中重度OSAS儿童腺样体扁桃体切除术后持续残留的预测模型构建与鉴定  

Construction and identification of predictive model for persistent OSAS after adenotonsillectomy in children with moderate to severe OSAS

作　　者：高淑蔚 冯国双[1] 吴云肖[1] 郑莉[1] 郭永丽[1] 许志飞[1] GAO Shuwei;FENG Guoshuang;WU Yunxiao;ZHENG Li;GUO Yongli;XU Zhifei(Beijing Children’s Hospitol,Capital Medical University,Beijing 100045,China)

机构地区：[1]首都医科大学附属北京儿童医院,北京100045

出　　处：《山东医药》2020年第22期18-22,共5页Shandong Medical Journal

基　　金：首都卫生发展科研专项项目(首发2018-1-2091);北京市医院管理局儿科学科协同发展中心“儿科专项”一般项目(XTYB201807)。

摘　　要：目的构建中重度阻塞性睡眠呼吸暂停综合征(OSAS)患儿腺样体扁桃体切除术后持续残留的预测模型,并鉴定其预测效能。方法选择行腺样体扁桃体切除术的中重度OSAS患儿75例,以术后6个月时整夜多导睡眠图(PSG)监测阻塞性呼吸暂停低通气指数(OAHI)≥1次/h为标准,将患儿分为持续性OSAS组44例、非持续性OSAS组31例。收集两组术前和术后6个月时整夜PSG监测指标,以五折交叉验证法筛选模型参数,通过分类树模型、节点增益分析构建最优模型,通过受试者工作特征(ROC)曲线评估预测模型的效能。结果以呼吸暂停低通气指数(AHI)、OAHI、混合性呼吸暂停指数(MAI)、快速动眼睡眠期氧减指数、非快动眼睡眠期氧减指数(ODINREM)、平均血氧饱和度(mean SpO2)、觉醒指数、总睡眠时间、睡眠效率及各个睡眠时期占总睡眠时间的百分比等PSG监测指标作为预测因素,构建分类树模型。该分类树模型共有5层、15个节点、4个终末节点,共筛选出MAI、min SpO2、AHI、ODINREM、睡眠效率、呼吸觉醒指数6个解释变量。分类树模型预测持续性OSAS的ROC曲线下面积为0.96(95%CI:0.90~0.98),其预测敏感性为83.3%、特异性为94.1%、准确性为88.6%。结论构建的分类树预测模型能够快速有效地筛选中重度OSAS患儿腺样体扁桃体切除术后持续残留的风险因素,从而为临床筛选术后需要随访治疗的人群提供依据。Objective To establish a predictive model of persistent OSAS after adenotonsillectomy in children with moderate to severe obstructive sleep apnea syndrome(OSAS)and to evaluate its predictive efficacy.Methods Seventy-five children with moderate to severe OSAS were collected in this study.According to the monitoring of obstructive apnea hypopnea index(OAHI)≥1 time/h by polysomnography(PSG)at 6 months after operation,the children were divided into persistent OSAS group(44 cases)and non-persistent OSAS group(31 cases).The PSG monitoring indexes of the two groups before and 6 months after operation were recorded.The parameters of the model were screened by 50%cross validation.The optimal model was constructed by classification tree model and node gain analysis.The effectiveness of the prediction model was evaluated by ROC curve.Results The PSG monitoring indexes such as AHI,OAHI,MAI,ODI of REM,ODI of NREM,mean SpO2,arousal index,total sleep time(TST),sleep efficiency and the percentage of each sleep period in the total sleep time were used as predictors to construct the classification tree model.The classification tree had 5 layers,15 nodes and 4 end nodes.Six explanatory variables MAI,min SpO2,AHI,ODI of NREM,sleep efficiency and respiratory arousal index were selected out.The area under ROC curve of classification tree model in predicting persistent OSAS ed was 0.96(95%CI:0.90-0.98),with the sensitivity of 83.3%,the specificity of 94.1%,and the accuracy of 88.6%.Conclusion The classification tree prediction model can screen out the risk factors of persistent OSAS after adenotonsillectomy in children with moderate or severe OSAS effectively,and provide the basis for clinical screening of patients who need follow-up treatment.

关 键 词：阻塞性睡眠呼吸暂停综合征 腺样体扁桃体切除术 术后持续残留 分类树模型 儿童 

分 类 号：R56[医药卫生—呼吸系统]