High-Flow Nasal Cannula Oxygen Therapy Can Be Effective for Patients in Acute Hypoxemic Respiratory Failure with Hypercapnia: a Retrospective, Propensity Score-Matched Cohort Study.

BACKGROUND: Usually, high-flow nasal cannula (HFNC) therapy is indicated for de novo acute hypoxemic respiratory failure (AHRF). Although only a few researches have examined the effectiveness of HFNC therapy for respiratory failure with hypercapnia, this therapy is often performed under such conditions for various reasons. We investigated the effectiveness of HFNC therapy for AHRF patients with hypercapnia compared to those without hypercapnia.
METHODS: All consecutive patients receiving HFNC therapy between January 2012 and June 2018 at a university hospital were enrolled and classified into nonhypercapnic and hypercapnic groups. We compared the outcomes of both groups and adjusted the outcomes with propensity score matching.
RESULTS: A total of 862 patients were enrolled, of which 202 were included in the hypercapnic group. HFNC weaning success rates were higher, and intensive care unit (ICU) and hospital mortality was lower in the hypercapnic group than in the nonhypercapnic group (all P < 0.05). However, no statistical differences in HFNC weaning success (adjusted P = 0.623, matched P = 0.593), ICU mortality (adjusted P = 0.463, matched P = 0.195), and hospital mortality (adjusted P = 0.602, matched P = 0.579) were noted from the propensity-adjusted and propensity-matched analyses. Additionally, in the propensity score-matched subgroup analysis (according to chronic lung diseases and causes of HFNC application), there was also no significant difference in outcomes between the two groups.
CONCLUSION: In AHRF with underlying conditions, HFNC therapy might be helpful for patients with hypercapnia. Large prospective and randomized controlled trials are required for firm conclusions.

INTRODUCTION

Oxygen administration is an essential supportive treatment for maintaining proper tissue oxygenation and alleviating breathlessness in patients with respiratory failure (RF). However, the choice of an appropriate form of oxygen therapy is unclear. The high-flow nasal cannula (HFNC) is a high-flow oxygen supply device that is increasingly used in the treatment of RF in intensive care unit (ICU) patients.1 HFNC therapy is usually used to treat de novo acute hypoxemic respiratory failure (AHRF) without hypercapnia and reduces the need for tracheal intubation compared to conventional oxygen therapies.23 In addition, HFNC therapy is also helpful for post extubation respiratory failure and invasive airway procedure such as intubation and bronchoscopy.456

In some patients with chronic lung disease (CLD), high-flow oxygen supply can aggravate hypercapnia,78 and noninvasive ventilation (NIV) is strongly recommended in patients with hypercapnic RF9; therefore, meticulous care is required in such cases when using an HFNC. However, in clinical practice, physicians have been using HFNC therapy frequently not only in patients with de novo AHRF but also in patients with CLD. Several studies have reported that HFNC therapy might also be helpful in patients with underlying CLD,10111213 while others reported no significant increase in carbon dioxide (CO2) levels and progression to NIV or mechanical ventilation in patients with hypercapnia compared to those without hypercapnia after HFNC therapy.141516

We compared the effectiveness of HFNC therapy for patients with or without hypercapnia in AHRF using a risk stratification model that adjusts for potential differences between the two groups. The primary outcome was the HFNC weaning success rate, and the secondary outcomes were ICU mortality, hospital mortality, length of ICU stay, length of hospital stay, and complications during ICU stay. We also compared vital signs, pH and partial pressure of CO2 (PCO2) changes between the two groups to evaluate the short-term effects after HFNC application.

METHODS

Study design and subjects

We performed a retrospective observational study of critically ill patients older than 18 years who received HFNC therapy for a hypoxemic respiratory problem (ratio of partial pressure arterial oxygen [PO2] and fraction of inspired oxygen [FiO2] ≤ 200 mmHg) at the ICU of Ulsan University Hospital, Korea, between January 2012 and June 2018. Using the PCO2 in the arterial blood gas analysis (ABGA) just before HFNC therapy, we classified the study patients into nonhypercapnic and hypercapnic groups. We compared the outcomes of both groups and adjusted our model using propensity score analysis. Additionally, we performed further analyses of the outcomes in selected subgroups: patients without CLD, patients with CLD, patients who received HFNC therapy after extubation, and patients who received HFNC therapy for causes other than extubation.

HFNC device application

All study patients were treated with high-flow oxygen using an HFNC (Optiflow™ or AIRVO™ 2; Fisher and Paykel Healthcare, Auckland, New Zealand). Each patient was kept under close observation in the ICU and was continuously monitored until their conditions stabilized. We routinely determined HFNC application based on the attending physician’s assessment and consulted with the respiratory medicine specialists or critical care intensivists if necessary.

Definitions

We defined “hypercapnia” as PCO2 ≥ 45 mmHg based on pre-HFNC ABGA. “HFNC weaning” was defined as spontaneous breathing with an oxygen flow ≤ 6 L/min via a nasal cannula for > 48 hours after stopping HFNC therapy. “Use of immunosuppressive agents” was defined as treatment with steroids, immunosuppressive medications, and/or chemotherapeutic agents within 6 months of HFNC therapy. “After extubation” was defined as a state of HFNC application within 6 hours after extubation. Acute physiology and chronic health evaluation (APACHE) II and sequential organ failure assessment (SOFA) scores were calculated using the worst variable within 24 hours of HFNC application. The study participants were classified into four groups using the modified classification criteria according to the causes of RF necessitating the use of HFNC17: de novo acute RF, acute-on-CLD, septic shock for reasons other than respiratory infection, pulmonary edema, or after extubation. In cases of more than one cause of HFNC therapy in a patient, we included the major cause for HFNC indication. We assessed the short-term effectiveness of HFNC therapy by checking vital signs, pH and PCO2 values immediately before and after HFNC application within 2 hours.

Data collection

Clinical and laboratory findings were obtained from the clinical data warehouse appliance (uICE, Ulsan University Hospital Information of Clinical Ecosystem) in connection with the electronic medical records at the Ulsan University Hospital. One critical care intensivist (also specializing in respiratory medicine) reviewed all the patient records collected from the uICE and checked any faults by identifying the patient records directly.

Statistical analysis

The data are reported as means ± standard deviation, medians (interquartile range) or numbers (percentages). To investigate comparisons between nonhypercapnic and hypercapnic groups, the independent t-test was performed for continuous variables and the χ2 test was used for categorical variables.

In our study, patients were not randomly assigned to the nonhypercapnic and hypercapnic groups. To reduce the effect of selection bias based on hypercapnia and a potential confounding factor in the study, we performed careful adjustment for differences in baseline characteristics except ABGA using a propensity score analysis.18 Propensity scores were estimated by multiple logistic regression analysis, and all covariables previously described in Tables 1 except ABGA were included in the final model. To check multicollinearity, we also calculated the variance inflation factor using multiple linear regression analysis (Supplementary Table 1). The discrimination and calibration of the model were assessed by c-statistics (c = 0.701) and Hosmer-Lemeshow statistics (χ2 = 5.991, df = 8, P = 0.648), respectively. Additionally, we compared the standardized mean difference in both groups before and after propensity score matching. We also performed multiple logistic regression and multiple linear regression analyses to estimate the propensity scores and evaluate the variance inflation factors in each subgroup (Supplementary Tables 2–5).

Table 1

Baseline characteristics in study patients who received HFNC therapy

Baseline characteristics		Nonhypercapnic group (n = 660)	Hypercapnic group (n = 202)	Standardized mean difference	P-valuea
Age, yr		66.1 ± 13.9	65.6 ± 14.6	0.0310	0.700
Gender, men		450/660 (68.2)	145/202 (71.8)	0.0946	0.333
Body mass index, kg/m2		23.8 ± 4.4 (n = 652)	23.9 ± 4.9	0.0162	0.841
Smoker		402/660 (60.9)	128/201 (63.7)	0.0651	0.479
Underlying diseases
	Diabetes mellitus	256/660 (38.8)	76/202 (37.6)	0.0272	0.766
	Hypertension	373/660 (56.5)	105/202 (52.0)	0.1008	0.256
	Solid malignancies	182/660 (27.6)	84/202 (41.6)	0.3450	< 0.001b
	Hematologic malignancies	35/660 (5.3)	4/202 (2.0)	0.5621	0.047b
	Heart failure	105/660 (15.9)	30/202 (14.9)	0.0448	0.717
	Ischemic heart disease	86/660 (13.0)	23/202 (11.4)	0.0847	0.538
	Chronic kidney disease/dialysis	57/660 (8.6)	12/202 (5.9)	0.2223	0.217
	Liver cirrhosis	49/660 (7.4)	17/202 (8.4)	0.0751	0.643
	Use of immunosuppressive agents	146/660 (22.1)	49/202 (24.3)	0.0662	0.525
Underlying chronic lung diseases
	Asthma	19/660 (2.9)	10/202 (5.0)	0.3108	0.153
	Chronic obstructive pulmonary disease	135/660 (20.5)	55/202 (27.2)	0.2068	0.042b
	Pulmonary tuberculosis history	84/660 (12.7)	30/202 (14.9)	0.0987	0.435
	Bronchiectasis	78/660 (11.8)	48/202 (23.8)	0.4653	< 0.001b
	Lung cancer	47/660 (7.1)	43/202 (21.3)	0.6950	< 0.001b
	Interstitial lung disease	23/660 (3.5)	9/202 (4.5)	0.1410	0.523
ICU type, medical ICU		387/660 (58.6)	91/202 (45.0)	0.3019	0.001b
HFNC initial setting
	FiO2, %	54.9 ± 14.4	51.7 ± 12.8	0.2271	< 0.001b
	Flow, L/min	43.0 ± 8.4	41.9 ± 9.0	0.1291	0.110
PO2/FiO2 before HFNC application, mmHg		124.5 ± 39.2	135.7 ± 38.8	0.2857	< 0.001b
APACHE II score		20.9 ± 7.3	18.8 ± 7.9	0.2785	0.001b
SOFA score		8.2 ± 3.6	7.3 ± 3.5	0.2527	0.002b
Time differences between ABGA and HFNC application, min		42.0 (13.0–115)	23.5 (6.8–76.0)	0.0616	0.004b
Causes of HFNC application				0.4669	< 0.001b
	De novo acute respiratory failure	168/660 (25.5)	28/202 (13.9)
	Acute-on-chronic lung disease	143/660 (21.7)	32/202 (15.8)
	Septic shock for reasons other than respiratory infection	52/660 (7.9)	8/202 (4.0)
	Pulmonary edema	109/660 (16.5)	29/202 (14.4)
	After extubation	188/660 (28.5)	105/202 (52.0)
ABGA results before HFNC application
	pH	7.44 ± 0.07	7.35 ± 0.09	-	< 0.001b
	PCO2, mmHg	35.2 ± 5.6	52.1 ± 8.5	-	< 0.001b
	PO2, mmHg	67.1 ± 17.5	71.3 ± 18.7	-	0.003b
	HCO3, mEq/L	24.1 ± 5.6	29.0 ± 5.5	-	< 0.001b
	SpO2, %	91.5 ± 6.7	90.7 ± 7.8	-	0.182
	Lactate, mmol/L	1.6 (1.0–2.6) (n = 635)	1.5 (0.9–2.5) (n = 196)	-	0.115

Data are presented as mean ± standard deviation or median (interquartile range) or number (%).

HFNC = high-flow nasal cannula, ICU = intensive care unit, FiO2 = fraction of inspired oxygen, PO2 = partial pressure of carbon dioxide, APACHE = acute physiology and chronic health evaluation, SOFA = sequential organ failure assessment, ABGA = arterial blood gas analysis, PCO2 = partial pressure of carbon dioxide, HCO3 = bicarbonate, SpO2 = oxygen saturation.

aStatistical comparisons of the data were performed by using the χ2 test for categorical variables and the independent t-test for continuous variables; bStatistically significant P values.

To compare outcomes between nonhypercapnic and hypercapnic groups, we performed a logistic regression analysis (HFNC weaning, ICU mortality, and hospital mortality) and a Cox proportional regression analysis (90-day mortality). To achieve the propensity score-adjusted outcome, the individual propensity score was incorporated into each outcome model as a covariable. We also performed propensity score matching. After we had completed all of the propensity score matches, we compared the baseline covariables between the two intervention groups using the paired t-test for continuous variables and the McNemar test or marginal homogeneity test for categorical variables. For a logit link that accounted for the clustering of propensity score-matched pairs, the risks of clinical end points were analyzed using a generalized estimating equation logistic regression model (HFNC weaning, ICU mortality, and hospital mortality). Cox proportional regression analysis (90-day mortality) that stratified the matched pairs was also performed for each of the two matched samples.

To compare the vital signs, pH and PCO2 changes before and after HFNC application between nonhypercapnic and hypercapnic groups, we conducted generalized estimating equations using linear regression to account for the clustering nature of matched pairs. In addition, survival curves for two groups were analyzed using the Kaplan-Meier method and compared by log rank test. All analyses were performed using the Statistical Package for the Social Sciences (SPSS) Version 24.0 (IBM Corporation, Armonk, New York, USA), and differences with a P value < 0.05 was considered statistically significant.

Ethics statement

The present study was approved by the Institutional Review Board of Ulsan University Hospital (No. 2018-11-001), and the informed consent was waived because of the retrospective design of the study.

RESULTS

During the study period, there were 18,635 ICU admitted patients at Ulsan University Hospital. We excluded 16,868 patients who had not received HFNC therapy or were under 18 years of age. After the review of electronic medical records, we further excluded 905 patients who were not suitable for the evaluation of the effectiveness of HFNC therapy. Subsequently, the patients were classified into a nonhypercapnic group (n = 660) and a hypercapnic group (n = 202) based on their PCO2 values (Fig. 1).

Fig. 1

Distribution of the study patients according to the presence of hypercapnia before high-flow nasal cannula.

ICU = intensive care unit, HFNC = high-flow nasal cannula, PCO2 = partial pressure of carbon dioxide.

Patient characteristics

The baseline characteristics of the study population are presented in Table 1. There were more solid malignancies and CLDs (chronic obstructive pulmonary disease [COPD], bronchiectasis, and lung cancer) in the hypercapnic group than in the nonhypercapnic group. The nonhypercapnic group also had a significantly higher oxygen requirements (FiO2 and PO2/FiO2), and severity indexes (APACHE II and SOFA scores) than the hypercapnic group. The most common cause of HFNC application was “after extubation” in both groups. In the ABGA results, the hypercapnic group had lower pH and higher PCO2, PO2, and bicarbonate (HCO3) than the nonhypercapnic group.

Hospital outcomes

The hospital outcomes of the hypercapnic group were mostly better than those of the nonhypercapnic group (Table 2). The hypercapnic group was significantly more likely to have better HFNC weaning success, lower ICU mortality, lower hospital mortality and a shorter length of ICU stay than the nonhypercapnic group. During the ICU stay, there were also fewer complications, such as heart failure and cardiopulmonary resuscitation in the hypercapnic group.

Table 2

Hospital outcomes in study patients who received HFNC therapy

Hospital outcomes		Nonhypercapnic group (n = 660)	Hypercapnic group (n = 202)	P valuea
HFNC weaning success rate		389/660 (58.9)	145/202 (71.8)	0.001b
ICU mortality		130/660 (19.7)	21/202 (10.4)	0.002b
Hospital mortality		187/660 (28.3)	38/202 (18.8)	0.007b
Length of ICU stay		6.0 (3.0–12.0)	4.0 (1.0–9.0)	0.015b
Length of hospital stay		23.0 (14.0–39.0)	20.0 (11.8–37.0)	0.851
Complications during ICU stay
	Pneumonia	155/660 (23.5)	37/202 (18.3)	0.122
	Myocardial infarction	15/660 (2.3)	6/202 (3.0)	0.603
	Biliary infection	2/660 (0.3)	0/202 (0.0)	0.999
	Gastrointestinal infection	17/660 (2.6)	1/202 (0.5)	0.090
	Gastrointestinal bleeding (required endoscopy)	14/660 (2.1)	2/202 (1.0)	0.385
	Pulmonary thromboembolism	2/660 (0.3)	0/202 (0.0)	0.999
	Ischemic stroke	6/660 (0.9)	0/202 (0.0)	0.345
	Heart failure	31/660 (4.7)	3/202 (1.5)	0.040b
	Pneumothorax	18/660 (2.7)	2/202 (1.0)	0.188
	Urinary tract infection	8/660 (1.2)	3/202 (1.5)	0.726
	Liver failure	3/660 (0.5)	2/202 (1.0)	0.334
	Catheter-related bloodstream infections	19/660 (2.9)	6/202 (3.0)	0.946
	Acute kidney injury (required continuous renal replacement therapy)	119/660 (18.0)	26/202 (12.9)	0.086
	Cardiopulmonary resuscitation	48/660 (7.3)	6/202 (3.0)	0.027b

Data are presented as mean ± standard deviation or median (interquartile range) or number (%).

HFNC = high-flow nasal cannula, ICU = intensive care unit.

aStatistical comparisons of the data were performed by using the χ2 test for categorical variables and the independent t-test for continuous variables; bStatistically significant P values.

Propensity score-adjusted and score-matched outcomes

After propensity score matching, we selected 162 pairs of patients with similar baseline characteristics except pre-HFNC ABGA. From the ABGA results, the hypercapnic group had a lower pH and higher PCO2 and HCO3 than the nonhypercapnic group (Table 3). Unlike the results before the propensity score matching, there were no significant differences between both groups in terms of hospital outcomes and complications during ICU stay (Table 4). When we performed a propensity score-adjusted and score-matched analysis, no statistically significant differences were found in HFNC weaning success (adjusted P = 0.623, matched P = 0.593), ICU mortality (adjusted P = 0.463, matched P = 0.195), hospital mortality (adjusted P = 0.602, matched P = 0.579), and 90-day mortality (adjusted P = 0.758, matched P = 0.808) (Table 5).

Table 3

Baseline characteristics in study patients who received HFNC therapy after propensity matching

Baseline characteristics		Nonhypercapnic group (n = 162)	Hypercapnic group (n = 162)	Standardized mean difference	P valuea
Age, yr		64.7 ± 14.3	64.9 ± 15.4	0.0116	0.906
Gender, men		116/162 (71.6)	114/162 (70.4)	0.0331	0.907
Body mass index, kg/m2		24.2 ± 4.7	24.3 ± 5.1	0.0091	0.930
Smoker		96/162 (59.3)	101/162 (62.3)	0.0714	0.657
Underlying diseases
	Diabetes mellitus	65/162 (40.1)	63/162 (38.9)	0.0285	0.909
	Hypertension	83/162 (51.2)	91/162 (56.2)	0.1096	0.440
	Solid malignancies	56/162 (34.6)	52/162 (32.1)	0.0613	0.703
	Hematologic malignancies	5/162 (3.1)	4/162 (2.5)	0.1265	0.999
	Heart failure	27/162 (16.7)	28/162 (17.3)	0.0241	0.999
	Ischemic heart disease	18/162 (11.1)	20/162 (12.3)	0.0658	0.868
	Chronic kidney disease/dialysis	9/162 (5.6)	10/162 (6.2)	0.0617	0.999
	Liver cirrhosis	16/162 (9.9)	15/162 (9.3)	0.0393	0.999
	Use of immunosuppressive agents	38/162 (23.5)	38/162 (23.5)	0.0000	0.999
Underlying chronic lung diseases
	Asthma	6/162 (3.7)	7/162 (4.3)	0.0885	0.999
	Chronic obstructive pulmonary disease	31/162 (19.1)	36/162 (22.2)	0.1039	0.590
	Pulmonary tuberculosis history	20/162 (12.3)	20/162 (12.3)	0.0000	0.999
	Bronchiectasis	22/162 (13.6)	27/162 (16.7)	0.1330	0.511
	Lung cancer	15/162 (9.3)	14/162 (8.6)	0.0418	0.999
	Interstitial lung disease	5/162 (3.1)	4/162 (2.5)	0.1265	0.999
ICU type, medical ICU		87/162 (53.7)	83/162 (51.2)	0.0546	0.720
HFNC initial setting
	FiO2, %	52.7 ± 14.0	53.0 ± 13.6	0.0170	0.870
	Flow, L/min	42.5 ± 8.0	42.0 ± 9.6	0.0615	0.589
PO2/FiO2 before HFNC application, mmHg		127.1 ± 38.9	129.7 ± 38.1	0.0684	0.509
APACHE II score		19.4 ± 6.9	19.6 ± 8.2	0.0211	0.841
SOFA score		7.4 ± 3.1	7.7 ± 3.5	0.1085	0.326
Time differences between ABGA and HFNC application, min		34.5 (12.0–104)	33.5 (9.8–104.3)	0.0218	0.841
Causes of HFNC application				0.0467	0.583b
	De novo acute respiratory failure	32/162 (19.8)	27/162 (16.7)
	Acute-on-chronic lung disease	28/162 (17.3)	30/162 (18.5)
	Septic shock for reasons other than respiratory infection	5/162 (3.1)	8/162 (4.9)
	Pulmonary edema	33/162 (20.4)	29/162 (17.9)
	After extubation	64/162 (39.5)	68/162 (42.0)
ABGA results before HFNC application
	pH	7.45 ± 0.07	7.36 ± 0.09	-	< 0.001c
	PCO2, mmHg	36.1 ± 5.7	51.5 ± 8.5	-	< 0.001c
	PO2, mmHg	66.4 ± 16.3	68.7 ± 17.8	-	0.218
	HCO3, mEq/L	25.6 ± 7.3	29.0 ± 5.7	-	< 0.001c
	SpO2, %	91.0 ± 9.5	90.0 ± 7.9	-	0.316
	Lactate, mmol/L	1.4 (0.9, 2.4)	1.4 (0.9, 2.4)	-	0.343

Data are presented as mean ± standard deviation or median (interquartile range) or number (%).

HFNC = high-flow nasal cannula, ICU = intensive care unit, FiO2 = fraction of inspired oxygen, PO2 = partial pressure of carbon dioxide, APACHE = acute physiology and chronic health evaluation, SOFA = sequential organ failure assessment, ABGA = arterial blood gas analysis, PCO2 = partial pressure of carbon dioxide, HCO3 = bicarbonate, SpO2 = oxygen saturation.

aStatistical comparisons of the data were performed by using the McNemar test for categorical variables and the paired t-test for continuous variables; bStatistical comparison of the data was performed by using Marginal homogeneity test; cStatistically significant P values.

Table 4

Hospital outcomes in study patients who received HFNC therapy after propensity matching

Hospital outcomes		Nonhypercapnic group (n = 162)	Hypercapnic group (n = 162)	P valuea
HFNC weaning success rate		107/162 (66.0)	111/162 (68.5)	0.598
ICU mortality		28/162 (17.3)	20/162 (12.3)	0.688
Hospital mortality		38/162 (23.5)	34/162 (21)	0.256
Length of ICU stay		6.0 (3.0–12.0)	5.0 (2.0–10.0)	0.677
Length of hospital stay		23.0 (13.0–40.5)	21.0 (12.0–38.3)	0.181
Complications during ICU stay
	Pneumonia	31/162 (19.1)	32/162 (19.8)	0.999
	Myocardial infarction	4/162 (2.5)	5/162 (3.1)	0.999
	Biliary infection	0/162 (0.0)	0/162 (0.0)	-
	Gastrointestinal infection	4/162 (2.5)	1/162 (0.6)	0.375
	Gastrointestinal bleeding (required endoscopy)	2/162 (1.2)	2/162 (1.2)	0.999
	Pulmonary thromboembolism	1/162 (0.6)	0/162 (0.0)	0.999
	Ischemic stroke	1/162 (0.6)	0/162 (0.0)	0.999
	Heart failure	6/162 (3.7)	3/162 (1.9)	0.508
	Pneumothorax	6/162 (3.7)	2/162 (1.2)	0.289
	Urinary tract infection	0/162 (0.0)	2/162 (1.2)	0.500
	Liver failure	0/162 (0.0)	2/162 (1.2)	0.500
	Catheter-related bloodstream infections	6/162 (3.7)	5/162 (3.1)	0.999
	Acute kidney injury (required continuous renal replacement therapy)	23/162 (14.2)	24/162 (14.8)	0.999
	Cardiopulmonary resuscitation	12/162 (7.4)	6/162 (3.7)	0.210

Data are presented as mean ± standard deviation or median (interquartile range) or number (%).

HFNC = high-flow nasal cannula, ICU = intensive care unit.

aStatistical comparisons of the data were performed by using the McNemar test for categorical variables and the paired t-test for continuous variables.

Table 5

Analysis of hospital outcomes in the hypercapnic group compared with the nonhypercapnic group (as reference)

Variables		Crude		Propensity-adjusteda		Propensity-matched
		Odds ratio (95% CI)	P valueb	Odds ratio (95% CI)	P valueb	Odds ratio (95% CI)	P valueb
All patientsc
	HFNC weaning success	1.772 (1.257–2.498)	0.001j	1.099 (0.755–1.601)	0.623	1.119 (0.741–1.688)	0.593
	ICU mortality	0.473 (0.290–0.773)	0.003j	0.823 (0.490–1.384)	0.463	0.674 (0.371–1.224)	0.195
	Hospital mortality	0.586 (0.396–0.867)	0.007j	0.895 (0.590–1.357)	0.602	0.867 (0.523–1.436)	0.579
	90-day mortality	0.645 (0.465–0.897)d	0.009e,j	0.948 (0.675–1.332)d	0.758e	0.943 (0.586–1.517)d	0.808e
Patients without chronic lung diseasef
	HFNC weaning success	1.342 (0.836–2.154)	0.224	1.004 (0.607–1.662)	0.987	0.813 (0.428–1.545)	0.527
	ICU mortality	0.550 (0.264–1.147)	0.111	1.001 (0.459–2.186)	0.997	0.799 (0.332–1.926)	0.617
	Hospital mortality	0.553 (0.295–1.037)	0.065	0.863 (0.446–1.670)	0.662	0.917 (0.407–2.066)	0.835
	90-day mortality	0.572 (0.327–1.000)d	0.050e	0.832 (0.469–1.473)d	0.527e	0.857 (0.396–1.853)d	0.695e
Patients with chronic lung diseaseg
	HFNC weaning success	2.591 (1.562–4.297)	< 0.001j	1.364 (0.769–2.420)	0.289	1.184 (0.605–2.318)	0.622
	ICU mortality	0.353 (0.181–0.687)	0.002j	0.673 (0.322–1.405)	0.292	0.657 (0.279–1.549)	0.337
	Hospital mortality	0.474 (0.282–0.796)	0.005j	0.829 (0.461–1.489)	0.530	0.941 (0.475–1.864)	0.862
	90-day mortality	0.570 (0.378–0.861)d	0.008e,j	0.936 (0.593–1.476)d	0.776e	0.909 (0.496–1.666)d	0.758e
Patients with HFNC application after extubationh
	HFNC weaning success	1.593 (0.817–3.105)	0.172	1.175 (0.567–2.439)	0.664	1.218 (0.536–2.768)	0.637
	ICU mortality	0.155 (0.020–1.216)	0.076	0.311 (0.037–2.642)	0.285	0.156 (0.018–1.360)	0.093
	Hospital mortality	0.398 (0.145–1.088)	0.072	0.452 (0.154–1.328)	0.149	0.329 (0.095–1.139)	0.079
	90-day mortality	0.597 (0.267–1.334)d	0.209e	0.631 (0.263–1.513)d	0.302e	0.455 (0.158–1.308)d	0.144e
Patients with HFNC application other causes than extubationi
	HFNC weaning success	1.235 (0.796–1.916)	0.347	1.137 (0.719–1.799)	0.583	1.044 (0.573–1.900)	0.889
	ICU mortality	0.770 (0.452–1.314)	0.339	0.822 (0.472–1.431)	0.488	0.881 (0.410–1.895)	0.746
	Hospital mortality	0.950 (0.600–1.507)	0.829	0.982 (0.608–1.586)	0.940	1.279 (0.679–2.408)	0.446
	90-day mortality	0.945 (0.659–1.357)d	0.761e	0.976 (0.672–1.416)d	0.898e	1.261 (0.729–2.179)d	0.406e

CI = confidence interval, HFNC = high-flow nasal cannula, ICU = intensive care unit.

aThe individual propensity score was integrated into each outcome model as a co-variable; bStatistical comparisons of the data were performed using logistic regression analysis; cOf the 862 patients, 162 pairs were matched; dHazard ratios analyzed by Cox proportional regression model; eStatistical comparisons of the data were performed using Cox proportional regression analysis; fOf the 518 patients, 95 pairs were matched; gOf the 344 patients, 84 pairs were matched; hOf the 293 patients, 78 pairs were matched; iOf the 569 patients, 95 pairs were matched; jStatistically significant P values.

Fig. 2 presented the short-term effectiveness of HFNC therapy by using HFNC application immediately before and after vital signs, pH and PCO2 values in before and after propensity score matching cohorts. There were no interactions associated with vital sign changes according to HFNC therapy between nonhypercapnic and hypercapnic groups in after propensity score matching cohort. However, there were interactions associated with pH (before propensity score matching P < 0.001, after propensity score matching P < 0.001) and PCO2 (before propensity score matching P < 0.001, after propensity score matching P = 0.001) values according to HFNC therapy between the two groups. Fig. 3 shows the survival curve of the matched groups stratified by the presence of hypercapnia, and there was no significant difference for the 90-day survival between the two groups (P = 0.468).

Fig. 2

Changes of vital signs and arterial blood gas analysis results between nonhypercapnic and hypercapnic groups after initiation of high-flow nasal cannula therapy. In before propensity score matching cohort: (A) SBP change after HFNC application, (B) DBP change after HFNC application, (C) PR change after HFNC application, (D) RR change after HFNC application, (E) pH change after HFNC application, (F) PCO2 change after HFNC application, And in after propensity score matching cohort: (G) SBP change after HFNC application, (H) DBP change after HFNC application, (I) PR change after HFNC application, (J) RR change after HFNC application, (K) pH change after HFNC application, (L) PCO2 change after HFNC application.

HFNC = high-flow nasal cannula, SBP = systolic blood pressure, SE = standard error, IQR = interquartile range, DBP = diastolic blood pressure, PR = pulse rate, RR = respiration rate, PCO2 = partial pressure of carbon dioxide.

Fig. 3

Kaplan-Meier survival curves stratified by the presence of hypercapnia before high-flow nasal cannula application.

Subgroup analysis

We performed subgroup analysis according to the presence of CLD and causes of HFNC application. Although several baseline characteristics of each subgroup were different between the nonhypercapnic and hypercapnic groups, the hypercapnic group had similar or better HFNC weaning success, similar or lower mortality, and similar or fewer complications in all subgroups (Supplementary Tables 6–9). After propensity score matching for each subgroup, there was no significant difference in HFNC weaning success and mortality between the two groups (Supplementary Tables 10–13). When the propensity score-adjusted and score-matched analysis were performed in each subgroup, both groups also did not differ significantly in terms of HFNC weaning success, ICU mortality, hospital mortality and 90-day mortality (Table 5).

DISCUSSION

Our current study findings indicated that the effectiveness of HFNC therapy in AHRF patients with hypercapnia was not inferior to the effectiveness of the therapy in those without hypercapnia under various conditions. In other words, AHRF patients could be treated with HFNC therapy regardless of the presence of hypercapnia before HFNC application. This finding was in accordance with the results from previous studies that reported the effectiveness of HFNC therapy in patients with acute RF,141519 acute exacerbation of COPD,2021 and after extubation422 when accompanied with hypercapnia. However, these studies involved small sample sizes, and there has been no prospective randomized controlled study for the evaluation of HFNC weaning success and mortality. However, the current study was well constructed and confirmed these results using a cohort of relatively large sample sizes.

The choice of the appropriate form of oxygen device in AHRF patients with hypercapnia is unclear. Conventional oxygen devices have limited efficacy in RF because of the significant decrease in the delivered FiO2. Invasive mechanical ventilation is a lifesaving technique, but it is also associated with various complications. NIV is recommended for use only in patients with hypercapnia who have an acidic pH but not in those who do not have an acidic pH.9 Additionally, NIV is not recommended in patients with altered mental states, unstable hemodynamic conditions, or an inability to protect the airway.23 Considering these reasons and our study results, HFNC therapy might overcome the limitations of the other devices and be a good alternative for patients in AHRF with hypercapnia.

Derangements in gas exchange may be developed in patients with CLD and cause hypercapnia.24 These results were consistent with the fact that the hypercapnia group had a greater number of underlying CLDs in our study. The hospital outcomes and complications were worse in the nonhypercapnic group than in the hypercapnic group before propensity matching. This was presumed to be because of the severity of the nonhypercapnic group being higher than that of the hypercapnic group in most of the baseline characteristics except for underlying CLD. There were more solid malignancies in the hypercapnic group. However, considering the number of lung cancer patients, nonlung cancerous solid malignancies were similar between the nonhypercapnic and hypercapnic groups (20.5% vs. 20.3%). Although these differences between both groups were compensated for after propensity matching, the outcomes of the hypercapnic group were not significantly different from those of the nonhypercapnic group in our study. The subgroup results were also similar to these results. Our findings indicate that HFNC therapy might be beneficial for AHRF patients with hypercapnia in various conditions.

Although HFNC therapy is safe and useful in different clinical situations, including RF,25 patients with hypercapnia were mostly excluded in the previous large-scale HFNC studies.226272829 Therefore, the efficacy of HFNC in patients with hypercapnia has not been sufficiently evaluated. One study has even asserted that hypercapnia has potentially deleterious effects in patients with CLD.24 Our study findings showed the effectiveness of HFNC therapy in AHRF patients with hypercapnia using AHRF patients without hypercapnia as a control group under various conditions. However, one thing to note is that the effectiveness of HFNC was evaluated in AHRF patients with hypercapnia only and not in patients with hypercapnic RF. Although it was reported that HFNC therapy was used to successfully manage hypercapnic RF,30 special attention in the treatment of hypercapnic RF using HFNC therapy and a large prospective randomized controlled study are required. We hope the findings from our study contribute to establishing an appropriate indication for HFNC therapy.

Our study had several limitations. First, it was a retrospective observational study. However, we used propensity score matching and evaluated the effectiveness of HFNC treatment in AHRF patients with or without hypercapnia in a group of carefully selected patients to identify a possible link between hypercapnia and hospital outcomes. Second, our study was conducted at a single center. Therefore, selection bias cannot be excluded, and the results must be interpreted with caution. Large-scale multicenter and randomized controlled studies are required to obtain more accurate and reliable results. Third, there are possible differences in the actual FiO2 supplied to the patients and the FiO2 setting of the HFNC device. Data, including FiO2, should be carefully interpreted because the delivered FiO2 cannot be measured.

Because hypercapnia might be associated with poor prognosis in RF, HFNC therapy was used very cautiously under this condition. Our study findings showed that the effectiveness of HFNC therapy in AHRF patients with hypercapnia was not significantly different from those without hypercapnia in terms of HFNC weaning success and mortality. HFNC therapy might be effective in AHRF patients with hypercapnia. We believe that our study can contribute to the establishment of appropriate indications for HFNC therapy. Further large-scale, prospective, and randomized controlled trials are warranted.