Machine learning algorithms to predict seizure due to acute tramadol poisoning

Area of Science:

• Medical Informatics
• Clinical Toxicology
• Machine Learning in Medicine

Background:

• Acute tramadol poisoning poses a significant risk for seizures.
• Effective prediction of seizures is crucial for timely clinical intervention.
• Identifying high-risk patients aids in proactive management and resource allocation.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for predicting seizures in acute tramadol poisoning.
• To identify key predictors associated with seizure occurrence.
• To enhance clinical decision-making in emergency departments.

Main Methods:

• Data from 909 acute tramadol poisoning cases (2013-2019) were analyzed.
• Variable selection was performed using the random forest method.
• Prediction models were built using Support Vector Machine (SVM), Naïve Bayes (NB), Artificial Neural Network (ANN), and K-Nearest Neighbor (K-NN) algorithms.
• Model performance was assessed using Area Under the Curve (AUC) and other diagnostic metrics.

Main Results:

• Seizures occurred in 59.8% of patients (544/909).
• Key predictors identified included sex, pulse rate, arterial blood oxygen pressure, blood bicarbonate level, and pH.
• NB (AUC=0.71), ANN (AUC=0.70), and SVM (AUC=0.68) models demonstrated better performance than K-NN (AUC=0.58).
• NB excelled in sensitivity and negative predictive value; K-NN in specificity and positive predictive value.

Conclusions:

• While SVM, ANN, and NB models showed comparable performance, validated logistic regression (LR) emerged as the superior model.
• Accurate prediction models can significantly improve clinical decision-making and patient care in emergency settings.
• The findings support the use of predictive modeling for managing tramadol poisoning cases.