Assessing Disease Comorbidity in Hospital Patients through Machine Learning and Network Analysis Techniques

Abstract

Accurate assessment of disease comorbidity in hospital patients has always been crucial for improving treatment strategies and healthcare outcomes. In the last few years, different machine learning (ML) and network analysis (NA) techniques have been developed to enhance the detection, prediction, and understanding of comorbid conditions in hospital patients. In many cases, these techniques allow efficient identification of comorbidity from medical records, without requiring time-consuming and expensive clinical annotation, which is also prone to inconsistencies. In this chapter, we survey the state-of-the-art ML and NA techniques to improve the detection, analysis and prediction of comorbid conditions in hospital patients. We describe and compare different approaches based on ML and NA algorithms to model the complex relationships between co-occurring diseases, analyzing how different features (such as patient demographics, medical records, and laboratory results) can be used to train and optimize the associated models. We also discuss the use of different advanced techniques applied to electronic health records---such as named entity recognition (NER)---that aim to provide better interpretability of the resulting models, with direct implications for improving patient management and healthcare delivery.