Health Informatics via Machine Learning for the Clinical Management of Patients

摘要

The modern generation of health informatics systems based on machine learning incorporates a range of technologies from (i) wearable sensors for the acquisition of data from patients, through to (ii) bio-medical signal processing methods for conditioning the high-resolution data that result, through to (iii) “big data” machine learning approaches to fuse the heterogeneous data that are currently collected from routine clinical care in many healthcare systems. The latter may include fusion of time-series data from sensors; categorical data from electronic health records (EHRs); and, increasingly, biomarkers derived from genomics, proteomics, and other sources. While this field holds substantial promise for the future of medicine, and for our ability to tailor care to the particular physiology of the individual, the penetration of such systems based on machine learning into actual clinical practice is in its infancy. This review surveys the recent literature in this rapidly-changing field, aiming to investigate how health informatics systems that employ machine-learning methods are affecting the clinical management of patients. While the range of literature presented is broad, underpinning each publication is the demonstration of machine learning methods for health informatics.This review describes health informatics systems that are based on machine learning, throughout the patient journey through a typical hospital healthcare system, from (i) the intensive care unit (ICU), to (ii) discharge and subsequent monitoring on acute wards and on general wards, to (iii) wider-scale tracking of patient condition using the EHR.