External validation and comparison of two prediction models for seizure recurrence after the withdrawal of antiepileptic drugs in adult patients

Abstract

Objective:The models currently available for predicting the risk of seizure recurrence after antiepileptic drug (AED) withdrawal in adult epilepsy patients include the prediction model developed by Lamberink et al (Lamberink model, 2017) and the Medical Research Council prediction model (MRC model, 1993). However, there was no external validation for the two models. The purpose of this study was to perform an independent external validation and a comparison of the Lamberink model and the MRC model in adult patients. Methods: The study population was recruited from the Wenzhou Epilepsy Follow-up Registry Database (WEFURD). All the predictors of the Lamberink and MRC models and the occurrence of seizure recurrence in the participants were collected based on the WEFURD. Participants' predicted probabilities of seizure recurrence were obtained by a Web-based tool and the prognostic index formula. The external validation of the Lamberink model and the MRC model were quantified by discrimination, calibration, and decision curve analysis (DCA). Results: Of 212 patients, 126 (59.4%) had seizure recurrence after AED withdrawal. The Lamberink 2-year model, the Lamberink 5-year model, the MRC 1-year model, and the MRC 2-year model had areas under the curve of 0.71 (95% confidence interval [CI] = 0.64-0.78), 0.68 (95% CI = 0.60-0.76), 0.60 (95% CI = 0.50-0.69), and 0.58 (95% CI = 0.50-0.66), respectively. Additionally, the Lamberink 2-year model had a significantly better integrated discrimination improvement than the MRC 2-year model (P < .001). Regarding calibration, the Lamberink 2-year model (P = .121) and the MRC 1-year model (P = .264) were well calibrated, but the Lamberink 5-year model (P = .022) and the MRC 2-year model (P = .008) were not. In the DCA, the Lamberink 2-year model performed well at threshold probabilities of 30%-65%. Significance: This external validation shows that the Lamberink 2-year model might be more accurate and has greater clinical benefit than others for guiding drug withdrawal in adult epilepsy clinics. 116 | LIN et aL. K E Y W O R D S adult epilepsy, drug withdrawal, external validation, prediction model, seizure recurrence Key Points • The Lamberink 2-year model had the highest AUC (0.71) among the models in the external validation • The Lamberink 2-year model (P = .121) and the MRC 1-year model (P = .264) were well calibrated, but the other prediction models were not • In the DCA, the Lamberink 2-year model performed well at threshold probabilities of 30%-65%

Prediction of the recurrence risk in patients with epilepsy after the withdrawal of antiepileptic drugs

“…The calculator for the risk of withdrawal is available at https://tinyurl.com/uxyu26p, which can be used to evaluate the 2year and 5-year recurrence risks after withdrawal and quantify the probability of achieving a seizure-free status at 10 years after withdrawal. In 2019, Lin et al [94] used a database to validate the prediction models proposed by Lamberink et al [33] and the MRC [88] with a cohort of 212 patients in all age groups who withdrew from AEDs and underwent long-term follow-up. The areas under the curves (AUCs) were 0.71 and 0.68 for the 2-year and 5-year models proposed by Lamberink et al, while AUCs were 0.60 and 0.58 for the 1-year and 2-year models from the MRC, respectively.…”

“…Current models for predicting the recurrence risk after withdrawal are accurate and convenient references; however, most models are only applicable to specific populations with epilepsy, and different variables are used to predict the recurrence risk after withdrawal in different patient populations. To date, only a few external validation studies have been conducted to independently validate the usefulness and accuracy of these prediction models [94]. In addition, no consensus has been established for withdrawal in patients with acute symptomatic epilepsy, and no predictive variables or models have been developed to predict the recurrence risk after AED withdrawal in these patients.…”

Driving eligibility: Implications of studies on seizure recurrence risk

“…Thus, neurologists can provide their patients with well‐reasonable data on an issue with a high impact in their lives. The results of Lamberink et al 8 for the 2‐year follow‐up have recently been confirmed in an independent cohort of patients 9 . An important limitation of the Lamberink model is that the risk of seizure relapse of patients staying on ASM could not be compared with the risk of those who were discontinued.…”

Prediction of seizure recurrence risk following discontinuation of antiepileptic drugs

“…A major obstacle to the practical implementation of LPM is represented by the lack of a single threshold probability value that separates high‐risk from low‐risk patients. Lin et al tried to overcome this issue by a decision curve analysis: This method, unfortunately, did not offer a specific cut‐off but it revealed that the model's usefulness resides in a specific probability range 20 . On the contrary, Chu et al proposed a cut‐off by calculating the largest Youden index on receiver‐operating characteristic (ROC) curves: at 2 years this value was 47% (with a sensitivity of 0.758 and a specificity of 0.410), whereas at 5 years it was 77% (sensitivity of 0.358 and a specificity of 0.979) 21 .…”