End-Of-Life: AI Fails to Forecast the Ultimate Choice.


End-of-life decisions are complex and often feared by both patients and healthcare practitioners. Although many people express a desire to die at home, they often end up in hospitals or acute care settings. This is partly due to the under-utilization of hospice facilities and healthcare professionals’ hesitancy to initiate conversations about end-of-life. Living wills and advanced directives, which can help individuals take control of the process, are generally uncommon or insufficiently detailed. Inaccurate predictions of a dying patient’s preferences by next-of-kin or surrogate decision-makers further complicate the decision-making process.

The article explores the potential for artificial intelligence (AI) to assist with end-of-life decision-making. While discussions about a “patient preference predictor” using AI algorithms have been gaining traction in the medical community, these algorithms are still under development and have not been clinically adopted. The challenge lies in determining which set of values the algorithms should be based on and establishing a “ground truth” against which to train or measure algorithm performance. Detailed datasets that include relevant information about a patient’s demographics, socioeconomic status, and religious or spiritual beliefs are currently lacking. Retrospective data and hypothetical scenarios may not be reliable enough to train end-of-life algorithms. Additionally, there are challenges related to determining the threshold of accuracy that would be acceptable and addressing issues of explainability and bias in the algorithms.

The article concludes by emphasizing the importance of individuals taking ownership of their end-of-life decisions and expressing their personal choices, rather than relying on AI algorithms. By proactively addressing these decisions, individuals can minimize the need for personalized algorithms in the future.