Predictive models for starting antiseizure medication withdrawal following epilepsy surgery in adults

• DOI: 10.1093/brain/awac437
• Publication Date: 2022-11-21T22:05:49Z
• Authors: Carolina Ferreira...; Jane de Tisi; Andrew W McEvoy; Anna Miserocchi; Jean Khoury; Ruta Yardi; Deborah T Vegh; James Butler; Hamin J Lee; Victoria Deli-Peri; Yi Yao; Feng-Peng Wang; Xiao-Bin Zhang; Lubna Shakhatreh; Pakeeran Siriratnam; Andrew Neal; Arjune Sen; Maggie Tristram; Elizabeth Varghese; Wendy Biney; William P Gray; Ana Rita Peralta; Alexandre Rainha-...; António J C Gonça...; José Pimentel; Juan Fernando Arias; Samuel Terman; Robert Terziev; Herm J Lamberink; Kees P J Braun; Willem M Otte; Fergus J Rugg-Gunn; Walter Gonzalez; Carla Bentes; Khalid Hamandi; Terence J O’Brien; Piero Perucca; Chen Yao; Richard J Burman; Lara Jehi; John S Duncan; Josemir W Sander; Matthias Koepp; Marian Galovic
• Keywords: Concordance

More than half of adults with epilepsy undergoing resective surgery achieve long-term seizure freedom and might consider withdrawing antiseizure medications. We aimed to identify predictors recurrence after starting postoperative medication withdrawal develop validate predictive models. performed an international multicentre observational cohort study in nine tertiary referral centres. included 850 who started following were free seizures other focal non-motor aware before withdrawal. developed a model predicting recurrent seizures, using Cox proportional hazards regression derivation (n = 231). Independent recurrence, the start [adjusted hazard ratio (aHR) 5.5, 95% confidence interval (CI) 2.7-11.1], history bilateral tonic-clonic (aHR 1.6, CI 0.9-2.8), time from 0.9, 0.8-0.9) number medications at 1.2, 0.9-1.6). Model discrimination showed concordance statistic 0.67 (95% 0.63-0.71) external validation cohorts 500). A secondary any (including seizures) was validated subgroup that did not have 639), showing 0.68 0.64-0.72). Calibration plots indicated high agreement predicted observed outcomes for both show simple algorithms, available as graphical nomograms online tools (predictepilepsy.github.io), can provide probabilities These multicentre-validated models may assist clinicians when discussing their patients.