SLECare: Explainable Machine Learning for Kidney Damage and Remission Prediction in Malaysian Lupus Nephritis

Abstract

Lupus Nephritis is a severe renal manifestation of Systemic Lupus Erythematosus that may progress to chronic kidney disease and delayed remission, yet early patient-level risk prediction remains challenging due to heterogeneous clinical patterns. This study developed and evaluated SLECare, an explainable machine learning-based clinical decision-support framework for predicting kidney damage and remission outcomes among Malaysian Lupus Nephritis patients. Retrospective clinical records from Hospital Canselor Tuanku Muhriz, Universiti Kebangsaan Malaysia, covering demographic, clinical, laboratory, biopsy, treatment, and outcome variables were analysed. A leakage-aware machine learning workflow was applied, including clinically guided feature removal, stratified train-test splitting, training-derived preprocessing, feature engineering, Recursive Feature Elimination, and time-based modelling across baseline, 6-month, and 12-month prediction scenarios. CatBoost, Random Forest, Multilayer Perceptron, Support Vector Machine, weighted soft voting, and stacked ensemble models were compared using precision, recall, F1-score, and confusion matrix. SHAP was used to explain both cohort-level and patient-level risk drivers. CatBoost provided the most clinically interpretable deployment model, while the stacked ensemble achieved the strongest 6-month kidney damage prediction performance with an F1-score of 0.8000. For remission prediction, the 12-month CatBoost model achieved the strongest overall performance with precision of 0.8095, recall of 0.7391, and F1-score of 0.7727. SHAP identified clinically meaningful predictors including baseline creatinine, ethnicity, Lupus Anticoagulant, chronicity index, CKD status, treatment delay, and global sclerosis. SLECare further translates model outputs into a dashboard prototype for cohort overview, individual risk assessment, high-risk patient prioritization, longitudinal monitoring, and global risk-driver review. These findings suggest that SLECare can support earlier, explainable, and locally relevant risk stratification for Malaysian Lupus Nephritis care.