Development and Internal Validation of a Gradient Boosting Model for Pressure Injury Risk in the ICU

Area of Science:

• Critical Care Medicine
• Artificial Intelligence in Healthcare
• Patient Safety

Background:

• Pressure injuries (PI) are a significant challenge in Intensive Care Units (ICUs).
• Existing risk assessment tools, like the Braden scale, have limitations in capturing PI risk comprehensively.
• There is a need for advanced predictive models using routinely collected ICU data.

Purpose of the Study:

• To develop and internally validate a machine-learning model for predicting new-onset pressure injuries in ICU patients.
• To assess the model's performance using routinely collected electronic health record data.
• To compare the machine learning model's predictive power against traditional risk assessment methods.

Main Methods:

• Retrospective analysis of adult ICU patients with length of stay ≥48 hours (2018-2023).
• Candidate predictors included albumin, lactate, SOFA, APACHE II, Braden score, age, BMI, nutrition score, and treatment indicators.
• A Gradient Boosting Model (GBM) was developed and validated using cross-validation, with Random Forest as a benchmark.

Main Results:

• 14.6% of ICU stays resulted in new-onset pressure injuries.
• The GBM achieved an Area Under the Curve (AUC) of approximately 0.69 with acceptable calibration.
• Key predictors identified included higher lactate, lower albumin, lower Braden scores, older age, CRRT, prone positioning, enteral nutrition, and analgesic exposure.

Conclusions:

• A Gradient Boosting Model utilizing routine ICU data offers moderate, well-calibrated discrimination for predicting new-onset pressure injuries.
• The developed model demonstrates decision-relevant net benefit, complementing existing screening tools like the Braden scale.
• This AI-driven approach can enhance risk stratification and optimize preventive interventions for ICU patients at high risk of pressure injuries.