Development and validation of a predictive model for end-stage renal disease risk in patients with diabetic...

This study was performed to develop and validate a predictive model for the risk of end-stage renal disease (ESRD) inpatients with diabetic nephropathy (DN) confirmed by renal biopsy. We conducted a retrospective study with 968 patients with T2DM who underwentrenal biopsy for the pathological confirmation of DNat the First Affiliated Hospital of Zhengzhou University from February 2012 to January 2015; the patients were followed until December 2018. The outcome was defined as a fatal or nonfatal ESRD event (peritoneal dialysis or hemodialysis for ESRD, renal transplantation, or death due to chronic renal failure or ESRD). The dataset was randomly split into development (75%) and validation (25%) cohorts. We used stepwise multivariablelogistic regression to identify baseline predictors for model development. The model's performance in the two cohorts, including discrimination and calibration, was evaluated by the C-statistic and the P value of the Hosmer-Lemeshow test. During the 3-year follow-up period, there were 225 outcome events (47.1%) during follow-up. Outcomes occurred in 187 patients (52.2%) in the derivation cohort and 38 patients (31.7%) in the validation cohort. The variables selected in the final multivariable logistic regression after backward selection were pathological grade, Log Urinary Albumin-to-creatinine ratio (Log ACR), cystatin C, estimated glomerular filtration rate (eGFR) and B-type natriuretic peptide (BNP). 4 prediction models were created in a derivation cohort of 478 patients: a clinical model that included cystatin C, eGFR, BNP, Log ACR; a clinical-pathological model and a clinical-medication model, respectively, also contained pathological grade and renin-angiotensin system blocker (RASB) use; and a full model that also contained the pathological grade, RASB use and age. Compared with the clinical model, the clinical-pathological model and the full model had better C statistics (0.865 and 0.866, respectively, vs. 0.864) in the derivation cohort and better C statistics (0.876 and 0.875, respectively, vs. 0.870) in the validation cohort. Among the four models, the clinical-pathological model had the lowest AIC of 332.53 and the best P value of 0.909 of the Hosmer-Lemeshow test. We constructed a nomogram which was a simple calculator to predict the risk ratio of progression to ESRD for patients with DN within 3 years. The clinical-pathological model using routinely available clinical measurements was shown to be accurate and validated method for predicting disease progression in patients with DN. The risk model can be used in clinical practice to improve the quality of risk management and early intervention.