Vetter Julius, Lim Kathleen, Dijkstra Tjeerd M H, Dargaville Peter A, Kohlbacher Oliver, Macke Jakob H, Poets Christian F
Machine Learning in Science, University of Tübingen and Tübingen AI Center, Tübingen, Germany.
Department of Computer Science, University of Tübingen, Tübingen, Germany.
PLOS Digit Health. 2024 Dec 13;3(12):e0000678. doi: 10.1371/journal.pdig.0000678. eCollection 2024 Dec.
Neonatal apneas and hypopneas present a serious risk for healthy infant development. Treating these adverse events requires frequent manual stimulation by skilled personnel, which can lead to alarm fatigue. This study aims to develop and validate an interpretable model that can predict apneas and hypopneas. Automatically predicting these adverse events before they occur would enable the use of methods for automatic intervention. We propose a neural additive model to predict individual occurrences of neonatal apnea and hypopnea and apply it to a physiological dataset from infants with Robin sequence at risk of upper airway obstruction. The dataset will be made publicly available together with this study. Our proposed model allows the prediction of individual apneas and hypopneas, achieving an average AuROC of 0.80 when discriminating segments of polysomnography recordings starting 15 seconds before the onset of apneas and hypopneas from control segments. Its additive nature makes the model inherently interpretable, which allowed insights into how important a given signal modality is for prediction and which patterns in the signal are discriminative. For our problem of predicting apneas and hypopneas in infants with Robin sequence, prior irregularities in breathing-related modalities as well as decreases in SpO2 levels were especially discriminative. Our prediction model presents a step towards an automatic prediction of neonatal apneas and hypopneas in infants at risk for upper airway obstruction. Together with the publicly released dataset, it has the potential to facilitate the development and application of methods for automatic intervention in clinical practice.
新生儿呼吸暂停和呼吸不足对健康婴儿的发育构成严重风险。治疗这些不良事件需要技术熟练的人员频繁进行人工刺激，这可能导致警报疲劳。本研究旨在开发并验证一种可解释的模型，该模型能够预测呼吸暂停和呼吸不足。在这些不良事件发生之前自动进行预测，将能够采用自动干预方法。我们提出一种神经加法模型来预测新生儿呼吸暂停和呼吸不足的个体发生情况，并将其应用于患有罗宾序列且有上呼吸道阻塞风险的婴儿的生理数据集。该数据集将与本研究一起公开提供。我们提出的模型能够预测个体的呼吸暂停和呼吸不足，在区分呼吸暂停和呼吸不足发作前15秒开始的多导睡眠图记录片段与对照片段时，平均曲线下面积（AuROC）达到0.80。其加法性质使该模型具有内在的可解释性，从而能够深入了解给定信号模态对预测的重要性以及信号中的哪些模式具有判别力。对于我们预测患有罗宾序列的婴儿的呼吸暂停和呼吸不足这一问题，呼吸相关模态先前的不规则情况以及血氧饱和度（SpO2）水平的下降尤其具有判别力。我们的预测模型朝着自动预测有上呼吸道阻塞风险的婴儿的新生儿呼吸暂停和呼吸不足迈出了一步。连同公开发布的数据集，它有可能促进临床实践中自动干预方法的开发和应用。
