Application of Machine Learning Methods for Epilepsy Risk Ranking in Patients with Hematopoietic Malignancies Using

Area of Science:

• Medical Informatics
• Neurology
• Oncology

Background:

• Epilepsy diagnosis in oncohematological and cardiovascular patients is often challenging.
• Machine learning (ML) shows promise for predicting epilepsy risk and outcomes.
• Accurate diagnosis is crucial to avoid unjustified treatment with antiseizure drugs.

Purpose of the Study:

• To develop and validate ML diagnostic models for epilepsy in oncohematological and cardiovascular patients.
• To improve the accuracy of epilepsy diagnosis in these specific patient groups.
• To differentiate between primary epilepsy and seizures secondary to underlying conditions.

Main Methods:

• Analysis of a large hospital database (66,723 vascular, 16,383 oncohematological episodes) from 2010-2020.
• Application of machine learning, specifically gradient boosting, for predictive modeling.
• Cross-validation with accuracy, precision, recall, and F1-score evaluation.

Main Results:

• Gradient boosting achieved a high mean accuracy of 0.96.
• The model demonstrated strong performance with an F1-score of 0.98 and weighted average recall of 0.96.
• Key predictors identified include atrial fibrillation, hypertension, and cerebrovascular conditions.

Conclusions:

• ML models, particularly gradient boosting, are effective for diagnosing epilepsy in oncohematological and cardiovascular patients.
• Identifying specific predictors like atrial fibrillation aids in understanding epilepsy pathogenesis in these groups.
• This approach can lead to more precise diagnoses and tailored treatment strategies.