Obstructive sleep apnoea (OSA) is caused by repetitive closure of the upper airway during sleep. While the retropalatal airway is reported to be the most collapsible site [1], any state-dependent segments within the upper airway are candidates for closure. Correct identification of the closure site in each OSA patient could lead to the development of individualised OSA treatment strategies [2]. In this issue of the European Respiratory Journal, Azarbarzin et al. [3] propose a model for prediction of epiglottic collapse for each breath by assessing the nasal airflow waveform in sleeping OSA patients. They employed a machine-learning approach to identify characteristic waveform features for constructing the predictive algorithm and validation of the final predictive model. They found that a nasal airflow signal with greater discontinuity index (rapid and marked reduction of inspiratory airflow immediately after achieving the maximum airflow) and greater jaggedness index (repeated deviations of the airflow from the mean value during inspiration) predicts epiglottic collapse. Furthermore, the non-calibrated nasal pressure signal is demonstrated to be equally usable for determining the epiglottic collapse, increasing applicability to clinical practice. However, it is noteworthy that this study does not completely clarify why the discontinuity and jaggedness features are produced by the epiglottic collapse. Whereas application of machine-learning and artificial intelligence technologies is a promising means for solving multifactorial difficult medical problems, as is proved in this study, these technologies should be utilised through careful selection of physiologically and clinically meaningful variables in order to avoid inclusion of meaningless variables in the prediction model. So, what is the physiology behind the success of the prediction? The epiglottis and soft palate function as one-way valves limiting flow during inspiration and expiration http://ow.ly/85JB30epWOC

中文翻译：

---

咽部有两个瓣膜

阻塞性睡眠呼吸暂停 (OSA) 是由睡眠期间上呼吸道反复关闭引起的。虽然据报道腭后气道是最易塌陷的部位 [1]，但上气道内的任何状态相关的段都是闭合的候选部位。正确识别每位 OSA 患者的闭合部位可能会导致制定个性化的 OSA 治疗策略 [2]。在本期欧洲呼吸杂志中，Azarbarzin 等人。[3] 通过评估睡眠 OSA 患者的鼻气流波形，提出了一种预测每次呼吸会厌塌陷的模型。他们采用机器学习方法来识别特征波形特征，以构建预测算法并验证最终预测模型。他们发现，具有较大不连续性指数（在达到最大气流后立即迅速且显着减少吸气气流）和较大锯齿状指数（吸气期间气流与平均值的反复偏差）的鼻气流信号可预测会厌塌陷。此外，未校准的鼻压信号被证明同样可用于确定会厌塌陷，增加了临床实践的适用性。然而，值得注意的是，这项研究并没有完全阐明为什么会厌塌陷会产生不连续性和锯齿状特征。而机器学习和人工智能技术的应用是解决多因素医学难题的有前途的手段，正如本研究所证明的那样，应通过仔细选择生理和临床上有意义的变量来利用这些技术，以避免在预测模型中包含无意义的变量。那么，预测成功背后的生理学是什么？会厌和软腭作为单向阀在吸气和呼气期间限制流量 http://ow.ly/85JB30epWOC