Impact of adherence with positive airway pressure therapy on hypercapnia in obstructive sleep apnea

Abstract

Background

Daytime hypercapnia associated with obstructive sleep apnea is common, particularly in the severely obese, and is associated with serious complications. Although positive airway pressure therapy improves daytime hypercapnia, the magnitude of benefit is unknown. Our objective was to quantify the effect of adherence with positive airway pressure on hypercapnia and hypoxia and to identify other factors that predict changes in PaCO₂ and PaO₂.

Methods

We performed a retrospective cohort study of seventy-five patients using a multivariable general linear model analysis to identify variables that predicted changes in PaCO₂ and PaO₂ after therapy. Bootstrap resampling methods were used to calculate confidence intervals for the effects of significant predictors and to internally validate the predictive models.

Results

The variables that predicted the change in PaCO₂ were: average daily hours of positive pressure therapy, FEV₁% of predicted, and baseline PaCO₂ (model R² = 0.70). The PaCO₂ dropped 1.84 mm Hg per hour of adherence and plateaued at 7 hours of average daily use. PaO₂ improved by approximately 3 mm Hg per hour of adherence and plateaued after 4.5 hours of therapy (model R² = 0.48). Patients who used therapy for more than 4.5 hours per day experienced significant improvements in PaCO₂ and PaO₂ compared to less adherent patients (ΔPaCO₂ 7.7±5 vs. 2.4±4 mm Hg, p<0.001; ΔPaO₂ 9.2±11 vs. 1.8±9 mm Hg, p<0.001). For adherent patients, the need for daytime home oxygen therapy decreased from 30% to 6% (p=0.02).

Conclusion

In hypercapnic patients with obstructive sleep apnea, adherence with positive airway pressure is an important modifiable predictor of improvements in PaCO₂ and PaO₂, and its benefit plateaus between 5 to 7 hours of daily therapy.

INTRODUCTION

The subset of patients with obstructive sleep apnea (OSA) who have chronic daytime hypercapnia (PaCO₂ > 45 mm Hg) experience a lower quality of life,^1,2 have greater healthcare expenses,³ and a higher risk of pulmonary hypertension.^4,5 Hypercapnia is associated with a four-fold increase in mortality among the severely obese after adjusting for age and body mass index.^6,7 Although the estimated prevalence of chronic daytime hypercapnia in patients with OSA varies widely, from 10% to 38% ^8-12 (35% in our clinic population), the prevalence will likely surge, at least in the United States, in parallel with the rising incidence of obesity.^12,13

The optimal management of patients with hypercapnia associated with OSA remains uncertain despite several small studies reporting improvement of hypercapnia with positive airway pressure (PAP) therapy.^14-19 The largest of these studies had a sample size of 23 patients¹⁷ and none of them measured the effect of average daily use of PAP on the improvement of hypercapnia.

We performed a retrospective cohort study of 75 patients with OSA and chronic daytime hypercapnia to describe the effect of adherence with positive airway pressure therapy on PaCO₂ and PaO₂ and to identify factors that modify response to therapy.

METHODS

Study Population

We assembled the cohort by first identifying patients with OSA and chronic daytime hypercapnia who received PAP therapy (n=169). All patients were followed at the sleep medicine clinic of a large public urban teaching hospital between April 2000 and April 2004. Our inclusion criteria were: age above 18 years, a stable daytime PaCO₂ above 45 mm Hg with pH less than 7.45 (without evidence of acute respiratory acidosis or metabolic alkalosis) measured in an ambulatory setting, apnea-hypopnea index (AHI) above 10 per hour, and therapy with a PAP machine equipped with a downloadable memory card. We excluded patients if their pre-treatment arterial blood gases were measured during a hospitalization or more than a year before starting therapy with PAP or if their post-treatment arterial blood gases were measured more than two years after starting therapy. We also excluded patients lacking adherence data or those who had undergone tracheostomy or bariatric surgery. We did not exclude patients with a history of obstructive airways disease. Table 1 provides the reasons for exclusion from the study cohort.

Table 1. Reasons for exclusion of subjects from the study^*.

Reasons for exclusion
Second PaCO2 > 2 years after PAP	32
First PaCO2 not available	19
First PaCO2 measured during a hospitalization	17
First PaCO2 > 1 year before PAP	10
Lost to follow-up	7
Adherence data not available	5
Tracheostomy	2
Gastric bypass	1
Second PaCO2 not available	1
Total included study subjects	75

^*The study cohort of 75 subjects was derived from 169 patients diagnosed with OSA and hypercapnia and treated from 2000-2004 at one sleep clinic. OSA refers to obstructive sleep apnea; PAP, positive airway pressure therapy.

Data Collected

Patients presenting to our sleep clinic for the suspicion of sleep apnea had the following data routinely collected: demographics, symptoms of OSA, co-morbid medical disorders, Epworth sleepiness scale,²⁰ Beck depression inventory,²¹ pulmonary function test (PFT), and results of polysomnography and positive airway pressure titration. Pulmonary function tests were ordered by internists or pulmonologists for evaluation of dyspnea or to determine the need for home oxygen therapy. Arterial blood samples were drawn from the radial artery with the patients sitting for 20 minutes and breathing room air. We recorded the results of arterial blood gas analysis performed before (first) and after at least 30 days of therapy with PAP (second). All arterial blood samples were analyzed by co-oximetry (ABL 725 System, Radiometer, Denmark). PFTs included spirometry and measurement of lung volume (SensorMedics, Loma Linda, CA) and were performed according to the American Thoracic Society standards.²² Weight was measured in the sleep laboratory on a scale with maximum limit of 340 Kg (Scale-Tronix, Wheaton, Illinois).

Polysomnography and positive airway pressure titration

Polysomnograms were recorded online and were monitored continuously in the laboratory using a 16-channel system (Sandman Elite version 7.0, Ontario, Canada). The polysomnographies were scored by experienced registered polysomnography technicians according to standard practices.^23,24 Inter-scorer agreement in our laboratory is routinely monitored and is 90% or higher. Titrations started in CPAP mode and supplemental oxygen or bi-level PAP was considered if CPAP had eliminated all respiratory events and SaO₂ remained below 88%. A split night protocol was performed in 92% of our patients: titration of positive airway pressure was initiated after 3 hours of baseline sleep in patients who exhibited repetitive episodes of obstructive respiratory events with AHI above 20 per hour.

Adherence with positive airway pressure therapy

Our main predictor variable was adherence with PAP therapy. We measured adherence by calculating the mean number of hours per day a patient used PAP at the prescribed pressure during the 30 days before the second (follow-up) measurement of arterial blood gases. Daily use at the prescribed pressure was monitored by Smartcard technology (Respironics Encore® Pro Smartcard^™, Murrysville, PA). We selected the 30-day period before the second measurement of arterial blood gases because all patients were on therapy for at least 30 days and because there was no significant difference in model results when we explored 60 and 90 day periods.

This study was approved by the hospital’s Institutional Review Board.

Statistical analysis

We compared the characteristics of the study sample with the group of patients who were excluded from the analyses to assess whether our sample was representative of the entire population of patients with hypercapnia and OSA seen in our clinic. We used chi-square, t-tests, and Mann-Whitney tests where appropriate.

We used general linear models to assess the effect of predictors on the change in PaCO₂, our primary outcome. The same modeling strategy was used to evaluate the change in PaO₂, our secondary outcome. Groups of potential predictors were entered into the model and tested using a hierarchical strategy: first, the baseline blood gas values were entered and tested; second, demographic variables; third, baseline clinical data, including PFT and polysomnography results; and, finally, adherence with PAP therapy was added and tested. Hypothesizing that adherence might have a nonlinear relationship with the change in PaCO₂ because of a likely ceiling or plateau effect, we tested whether quadratic or cubic terms for adherence were also significant. At each step, we retained those variables that remained significant at the 5% level. We modeled the change in PaCO₂ while controlling for the baseline PaCO₂ value because controlling for the baseline value makes the analysis less subject to bias from regression to the mean and usually yields more precise estimates.²⁵ We used a bootstrap resampling procedure to internally validate the two predictive models and generate confidence intervals for the effect size of significant predictors.²⁶

All analyses were conducted using the statistical software of SPSS, version 12 (SPSS, Inc, Chicago, IL) or STATA, version 8 (StataCorp, College Station, TX).

RESULTS

Comparison of patients with hypercapnia included in the analysis with patients who were excluded

Among the 169 subjects with hypercapnia and OSA who were treated with PAP therapy, 94 met one of the exclusion criteria (Table 1), leaving 75 subjects eligible for study. The 75 study subjects were similar to the 94 excluded subjects on the 38 variables measured at baseline, except for three: study subjects spent more time on positive airway pressure titration (363 vs. 240 minutes, p<0.001), were more likely to be prescribed nocturnal oxygen (37% vs. 20%, p=0.01), and a higher percentage received bi-level PAP (20% vs. 9%, p=0.03) (Table 2). For the entire cohort of 75 patients, the first measurement of arterial blood gases was performed at a median of 84 days (IQR 32-162 days) before initiating PAP therapy. The second measurement of arterial blood gases was performed at a median of 91 days (IQR 42-253 days) after initiating therapy.

Table 2. Baseline characteristics of the 75 study subjects.

Age	48±13
Gender (% men)	49
Body mass index	51±13
Race (%)
African American	72
White	13
Hispanic	12
Other	3
Co-morbidities (%)
Hypertension	79
Diabetes Mellitus	32
Obstructive Airways Disease	37
Congestive heart failure	21
Excessive daytime sleepiness (%)	92
Loud snoring (%)	92
Witnessed apneas (%)	56
Epworth sleepiness scale	14±6
Beck depression inventory	14±10
Polysomnographic parameters
Sleep efficiency (%)	80±12
Apnea-hypopnea index (median, IQR)	95, 50-125
Apnea-hypopnea index categories (%)
AHI 10-15	7
AHI 16-30	11
AHI 31-90	25
AHI ≥ 91	57
Arousal index (median, IQR)	31, 21-44
Lowest oxygen saturation (%)	60±13
Positive airway pressure titration
Time spent on titration (minutes)	363±82
AHI on final pressure (median, IQR)	3, 0-10
Type of PAP, %
Continuous	80
Bi-level	20
PAP pressure, cm H2O
Continuous	14±3
IPAP/EPAP*	18±2/13±2
Nocturnal oxygen (%)	37
Spirometry
FVC (% of predicted)	58±15
FEV1 (% of predicted)	55±17
FEV1/FVC (%)	76±10
Lung volumes
TLC (% of predicted)	76±18
ERV (% of predicted)	42±22
RV (% of predicted)	107±49
First arterial blood gas
pH	7.39±0.03
PaCO2, mm Hg	54±7
PaCO2 categories (%)
45-50 mm Hg	37
51-55 mm Hg	23
≥ 56 mm Hg	40
PaO2, mm Hg	59±11
Hemoglobin, g/dl	14±3

Data presented as mean ± SD unless otherwise specified.

^*Pressures for patients on bi-level positive airway pressure (PAP) therapy. AHI refers to apnea-hypopnea index; IQR, interquartile range; FVC, forced vital capacity; FEV₁, forced expiratory volume at 1 second; TLC, total lung capacity; ERV, expiratory reserve volume; RV, residual volume; IPAP, inspiratory positive airway pressure; EPAP, expiratory positive airway pressure.

Adherence with positive airway pressure therapy and change in PaCO₂

For the 34 patients (45%) whose adherence was more than 4.5 hours per day, the average drop in PaCO₂ was 7.7±5 mm Hg, compared to 2.4±4 mm Hg in patients whose adherence was less than 4.5 hours per day (P<0.001). The second PaCO₂ normalized (< 45 mm Hg) in 34 percent of patients with adherence of more than 4.5 hours per day compared to 17 percent of patients with adherence below 4.5 hours per day (p=0.001). For the entire cohort, PaCO₂ decreased from 54±7 mm Hg to 49±7 mm Hg (p<0.001).

Using multivariable general linear modeling, we identified four variables that independently predicted the change in PaCO₂ and explained 70% of its variance: 1) average daily use of PAP therapy (adherence), 2) adherence squared, 3) FEV₁% of predicted, and 4) first PaCO₂ (Table 3, Figure 1). The significance of the quadratic term—adherence squared—indicates that the relationship between adherence and change in PaCO₂ is nonlinear, with a plateau in improvement after 7 hours of daily PAP therapy. Thus, an average use of two hours daily produces a decline of 3 mm Hg in PaCO₂, four hours daily produces a decline of 6 mm Hg, and more than seven hours daily use produces a decline of 8 mm Hg.

Table 3. General linear model for estimating the change in PaCO₂.

Parameter	Value	Coefficient	95% CI	P value	Explained variance (95% CI)
Adherence	Hrs of average daily PAP use	−1.84	(−2.83, −0.98)		34% (19-53%)
Adherence squared	Hrs2 of average daily PAP use	0.11	0.03
FEV1% of predicted	Percent (0 to 100)	−0.08	(−0.15, −0.03)	0.004	13% (2-29%)

All coefficients are adjusted for baseline PaCO₂ in addition to the other variables in the model listed above. Model R² = 0.70 (95% CI 0.55-0.82). All 95% CI’s were constructed by bootstrap.

CI = confidence interval, PAP= Positive airway pressure.

Figure 1.

Relationship between mean daily adherence with positive airway pressure therapy in the last 30 days prior to the measurement of second PaCO₂ and change in PaCO₂ (ΔPaCO₂). The curves represent estimated ΔPaCO₂ with 95% confidence interval for the entire cohort adjusted for baseline PaCO₂ and FEV₁% of predicted.

The change in PaCO₂ was unrelated to age, gender, race, BMI, change in weight, presence of obstructive airways disease or other co-morbidities, PFT variables other than FEV₁% of predicted, symptoms of OSA, AHI, AHI on prescribed pressure, arousal index, or lowest oxygen saturation during polysomnography. The type of positive airway pressure (bi-level or continuous), use of nocturnal oxygen, and total duration of therapy had no impact on the change in PaCO₂. Results did not differ significantly if we calculated adherence using other periods (for example, the 60-day period preceding the last blood gas measurement) or if we used the patient’s cumulative use of positive airway pressure.

Adherence with positive airway pressure therapy and change in PaO₂

For the entire cohort, the PaO₂ increased from 59±11 mm Hg to 64±11 mm Hg (p<0.001). For the 34 patients (45%) whose adherence was more than 4.5 hours per day, the average increase in PaO₂ was 9.2±11 mm Hg compared to 1.8±9 mm Hg in less adherent patients (P<0.001). In adherent patients, the proportion of patients with PaO₂ of less than 55 mm Hg decreased from 30% to 6% (p=0.02) in contrast to an insignificant decrease of 34% to 32% in less adherent patients (p=0.01 for comparison between groups).

The change in PaO₂ was independently predicted by adherence, adherence squared, first PaCO₂, baseline PaO₂, and age (R²=0.48) (Table 4). Based on the model, the change in PaO₂ reached a plateau at approximately 4.5 hours of average daily therapy with PAP. The change in PaO₂ was unrelated to gender, race, BMI, change in weight, presence of obstructive airways disease or other co-morbidities, PFT variables, symptoms, AHI, arousal index, or the lowest oxygen saturation during polysomnography. The type of positive airway pressure (bi-level or continuous), use of nocturnal oxygen, and duration of therapy also had no impact on the change in PaO₂.

Table 4. General linear model for estimating the change in PaO₂.

Parameter	Coefficient	95% CI	p-value	Explained variance (95% CI)
Adherence	3.0	1.07, 5.24	<0.001	14% (2 to 35%)
Adherence squared	−0.29	−0.52, −0.05	0.002	10% (0.3 to 29%)
First PaCO2	−0.19	−0.62, 0.32	0.02	2% (0.001 to 17%)
Age	−0.23	−0.37, −0.10	0.001	14% (3 to 29%)

Model coefficients are adjusted for baseline PaO₂ in addition to all other predictor variables listed above. Model R² = 0.48 (95% CI 0.22-0.63). All 95% CI’s constructed by bootstrap.

Failure to improve hypercapnia despite adequate adherence to therapy

Among the 34 patients who used PAP for at least 4.5 hours per day, 8 patients (23%) did not have a significant improvement in their PaCO₂ — decrease in PaCO₂ of less than 4 mm Hg. These non-responders had less severe disease, represented by a lower AHI, compared to responders (44±45 vs. 86±47 per hour, p=0.03). Non-responders and responders were similar in age, race, gender, BMI, final AHI on prescribed pressure, time spent during polysomnography on titration of positive airway pressure, type of PAP therapy (continuous vs. bi-level), FEV₁% of predicted, FVC% of predicted, FEV₁/FVC ratio, baseline PaCO₂ and PaO₂, and adherence. Mean adherence with PAP therapy was 7.2±2.1 hours per day for non-responders vs. 6.0±1.7 hours per day for responders (p=0.1).

DISCUSSION

We found that among hypercapnic patients with obstructive sleep apnea, the magnitude of improvements in PaCO₂ and PaO₂ were directly related to adherence with PAP therapy but unrelated to BMI, change in weight, pulmonary function parameters (except FEV₁% of predicted), duration of therapy, type of PAP therapy (bi-level vs. CPAP), or polysomnography characteristics.

In this cohort, the only modifiable determinant of the change in PaCO₂ and PaO₂ was adherence with PAP therapy. Initially, the PaCO₂ dropped by 1.84 mm Hg for each hour of daily therapy with PAP. The beneficial effects of PAP therapy on PaCO₂ plateaued after 7 hours of average daily use and the effect on PaO₂ plateaued after 4.5 hours, hence the significance of the “adherence squared” term. The magnitude of benefit was greatest among patients with the highest baseline PaCO₂. While some of the improvement might be a regression to the mean, patients with the highest baseline PaCO₂ were probably more responsive to therapy. Patients with a higher percentage of predicted FEV₁ experienced a greater reduction in PaCO₂. Percentage of predicted FEV₁ was, however, weaker than adherence with therapy in predicting change in PaCO₂. For example, in 2 patients with identical adherence with therapy, the patient with an FEV₁ of 70% of predicted will experience a drop in PaCO₂ that will be only 1.6 mm Hg higher than the patient with an FEV₁ of 50% of predicted (0.08 mm Hg per 1% of increase in percentage of predicted FEV₁).

Although morbid obesity, obstructive ventilatory defects, restrictive ventilatory defects, daytime hypoxemia, and severe and prolonged nocturnal oxygen desaturation have been proposed as mechanisms behind the pathogenesis of hypercapnia in patients with OSA, none of these variables were related to the change in PaCO₂ in our model.^{8-10,19,27-33} It is possible that we were unable to demonstrate a relationship between BMI or change in BMI and the change in PaCO₂ because our cohort was homogeneously obese and few patients lost weight between the two blood gas measurements.

PaO₂ initially improved by approximately 3 mm Hg for each hour of daily therapy with PAP but quickly plateaued after 4.5 hours. With this degree of improvement in oxygenation, daytime home oxygen therapy can be discontinued for many patients.

The number of days patients had the positive airway pressure machine and the cumulative use of positive airway pressure beyond the last 30 days did not predict the change in PaCO₂. These findings reflect the importance of the most recent use of PAP therapy on PaCO₂ and that the full benefits of therapy—at least regarding arterial blood gases—will be realized in one month. This finding is consistent with prior reports in which PaCO₂ normalized after two weeks of therapy with either noninvasive ventilation or tracheostomy.^16,19 In both reports, however, the baseline PaCO₂ (47-49 mm Hg), BMI, or AHI were lower than our cohort.

The twenty-three patients in the retrospective study by Berger et al.¹⁷ are the most comparable to ours in terms of BMI, PaCO₂ and AHI. Adherence with positive airway pressure therapy in their study was assessed subjectively. PaCO₂ dropped 10 mm Hg after a period of therapy that ranged from four days to seven years. PaCO₂ did not change in patients who reported non-adherence and dropped by 16 mm Hg in patients who reported adherence with therapy. This study was limited by a wide range of follow-up period, subjective assessment of adherence, significant weight loss in adherent patients, and the large proportion of patients who had a tracheostomy (one third). We avoided these limitations by using more stringent inclusion criteria, though arbitrary, such as excluding patients who lacked objective adherence data or who had post-treatment measurement of arterial blood gases performed beyond two years of therapy, or underwent tracheostomy. Our cohort had a median percent weight loss of 1.0% (IQR, 1%, 3.9%) between the measurements of the two arterial blood gases. The maximum weight loss was 16%. Peppard et al.³⁴ reported that a 10% weight loss predicted a 26% decrease in the AHI. We believe that the percent weight loss in our cohort was clinically inconsequential. In the general linear model analysis, the change in weight was unimportant in predicting the change in PaCO₂.

Our study has several limitations. First, it was a retrospective analysis that, by design, allows the introduction of selection bias. However, we obtained data on all 169 patients with hypercapnia and OSA seen in our clinic during a four-year period and documented the similarity between excluded and included patients, supporting the assumption that this was a representative sample without significant selection bias. Second, multiple measurements of arterial blood gases were not available for more precise estimates of PaCO₂ and PaO₂ at baseline and follow-up. Third, there was significant variability in the timing of arterial blood gases before and after therapy. Fourth, we included some patients with obstructive airways disease; however, their inclusion enabled us to develop a more inclusive and generalizable predictive model. Finally, the patients served at our clinic are predominantly from an uninsured, urban, minority population; however, it is just this population that appears to be at highest risk of developing severe OSA.^35-38

Implications in pathogenesis of hypercapnia in patients with OSA

The improvement in hypercapnia in patients adherent to PAP therapy was neither universal nor complete as demonstrated by two patient subgroups. In the first subgroup, PaCO₂ did not change in eight patients with baseline PaCO₂ of 51 mm Hg despite seven hours of daily therapy. In the second subgroup, PaCO₂ decreased from 54±7 mm Hg to 45±4 mm Hg in 26 patients that used PAP for six hours daily; however, PaCO₂ remained elevated (>45 mm Hg) in eleven of them. We hypothesize that in the first subgroup OSA did not play a role in the pathogenesis of hypercapnia. This hypothesis can explain why these non-responders had lower AHI suggesting that obesity was the main mechanism for hypercapnia in this subgroup. Morbid obesity can contribute to daytime hypercapnia through mechanisms other than severe OSA. For example, it has been suggested that hyperleptinemia with leptin resistance plays a role in the pathogenesis of hypercapnia.^39,40 The incomplete response to PAP therapy in the second subgroup suggests that, in addition to obesity, OSA plays a mechanistic role in the development of hypercapnia. It is possible that prolonged untreated OSA resets the central chemoreceptors and thereby makes the response to therapy incomplete.^41-43 The resetting of central chemoreceptors might be partially reversible within 30 days of therapy. It is important to emphasize that our definition of non-responders is arbitrary and might miss small genuine response in cases of mild hypercapnia.

In summary, the improvements in PaO₂ and PaCO₂ were directly related to adherence with PAP therapy and plateaued after 5 to 7 hours of daily use, respectively. Hypercapnia did not improve in a fourth of patients adherent to therapy. Further research is needed to explain this phenomenon and to evaluate alternative therapies in this group of patients. Patients can overestimate their adherence with PAP therapy;^44-46 therefore clinicians should rely on objective evidence of adherence. Physicians could provide more effective care if they promptly recognized and addressed obstacles to poor adherence.