腺病毒感染患儿急性胃肠炎表型列线图预测模型的建立及验证

摘要：目的　构建腺病毒感染患儿发生急性胃肠炎表型的列线图预测模型，并验证其可行性。方法　选取腺病毒感染患儿144例为训练集，其中40例（27.8%）为急性胃肠炎表型，104例（72.2%）为呼吸道表型。收集患儿年龄、性别、发热天数、扁桃体渗出情况、结膜充血状态、入院后丙氨酸转氨酶（ALT）、天冬氨酸转氨酶（AST）、γ-谷氨酰转肽酶（γ-GT）、白细胞介素（IL）-6、乳酸脱氢酶（LDH）、CD4＋T细胞百分比、CD8＋T细胞百分比及CD4＋T细胞/CD8＋T细胞比值。采用最小绝对值收敛和选择算子（LASSO）回归筛选急性胃肠炎表型的预测因子，多因素Logistic回归分析建立风险预测模型并绘制列线图。采用受试者工作特征（ROC）曲线、Brier评分分别评价模型的判别能力和校准能力。另选取35例腺病毒感染患儿对预测模型进行外部验证。结果　与呼吸道表型组相比，胃肠炎表型组患儿年龄偏小，女性比例、ALT水平较高，发热天数较短，IL-6、LDH水平较低（P＜0.05）。LASSO回归分析显示年龄、性别、发热天数、LDH是急性胃肠炎表型的预测因子。多因素Logistic回归分析显示女性、年龄小、发热时间短、LDH降低是儿童HAdV感染胃肠炎表型的危险因素。纳入上述因素构建的列线图模型的ROC曲线下面积（AUC）为0.951（95% CI：0.923~0.984），敏感度为82.50%，特异度为95.19%。外部验证集中，35例患儿中8例为急性胃肠炎表型，AUC为0.925（95% CI：0.821~0.998），敏感度为75.42%，特异度为95.90%。结论　本研究建立的预测模型适用于腺病毒感染患儿出现急性胃肠炎表型的早期诊断，预测效果较好，具有一定的临床价值。

Establishment and validation of phenotypic line graph model for predicting acute gastroenteritis in children with adenovirus infection

Abstract: Objective　To construct a line graph prediction model for phenotype of acute gastroenteritis in pediatric patients with adenovirus infection and to verify its feasibility. Methods　A total of 144 children infected with adenovirus were selected as the training set, of which 40 (27.8%) showed acute gastroenteritis phenotype and 104 (72.2%) showed respiratory phenotype. The age, sex, days of fever, tonsil exudation, conjunctival congestion, alanine aminotransferase (ALT), aspartate aminotransferase (AST), γ-glutamyltransferase (γ-GT), interleukin (IL-6), lactate dehydrogenase (LDH), percentage of CD4+T cells, CD8+T cells and CD4+cell /CD8+T cell ratio were collected after admission. Predictors of acute gastroenteritis phenotypes were screened using least absolute shrinkage selection (LASSO) regression. Multifactorial Logistic regression analysis was used to establish a risk prediction model to plot column line graphs. Subject working characteristic (ROC) curves and Brier scores were used to evaluate the discriminatory ability and calibration ability of the model, respectively. An additional 35 pediatric patients with adenovirus infection were selected for external validation of the prediction model. Results　Among 144 children infected with adenovirus, 40 cases (27.8%) had acute gastroenteritis phenotype and 104 cases (72.2%) had respiratory phenotype. Compared with the respiratory tract phenotype group, the children in the gastroenteritis phenotype group were younger, with higher proportion of female patients, higher ALT levels, shorter fever days, and lower levels of IL-6 and LDH (P＜0.05). LASSO regression analysis revealed that the age, sex, days of fever and LDH were predictors of acute gastroenteritis phenotype. Multivariate Logistic regression analysis showed that female, young age, short fever time and low LDH were the risk factors of gastroenteritis phenotype of HAdV infection inchildren. The area under the ROC (AUC) curve of the column diagram model built with the above factors was 0.951 (95%CI: 0.923-0.984), the sensitivity was 82.50%, the specificity was 95.19%. In the external validation set, 8 of the 35 patients had acute gastroenteritis phenotype, the AUC was 0.925 (95%CI: 0.821-0.998), sensitivity was 75.42%, and specificity was 95.90%. Conclusion　The prediction model established in this study is suitable for the early diagnosis of acute gastroenteritis phenotype in children with adenovirus infection, with good prediction effect and certain clinical value.