Pressure Injury Risk Assessment in Nursing Practice: A Head-to-Head Comparison of the Braden Scale and Machine...

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Healthcare
• Predictive Analytics

Background:

• Pressure injury (PI) prevention heavily relies on the Braden Scale, but its predictive accuracy is limited by subjective measures and fixed thresholds.
• This can lead to errors like false alarms and inefficient workforce allocation.
• Machine learning (ML) offers a potential solution for improved PI risk discrimination, yet clinical comparisons are scarce.

Purpose of the Study:

• To compare the predictive performance of five classic ML models against the Braden Scale for pressure injury risk.
• To evaluate ML models trained on routine clinical data, excluding Braden Scale inputs.
• To assess ML performance using a matched operating-point framework against Braden Scale cutoffs.

Main Methods:

• Utilized data from 446 hospitalized patients in a tertiary Chilean hospital.
• Trained and compared five ML classification models (including XGB) using routinely collected clinical and nursing variables.
• Applied a matched operating-point framework to align ML thresholds with Braden Scale cutoffs based on equivalent recall or specificity.

Main Results:

• The XGB model demonstrated the highest discrimination performance with an AUC of 0.835.
• When matched for recall, XGB achieved a 17% increase in specificity, reducing false positives.
• When matched for specificity, ML models showed recall improvements of 13%–25%.

Conclusions:

• ML models, especially XGB, outperform the Braden Scale under equivalent clinical operating conditions.
• ML enhances precision and resource allocation while maintaining patient safety.
• ML serves as a valuable adjunct to, rather than a replacement for, the Braden Scale in PI prevention.