Beyond the apnea–hypopnea index

We have read with great interest the paper “Metrics of Sleep Apnea Severity: Beyond the AHI” written by Malhotra et al. [1], which was published in the Sleep on March 9, 2021. The authors presented research-based recommendations to enhance and optimize the diagnosis and severity estimation of obstructive sleep apnea (OSA). The authors also presented alternative metrics, such as the Hypoxic Burden and the apnea–hypopnea event duration, for estimation of OSA severity. They discussed the potential future directions of OSA diagnostics such as machine learning and wearable technologies. While we certainly agree that this is an important topic and the apnea–hypopnea index (AHI) is not the best metric to describe the severity of OSA, at the same time, we wonder how the authors have completely missed a vast amount of relevant research papers that have been published during the past decade. Albeit the authors state that the paper is not intended to be “an exhaustive review of the available literature,” the paper lacks numerous highly relevant references, which are very strongly related to the topic. For example, our research group has published over 50 scientific papers on this topic (20 most relevant ones listed in Table 1). However, none of these scientific papers were cited in the Research Statement.

Open multimedia modal

First, the authors did not mention the Desaturation Severity parameter [2] when talking about the Hypoxic Burden. Instead, the authors claimed that “now investigators have captured the area under the oxyhemoglobin saturation curve as a metric of hypoxic burden” referring to the paper by Azarbarzin et al. [3] published in 2019. However, the idea of calculating the area of blood oxygen desaturation events has been introduced many years before that paper [2]. In addition, a letter to editor [4] of the European Hearth Journal was published related to the paper by Azarbarzin et al. [3], in which an incomplete literature review was criticized; they also missed the previously published papers related to the calculation of area under the desaturation events. Furthermore, when Malhotra et al. discussed the apnea–hypopnea event duration they stated that “Event duration has been quantified by Butler et al.” [5]. Albeit being true, the paper by Butler et al. [5] is most certainly not the first one presenting the idea of taking apnea and hypopnea event duration into account in the severity estimation of OSA. At least Kulkas et al. introduced this idea already in 2013 [2].

Second, we agree with the authors that novel wearable technologies have a great potential to enhance the diagnosis of sleep disorders and evaluation of sleep quality in the future. Albeit there is lots of ongoing research related to, for example, simplification of devices intended for diagnosis of OSA, many of these novel approaches are not yet accurate and reliable enough compared to conventional polysomnography or home sleep apnea test devices. Furthermore, certain devices used in sleep research lack clinical validation according to medical device regulations and are considered consumer-grade devices. However, the authors discussed wearable technologies on a very general level providing an example on only one device while citing the paper written by de Zambotti et al. [6] presenting a much broader discussion on the available wearable sleep technologies. The authors could have even mentioned devices like Fitbit, Jawbone, Misfit, and ŌURA ring discussed by de Zambotti et al. [6]. Leaving the most promising technologies out of the discussion in the review paper does not provide the readers the correct overall picture of what is happening in the field of wearable sleep technology.

Third, when the future directions and machine learning were discussed, it was again very surprising to see how many relevant research papers were not cited. Just to pinpoint a couple, Phan et al. [7] and Gutiérrez-Tobal et al. [8] have done lots of research involving machine learning. They have, for example, introduced deep learning solutions for the classification of sleep stages from polysomnographic signals [7] and estimated OSA severity class from a blood oxygen saturation signal [8]. In addition, we have also recently published several research papers on how to utilize machine learning to analyze polysomnography data. In these publications, deep learning methods were developed to estimate the value of AHI from blood oxygen saturation signal [9] and automatically classified sleep stages from a single electroencephalography signal [10]. Mirroring the results and topics of the papers by us (Table 1) and others [7^, 8] to the topic of the Research Statement, the authors’ selection of the references is not appropriate.

To conclude, the topic of the Research Statement is highly relevant and timely tackling the very important issue. However, as the aim of the Research Statement was to discuss novel polysomnography metrics and future directions, and provide “recommendations for research needed to optimize OSA diagnostics,” we feel that the literature review included in the Research Statement is very incomplete. This is worrying especially considering the influential positions of the authors in the field of sleep medicine. Furthermore, we feel that the pioneering work should always be respected in science. We believe that in the spirit of fair play and collaboration, researchers worldwide have great opportunities to make major technological advances in the field of sleep medicine.