Grant: Travel Awards - Institute Community Support

The development of Artificial Intelligence (AI) systems to support diagnostic decision making is rapidly expanding in healthcare. Yet, several challenges exist to execute algorithmic systems at the frontlines of clinical practice, where, most often, they have not been trained to fit local patterns of care. This research analyses the implementation of an AI-based decision support system (DSS) in the emergency department of a large Academic Health Center. The DSS presents as a mobile application that questions patients prior to their encounter with physicians. Using Natural Language Processing, the DSS first identifies the patient chief complaint; then, it adapts the questionnaire according to patient answers. The DSS finally outputs a medical history that is given to physicians to streamline patient questioning and medical note-taking, and help them determine the more accurate diagnostic. This research focuses on the DSS implementation with the aim to identify the main barriers to physician adoption of the AI-generated medical history. To achieve this objective, this research relies on complementary qualitative methods: 20 semi-directed interviews with managers, developers, physicians and nurses; informal conversations; non-participant observation; and document analysis. Data were collected from January to September 2019 and analyzed using thematic analysis methodology with NVIVO 12 software. This research evidences several interrelated technical, social and organizational factors that, according to research participants, limited physician adoption of the AI-based DSS. In particular, the research shows how the DSS was perceived by physicians as introducing additional risk in clinical practice, which led to resistance to adoption. For healthcare managers, these results point out to the importance of considering the unique challenges associated with the implementation of AI-based systems, such as physicians' distrust in the effectivity of automated decision making.