[Predictive model of curative effect of mite subcutaneous immunotherapy in 5-18 years of age patients with allergic asthma].

Objective: To analyze the factors affecting the efficacy of mite subcutaneous immunotherapy (SCIT) in allergic asthma patients aged 5-18 years, and to find the best predictive model for the curative effect. Methods: The data of 688 patients aged 5-18 years with allergic asthma who completed more than 3 years of mite SCIT from December 2006 to November 2021 in the Department of Respiratory Medicine, Children's Hospital Affiliated to Nanjing Medical University were retrospectively analyzed. Male, results of skin prick test (SPT), age, daily medication score (DMS), visual analogue scale (VAS) score, and enrollment season were defined as independent variables. R language models, including Logistic regression model, random forest model and extreme gradient boosting (XGboost) model, were used to analyze the impact of these independent variables on the outcomes. The receiver operating characteristic curve was applied to compare the predictive ability of the models. Hypothesis testing of the area under curve (AUC) of the 3 models was performed using DeLong test. Results: There were 435 males and 253 females in the 688 patients. There were 349 patients aged 5-<8 years, 240 patients aged 8-<11 years, and 99 patients aged 11-18 years. SPT showed that 429 cases (62.4%) were only allergic to mite, and 259 cases (37.7%) were also allergic to other allergens. According to the efficacy after 3 years of SCIT, 351 cases (51.0%) discontinued the treatment and 337 cases (49.0%) required continued treatment. The DMS was 4 (3, 6) at initiation, 3 (2, 5) at 3 months, 3 (2, 5) at 4 months, 2 (1, 3) at 12 months, and 0 (0, 1) at 3 years of SCIT treatment. The VAS was 3.5 (2.5, 5.2) at initiation, 3.2 (2.2, 4.8) at 3 months, 2.6 (1.4, 4.1) at 4 months, 1.0 (0.6, 1.8) at 12 months, and 0.5 (0, 1.2) at 3 years of treatment. At 3, 4, and 12 months, the rate of decline in DMS was 0 (0, 20%), 16.7% (0, 33.3%), and 50.0% (31.0%, 75.0%), respectively; and the VAS decreased by 7.1% (3.2%,13.8%), 27.6% (16.7%,44.4%), and 70.2% (56.1%, 82.3%), respectively. Regarding the enrollment season, 99 cases were in spring, 230 cases in summer, 171 cases in autumn, and 188 cases in winter. The R language Logistic regression model found that DMS>3 points at 3 months (OR=-3.5, 95%CI:-4.3--2.7, P<0.01), male (OR=-1.7, 95%CI:-2.3--1.0), P<0.01), DMS decline rate>16.7% at 4 months (OR=-1.6, 95%CI:-2.3--0.8, P<0.01) and DMS decline rate>0 at 3 months (OR=-0.7, 95%CI:-1.3--0.2, P<0.05) had higher possibility of drug discontinuation; whereas, the decline rate of DMS at 12 months>50.0% (OR=0.7, 95%CI: 0.1-1.3, P<0.05), VAS at 12 months>1.0 points (OR=0.9, 95%CI: 0.3-1.6, P<0.05), and initial VAS<4.0 points (OR=1.0, 95%CI: 0.4-1.6, P<0.01) had lower possibility of drug discontinuation. Both the random forest model and the XGboost model showed that DMS>3 points at 3 months (mean decrease accuracy=30.9, importance=0.45) had the greatest impact on drug discontinuation. The AUC of the random forest model was the largest at 0.900, with an accuracy of 78.2% and a sensitivity of 84.5%. Logistic regression model had AUC of 0.891, accuracy of 80.0%, and sensitivity of 80.0%; XGboost model had AUC of 0.886, accuracy of 76.9%, and sensitivity of 84.5%. The AUC of the pairwise comparison model by DeLong test found that all three models could be used for the prediction of this data set (all P>0.05). Conclusions: The more drugs used to control the primary disease, and the more careful reduction of the control medicine after starting SCIT treatment, the more favorable it is to stop all drugs after 3 years. The random forest model is the best predictive model for the efficacy of mite SCIT in asthmatic children.目的： 分析影响5~18岁过敏性哮喘患儿螨皮下特异性免疫治疗（SCIT）效果的因素并寻找最佳预测模型。 方法： 回顾性分析南京医科大学附属儿童医院呼吸科2006年12月至2021年11月完成3年以上螨SCIT的688例5~18岁过敏性哮喘患儿的资料，根据疗效分为停药组和未停药组，定义男性、皮肤点刺结果、年龄、每日药物评分（DMS）、视觉模拟量表（VAS）得分和入组季节为自变量，使用R语言建模（Logistic回归、随机森林和极端梯度上升模型）分析自变量对结局的影响，使用受试者工作特征曲线比较3种模型预测能力，应用德隆检验进行3种模型曲线下面积（AUC）的假设检验。 结果： 688例过敏性哮喘患儿中男435例、女253例。就诊年龄5~<8岁349例，8~<11岁240例，11~18岁99例。皮肤点刺试验单一螨过敏429例（62.4%）、螨为主且合并其他过敏259例（37.7%）。3年后停药组351例（51.0%）、未停药组337例（49.0%）。DMS 初始时4（3，6）分，3个月3（2，5）分，4个月3（2，5）分，12个月2（1，3）分，3年0（0，1）分。VAS 初始时3.5（2.5，5.2）分，3个月3.2（2.2，4.8）分，4个月2.6（1.4，4.1）分，12个月1.0（0.6，1.8）分，3年0.5（0，1.2）分。3、4、12个月时DMS较初始时下降率分别为0（0，20%），16.7%（0，33.3%），50.0%（31.0%，75.0%）；VAS则分别为7.1%（3.2%，13.8%），27.6%（16.7%，44.4%），70.2%（56.1%，82.3%）。688例过敏性哮喘患儿起始治疗季节中春季99例、夏季230例、秋季171例、冬季188例。R语言Logistic回归模型发现DMS 3个月>3分（OR=-3.5，95%CI：-4.3~-2.7，P<0.01）、男性（OR=-1.7，95%CI：-2.3~-1.0，P<0.01）、DMS 4个月下降率>16.7%（OR=-1.6，95%CI：-2.3~-0.8，P<0.01）、DMS 3个月下降率>0（OR=-0.7，95%CI：-1.3~-0.2，P<0.05）对停药可能性大；DMS 12个月下降率>50.0%（OR=0.7，95%CI：0.1~1.3，P<0.05）、VAS 12个月>1.0分（OR=0.9，95%CI：0.3~1.6，P<0.05）、初始VAS<4.0分（OR=1.0，95%CI：0.4~1.6，P<0.01）对停药可能性小。随机森林模型和极端梯度上升模型均显示DMS 3个月>3分（平均减少准确度=30.9、重要性=0.45）对停药影响力最大。随机森林模型AUC 0.900、精确度78.2%、灵敏度84.5%，Logistic回归模型分别为0.891、80.0%、80.0%；极端梯度上升模型0.886、76.9%、84.5%。德隆检验分别比较3种模型AUC均可用于该数据集的预测（均P>0.05）。 结论： 过敏性哮喘需要使用越多的药物控制原发病、开始螨SCIT后越缓慢减少原发病的治疗药物，越有利于3年后停用所有药物。随机森林模型为过敏性哮喘螨皮下免疫治疗效果的最佳预测模型。.