Heart failure is common in patients receiving hemodialysis. A high flow arteriovenous fistula (AVF) may represent a modifiable risk factor for heart failure and death. Currently, no tools exist to assess the risk of developing a high flow AVF (>2000 mL/min). The aim of this study was to use machine learning to develop a predictive model identifying patients at risk of developing a high flow AVF and to examine the relationship between blood flow, heart failure, and death. Between 2011 and 2020, serial AVF blood flows were measured in 366 prevalent hemodialysis patients at two tertiary hospitals in Australia. Four prediction models (deep neural network, and 3 separate tree-based algorithms) employing age, first AVF flow, diabetes, and dyslipidemia were compared to predict high flow AVF development. Logistic regression was used to assess the relationship between AVF blood flow, heart failure, and death. High flow AVFs were present in 31.4% of patients. The bootstrap forest predictive model performed best in identifying those at risk of a high flow AVF (AUC 0.94, sensitivity 86%, specificity 83%). Heart failure prior to vascular access creation was identified in 10.2% of patients with an additional 24.9% of patients developing heart failure after AVF creation. Long-term mortality after access formation was 27%, with an average time to death after AVF creation of 307.5 ± 185.6 weeks. No univariable relationship using logistic regression was noted between AVF flow and incident heart failure after AVF creation or death. Age, flow at first measurement >1000 mL/min, time to highest AVF flow, and heart failure predicted death after AVF creation using a general linear model. Predictive modelling techniques can identify patients at risk of developing high flow AVF. No association was seen between AVF blood flow rate and incident heart failure after AVF creation. In those patients who died, time to highest AVF flow was the most important predictor of death after AVF creation.
Read full abstract