Artificial intelligence-based approaches for advance care planning: a scoping review

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Palliative Care Research

Background:

• Advance Care Planning (ACP) is crucial for informed healthcare decisions but faces implementation barriers like time constraints and unclear professional roles.
• Artificial intelligence (AI) offers potential solutions for optimizing ACP by identifying patients and aiding decision-making.
• The current application of AI in palliative care for ACP remains unclear.

Purpose of the Study:

• To explore the utilization of AI models in Advance Care Planning (ACP).
• To identify key factors influencing AI model performance, data transparency, code availability, and generalizability in ACP.
• To understand the current landscape of AI applications within palliative care for ACP.

Main Methods:

• A comprehensive scoping review was conducted following the Arksey and O'Malley framework and PRISMA-ScR guidelines.
• Searches were performed on Scopus, Web of Science, and seven preprint servers for English, German, and French publications over the last 10 years.
• Search terms combined concepts of ACP and AI models, with study quality assessed using the GRADE approach.

Main Results:

• Forty-one studies were included, primarily using retrospective electronic health record data.
• Most studies focused on identifying patients for ACP (n=39), with fewer addressing discussion initiation (n=10) or documentation (n=8).
• Logistic regression was the most common AI method (n=15); models generally showed good performance (n=28), but transparency and reproducibility were lacking (n=17, n=36).

Conclusions:

• AI models demonstrate promising results for predicting outcomes and supporting decision-making in ACP.
• Significant challenges persist regarding data and code availability, impacting transparency and reproducibility.
• Future research should prioritize open-source code and transparency to enable rigorous evaluation and explore novel AI approaches for ACP processes.