A new paper by Nikola Biller-Andorno and Armin Biller on the use of algorithms to predict patient preferences has recently been published in the New England Journal of Medicine.
The use of artificial intelligence (AI) is increasingly common in medicine, as it is in other fields. Just as AI systems using natural language processing answer legal questions, predict court decisions, and prepare contracts, and as self-driving cars gather traffic information so they can make and act on decisions, diagnostic algorithms already outperform radiologists and dermatologists, estimate life expectancy, and detect health risks. Health care AI alone is expected to become a $20 billion market within the next 5 years. Training algorithms to predict people’s advance health care choices seem to be in keeping with the stream of emerging applications, and it would be surprising if we had long to wait before products based on such technology became commercially available.
Advance decisions about such matters as do not-attempt-resuscitation (DNAR) status, organ donation, and curative versus palliative care are based on individual preferences and values and frequently reflect moral choices. They are notoriously difficult for others to make for the person in question, especially if no clear instructions have been provided through an advance directive or care plan. It is a well-known problem in clinical ethics that surrogates may make decisions that are inconsistent with the person’s preferences and values. Readily available intelligent decision support might improve the process, which must often take place in suboptimal conditions involving time constraints, stress, unclear advance directives, unavailability of surrogates, personal bias, or conflicts of interest.