Influence factors of serum sodium and prediction of hyponatremia using back propagation artificial neural network model (BP-ANN) model in cirrhosis patients

Hyponatremia has long been considered related to high risk of hepatorenal syndrome and hepatic encephalopathy. By using artificial neural network(ANN), we investigated the influencing factors of serum sodium and established a model for predicting hyponatremia in cirrhosis patients A total of 424 cirrhosis patients (302 males and 122 females) were recruited. Correlation between serum sodium levels and clinical parameters include age, gender, serum ALT, AST, TBIL, LDL, TG, GLU, BUN, CRE, serum potassium, chlorine, albumin, platelet, cholinesterase and Child-Pugh score were analyzed. These indexes were input into the BP-ANN model as the input layer, serum sodium levels was set as the output layer. Results showed that serum sodium were positively related to serum chlorine, serum potassium, albumin, cholinesterase, LDL, age, ALT and TG, negatively related to Child-pugh score, CRE, platelet, fasting glucose, TBIL, gender, AST and BUN according to the above order. A BP-ANN model was established with the software Matlab to predict hyponatremia. In conclusions: potassium supplement is of great significance for the prevention of hyponatremia. Albumin supplementation may also helps to improve the level of serum sodium to a certain extent. Patients with higher CHILD score and serum creatinine level should be paid attention to the prevention of hyponatremia occurrence. BP-ANN model has clinical value with respect to prediction of hyponatremia based on routinely available clinical and laboratory data in cirrhosis patients.