Data Mining Performance in Identifying the Risk Factors of Early Arteriovenous Fistula Failure in Hemodialysis Patients

Abstract

Background and Objectives: Arteriovenous fistula is a popular vascular access method for surgical treatment of hemodialysis patients. The method, however, is associated with a high rate of early failure varying in the range of 20-60%. Predicting early Arteriovenous fistula failure and its risk factors can help reduce its incidence, its hospitalization rate, and associated costs. In this study, we examined performance of data mining in the prediction of early AVF failure and identification of its risk factors. Methods: The data of 193 patients who underwent homodialysis in Hasheminejad Kidney Center were explored. Eight common attributes of the patients including age, sex, hypertension level, Diabetes Mellitus state, hemoglobin level, smoking behavior, location of Arteriovenous fistula, and thrombosis state were used in the machine learning process. Two learning operators including W-Simple Cart and WJ48 tree were used in data mining process. Findings: Smoking was identified as a factor influencing the relationship between the outcome of vascular access surgery and hemoglobin level. Prediction accuracy varied within the range of 69.15-85.11%. Conclusions: According to our results smoking is a crucial risk factor for early Arteriovenous fistula failure, even at normal levels of hemoglobin. Our results provide further supports for the notion that data mining can help medical decision-making process by deciphering the complex interactions between various biological variables and translating the hidden patterns in data into detailed decision-making criteria.