Relationship of stroke volume to different patterns of Cheyne-Stokes respiration in heart failure

Abstract

Study Objectives

In patients with heart failure (HF) and reduced left ventricular ejection fraction (HFrEF), stroke volume (SV) falls during hyperpnea of Cheyne-Stokes respiration with central sleep apnea (CSR-CSA). We have identified two distinct patterns of hyperpnea: positive, in which end-expiratory lung volume (EELV) remains at or above functional residual capacity (FRC), and negative, in which EELV falls below FRC. The increase in expiratory intrathoracic pressure generated by the latter should have effects on the heart analogous to external chest compression. To test the hypotheses that in HFrEF patients, CSR-CSA with the negative pattern has an auto-resuscitation effect such that compared with the positive pattern, it is associated with a smaller fall in SV and a smaller increase in cardiac workload (product of heart rate and systolic blood pressure).

Methods

In 15 consecutive HFrEF patients with CSR-CSA during polysomnography, hemodynamic data derived from digital photoplethysmography during positive and negative hyperpneas were compared.

Results

Compared to the positive, negative hyperpneas were accompanied by reductions in the maximum and mean relative fall in SV of 30% (p = 0.002) and 10% (p = 0.031), respectively, and by reductions in the degree of increases in heart rate and rate pressure product during hyperpnea of 46% (p < 0.001) and 13% (p = 0.007), respectively.

Conclusions

Our findings suggest the novel concept that the negative pattern of CSR-CSA may constitute a form of auto-resuscitation that acts as a compensatory mechanism to maintain SV in patients with severe HF.

Statement of Significance.

In patients with heart failure and reduced left ventricular ejection fraction (HFrEF), stroke volume (SV) falls during the hyperpnea of Cheyne-Stokes respiration with central sleep apnea (CSR-CSA). Two patterns of hyperpnea were identified: positive, during which end-expiratory lung volume (EELV) remains at or above functional residual capacity (FRC), and negative, during which EELV falls below FRC, generating an increase in expiratory intrathoracic pressure (P_IT). We provide evidence for the novel concept that compared to the positive hyperpnea pattern, the negative pattern during CSR-CSA has a cardiac auto-resuscitative effect via generation of positive P_IT that reduces the degree of fall in SV. This effect may act as a compensatory mechanism to maintain SV during hyperpnea of CSR-CSA in patients with HFrEF.

Introduction

Cheyne-Stokes respiration with central sleep apnea (CSR-CSA) is a form of periodic breathing in which central apneas and hypopneas alternate with hyperpneas having a prolonged waxing-waning pattern of tidal volume [1]. CSR-CSA is commonly observed in patients with heart failure (HF) and reduced left ventricular ejection fraction (HFrEF). In such patients, the durations of the hyperpnea and apnea-hyperpnea cycle are directly proportional to circulation time and inversely related to stroke volume (SV) [2, 3]. CSR-CSA is accompanied by cyclic surges in sympathetic nervous activity, blood pressure (BP) and heart rate (HR) [4]. Although, long-term exposure of the heart to excessive sympathetic nervous activity is associated with worse prognosis, whether CSR-CSA is an independent risk factor for mortality or is just a manifestation of HF severity is a question that has yet to be resolved [5–7]. Indeed, it has been proposed that CSR-CSA might act as a compensatory mechanism in patients with HFrEF [8, 9]. Consistent with this possibility is the observation that in contrast to obstructive sleep apnea (OSA), during CSR-CSA, SV rises during central apneas but falls during hyperpneas [10].

During the hyperpneic phase of CSR-CSA, we have described two different patterns of hyperpnea: one in which end-expiratory lung volume (EELV) remains at, or increases above functional residual capacity (FRC), and a second in which EELV falls below FRC [11]. A fall in EELV below FRC implies activation of the expiratory muscles and generation of positive intrathoracic pressure (P_IT). There are two circumstances in which an increase in P_IT augments SV: during external chest compression in cardiopulmonary resuscitation [12, 13] and during voluntary coughing following a life-threatening cardiac arrhythmia [14, 15]. Thus CSR-CSA during which EELV falls below FRC could have an auto-resuscitating effect whereby the positive P_IT applied to the left ventricle (LV) reduces afterload by reducing the difference between intrathoracic and intra-cardiac pressure, and by augmenting SV [16–18].

Based on these observations, we hypothesized that in HFrEF patients, CSR-CSA with a fall in EELV below FRC during hyperpnea (negative pattern) would be associated with a smaller fall in SV than occurs in those with CSR-CSA during which EELV remains at or rises above FRC (positive pattern). We further hypothesized that CSR-CSA with the negative hyperpnea pattern would reduce cardiac workload expressed as the product of heart rate and systolic BP (i.e. rate × pressure product) [19].

Methods

See the Supplementary Material for detailed methods.

Participants

Inclusion criteria were patients with HFrEF (LV ejection fraction ≤ 45%) and CSR-CSA at least 18 years of age, and with a New York Heart Association class of I–III. Exclusion criteria were: (1) treated sleep apnea; (2) unstable angina, myocardial infarction, or cardiac surgery within the previous 3 months; (3) cardiac pacing; and (4) a history of asthma or chronic obstructive airways disease. The protocol was approved by the research ethics board of the Toronto Rehabilitation Institute, and all participants provided written consent before participation.

Polysomnography

All participants underwent polysomnography using standard techniques and scoring criteria for sleep stages and arousals from sleep [20, 21]. Thoracoabdominal motion was monitored using respiratory inductance plethysmography, the electronic sum of which was used to assess relative changes in tidal volume. Nasal pressure and arterial oxyhemoglobin saturation (SaO₂) were monitored. Central apneas and hypopneas were defined as previously described [10]. A CSR-CSA disorder was defined as an AHI of at least 15 and with a central AHI > 10 that alternated with hyperpnea having a waxing-waning pattern of tidal volume.

In those who had SaO₂ assessed by an oximeter placed on the ear, lung to ear circulation time was measured from the end of the apnea to subsequent nadir of SaO₂. The degree of oxygen desaturation (∆SaO₂) was calculated as the difference between the peak SaO₂ and the subsequent nadir for that event.

Hyperpnea analysis

To quantify the change of EELV above or below FRC during hyperpnea, a mathematical computer algorithm tool was developed as previously described [11] that calculated a variable that described the relative change in EELV from FRC; EELV index (EELVI).

To quantify EELVI, the baseline FRC was calculated as the average of the two reference points as shown in Figure 1. The amplitude of voltage changes in end-inspiratory lung volume (Ains) and EELV (Aexp) from baseline were measured for each breath. Subsequently, the EELVI as a single value for a hyperpneic episode was calculated as:

Figure 1.

Panel A illustrates how baseline end-expiratory lung volume (EELV) was determined for each hyperpnea. The closed circles represent the first reference point at the end of the preceding central apnea and the open circles represent the second reference point at the onset of the subsequent central apnea. The baseline EELV is determined as the average of these two reference points. The solid arrows represent the amplitude of voltage changes from baseline to end-inspiratory lung volume (Ains), while the dashed arrows represent the amplitude of voltage changes from baseline to EELV (Aexp). Panels B and C show actual polysomnographic recordings from one subject showing hyperpneas with positive and negative patterns, respectively. ECG = electrocardiography, EEG = electroencephalography, EMG = electromyography, EOG = electrooculography, SaO₂ = oxyhemoglobin saturation.

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{E}\mathrm{E}\mathrm{L}\mathrm{V}\mathrm{I}=\frac{\mathrm{A}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{g}\mathrm{e}\mathrm{o}\mathrm{f}\mathrm{A}exp}{\mathrm{A}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{a}\mathrm{g}\mathrm{e}\mathrm{o}\mathrm{f}\mathrm{A}ins}\times 100(\mathrm{\%})}\end{array}}$$

We classified each hyperpnea into the two patterns: the positive pattern in which hyperpneas had an EELVI of 0% or more, and the negative pattern in which hyperpneas had an EELVI of less than 0%. We analyzed only episodes of CSR-CSA that occurred in the supine position during stage 2 non-REM sleep to exclude potential influence of body position and sleep stages on the respiratory pattern and hemodynamics.

Hemodynamic measurements and analysis

During polysomnography, we obtained beat-to-beat noninvasive measurements of BP, HR, and SV using digital photoplethysmography (DPP) [10, 22, 23] via finger cuffs during stage 2 sleep. A splint was applied to the patient’s arm to prevent movement and maintain it in an extended position. SV was computed using the well-validated Modelflow method through pulse contour analysis software (BeatScope^@1.1a; Finapres Medical Systems BV, Amsterdam, The Netherlands) [24–26]. As an index of myocardial workload, rate pressure product was calculated by multiplying systolic BP by HR [19]. We calculated the difference in systolic BP, HR, SV, and rate pressure product expressed as percentage change (%∆SBP, %∆HR, %∆SV, and %∆rate pressure product, respectively) between values during hyperpnea and the average values during the 10 seconds of central apnea immediately before the beginning of hyperpnea (baseline). We evaluated both the average change in each variable from baseline to the average during the entire hyperpnea and the change in SV and SBP between baseline and the heartbeat during hyperpnea at which the lowest SV was observed.

Statistical analysis

Comparisons between the positive and negative hyperpneas were performed using unpaired t-test for normally distributed data and Mann-Whitney U test for non-normally distributed data. To describe the beat-to-beat change in each hemodynamic index during the entire hyperpnea, we used percentage of total hyperpnea time for each cardiac beat, which was calculated as the difference in time from the onset of hyperpnea to each cardiac beat divided by hyperpnea time multiplied by 100. Data are presented as mean ± SD, median (interquartile range) or frequencies. A two-tailed p value of less than 0.05 was considered statistically significant. All analyses were performed using IBM SPSS, Version 20.0 (IBM Corp., Armonk, NY).

Results

Characteristics of the participants

Fifteen participants met the eligibility criteria. Their characteristics are presented in Table 1. Participants were predominantly male, were generally non-obese, and had severely depressed LV ejection fraction and severe CSR-CSA.

Table 1.

Characteristics of the participants

Variables	n = 15
Age, year	63.3 ± 12.0
Male, n (%)	13 (87)
Body mass index, kg/m2	26.9 ± 4.0
New York Heart Association class	2.0 (2.0–2.5)
Ischemic etiology, n (%)	6 (40)
Left ventricular ejection fraction, %	29.6 ± 9.9
Atrial fibrillation, n (%)	3 (20)
Total sleep time, hours	4.0 ± 1.7
Apnea-hypopnea index, events/hour of sleep	46.6 ± 14.8
Obstructive events, no	36 (8–89)
Central events, no	87 (38–170)
Minimum SaO2, %	81.1 ± 6.2

SaO₂ = arterial oxyhemoglobin saturation. Values are expressed as mean ± SD or median (interquartile range), unless otherwise indicated.

Hemodynamic and respiratory indices during the entire hyperpnea

Two hundred fifty-four cycles of CSR-CSA were identified. Of these, 27 were excluded because of frequent ventricular ectopy. Of the remaining 227 CSR-CSA cycles, 180 were classified as positive (from 14 participants) and 47 as negative (from seven participants). Six participants displayed both positive and negative patterns.

Figure 2 shows the composite time course for the positive and negative patterns for %∆SV, %∆SBP, %∆HR, and %∆rate pressure product during the entire hyperpnea. %∆SBP in the negative pattern increased more than in the positive pattern (p = 0.012). %∆SV fell significantly less in the negative than in the positive pattern (p = 0.031). Increases in %∆HR and %∆rate pressure product during hyperpnea were significantly less in the negative than in the positive pattern by 46% (p < 0.001) and 13% (p = 0.007), respectively. At the time of the lowest SV, relative percentage fall in SV during the negative pattern was substantially smaller by 30% (−14.1 [−19.4 to −11.8] % vs. −20.1 [−29.3 to −14.0] %, p = 0.002) than during the positive pattern, and the relative percentage increase in SBP during the negative pattern was greater than during the positive pattern by 36% (13.6 [8.1 to 18.0] vs. 10.0 [3.6 to 15.1] %, p = 0.036).

Figure 2.

Left panels show time-courses of relative changes in hemodynamic indices during the entire hyperpnea. Dashed lines represent changes for positive patterns and solid lines for negative patterns. Right panels show box plots of the median and interquartile range while whiskers indicate 1.5 times the interquartile range. Open boxes represent changes for positive patterns and grey for negative patterns. Panels A and B show relative change in stroke volume (%∆SV) which fell significantly less in the negative than the positive pattern: −5.2 (−7.3 to −2.3) vs. −5.8 (−7.9 to −3.6) % (p = 0.031). Panels C and D show relative changes in systolic blood pressure (%∆SBP) which increased significantly more in negative than positive patterns: 10.3 (7.7–13.0) vs. 9.0 (6.3–12.3) % (p = 0.012). Panels E and F show relative changes in heart rate (%∆HR) that rose significantly less in the negative compared to the positive pattern: 3.7 (2.1–5.2) vs. 6.8 (4.3–10.1) % (p < 0.001). Panels G and H show relative changes in rate pressure product (%∆rate pressure product) that rose significantly less during negative compared to positive hyperpneas: 14.6 (9.7–18.9) % vs. 16.7 (11.4–23.1) % (p = 0.007).

As displayed in Table 2, negative hyperpneas had lower EELVI than the positive ones. The negative pattern was also associated with longer cycle and hyperpnea times, a lower respiratory rate and a lesser fall in SaO₂ (p < 0.001) than the positive pattern. In the 12 participants in whom an oximeter was placed on the ear, there was a nonsignificant tendency for longer lung-to-ear circulation time (LECT) in the negative than the positive pattern (p = 0.088).

Table 2.

Comparison of respiratory variables between the positive and negative hyperpnea patterns

Variables	Positive cyclesN = 180	Negative cyclesN = 47	P
End-expiratory lung volume index, %	16.1 (5.6–28.5)	−9.2 (−13.8 to −3.6)	<0.001
Cycle time, seconds	53 ± 9	64 ± 10	<0.001
Lung-to-ear circulation time, seconds	20.8 ± 4.2a	22.0 ± 3.4b	0.088
Hyperpnea time, seconds	32 ± 7	37 ± 5	<0.001
Apnea time, seconds	21 ± 6	25 ± 7	<0.001
Peak SaO2, %	98 (97–99)	97 (97–98)	0.001
∆SaO2, %	−7 (−9 to −6)	−4 (−8 to −3)	<0.001
Respiratory rate during hyperpnea, bpm	20 (17–22)	15 (14–17)	<0.001

bpm = breaths per minute; SaO₂ = oxyhemoglobin saturation; ∆SaO₂ = fall in oxyhemoglobin saturation. Values are expressed as mean ± SD or median (interquartile range), unless otherwise indicated.

^aData from 117 cycles.

^bData from 47 cycles.

Discussion

This study has given rise to several novel findings on the influence of the two different patterns of hyperpnea on SV during sleep in patients with HFrEF. First, we found that the negative pattern of hyperpnea was associated with a 30% reduction in the maximum and 10% reduction in the mean fall in SV compared to the positive pattern of hyperpnea. Second, the negative pattern of hyperpnea was associated with a 13% lower increase in cardiac workload compared to the positive pattern. Third, the degree of O₂ desaturation per respiratory event was less during the negative than the positive pattern. Finally, both cycle and hyperpnea times were longer in the negative than the positive pattern suggesting lower SV during the former. These findings support our hypotheses that the negative pattern of hyperpnea during CSR-CSA is accompanied by a smaller fall in SV and a smaller increase in cardiac workload than the positive pattern. Taken together, they suggest an auto-resuscitating effect during negative hyperpneas that may be a compensatory mechanism to maintain SV and reduce cardiac energy expenditure in the face of worse cardiac function than during positive hyperpneas. Moreover, the lesser degree of hypoxia during negative hyperpneas suggests a further cardio-protective effect by reducing the degree of hypoxia to which the cardiovascular system is exposed during CSR-CSA.

We have demonstrated in previous studies that SV and circulation time can change overnight in patients with HFrEF and co-existing sleep-disordered breathing [23]. We have also shown that such overnight hemodynamic changes can be accompanied by changes in circulation time, CSR-CSA cycle length and the type of sleep apnea [27]. As displayed in Table 2, CSR-CSA cycle and hyperpnea lengths were longer during negative than positive CSR-CSA cycles, indicating worse hemodynamic compromise in negative than positive cycles. Accordingly, the most likely explanation for the occurrence of both positive and negative CSR-CSA cycles in the same subject was overnight alterations in SV and circulation time, such that lower SVs and longer circulation times were accompanied by negative CSR-CSA cycles, and vice versa.

During hyperpnea in CSR-CSA, negative inspiratory P_IT is generated. Such negative P_IT would increase venous return and cause leftward shift of the interventricular septum that could impede LV filling and reduce SV [28, 29]. In addition to this adverse ventricular interaction, negative P_IT would increase LV afterload by increasing the difference between P_IT and intra-LV pressure, further reducing SV [30, 31]. One physiological means of counteracting these potential adverse effects would be to generate positive P_IT. Positive P_IT generated during both the positive and negative hyperpnea patterns, should have the opposite effects of negative inspiratory P_IT. However, when EELV remains at or above FRC, as in positive hyperpneas, expiration is mainly passive due to the elastic recoil of the lungs and chest wall with only a small degree of positive P_IT generated [32]. On the other hand, if EELV falls below FRC, as in negative hyperpneas, expiratory muscles must be activated and, as a consequence, greater positive P_IT would be generated than during positive hyperpneas. Furthermore, generation of positive expiratory P_IT during negative hyperpneas could have a direct compressing effect on the myocardium. This effect would be similar to that arising from external chest compression during cardiac arrest or coughing in participants following life-threatening cardiac arrhythmias, both of which augment SV due to cardiac compression by positive P_IT [12–15]. However, it is possible that the rise in EELV during positive hyperpneas could be due to dynamic hyperinflation with generation of intrinsic positive end-expiratory pressure (PEEP) related to airflow obstruction. If so, end-expiratory pressure may be as positive as in negative hyperpneas. However, against this possibility is that patients with obstructive airways disease were excluded from the study, and when lung function abnormalities are present in patients with chronic HFrEF, these are usually restrictive, not obstructive [33], and would not give rise to dynamic hyperinflation or intrinsic PEEP.

In a previous study involving a separate group of HFrEF patients with CSR-CSA, we demonstrated longer hyperpnea and cycle times and higher plasma N-terminal pro-brain natriuretic peptide concentrations in those with mainly negative than those with mainly positive hyperpneas [11]. Those data suggested that patients with the negative hyperpnea pattern have lower SV and higher LV filling pressures. In the present study, we also found longer cycle and hyperpnea times during negative than during positive hyperpneas, consistent with lower SV during the former. It is, therefore, possible that the generation of the negative hyperpnea pattern is, at least in part, an adaptation to impaired hemodynamics that stimulates engagement of the expiratory muscles to generate positive P_IT to prevent large falls in SV during hyperpnea.

We found greater attenuation in maximum and mean percentage falls in SV during negative than positive hyperpneas, suggesting the possibility that responses to various forms of positive airway pressure may differ between the positive and negative CSR-CSA patterns. The Adaptive Servo-Ventilation for Central Sleep Apnea in Systolic Heart Failure (SERVE-HF) trial found that minute ventilation triggered ASV increased mortality in HFrEF patients [34]. However, patients were not divided into those with positive and negative hyperpnea patterns. Nevertheless, they did find that those at greatest risk of death among those randomized to ASV were patients with worse cardiac function as evidenced by lower LVEF. We have demonstrated in a previous, and in the present study, that among patients with HFrEF and CSR-CSA, those with the negative pattern have worse cardiac function, as evidenced lower LVEF and higher N-terminal pro-brain natriuretic peptide as well as longer CSR-CSA cycle and hyperpnea lengths, than those with the positive pattern of hyperpnea [11]. We would, therefore, speculate that those with the negative pattern of hyperpnea would be at greater risk of adverse effects from ASV than those with the positive pattern. It is difficult to estimate the proportion of participants in the SERVE-HF trial who had a negative hyperpnea pattern. However, in a small sample of participants in the Adaptive Servo-ventilation (ASV) for the Treatment of Obstructive and Central Sleep Apnea in Patients with Heart Failure (ADVENT-HF) trial with CSR-CSA, we found that 27% had predominantly negative patterns of hyperpnea. It will be possible to test prospectively whether responses to ASV therapy differ between the positive and negative forms of CSR-CSA in the setting of the ADVENT-HF trial [11, 35]. This randomized trial is examining the effects of peak flow triggered ASV on morbidity, mortality and cardiac structure and function [36] in HFrEF patients with either CSR-CSA or OSA.

Reduction in LV workload is a mainstay of HFrEF therapy. The combination of BP and HR lowering by angiotensin-converting enzyme inhibitors and beta blockers, reduce LV workload. In the present study, we found that the negative hyperpneas were accompanied by smaller percentage increases in rate pressure product than positive hyperpneas. This was due to a lesser increase in HR, which would allow more time for LV filling. Thus it would appear that not only does the negative hyperpnea pattern preserve SV more than the positive pattern, but it does so at lower HR and LV energy expenditure. This suggests the novel concept that among HFrEF patients with CSR-CSA, energy transfer from the respiratory to the cardiovascular system during negative hyperpneas improves cardiac efficiency compared to positive hyperpneas.

HF patients with CSR-CSA intermittently hyperventilate which increases the load on weak inspiratory muscles and causes PaCO₂ to fall below the apnea threshold [37]. We previously demonstrated that CPAP unloaded weakened inspiratory muscles in HFrEF patients with CSR-CSA and increased their strength, providing evidence for a state of chronic inspiratory muscle fatigue [38]. Recruitment of expiratory muscles causing a fall in EELV below FRC during negative hyperpneas may be a way of unloading weak inspiratory muscles since the initial phase of inspiration is passive due to outward elastic recoil of the chest wall. In addition, the observation that SaO₂ fell 3% less during negative than positive hyperpneas strongly suggests better gas exchange during the former.

Limitations

First, we did not measure P_IT using esophageal manometry. Because of this, the absolute differences in P_IT and respiratory workload between the two hyperpnea patterns could not be determined. There is a limitation in how many physiological variables can be measured in HFrEF patients without disturbing their sleep. Wearing the DPP can be quite cumbersome and necessitates wearing a forearm splint to keep the arm straight. These conditions can be quite uncomfortable and may have limited the amount of sleep, since the average sleep time was only 4 hours (Table 1). Accordingly, we were concerned that if we added an esophageal balloon, sleep time would have been even less and there would not have been enough sleep to obtain sufficient data. Second, it was not possible to measure absolute lung volume. Thus, we cannot determine the effect of absolute differences in inspiratory and expiratory lung volume between the two hyperpnea patterns on SV. Third, we made hemodynamic measurements noninvasively using DPP. Invasive monitoring of hemodynamics was not feasible and it is also not possible to obtain beat-to-beat changes in SV by usual invasive means. Since DPP measures hemodynamic variables at the finger, arterial waveform and absolute values of hemodynamic variables may differ from those measured from the heart or radial artery. Thus, we focused on the percent changes rather than absolute values. We have previously validated relative change in SV by DPP in comparison with that by echocardiographic-Doppler during various respiratory maneuvers, confirming that DPP can provide valid estimates of relative changes in SV that occur over time [10]. In addition, DPP has been used to evaluate SV responses to drug intervention in patients with low SV due to decompensated HFrEF [39].

Conclusions

In conclusion, our results suggest that among HFrEF patients with CSR-CSA, the negative pattern of hyperpnea may have compensatory effects during hyperpnea with more severe degrees of myocardial compromise that recruits the respiratory system to preserve SV at a lower myocardial workload compared to positive hyperpneas. Accordingly, these findings suggest the novel concept that the negative pattern of hyperpnea may have a cardiac auto-resuscitative effect generated by activation of expiratory muscles. Whether this has prognostic significance, or whether application of various forms of positive airway pressure devices to HFrEF patients with CSR-CSA have different effects on outcomes depending on the pattern of hyperpnea remains to be examined in future studies.

Funding

This study was supported by operating grant MOP-82731 from the Canadian Institutes of Health Research. T.I. was supported by unrestricted research fellowships from Nippon Medical School, Phillips Respironics, Japan and Fukuda Foundation for Medical Technology, Japan. T.K. and D.Y. were supported by unrestricted research fellowships from Fuji-Respironics Inc, and Toronto Rehabilitation Institute. E.P. was supported by a fellowship from the University of Brescia, Italy, J.S.F. by a Canada Research Chair in Integrative Cardiovascular Biology, and T.D.B. by the Clifford Nordal Chair in Sleep Apnea and Rehabilitation Research from Toronto Rehabilitation Institute and the Godfrey S. Pettit Chair in Respiratory Medicine from the University of Toronto. Relationships with Industry: T.I. was supported by unrestricted research fellowships from Philips Respironics, Japan and Fukuda Foundation for Medical Technology, Japan. T.K. and D.Y. were supported by unrestricted research fellowships from Fuji-Respironics Inc.

Conflict of interest statement. None declared.