Development and Validation of a Deep Learning Model for Predicting Treatment Response in Patients With Newly...

Area of Science:

• Neurology
• Artificial Intelligence
• Pharmacogenomics

Background:

• Epilepsy treatment selection is often a trial-and-error process, leading to delays in effective therapy.
• Many patients experience sequential trials of ineffective antiseizure medications (ASMs).

Purpose of the Study:

• To develop and validate a deep learning model for predicting individual patient response to the first ASM.
• To utilize readily available clinical information for personalized epilepsy treatment prediction.

Main Methods:

• A transformer deep learning model was developed using 16 clinical factors and ASM data from a large, multi-national cohort.
• The model was trained on pooled data and validated externally on separate cohorts.
• Performance was assessed using area under the receiver operating characteristic curve (AUROC) and balanced accuracy.

Main Results:

• The transformer model trained on pooled data achieved an AUROC of 0.65 and weighted balanced accuracy of 0.62.
• Models trained on individual cohorts showed lower predictive performance upon external validation.
• Key predictors included seizure frequency, psychiatric disorders, EEG, and brain imaging.

Conclusions:

• A deep learning model demonstrates feasibility in predicting ASM response using clinical data.
• This approach shows potential to personalize epilepsy treatment and improve initial drug selection.
• Future improvements may involve integrating genetic and advanced imaging data.