FlashMed

Chronic renal failure (CRF) patients experience a 30% higher risk of cardiovascular (CV) death compared to the general population. In this study data of 3581 incident hemodialysis (HD) patients were considered, i.e. patients who started for the first time the HD treatment. In this work supervised SOM were used with an innovative strategy to built a predictive model to estimate the probability that incident CRF patients experience a CV event in the second semester after a semester of HD treatment. A feature selection approach based on the minimum redundancy maximum relevance (mRMR) algorithm, was wrapped on the self-organizing maps (SOMs) model and a subset of 17 physiological variables with higher performance capability than the complete set of 39 variables was identified. AUC of the ROC curve of the shrunk model was 67±4%. The obtained model permits to investigate non-linear relationships among features related to an increased CV risk condition.