A prediction model of arteriovenous fistula dysfunction in hemodialysis patients was established based on a few sample synthesis oversampling technique algorithm

Abstract

Objective To explore the risk factors of arteriovenous fistula (AVF) dysfunction in hemodialysis (HD) patients and validate a risk warning model based upon synthetic minority oversampling technique (SMOTE). Methods From January 1, 2019 to December 31, 2021, 400 HD patients with AVF as dialysis access were selected as study subjects. According to AVF function of HD, they were assigned into two groups of poor AVF function (n=81) and normal AVF function (n=319). The relevant clinical data of selected HD patients were reviewed and the risk factors of poor AVF function screened by univariate and multivariate Logistic regression. Then the above risk factors were reconstructed by SMOTE algorithm for established an early warning model of poor AVF function. The prediction efficiency of two models was compared.Results Females, diabetes mellitus (DM), albumin 2 mmol/L and AVF stenosis were the risk factors for AVF dysfunction in HD patients (P<0.05). According to the above risk factors and regression coefficients, area under the ROC curve (AUC) of original warning model P1 was 0.787(95%CI: 0.743-0.831). And area under the ROC curve (AUC) of P2 warning model was 0.870(95%CI: 0.743-0.831). TPR (0.731) of early warning model based upon SMOTE algorithm was lower than original warning model (0.763) while PPV (0.742 vs 0.866) and F-score (0.729 vs 0.886) were higher than original warning model. Conclusions Females, DM, albumin <35 g/L, CRP and AVF stenosis are risk factors for poor AVF function in HD patients. And SMOTE early warning model based upon the above risk factors has higher predictive value as compared with traditional Logistic regression model.

Keywords

• synthetic minority over-sampling technique
• renal dialysis
• arteriovenous fistula
• poor function
• prediction model
• hemodialysis