Predicting Pressure Injury in Intensive Care Patients With Ensemble-Based Machine Learning Methods.

Area of Science:

• Medical Informatics
• Computational Biology
• Nursing Science

Background:

• Pressure injuries pose a significant risk to patients in intensive care units (ICUs).
• Predicting and managing pressure injuries requires sophisticated analytical methods.
• Current methods may not fully leverage patient data for personalized treatment strategies.

Purpose of the Study:

• To predict pressure injury risk and treatment methods in ICU patients using machine learning.
• To evaluate the efficacy of ensemble learning models for pressure injury management.
• To develop a data-driven approach for personalized nursing care recommendations.

Main Methods:

• Collected data from various ICU patient populations.
• Applied data preprocessing and oversampling techniques to address data imbalance.
• Utilized ensemble learning methods, including LightGBM, Random Forest, and AdaBoost, with optimization.
• Random Forest achieved the highest classification accuracy for applied treatments.

Main Results:

• Random Forest model demonstrated a 0.76 overall accuracy in classifying pressure injury treatments.
• Ensemble learning methods successfully predicted treatment methods for pressure injuries.
• The study analyzed pressure ulcer stages and linked them to patient characteristics.

Conclusions:

• Ensemble learning-based machine learning models can effectively predict pressure injury treatments in ICUs.
• Data preprocessing and optimization are crucial for high-accuracy clinical machine learning models.
• This approach supports clinical decision-making and enables personalized nursing care for pressure injury management.