Implementing AI Models for Prognostic Predictions in High-Risk Burn Patients

Area of Science:

• Medical Informatics
• Computational Medicine
• Artificial Intelligence in Healthcare

Background:

• Burn injuries present a spectrum of severity, from minor to life-threatening.
• Treatment varies based on burn severity and location, ranging from home care to specialized burn centers.
• Predicting adverse outcomes in burn patients is crucial for effective management.

Purpose of the Study:

• To develop and evaluate artificial intelligence (AI) and machine learning (ML) models for forecasting adverse effects in burn patients.
• To predict the likelihood of graft surgery, prolonged hospital stays, and overall complications.
• To integrate predictive AI models into hospital information systems for clinical decision support.

Main Methods:

• Retrospective analysis of 224 burn patients admitted between 2010 and 2019.
• Utilized 14 features including comorbidities and laboratory results for model training and testing (70% train, 30% test).
• Employed Random Forest, LightGBM, and Logistic Regression algorithms, evaluating with accuracy, sensitivity, specificity, and AUC.

Main Results:

• Random Forest model achieved the highest AUC (81.1%) for predicting prolonged hospital stays (>14 days).
• Random Forest model showed the highest AUC (78.8%) for predicting the need for skin grafts.
• Random Forest and XGBoost models demonstrated the highest AUC (87.2%) for predicting overall adverse complications.

Conclusions:

• AI and ML models effectively predict prolonged hospital stays, skin graft requirements, and adverse complications in burn patients.
• The developed AI prediction system can be integrated into hospital systems to enhance clinical decision-making.
• This approach supports improved physician-patient communication and care planning for burn survivors.