The Impact of Information Relevancy and Interactivity on Intensivists' Trust in a Machine Learning-Based Bacteremia...

Area of Science:

• Medical Informatics
• Human-Computer Interaction
• Artificial Intelligence in Medicine

Background:

• Machine learning (ML) algorithms are advancing rapidly but are often "black boxes," leading to distrust.
• In critical medical fields, reluctance to trust ML systems is high due to potential fatal outcomes.

Purpose of the Study:

• To investigate how user interface (UI) design elements influence intensivists' trust in ML-based clinical decision support systems (CDSS).

Main Methods:

• 47 critical care physicians evaluated 3 bacteremia patient cases using an ML-based simulation system.
• UI conditions varied in information relevancy and interactivity.
• Trust was measured by agreement with ML predictions and post-experiment questionnaires.

Main Results:

• Physician agreement with ML predictions was not affected by UI conditions.
• Higher perceived information relevancy and interactivity in the UI significantly increased trust (P<.001).
• Explicitly displaying ML algorithm features on the UI led to decreased trust (P=.05).

Conclusions:

• UI design incorporating information relevancy and interactivity is crucial for building intensivist trust in ML CDSS.
• Findings highlight the importance of UI design in human-ML interaction within intensive care units.