Development of a machine learning model for precision prognosis of rapid kidney function decline in people with diabetes and chronic kidney disease

Aims: To develop a machine learning model for predicting rapid kidney function decline in people with type 2 diabetes (T2D) and chronic kidney disease (CKD) and to pinpoint key modifiable risk factors for targeted interventions. Methods: We conducted a retrospective cohort study on 6,924 individuals with T2D and CKD at Seoul National University Hospital. Kidney function decline was assessed using estimated glomerular filtration rate slopes. The performance of the eXtreme Gradient Boosting (XGBoost) model was evaluated through model diagnosis and time-to-event analyses. Copula simulation was conducted to stratify risk subgroups using modifiable risk factors. Results: A total of 906 (13.1 %) individuals experienced rapid kidney function decline. The XGBoost model demonstrated optimal performance (area under the receiver operating characteristic curve: 0.826). The hazard of end-stage kidney disease within eight years increased across risk quartiles, with statistically significant hazard ratios in Q3 (2.06; 95 % confidence interval [CI]: 1.29-3.29) and Q4 (10.9; 95 % CI: 7.36-16.2). Simulation analysis identified high-risk subgroups by stage A3 albuminuria and at least two of the following: haematocrit < 39.0 %, systolic blood pressure > 120 mmHg, and glycated hemoglobin A1c > 6.5 %. Conclusions: The XGBoost model, augmented by copula simulation, successfully stratified kidney prognosis in individuals with T2D and CKD.