Development of a machine learning-based acuity score prediction model for virtual care settings.

Area of Science:

• Digital health innovation
• Machine learning in healthcare
• Clinical decision support systems

Background:

• Healthcare digitization is advancing, but remote triage prediction systems using machine learning (ML) are limited.
• The Canadian Triage and Acuity Scale (CTAS) is the standard for in-person triage in Canada.

Purpose of the Study:

• To develop an ML-based acuity score system modeled after the CTAS.
• To address the gap in remote and automated triage prediction for virtual urgent care.

Main Methods:

• Utilized 2,460,109 de-identified patient records from three Canadian healthcare organizations.
• Trained five ML models (decision tree, k-NN, random forest, gradient boosting, neural net) using presenting complaint, modifiers, age, sex, and pain.
• Compared ML-predicted acuity scores against nurse-assigned CTAS scores.

Main Results:

• Gradient boosting regressor achieved the highest prediction accuracy.
• The model was optimized for up-triage to enhance patient safety.
• The algorithm predicted the same acuity score in 47.4% of cases and an equal or higher score in 95.0% of cases.

Conclusions:

• The ML algorithm exhibits strong predictive accuracy and safety.
• This represents the largest dataset used for such a model to date.
• Future validation through a pilot study is planned for remote acuity score assignment.