Machine Learning Approach to HHD Management

Abstract

We examined whether a machine learning-based analysis of home hemodialysis (HHD) problems contributes to improving the efficiency of operations and the quality of responses. We analyzed 424 problems that could be handled by telephone during the period from November 2011 to July 2022. Regarding the machine learning environment, we used Windows as the as the operating system (OS), with Python 3.7.12 on Google Colaboratory. The features used were personal identification numbers (IDs), the occurrence of an alarm, the situation when a problem occurs, and the dialyzer mode at the onset of a problem. The decision coefficient was classified in binary terms, based on whether the problem was caused by a human error, such as an error when connecting to the blood circuit, or other problems, choosing from a total of 24 patterns of problems. The results were showed an accuracy of 0.9124, a recall of 0.9191, a precision of 0.9255, an F1 score of 0.9195, and an area under the curve (AUC) of 0.9639. This time, we constructed a model that classifies problems into those caused by human errors and those caused by other factors, based on the situation when a problem occurs, etc. The test data demonstrated that we were able to construct a highly accurate model with an overall accuracy of 0.9449. It is effective to use machine learning for HHD management.