# Prognostic factors and a preliminary prognostic model in anti-GAD antibody-associated epilepsy

## TL;DR

The study identifies factors influencing outcomes in anti-GAD antibody-associated epilepsy and develops a predictive model for prognosis.

## Contribution

A novel prognostic model using TLE and seizure frequency to predict outcomes in anti-GAD antibody-associated epilepsy is developed and validated.

## Key findings

- Temporal lobe epilepsy (TLE) and higher seizure frequency independently predict poor prognosis.
- Early diagnosis and treatment, along with higher CD8+T-cell counts, are linked to better outcomes.
- The developed model effectively stratifies patients into low, intermediate, and high-risk groups for poor prognosis.

## Abstract

Prognostic determinants in anti-glutamic acid decarboxylase (GAD) antibody-associated epilepsy remain unclear, and no validated predictive model exists. We aimed to identify prognostic factors and develop a predictive model.

This multicenter cohort included patients diagnosed with anti-GAD antibody-associated epilepsy before September 2024. Data encompassed demographics, seizure semiology, cellular and serological parameters, neuroimaging and electrophysiological findings, and treatment regimens. Favorable outcome was defined as seizure-free for ≥12 months following immunotherapy and antiseizure medications, poor outcome was defined as persistent seizures. Prognostic factors were analyzed and a predictive model was constructed.

Among 91 patients, 22 (24%) achieved seizure freedom, whereas 69 (76%) continued to experience seizures despite appropriate treatment. Poor prognosis was associated with focal seizures (50% vs. 81%, p = 0.004), temporal lobe epilepsy (TLE) (23% vs. 75%, p < 0.001), musicogenic epilepsy (n = 5, all with poor seizure control), and higher seizure frequency [≥1 seizure/month (67% vs. 97%, p < 0.001)]. In contrast, a shorter disease duration from symptom onset to diagnosis [3 (IQR 0.9–26.0) vs. 8 (IQR 1.5–36.0) months, p = 0.025], a shorter interval to initiation of immunotherapy [3 (IQR 1.0–14.0) vs. 7 (IQR 1.9–27.3) months, p = 0.005], higher CD8+T-cell counts (829.5 ± 473.9 vs. 619.5 ± 338.6 cells/µL, p = 0.035) were associated with favorable outcomes. Multivariate logistic regression identified TLE (OR = 0.098, 95% CI: 0.028–0.341, p < 0.001) and seizure frequency (OR = 0.067, 95% CI: 0.010–0.450, p = 0.005) as independent predictors of prognosis. The prognostic model based on these two variables demonstrated good discrimination (AUC = 0.807, 95% CI: 0.696–0.919, p < 0.001) and calibration (Hosmer–Lemeshow χ² = 0.124, p = 0.740), with sensitivity of 81.8%, specificity of 72.5%, and overall accuracy of 74.7%. Internal validation with bootstrapping confirmed model stability. Risk stratification further classified patients into low- (8.7%), intermediate- (49.3–58.9%), and high-risk (93.6%) groups for poor prognosis.

Focal seizures, TLE, and higher seizure frequency were associated with poor prognosis, whereas early diagnosis, timely treatment, and higher peripheral CD8+T-cell counts were associated with favorable outcomes. TLE and seizure frequency independently predicted clinical outcomes in anti-GAD antibody-associated epilepsy. The logistic regression model effectively stratified patients, identifying those likely to achieve seizure freedom versus refractoriness.