Wang Yichen, Huang Yuting, Yeo Yee Hui, Pang Songhan, Ramai Daryl, Zheng Ting, Wang Yiming, Yan Yan, DeVault Kenneth R, Francis Dawn, Antwi Samuel O, Pang Maoyin
Division of Hospital Medicine, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PA, USA.
Division of Gastroenterology and Hepatology, Mayo Clinic, 4500 San Pablo Rd, Jacksonville, FL, USA.
Dig Dis Sci. 2025 Feb;70(2):675-684. doi: 10.1007/s10620-024-08823-w. Epub 2025 Jan 8.
Eosinophilic esophagitis (EoE) is an increasingly common cause of food impaction.
This study aims to provide a nationwide analysis of food impaction in patients with or without EoE diagnosis, concentrating on patient demographics, interventions, outcomes, and development of predictive machine-learning models.
A retrospective assessment was conducted using Nationwide Emergency Department Sample data from January 1, 2018, to December 31, 2019. We compared patients with food impaction with an associated EoE diagnosis to those without EoE and derived machine-learning models to predict EoE using International Classification of Diseases codes at discharge for identification.
Of 286,886,714 emergency department visits, 146,084 were for food impaction, with 7093 cases coinciding with an EoE diagnosis (4.9%). Patients with EoE were more commonly young men with fewer overall comorbidities but higher incidences of obesity, asthma, gastritis, and allergic rhinitis. A significantly larger proportion in the EoE group (89.6%) underwent esophagogastroduodenoscopy compared to the non-EoE group (51.1%; P < 0.001) and had a higher rate of biopsy during esophagogastroduodenoscopy in the emergency department (54.9% vs 13.4%; P < 0.001). Our machine-learning models, incorporating patient demographics, hospital attributes, and comorbidities, had a sensitivity of 86.1% and an area under the receiver operating characteristic curve of 0.828.
This nationwide study demonstrates that EoE in food impaction is associated with specific patient demographics, comorbidities, and elevated interventions. Our machine-learning models hold promise as screening tools for EoE, aiding medical practitioners in determining the need for biopsy.
嗜酸性粒细胞性食管炎（EoE）是食物嵌塞越来越常见的原因。
本研究旨在对有或无EoE诊断的患者的食物嵌塞情况进行全国性分析，重点关注患者人口统计学特征、干预措施、结局以及预测性机器学习模型的开发。
使用2018年1月1日至2019年12月31日的全国急诊科样本数据进行回顾性评估。我们将有相关EoE诊断的食物嵌塞患者与无EoE的患者进行比较，并使用出院时的国际疾病分类代码推导机器学习模型以预测EoE用于识别。
在286,886,714次急诊科就诊中，146,084次是因食物嵌塞，其中7093例与EoE诊断相符（4.9%）。EoE患者更常见于年轻男性，总体合并症较少，但肥胖、哮喘、胃炎和过敏性鼻炎的发病率较高。与非EoE组（51.1%；P < 0.001）相比，EoE组中接受食管胃十二指肠镜检查的比例显著更高（89.6%），且在急诊科进行食管胃十二指肠镜检查时活检率更高（54.9%对13.4%；P < 0.001）。我们纳入患者人口统计学特征、医院属性和合并症的机器学习模型，灵敏度为86.1%，受试者工作特征曲线下面积为0.828。
这项全国性研究表明，食物嵌塞中的EoE与特定的患者人口统计学特征、合并症和更高的干预率相关。我们的机器学习模型有望作为EoE的筛查工具，帮助医生确定活检的必要性。
