Abstract

Substantial advancement in the diagnosis and treatment of psychiatric disorders may come from assembling diverse data streams from clinical notes, neuroimaging, genetics, and real-time digital footprints from smartphones and wearable devices. This is called “deep phenotyping” and often involves machine learning. We argue that incidental findings arising in deep phenotyping research have certain special, morally and legally salient features: They are specific, actionable, numerous, and probabilistic. We consider ethical and legal implications of these features and propose a practical ethics strategy for managing them.