Objective: To construct a diagnostic prediction model for childhood asthma and conduct a preliminary evaluation based on the test results of specific IgE (sIgE) for airborne allergens and in combination with clinical data. Methods: This study is a case-control study. A total of 4 338 cases that completed the sIgE test for airborne allergens in the Allergy Department of Beijing Children's Hospital Affiliated to Capital Medical University from January to December 2023 were selected as the research subjects. They were divided into the asthma group and the non-asthma group based on the diagnostic information. Age, gender, cough and wheezing symptoms, and the classification results of sIgE concentrations of 15 airborne allergens were collected as the predictor variables of the asthma diagnostic prediction model. Differential analysis and LASSO regression were employed for the screening of predictor variables. The multivariate logistic regression method was applied to construct the nomogram prediction model. The data set was randomly split at a ratio of 7∶3 into a training set (3 036 cases) for constructing the prediction model and a validation set (1 302 cases) for testing the predictive efficacy of the model. The area under the receiver operating characteristic (ROC) curve (AUC), the Hosmer-Lemeshow calibration curve were utilized to assess the discrimination and goodness of fit of the model, and the clinical decision curve (DCA) was adopted to evaluate the clinical application value of the model. Results: Among 4 338 pediatric cases, children aged 0 to <3 years accounted for 10.17% (441 cases), those aged 3 to <6 years accounted for 36.49% (1 583 cases), those aged 6 to <12 years accounted for 46.98% (2 038 cases), and those aged 12 to 18 years accounted for 6.36% (276 cases). Males constituted 65.17% (2 827 cases), and females 34.83% (1 511 cases). The proportion of children without wheezing symptoms was 41.47% (1 799 cases), while those with wheezing symptoms was 58.53% (2 539 cases). The asthma group accounted for 41.77% (1 812 cases), and the non-asthma group for 58.23% (2 526 cases). Statistically significant differences were observed between the asthma group and the non-asthma group in 18 predictive variables including age, gender, wheezing symptoms, d1, d2, e1, e5, g2, g6, m6, t11, t3, t6, w1, w22, w6, wx5, and m3 (P<0.05). LASSO regression analysis identified six predictor variables: age (calculated in months), cough and wheezing symptoms, and sIgE of four airborne allergens, namely, Dermatophagoides pteronyssinus (d1), Canis familiaris dander (e5), Aspergillus fumigatus (m3), and Artemisia vulgaris pollen (w6).Multifactorial regression analysis revealed that the contribution degrees of the above-mentioned predictor variables to the asthma diagnosis prediction model were ranked as follows: cough and wheezing symptoms (OR=24.37, P<0.001), m3 (OR=1.34, P<0.001), d1 (OR=1.22, P<0.001), e5 (OR=1.12, P=0.028), w6 (OR=1.11, P<0.001), and age (OR=1.01, P<0.001).The AUCs of the nomogram prediction model for the training set and the validation set were 0.853 (95%CI: 0.840-0.866) and 0.838 (95%CI: 0.817-0.860), respectively. The Hosmer-Lemeshow calibration curve indicated a good fit (P=0.215 for the training set; P=0.352 for the validation set). The DCA of the validation set demonstrated that when the probability threshold for predicting the occurrence of childhood asthma was 8%-92%, the model had the best applicability. Conclusion: By combining age, cough and wheezing symptoms, and sIgE of the four airborne allergens (d1, e5, m3, and w6) selected from 15 airborne allergens, a childhood asthma diagnosis prediction model with good predictive performance and clinical practicability was constructed. It can serve as a simple and convenient tool for accurately identifying asthma and provides a practical basis for the application of artificial intelligence big data analysis models in the prevention, treatment, and management of childhood asthma.
目的： 基于气传过敏原特异性IgE（sIgE）检测结果，结合临床资料，构建儿童哮喘诊断预测模型并进行初步评价。 方法： 采用病例对照研究，选取2023年1—12月在首都医科大学附属北京儿童医院过敏反应科完成气传过敏原sIgE检测的4 338例病例作为研究对象（男性2 827例，女性1 511例），根据诊断信息将其分为哮喘组（1 812例）和非哮喘组（2 526例），收集其年龄、性别、咳喘症状、15种气传过敏原sIgE浓度分级结果作为哮喘诊断预测模型的预测变量。采用差异性分析及LASSO回归进行预测变量的筛选。应用多因素logistic回归方法构建列线图预测模型，通过简单随机拆分法以7∶3随机拆分数据集为训练集（3 036例）用于构建预测模型，验证集（1 302例）用于测试模型的预测效能。分别采用受试者工作特征（ROC）曲线下面积（AUC）、Hosmer-Lemeshow校准曲线评估模型的区分度、拟合度，并应用临床决策曲线（DCA）评估模型的临床应用价值。 结果： 4 338例患儿中，0~<3岁患儿占10.17%（441例），3~<6岁患儿占36.49%（1 583例），6~<12岁患儿占46.98%（2 038例），12~18岁患儿占6.36%（276例）。男性占65.17%（2 827例），女性占34.83%（1 511例）。无咳喘症状者占41.47%（1 799例），有咳喘症状者占58.53%（2 539例）。哮喘组占41.77%（1 812例），非哮喘组占58.23%（2 526例）。哮喘组与非哮喘组在年龄、性别、咳喘症状、d1、d2、e1、e5、g2、g6、m6、t11、t3、t6、w1、w22、w6、wx5、m3等共18个预测变量间差异均有统计学意义（P<0.05）。LASSO回归分析筛选出6个预测变量：年龄（以月龄计算）、咳喘症状，以及户尘螨（d1）、狗皮屑（e5）、烟曲霉（m3）、艾蒿花粉（w6）4种气传过敏原的sIgE。多因素回归分析显示上述6个预测变量对哮喘诊断预测模型贡献度排序依次为咳喘症状（OR=24.37，P<0.001）、m3（OR=1.34，P<0.001）、d1（OR=1.22，P<0.001）、e5（OR=1.12，P=0.028）、w6（OR=1.11，P<0.001）、年龄（OR=1.01，P<0.001）。列线图预测模型的训练集和验证集的AUC分别为0.853（95%CI=0.840~0.866）和0.838（95%CI=0.817~0.860），Hosmer-Lemeshow校准曲线显示拟合度良好（训练集P=0.215，验证集P=0.352）。验证集DCA显示，列线图模型预测儿童哮喘发生的概率阈值为8%~92%时，该模型的适用性最佳。 结论： 结合年龄、咳喘症状，从15种气传过敏原中筛选出的d1、e5、m3、w6这4种气传过敏原sIgE，构建了预测性能和临床实用性较好的儿童哮喘诊断预测模型，可作为早期准确识别哮喘的简易便捷的工具，为人工智能大数据分析模型在儿童哮喘防治及管理中的应用提供了实践基础。.