Introduction: A good prediction model plays an important role in determining the progression to diabetic kidney disease. We aimed to create a model to predict progression to kidney failure in patients with diabetic kidney disease.Methods: We retrospectively assessed 641 patients with type 2 diabetic kidney disease as derivation cohort and 280 patients as external out time validation cohort. We used a combination of clinical guidance and univariate logistic regression to select the relevant variables. We calculated the discrimination and calibration of different models. The best model was selected according to the optimal combination of discrimination and calibration.Results: During the 3 years follow up, there were 272 outcomes (42%) in derivation cohort and 138 outcomes (49%) in external validation cohort. The final variables selected in the multivariate logistics regression were age, gender, hemoglobin, NLR, serum cystatin C, eGFR, 24-h urine protein, and the use of oral hypoglycemic drugs. We developed four different models as clinical, laboratory, lab-medication, and full models according to these independent risk factors. Laboratory model performed well in both discrimination and calibration among all the models (C-statistics: external validation 0.863; p value of the Hosmer-Lemeshow, .817). There was no significant difference in NRI among laboratory model, lab-medication model, and full model (p > .05). So, we chose the laboratory model as the optimal model.Conclusion: We constructed a nomogram which contained hemoglobin, NLR, serum cystatin C, eGFR, and 24-h urine protein to predict the risk of patients with diabetic kidney disease initiating renal replacement in 3 years.
Keywords: Diabetic kidney disease; predictive model; progression; renal replacement.