Predicting Hypoproteinemia Among Patients Undergoing Maintenance Hemodialysis: A Development and Validation Study Based on Machine Learning Algorithms

Abstract

Maintenance hemodialysis is a common alternative therapy for patients with end-stage renal disease (ESRD). We designed a prediction model of hypoproteinemia among the patients based on machine learning algorithms. The “hypoproteinemia risk factor data” were obtained, and univariate analysis was used to screen independent risk factors as prediction variables. A total of 468 patients were recruited. The incidence of hypoproteinemia in total was 30.8%. A difference between the hypoproteinemia and non-hypoproteinemia groups was significant in 18 aspects, including age, weight, dialysis duration, and dialysis frequency. In the training set, after hyper-parameter adjustment by k-fold (k = 5) cross-validation and grid search, random forest (RF), support vector machine, and logistic regression (LR) prediction models were greater than 0.8. The RF model had the highest value (0.924). The specificities of the LR and RF models were similar (0.846 and 0.839), whereas the RF model had the best accuracy (0.924). The prediction model provided by the results of this study is likely to recognize the characteristics related to hypoproteinemia. The clinical significance of the findings is a prediction of the risk of hypoproteinemia in ESRD, thus helping risk observation to nurses and improving accurate screening, primary prevention, and early intervention.