To Recommend or Not to Recommend: Designing and Evaluating AI-Enabled Decision Support for Time-Critical Medical...

Area of Science:

• Artificial Intelligence in Medicine
• Human-Computer Interaction
• Medical Decision Support

Background:

• AI-enabled decision-support systems are crucial for rapid medical decision-making in emergencies.
• A key challenge is enabling providers to effectively interpret AI outputs for optimal treatment choices.

Purpose of the Study:

• To design and evaluate an AI-enabled decision-support system for treating traumatic injuries.
• To assess the impact of different human-AI interaction strategies on provider decision-making.

Main Methods:

• User research with physicians to define information needs and AI outputs.
• An online experiment involving 35 medical providers evaluating AI information synthesis versus AI information and recommendations.

Main Results:

• Providers were more likely to make correct decisions with AI information and recommendations compared to no AI support.
• Identified socio-technical barriers: accuracy-time trade-offs and polarizing perceptions of AI recommendations.

Conclusions:

• AI-enabled decision support, particularly with recommendations, can enhance provider accuracy in critical care.
• Addressing socio-technical barriers is essential for effective AI implementation in time-critical medical events.