Using Visual Patient to Show Vital Sign Predictions, a Computer-Based Mixed Quantitative and Qualitative Simulation...

Area of Science:

• Medical Informatics
• Human-Computer Interaction
• Machine Learning in Healthcare

Background:

• Machine learning models can predict future patient vital signs.
• Integrating these predictions into an avatar visualization (Philips Visual-Patient-avatar) aids human caregivers by presenting data non-numerically.
• This approach aims to improve the usability of complex patient monitoring data.

Purpose of the Study:

• To integrate machine learning-based vital sign predictions into a visual patient avatar.
• To evaluate the understandability and usability of these visualizations among healthcare professionals.
• To gather user feedback for future development of the visualization system.

Main Methods:

• A simulation study involved 70 participants (anesthesiologists and intensivists) across 3 European hospitals.
• Participants identified prediction visualizations comprising a condition and an urgency.
• Qualitative feedback was collected via interviews and rated in an online survey.

Main Results:

• Correct identification of conditions alone was 93.8%; accuracy decreased when urgency was included (77.9%).
• 65.3% found the visualizations fun, and 61.2% could imagine using them, but 65.3% found urgencies difficult to identify.
• User feedback indicated a need for improvement in visualizing prediction urgency.

Conclusions:

• Care providers accurately identified over 90% of predicted conditions without urgency.
• The accuracy of identifying both condition and urgency was lower, highlighting a usability challenge.
• Future development will focus on displaying conditions only or enhancing urgency visualization for better clinical integration.