Hospice Care in the Era of <scp>AI</scp> : Hospices' Views on Data‐Driven Tools to Support Live Discharge Decisions

BACKGROUND: Live discharge occurs for 20% of hospice enrollees, resulting in loss of support and disruptive care transitions, with higher risk for patients with Alzheimer's disease and related dementias (ADRD). Little is understood about how data-driven clinical decision support tools (e.g., predictive algorithms) might support decision making regarding live discharge. As hospices adopt value-based care, identifying opportunities and challenges for data-driven tools to predict and support live discharge holds great potential to support hospice patients and their caregivers. METHODS: Semi-structured interviews were conducted with 20 hospice leaders in clinical care, quality, and data science at seven non-profit United States hospices. Four-step rapid analysis and deductive approaches were used to summarize interview content in response to research questions and identify cross-organizational themes. RESULTS: Participants identified multilevel-individual and family, organizational, community, and system-challenges and facilitators to support patients following live discharge. Families dealing with ADRD face a heavier care burden but also access ADRD-specific programs. Participants expressed strong interest in using predictive tools to identify patients at increased risk for live discharge and either support them to remain in hospice or facilitate robust discharge planning. Participants emphasized the importance of tool specification and clinical workflow integration to make predictive tools useful and impactful. CONCLUSIONS: Hospices face barriers to support hospice patients and caregivers experiencing burden and suboptimal outcomes following live discharge. Predictive modeling could be a potentially powerful tool to facilitate support for patients discharged alive, provided they are accurately specified and thoughtfully integrated into clinical workflows.