The use of patient-specific modelling in the assessment of a clinical indicator for arteriovenous fistula failure

Abstract

The arteriovenous fistula (AVF) is a surgically-made vascular structure connecting an artery to a vein. It is the optimal form of vascular access for haemodialysis-dependent end-stage renal disease patients. However, AVF are prone to access dysfunction through the formation of stenoses, which compromise the structure’s utility. To date, a plethora of clinical models are used to predict AVF formation failure based on patient factors and other models predicting late AVF failure by assessing haemodynamics and quantifying disturbed flow behaviours and wall shear stress metrics with stenosis formation. That said, inconsistencies were identified in the correlation between these metrics and diseased AVFs. This thesis aims to assess the suitability of another haemodynamic-related metric, resistance, derived from pressure drop and flow rates through patient-specific CFD modelling, for diagnosing and predicting AVF failure. A three-dimensional ultrasound scanning system was used to obtain patient-specific geometry and flow profiles, used for CFD models which were then analysed, with resistance calculated for each patient. The significance of patient-specific CFD modelling was demonstrated in its usefulness to generate a patient-targeted indicator of diseased AVF. To study the effectiveness of resistance as a metric, the relationship between CFD-derived resistance and the potential for AVF failure was evaluated, starting with classification of resistance results among patients who had undergone treatment for stenosis. An exploratory study into the suitability of CFD-derived resistance and its association with patients’ AVF conditions was further conducted by classifying data from a larger patient dataset and fitting the classified data to a multilevel regression model. CFD-derived resistance was found to be higher at the proximal vein of problematic AVF, however this figure was 76% lower among patients who had undergone stenosis treatment. Meanwhile, no correlation was found between resistance at the proximal artery and patency status. An area under curve of 92.1% was found from the receiver operating characteristic analysis, noting an outstanding discrimination of the classification. CFD-derived resistance appears to be a promising metric in the assessment of a suitable diagnostic marker for AVF failure. This research concludes with aspirations for clinical implementation of a related system, alongside routine surveillance of AVF.