model for IgAN was successfully developed using machine learning.
• The RF-based model demonstrated superior accuracy and clinical applicability.
• ML approaches offer potential for early IgAN diagnosis and reduced invasive procedures.