Equity in Using Artificial Intelligence Mortality Predictions to Target Goals of Care Documentation

Area of Science:

• Medical Informatics
• Health Equity
• Clinical Decision Support

Background:

• Artificial intelligence (AI) is increasingly utilized to identify patients at high risk of mortality for goals-of-care (GOC) discussions.
• This practice aims to proactively engage patients in end-of-life care planning.

Purpose of the Study:

• To assess the association between the presence or absence of AI-generated mortality risk scores and the documentation of GOC conversations.
• To investigate potential disparities in GOC documentation based on AI risk score assignment.

Main Methods:

• Retrospective cross-sectional study of 13,710 hospitalized adult patients.
• Utilized the Serious Illness Risk Indicator (SIRI) score to define elevated (>30% 90-day mortality risk) or absent scores.
• Employed propensity score matching and logistic regression to compare GOC documentation rates.

Main Results:

• Patients with elevated SIRI scores were more likely to have GOC documentation compared to those without scores (aOR 2.1, p<0.0001).
• Patients lacking SIRI scores were more likely to be younger, non-White, and have Medicaid insurance.
• The intervention targeted increased GOC documentation for patients with elevated AI risk scores.

Conclusions:

• AI-driven targeting for GOC documentation may lead to disparities, potentially disadvantaging patients without AI scores.
• Severely ill patients lacking AI-generated scores, often from minority and lower socioeconomic groups, may receive less targeted GOC documentation.
• This highlights the need for equitable AI implementation in clinical care pathways.