PROJECT ABSTRACT Artificial Intelligence (AI) - computer-based algorithms capable of learning from enormous data sets, including electronic health records and chart notes, in order to carry out tasks typically reserved for humans – is poised to dramatically affect medical research and practice, including end-of-life and palliative care (EOLPC). Recent AI-based algorithms seem capable of accurately predicting a patient’s prognosis or probability of death years in advance. These algorithms can do so in an automated fashion, without the input of clinicians, and they are starting to move from research into practice. For the millions of Americans who experience the physical, psychological, and social effects of severe and chronic illness, knowing a prognosis could promote earlier access to palliative care and to support medical decision-making that is consistent with patients’ and families’ goals and preferences. However, AI also raises concerns about loss of autonomy in patient or clinician decision-making, depersonalized or unempathetic care, racially biased algorithms, distrust of “black box” machines, and an over-emphasis on survival statistics in decision-making. Studies consistently show that patients and caregivers may be unaware of their prognosis, that physicians are often inaccurate in predictions, and that patients of certain socioeconomic statuses or races may be less aware of their prognosis; however, the need for an accurate prognosis may vary by disease state, individual preference, or other sociocultural factors. Thus, how AI-based prognostication will affect our basic scientific understanding of the role of prognostic awareness in medical decision-making in support of high quality, goal concordant EOLPC is a critical knowledge gap. Before AI becomes more widely used in EOLPC, spreads to other uses (e.g., virtual nurse assistants and caregiver robots), or becomes necessary as proof o f eligibility for services (e.g., hospice), there is an urgent need to understand its potential impact on patient- and family-centered care and to develop practical ethics guidance for its use. The goal of this project is to ensure AI is developed and implemented in ways that support high quality EOLPC. With a unique team of experts in palliative care, artificial intelligence, bioethics, and patient engagement, we will: (1) use semi-structured interviews to obtain rich insights into the experiences and beliefs of all EOLPC team members, patients, and family caregivers regarding AI-based prognostication at 4 purposefully chosen sites across the United States; (2) conduct a nationally representative survey of palliative care physicians regarding the anticipated benefits and challenges of using AI-based prognostication; and (3) convene a Delphi panel of experts to create practical recommendations for the use of AI in EOLPC. The project will be supported within the Palliative Care Research Cooperative Group (PCRC) (U2C NR014637), a robust interdisciplinary research community comprised of more than 500 members at more than 180 sites.

PROJECT NARRATIVE Millions of Americans could benefit from palliative care to address the physical, psychological, and social effects of severe and chronic illness. Artificial intelligence (AI) - computer-based algorithms that can carry out tasks typically reserved for humans – has the potential to give patients and families more detailed and more accurate information about prognosis than ever before, but whether or how to use this information ethically in support of patient- and family-centered care is unknown. To ensure AI supports high quality end-of-life and palliative care that is consistent with patients’ and families’ beliefs, values, and preferences, this project will identify the benefits and challenges of AI for palliative care and then develop practical guidance for its use.