ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005


The Latest on X

Kidney Week

Abstract: TH-PO029

Artificial Intelligence Prediction of Dialysis Access Complications

Session Information

Category: Augmented Intelligence, Digital Health, and Data Science

  • 300 Augmented Intelligence, Digital Health, and Data Science


  • Gaweda, Adam E., University of Louisville Health Sciences Center, Louisville, Kentucky, United States
  • Alsauskas, Zygimantas C., University of Louisville Health Sciences Center, Louisville, Kentucky, United States
  • Brier, Michael E., University of Louisville Health Sciences Center, Louisville, Kentucky, United States

The single most important factor determining quality of dialysis is adequate access. This begins with the placement of an access that is optimized for the individual patient and may include a fistula, graft, or in some cases a catheter. Constant monitoring of the fistula or graft is needed to avoid future complications and with the EMR we can provide this in real time using the dialysis machine / EMR interface. We evaluated an AI approach to prediction venous stenosis at different locations with sufficient warning that an intervention can be performed prior to interruption of dialysis.


Data were obtained from the University of Louisville dialysis electronic medical record system and the Interventional Nephrology program for the years 2018 to 2020. Routinely collected information at the time of initiation as well as intradialytic data were obtained. Events were determined by direct examination within the interventional facility and segregated into inflow, outflow, central, and other stenosis. Event classification was performed using sequential vector of predictors consisting of target and achieved blood flow rate, arterial pressure, arterial resistance, venous pressure, venous resistance, and mean arterial pressure. Resistance was defined as the ratio of pressure to blood flow rate. Events were predicted 1 week in advance. Classification was performed using tree ensembles (AdaBoost) and Artificial Neural Network.


Data set contained 2369 samples with 27 inflow, 133 outflow, 54 central, and 59 other stenosis events. Classifiers were trained using 10-fold cross-validation. Additionally, 10% of the data were held out for testing. Table 1 shows comparison of the classifier performance on the test data between the two methods.


Sufficient information exists during the dialysis procedure that can be leveraged using advanced mathematical techniques to adequately predict dialysis access complications. The methodology applied also allows for the explanation of the observations seen and can provide predictions through a mathematical model or an expert system in the form of a decision tree.

True Positive Rate10010066.498.710010078.7100
True Negative Rate73.697.477.189.242.997.485.392.2
False Positive Rate26.42.622.910.857.12.614.77.8
False Negative Rate0033.61.30021.30