A Machine Learning Approach Towards Early Detection of Frequent Health Care Users

Abstract

In primary health care, a small number of frequent users incur a large portion of the total health care expenditures. In this work, we study whether it is possible to recognize these frequent users early on, through the application of machine learning based text mining techniques on clinical notes. We implement our study on a data set of 147 Finnish primary health care users, using a regularized least-squares based ranking method. The method achieves a ranking accuracy of 0.68 when making predictions based on the recorded text and observed visitation frequency after 20 visitations by a patient, demonstrating that it is possible to make useful predictions about the future rate of visitations.