Detecting central sleep apnea in adult patients using WatchPAT—a multicenter validation study

In order to have a substantial representation of patients with CSA within the study population, we selectively recruited heart-failure patients in this study. It was previously reported in the literature that there is a high prevalence (between 33 and 40%) of CSA in heart-failure patients [26, 27]. The total number of cardiac patients (CHF and/or AFib) included in the analysis was 50 (CHF only, n = 33, A. fib. only n = 9, and 8 patients with both conditions), comprising 59.5% of the study population. The local staff at each participating center attempted recruiting participants in whom they estimated that there was a relatively high risk of having central apneas. The contribution of the various centers was not equally distributed, and the various sites had recruited the following number of participants: 1, 2, 3, 5, 5, 8, 8, 8, 11, 12, 21 (total of 84).

The following were the inclusion and exclusion criteria in all sites:

Eighty-four (84) patients suspected of having SDB with selective bias toward recruiting patients with congestive heart failure (CHF) were recruited (54 males), age 57 ± 16 (range 22–83) years, BMI 30 ± 5.9 kg/m² (range 17–45). They consented to and successfully completed the study (with at least 4 h of actual sleep time with valid signals in PSG, and at least 1.5 h of valid WP signals during actual recorded sleep).

Patient anthropometric data and medical history information are summarized in Table 1.

The reference used for comparison in this study was an FDA-approved in-lab PSG from multiple manufacturers used in the eleven study sites. Multichannel PSG configuration compliant with accepted standards and manual scoring according to the American Academy of Sleep Medicine (AASM) directives [28] is considered to be the “gold standard” for sleep testing, and was used to derive standard indicia including the apnea/hypopnea index (AHI) and central sleep apnea-hypopnea index (AHIc) [29, 30]. Apnea with the absence of respiratory effort was scored as central, while obstructive apnea was scored when respiratory effort was noted. In the PSG, hypopneas were scored when there was a reduction in tidal ventilation presented by a 30% or greater reduction in NAF associated with either oxygen desaturation of at least 3% or an EEG change that indicate arousal. It was classified as central or obstructive based on the presence or absence of flow limitation in the NAF signal. Obstructive hypopneas were scored when the reduction in airflow was mainly due to increased upper airway resistance (i.e., with flow limitation in the NAF signal: flattening of the inspiratory portion of the nasal pressure), while central hypopneas were scored when the reduction in flow was mainly due to a reduction in ventilatory effort (and absence of flattening of the inspiratory portion of the nasal pressure).

Enrolled patients underwent a standard sleep study using the PSG system simultaneously with the WP in the clinical sleep lab. The WP provided a series of synchronizing pulses to an auxiliary input channel of the PSG to time-synchronize-scored epochs between the two systems.

The WP system used in this study is a home sleep testing (HST) system based on a wrist-worn device and a finger probe which acquires Peripheral Arterial Tone (PAT) signals and arterial oxygen saturation levels, together with actigraphy data from a 3D accelerometer that is embedded in the wrist unit, and an optional snoring and body position (SBP) sensor that is positioned under the sternal notch. The WP algorithm detects offline apnea/hypopnea events, respiratory effort-related arousals, and sleep/wake status, and determines sleep stages [8, 21‐24].

The WP finger probe is designed to optimize the dynamic range of peripheral arterial volume changes by applying a near diastolic level of counter pressure sufficient to unload arterial wall tension, while preventing the occurrence of venous blood pooling in the finger so as to avoid the possible induction of venoarteriolar reflex–mediated vascular responses, and other measurement-related artifacts, as previously described [8, 31].

The probe measurement site at the distal phalanx of the finger is characterized by a large dynamic blood flow response range by virtue of a dense vascular network of predominantly arteriovenous shunt vessels, which are controlled by α-adrenergic receptors of vascular smooth muscle of the finger vascular bed and thus reflect sympathetic nervous system activity of the patient [10, 32, 33].

The WP assesses changes in vascular tone at the fingertip, with transient vasoconstrictory events being associated with sympathetic activation that typically terminates sleep-disordered breathing events [31]. The autonomic basis for the transient vasoconstrictory events provide a sensitive and accurate means for detecting arousals, which may in fact be considerably more sensitive than electroencephalographically detected cortical arousals. Such autonomic, brainstem, or subcortical arousals are particularly more sensitive markers of arousal even in children than EEG, as reported by Tauman et al. [34]. In adults too, it has been reported that a substantial percentage of respiratory events are not associated with cortical arousals but are related to subcortical (autonomic) arousals. This is consistent with findings in adults of absent cortical arousals in 10–25% of actual apneic events, as reported by Eckert et al. [35]. In the current study, the WP system included also the SBP placed below the sternal-notch [36].

The SBP sensor encapsulates together a microphone and a 3D accelerometer used to derive the spatial body orientation relatively to the vertical gravity from which district position can be calculated (supine, left side, right side, prone, or sitting). The microphone and a proprietary transfer function provide snoring level in decibels calibrated to an estimated recording from a 1-m distance from the patient’s head.

The current WP used in this study is a further development that allows the detection of central sleep apnea events. This development essentially comprised two main independent sources of physiological information that are used to discriminate between central and obstructive apneas. The first involves the use of the 3D accelerometer of the SBP sensor to derive upper chest wall respiratory movements that are diminished during a central sleep apnea event. The second is based on information derived solely from the PAT signal waveform by analyzing dynamic changes of measured systolic upstrokes as described below. These two sources of data are further supported by data from the other WP channels—oximetry, wrist movement, and snoring.

The SBP sensor that contains a triaxial accelerometer was applied to the upper patient’s chest, below the supra sternal notch, and connected to the wrist-mounted WP device. Signal acquisition, signal conditioning, and signal processing yielded accelerometric data indicative of the level of respiratory effort-related chest wall motion. This information was used to determine an adaptive threshold value indicative of the presence or absence of respiratory effort during apneic events determined by the WP system, and thus allows for the identification of central events. The detection was done by a multidimensional classifier, which involved a cumulative indication from several signals, and a few characteristics that disqualified an event from being central.

A representative 10-min example of the output signal of this sensor is shown in Fig. 1. This clearly shows the recurrent absence of respiratory effort during central apneic events, followed by continuous respiratory movement during an apnea-free period.

The use of an accelerometer for determining the presence, stability, and magnitude of chest wall motion indicative of respiratory effort can provide a robust method for differentiation of central and obstructive apneas, as has been previously reported [37‐39].

The WP central apnea detection method which is based on plethysmographic information derived from the finger sensor provides a method for determining the effect of respiratory effort on the PAT signal’s pulse wave.

This is based on an interaction between breathing-related changes of intrathoracic pressure and the dynamic changes occurring during ventricular systole on the detected pulse-wave signals, which occur for amongst other reasons, because the thoracic pressure changes directly impinge on the heart.

The method entails (a) the detection of successive systolic upstrokes; (b) the determination of an index of upstroke dynamics, for example, the fraction of the upstroke occurring within a prescribed time interval commencing at a prescribed fraction of the upstroke or the slope between preset fractions of the upstroke; (c) the normalization to changes within a certain time window; and (d) the determination of the variability of a series of upstroke dynamic index values in time corresponding to the patients breathing cycle.

Essentially, this enables a clear distinction to be made between central and non-central apneas as the two respective conditions represent polar opposites in the degree of breathing-related thoracic pressure–induced changes on the pattern of the pulse-wave upstroke. In the case of central apnea, the absence of respiratory effort means that the stereotypical pattern of the systolic upstroke is unaffected by superimposed pressure fluctuations due to respiratory effort.

In contrast, the large intrathoracic pressure swings associated with resistive breathing during upper airway obstruction or partial obstruction result in changes in the pattern of the systolic upstroke, which vary throughout the course of each breathing cycle, and result in a characteristic, cyclic pattern of changes in the upstroke dynamics over time.

The contrasting patterns of PAT upstrokes during central and non-central apneas can be seen in a montage of superimposed pulses in Fig. 2.

Figure 3 shows the time course during central apnea events in a patient. The signals shown from top to bottom are the respiratory chest movement and upstroke index, which are derived from the WatchPAT device, as well as flow and thoracic and abdominal belts taken from the PSG (marked with darker background).

As can be seen, the respiratory movement is similar to the belt and shows reduction in the movement during apneic events only. Both respiratory chest movement signal and PAT upstroke index pattern variability are substantially reduced during the central event (marked by blue box). The red boxes mark normal breathing before and after the event.

Both the PAT signal upstroke variations associated with intrathoracic pressure changes, and the respiratory movements derived from the enhanced SBP sensor are used to identify central apnea/hypopnea events in a newly upgraded algorithms of the WP software. In addition to these two inputs, additional WP information is used in the algorithm to support the detection: (a) the absence of snoring and (b) decrease in breathing movement as reflected in hand movement that is measured in the wrist actigraphy, if such movement was present before and after the event. Prior to the present study, the algorithm parameters utilizing these parameters were tuned in an optimizing process in a training set of 69 cardiac and OSA patients and were shown to give robust correlation with polysomnography CSA detection results. This algorithm was used in the present study.

The WP determination of SDB events is performed in hierarchical order:

Based on simultaneous comparisons to the PSG gold standard in 84 patients in multiple centers, our findings support our hypothesis that the WatchPAT, when enhanced with the two new features, PAT signal upstroke analysis and RESBP, can accurately detect overall AHI and AHIc, all with correlations of 0.80 or higher compared with in-lab PSG.

In addition, the WP further benefits from the well-established advantages of home sleep testing, including the realistic option of tracking patient condition over time as well as in response to therapeutic interventions. In particular, the ability to accurately diagnose CSA using a very simple and patient-friendly system may offer considerable advantages for a patient population when considering treatment.

Although accelerometric-based detection of chest wall movement is a well-known method for sensing respiratory effort [37‐39], it has not been widely integrated into clinical use, and respiratory belts are the common standard. Although placement of the sensor makes patient setup somewhat complicated, it still provides a more robust and comfortable attachment than respiratory belts which tend to slide and become displaced and may therefore be advantageous for home sleep testing. The use of SBP in which the sensing element (accelerometer) is already in place obviates the need for a dedicated effort device, whether a belt or another accelerometric sensor and as such does not add any additional complexity. A potential disadvantage of the SBP in comparison with the other methods is its location on the very upper chest where respiratory movement is subtle and therefore the signal-to-noise ratio is lower.

In a somewhat similar approach to using the peripheral pulse wave to differentiate between central and non-central events during sleep to that used in this study, Sommermeyer et al. [40] also used a peripheral pulse signal to determine the degree of respiratory effort during apneic events in order to determine the nature of the events, and to differentiate central from non-central events.

These authors determined the signal power in the typical frequency band of breathing (0.15–0.4 Hz) during the apnea phase and compared it with the signal power of the same frequency range 15 s before the apnea event to thereby define an effort ratio index.

In the present study, we restricted the pulse analysis to the duration of the systolic upstroke and normalized the amplitude of the pulse waves in order to avoid the potentially confounding influence of spontaneous modulation of the pulse amplitude, which may occur at variable frequencies and with large and variable amplitude changes, as observed by Burton [41].

In addition, variations in the finger pulse amplitude frequency domain occur during sleep and have in fact been used by the WP to discriminate between REM and non-REM sleep [22] and between light and deep non-REM sleep stages [23, 24].

We used 2 AHIc thresholds to demonstrate the accuracy of WP’s diagnosis, 10 events/h and 15 events/h. Gabor et al. [42] has chosen 10 events/h cut-off point stating that “Although no formal determination regarding the definitive threshold has been set, the AHI ≥ 10/h is commonly used to establish a sleep apnea diagnosis, and AHIc > 10 to differentiate the diagnosis as central,” and same approach was suggested by Cowie et al. [43, 44] in the SERVE-HF study who also used this threshold to enroll predominantly CSA participants defined as having AHI ≥ 15 events per h, with > 50% central events and a central AHI of ≥ 10 events per h. However, an AHIc of 15-events/h threshold is more commonly used in the clinical setting and serves to approve treatment, and may represent a clinical threshold of amenability to improvement following treatment [45‐47]. Furthermore, Arzt et al. [46] reported that “early suppression of CSA by CPAP to an AHI below 15 per h may improve both LVEF and transplant-free survival.”, and likewise, Momomura [47] found that in CSA patients in whom AHI was improved to below 15, CPAP treatment resulted in a greater increase in LV ejection fraction at 3 months and significantly better transplant-free survival than control subjects, whereas in those whose AHI remained above 15, neither LV ejection fraction or survival was improved.

The two approaches for determining central apneas used in this study have limitations. Finger vascular responses may be adversely affected by the presence of cardiac arrhythmia, which can affect the effectiveness of myocardial contraction and thus introduce upstroke index variability unrelated to the changes in intrathoracic pressure associated with respiratory effort. Such changes would however be easily recognized on the basis of the inter-beat timing, and in the case of reproducible arrhythmias, the effect on upstroke dynamics could be possibly be learned, and adjusted for.

In the case of the accelerometric evaluation of chest wall motion indicative of respiratory effort, similarly to other surface-attached technologies in use, it is possible that chest wall motion unrelated to breathing or respiratory effort may confound the analysis. Similarly, variations in patient posture may alter the basic conditions for signal transduction, and change the signal gain and/or frequency characteristics. The accuracy of sensor placement and coupling to the body surface may also be susceptible to the limitations imposed by patient self-application.

The combined use of both independent methods in this study may have reduced the potential inaccuracies of each respective method.

The selection of a patient population with high percentage of heart failure in this study may have resulted in an unrepresentative sample of the central sleep apnea patient population at large. However, irrespective of the etiology of CSA, the functional outcome of absent respiratory effort remains a common feature, and the diagnostic principles employed in this study were explicitly directed to identifying absent respiratory effort.