Predicting CPAP failure in patients with suspected sleep hypoventilation identified on ambulatory testing

Abstract

Study Objectives:

Home sleep apnea testing (HSAT) is commonly used to diagnose obstructive sleep apnea, but its role in identifying patients with suspected hypoventilation or predicting their response to continuous positive airway pressure (CPAP) therapy has not been assessed. The primary objective was to determine if HSAT, combined with clinical variables, could predict the failure of CPAP to correct nocturnal hypoxemia during polysomnography in a population with suspected hypoventilation. Secondary objectives were to determine if HSAT and clinical parameters could predict awake or sleep hypoventilation.

Methods:

A retrospective review was performed of 142 consecutive patients who underwent split-night polysomnography for suspected hypoventilation after clinical assessment by a sleep physician and review of HSAT. We collected quantitative indices of nocturnal hypoxemia, patient demographics, medications, pulmonary function tests, as well as arterial blood gas data from the night of the polysomnography . CPAP failure was defined as persistent obstructive sleep apnea, hypoxemia (oxygen saturation measured by pulse oximetry < 85%), or hypercapnia despite maximal CPAP.

Results:

Failure of CPAP was predicted by awake oxygen saturation and arterial blood gas results but not by HSAT indices of nocturnal hypoxemia. Awake oxygen saturation ≥ 94% ruled out CPAP failure, and partial pressure of oxygen measured by arterial blood gas ≥ 68 mmHg decreased the likelihood of CPAP failure significantly.

Conclusions:

In patients with suspected hypoventilation based on clinical review and HSAT interpretation by a sleep physician, awake oxygen saturation measured by pulse oximetry and partial pressure of oxygen measured by arterial blood gas can reliably identify patients in whom CPAP is likely to fail. Additional research is required to determine the role of HSAT in the identification and treatment of patients with hypoventilation.

Citation:

Braganza MV, Hanly PJ, Fraser KL, Tsai WH, Pendharkar SR. Predicting CPAP failure in patients with suspected sleep hypoventilation identified on ambulatory testing. J Clin Sleep Med. 2020;16(9):1555–1565.

BRIEF SUMMARY

Current Knowledge/Study Rationale: Clinical pathways using home sleep apnea testing are increasingly common in the diagnosis of obstructive sleep apnea, but their utility in patients with suspected hypoventilation syndromes has not been determined. Specifically, it is unknown if home sleep apnea testing data can be used to differentiate patients whose nocturnal hypoxemia can be treated with continuous positive airway pressure from those who require noninvasive ventilation with or without supplemental oxygen.

Study Impact: Quantitative indices of nocturnal hypoxemia on home sleep apnea testing do not predict the response to continuous positive airway pressure therapy during split-night polysomnography in patients with suspected hypoventilation. Measures of awake hypoxemia were predictive and may be useful in algorithms to identify patients who are unlikely to require noninvasive ventilation.

INTRODUCTION

Obstructive sleep apnea (OSA) is common and associated with several clinical consequences, including cardiometabolic disease, motor vehicle crashes, cognitive impairment, and decreased quality of life.^1,2 Sleep hypoventilation, defined as nocturnal hypoventilation with or without daytime hypercapnia, can occur in up to 50% of those with OSA.^3–6 Patients with sleep hypoventilation are at higher risk of cardiopulmonary complications, are more likely to be admitted to hospital or intensive care, and have higher health care costs compared with patients with eucapnia and obesity or those with uncomplicated OSA.^7,8 Treatment of OSA, with or without hypoventilation, provides clinical benefit and is cost-effective.^9,10

Although laboratory-based polysomnography (PSG) is the gold standard for diagnosis of OSA and titration of continuous positive airway pressure therapy (CPAP), clinical pathways incorporating home sleep apnea testing (HSAT) and auto-titrating CPAP are efficacious and cost-effective.^11,12 Furthermore, HSAT may improve timely access to diagnosis and treatment of OSA and may be more convenient for patients.¹²

Clinical guidelines suggest that HSAT should only be used in patients with a high pretest probability of moderate to severe OSA and without alternative sleep diagnoses or cardiopulmonary disease.^13,14 Polysomnographic diagnosis and positive airway pressure (PAP) titration is recommended to manage patients with suspected hypoventilation¹⁵; however, many patients may initially undergo HSAT due to the frequent overlap with uncomplicated OSA and similarities in clinical presentation. Furthermore, recent studies and clinical practice guidelines have suggested that a large proportion of patients with sleep hypoventilation can be effectively treated with CPAP rather than bilevel positive airway pressure (BPAP).^8,16–20

Although previous studies have evaluated clinical predictors of hypoventilation,^16,21,22 it is unclear whether clinical variables can identify patients who can be treated safely with CPAP and do not require BPAP. Furthermore, the role of HSAT in predicting the response to CPAP has not been investigated. The purpose of this study was to determine, in patients undergoing PSG for suspected hypoventilation, whether HSAT and clinical assessment could identify patients who require BPAP due to failure of CPAP therapy to resolve nocturnal hypoxemia.

METHODS

Study design

This was a single-center, retrospective study of adults referred for split-night PSG for suspected sleep hypoventilation. The primary objective was to identify clinical predictors of CPAP failure to correct hypoxemia due to hypoventilation during split-night PSG, with subsequent need for BPAP with or without supplemental oxygen. The study was performed according to the Declaration of Helsinki and was approved by the University of Calgary Conjoint Health Research Ethics Board [REB 15-3130].

Study setting

The study was conducted at the Foothills Medical Centre (FMC) Sleep Centre, in Calgary, Alberta. The FMC Sleep Centre is the major publicly funded adult sleep laboratory in Southern Alberta and serves a population of approximately 1.2 million people. Newly referred patients with suspected OSA undergo HSAT as part of the triage process. At a subsequent clinic visit, a sleep physician obtains the patient’s history, performs a physical examination, and reviews HSAT results (including review of raw data) and other available clinical data. The physician uses this clinical information to determine whether the patient can safely undergo an ambulatory CPAP titration using autotitrating CPAP or whether PSG is required to evaluate sleep hypoventilation and initiate therapy in a monitored setting.

Polysomnograms for evaluation of sleep hypoventilation are routinely ordered as split-night studies at the FMC Sleep Centre. Once the diagnosis of sleep-disordered breathing (SDB) has been established on the diagnostic portion, a standardized protocol for PAP titration is followed to determine optimal PAP settings and add supplemental oxygen if required (Figure 1). This protocol requires maximal CPAP titration with conversion to BPAP for persistent hypoxemia, uncontrolled OSA, or an increase in transcutaneous partial pressure of carbon dioxide (TcCO₂) by 10 mmHg from baseline awake value (“CPAP failure”). Nocturnal oxygen may be added for persistent hypoxemia despite maximal BPAP titration.

Figure 1. Positive airway pressure titration protocol during split-night polysomnography.

BPAP = bilevel positive airway pressure; CPAP = continuous positive airway pressure; OSA = obstructive sleep apnea; PSG = polysomnogram, SpO₂ = oxygen saturation measured by pulse oximetry, TcCO₂ = transcutaneous partial pressure of carbon dioxide.

Study population

Study patients were identified from the FMC Sleep Centre database. Patients were included if they underwent split-night PSG for suspected hypoventilation with both arterial blood gas (ABG) measurement on the night of the PSG and continuous monitoring of TcCO₂ throughout the study. The suspicion of hypoventilation was based on a clinical assessment and qualitative review of HSAT raw data by the sleep physician; typical indicators of hypoventilation on HSAT included oxygen saturation persistently below 90% and periods of sustained hypoxemia (eg, below 85% for at least 5 consecutive minutes) (Figure 2). Patients were excluded if any of the following criteria were met: 1) nondiagnostic PSG, 2) known neuromuscular disease with respiratory muscle weakness, 3) previous diagnosis of SDB, 4) previous use of PAP therapy, or 5) current treatment with supplemental oxygen.

Figure 2. Home sleep apnea test fragments.

(A) Obstructive sleep apnea without sustained hypoxemia. (B) Obstructive sleep apnea with sustained hypoxemia (possible hypoventilation). Each fragment represents 10 minutes of recording time. SpO₂ = oxygen saturation measured by pulse oximetry.

Data sources

The medical record of each patient was reviewed to obtain demographic and anthropometric data, medications prescribed within the previous year, Epworth Sleepiness Scale (ESS) score, pulmonary function test, ABG results from the night of PSG, and data from both HSAT and PSG.

Arterial blood gas

Samples were collected by a registered respiratory therapist in the sleep laboratory immediately prior to the PSG, with patient supine, awake and breathing room air.

Home sleep apnea test

HSATs that were done using the Remmers Sleep Recorder (Model 4.2, Saga Tech Electronics Inc., Calgary, AB, Canada) or ApneaLink (ResMed Inc., San Diego, CA) monitor were included. Both the Remmers Sleep Recorder and ApneaLink measure oxygen saturation (SpO₂) and heart rate by pulse oximetry, nasal airflow using a pressure transducer, snoring via microphone, and body position; the ApneaLink also includes pneumatic effort sensors within a chest wall band to record chest wall excursion. The severity of OSA was determined on HSAT using the oxygen desaturation index (ODI) for Remmers Sleep Recorder or the respiratory event index (REI) for ApneaLink. Nocturnal hypoxemia was quantified using the mean SpO₂ during the recording, as well as the duration of total recording time (TRT) below prespecified SpO₂ thresholds (TRT < 90%, TRT < 85%, TRT < 80%). All HSATs were interpreted by board-certified or board-eligible sleep medicine physicians at the FMC Sleep Centre.

Polysomnography

All patients underwent overnight attended split-night PSG, which included continuously monitored three channel electroencephalogram (C3C4, 0102; 2-channel electrooculogram, right eye and left eye), submental electromyogram, single-lead electrocardiogram, oral thermistor (Protech; Philips Respironics, Murraysville, PA), nasal pressure (Braebon, Ottawa, ON, Canada), abdominal and thoracic respiratory excursion (Respitrace, Ambulatory Monitoring, Ardsley, NY), and pulse oximetry (NATUS Embla systems, Tonawanda, NY).

PSG was performed using a standardized hypoventilation protocol which included continuous measurement of TcCO₂ (TCM4; Radiometer Medical, Copenhagen, Denmark) and tidal volume derived from expiratory flow within the PAP device. If OSA was identified during the diagnostic portion, CPAP was titrated to a maximum of 18 cm H₂O as per the American Academy of Sleep Medicine (AASM) guidelines.²³ BPAP was initiated if one of the following criteria for “CPAP failure” were met: 1) persistent OSA on CPAP 18 cm H₂O in the absence of mask leak; 2) persistent nocturnal hypoxemia (SpO₂ < 85% for 5 consecutive minutes) on CPAP 18 cm H₂O with OSA controlled; 3) increase in TcCO₂ by at least 10 mm Hg from baseline on CPAP 18 cm H₂O. Once BPAP was initiated, expiratory PAP was titrated to control OSA and inspiratory PAP titrated to achieve tidal volumes of 6–8 ml/kg of ideal body weight. If tidal volume was sufficient and hypoxemia persisted, supplemental oxygen was added to maintain SpO₂ greater than 85% (Figure 1).

Polysomnograms were manually scored by experienced registered PSG technologists according to AASM criteria.²⁴ An obstructive apnea was defined as a fall in the amplitude of the oronasal thermal sensor recording by at least 90% for at least 10 seconds. An obstructive hypopnea was defined by a fall in the amplitude of the oronasal pressure recording by at least 30% for at least 10 seconds followed by an arousal and/or at least a 3% fall in oxygen saturation. All PSGs were interpreted by respiratory physicians who were board certified or board eligible in sleep medicine.

Study outcomes

The primary objective of the study was to identify predictors of CPAP failure during split-night PSG. Secondary objectives included the identification of predictors of awake hypoventilation and of sleep hypoventilation. Awake hypoventilation was defined as an awake arterial partial pressure of carbon dioxide (PaCO₂) greater than 45mm Hg. Sleep hypoventilation during diagnostic PSG was defined as either: 1) 10 mm Hg or greater increase in TcCO₂ on transition from wakefulness to sleep to a value of at least 50 mm Hg or 2) an absolute TcCO₂ value of at least 55 mm Hg during sleep.²⁵

Statistical analysis

Means and standard deviations or median and interquartile ranges were calculated for parametric and nonparametric continuous variables, respectively. Regression analysis was used to identify univariate predictors of CPAP failure. The following candidate variables were specified a priori: sex, age, body mass index, SpO₂, ESS score, and use of opioids or benzodiazepines. Pulmonary function measurements were also included if available. From HSAT, candidate variables included REI, ODI, mean SpO₂, nadir SpO₂, and TRT < 90%, TRT < 85%, and TRT < 80%. Variables from PSG were not included in the statistical models as one of the aims of this study was to identify patients that might avoid polysomnographic PAP titration. The exception was change in TcCO₂, because this measurement is increasingly used in ambulatory settings and could be used as an adjunct to HSAT. Variables that were predictive on univariate analysis were included in multivariable logistic regression models, which were then reduced by stepwise regression. The same variables were used to identify predictors of awake and sleep hypoventilation. Operating characteristics were calculated for multivariable predictors of CPAP failure, using clinically relevant threshold values. STATA IC 13.1 (StataCorp, College Station, TX) was used for all statistical calculations.

RESULTS

Five hundred ninety-eight patients underwent split-night PSG between November 2016 and July 2017, of which 456 patients were excluded (Figure 3). The main reasons for exclusion were previous diagnosis of SDB, absence of an interpretable ABG on the night of the PSG, or if the patient was an inpatient on the night of PSG. Table 1 outlines the characteristics of the remaining 142 patients that were included in the study. The study cohort generally consisted of middle-aged men with obesity and women with severe OSA and nocturnal hypoxemia. In addition, 52 patients (37%) had awake hypoxemia (PaO₂ < 60 mm Hg) and 26 patients (18%) had awake hypercapnia (PaCO₂ > 45 mm Hg). Pulmonary function test data were available for 102 patients, and the results demonstrated that most patients did not have significant lung disease.

Figure 3. Study flow diagram.

Note that the study flow does not represent patient flow; HSAT preceded PSG in all patients. ABG = arterial blood gas, CPAP = continuous positive airway pressure, HSAT = home sleep apnea test, OSA = obstructive sleep apnea, PSG = polysomnogram.

Table 1.

Baseline patient characteristics.

	All Patients N = 142	CPAP Sufficient N =112	CPAP Failure N = 30	P Value
Age, years	56 (13)	55 (13)	59 (12)	.2073
Sex				.8928
Male, n (%)	82 (58%)	65 (58%)	17 (57%)
Female, n (%)	60 (42%)	47 (42%)	13 (43%)
BMI, kg/m2†	40 (35,48)	40 (35,47)	44 (38,50)	.0540
Epworth Sleepiness Scale score*†	9 (6,13)	9 (6,13) (N=103)	10 (6,13) (N=28)	.9126
Home sleep apnea test				.0935
Remmers Sleep Recorder, n (%)	94 (66%)	78 (69%)	16 (53%)
ApneaLink, n (%)	48 (34%)	34 (31%)	14 (47%)
Remmers Sleep Recorder
ODI, events/h†	46 (24,70)	47 (24,75)	38 (24,60)	.2641
Mean oxygen saturation, % †	85 (81,87)	85 (82,87)	82 (79,84)	.0034
ApneaLink Recorder
REI, events/h	50 (22)	49 (22)	51 (25)	.8131
Mean oxygen saturation, % †	82 (79,85)	83 (81,85)	77 (72,82)	.0108
Arterial blood Gas
pH†	7.43 (7.40,7.45)	7.43 (7.41,7.45)	7.41 (7.39,7.43)	.0491
PaCO2, mm Hg†	42 (39,44)	41 (38,44)	45 (43,50)	.0000
PaO2, mm Hg	63 (10)	66 (10)	54 (8)	.0000
Bicarbonate, mmol/L†	27 (25,29)	26 (25,28)	28 (27,32)	.0000
SaO2, % †	91 (88,93)	91 (89,93)	87 (83,90)	.0000
Spirometry**
FEV1, L†	2.12 (1.73, 2.75)	2.28 (1.87,2.95)	1.84 (1.54,2.39)	.0215
FEV1, % predicted	78 (20)	82 (20)	65 (15)	.0002
FVC, L†	3.11 (2.46, 3.77)	3.11 (2.50,3.90)	3.06 (2.34,3.65)	.1893
FVC, % predicted†	83 (72,94)	86.5 (77, 95)	72.0 (67,76)	.003
FEV1/FVC†	75 (68,79)	75 (70,79) (N=77)	71 (58,78) (N=25)	.0472
Medication use
Opioids, n (%)	16 (11%)	11 (10%)	5 (17%)	.2923
Benzodiazepines, n (%)	13 (9%)	8 (7%)	5 (17%)	.1082
Diagnostic PSG
TST, min†	109 (90,137)	114 (91,142)	100 (80,116)	.0407
Sleep efficiency, % †	71 (58,82)	69 (59,82)	76 (55,82)	.5095
Sleep latency, min†	11 (6,24)	12 (6,30)	8 (6,20)	.2198
AHI, events/h†	64 (31,101)	64 (32,99)	85 (25,110)	.6385
Mean oxygen saturation, % †	86 (82,89)	87 (83,90)	82 (80,84)	.0000
% of TST with SpO2 90–100% †	11 (1,36)	16 (4,46)	1 (0,3)	.0000
% of TST with SpO2 80–89% †	62 (37,81)	61 (37,80)	65 (38,84)	.5706
% of TST with SpO2 70–79% †	5 (0,30)	3 (0,20)	30 (10,41)	.0000

^†Nonparametric data, presented as median (interquartile range); all other data are presented as mean (standard deviation). *Epworth score available for 131 patients. **Spirometry available for 102 patients. AHI = apnea-hypopnea index, BMI = body mass index, FEV₁ = forced expiratory volume in 1 second, FVC = forced vital capacity, NREM = nonrapid eye movement sleep, ODI = oxygen desaturation index, PaCO₂ = partial pressure of carbon dioxide (arterial blood gas), PaO₂ = partial pressure of oxygen (arterial blood gas), PSG = polysomnography, REI = respiratory event index, REM = rapid eye movement sleep, SaO₂ = measured oxygen saturation (arterial blood gas), SpO₂ = oxygen saturation (pulse oximetry), TST = total sleep time.

Primary outcome

Failure of CPAP occurred in 30 patients (21%) and was due to persistent nocturnal hypoxemia in all patients. No patients required BPAP for uncontrolled OSA. Patients who failed CPAP had more severe awake and nocturnal hypoxemia, higher PaCO₂, and lower percent predicted FEV₁, FVC, and FEV₁/FVC than those patients whose hypoxemia was corrected by CPAP. However, there were no intergroup differences in age, sex, body mass index, ESS, ODI, or REI or benzodiazepine or opioid use.

Univariate predictors of CPAP failure included awake SpO₂ on PSG, percent predicted FEV1, FVC, and FEV1/FVC ratio (Table S1 in supplemental material). Predictors of CPAP failure from ABG included oxygen saturation, partial pressure of oxygen (PaO₂), partial pressure of carbon dioxide (PaCO₂), and bicarbonate. Variables from HSAT that were predictive of CPAP failure included mean nocturnal SpO₂, TRT < 90%, TRT < 85%, and TRT < 80%. Measures of OSA severity derived from HSAT (ODI and REI) were not predictive of CPAP failure. Change in TcCO₂ was not associated with CPAP failure on univariate analysis.

Multivariable analysis revealed that HSAT indices of nocturnal hypoxemia did not predict CPAP failure. However, awake saturation measured by pulse oximetry (SpO₂), PaO₂ and PaCO₂ predicted CPAP failure. Operating characteristics for threshold values of these multivariable predictors are presented in Table 2. No patient with awake resting SpO₂ ≥ 94% failed CPAP. Similarly, awake PaO₂ ≥ 68 mm Hg or PaCO₂ < 40 mm Hg virtually excluded CPAP failure (positive predictive value 0.98).

Table 2.

Threshold values for excluding CPAP failure.

Threshold	Sensitivity (%)	Specificity (%)	Positive Predictive Value (%)	Negative Predictive Value (%)
SpO2, %
≥ 95	6	100	100	22
≥ 94	21	100	100	25
≥ 92	46	80	90	29
≥ 90	71	70	90	40
≥ 88	86	57	88	52
PaO2, mmHg
≥ 70	32	97	97	28
≥ 68	39	97	98	30
≥ 65	52	87	94	33
≥ 60	75	80	93	46
PaCO2, mmHg
≤ 40	38	97	98	29
≤ 45	82	50	86	43
≤ 48	94	43	86	65
≤ 50	97	27	83	73

PaCO₂ = partial pressure of carbon dioxide (arterial blood gas), PaO₂ = partial pressure of oxygen (arterial blood gas), SpO₂ = oxygen saturation (pulse oximetry).

Secondary outcomes

Twenty-six patients (18%) had awake hypoventilation (Table 3). These patients had more severe nocturnal and awake hypoxemia. However, there were no intergroup differences in age, sex, body mass index, ESS, ODI, REI, or measurements of lung function. Univariate predictors of awake hypoventilation included oxygen saturation and awake SpO₂ (Table S2). Mean SpO₂, TRT < 90%, TRT < 85%, and TRT < 80% on HSAT were found to be univariate predictors of awake hypoventilation, but none were predictive on multivariable analysis. Fifty-nine patients (42%) had sleep hypoventilation (Table 3). These patients had higher body mass index, more severe awake and nocturnal hypoxemia, and higher PaCO₂ than patients without nocturnal hypoventilation. In addition, percent predicted FEV₁ and FVC were significantly lower. However, there were no intergroup differences in age, sex, ESS, ODI, or REI. Univariate predictors of sleep hypoventilation included mean SpO₂, nadir SpO₂, TRT < 90%, TRT < 85%, and TRT < 80% on HSAT (Table S3). Only PaCO₂ on ABG remained predictive of sleep hypoventilation on multivariable analysis.

Table 3.

Characteristics of patients with awake and sleep hypoventilation.

	Awake PaCO2 ≤ 45 mm Hg N = 116	Awake PaCO2>45 mm Hg N = 26	P Value	No Sleep Hypoventilation N = 80	Sleep Hypoventilation N = 59	P Value
Age, years	56 (13)	59 (12)	.1817	56 (12)	57 (13)	.6276
Sex			.9951			.6307
Male, n (%)	67 (58%)	15 (58%)		48 (60%)	33 (56%)
Female, n (%)	49 (42%)	11 (42%)		32 (40%)	26 (44%)
BMI, kg/m2†	40 (35,47)	43 (36,51)	.1811	39 (34,47)	43 (37,49)	.0157
Epworth Sleepiness Scale score*†	9 (6,13)	8 (4,13)	.5871	10 (7,14)	8 (6,13)	.2962
	(N=106)	(N=25)		(N=73)	(N=55)
Home Sleep Apnea test			.0534			.0282
Remmers Sleep Recorder, n (%)	81 (70%)	13(50%)		59 (74%)	33 (56%)
ApneaLink, n (%)	35 (30%)	13(50%)		21 (26%)	26 (44%)
Remmers Sleep Recorder
ODI, events/h†	46 (24,70)	42 (26,82)	.5653	55 (26,77)	38 (21, 61)	.0646
Mean oxygen saturation, % †	85 (82,87)	82 (80,85)	.0506	86 (83,87)	83 (80,87)	.0829
ApneaLink Recorder
REI, events/h	49 (21)	52 (27)	.7849	45 (22)	54 (23)	.1927
Mean oxygen saturation, % †	83 (80, 85)	77 (72,83)	.0405	85 (82, 87)	81 (77,83)	.0049
Arterial blood gas
pH†	7.43 (7.41,7.45)	7.40 (7.38,7.42)	.0000	7.43 (7.41,7.45)	7.42 (7.40,7.45)	.3113
PaCO2, mm Hg†	41 (38,43)	49 (48,51)	.0000	41 (38,44)	44 (40,47)	.0006
PaO2, mm Hg	64 (10)	56 (10)	.0001	65 (11)	61 (10)	.0222
Bicarbonate, mmol/L†	26 (25,28)	30 (28,32)	.0000	26 (25,28)	28 (26,31)	.0006
SaO2, % †	91 (89,93)	88 (84,89)	.0000	91 (88,93)	90 (87,92)	.2315
Spirometry**
FEV1, L†	2.13 (1.76,2.89)	2.12 (1.54,2.61)	.3084	2.26 (1.80,3.10)	2.08 (1.57,2.61)	.1513
FEV1, % predicted	79 (19.5)	72 (22)	.2074	82 (22)	74 (18)	.0439
FVC, L†	3.10 (2.48,3.90)	3.16 (2.45, 3.69)	.4418	3.11 (2.54,3.96)	3.15 (2.45,3.69)	.3339
FVC, % predicted†	83 (74,94)	74 (69,88)	.0758	88 (76,87)	78 (72,89)	.0423
FEV1/FVC†	75 (68,79)	72 (62,80)	.3014	76 (70,79)	74 (62,79)	.2676
	(N=83)	(N=19)		(N=50)	(N=49)
Medication use
Opioids, n (%)	14 (12%)	2 (8%)	.5335	8(10%)	7 (12%)	.7262
Benzodiazepines, n (%)	11 (9%)	2 (8%)	.7748	8 (10%)	4 (7%)	.040
Diagnostic PSG
TST, min†	112 (91,141)	97 (40,123)	.0250	118 (92, 148)	102 (88, 126)	.0170
Sleep efficiency, % †	70 (59,82)	73 (51,83)	.8826	71 (60, 84)	71 (56,82)	.6653
Sleep Latency, min†	11 (6,26)	11 (8,20)	.6769	11 (5,24)	11 (7,23)	.6211
AHI, events/h†	63 (27,96)	85 (40,111)	.1248	64 (32,107)	64 (26,98)	.6316
Mean oxygen saturation, % †	86 (82,89)	83 (78,85)	.0003	88 (84,91)	83 (80,87)	.0000
% of TST with SpO2 90–100% †	14 (3,41)	2 (0,10)	.0005	19 (4,58)	5 (0,16)	.0000
% of TST with SpO2 80–89% †	61 (38,81)	66 (36,81)	.7297	58 (36,80)	65 (38,82)	.3964
% of TST with SpO2 70–79% †	3 (0, 20)	30 (10,41)	.0007	2 (0,14)	17 (2,39)	.0002

^†Nonparametric data, presented as median (interquartile range); all other data are presented as mean (standard deviation). *Epworth score available for 131 patients. **Spirometry available for 102 patients. AHI = apnea-hypopnea index, BMI = body mass index, FEV₁ = forced expiratory volume in 1 second, FVC = forced vital capacity, NREM = nonrapid eye movement sleep, ODI = oxygen desaturation index, PaCO₂ = partial pressure of carbon dioxide (arterial blood gas), PaO₂ = partial pressure of oxygen (arterial blood gas), PSG = polysomnography, REI = respiratory event index, REM = rapid eye movement sleep, SaO₂ = measured oxygen saturation (arterial blood gas), SpO₂ = oxygen saturation (pulse oximetry), TST = total sleep time.

DISCUSSION

The results of this study indicate that in a population of patients with suspected hypoventilation based on interpretation of the HSAT, there were several clinical and physiologic variables that were associated with failure of CPAP to resolve nocturnal hypoxemia during split-night PSG. However, only awake SpO₂, PaO₂, and PaCO₂ were predictive of this outcome on multivariable analysis. Although HSAT was part of the clinical assessment and often demonstrated oxygenation profiles, such as sustained hypoxemia, that were suspicious for hypoventilation, quantitative indices of nocturnal hypoxemia were not predictive of the response to CPAP. The analysis also revealed that physiologic cut points for SpO₂ and PaO₂ could be used to exclude failure of CPAP. This is the first study to investigate clinical predictors of the response to CPAP therapy during PSG in patients with suspected hypoventilation; furthermore, it is the first study to incorporate HSAT data into this analysis.

The definition of CPAP failure is variably described in the literature.^23,26–29 Most authors have recommended conversion from CPAP to BPAP if OSA is not controlled or if there is evidence of persistent nocturnal hypoxemia on maximal CPAP, defined as 15–20 cm H₂O.^8,23 Previous studies have assessed the effectiveness of CPAP over weeks to years in patients with awake hypoventilation.^{17,18,20,26–29} In this study, we used the AASM criteria for CPAP failure using a single split-night PSG to efficiently determine the most effective initial PAP therapy to treat SDB.²³ Despite different timelines for assessing response to CPAP therapy, our findings are consistent with prior studies in demonstrating that measurements of awake gas exchange are the most powerful predictors of CPAP failure. This study extends previous findings by providing possible threshold values that clinicians could use to exclude CPAP failure during polysomnographic PAP titration.

Prior studies have examined clinical predictors of awake hypoventilation (PaCO₂ > 45 mmHg) in patients referred for evaluation of sleep apnea.^{16,22,30–33} Only 2 of these studies used ambulatory alternatives to PSG to evaluate SDB. In a retrospective study, Macavei et al²² used overnight oximetry or limited portable polygraphy in patients with obesity referred for diagnostic sleep apnea testing. Mandal and colleagues²¹ prospectively evaluated whether findings from overnight oximetry could identify patients with more severe SDB. These studies used oximetry or limited polygraphy as inclusion criteria and not to evaluate whether they predicted awake hypercapnia. Our study examined specific quantitative indices of nocturnal hypoxemia on HSAT but did not find they were predictive of hypoventilation on multivariable analysis. Similar to our study, however, Macavei’s and Mandal’s groups^21,22 found a high proportion of patients with awake hypoventilation. Additionally, our study demonstrated a high prevalence of nocturnal hypoventilation. These results demonstrate that ambulatory tests for SDB, such as oximetry and HSAT, can be used to identify patients with more severe disease.

Polysomnographic variables have been used to predict obesity hypoventilation syndrome. The severity of nocturnal hypoxemia characterized by percent of TST with SpO₂ less than 90% (TST < 90%) has consistently predicted awake hypercapnia.^{22,30,31,34–36} The analogous variable to TST < 90% on HSAT is TRT < 90%; this variable was predictive in univariate analysis for CPAP failure and hypoventilation, but not in multivariable analysis. Since HSAT does not record sleep duration, it is possible that sleep-related hypoxemia was underestimated in this study. Apnea-hypopnea index greater than 50 events/h has also demonstrated predictive ability for hypoventilation.^3,36 Mandal’s study using overnight oximetry found that an ODI using a 4% oxygen desaturation threshold predicted awake hypoventilation,²¹ but the current study did not reveal similar results. This finding may be explained by the known underestimation of apnea-hypopnea by HSAT¹³ or may be related to the model of SDB care at the FMC Sleep Centre; specifically, patients with uncomplicated OSA typically undergo ambulatory CPAP titration using autotitrating equipment and are not referred for PSG. Thus, in this study cohort of patients with suspected hypoventilation based on HSAT and clinical assessment, the predictive ability of ODI or REI may have been weakened.

Although PaCO₂ was associated with sleep hypoventilation on multivariable analysis, changes in TcCO₂ did not predict CPAP failure. The lack of a relationship between sleep hypoventilation and CPAP failure is consistent with the clinical observation that CPAP will often correct gas exchange abnormalities. Furthermore, randomized controlled trials have demonstrated that in patients with obesity hypoventilation syndrome, treatment outcomes are similar between CPAP and BPAP.^17,18 Given this uncertainty about the clinical utility of isolated sleep hypoventilation, it is more clinically useful to identify patients who will fail CPAP and require polysomnographic BPAP titration.

The results of this study could be used to identify patients with OSA who could forego polysomnography and proceed to ambulatory CPAP titration. For example, a finding of awake SpO₂ ≥ 94% would support proceeding to ambulatory CPAP titration, whereas SpO₂ < 94% would prompt ABG measurement. If PaO₂ ≥ 68 mm Hg, patients could proceed to ambulatory CPAP titration and all other patients would be referred for polysomnographic PAP titration since CPAP failure could not be excluded. Although PaCO₂ was significant on multivariable analysis, it did not perform better than indices of oxygenation. Furthermore, a threshold value of 40 mm Hg is unlikely to be clinically relevant in this population. Importantly, any algorithm derived from these results should be prospectively tested using a validation cohort before implementation in clinical practice.

This study has several limitations. First, the study was retrospective and, consequently, complete data were not available for all patients. Although the majority of data were available, it was incomplete for ESS (n = 131, 92%) and pulmonary function tests (n = 102, 72%). This issue was most relevant for pulmonary function tests, although statistical significance of spirometric parameters on univariate analysis suggests that power was adequate even with limited data. Prospective validation studies with comprehensive data collection are indicated to confirm the accuracy of predictive variables identified in this study. Second, the clinical suspicion of hypoventilation was based primarily on a sleep physician’s qualitative interpretation of oximetry raw data from HSAT, in the context of the patient presentation. Without clear thresholds for sustained hypoxemia, generalizability to other settings may be limited. Furthermore, the use of qualitative assessment for patient selection and quantitative measures in our analysis may have restricted the power of nocturnal hypoxemia to predict clinical outcomes. Further analysis of oximetry patterns from HSAT could be explored in future studies.

A third limitation relates to selection bias, which could have arisen through 2 aspects of the study design. We selected patients with more severe SDB, which could limit the applicability of the findings to clinical populations with a wider range of SDB severity. Notwithstanding that limitation, the research question is most relevant for patients with more severe disease. While a similar study in a broader range of patients might identify stronger associations between HSAT variables and polysomnographic CPAP titration failure, we believe that the negative associations demonstrated in this study population are still valuable. A related limitation is the large number of patients who were excluded in the analysis (456/598), mostly due to a previous diagnosis of SDB or missing test data. Although the decision to exclude these patients was intended to strengthen the statistical analysis, we acknowledge that it limits generalizability. Prospective validation studies are required before these findings can be applied in actual clinical practice.

CONCLUSIONS

This study demonstrated that awake SpO₂ and ABG analysis can be used to predict CPAP failure on split-night PSG in patients whose clinical assessment and HSAT suggest hypoventilation. In an era of constrained health care resources, an algorithm to differentiate patients who can safely initiate CPAP therapy in an ambulatory setting from those who require titration using PSG could promote more efficient and cost-effective use of resources. However, further prospective studies are required to validate these findings before implementing them into clinical management pathways.

DISCLOSURE STATEMENT

All authors have seen and approved this manuscript. Work for this study was performed at Sleep Centre, Foothills Medical Centre, Cumming School of Medicine, University of Calgary. This study was funded by the Alberta Health Services Respiratory Health Strategic Clinical Network Summer Studentship for Minyoung Lee. Dr. Braganza was supported by the Canadian Sleep and Circadian Network (CSCN) Canadian Sleep Medicine Fellowship and the Cumming School of Medicine Sleep Research Program at the University of Calgary. Dr. Hanly has received financial support and positive airway pressure equipment for clinical research from Philips Respironics. The authors report no conflicts of interest.

ABBREVIATIONS

AASM

American Academy of Sleep Medicine

ABG

arterial blood gas

BPAP

bilevel positive airway pressure

CPAP

continuous positive airway pressure

ESS

Epworth Sleepiness Score

FEV₁

forced expiratory volume in 1 second

FMC

Foothills Medical Centre

FVC

forced vital capacity

HSAT

home sleep apnea test

ODI

oxygen desaturation index

OSA

obstructive sleep apnea

PaCO₂

partial pressure of carbon dioxide measured by arterial blood gas

PaO₂

partial pressure of oxygen measured by arterial blood gas

PAP

positive airway pressure

PSG

polysomnography

REI

respiratory event index

SDB

sleep-disordered breathing

SpO₂

oxygen saturation measured by pulse oximetry

TcCO₂

transcutaneous partial pressure of carbon dioxide

TRT

total recording time on home sleep apnea test

TST

total sleep time on polysomnography