Multi expert integrated algorithm for kidney biopsy triage

Clinical decision-making often exhibits substantial inter-physician variability when evaluating identical patient data, limiting the reliability of conventional one data-one outcome clinical decision support systems. We developed and validated a Multi Expert Integrated Algorithm (MEIA) designed to preserve and integrate diverse expert decision patterns for kidney biopsy triage. The study included 9598 patients across three cohorts, comprising a developmental cohort of 8228 patients and two external validation cohorts. Three board-certified nephrologists independently annotated biopsy decisions, and expert-specific machine learning models were trained using identical feature sets to replicate each physician's labeling pattern. These models were integrated through a predefined majority voting framework. Individual models closely reproduced expert decisions in internal validation, while MEIA demonstrated strong performance (accuracy 95.3%, F1-score 84.4%). In external validation, MEIA achieved an AUC of 0.933, with significantly higher discrimination than Expert model C (P< 0.001) and comparable performance to Expert models A and B. SHAP analysis revealed heterogeneity in feature importance across experts. In a pathology-confirmed cohort, all MEIA-recommended cases demonstrated histopathological abnormalities. MEIA provides a structured framework for modeling expert variability; prospective validation is required to confirm clinical utility.