A Personalized Patient Preference Predictor for Substituted Judgments in Healthcare: Technically Feasible and...

Area of Science:

• Bioethics
• Artificial Intelligence
• Medical Decision-Making

Background:

• Surrogates face challenges in determining incapacitated patients' treatment preferences.
• Existing Patient Preference Predictors (PPP) use population data, raising autonomy concerns.
• Critics argue PPPs lack individual specificity, potentially misrepresenting patient values.

Purpose of the Study:

• To propose a Personalized Patient Preference Predictor (P4) that addresses autonomy concerns.
• To leverage machine learning for inferring patient preferences from person-specific data.
• To enhance accuracy and respect for individual patient values in substituted judgment.

Main Methods:

• Utilizing recent advances in machine learning and large language models.
• Fine-tuning AI models on individual patient-specific data (e.g., prior decisions).
• Comparing the proposed P4 approach with existing PPP methods.

Main Results:

• The P4 is technically feasible due to advancements in AI.
• P4 predictions are expected to be more accurate at the individual level.
• P4 is designed to more directly reflect a patient's unique reasons and values.

Conclusions:

• The Personalized Patient Preference Predictor (P4) offers a promising solution for substituted judgment.
• P4 has the potential to alleviate autonomy-based criticisms of earlier preference prediction models.
• Further development and deployment of P4 require careful consideration of ethical objections.