Development and Validation of a Machine Learning Model for Differentiating Diabetic Kidney Disease and Non-Diabetic Kidney Disease in Type 2 Diabetes

Overview

This multicenter retrospective observational study aims to develop and validate an interpretable machine learning model for differentiating diabetic kidney disease (DKD) from non-diabetic kidney disease (NDKD) in patients with type 2 diabetes mellitus. Clinical, laboratory, and pathological data from biopsy-confirmed patients were collected from 14 medical centers in China. Multiple machine learning algorithms were evaluated and externally validated. The final model was implemented as a web-based clinical decision support tool.

This study retrospectively collected clinical and pathological data from adult patients with type 2 diabetes who underwent kidney biopsy between January 2019 and December 2022 at 14 medical centers in China. Patients were classified as having diabetic kidney disease (DKD), non-diabetic kidney disease (NDKD), or mixed pathology according to kidney biopsy findings. Demographic characteristics, diabetic complications, laboratory measurements, and renal function parameters were extracted from electronic medical records. Six machine learning algorithms were trained and compared for discriminating DKD from NDKD. Recursive feature elimination was used for feature selection. The best-performing model was externally validated using an independent cohort enrolled between January 2022 and December 2024. Model interpretability was assessed using SHapley Additive exPlanations (SHAP). The primary objective was to develop a noninvasive and interpretable diagnostic model capable of distinguishing DKD from NDKD using routinely available clinical variables.

Conditions