Department of Microbiology & Immunology, University of Iowa, Iowa City, IA, USA.
Stead Family Department of Pediatrics, University of Iowa, Iowa City, IA, USA.
Pediatr Res. 2022 Feb;91(3):590-597. doi: 10.1038/s41390-021-01570-y. Epub 2021 May 21.
Necrotizing enterocolitis (NEC) is a devastating intestinal disease of premature infants, with significant mortality and long-term morbidity among survivors. Multiple NEC definitions exist, but no formal head-to-head evaluation has been performed. We hypothesized that contemporary definitions would perform better in evaluation metrics than Bell's and range features would be more frequently identified as important than yes/no features.
Two hundred and nineteen patients from the University of Iowa hospital with NEC, intestinal perforation, or NEC concern were identified from a 10-year retrospective cohort. NEC presence was confirmed by a blinded investigator. Evaluation metrics were calculated using statistics and six supervised machine learning classifiers for current NEC definitions. Feature importance evaluation was performed on each decision tree classifier.
Newer definitions outperformed Bell's staging using both standard statistics and most machine learning classifiers. The decision tree classifier had the highest overall machine learning scores, which resulted in Non-Bell definitions having high sensitivity (0.826, INC) and specificity (0.969, ST), while Modified Bell (IIA+) had reasonable sensitivity (0.783), but poor specificity (0.531). Feature importance evaluation identified nine criteria as important for diagnosis.
This preliminary study suggests that Non-Bell NEC definitions may be better at diagnosing NEC and calls for further examination of definitions and important criteria.
This article is the first formal head-to-head evaluation of current available definitions of NEC. Non-Bell NEC definitions may be more effective in identifying NEC based on findings from traditional measures of diagnostic performance and machine learning techniques. Nine features were identified as important for diagnosis from the definitions evaluated within the decision tree when performing supervised classification machine learning. This article serves as a preliminary study to formally evaluate the definitions of NEC utilized and should be expounded upon with a larger and more diverse patient cohort.
坏死性小肠结肠炎（NEC）是一种严重的早产儿肠道疾病，幸存者的死亡率和长期发病率都很高。目前存在多种 NEC 定义，但尚未进行正式的直接比较评估。我们假设，与 Bell 分期相比，现代定义在评估指标上表现更好，范围特征比是/否特征更常被确定为重要特征。
从 10 年回顾性队列中确定了来自爱荷华大学医院的 219 名患有 NEC、肠穿孔或 NEC 疑似的患者。由盲法研究者确认 NEC 的存在。使用统计学和 6 种监督机器学习分类器计算当前 NEC 定义的评估指标。对每个决策树分类器进行特征重要性评估。
新定义在使用标准统计学和大多数机器学习分类器时均优于 Bell 分期。决策树分类器具有最高的整体机器学习评分，导致非 Bell 定义具有高敏感性（0.826，INC）和特异性（0.969，ST），而改良 Bell（IIA+）具有合理的敏感性（0.783），但特异性较差（0.531）。特征重要性评估确定了 9 个标准对诊断很重要。
这项初步研究表明，非 Bell NEC 定义可能更有助于诊断 NEC，并呼吁进一步检查定义和重要标准。
本文是首次对当前可用的 NEC 定义进行直接比较评估。基于传统诊断性能测量和机器学习技术的发现，非 Bell NEC 定义在识别 NEC 方面可能更有效。在执行监督分类机器学习时，从评估的定义中，决策树确定了 9 个特征对诊断很重要。本文作为正式评估 NEC 定义的初步研究，应该在更大、更多样化的患者队列中进行扩展。
