Sleep Hypoventilation is Common in Diurnal Normocapnic COPD Patients with Severe or Very Severe Obstruction and is Not Associated with Body Mass Index

Abstract

Sleep hypoventilation (SH) is common in COPD patients with diurnal hypercapnia, however there are little data on the presence of SH in COPD patients with diurnal normocapnia. In this study the prevalence of SH in stable normocapnic COPD patients with severe or very severe obstruction (GOLD stages III and IV) was evaluated across body mass index (BMI) classes and associations between SH and body composition measures were explored. A total of 56 diurnal normocapnic COPD patients, of whom 17 normal-weight (COPD_NW), 18 overweight (COPD_OW) and 21 obese (COPD_OB), underwent polysomnography to objectify SH and bioelectrical impedance analysis to assess body composition. The overall prevalence of SH was 66.1% and was not different across BMI classes. Logistic regression models indicated that SH was not associated with waist-to-hip ratio, body fat percentage and fat-free mass index. Our data indicate that SH is common in diurnal normocapnic COPD patients with severe or very severe obstruction and is not associated with BMI or body composition.

Keywords:

• COPD
• sleep hypoventilation
• obesity

Introduction

During sleep, COPD patients may have impairments in gas exchange (hypoxemia with or without hypercapnia). Nocturnal gas exchange impairments in COPD have been associated with adverse effects, such as arrhythmias during sleep, pulmonary hypertension, exacerbation frequency and mortality [1,2]. Nocturnal hypoxemia in COPD patients is well investigated [3,4], but data on sleep hypoventilation (SH) is limited. Even though noninvasive ventilation for (daytime) chronic hypercapnic respiratory failure in COPD usually takes effect by alleviating hypercapnia during sleep [5], there is no clarity on the prevalence of SH in patients with COPD. This applies particularly to patients with hypoventilation only during sleep while displaying normocapnia during wakefulness [6]. Identification of the latter seems important, as SH is believed to precede daytime hypercapnia and chronic hypercapnic respiratory failure [7,8].

Sleep hypoventilation has been defined by the American Association of Sleep Medicine (AASM) as an increase in PaCO₂ (or surrogate) to a value >55 mmHg (7.3 kPa) for ≥10 min, or an increase in PaCO₂ ≥10 mmHg (1.3 kPa) above the awake supine value to a value exceeding 50 mmHg (6.7 kPa) for ≥10 min [9]. Whereas hypoxemia in COPD may be a result of pulmonary failure only, (sleep) hypoventilation results from failure of the respiratory pump and thus ventilatory failure [10]. Sleep itself leads to a reduction of alveolar ventilation: the loss of the wakefulness drive, increased airway resistance and diminished chemosensitivity during non-Rapid Eye Movement (NREM) sleep result in a mild increase of PaCO₂ and decrease of PaO₂ [11]. During REM sleep these changes are further enhanced by rapid shallow breathing and reduced activity of accessory respiratory muscles [12]. In contrast to healthy subjects, in whom these changes are mild, several factors may contribute to increased work of breathing (and subsequently ventilatory failure) in patients with COPD, including increased upper airway resistance and mechanical disadvantages imposed by hyperinflation, as well as respiratory muscle dysfunction and ventilation-perfusion mismatch [7,13].

Obesity, which on its own may lead to alveolar hypoventilation, may add to the respiratory load and further aggravate alveolar hypoventilation in patients with COPD. In fact, studies in COPD patients with diurnal hypercapnia indicate that SH is common and correlates with higher BMI [14,15].

Studies on the prevalence of SH in COPD patients with diurnal normocapnia are more scarce. Nevertheless, some data are available suggesting the presence of SH in COPD patients with diurnal normocapnia [16,17]. Data from COPD and obstructive sleep apnea (OSA) studies indicate a higher BMI as one of the risk factors of developing SH, but this was not found in a study including normocapnic COPD patients [17]. However, the median BMI in the latter study was 25 kg/m² with only 13% of the study population presenting with a BMI of ≥30 kg/m². It is thus unclear, whether obesity is a risk factor for SH in patients with diurnal normocapnic COPD. Also, it is unknown whether body composition, independent of BMI, is associated with SH in these patients.

The primary objective in this study was to evaluate the association between BMI and prevalence of SH in stable normocapnic COPD patients with severe or very severe obstruction (GOLD stages III and IV). Our secondary objective was to explore associations between SH and body composition measures. We hypothesized that higher BMI and higher measures of adiposity were associated with SH.

Methods

Study design

This was an observational study performed at Rijnstate Hospital, Arnhem, The Netherlands. Patients were recruited between February and December 2016. The study was approved by the regional ethics committee (CMO Arnhem-Nijmegen) and Rijnstate Hospital local ethics committee (2015-1750). COPD patients who met the inclusion criteria were invited to participate by their pulmonologist. All subjects signed informed consent and spent one night at the hospital for polysomnography (PSG).

Subjects

We included normal-weight (COPD_NW), overweight (COPD_OW) and obese (COPD_OB) patients with COPD. All patients had a diagnosis of severe to very severe COPD, defined as an obstructive lung function with forced expiratory volume in 1 s/forced vital capacity (FEV₁/FVC) < lower limit of normal and FEV₁ < 50% predicted (COPD stages GOLD III and IV) according to the GOLD report [18]. Normal weight was defined as BMI 18.5-24.99 kg/m², overweight as BMI 25.00-29.99 kg/m², and obese as BMI ≥ 30 kg/m². All patients were clinically stable, had no exacerbation history in the preceding 4 weeks, and included men and women between 18-80 years of age.

Exclusion criteria were 1) presence of significant comorbidity potentially interfering with outcomes of interest (i.e. cardiovascular, neuromuscular or other respiratory diseases); 2) inability to perform pulmonary function tests or fill in questionnaires; 3) previous diagnosis (or diagnosed during study PSG) of moderate to severe obstructive sleep apnea (apnea hypopnea index (AHI)≥15 events/hour) and/or active positive airway pressure treatment; 4) daytime hypercapnia, defined by awake capillary pCO₂ >48 mmHg (6,4 kPa) during a stable phase of disease.

Measurements

Information on demographics, medication use, smoking habits, known comorbidities, and anthropometric measurements (weight, height, waist and hip circumference) were obtained. Body composition was measured with bioelectrical impedance analysis (Bodystat 1500, Bodystat, UK) yielding body fat percentage and fat-free mass (FFM) from which fat-free mass index (FFMI) was calculated as FFM/height². Spirometry and body plethysmography data were either collected from medical records if assessed <1 year prior to inclusion, or were assessed as part of study measurements according to international standards [19]. Daytime capillary blood gasses were taken in all subjects. PSG set-up and PSG analysis was performed according to the 2012 AASM specifications (Smart-PSG, Cidelec, France) [9]. Sleep scoring was performed manually by an experienced sleep technologist.

Transcutaneous pCO₂ (P_tcCO₂, Tosca, Radiometer, Australia) was attached to the earlobe between 9:00-9:30 p.m. Tosca was calibrated with CO₂-gas at the beginning and at the end of the night to correct for overnight pCO₂-drift. At 10:00 p.m. PSG measurements were started. All PSG data (total sleep time, sleep efficiency, duration of REM and NREM sleep and sleep apnea/hypopneas), O₂- saturation and P_tcCO₂ were stored and analyzed using Cidelec software. SH was defined according to AASM criteria: an increase in PaCO₂ (or surrogate) to a value >55 mm Hg for ≥10 min, or an increase in PaCO₂ ≥10 mmHg above the awake supine value to a value exceeding 50 mmHg for ≥10 min [9]. The severity of SH was assessed by comparing the highest P_tcCO₂ during sleep, mean P_tcCO₂ during sleep, change in P_tcCO₂ between awake to the level of mean P_tcCO₂ during sleep (Δ P_tcCO₂ Mean) and change in P_tcCO₂ between awake to the highest P_tcCO₂ during sleep (Δ P_tcCO₂ Max) in the different weight groups. Sleep quality was assessed using the Pittsburgh Sleep Quality Index (PSQI) [20].

Statistics

Descriptive statistics were used to characterize the study population. Continuous variables were expressed as mean ± SD while discrete variables were shown as percentages. A two-tailed p value < 0.05 was considered statistically significant. For between-group comparisons, the independent t-test (for 2 groups) or the ANOVA (for >2 groups) were utilized. Post hoc analyses were performed with Bonferroni or Games-Howell (depending on homogeneity of variance according to Levene’s test) to correct for multiple comparisons. Proportions of categorical variables were compared by Chi-squared test (two-tailed). Binary logistic regression models were used to assess the associations between SH and its potential determinants (pulmonary function, anthropometric measures etc.). Analyses were performed with SPSS version 21.0 (IBM, USA).

Results

A total of 75 eligible patients completed PSG. Five patients were excluded from further analyses because of newly diagnosed OSA and another 14 because of technical failures (8 failures for transcutaneous PCO₂-measurement, 6 PSG failures). Finally, 17 COPD_NW, 18 COPD_OW and 21 COPD_OB were included in the analyses.

The characteristics of the study participants are presented in Table 1. There were no statistically significant differences in age, sex, pulmonary medication use or comorbidities between BMI classes. There were significantly less current smokers in COPD_OB compared to the COPD_OW and COPD_NW while smoked pack years were comparable. As expected, weight, hip and waist circumference, body fat percentage and fat-free mass index gradually increased with increasing BMI class. Pulmonary function parameters are presented in Table 2. COPD_OB had significantly higher FEV₁ values compared to COPD_NW. FRC and residual volume (RV) decreased and inspiratory capacity (IC) increased with increasing BMI class. Diffusion capacity of the lungs for carbon monoxide (DLCO) was comparable between BMI classes. Mean diurnal capillary pCO₂ was not significantly different between BMI classes, and averaged at 41.0 mmHg for the group as a whole.

Table 1. Characteristics of the study participants. Data are presented as mean (SD) unless otherwise stated.

	NW	OW	OB	Total group	P between groups	P NW vs OW	P NW vs OB	P OW vs OB
N	17	18	21	56	–	–	–	–
Age, mean	63.1 (9.6)	68.4 (5.5)	67.0 (8.0)	66.3 (8.0)	0.13	–	–	–
Sex, % males	47	61	43	50	0.50	–	–	–
Weight, kg	64.6 (7.6)	79.1 (9.0)	90.6 (10.9)	79.0 (14.2)	<0.01	<0.01	<0.01	<0.01
Hight, cm	170 (8)	168 (7)	167 (10)	168 (9)	0.45	–	–	–
BMI	22.3 (1.7)	28.0 (1.3)	32.5 (1.2)	28.0 (4.5)	<0.01	<0.01	<0.01	<0.01
Hip circumference, cm	92 (5)	100 (7)	108 (11)	101 (11)	<0.01	<0.01	<0.01	0.02
Waist circumference, cm	90 (6)	105 (9)	114 (9)	104 (13)	<0.01	<0.01	<0.01	<0.01
Waist/hip ratio	0.98 (0.08)	1.06 (0.12)	1.07 (0.13)	1.04 (0.11)	0.05	–	–	–
Fat, %	31 (8)	36 (7)	42 (7)	37 (9)	<0.01	0.12	<0.01	0.10
FFMI, kg/m2	15.3 (2.3)	17.8 (2.4)	19.0 (2.3)	17.5	<0.01	<0.01	<0.01	0.40
Smoked pack years	37 (23)	41 (30)	33 (16)	37 (23)	0.57	–	–	–
Smoking status
% current smokers	41	28	5	23	0.02	0.40	0.01	0.08
Use of inhalers
% using SABA	71	61	61	64	0.80	–	–	–
% using SAMA	6	17	10	11	0.58	–	–	–
% using LABA	100	94	95	96	0.47	–	–	–
% using LAMA	100	83	86	89	0.10	–	–	–
% using ICS	82	78	76	79	0.89	–	–	–
Comorbidities
% Diabetes	0	22	14	13	0.05	–	–	–
% Hypertension	18	50	48	39	0.09	–	–	–
% Atrial fibrillation	12	6	5	7	0.69	–	–	–
% Myocardial infarction	0	17	10	9	0.12	–	–	–
% Congestive heart failure	0	0	0	0	–	–	–	–

Abbreviations: BMI, body mass index; FFMI, free fat mass index; SABA, short-acting beta agonist; SAMA, short-acting muscarinic antagonist; LABA, long-acting beta agonist; LAMA, long-acting muscarinic antagonist; ICS, inhaled corticosteroid.

Table 2. Pulmonary function tests. Data are presented as mean (SD) unless otherwise stated.

	NW	OW	OB	Total group	P between groups	P NW vs OW	P NW vs OB	OW vs OB
FEV1, % pred	32.2 (8.5)	38.1 (6.6)	43.9 (9.5)	38.5 (9.5)	<0.01	0.13	<0.01	0.10
FEV1, L	0.90 (0.35)	0.96 (0.20)	1.08 (0.35)	0.99 (0.31)	0.21	–	–	–
VC, % pred	88.5 (18.6)	93.4 (15.2)	94.6 (17.8)	92.4 (17.1)	0.53	–	–	–
VC, L	3.15 (0.99)	3.14 (0.84)	3.02 (0.92)	3.10 (0.90)	0.89	–	–	–
IC, % pred	81.9 (18.9)	88.9 (13.1)	98.0 (19.8)	90.2 (18.6)	0.03	0.74	0.02	0.35
IC, L	2.14 (0.67)	2.19 (0.52)	2.30 (0.73)	2.22 (0.64)	0.72	–	–	–
FRC, % pred	191.6 (29.1)	162.9 (28.7)	155.3 (21.3)	168.3 (30.0)	<0.01	<0.01	<0.01	1.00
FRC, L	6.05 (1.16)	5.16 (1.08)	4.74 (1.04)	5.26 (1.20)	<0.01	0.06	<0.01	0.70
ERV, % pred	107.7 (33.4)	106.7 (39.9)	87.1 (39.6)	99.7 (38.5)	0.17	–	–	–
ERV, L	1.02 (0.42)	0.94 (0.44)	0.72 (0.35)	0.88 (0.41)	0.06	–	–	–
RV, % pred	227.9 (42.6)	203.2 (45.3)	185.3 (31.0)	203.1 (42.4)	0.01	0.24	<0.01	0.49
RV, L	5.11 (1.03)	4.61 (0.90)	4.04 (0.67)	4.53 (0.95)	<0.01	0.30	<0.01	0.13
TLC, % pred	130.9 (15.5)	121.6 (17.9)	120.3 (13.8)	123.8 (16.1)	0.11	–	–	–
TLC, L	7.82 (1.38)	7.09 (1.29)	6.77 (1.41)	7.18 (1.41)	0.07	–	–	–
RV/TLC, %	65.9 (11.7)	65.6 (10.0)	60.7 (7.9)	63.8 (9.9)	0.19	–	–	–
DLCO SB, % pred	46.5 (12.1)	53.7 (15.2)	55.7 (11.4)	52.3 (13.3)	0.10	–	–	–
DLCO/VA, % pred	56.6 (14.7)	64.4 (18.3)	71.6 (20.9)	64.8 (19.1)	0.06	–	–	–
PcCO2, mmHg	40.3 (3.2)	41.4 (3.9)	41.3 (4.5)	41.0 (3.9)	0.67	–	–	–

Abbreviations: FEV₁, forced expiratory volume in 1 s; VC, vital capacity; TLC, total lung capacity; FRC, functional residual capacity; ERV, expiratory reserve volume; RV, residual volume; DLCO, diffusing capacity for carbon monoxide; VA, alveolar volume; PcCO₂, capillary pCO₂.

PSG and nocturnal transcutaneous CO₂ data are shown in Table 3. Overall SH prevalence was 66.1%. SH prevalence was similar across BMI classes. There was no significant difference between the mean BMI in the group with SH (28.1 ± 4.5 kg/m²) compared to the group without SH (27.7 ± 4.4 kg/m²; p = 0.75). In patients with SH, the highest P_tcCO₂ occurred in almost all (36 of 37 patients) during REM-sleep, while the remaining patient did not present REM-sleep during the sleep study. All the measures which assessed the severity of SH (highest P_tcCO₂ during sleep, mean P_tcCO₂ during sleep, Δ P_tcCO₂ MEAN and Δ P_tcCO₂ Max) were slightly higher in the overweight and obese group compared to the normal-weight, but these differences were statistically not significant. Furthermore, explorative logistic regression models performed in the whole study population, showed no statistically significant associations between SH with waist circumference, hip circumference, waist-to-hip ratio, body fat percentage, FFMI, pulmonary function parameters, smoking status, class of inhalers, mean and lowest SpO₂ during sleep, capillary pCO₂ during wakefulness or AHI (data not shown).

Table 3. Polysomnography and sleep hypoventilation. Data are presented as mean (SD) unless otherwise stated.

	NW	OW	OB	Total group	P between groups
TST, min	369 (87)	380 (56)	391 (62)	381 (68)	0.63
Sleep efficiency, % TST/PST	64.5 (13.5)	66.3 (8.3)	69.3 (8.7)	66.9 (10.3)	0.36
% REM of TST	15.3 (7.6)	16.9 (6.4)	17.2 (7.6)	16.6 (7.1)	0.69
AI, events/hour, mean	0.65 (1.6)	0.67 (1.3)	0.81 (1.1)	0.71 (1.3)	0.92
AHI, events/hour, mean	10.2 (5.4)	9.9 (5.4)	11.1 (6.6)	10.4 (5.8)	0.81
MeanSpO2 during sleep	90.3 (2.1)	89.6 (2.2)	88.8 (3.1)	89.5 (2.6)	0.21
Lowest SpO2 during sleep	81.5 (5.2)	79.6 (7.1)	81.1 (5.5)	80.8 (5.9)	0.58
Mean sleep PtcCO2, mmHg	53.4 (4.9)	54.6 (5.8)	54.9 (6.4)	54.3 (5.7)	0.70
Highest PtcCO2, mmHg	58.5 (6.4)	60.3 (6.0)	60.5 (7.7)	59.8 (6.8)	0.62
Highest PtcCO2 REM, mmHg	59.9 (6.5)	60.1 (6.3)	60.5 (7.8)	60.2 (6.8)	0.96
Δ PtcCO2 Mean, mmHG	5.4 (2.9)	6.8 (4.0)	6.5 (3.7)	6.3 (3.6)	0.49
Δ PtcCO2 Max, mmHG	10.5 (4.3)	12.5 (4.2)	12.1 (4.8)	11.8 (4.5)	0.40
Sleep hypoventilation, %	64.7	66.7	66.7	66.1	0.99

Abbreviations: TST, total sleep time; REM, rapid eye movement sleep; AI, arousal index; AHI, apnea/hypopnea index; SpO2, oxygen saturation; P_tcCO₂, transcutaneous pCO₂; Δ P_tcCO₂ Mean, delta P_tcCO₂ between awake and mean P_tcCO₂ during sleep; Δ P_tcCO₂ Max, delta P_tcCO2 between awake and highest P_tcCO₂ during sleep.

In the group as a whole, PSQI data indicated that overall sleep quality was poor (global score >5). There were no differences between BMI classes in the total score and in subdomains of the questionnaire in the unadjusted analysis. Also in further analysis, considering adjustments for potential cofounders (age, sex, smoking status, packyears, comorbidities, FEV₁% predicted, FRC% predicted and DLCO), by using one-way analysis of covariance (ANCOVA), there were no significant differences between BMI classes.

Discussion

In this study we demonstrated that SH as defined by AASM criteria, is a common finding in stable COPD patients (GOLD stages III-IV) with normocapnia during wakefulness. In our study population 66% of the COPD patients complied with the AASM definition of SH. Furthermore, we observed that the BMI did not influence the prevalence of SH.

Only few studies have evaluated the prevalence of nocturnal hypercapnia in COPD patients with diurnal normocapnia. Similar to our results, Kitajima et al. found a high prevalence (48%) of nocturnal episodic hypercapnia [16] in COPD patients. They used a mild definition of nocturnal hypercapnia namely an episodic increase of ≥ 5 mmHg from baseline PtcCO₂ accompanied by an episodic oxygen desaturation of < 90% for ≥ 5 min continuously, at least once during the night. Furthermore, no evaluation of an association with weight or BMI was performed by Kitajima et al.

Another study by Holmedahl et al. did apply the AASM SH criteria in normocapnic COPD patients [17] and found that only 6 out of the 76 normocapnic patients (8%) studied appeared to have SH. The mean BMI of their study population was lower (25.5 kg/m²) compared to the present study (28.0 kg/m²), while the mean FEV₁ predicted was higher (43% vs. 38.5%). These differences might (partially) explain the lower prevalence of SH reported by Holmedahl et al. This assumption is supported by the following: Firstly, the severity of airflow obstruction in COPD has been associated with a higher risk for hypoventilation/hypercapnia [21]. In our study, mean FEV₁ was considerably lower, especially in our NW (32%) and OW (38%) group compared to the group of Holmedahl et al. providing a possible explanation for the higher prevalence in our study. Secondly, previous studies evaluating COPD patients with diurnal hypercapnia, indicate that the prevalence and severity of SH is correlated with higher BMI values [14,15]. Also, studies in patients with OSA without COPD indicate that obesity is an independent risk factor for hypoventilation [22–24]. Hence, even though we did not find an association between BMI classes and SH, the higher overall BMI in our study population may have played a role in the higher overall prevalence of SH. It is also important to mention that differences in results with Kitajima et al. and Holmedahl et al. might be due to the use of different devices for P_tcCO₂ measurement (Kitajima: SenTec; Holmedahl: Radiometer) and placement of the transcutaneous electrodes, while we have placed this to the earlobe it is unknown at which site the electrodes were placed in the other studies.

In our study we found no association between BMI class and (severity of) SH. Also, parameters of sleep quantity and sleep quality as measured by PSG and PSQI, were comparable between BMI groups. Although this result is in line with a previous study in normocapnic COPD patients [17], indicating no negative association between increasing weight and SH, it is in contrast with data from hypercapnic COPD patients [14,15], where higher BMI was correlated with nocturnal hypoventilation. A potential explanation might be that the NW group had more airflow obstruction compared to the OB group, thus masking the negative effect of higher BMI on SH prevalence. However, we could not find a correlation between the prevalence of SH and BMI or FEV₁. Similarly, no correlation was found between the presence of SH and body fat distribution or body composition markers. This lack of correlations may be due to the bias of only including patients with severe or very severe COPD (GOLD stages III-IV) in combination with relatively small sample sizes in the various BMI groups due to a relatively high number of excluded patients. More studies, including COPD patients of all GOLD stages and larger study populations are needed in the future to address these issues.

Our data indicate that a large portion of COPD patients with diurnal normocapnia have SH as defined by the AASM. In COPD, lung hyperinflation may be protective against certain sleep-related breathing disorders such as obstructive sleep apnea. However, by diminishing the efficacy of diaphragm and in conjunction with disturbed ventilation-perfusion, worsening airflow obstruction during sleep and decreased skeletal muscle contraction, it contributes to disturbed gas exchange during sleep [7]. Earlier reports indicate that nocturnal gas exchange impairments are associated with arrhythmias during sleep, pulmonary hypertension, higher risk of COPD exacerbation and worse survival. [1,2,16]. A recent study in patients with COPD GOLD stages III-IV, indicated reduced hospital admissions, improved symptoms and quality of life measures by using low pressure domiciliary noninvasive ventilation in this group [25]. Interestingly, all these benefits were not only seen in patients with diurnal hypercapnia, but also in normocapnic patients. This study did not assess SH in their subjects as a possible explanation of this finding. One could hypothesize that the relative high prevalence of SH in the normocapnic group, as shown in our study, might have played a role. Taking all this into consideration, our findings emphasize the urgency of further studies dealing with early identification of patients with SH to explore the potential of treating SH.

To our best knowledge, this is the first study assessing the prevalence and correlation of SH with obesity as a primary objective in stable normocapnic COPD patients using the AASM criteria for SH. However, this study has some limitations: The final study sample included less patients than initially anticipated, as more than 25% of patients had to be excluded from analysis, mainly because of technical PSG failures. The airflow limitation was unexpectedly significantly better in COPD_OB. Theoretically, this might partially explain why SH was not more prevalent in this group. However, also after adjustment for this confounder we could not find a negative correlation between obesity and SH. It should be noted that our results are not generalizable for the whole COPD population, since only patients with severe or very severe COPD were included. We did not asses bicarbonate levels with arterial blood gasses during wakefulness. High levels of bicarbonate can be suggestive for SH and it would be interesting to correlate this with the more direct measurement of transcutaneous PCO₂ during sleep. Furthermore, while de AASM definition is usually used to define SH, some could argue the clinical relevance of this definition because this only discriminates between having SH and not having SH and does not provide insight into the severity of SH. Finally, this was a single center study with no control group of healthy individuals to compare for the occurrence and severity of SH.

In conclusion, we observed that SH was common in our sample of stable COPD Gold III and IV patients with daytime normocapnia and was not associated with BMI or body composition.

Declaration of interest

The authors declare that there is no conflict of interest.

Additional information

Funding

This work was supported by an unrestricted grant from GlaxoSmithKline. The funding agency had no involvement in study design, data collection, data analysis, interpretation of data, or writing of the report.