External validation of the AntiEpileptic Drug Monitoring in PREgnancy (EMPiRE) model for predicting seizures in pregnant women with epilepsy

BACKGROUND: The AntiEpileptic Drug Monitoring in PREgnancy (EMPiRE) model is the only available tool for predicting seizures in pregnant women with epilepsy (WWE) using anti-seizure medications (ASMs); however, its predictive performance requires validation. This study aimed to evaluate the predictive ability of this model in pregnant Chinese WWE and its potential usefulness in clinical practice. METHODS: Data of the EMPiRE model were derived from the EMPiRE study, a prospective multicenter cohort study that recruited women on ASM monotherapy (lamotrigine, carbamazepine, phenytoin or levetiracetam) or polytherapy (lamotrigine with either carbamazepine, phenytoin or levetiracetam). Based on the applicable population of the EMPiRE model, we evaluated 280 patients registered in the Wenzhou Epilepsy Follow-up Registry Database from January 1, 2010, to December 31, 2020. A total of 158 eligible patients were included in the validation cohort. We collected data on the baseline characteristics of patients, eight predictors of the EMPiRE model and outcome events. The outcome was the occurrence of tonic-clonic or non-tonic-clonic seizures at any time in pregnancy up to 6 weeks postpartum. We used the equation of the EMPiRE model to obtain the predicted probabilities of seizures. The predictive ability of the EMPiRE model was quantified by the C-statistic (scale 0–1, values > 0.5 show discrimination), GiViTI calibration test and decision curve analysis (DCA). RESULTS: Of 158 eligible patients, 96 patients (60.8%, 96/158) experienced one or more seizures at any time between pregnancy and 6 weeks postpartum. The EMPiRE model showed good discrimination with a C-statistic of 0.76 (95% confidence interval [CI] 0.70–0.84). The GiViTI calibration belt showed that the predicted probabilities, which ranged from 16 to 96% (95% CI), were lower than the actual probabilities. DCA indicated that the highest net proportional benefit was obtained for predicted probability thresholds of 15–18% and 54–96%. CONCLUSIONS: The EMPiRE model could discriminate well between WWE with and without seizures during pregnancy and 6 weeks postpartum, but the risk of seizures may be underestimated. The limitations of the model for specific medication regimens may limit its real-world application. If the model is further improved, it will be incredibly valuable. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12884-023-05822-z.