Predicting delayed neurological sequelae in patients with carbon monoxide poisoning using machine learning models.

Area of Science:

• Neurology
• Medical Informatics
• Toxicology

Background:

• Delayed neurological sequelae (DNS) are a common complication of carbon monoxide (CO) poisoning.
• DNS significantly impacts patient quality of life.
• Predictive models are needed to identify at-risk individuals.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting DNS in CO poisoning patients.
• To identify key factors contributing to DNS prediction.

Main Methods:

• Retrospective analysis of 360 CO poisoning patients.
• Development and evaluation of 16 ML models.
• Utilized Synthetic Minority Oversampling Technique (SMOTE) with Random Forest (RF).
• Performance assessed using accuracy, sensitivity, specificity, and AUC.

Main Results:

• The SMOTE-Random Forest model achieved an AUC of 0.89.
• The model demonstrated high accuracy (0.83), sensitivity (0.9), and specificity (0.8).
• Key predictors included Glasgow Coma Scale, hyperbaric oxygen therapy, kidney function, immune response, liver function, and blood clotting.

Conclusions:

• A robust ML model (SMOTE-Random Forest) can accurately predict DNS in CO poisoning patients.
• The model aids in early identification of individuals at high risk for DNS.
• Explainable AI (SHAP) highlighted critical clinical factors influencing DNS risk.