“Screening” for End of Life Using Artificial Intelligence: A Qualitative Study of Palliative Care Team Members’ Perspectives on Ethical Use

Abstract

Background:

Artificial intelligence (AI) tools for health care applications are rapidly emerging. Some AI-based prognostic tools can predict patient mortality automatically and with accuracy that outperforms clinicians and other available tools. In palliative care (PC), these tools may change practice in ways we do not fully understand and raise important ethical and implementation questions.

Objective:

Identify the ethical challenges that could shape the implementation of AI-based prognostication in PC.

Design:

We conducted a qualitative study using semistructured interviews and a modified grounded theory approach.

Setting/Subjects:

We interviewed 45 PC team members (physicians, nurses, social workers, spiritual care providers, psychologists) from four U.S. academic medical centers.

Results:

PC team members viewed AI-based prognostication as a form of “screening” for end of life (EOL) and believed it could help identify patients with PC or other EOL needs. By drawing on an established model of screening ethics as a framework for our analysis, we identified four principles to guide the implementation of AI-based prognostication as screening: (i) screening for EOL should be evidence-based, (ii) screening for EOL should take opportunity cost into account, (iii) screening for EOL should distribute costs and benefits fairly, and (iv) screening for EOL should offer respect for persons.

Conclusions:

Our findings help us understand how PC team members view AI-based prognostic tools and offer guidance for their implementation. In the future, it will be important to define more precisely the role of screening in this context and understand how it affects decision making for patients, families, and care teams.