A multi-site study of clinician perspectives in the lifecycle of an algorithmic risk prediction tool

Recent advancements in the performative capacities of artificial intelligence (AI), machine learning (ML), and algorithmic-based tools open up numerous applications in modern medicine. There are, however, few studies that track the whole lifecycle of a digital healthcare tool as it evolves from conception, to design, and deployment in real world settings-especially with a focus on the social dynamics amongst the end-users of the tool: clinicians. In this paper, we present data from a multi-site, 5-year study focused on the development and deployment of an algorithmic risk calculator (HeartMate 3 Risk Score) into a validated and efficacy tested clinical decision support system (CDSS) for patients and clinicians engaging in shared decision making about left ventricular assist device (LVAD) therapy for advanced heart failure. We conducted a total of 76 interviews with 20 advanced heart failure cardiologists and 14 nurse coordinators with LVAD expertise (n=34) across different timepoints during the lifecycle of this digital healthcare tool. Results from Thematic Analysis revealed an array of social factors at play at each stage of the tool's development and implementation, from finding social consensus around risk messaging in the conception and design phases, to various social contingencies that served as facilitators and barriers to the successful integration of the tool in its later stages. Our findings confirm many previously raised issues with introducing new medical and digital healthcare tools into clinical care, and highlight new issues specific to the rapidly advancing technology in CDSS.