Transforming health policy through machine learning
Hutan Ashrafian, Ara Wardkes Darzi · PLoS Medicine · 2018
Machine learning (ML) is one of the most prominent applications of artificial intelligence (AI) technology and offers multiple routes to support the core objectives of health policy.These include 'creating the conditions that ensure good health' [1] and social care for an entire population through preventive strategies, protection from disease, promotion of healthy lifestyles, and population screening through knowledge capture (typically in the form of big data).Overall governance will offer a patient-centred approach with the consideration of patient advocacy and workforce and resource management [2] (Fig 1).Herein, we will break down the role of ML in each of these areas.The most prominent contribution of AI to health policy knowledge currently resides within the application of ML to large, population-level datasets such as those from medical imaging, electronic health records (EHRs), and whole-genome studies.This information can guide interventions for high-risk individuals.Current applications can outperform established risk scores to predict clinical outcomes.These include in-hospital mortality, 30-day unplanned readmission, prolonged length of stay, and final discharge diagnoses for patient populations numbering in the several hundred thousands [3].Here, the major limitation is access to large and high-quality population-level datasets with which to apply ML approaches.We feel that the unification of the United Kingdom National Health Service (NHS) dataset of over 66 Fig 1.The components of health policy.