Potential clinical utility of a KDIGO 2021-based clinical decision support system for glomerular diseases: a retrospective two-center pilot study.

Abstract

Glomerulopathies are a heterogeneous group of diseases characterized by complex immunopathogenesis and wide clinical variability. A wide range of laboratory and histopathological tests is used to assess glomerular diseases. However, the volume of data is often too great to analyze and to provide a clear diagnosis. The rapid development of artificial intelligence (AI) methodologies, including machine learning (ML) and deep learning, facilitates the integration of histopathological, clinical, and molecular data. This foundation enables the creation of more precise and objective diagnostic models. Modular expert systems based on ML algorithms (XGBoost, Random Forest, and natural language processing were implemented. A retrospective comparative analysis was performed comparing physician decisions with algorithmic recommendations in a patient cohort from the Opole University Clinical Hospital and the Wroclaw Medical University Hospital (2018-2025), comprising idiopathic membranous glomerulonephritis (iMN, n = 56), focal segmental glomerulosclerosis (FSGS, n = 127), lupus nephritis (LN, n = 31), and minimal change disease (MCD, n = 187). The iMN module achieved 100% accuracy in risk categorization. Retrospective analysis revealed significant gaps in clinical practice: 13.4% of decisions were suboptimal in the FSGS group, 25.8% of diagnoses were irregular in the LN cohort, and 10.8% of diagnoses were irregular in the MCD group. The integration of AI into clinical practice is a key element in the evolution of "precision nephrology," supporting the detection of subtle disease patterns and the standardization of management.