On the ethics of algorithmic decision-making in healthcare

Machine learning is increasingly being conceived as a technology with the potential to transform professional healthcare. Recently, there has been a surge of interest in machine learning for medical decision-making (reviewed by Esteva and Topol1 2), fuelled by a series of studies demonstrating'expert-level' accuracy of machine learning algorithms, for example, in diagnosing eye diseases from fundus images,3 and different types of skin cancer from images of skin lesions.4 Moreover, a study made by Walsh and colleagues found that machine learning algorithms managed to predict the risk of imminent suicide attempts at high accuracy based on a large repository of clinical electronic health data (Walsh, p. 460).5 In contrast, for clinicians, the ability to predict suicide attempts has been near chance for decades. Hence, machine learning algorithms promise to enhance the diagnostic as well as the predictive abilities of clinicians by assessing health risks of individual patients based on complex diagnostic data sets.