Predicting Pressure Injury in Critical Care Patients: A Machine-Learning Model.

Area of Science:

• Medical Informatics
• Clinical Prediction Models
• Machine Learning in Healthcare

Background:

• Hospital-acquired pressure injuries are a significant concern for critical care patients.
• Preventive measures like specialty beds are costly and not universally applicable.
• Existing risk assessment tools often classify most critical care patients as high-risk, complicating targeted interventions.

Purpose of the Study:

• To develop a predictive model for pressure injuries in surgical critical care patients.
• To leverage electronic health record data for improved risk stratification.
• To create a more efficient and accessible tool for identifying patients at risk.

Main Methods:

• A random forest algorithm was employed using data from electronic health records.
• The dataset was split into training (67%) and testing (33%) subsets for model development and validation.
• The model was designed to predict the development of stage 1 or greater, and stage 2 or greater pressure injuries.

Main Results:

• The study analyzed 6376 patients, with 8.1% developing stage 1 or greater pressure injuries and 4.0% developing stage 2 or greater injuries.
• The random forest models achieved an area under the receiver operating characteristic curve of 0.79 for both prediction outcomes.
• The models demonstrated effective classification performance in the testing dataset.

Conclusions:

• This machine learning model utilizes readily available electronic health record data, unlike traditional tools requiring manual input.
• The approach offers a novel method for predicting pressure injuries without relying on clinician-inputted data.
• Future research will focus on independent sample testing and calibration to enhance model specificity.