# From inferring preferences to enabling choice: potentials of digital tools to improve substitute decision-making

## TL;DR

This paper discusses how digital tools and AI can help make healthcare decisions for patients who cannot decide themselves, but emphasizes the importance of advance care planning over preference prediction.

## Contribution

The paper highlights the need to prioritize digital tools that strengthen advance care planning over those that predict patient preferences.

## Key findings

- Digital AI tools can infer patient treatment preferences but lack the ethical robustness of advance directives.
- Current enthusiasm for AI preference prediction risks overshadowing the potential of digital tools to improve advance care planning.
- Investing in digital technologies to enhance advance care planning is crucial for respecting patient autonomy.

## Abstract

Respect for patient autonomy is a foundational principle in healthcare ethics, which holds that patients can make their own treatment decisions. However, sometimes patients lack the capacity to do so and surrogates must decide on their behalf in the sense of substitute decision-making. This is challenging, as guidance for these decisions is often lacking due to limited engagement in advance care planning (ACP) and the low prevalence of advance directives (ADs), which allow patients to pre-determine their treatment preferences. In response to these challenges, digital technologies employing artificial intelligence—particularly so-called (Personalized) Patient Preference Predictors (PPP or P4)—have recently received comprehensive scholarly attention, with initial studies exploring their technical feasibility. These tools aim to leverage AI's capacity to process large datasets to infer individual patients’ likely treatment preferences, thereby hoping to alleviate surrogates’ burden and to promote patient autonomy by facilitating treatment decisions more in line with patients’ preferences. In this article, we emphasize that autonomy is more robustly respected when substitute decisions rely on deliberate expressions of will formulated through ACP or documented in ADs rather than on even highly accurate predictions of treatment preferences. While we acknowledge the potential of PPPs/P4s to improve substitute decision-making when no explicit guidance exists, we caution against allowing current enthusiasm for AI-driven preference prediction to overlook the considerable potential that digital tools and AI offer for strengthening ACP and increasing completion of ADs. We therefore call for greater investment in using digital technologies to enhance ACP processes.