Monitoring mandibular movements to detect Cheyne-Stokes Breathing

Cheyne Stokes Breathing (CSB) is a poor prognosis sleep condition that should be screened for in patients with heart failure or other severe neurological or kidney diseases. A simple device that detects mandibular movements can accurately identify patients with CSB.

Central apnea/hypopnea assuming the pattern of Cheyne Stokes breathing (CSB) is an independent risk condition for mortality in patients with chronic heart failure with a prevalence similar to obstructive sleep apnea or hypopnea (OSA) reaching 30% [1‐3]. CSB has also been described in patients who have experienced cerebrovascular accidents and among those with end-stage renal disease [4]. In addition, the presence of idiopathic central apneas carries an increased risk of atrial fibrillation [5]. Thus, screening and early detection of CSB could be valuable in large populations of chronic disease at-risk patients.

CSB is a particular form of periodic waxing and waning respiration. The diagnosis is made in presence of periods of central apnea or hypopnea alternating with a crescendo/decrescendo pattern of ventilation. The International Classification of Sleep Diseases diagnostic criteria for CSB requires 10 central apneas per hour of sleep. The cycle length (the time from one zenith in airflow during the respiratory phase to the next zenith in airflow) varies with the underlying disease. In systolic heart failure, cycle lengths are longer (between 45 and 90 s), when compared to the cycle lengths reported for other disorders associated with CSB (~35 s) [6]. During the crescendo/decrescendo phases, respiratory efforts are driving changes in ventilation reflecting variations in respiratory drive. Recurrent arousals occur usually at ventilation peaks, promote ventilatory instability, and perpetuate CSB breathing patterns [4, 7]. The diagnosis requires expensive and poorly accessible type 1 or 2 polysomnography but could be addressed with a portable monitoring in a high pre-test probability population of CSB for early detection and intervention.

We have previously showed that respiratory effort are well characterized by an increase in the amplitude of mandibular movements (MM) > 0.3 mm during episodes of OSA [8, 9]. In addition, we have also documented that MM reliably identify cortical arousals ending respiratory events and closing the mouth [10, 11]. The MM patterns observed are 1) large-amplitude MM (MML), and 2) “sharp” large-amplitude MM (MMS), and these two specific and easily recognizable features readily identify respiratory efforts and cortical arousals, respectively [11]. Because MM are simple to acquire using magnetometry, and since the quality of MM signals remains more stable than nasal flow, throughout the night, measurements of MM could be used as a single channel monitor or screener, particularly for use in out of sleep laboratory conditions use [11].

Based on early empirical observations of magnetometry recordings during CSB events, we hypothesized that periodic MML (LPM) and periodic MMS (SPM) would identify classical CSB which is characterized by periods of respiratory effort with increased ventilation alternating with periods of no (or substantially reduced) effort during apneas (or hypopneas). We therefore evaluated the concordance of LPM and SPM with CSB identified during in-lab PSG.

A group of 22 patients was identified with CSB in 573 consecutive recorded subjects; in this group, a total of 1,970 3-min sleep segments were evaluated (CSB was confirmed in 1,060 fragments). Patient characteristics are summarized in Table 2. Twelve patients had congestive heart failure (mean left ventricular ejection fraction < 30%), five patients has sustained a cerebrovascular accident (CVA), and in five patients, the etiology of CSB was unknown. No patient was on opioids. Compared to the original dataset, the training set of 985 5-min fragments had the same distribution pattern of CSB, LPM and SPM.

In the present study, scoring of the MM signals reliably indicated the presence or absence of CSB in patients being evaluated for suspected SDB. Our findings show that the presence of LPM and/or SPM can accurately detect CSB because the periodicity of these typical movements of the mandible. In contrast, MM ruled out CSB in absence of LPM and SPM. These findings support conducting a prospective validity assessment studies of MM recordings in a home-based type 4 device to detect CSB in chronic disease at-risk populations.

Use of MM signals has emerged as sensitive and specific for identifying underlying respiratory events during OSA [9]. Furthermore, sharp large-amplitude movements of the mandible (MMS) detect cortical arousals that occur after OSA events [12]. In the presence of respiratory efforts, even without pharyngeal obstruction, the forces produced by the displacement of the thorax are transmitted through the mediastinum to the upper airway. During each breath, the negative swings of intrathoracic pressure exert caudal traction stretching and improving upper airway patency [18]. Accordingly, similar to the MM patterns observed in OSA, changes in mandibular position directly reflect changes in respiratory effort during the hyperventilation phase of CSB. During the apneic period, no MM is observed.

In the present study, classification approaches using two binary predictors (LPM and SPM) reliably detected a single binary response variable (presence of CSB detected by PSG). Although different statistical methods may differ in their algorithms, their level of complexity and their overall performance, decision rules in classifying and analyzing the relationship between the binary predictors (LPM, SPM) and the predicted in-laboratory PSG CSB were clearly apparent and involved interpretable models (a traditional logistic model and a decision tree (CART model). Compared to a traditional logistic model and odds-ratio, the CART algorithm has more advantages as the decision tree can be visualized and easily interpreted by straightforward prediction rules, and accordingly is simple and robust for clinical decision making [17]. As the validation results were identical for both logistic and CART models, any of them could be used to predict the probability of CSB using MM signals.

Notwithstanding the unique predictability of CSB from MM recordings, we should mention some potential limitations of MM signals. MM signals may fade or become less easily detectable in presence of the muscle atonia during REM sleep. However, CSB is more common during NREM sleep and usually abolished during REM sleep [19, 20].

MM signal artefacts from magnetic interferences have only been observed in CT scan rooms or in presence of metal parts placed very close to the probes, or moving in the space around or between the probes. Steady state magnetic signals are filtered out [21].

Some drugs such as sedatives could increase the threshold for arousals and accordingly reduce the frequency of arousal events, and therefore of SPM, affecting the ability of the signal to detect CSB [22]. This risk is limited in a model where two independent variables (LPM and/or SPM) are employed. Others movements of the mandible (due to bruxism, chewing or swallowing) have been described during sleep but the latter are void of the typical CSB periodicity. They are not driven at the breathing frequency [19]. The false negative and positive rates could have been affected by scoring error and the simplicity of the classification rule (exclusive use of binary predictor). However, the study is strengthened by the blinded scoring approach, the large sample size used with random assignments into a training and validation sets, a large and comprehensive range of classifiers with potential machine solutions, a repeated K-fold cross validation on random data subsets, and an independent and random model of validation. Based on the present findings, we posit that the models developed herein could contribute the development of an algorithm to evaluate CSB diagnosis based on MM manual analysis, particularly considering that the CART model classification tree is appropriate for human decision making [23]. Of note, even more complex algorithms may have potential applications when used in the development of automated interpretation software.

Isolated analysis of mandibular movements enables accurate identification of CSB in patients during in lab PSG. MM may be a useful simple tool for screening and home monitoring of patients at risk for CSB.