BACKGROUND Most children with Benign epilepsy with centro-temporal spikes (BECTS) undergo remission during late adolescence and do not require treatment. In a small group of patients, the condition may evolve to encephalopathic syndromes including epileptic encephalopathy with continuous spike-and-wave during sleep (ECSWS), or Landau-Kleffner Syndrome (LKS). Development of prediction models for early identification of at-risk children is of utmost importance. AIM To develop a predictive model of encephalopathic transformation using data-driven approaches, reveal complex interactions to identify potential risk factors. METHODS Data were collected from a cohort of 91 patients diagnosed with BECTS treated between the years 2005-2017 at a pediatric neurology institute. Data on the initial presentation was collected based on a novel BECTS ontology and used to discover potential risk factors and to build a predictive model. Statistical and machine learning methods were compared. RESULTS A subgroup of 18 children had encephalopathic transformation. The least absolute shrinkage and selection operator (LASSO) regression Model with Elastic Net was able to successfully detect children with ECSWS or LKS. Sensitivity and specificity were 0.83 and 0.44. The most notable risk factors were fronto-temporal and temporo-parietal localization of epileptic foci, semiology of seizure involving dysarthria or somatosensory auras. CONCLUSION Novel prediction model for early identification of patients with BECTS at risk for ECSWS or LKS. This model can be used as a screening tool and assist physicians to consider special management for children predicted at high-risk. Clinical application of machine learning methods opens new frontiers of personalized patient care and treatment.

中文翻译：

---

伴有中央颞区棘波的良性癫痫患儿脑病转化的早期预测

背景 大多数患有中心颞叶尖峰 (BECTS) 良性癫痫的儿童在青春期后期会得到缓解，不需要治疗。在一小部分患者中，病情可能会演变为脑病综合征，包括癫痫性脑病伴睡眠期间持续的棘波 (ECSWS) 或 Landau-Kleffner 综合征 (LKS)。开发早期识别高危儿童的预测模型至关重要。目的 使用数据驱动的方法开发脑病转化的预测模型，揭示复杂的相互作用以识别潜在的风险因素。方法 从 2005 年至 2017 年间在儿科神经病学研究所接受治疗的 91 名诊断为 BECTS 的患者中收集数据。初始演示的数据是基于新的 BECTS 本体收集的，用于发现潜在的风险因素并构建预测模型。比较了统计方法和机器学习方法。结果 一个由 18 名儿童组成的亚组患有脑病转化。使用 Elastic Net 的最小绝对收缩和选择算子 (LASSO) 回归模型能够成功检测出患有 ECSWS 或 LKS 的儿童。敏感性和特异性分别为 0.83 和 0.44。最显着的危险因素是癫痫病灶的额颞叶和颞顶叶定位、涉及构音障碍或躯体感觉先兆的癫痫发作的符号学。结论 用于早期识别有 ECSWS 或 LKS 风险的 BECTS 患者的新型预测模型。该模型可用作筛查工具并协助医生考虑对预测为高危儿童的特殊管理。机器学习方法的临床应用开辟了个性化患者护理和治疗的新领域。