Can epileptic seizures be predicted in pregnancy?

Women with epilepsy are more than 10 times likelier to die during pregnancy than those who do not have the disorder. A new study published in PLOS Medicine suggests that a prognostic model may help predict which women are likely to suffer seizures and be at high risk of adverse events from epilepsy.

Five hundred twenty-seven pregnant women with epilepsy and on medication for it were recruited from 50 hospitals in the UK. Of the women, 399 had their antiepileptic drug dosage adjusted based on clinical features only (model development cohort) and 128 women had adjustments informed by serum drug levels (validation cohort). 

Average gestational age at baseline was 16.6 weeks (SD 4.0) in the development cohort and 14.9 weeks (SD 4.4) in the validation cohort. Mean age at first seizure was approximately 16 years in both cohorts and 10% to 15% of women had a learning difficulty or mental health disorder. Forty-six percent of women in the development cohort and 39% of women in the validation cohort had experienced seizures in the 3 months prior to pregnancy.

The primary outcome of the study was epileptic (non-eclamptic) seizure, documented using diary records. The model’s predictive performance was assessed using measures of discrimination (C-statistic) and accuracy (calibration slope).

Overall, 46% of women in the development cohort experienced one or more seizures at any time from baseline until 6 weeks after delivery. Approximately half of all seizures were tonic-clonic (90/183, 49%). The authors observed eight predictors significantly associated with seizures, which were included in the final multivariable model: age at first seizure, history of mental health disorder or learning difficulty, baseline seizure classification (tonic-clonic, non-tonic-clonic, unspecified), hospital admission for seizures in a previous pregnancy, tonic-clonic seizure in the 3 months before pregnancy, non-tonic-clonic seizure in the 3 months before pregnancy, baseline doses of lamotrigine and levetiracetam.

The prediction model had a C-statistici of 0.79 (95% CI 0.75-0.84) for discriminating between women with and without seizures. (The C-statistic can be summarized as the area under the receiver operating characteristic curve and is a measure of goodness of fit for binary outcomes in a logistic regression model.) On external validation, the model showed good performance (C-statistic 0.76, 95% CI 0.66-0.85; calibration slope 0.93, 95% CI 0.44, 1.41) although the estimates were imprecise. The EMPiRE model showed the highest net proportional benefit for predicted probability thresholds between 12% and 99%. 

The authors believe the EMPiRE model performed well in predicting risk of seizures in pregnant women with epilepsy who are prescribed antiepileptic medication. It is clinically useful over a range of threshold probabilities and is relevant to general practitioners, epilepsy specialists, obstetricians, and midwives. An online risk calculator is available for use here.