| Objective. To investigate the predictive value of four risk models in evaluating the progression risk of end-stage renal disease among individuals with Ig A nephropathy, and to establish a new prediction model for the risk of ESRD among Ig AN patients in China.Methods. 535 patients diagnosed as Ig AN in Xin Hua Hospital(June 1998 to June 2014) were involved in the study. We reviewed the retrospective clinical baseline data of the cohort and carried out a further follow-up until the end of 2014. 356 of them with clear outcome were involved in the validation cohort. We evaluated the predictive value of each model for 2-year, 5-year and 10-year risks of ESRD. All of 535 patients were involved in the modeling cohort. Univariate and multivariate analysis were applied to identify independent risk factors. We utilized Kaplan-Meier survival analysis and Cox proportional hazard regression to establish the new model. The predictive value of model was assessed by discrimination and calibration. The discrimination was assessed by the area under receiver operating characteristic curve and p<0.05 indicated statistically significant. The calibration of model was assessed by Hosmer-Lemeshow test and p>0.05 indicated good fitting.Results. 38 patients of validation cohort progressed to ESRD during the follow-up period. Goto model had a good performance in predicting 2-year(AUC=0.958, p<0.05) and 5-year risk(AUC=0.950, p<0.05) of ESRD, but was fair for 10-year risk(AUC=0.876, p<0.05). Utsunomiya model had a good performance for 2-year(AUC=0.923, p<0.05), 5-year(AUC=0.933, p<0.05) and 10-year risk(AUC=0.915, p<0.05) overall. The discrimination of Berthoux model was not as well as the others(2-year: AUC=0.819, p<0.05; 5-year: AUC=0.808, p<0.05; 10-year: AUC=0.780, p<0.05). Ruijin model had an excellent performance over the other three for 10-year risk(AUC=0.931, p<0.05), and the discrimination was also appealing for 2-year(AUC=0.946, p<0.05) and 5-year risk(AUC=0.939, p<0.05). All the above four models had good calibration(p>0.05).40 patients of the modeling cohort progressed to ESRD and 151 patients were lost during the follow-up period. Male, growth of age, decline of e GFR, elevation of serum creatinine, high systolic blood pressure, high diastolic blood pressure, proteinuria, anemia, hypoalbuminemia, hyperuricemia, low level of total bilirubin, low level of serum C3 and high histological grade with tubular atrophy, interstitial fibrosis, global sclerosis or cellular or fibrocellular crescent were identified to be associated with the progression to ESRD. A new predictive model was established with three variables significant in both univariate and multivariate analysis, including global sclerosis percentage(β=2.204,HR=9.062,p=0.004), cellular or fibrocellular crescent percentage(β=2.004,HR=7.422,p=0.039) and e GFR(β=-0.069,HR=0.934,p<0.001). ROC analysis revealed that the new model had a good predictive applicability for 2-year(AUC=0.964), 5-year(AUC=0.965) and 10-year risk(AUC=0.959) of ESRD. The calibration of the new model was also good enough(?2=9.905,p=0.272).Conclusion. Based on the validation cohort, Goto model had the best predictive value in predicting 2-year and 5-year risk of ESRD over the other three, while Ruijin model performed best for 10-year risk prediction. Overall, Utsunomiya model was validated to have a good predictive ability for 2-year, 5-year and 10-year risk. Berthoux model could not give an accurate prediction generally.Global sclerosis percentage, cellular or fibrocellular crescent percentage and e GFR were strong predictive factors in evaluating the progression to ESRD in Ig AN patients. The 3-variable model had an appealing risk predictive value in evaluating the renal outcome for 2 years, 5 years and 10 years with a fairly good discrimination and calibration. |
PDF Full Text Request