Abstract
Background: The differential diagnosis of diabetic kidney disease (DKD) from non-diabetic kidney disease (NDKD) presents significant challenges in clinical practice, as current diagnostic methods, such as renal biopsy, are invasive and lack specificity. This study aims to develop a non-invasive predictive model based on clinical and biochemical indicators to enhance diagnostic accuracy in distinguishing DKD from NDKD. The model is designed to serve as a decision-support tool for clinicians, improving renal health management in patients with type 2 diabetes mellitus (T2DM). Methods: A retrospective examination of data was executed. Clinical characteristics and laboratory data of T2DM patients who underwent renal biopsy at The First Affiliated Hospital of Sun Yat-sen University, spanning January 2015 to September 2023, were collated and stratified into a training cohort (January 2020 to September 2023) and an internal validation cohort (January 2015 to December 2019). A distinct analysis was conducted for patients with renal transplants within the validation cohort. Partial case data from Shenzhen Third Hospital (January 2022 to July 2025) and Southern Hospital (January 2018 to December 2023) were used as external validation cohort. The training cohort data underwent both univariate and multivariate regression analyses to formulate a predictive probability model, which was subsequently subjected to validation against the validation cohort. The efficacy of the model was meticulously assessed through metrics such as the area under the ROC curve, calibration plots, DAC, and the Hosmer-Lemeshow goodness-of-fit test. Results: The study encompassed a total of 1091 T2DM patients, including 385 with DKD, 585 with NDKD, and 121 with a concomitant diagnosis of DKD and NDKD, denoted as MIX. Membranous nephropathy was identified as the predominant pathological entity in both NDKD and MIX cases. The probability model incorporated six variables: gender, age, diabetes duration, diabetic retinopathy status, serum uric acid, and low-density lipoprotein levels. The model demonstrated robust discrimination and calibration capabilities for patients without renal transplants but exhibited diminished applicability for those with renal transplants. Conclusion: The research successfully established a model capable of accurately forecasting the likelihood of NDKD in the renal biopsy findings of T2DM patients. However, the model's applicability to patients with renal transplants is constrained, suggesting that future research endeavors should focus on enhancing the model to encompass a more diverse patient demographic.