Abstract
OBJECTIVE: Kidney fibrosis is a key pathological feature in the progression of chronic kidney disease (CKD), traditionally diagnosed through invasive kidney biopsy. This study aimed to develop and validate a noninvasive, multi-center predictive model incorporating machine learning (ML) for assessing kidney fibrosis severity using biochemical markers. METHODS: This multi-center retrospective study included 598 patients with kidney fibrosis from four hospitals. A training cohort of 360 patients from Shanghai Tongji Hospital was used to develop a predictive nomogram and ML model, with fibrosis severity classified as mild or moderate-to-severe based on Banff scores. Logistic regression identified key predictors, which were incorporated into a nomogram and ML model. An external validation cohort of 238 patients from three additional hospitals was used for model evaluation. RESULTS: Serum creatinine (Scr), estimated glomerular filtration rate (eGFR), parathyroid hormone (PTH), brain natriuretic peptide (BNP), and sex were identified as independent predictors of kidney fibrosis severity. The nomogram demonstrated superior discriminative ability in the training cohort (AUC: 0.89, 95% CI: 0.85-0.92) compared to eGFR (AUC: 0.83, 95% CI: 0.78-0.87) and Scr (AUC: 0.87, 95% CI: 0.83-0.91). Among ML models, the Random Forest (RF) model achieved the highest AUC (0.98). In external validation, the nomogram and RF models maintained robust performance with AUCs of 0.86 and 0.79, respectively. CONCLUSION: This study presents a validated, noninvasive, multi-center Scr-based machine learning model for assessing kidney fibrosis severity in CKD. The integration of a clinical nomogram and ML approach offers a novel, practical alternative to biopsy for dynamic fibrosis evaluation.