Development Of The Risk Prediction Model Of Kidney Injury For Elderly Patients With T2DM Based On RWD

Objective: To explore the risk factors of kidney injury in elderly patients with Type2 Diabetes Mellitus(T2DM),and to establish models for predicting the risk of kidney injury based on Real World Data(RWD).Methods: Data were collected from elderly T2 DM patients at 10 medical centers,including Sichuan Provincial People’s Hospital.The data were standardized through cleaning procedures to prepare them for further analysis.K-Means clustering was used to classify patients into different risk categories for kidney injury,and the resulting clusters were included as predictor variables.Random Forest(RF)was used for filling missing values,Borderline Synthetic Minority Over Sampling Technique(Borderline SMOTE)was used for data sampling,and Lasso regression was used for variable screening.Seven machine learning methods were used to establish prediction models for renal injury in elderly T2 DM patients after 6-and 12-month intervals.The performance of these models was evaluated using Area Under Curve(AUC),accuracy,precision,recall,and F1 score.Calibration curves were used to evaluate the consistency of the models,and Shapley Additive Explanation(SHAP)was used to identify the contribution of each variable to the predicted value.Decision Curve Analysis(DCA)was used to access the clinical utility of the model.Finally,the sample size was verified to ensure validity.Results: 1.The study established two datasets for predicting kidney injury in elderly patients with type 2 diabetes mellitus(T2DM): Dataset 1 for predicting kidney injury after 6 months and dataset 2 for predicting kidney injury after 12 months.Dataset 1included a total of 1818 patients and 90 variables,while dataset 2 included 1001 patients and 90 variables.2.The rates of kidney injury after 6 months across different risk layers were 0.0%,11.9%,15.8% and 100.0%,and the rates of kidney injury after 12 months were 0.0%,13.4%,17.8% and 100.0%.The variables associated with a high risk of kidney injury include: multiple comorbidities,concurrent use of multiple drugs,a decrease in estimated glomerular filtration rate(eGFR)and endogenous creatinine clearance rate,and an increase in urea and serum creatinine(SCr).3.The optimal model for predicting the risk of kidney injury after 6 months is Categorical Boosting(CB)with an AUC of 0.9505,while the optimal model for predicting the risk of kidney injury after 12 months is also CB with an AUC of 0.9297.4.SHAP results revealed that the most important variables in the prediction model of kidney injury after 6 months were eGFR,diabetes complications,clustering results,cardiovascular disease,and the number of cardiovascular diseases,while the most important variables for predicting kidney injury after 12 months were eGFR,lactate dehydrogenase,diabetes complications,aspartate aminotransferase,and red blood cell count.Conclusion: This study established a risk stratification method and prediction models for kidney injury after 6 months and 12 months in elderly patients with T2 DM applying CB algorithm.The models showed strong predictive performance and provided a useful theoretical foundation for the early identification and treatment of kidney injury in this patient population.