Using Multi-Anchors To Identify Patients Suffering From Multimorbidities

Chronic diseases, and in particular the cooccurrence of more than one chronic disease in an patient, which is known as multimorbidity, represent an increasing problem in modern society. For the individual patients the consequences are potentially serious, especially if not diagnosed and/or treated. In order to allow for an efficient allocation of health resources, and slow down the progression of these diseases, methods for predicting the health status of chronic patients have been developed. In this work, we propose a data-driven approach to identify the health status of chronically ill patients using data extracted from their electronic health records. For this purpose we take advantage of recent advances in machine learning for healthcare and use anchor learning that exploits vast amounts of unlabeled data. Moreover, in order to identify patients suffering from multimorbidities, we adapt the anchor method to a multi-anchor learning framework. The experiments show that using multi-anchor learning one can accurately identify patients who suffer from one or more chronic conditions. In fact, the performance is almost comparable to a completely supervised baseline.