Nocturnal Oxygen Therapy for Central Sleep Apnea in Patients with Heart Failure: A Multisite, Double-Blind, Sham-controlled Randomized Clinical Trial (LOFT-HF)

Abstract

Rationale

There are insufficient data to inform the management of central sleep apnea (CSA) in patients with heart failure with reduced ejection fraction (HFrEF). Nocturnal oxygen therapy (NOT) has been postulated to benefit CSA patients with HFrEF but has not been rigorously studied.

Objectives

To compare NOT with sham NOT (control) in heart failure (HF) patients receiving guideline-based HF therapy on the composite outcome of first occurrence of either mortality due to any cause, a lifesaving cardiovascular intervention, or an unplanned hospitalization for worsening HF, together with other secondary outcomes.

Methods

A multisite, double-blind, sham-controlled randomized clinical trial was conducted from September 2019 to December 2021, when the study was terminated prematurely because of slow enrollment. Cox proportional-hazards regression models were used to analyze time-to-event outcomes.

Results

Ninety-eight participants (mean left ventricular ejection fraction, 27.8 ± 9.6%; mean central apnea–hypopnea index, 30.6 ± 18.2 events/h) were randomized and followed for an average of 10.8 ± 6.3 months. A total of 22 events met the criteria for the primary composite endpoint. The hazard ratio comparing the NOT group with the control group according to time to first event, adjusted for the stratification factor (hospitalization for HF in the past 12 mo and/or elevated outpatient brain natriuretic peptide or N-terminal pro–B-type natriuretic peptide concentration) was 1.46 (95% confidence interval, 0.65–3.29). No group differences in changes in patient-reported outcomes (HF-specific quality of life [Kansas City Cardiomyopathy Questionnaire], sleep disturbance and sleep-related impairment [Patient-reported Outcomes Measurement Information System], generic health [EQ-5D], or mood [Patient Health Questionnaire-8]) were observed at 6 months. Polysomnography showed improved indices of sleep-disordered breathing (apnea–hypopnea index, central apnea–hypopnea index, and time at oxygen saturation < 90%) with oxygen compared with room air.

Conclusions

Although NOT improves CSA and overnight oxygenation, this prematurely terminated study does not provide support for the clinical effectiveness of NOT in patients with CSA and HFrEF.

Clinical trial registered with www.clinicaltrials.gov (NCT 03745898).

Patients with heart failure with reduced ejection fraction (HFrEF) and central sleep apnea (CSA) experience higher heart failure (HF) hospitalization and mortality rates than patients without CSA (1–4). CSA is a marker of poor cardiac function and also may contribute to worse outcomes by causing tissue hypoxia and sleep fragmentation and amplifying sympathetic nervous system activity, an important predictor of mortality in HF (5). Although effective treatment of CSA may improve outcomes in patients with HF, there are insufficient data to inform the management of CSA in patients with HFrEF that persists despite optimal medical management. Although interventions such as pressure-based therapies and phrenic nerve stimulation have been shown to reduce the apnea–hypopnea index (AHI) and several intermediate outcomes (6–8), no prospective randomized controlled trial (RCT) targeting CSA has demonstrated improved clinical endpoints such as reduced mortality or time to first HF hospitalization. Moreover, minute ventilation–triggered adaptive servoventilation in patients with HFrEF was reported to increase mortality (7).

Nocturnal oxygen therapy (NOT) has been proposed as a treatment for CSA in HFrEF. NOT may directly reduce ventilatory drive, thereby lowering loop gain and stabilizing breathing, as well as improve oxygen delivery. Reducing hypoxia during sleep also may blunt sympathetic hyperactivation (9). The potential benefit of NOT is supported by studies demonstrating that low-flow oxygen reduces AHI and improves Hunter-Cheyne-Stokes breathing (10–13). Several small short-term studies also suggest that NOT may improve left ventricular ejection fraction (LVEF), exercise capacity, sleep quality, quality of life, and sympathetic activity (9, 10, 14–17). However, small randomized controlled studies have not replicated benefits on cardiac function or neuropeptide concentrations (10, 18). No study has yet addressed the safety of long-term oxygen use despite its potential to depress hypoxic ventilatory responses and prolong apneas and, at higher flow rates, to result in hyperoxia-related oxygen free radical release, vasoconstriction, and tissue injury (19). Given these evidence gaps, there are no guidelines that address the use of NOT in patients with HFrEF.

A phase III RCT, LOFT-HF (Impact of Low Flow Nocturnal Oxygen Therapy on Hospital Admissions and Mortality in Patients with Heart Failure and Central Sleep Apnea) (NCT 03745898) was designed to evaluate the long-term outcomes of NOT in patients with HFrEF and CSA. Unfortunately, the trial was closed early because of challenges in study operations related to slow enrollment, exacerbated by the coronavirus disease (COVID-19) pandemic. In this paper we report data from 98 participants who were randomized before study closure, providing unique data that could inform future research.

Methods

Study Design and Eligibility

LOFT-HF was a double-blind, sham-controlled RCT designed to compare NOT with sham oxygen (control group) in HFrEF patients on guideline-based medical therapy. Inclusion criteria were CSA (AHI > 15 events/h with ≥50% central events), stable HF on guideline-based medical therapy (as determined by the latest guidelines and confirmed by the site investigator), LVEF ≤ 50%, and New York Heart Association (NYHA) functional class III or IV or NYHA functional class II with high-risk markers. Patients were ineligible for study participation if they regularly used positive airway pressure or oxygen during the day or night; were severely hypoxemic during the day (<90% saturation) or during sleep (<88% saturation for >5 min unaccompanied by respiratory events); had severe chronic obstructive pulmonary disease (COPD); had cardiac surgery, percutaneous coronary intervention, myocardial infarction, unstable angina, transient ischemic attack, or stroke within the prior 3 months; had cardiac resynchronization therapy implantation planned or performed within 3 months before randomization; or had hemodynamically significant uncorrected valvular heart disease, reversible cardiomyopathy within the previous 6 months (e.g., postpartum cardiomyopathy, tachycardia-induced cardiomyopathy), or end-stage HF (stage D HF requiring continuous outpatient intravenous inotropic therapy, placement of a ventricular assist device, listing for cardiac transplantation, or end-of-life care).

Recruitment and Data Collection

Participants were identified from cardiology and sleep clinical settings. Before randomization, CSA eligibility was determined according to central scoring of a clinically obtained sleep study within 6 months before randomization. If not available, an unattended type 2 polysomnogram (Nox Medical A1 polysomnograph) was obtained. After sleep studies were scored, participants who met the study eligibility criteria were invited to participate in an in-person or remote (implemented as an option after the beginning of the COVID-19 pandemic) baseline visit for completion of questionnaires and anthropometry (in-person visits only) and to receive education on healthy sleep and the use of study concentrators. Participants were then randomized to one of the study arms using a web-based randomization module, stratified by the presence of high-risk markers, defined as hospitalization for HF in the past 12 months and/or elevated outpatient brain natriuretic peptide (BNP) or N-terminal pro-BNP concentration, within each site. In-person or telephone-based follow-up visits were conducted to reinforce use of study interventions and to ascertain adverse events (AEs) and healthcare utilization (at Months 1, 3, 6, 9, and 12 and then every 6 mo). At 6 and 12 months, participants also completed patient-reported outcomes questionnaires and were evaluated for NYHA functional classification. Before closure of the study, participants were asked to undergo a repeat unattended type 2 polysomnogram using their assigned therapy. At a final debriefing visit, participants completed a patient satisfaction survey before being told to which arm they had been randomized.

Endpoints and Effect Modifiers

The primary endpoint was first occurrence of mortality due to any cause, a lifesaving cardiovascular intervention, cardiac transplantation, or an unplanned hospitalization for worsening HF. Secondary composite endpoints included first occurrence and recurrent risk of any of the prior events plus hospitalization for myocardial infarction or stroke. Potential clinical endpoints were adjudicated by a blinded clinical endpoints committee using standard criteria (20).

Other secondary endpoints included patient-reported HF-specific quality of life (Kansas City Cardiomyopathy Questionnaire [21]), sleep disturbance and sleep-related impairment (Patient-reported Outcomes Measurement Information System Sleep Disturbance 8a and Sleep-related Impairment 8a questionnaires [21]), generic quality of life (EQ-5D), and mood (Patient Health Questionnaire-8 [22]).

Sleep apnea endotypes, prespecified as potential effect modifiers, were derived from baseline polysomnography using previously reported methods (23).

Interventions

After randomization, participants received either sham or oxygen concentrators (EverFlo Concentrator; Philips Respironics) with humidifiers and remote monitoring capabilities. Sham devices were configured to produce the same flow rates as oxygen concentrators. The initial plan was to titrate oxygen flow using remote oximetry monitoring over several nights with flow rates of 1–4 L/min. This procedure proved to be cumbersome, so early in the trial both groups were assigned to receive air or oxygen at 3 L/min, a commonly used oxygen flow rate (10, 15, 18, 24). Adherence was monitored using a cloud monitoring platform that continuously recorded flow rates and hours of concentrator use (Care Orchestrator cloud-based patient management solution; Philips Respironics). Study coordinators contacted participants when adherence was low to troubleshoot barriers to use and encouraged device use through motivational enhancement techniques adapted from materials used for continuous positive airway pressure (25).

Statistical Analyses and Power

We followed the initial statistical analysis plan, but because of early termination, the prespecified interim efficacy/futility analyses were not performed. The primary analysis was a time–to–first event analysis conducted using Cox proportional-hazards regression, adjusted for the stratification factor used in randomization (presence of high-risk markers: elevated BNP or N-terminal pro-BNP or hospitalized in the past 12 mo). Covariate adjustment was made through inverse probability of treatment weighting. Stabilized weights were used to reduce the variability introduced by study participants with extreme weights (26). Secondary models incorporated additional covariates that showed group imbalance, including self-reported race (Black or other), continuously measured LVEF, and NYHA functional class, into a propensity score model. For continuous outcomes, between-group comparisons and within-group changes over time were assessed using Wilcoxon rank sum tests and signed rank tests, respectively. Two-sided P values and 95% confidence intervals (CIs) are reported.

The projected enrollment was 858 patients (429 per arm), with an average of 2.49 years of follow-up duration. On the basis of a 1-year estimated event rate in the control group of 40%, this sample was estimated to provide 87.9% power at a 5% significance level to detect a hazard ratio (HR) of 0.755, using a two-sided log-rank test, assuming a 5% crossover and 10% loss to follow-up throughout the study.

Results

Figure 1 shows the participant flow diagram. The first participant was enrolled in September 2019 and the last one on June 4, 2021, when the study was closed to enrollment. Follow-up continued through December 15, 2021, providing a minimum 6-month follow-up period for active participants. At the time the study was closed, 192 patients were consented, of whom 167 were evaluated with polysomnography. Of those studies, 62 (37.1%) did not meet criteria for CSA (32 with predominantly obstructive sleep apnea, 30 with AHI < 15 events/h, and 1 with prolonged sleep hypoxemia). Of the 104 (62.3%) participants who met CSA eligibility criteria, 98 were randomized (50 to control, 48 to NOT) from 18 sites (6 eligible participants were not randomized because their eligibility status was identified after enrollment was stopped). Mean follow-up times for clinical endpoint adjudication from randomization to final follow-up were 11.3 ± 6.4 and 10.3 ± 6.2 months for the control and NOT arms, respectively.

Figure 1.

Study flow diagram depicting recruitment, randomization, and follow up. A total of 98 subjects were randomized between October 1, 2019, and June 4, 2021. HF = heart failure; PSG = polysomnography.

Table 1 shows participant characteristics by intervention arm. The sample was 29% Black, 7% Hispanic, and 23% female. The majority of individuals (82%) had high-risk HF markers. The mean LVEF was 27.8 ± 9.6%, and a majority of participants (67%) were classified in NYHA functional class III or IV. The group had severe CSA (mean central AHI 30.6 ± 18.2 events/h) and few obstructive apneas. Sleep architecture was consistent with poor sleep, with increased stage N1 and decreased stage N3. The sample randomized to NOT had a lower average LVEF and included a higher proportion of individuals who were Black or were in NYHA functional class III or IV compared with control subjects.

Table 1.

Participant characteristics by randomization arm

	Control (n = 50)	NOT (n = 48)
Age, yr	62.4 ± 11.6	63.6 ± 11.0
Male	39 (78)	36 (75)
Race
Black or African American	10 (20)	18 (38)
White	37 (74)	29 (60)
Other, multiracial	3 (6)	1 (2)
Hispanic ethnicity	3 (6)	4 (8)
Body mass index, kg/m2	32.6 ± 6.7	30.9 ± 7.5
Left ventricular ejection fraction, %	30.3 ± 10.3	25.2 ± 8.2
NYHA functional class
II	22 (44)	10 (21)
III	27 (59)	38 (79)
IV	1 (2)	0 (0)
Years treated for HF*
≤2	14 (28)	12 (25)
3–5	11 (22)	9 (19)
≥5	25 (50)	26 (54)
Kansas City Cardiomyopathy Questionnaire summary score	68.4 ± 22.1	60.2 ± 24.3
PROMIS SD T score	54.2 ± 8.8	53.8 ± 8.7
PROMIS SRI T score	56.3 ± 11.2	58.2 ± 10.3
Hospitalized for HF in past 12 mo	26 (52)	22 (46)
Elevated BNP or NT-proBNP	34 (68)	38 (79)
BNP,† pg/ml	321.5 (129.2–1,113.5)	324.0 (151.0–871.0)
NT-proBNP,† pg/ml	1,490 (907–2,538)	1,602 (930.5–3,568.5)
History of emphysema or COPD	2 (4)	8 (17)
History of atrial fibrillation	26 (52)	17 (36)
History of diabetes	19 (38)	21 (45)
SGLT2 inhibitors	15 (30)	12 (25)
PSG
Total sleep time	356.7 ± 104.8	325.1 ± 106.1
N1 duration, %	19.6 ± 12.3	17.2 ± 11.7
N3 duration, %	4.5 ± 5.9	4.0 ± 4.7
REM duration TST, %	13.8 ± 8.4	13.2 ± 9.6
Apnea–hypopnea index, events/h	39.5 ± 19.1	38.9 ± 19.0
Central apnea–hypopnea index, events/h	32.3 ± 19.6	28.8 ± 16.5
Obstructive apnea index, events/h	2.9 ± 3.7	4.1 ± 6.4
Hunter-Cheyne-Stokes breathing	48 (96)	44 (92)

Definition of abbreviations: BNP = B-type natriuretic peptide; COPD = chronic obstructive pulmonary disease; HF = heart failure; NOT = nocturnal oxygen therapy; NT-proBNP = N-terminal pro–B-type natriuretic peptide; NYHA = New York Heart Association; PROMIS SD = Patient-reported Outcomes Measurement Information System Sleep Disturbance Scale 8a; PROMIS SRI = Patient-reported Outcomes Measurement Sleep-related Impairment Scale 8a; PSG = polysomnography; REM = rapid eye movement; SGLT2 = sodium-glucose cotransporter 2; TST = total sleep time.

Data are expressed as mean ± standard deviation, n (%), or median (interquartile range).

^*Missing for one participant in the NOT group.

^† Measured values reflect assays available from the medical record.

Clinical Endpoints

A total of 41 clinical events from 25 participants were adjudicated to meet the criteria for clinical endpoints (22 events contributing to the primary composite endpoint, 25 first events that also included stroke and myocardial infarction events). A larger number of primary clinical endpoints occurred in the NOT group (n = 13) than the control group (n = 9). The primary analysis adjusting for the stratification factor using inverse probability of treatment weighting Cox regression yielded an HR of 1.46 (95% CI, 0.65–3.29). This suggestive elevated risk associated with NOT persisted after also controlling for race, LVEF, and NYHA functional class (Table 2, Figure 2).

Table 2.

Time–to–first event primary composite outcome: estimated hazard ratio for unadjusted and adjusted models

Model	HR (95% CI)	P Value
Unadjusted	1.57 (0.67–3.67)	0.30
IPTW* (stratification factor adjusted)	1.46 (0.65–3.29)	0.36
IPTW† (stratification factor and additional covariates adjusted)	1.47 (0.65–3.33)	0.36

Definition of abbreviations: BNP = brain natriuretic peptide; CI = confidence interval; HF = heart failure; HR = hazard ratio; IPTW = inverse probability of treatment weighting; LVEF = left ventricular ejection fraction; NT-proBNP = N-terminal pro–brain natriuretic peptide; NYHA = New York Heart Association.

The reference group is the sham concentrator group. An event was defined as the first occurrence of mortality due to any cause, a lifesaving cardiovascular intervention, cardiac transplantation, or an unplanned hospitalization for worsening HF measured over the duration of the trial, adjudicated by a central clinical endpoints committee. The stratification factor is the presence of high-risk markers: elevated BNP/NT-proBNP or hospitalization in the past 12 months.

^*IPTW model using stabilized weight; the propensity score model adjusted for the stratification factor (presence of high-risk markers: elevated BNP or hospitalization in the past 12 mo).

^† IPTW model using stabilized weight, the propensity score model adjusted for stratification factor and race, LVEF, and NYHA functional class.

Figure 2.

Survival curve: time–to–first event primary composite outcome (the first occurrence of mortality due to any cause, a lifesaving cardiovascular intervention, cardiac transplantation, or unplanned hospitalization for worsening heart failure); unadjusted hazard ratio for control (blue) versus active oxygen treatment (gold) groups. CI = confidence interval.

In secondary analyses of the composite time-to-event endpoint inclusive of stroke and myocardial infarction, the stratification adjusted HR was 1.97 (95% CI, 0.89–4.35) (see Table E1 and Figure E1 in the data supplement). Event rates by the event category and recurrent event rates are shown in Tables E2–E4.

Adherence

Concentrators were used on approximately 50% of nights, with an average of 7.6 ± 2.4 h/night used. Over all nights, concentrator usage averaged 3.5 ± 2.9 h/night. No differences were observed by intervention arm (Table 3).

Table 3.

Patterns of adherence by treatment arm

	Control (n = 47)	NOT (n = 42)	Total (n = 93)	P Value
Percentage of nights used*
Mean	47	55	51	0.201
Range	0.01–0.99	0.07–1.00	0.01–1.00
Mean hours of use over all nights*
Mean ± SD	3.4 ± 3.0	3.6 ± 2.9	3.5 ± 2.9	0.753
Range	0.00–8.89	0.00–9.66	0.00–9.66
Mean hours of use over nights used*
Mean ± SD	7.5 ± 2.9	7.6 ± 1.6	7.6 ± 2.4	0.664
Range	0.30–14.62	4.27–10.59	0.30–14.62

Definition of abbreviations: NOT = nocturnal oxygen therapy; SD = standard deviation.

^*Adherence data could not be retrieved for nine participants (six in the NOT group and three in the control group) because of connectivity or other technical difficulties.

In sensitivity analyses eliminating participants with <2 h/night of average use (n = 56), the fully adjusted HR for the primary outcome was 1.25 (95% CI, 0.45–3.44).

Patient-reported Outcomes

Six-month patient-reported outcomes were available for 72 participants (38 in the NOT arm and 34 in the control arm). Improved outcomes were observed in both arms, but no between-group difference reached statistical significance (Table 4).

Table 4.

Patient-reported outcomes after 6-month follow up by treatment group (n = 72)

	Control (n = 38)		P Value (Within Arm)	NOT (n = 34)		P Value (Within Arm)	P Value for Between-Arm Differences
	Baseline	Change		Baseline	Change
KCCQ score	62.0 ± 23.9	4.8 ± 13.4	0.011	53.6 ± 23.1	7.2 ± 22.3	0.074	0.681
PROMIS SD T score	54.2 ± 8.9	−5.0 ± 8.3	0.002	59.2 ± 8.9	−6.3 ± 7.7	<0.001	0.516
PROMIS SRI T score	54.2 ± 9.6	−4.2 ± 12.6	0.015	57.0 ± 11.0	−5.7 ± 9.4	0.003	0.587
PHQ-8 score	7.5 ± 6.6	−1.6 ± 5.9	0.120	9.1 ± 6.4	−2.2 ± 5.0	0.006	0.199
EQ-5D score	0.8 ± 0.2	0.02 ± 0.1	0.162	0.7 ± 0.2	0.1 ± 0.2	0.056	0.308

Definition of abbreviations: KCCQ = Kansas City Cardiomyopathy Questionnaire; NOT = nocturnal oxygen therapy; PROMIS SD = Patient-reported Outcomes Measurement Information System Sleep Disturbance Scale 8a; PROMIS SRI = Patient-reported Outcomes Measurement Information System Sleep-related Impairment Scale 8a; PHQ-8 = Patient Health Questionnaire-8.

Data are expressed as mean ± standard deviation. For the KCCQ and EQ-5D, higher scores are better; for PROMIS SD, PROMIS SRI, and the PHQ, higher scores are worse.

Polysomnographic Changes

Thirty-four repeat sleep studies (18 on NOT, 16 on control) were performed a median of 11.3 months after randomization, each scored using standardized scoring criteria (but without an airflow channel), blinded to assignment. The group that underwent follow-up polysomnography was similar to the total randomized sample other than fewer endpoints (reflecting censorship related to death or cardiac transplantation). Participants in the NOT arm compared with control participants experienced greater improvements in nocturnal hypoxemia and AHI (AHI change, −20.1 ± 16.4 vs. −5.4 ± 16.3 events/h), which was driven by improvement in central AHI (Table 5). No significant group difference in change in the average duration of apneas or hypopneas was observed.

Table 5.

Follow up polysomnography changes on assigned intervention by treatment arm (n = 34)

	Control (n = 16)		NOT (n = 18)		P Value for Between-Arm Differences
	Baseline	Change (P Value)	Baseline	Change (P Value)
AHI, events/h	38.2 ± 20.3	−5.4 ± 16.3 (0.144)	38.8 ± 19.1	−20.1 ± 16.4 (<0.001)	0.012
CAHI, events/h	31.2 ± 20.6	−6.4 ± 17.0 (0.117)	28.7 ± 16.6	−18.2 ± 15.1 (<0.001)	0.040
OAI, events/h	2.8 ± 3.7	−1.1 ± 2.5 (0.126)	4.1 ± 6.4	−3.0 ± 7.4 (0.007)	0.314
Percentage sleep time at <90% saturation	25.8 ± 30.7	3.9 ± 13.9 (0.583)	17.3 ± 21.6	−13.4 ± 23.0 (0.013)	0.019
Duration of apnea, s	23.6 ± 8.5	−4.6 ± 5.9 (0.021)	21.1 ± 5.6	−2.5 ± 6.0 (0.068)	0.686
Duration of hypopnea, s	22.6 ± 5.7	0.5 ± 6.5 (0.330)	21.6 ± 4.0	4.9 ± 7.3 (0.022)	0.311
New OSA (AHI > 15 events/h, CAHI < 10 events/h)	1 (6.3)		3 (16.7)		0.347
Resolved central apnea (CAHI < 10 events/h)	7 (43.8)		12 (66.7)		0.179
Persistent sleep apnea (AHI > 5 events/h)	15 (94)		16 (89)		0.618

Definition of abbreviations: AHI = apnea–hypopnea index; CAHI = central apnea–hypopnea index; NOT = nocturnal oxygen therapy; OAI = obstructive apnea index; OSA = obstructive sleep apnea.

Data are expressed as mean ± standard deviation or n (%).

Patient Satisfaction Survey

A survey was completed by 57 participants at a debriefing visit. Approximately 46% of participants in each arm reported that they could not guess to which arm they were randomized, while another 16% guessed incorrectly, with no differences by intervention arm. The most common treatment-related problems were related to wearing the nasal cannula and noise from the concentrator. Differences in satisfaction by treatment arm are reported in Table E5.

Effect Moderation

Stratified analyses by endotypes and overnight oxygen saturation suggested that NOT was associated with a higher HR for the composite endpoint including all adverse clinical events in individuals with less baseline hypoxemia (HR, 4.37 [95% CI, 0.96–19.8]; P = 0.056) and higher loop gain (HR, 3.51 [95% CI, 1.01–12.16]; P = 0.047) (see Table E6).

AEs

There was a total of 211 AEs, 106 in the NOT arm and 105 in the control arm. Of these, 4 were adjudicated to be study related (2 in each study arm), and 117 were serious unrelated AEs (60 in the NOT arm, 57 in the control arm) (see Table E7).

Discussion

NOT has been posited to be a physiologically based intervention for reducing the high morbidity and mortality common in patients with CSA and HFrEF through effects on improving ventilatory instability, chemoreceptor overactivity, and hypoxia–reoxygenation tissue injury (27). Despite more than 25 years of interest in targeting hypoxia as a mediator of adverse outcomes, the long-term effectiveness and safety of NOT in patients with CSA and HFrEF has not been evaluated definitively. LOFT-HF was designed to fill this critical knowledge gap. Unfortunately, this trial was stopped prematurely after 98 patients were randomized and followed for an average of 10.8 ± 6.3 months. Although underpowered, LOFT-HF is the largest controlled multicenter trial to date of NOT with the longest follow-up period conducted in patients with HF; moreover, it is the only double-blind, placebo-controlled trial designed to address long-term clinical outcomes in this patient group. As expected, a large improvement in indices of sleep-disordered breathing (AHI, central AHI, and oxygenation) was observed in individuals randomized to NOT compared with control. Unexpectedly, a trend for a higher occurrence of clinical events was observed in the NOT treatment arm, though this did not reach statistical significance. Similar patterns were found in sensitivity analyses that adjusted for baseline participant characteristics, excluded individuals with documented nightly oxygen use of <2 h/night, and in recurrent event analysis. Although underpowered, the findings do not provide support for the clinical effectiveness of low-flow oxygen in patients with CSA and HFrEF despite improvement in sleep-disordered breathing.

Baseline scores for measures of HF-specific quality of life, sleep disturbances, and sleep-related impairment each reflected mild to moderate impairment. Scores in both groups improved, highlighting the potential for placebo effects and the value of controlled study designs. Notably, no between-group differences in scores were observed despite greater improvement in polysomnographic parameters in the NOT group, highlighting the divergence in patient-reported outcomes and physiological measurements.

Support for NOT in CSA was suggested by small short-term studies in patients with HFrEF that demonstrated that low-flow oxygen during sleep virtually eliminated overnight hypoxemia and reduced the number of central apneas (9, 10, 14–16). In LOFT-HF, polysomnography performed at least 3 months after the initiation of intervention also demonstrated that NOT improved overnight hypoxemia and led to an approximately 50% reduction in AHI that reflected a decrease in central events. Oxygen therapy may cause a shift from central to obstructive apneas and prolong respiratory events. Although we observed no effects of the intervention on new development of obstructive sleep apnea (three in the NOT group, one in the control group) or on respiratory event duration, we cannot exclude the possibility that oxygen contributed to adverse effects on ventilatory control.

A cardioprotective role for NOT has been suggested by small, largely uncontrolled studies in patients with HFrEF that reported that low-flow oxygen during sleep positively affected predictors of survival in HF, such as sympathetic activity, peak exercise oxygen consumption, and LVEF (9, 14–16). However, the three largest prior trials of NOT produced conflicting results. A clinical trial of 51 patients followed for 6 months randomized to NOT compared with usual care did not show significant group differences in LVEF (10). Two trials from Japan (56 patients followed for 12 weeks [24] and 52 patients followed for 52 weeks [18]) reported improved LVEF with NOT compared with control. However, a more rigorous comparison did not demonstrate significant between-group differences in LVEF or improved clinical outcomes.

The lack of evidence for a beneficial effect of oxygen in LOFT-HF is similar to the overall null findings from large clinical trials of low-flow oxygen in patients with COPD. Although trials from the 1980s identified survival benefit for treating patients with severe, sustained hypoxemia with continuous oxygen therapy (28), later trials testing NOT in patients with COPD with hypoxemia only during sleep have not identified benefit (29). Notably, the INOX (International Nocturnal Oxygen) trial, although terminated early because of recruitment delays, did not identify evidence for benefit or harm (mortality HR, 0.98 [95% CI, 0.60–1.63]) (29).

The observed trend for an increased rate of clinical endpoints in the NOT arm in LOFT-HF was unexpected. It is possible that this trend would have reversed with further follow-up and patient accrual or that the imbalance in baseline participant characteristics between intervention arms that occurred despite randomization contributed to study findings. As discussed by Shuvy and colleagues (30), the plausibility for a harmful effect of oxygen administration is supported only by studies that evaluated high concentrations of oxygen (≥6 L/min), which can result in formation of oxygen free radicals, altered adaptive redox state, and inflammation. Multiple studies have evaluated high-flow oxygen in normoxic patients presenting with acute myocardial infarction, acute HF, or stroke (31–33). None of these studies demonstrated benefit from NOT, and several studies suggested adverse effects, including increased infarction size (34) and prolonged intensive care unit stay (35). However, LOFT-HF tested NOT at a low flow rate that was monitored in real time and would be expected to deliver oxygen at fractional inspired concentrations lower than 0.4, a threshold considered to denote possible risk if exceeded. We cannot rule out that some individuals, particularly those with less baseline hypoxia, may have been oversupplemented at least for part of their sleep period. The trend for NOT to be associated with higher rates of adverse clinical outcomes in individuals with less baseline hypoxia suggests that may have been the case and that the use of NOT may require more precise oxygen dosing and careful consideration of individual differences in treatment responses.

Although imprecise, the elevated observed HR for NOT in LOFT-HF raises the possibility that there are clinical scenarios in which mild intermittent hypoxemia may be protective and/or low-flow oxygen may be deleterious. Hypoxia has both adverse and beneficial effects that vary by dose and duration of prior and current exposures (36). The complexity of multiple biological systems that are modulated by changes in oxygen tension and that mediate tissue responses to oxygen support the need to study individual variation of responses to supplemental oxygen. This may be relevant to patients with CSA and HF, a heterogeneous group of patients with different sleep apnea endotypes who may respond differently to oxygen (37). LOFT-HF was originally designed to test the latter hypothesis. Although underpowered, prespecified analyses of groups defined by endotypic markers are reported in the supplement.

Lessons Learned

LOFT-HF offers several lessons regarding the implementation of a clinical trial in patients with HF and CSA and the adaptations made during the COVID-19 pandemic. Many procedures were successfully streamlined during the pandemic, such as the use of remote consent procedures and electronic data collection. When a recent clinical sleep study was unavailable, the study used home-based polysomnography with self-applied sensors. Despite the need to often mail devices to participants and provide instructions by phone or video because of the pandemic, high-quality polysomnography data were collected, and most patients reported that this procedure preferrable to a hospital-based sleep study.

Enrollment was slower than anticipated because of delays in activating sites and lower site-specific enrollment. In addition to the challenges of the pandemic, site activation was delayed because of the complexity of coordinating single and local institutional review board agreements, highlighting the need to ensure that all parties fully understand changing regulatory approval processes. Efforts to include new research networks were delayed because of requirements for distinct regularity approvals. The study used a fixed-dollar capitated reimbursement model that was often cited as insufficient for fully supporting the needed personnel to achieve recruitment goals and for supporting patients with multiple health and social needs. Concerns by some patients that oxygen use would result in being labeled with a chronic disease negatively affected enrollment, emphasizing the need to destigmatize device-based therapies.

It was anticipated that oxygen adherence would be higher than what is reported for positive airway pressure. Concentrators, however, were used on only about 50% of nights. Approximately 30% of participants reported problems sleeping because of the noise of the machine and nasal dryness, underscoring the need for both improved technology and individualized patient support. As therapeutic effectiveness requires that patients adhere to the prescribed treatment, our data underscore the importance in addressing and monitoring oxygen adherence when oxygen is prescribed.

Strengths and Limitations

The main study limitation relates to its early closure and recruitment of only 11% of its targeted sample size. The sample had a high prevalence of high-risk markers for HF and may not be broadly representative of patients with HF and CSA. Although rigorous data quality procedures were used for the study’s primary and secondary endpoints, budgetary constraints and the impact of the pandemic on data collection limited other data collection. For example, data from blood assays and echocardiograms were extracted from the medical charts and not collected by the research study.

The study had several strengths, including the enrollment of an ethnically and racially diverse sample of men and women from multiple geographic sites; the use of a sham concentrator that allowed strict double blinding; remote, real-time monitoring of oxygen adherence; high-quality polysomnography at screening and follow-up; well-validated patient-reported outcome instruments; and rigorous endpoint adjudication by an experienced committee.

Conclusions

In summary, this prematurely terminated, multisite, randomized, controlled, double-blind trial of NOT did not provide evidence for the hypothesized reduction in important clinical endpoints, but rather suggested a trend toward the possibility of harm. There also was no evidence for clinically significant differences in patient-reported outcomes between the intervention and control groups. Follow-up polysomnography showed improvements in sleep-disordered breathing and suggested that baseline phenotypic characteristics may influence response to therapy. The challenges experienced highlight the need to use pragmatic clinical trial designs informed by patient input and that are implemented with adequate support for recruiting patients with high disease burden.