Developing a multivariable deep learning model to predict psychiatric illness in patients with epilepsy

Area of Science:

• Neurology
• Psychiatry
• Artificial Intelligence

Background:

• Psychiatric disorders are highly prevalent in epilepsy patients, worsening quality of life and treatment outcomes.
• Early identification of psychiatric comorbidities in epilepsy is essential for timely intervention.

Purpose of the Study:

• To develop and validate neural network models for predicting psychiatric disorders in epilepsy patients.
• To leverage clinical and demographic data for risk assessment.

Main Methods:

• Retrospective analysis of 2,258 epilepsy patients' data (2013-2023).
• Development of neural network models using keras and neuralnet in R.
• Evaluation of model performance using accuracy, sensitivity, specificity, and ROC-AUC; feature importance assessed via SHAP values.

Main Results:

• 27.6% of epilepsy patients exhibited psychiatric disorders.
• The neuralnet model achieved 97.16% accuracy (ROC-AUC: 0.974), outperforming the keras model (92.4% accuracy, ROC-AUC: 0.973).
• Key predictors included age of onset, seizure duration, and antiseizure drug profiles.

Conclusions:

• Neural network models accurately predict psychiatric comorbidities in epilepsy.
• These predictive tools can facilitate early identification of at-risk individuals.
• The models support personalized care strategies to enhance epilepsy management.