The patient with kidney failure who requires hemodialysis needs a connection to the machine from his vascular system. This may be a catheter or an arteriovenous fistula, which is a surgical connection between an artery and a vein in the arm to generate sufficient flow. As the catheter can frequently generate complications associated with infections or vascular damage, the fistula is the ideal option. However, the fistula can present problems due to deficit or excess flow. Stenosis (narrowings in the vein) cause loss of flow and thrombosis, with important consequences due to the need for catheters, and high flows overload the heart. It is necessary, therefore, to be able to predict the function of the fistula, although today there are no methods available that allow this, explains José Ibeas, head of the Clinical, Interventional and Computational Nephrology group at the Institut d'Investigació i Innovació Parc. Taulí, from Sabadell.
The Ibeas group is working on a project that aims to generate a clinical decision support system based on machine learning. Through the massive use of data from various sources - clinical, biometric, analytical or image data - a personalized prediction could be made for each patient, on the one hand, of the risk of fistula failure and, on the other, of its potential repercussion cardiac.
Based on information from a hundred patients, the team has developed a prototype with promising results: accuracy of 0.82 and precision of 0.86. The objective is to improve the training of artificial intelligence algorithms with more patients and to develop a tool that is prepared for clinical trials and the relevant regulatory procedures. And, later, if the results are optimal, Ibeas advances, think about creating a spin-off.
Transparency statement: This research is funded by the "la Caixa" Foundation, an entity that supports the Big Vang scientific information channel.
