Predicting open wound mortality in the ICU using machine learning.

Area of Science:

• Critical Care Medicine
• Data Science in Healthcare
• Wound Management

Background:

• Open wounds significantly impact patient health, leading to pain, functional loss, and increased mortality.
• Open wounds are a prevalent comorbidity, affecting a large US population and increasing ICU stay risk.
• Limited research exists on mortality prediction for ICU patients with open wounds.

Purpose of the Study:

• To develop a predictive model for mortality risk in intensive care unit (ICU) patients with open wounds.
• To assess the efficacy of machine learning models in predicting outcomes for this patient group.

Main Methods:

• Utilized the Medical Information Mart for Intensive Care III (MIMIC-III) database for de-identified patient data.
• Developed and compared random forest and binomial logistic regression models.
• Included variables such as wound location, demographics, admission type, platelet count, and hyperphosphatemia.

Main Results:

• The study included 3,937 patients; 15% died during their ICU stay.
• The random forest model demonstrated high predictive accuracy with an Area Under the Curve (AUC) of 0.924.
• Traditional comorbidity indices showed lower predictive power (AUCs of 0.528 and 0.565).

Conclusions:

• Machine learning models show promise for predicting mortality in ICU patients with open wounds.
• These models can potentially enhance clinical decision-making and patient management.
• Further research may refine these predictive tools for improved wound care outcomes.