Multidimensional Early Prediction Score for Drug-Resistant Epilepsy.

Area of Science:

• Neurology
• Genetics
• Medical Informatics

Background:

• Early referral for preoperative examinations is crucial for favorable outcomes in drug-resistant epilepsy (DRE).
• Developing user-friendly prediction models is essential for nonexpert utilization in clinical practice.

Purpose of the Study:

• To investigate the feasibility of a user-friendly early prediction model for drug-resistant epilepsy (DRE).
• To create a model that is easily applicable by nonexperts for DRE risk assessment.

Main Methods:

• A two-step genotype analysis involving whole-exome sequencing (WES) and target sequencing was performed on initial (n=243) and validation (n=311) sets.
• A DRE risk prediction model was developed using 11 genetic and 2 clinical predictors from a multicenter case-control study.
• Early prediction scores for DRE (EPS-DRE) were calculated for genetic (EPS-DREgen), clinical (EPS-DREcln), and mixed (EPS-DREmix) predictors.

Main Results:

• The multidimensional EPS-DREmix model demonstrated a better fit to outcome data compared to unidimensional EPS-DREgen or EPS-DREcln.
• The EPS-DREmix model, using 11 genetic and 2 clinical predictors, showed good discrimination between DRE and drug-responsive epilepsy.
• Model performance was validated using an independent, unrelated dataset.

Conclusions:

• The EPS-DREmix model shows promise for early DRE prediction.
• This tool is user-friendly and suitable for real-world clinical application, particularly for nonexperts.
• Further research is recommended to enhance the performance of the EPS-DREmix model.