OP.LB01.09] PREDICTION OF CV EVENT IN DIABETIC NEPHROPATHY

Abstract

Objective: Prediction of cardiovascular events in type 2 diabetes nephropathyDesign and method: The data come from the multicentre prospective observational ALICE -Protect Study. Between January 2010 and February 2011, 153 nephrologists recruited 986 adults with type 2 diabetes nephropathy defined by eGFR> 15 mL/min/1.73m2 and significant proteinuria. Twenty six variables related to the characteristics of the subject, treatment, history, biological characteristics were recorded.The prediction model used a Bayesian network (Tree augmented naive algorithm) calibrated with 2000 simulated individual data (1000 with a CV event and 1000 without CV event) generated from the ALICE database. Results: The used original database included data from the 695 subjects who had a final visit. A CV event (n = 186 including 26 deaths) occurred in 26.8 % of the subjects. The population was composed of 73.8 % men, had a mean age of 70 ± 10 years, a mean eGFR (MDRD) of 40.0 ± 20.3 mL/min/1,73m2. The 10 fold cross validation of the generated model had an AUC under the ROC curve of 70.8 %. The generated model had an AUC under the ROC curve of 71.6% % using the original 695 dataset. Sensibility reached 67.2% specificity 65.0%, negative predictive value was 84.4%, and positive predictive value 41.3%. The individual 2 years CV risk can be computed online at https://www.hed.cc/?s=cvevent&t=CV%20Event Conclusions: In this population at a very high risk of CV event, Bayesian models can compute the two year individual CV risk to optimize the management diabetic nephropathy. The properties of the model must be confirmed in a prospective study that is under consideration.