Integrating Artificial Intelligence with Urinary Biomarkers for Early Prediction and Prognostication of Lupus Nephritis Flares

Abstract

Lupus nephritis flares significantly contribute to morbidity and mortality in systemic lupus erythematosus patients, necessitating improved predictive tools for early intervention. This study developed and validated an artificial intelligence framework integrating urinary biomarkers for early prediction and prognostication of lupus nephritis flares. A prospective cohort of 108 biopsy-proven lupus nephritis patients was enrolled between January 2021 and December 2022. Monthly urine samples were collected to measure neutrophil gelatinase-associated lipocalin, monocyte chemoattractant protein-1, tumor necrosis factor-like weak inducer of apoptosis, and vascular cell adhesion molecule-1. Machine learning algorithms including random forest, extreme gradient boosting, and logistic regression were developed incorporating temporal biomarker dynamics and clinical variables. Model performance was evaluated through time-dependent receiver operating characteristic curves and decision curve analysis. During the 12-month follow-up period, 38 patients experienced renal flares, representing 35.2% of the cohort. The XGBoost-based integrated model achieved superior predictive performance with area under the curve values of 0.82 at 30 days, 0.79 at 60 days, and 0.76 at 90 days before flare onset, substantially outperforming individual biomarkers and conventional combined approaches. The model provided a median lead time of 42 days for flare prediction, compared to 18 days using traditional biomarker assessment. Risk stratification successfully categorized patients into three groups with flare rates of 78.6% for high-risk, 35.7% for intermediate-risk, and 7.9% for low-risk patients. This integrated approach enables personalized risk assessment and early intervention strategies, potentially transforming reactive management into proactive care for lupus nephritis patients. The framework offers a non-invasive, clinically applicable tool for optimizing resource allocation while improving patient outcomes.