Safe Implementation of Agentic Artificial Intelligence in Palliative Care: A Systematic Review

This research project is a systematic review examining the safe implementation of agentic artificial intelligence in palliative care. Agentic AI refers to artificial intelligence systems capable of goal-directed recommendation, planning, adaptive interaction, workflow initiation, or partially autonomous decision support. While such systems may help improve patient identification, prognostication, advance care planning, documentation, and symptom-management support, their use in palliative care raises important safety, ethical, clinical, and governance concerns. The purpose of this review is to synthesize and critically appraise existing evidence on the benefits, harms, implementation conditions, and ethical considerations associated with agentic or partially autonomous AI systems in palliative, hospice, end-of-life, and supportive care contexts. The review follows PRISMA guidance and uses the SPIDER framework to structure the research question. Searches were conducted across biomedical, computer science, engineering, and multidisciplinary databases. Eligible studies include empirical, qualitative, mixed-methods, design-oriented, implementation-focused, peer-reviewed conference, systematic review, and theoretical or conceptual work that addresses agentic AI in palliative care with explicit attention to safety, ethics, governance, clinical integration, workflow implications, or human oversight. The review uses narrative synthesis because the included studies vary substantially in design, population, AI system type, clinical context, and outcome reporting. Studies are grouped into application areas, including AI for prognostication, screening, and triggering palliative care referrals, as well as broader applications such as advance care planning, documentation support, expert deliberation, and symptom-management decision support. Quality appraisal is conducted using study-design-appropriate tools, including Joanna Briggs Institute checklists, the Critical Appraisal Skills Programme checklist, and the Mixed Methods Appraisal Tool. Expected outcomes of this project include a clearer understanding of how agentic AI is currently being applied in palliative care, what benefits have been reported, and what safety risks remain unresolved. The review is expected to identify key implementation requirements, including transparent and explainable system design, human-in-the-loop oversight, ethical governance, workflow integration, accountability structures, confidentiality safeguards, and equity evaluation. The project also aims to highlight evidence gaps, particularly the limited availability of longitudinal patient-centered outcomes, family experience data, equity analyses, workforce impact measures, and real-world safety evaluations. By registering this systematic review, the project seeks to promote transparency, reduce duplication, and provide a structured foundation for future research and policy development. The findings are intended to inform clinicians, palliative care nurses, researchers, healthcare leaders, AI developers, and policymakers considering the responsible use of agentic AI in palliative care. Ultimately, this review positions agentic AI as a potentially useful but carefully governed augmentative tool that should support, rather than replace, human clinical judgment, compassionate communication, and person-centered decision-making near the end of life.