Predicting the Incidence of Pressure Ulcers in the Intensive Care Unit Using Machine Learning.

Area of Science:

• Medical Informatics
• Clinical Quality Improvement
• Patient Safety

Background:

• Hospital-acquired pressure ulcers (PUs) are a critical quality metric in intensive care units (ICUs) globally.
• Limited research exists on PU profiles using electronic health record (EHR) data.
• Current risk assessment often relies on manual scores like the Braden score, with few EHR-based tools available.

Purpose of the Study:

• To develop and evaluate an EHR-based predictive model for pressure ulcer development in the ICU.
• To compare the performance of the EHR model against the Braden score for PU risk stratification.

Main Methods:

• Utilized EHR data from 50,851 ICU admissions (MIMIC-III database).
• Trained machine learning algorithms using demographic, diagnosis, laboratory, and vital sign data within the first 24 hours.
• Assessed the prognostic value of the Braden score (<=12) for PU development.

Main Results:

• The prevalence of stage 2 or higher PUs was 7.8% in the study cohort.
• The Braden score demonstrated low precision (0.09) and recall (0.50) for predicting future PUs.
• A weighted linear regression model using EHR data achieved higher recall (0.71) without incorporating Braden score elements.

Conclusions:

• EHR-based models can outperform the Braden score in identifying patients at risk for PUs.
• This automated risk stratification tool can enhance early intervention strategies and quality protocols in ICUs.