Prognostic prediction of idiopathic membranous nephropathy using interpretable machine learning

Area of Science:

• Nephrology
• Medical Informatics
• Machine Learning

Background:

• Established prognostic models for idiopathic membranous nephropathy (IMN) lacked comprehensive data integration.
• Traditional methods did not fully utilize clinical and pathological patient data.
• Electronic medical record (EMR) systems offer a rich data source for improved prognostic modeling.

Purpose of the Study:

• To develop a machine learning (ML)-based risk prediction model for IMN prognosis.
• To leverage comprehensive clinical and pathological data from EMRs.
• To improve the accuracy and personalization of IMN patient management.

Main Methods:

• Utilized data from 418 IMN patients diagnosed via renal biopsy.
• Extracted 59 medical features from EMRs.
• Developed and compared five ML algorithms, evaluating performance using AUC, recall, accuracy, and F1 scores.
• Employed Shapley additive explanation (SHAP) for model interpretability.

Main Results:

• The LightGBM model demonstrated superior performance with the highest AUC (0.892), accuracy (0.909), recall (0.741), and F1 score (0.905).
• Key predictors identified by SHAP include anti-phospholipase A2 receptor (anti-PLA2R), immunohistochemical immunoglobulin G4 (IHC IgG4), D-dimer (D-DIMER), triglyceride (TG), serum albumin (ALB), and others.
• High anti-PLA2R and low IHC IgG4 levels correlated with an increased risk of adverse events in IMN patients.

Conclusions:

• A novel ML-based risk prediction model for IMN prognosis was successfully established using clinical and pathological data.
• The LightGBM model shows promise as a tool for personalized management of IMN patients.
• Accurate identification of risk factors can guide clinical decision-making and improve patient outcomes.