Objective To explore the predictive value of a predictive model based on bioelectrical impedance vector analysis (BIVA) and machine learning algorithm for anemia and nutritional status in maintenance hemodialysis (MHD) patients.
Methods The bioelectrical signial data of MHD patients measured by body composition monitor (BCM) and albumin (Alb), hemoglobin (Hb), low density lipoprotein cholesterol (LDL-C), total cholesterol (TC) and other blood biochemical indexes data were collected. Three prediction models were established based on BIVA and three machine learning algorithms (random forest, support vector machine and Adaboost algorithm) respectively, and the prediction efficiency of the three models on Alb, LDL-C, Hb and TC indexes was compared.
Results The results of individual correlation analysis showed that the bioelectrical indexes were significantly correlated with the nutritional indexes (Alb, LDL-C, Hb, TC) (P < 0.05 or P < 0.01). Among the three models, the model based on random forest algorithm had the best performance, and the accuracies of predicting Alb, LDL-C, Hb and TC were 0.880, 0.879, 0.904 and 0.937, respectively.
Conclusion Predictive models based on BIVA and machine learning algorithms (random forest algorithms) have high value in the assessment of anemia and nutritional status in MHD patients and can assist clinical decision making.
DownLoad: