Artifical Intelligence in Kidney Transplantation

UNSTRUCTURED The Covid-19 pandemic has put new demands on the medical systems worldwide. The pressure of taking far-reaching decisions within multiply limited resources under the constraint that personal contact must be minimized has evoked the question if technical support in the form of Artificial Intelligence (AI) could help leverage these challenges. At the same time, AI comes with its own issues such as limited transparency that cannot be neglected especially in a medical context. We will deliberate this in the domain of specialized outpatient care of kidney transplant recipients. In order to improve long-term care for these patients, we implemented a telemedicine functionality monitoring vital signs, medication adherence and symptoms at Charité – Universitätsmedizin Berlin. This paper seeks to combine this established telemonitoring approach with methods from Artificial Intelligence proposing an AI-based clinical decision support system (AI-CDSS) that aims to detect Covid-19 and other severe diseases in this high-risk population. After analyzing medical needs and difficulties and suggesting possible technical solutions, we argue that AI-supported telemonitoring in outpatient care can play a valuable role in managing resources and risks in kidney transplant patients in times of Covid-19 and beyond. Additionally, regarding the multitude of ethical and legal questions arising when integrating AI into workflows, we exemplarily discuss the concept of meaningful human control and whether it is achievable with the proposed AI-CDSS.

Patient care after kidney transplantation requires integration of complex information to make informed decisions on risk constellations. Many machine learning models have been developed for detecting patient outcomes in the past years. However, performance metrics alone do not determine practical utility. Often, the actual performance of medical professionals on the given task is not known. We present a newly developed clinical decision support system (CDSS) for detection of patients at risk for rejection and death-censored graft failure. The CDSS is based on clinical routine data including 1516 kidney transplant recipients and more than 100 000 data points. Additionally, we conduct a reader study to compare the performance of the system to estimations of physicians at a nephrology department with and without the CDSS. Internal validation shows AUC-ROC scores of 0.83 for rejection, and 0.95 for graft failure. The reader study shows that although the predictions by physicians converge towards the suggestions made by the CDSS, performance in terms of AUC-ROC does not improve (0.6413 vs. 0.6314 for rejection; 0.8072 vs. 0.7778 for graft failure). Finally, the study shows that the CDSS detects partially different patients at risk compared to physicians without CDSS. This indicates that the combination of both, medical professionals and a CDSS might help detect more patients at risk for graft failure. However, the question of how to integrate such a system efficiently into clinical practice remains open.

Scientific publications about machine learning in healthcare are often about implementing novel methods and boosting the performance - at least from a computer science perspective. However, beyond such often short-lived improvements, much more needs to be taken into consideration if we want to arrive at a sustainable progress in healthcare. What does it take to actually implement such a system, make it usable for the domain expert, and possibly bring it into practical usage? Targeted at Computer Scientists, this work presents a multidisciplinary view on machine learning in medical decision support systems and covers information technology, medical, as well as ethical aspects. Along with an implemented risk prediction system in nephrology, challenges and lessons learned in a pilot project are presented.