Zhang J, Zhao D, Zhou Z X, Wang Y, Chen B Y
Biomedical Engineering, School of Precision Instrument and Opto-Electronics Engineering, Tianjin University, Tianjin 300072, China.
Department of Respiratory and Critical Care Medicine, Tianjin Medical University Central Hospital, Tianjin 300052, China.
Zhonghua Jie He He Hu Xi Za Zhi. 2021 Feb 12;44(2):101-107. doi: 10.3760/cma.j.cn112147-20200724-00843.
To explore the value of night pulse oximetry monitoring in the prediction and classification of obstructive sleep apnea hypopnea syndrome (OSAHS). From January 2018 to December 2019, 580 snoring patients admitted to the Sleep Center of Tianjin Medical University General Hospital were analyzed retrospectively. There were 418 males and 162 females, aging 13-85(49±14) years. All subjects underwent polysomnography, and the apnea hypopnea index (AHI)was 0-101.4(43.06±27.47) times/hour. There were 52 cases in the non-OSAHS group (AHI<5 times/h), 69 cases in the mild OSAHS group (5 times/h<AHI≤15 times/h), 98 cases in the moderate OSAHS group (15 times/h<AHI≤30 times/h), and 361 cases in the severe OSAHS group (30 times/h<AHI).Correlation analysis was performed between indicators extracted from SpO signal and AHI, and 11 blood oxygen indicators related to AHI were selected (3% oxygen reduction recovery index, the area of SpO under the 90% curve, average lowest SpO, lowest SpO, the average SpO, the percentage of time SpO under 95%, 90%, 85%, 80%, 75%, 70%). Finally, gender, age and body mass index (BMI) were added. We ysed multiple linear regression (MLR) method to achieve AHI prediction, and back propagation neural network (BPNN) multi-classification method to achieve OSAHS severity classification. Statistical analysis was performed based on SPSS 25.0. The measurement data were analyzed using Pearson correlation test. The MLR method achieved high prediction performance, with a prediction correlation coefficient =0.901 (<0.05) and a goodness of fit = 0.848 (<0.05).The specificity and negative prediction rate of BPNN method classification results were both around 90%, and the sensitivity and positive prediction rates were also high. Among them, the sensitivity of the non-OSAHS group (AHI<5 times/h) was 88.46%±4.50%, and the sensitivity of the severe OSAHS group (AHI>30 times/h) was 94.74%±0.76%. Based on the signals recorded by the SpO monitor, the methods of using MLR model for AHI prediction and using BPNN model for multi-classification may have higher value for the prediction and classification of OSAHS.
探讨夜间脉搏血氧饱和度监测在阻塞性睡眠呼吸暂停低通气综合征（OSAHS）预测及分级中的价值。回顾性分析2018年1月至2019年12月天津医科大学总医院睡眠中心收治的580例打鼾患者。其中男性418例，女性162例，年龄13 - 85（49±14）岁。所有受试者均接受多导睡眠图检查，呼吸暂停低通气指数（AHI）为0 - 101.4（43.06±27.47）次/小时。非OSAHS组（AHI＜5次/小时）52例，轻度OSAHS组（5次/小时＜AHI≤15次/小时）69例，中度OSAHS组（15次/小时＜AHI≤30次/小时）98例，重度OSAHS组（AHI＞30次/小时）361例。对从SpO信号中提取的指标与AHI进行相关性分析，选取11个与AHI相关的血氧指标（3%氧降恢复指数、SpO在90%曲线下面积、平均最低SpO、最低SpO、平均SpO、SpO低于95%、90%、85%、80%、75%、70%的时间百分比）。最后加入性别、年龄和体重指数（BMI）。采用多元线性回归（MLR）方法实现AHI预测，采用反向传播神经网络（BPNN）多分类方法实现OSAHS严重程度分级。基于SPSS 25.0进行统计分析。计量资料采用Pearson相关检验进行分析。MLR方法具有较高的预测性能，预测相关系数 = 0.901（＜0.05），拟合优度 = 0.848（＜0.05）。BPNN方法分类结果的特异度和阴性预测率均在90%左右，敏感度和阳性预测率也较高。其中，非OSAHS组（AHI＜5次/小时）敏感度为88.46%±4.50%，重度OSAHS组（AHI＞30次/小时）敏感度为94.74%±0.76%。基于SpO监测仪记录的信号，使用MLR模型进行AHI预测和使用BPNN模型进行多分类的方法可能对OSAHS的预测和分级具有较高价值。
