Electronic-Medical-Record-Driven Machine Learning Predictive Model for Hospital-Acquired Pressure Injuries:...

Area of Science:

• Bioengineering
• Health Informatics
• Clinical Prediction Models

Background:

• Hospital-acquired pressure injuries (HAPIs) impact millions of patients annually in the US, increasing morbidity and healthcare expenses.
• Existing screening tools like the Braden Scale exhibit limitations in sensitivity for accurate HAPI risk assessment.
• There is a critical need for enhanced predictive methodologies to mitigate HAPI incidence.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting HAPI risk.
• To leverage longitudinal electronic medical record (EMR) data for dynamic risk assessment.
• To establish a generalizable and scalable model for cross-institutional application.

Main Methods:

• Development of an automated pipeline for EMR data curation, labeling, and integration.
• Utilized XGBoost with recursive feature elimination to identify 35 key clinical variables.
• Employed time-series analysis for dynamic HAPI risk prediction in adult inpatients (2018-2023).

Main Results:

• The ML model achieved strong performance with AUROC values of 0.83-0.85 during internal and multi-center external validation.
• The model demonstrated superior sensitivity and F1-score compared to the Braden Scale.
• Performance surpassed previous predictive models for HAPI risk.

Conclusions:

• This study presents the first externally validated, cross-institutional HAPI prediction model utilizing longitudinal EMR data and automated pipelines.
• The model exhibits significant generalizability, scalability, and real-time applicability for clinical settings.
• This bioengineering approach offers a novel strategy to enhance HAPI prevention, patient outcomes, and operational efficiency.