Artificial intelligence (AI) predictions are slowly but surely making their way into end-of-life medical decisions, raising important ethical questions about the role of AI in such sensitive matters. Recent studies have shown the potential benefits of using AI algorithms to predict patient preferences and guide end-of-life care planning.
One study conducted by European researchers focused on building an AI “patient preference predictor” based on detailed surveys of patient preferences for end-of-life interventions. The personalized model, which analyzed 61 characteristics of each patient over the age of 50, accurately predicted end-of-life preferences 71% of the time. This outperformed the accuracy of human surrogates and couples predicting their partner’s wishes.
Another study involved sharing AI-generated predictions of a patient’s death risk with clinicians to facilitate end-of-life planning. This initiative led to a significant increase in the use of palliative care and hospice, resulting in a lower death rate within 30 days of discharge than expected.
Both sets of researchers emphasized the importance of human involvement in the decision-making process alongside AI predictions. While AI algorithms can provide valuable insights, they should not replace human judgment and values. The European team suggested integrating human surrogates into the AI training process to ensure a more holistic approach to end-of-life care decisions.
In contrast to previous attempts to use predictive algorithms for end-of-life discussions, the St. Louis group implemented a careful review process to avoid overwhelming clinicians with alerts. Alerts generated by an algorithmic mortality score were first reviewed by another clinician before being forwarded if necessary, reducing the risk of “alert fatigue.”
As the use of AI in medicine continues to grow, concerns about overestimating its capabilities in end-of-life care persist. A commentary on the European research findings highlighted the importance of recognizing that preferences can change over time and that AI should serve as a decision aid rather than a decision maker.
Looking ahead, researchers are exploring the development of AI surrogates that not only predict outcomes but also consider the moral adequacy of representation. These AI models would aim to honor patients’ values, relationships, and cultural worldviews while providing transparent reasoning for their predictions.
Ultimately, while AI can offer valuable insights into end-of-life care decisions, individuals are encouraged to take an active role in advance care planning to ensure their preferences are accurately reflected. By balancing the benefits of AI predictions with human judgment, we can create a more comprehensive and ethical approach to end-of-life care.
