Academic Critical Care & Neurosurgery, Aberdeen Royal Infirmary, NHS Grampian.
Department of Psychiatry, Oxford Health NHS Foundation Trust.
Epileptic Disord. 2022 Apr 1;24(2):302-314. doi: 10.1684/epd.2021.1391.
Stroke is the commonest cause of epileptic seizures in older adults. Risk factors for post-stroke seizure (PSS) are well known, however, predicting PSS risk is clinically challenging. This study aimed to evaluate the predictive accuracy of PSS risk prediction models developed to date.
We performed a systematic review and meta-analysis of studies using MEDLINE and EMBASE from database inception to 28 December 2020. The search criteria included all peer-reviewed research articles, in which PSS risk prediction models were developed or validated for ischaemic and/or haemorrhagic stroke. Random-effects meta-analysis was used to generate summary statistics of model performance and receiver operating characteristic curves. Quality appraisal of studies was conducted using PROBAST.
Thirteen original studies involving 182,673 stroke patients (mean age: 38-74.9 years; 29.4-60.9% males), reporting 15 PSS risk prediction models were included. The incidence of early PSS (occurring ≤one week from stroke onset) and late PSS (occurring >one week from stroke onset) was 4.5% and 2.1%, respectively. Cortical involvement, functional deficits, increasing lesion size, early seizures, younger age, and haemorrhage were the commonest predictors across the models. SeLECT demonstrated greatest predictive accuracy (AUC 0.77 [95% CI: 0.71-0.82]) for late PSS following ischaemic stroke, and CAVE for predicting late PSS following haemorrhagic stroke (AUC 0.81 [0.76-0.86]). Fourteen of 15 studies demonstrated a high risk of bias, with lack of model validation and reporting of performance measures on calibration and discrimination being the commonest reasons.
Although risk factors for PSS are widely documented, this review identified few multivariate models with low risk of bias, synthetising single variables into an individual prediction of seizure risk. Such models may help personalise clinical management and serve as useful research tools by identifying stroke patients at high risk of developing PSS for recruitment into studies of anti-epileptic drug prophylaxis.
中风是老年人癫痫发作最常见的原因。中风后癫痫发作（PSS）的危险因素众所周知，但预测 PSS 风险在临床上具有挑战性。本研究旨在评估迄今为止开发的 PSS 风险预测模型的预测准确性。
我们对从数据库建立到 2020 年 12 月 28 日使用 MEDLINE 和 EMBASE 进行的研究进行了系统评价和荟萃分析。搜索标准包括所有同行评审的研究文章，其中开发或验证了缺血性和/或出血性中风的 PSS 风险预测模型。使用随机效应荟萃分析生成模型性能的汇总统计数据和接收者操作特征曲线。使用 PROBAST 对研究进行质量评估。
共纳入 13 项原始研究，涉及 182673 例中风患者（平均年龄：38-74.9 岁；男性占 29.4-60.9%），报告了 15 个 PSS 风险预测模型。早期 PSS（中风发作后≤1 周发生）和晚期 PSS（中风发作后>1 周发生）的发生率分别为 4.5%和 2.1%。皮质受累、功能缺陷、病变增大、早期发作、年龄较小和出血是各模型中最常见的预测因素。SELECT 模型对缺血性中风后晚期 PSS 的预测准确性最高（AUC 0.77[95%CI：0.71-0.82]），CAVE 模型对出血性中风后晚期 PSS 的预测准确性最高（AUC 0.81[0.76-0.86]）。15 项研究中有 14 项存在高偏倚风险，缺乏模型验证和报告校准和区分性能的措施是最常见的原因。
尽管 PSS 的危险因素已广泛记录，但本综述发现很少有低偏倚风险的多变量模型，将单一变量综合成个体癫痫发作风险预测。这些模型可以通过识别出发生 PSS 风险较高的中风患者，将其纳入抗癫痫药物预防研究中，从而帮助进行个体化临床管理，并作为有用的研究工具。
