Periodic breathing in patients with stable obstructive sleep apnea on long-term continuous positive airway pressure treatment: a retrospective study using CPAP remote monitoring data

We conducted a retrospective cohort study on 775 patients who had used the Dream Station Auto CPAP machine (Philips Respironics, Murrysville, PA) for at least 1 year at Mie Sleep Clinic as of September 1, 2020. All patients were diagnosed as having OSA, with an apnea–hypopnea index (AHI) ≥ 20 by overnight PSG, or having sleep apnea syndrome (SAS), predominantly of obstructive type, with AHI ≥ 40 by out-of-center sleep testing (OCST), and have been continuously treated with CPAP.

Exclusion criteria were as follows: age ≥ 90 years as of September 1, 2020; poorly adherent patients with a percentage of days of CPAP use of < 50% or an average CPAP use time of < 4 h between September 1 and 30, 2020; and new onset of heart failure, worsening of chronic heart failure, or new onset or relapse of atrial fibrillation (Af) or other cardiovascular diseases in the 3 months before or after September 1, 2020 (i.e., between June 1, 2020, and November 30, 2020).

This study was approved by the ethics committee of the Medical Corporation MSC (#20,002). Written consent was obtained from all patients. This study is registered in the UMIN Clinical Trials Registry (UMIN000042555).

The primary endpoint was the calculation of standard data, such as the monthly median PB%, and the analysis of factors associated with PB in patients with stable OSA on CPAP. The secondary endpoint was subgroup analyses of factors that were strongly associated with PB.

The following baseline characteristic data were extracted from the Medical Corporation MSC Mie Sleep Clinic database, which contained data from October 1, 2007, to November 30, 2020, with the base point set at September 1, 2020 (Table 1):

Data on comorbidities and patients’ medication status were obtained from the database or from the medical records of patients following physician referral. The diagnoses of heart failure, ischemic heart disease, and Af were based on diagnoses by cardiologists at neighboring flagship hospitals.

Table 2 shows the further classification of Af type (paroxysmal, persistent, or permanent), QRS complex, and central nervous system (CNS) agents. We included four medication classes as CNS agents: opioids, benzodiazepines, sedatives, and non-benzodiazepine agonists.

Full attended overnight polysomnography (PSG) was conducted in a laboratory setting, which was classified as type 1 sleep study by the American Academy of Sleep Medicine (AASM). PSG was performed using the Alice® 4 or Alice® 6 system (Philips Respironics, Murrysville, PA). Sleep and respiratory event scoring was carried out according to the AASM manual. Our clinic is specialized in pulmonary medicine. All patients with a history of heart failure underwent diagnostic PSG in the stable compensatory phase of heart failure.

The AASM recommends scoring a respiratory event as Cheyne-Stokes breathing if both of the following criteria are met:

According to the algorithm of the CPAP device, the device detects PB that meets only criterion a. Thus, in this study, we defined PB as follows:

Patients with PB%Device ≥ 2% were divided into the emergence group that included those with PB%PSG < 1% and the persistence group that included those with PB%PSG ≥ 1%.

The clinical factors associated with the emergence or persistence of PB were investigated in these groups.

At our clinic, CPAP treatment is initiated in the Auto CPAP mode on an outpatient basis (Supplemental Figure). When the treatment is initiated, the upper limit of pressure is set at a value estimated to not be excessive, according to the body weight of each patient. Within 1 week of treatment initiation, daily remote monitoring was initiated. The association between pressure variation and respiratory events, as well as leak patterns, is closely monitored, in addition to the AHI calculated using the CPAP device. The patient was re-examined at the outpatient clinic within 2–4 weeks after treatment initiation. If obstructive events persist (AHI ≥ 10, PB%Device < 2%) after the set pressure upper limit was reached, we would raise the pressure limit further. When a patient exhibits PB and a large leak with pressure reaching the upper limit or increasing despite an AHI ≥ 10, PB%Device ≥ 2%, and a decreased incidence of obstructive events, the patient is regarded to have treatment-emergent central sleep apnea (TECSA). The pressure at the occurrence of PB was assumed to be the break point, and the upper limit of the pressure was adjusted to be lower than this point. This adjustment leads to resolution of the TECSA pattern within 3 months in most patients without comorbid heart or cerebrovascular disease. When the TECSA pattern persists for 3 months, patients are admitted for manual CPAP titration under PSG. Meanwhile, patients suspected of having comorbid heart failure or other heart or cerebrovascular diseases, and who provided consent, underwent measurement of brain natriuretic peptide (BNP), electrocardiography, and echocardiography, according to different criteria (AHI15 and CSR > 2%). The patients were then admitted for manual adaptive servo-ventilation (ASV) titration under PSG within 3 to 6 months. Patients indicated for ASV are transitioned to ASV.

CHF patients with TECSA pattern sets the minimum pressure support of the ASV to 0 cmH2O. This study included patients who had been continuously treated with CPAP for 1 year or longer, after undergoing the procedure described above at our clinic.

The CPAP device used was Dream Station Auto. The device periodically records the user’s breathing signals and analyzes the data in near real time. It also records raw respiratory data at random. Analysis results are automatically uploaded to a central database. The data were analyzed online using the dedicated analysis system (EncoreAnywhere, ver. 2.44; Philips Respironics) and cited using the CareExporter software (version 1.11.1.0; Philips Respironics). The data cited were average apnea/hypopnea index (AHI), average central apnea (clear airway) index (CAI), PB%Device, average percentage of large leak on days of CPAP use (LL%), percentage of days of CPAP use (%), and average CPAP use time (on days of use, min). The monitoring data from the CPAP device was averaged over a 30-day period starting on September 1, 2020.

The Kolmogorov–Smirnov test was used to test the normal distribution of continuous variables. Data are presented as average ± SD, median (range), 75th percentile, or n (%). The data for AHI, CAI, PB%Device, and LL% showed a pointed distribution, expressed as median (full range), and transformed to 10-based logarithmic scale to obtain a normal distribution. The correlation among log AHI, log CAI, and log PB%Device was analyzed using the Pearson product-moment correlation coefficient. For the analysis of factors associated with PB, multiple regression analysis was used to create a prediction formula for log PB%Device. For explanatory variables, those that were considered to be clinically important based on previous studies were added to the model. In addition to age and sex, smoking history and obesity, which are cardiovascular risk factors, as well as factors previously identified as being associated with PB, including use of CNS agents, such as alcohol and opioid [8, 9], history of congestive heart failure, history of stroke/TIA, history of pacemaker implantation, Af, hypertension, concomitant stage ≥ 3b CKD [10], QRS duration on ECG [11‐13], percent large leak calculated from CPAP device data [7, 14], and average CPAP use time [15], were considered to be related to PB onset and analyzed as binary variables, where BMI ≥ 30, CKD stage ≥ 3b, QRS ≥ 110 ms, and LL% ≥ 15% were coded as 1. The difference in average log PB%Device between different Af types was analyzed using one-way analysis of variance (ANOVA), followed by multiple comparison using Tukey’s post hoc test. The difference in QRS duration between the two groups was analyzed using Welch’s t-test.

Probability values ≤ 0.05 were considered significant. All statistical analyses were performed with EZR Ver.1.52 (Saitama Medical Center, Jichi Medical University, Saitama, Japan), which is a graphical user interface for the R version 4.02 (The R Foundation for Statistical Computing, Vienna, Austria). Specifically, it is a modified version of the R commander designed to add statistical functions frequently used in biostatistics.

A total of 775 patients were enrolled, of whom 3 were aged ≥ 90 years, 149 were poorly adherent to CPAP treatment, and 5 had cardiovascular events within 3 months before or after analysis. Consequently, 618 patients were considered as having stable OSA and were included in the analysis (Fig. 1). Table 1 lists the baseline characteristics of the analyzable patients. The average duration of CPAP use by the 618 patients was 7.5 ± 4.0 years (1–13 years). The average age was 61.7 ± 12.2 years, men accounted for 88.8% of all patients, and the average BMI was 27.2 ± 4.9 kg/m². A history of congestive heart failure was seen in 5% of patients, while 5% had prior stroke/TIA and 5% had stage ≥ 3b CKD. Of the 60 (10%) patients with Af, 25 had paroxysmal, 10 had persistent, and 25 had permanent type of Af. The QRS cut-off of 110 ms was based on the upper limit of the normal range for the Japanese population. Of the 16 patients with a QRS duration ≥ 120 ms, 14 had complete right bundle branch block (CRBBB), including 1 patient with Wolff-Parkinson-White syndrome. Among CNS agents, opioids were used only by 2 patients, both for pain relief from orthopedic diseases (Supplemental Table).

Sleep variables derived from the CPAP device are shown in Table 2. The average CPAP use time was 381.0 ± 74.4 min, and the average percentage of days of CPAP use was 95.5 ± 9.4%, indicating excellent adherence to CPAP treatment. The median AHI of 3.1 times/h also indicates favorable control of SAS in almost all patients. The median PB%Device in stable OSA patients on CPAP was low at 0.32% with a 75th percentile of 0.94%. The average log PB%Device was − 0.43 ± 0.64, which corresponds to a PB%Device of 0.37% ± 2.360 (0–32.8%). Thus, of the 618 patients, 149 (24%) had PB% of ≥ 1%.

Figure 2a and b show the correlations between log AHI and log PB%Device and between log CAI and log PB%Device, respectively. Patients with AHI ≥ 5 with poorly controlled OSA on CPAP tended to have higher PB%Device values and showed a moderate correlation between AHI and PB%Device with r = 0.562 (95% CI: 0.51–0.61). Of the 466 patients whose SAS was well controlled with AHI < 5, 66 (14.2%) had PB%Device ≥ 1%, with the minimum being 1% and the maximum being 14.3%. The correlation between log CAI and log PB%Device was moderate with r = 0.42, and only 2 patients had CAI ≥ 5.

The results of linear multiple regression analysis of factors associated with PB (log PB%Device) are shown in Table 3. Eleven patients had a PB% of 0, which was considered as missing data. The p value of F-test was < 0.001, indicating that the current model was effective. The most important factor was atrial fibrillation (Af) with a coefficient of 0.693 (95% CI; 0.536 to 0.851), followed by QRS duration coefficient of 0.445 (95% CI; 0.304 to 0.586), followed by history of heart failure, male, comorbid hypertension, obesity, and age. Contrastingly, CKD, prior stroke/TIA, and CNS agents were not strongly associated with CSB onset. Finally, the following regression equation was obtained.

Equation for multiple regression: [log PB%Device =  − 1.464 + 0.011 × age + 0.213 × male (1) + 0.136 × BMI ≥ 30 (1) + 0.693 × Atrial fibrillation (1) + 0.227 × History of CHF (1) + 0.143 × Hypertension (1) + 0.445 × QRS duration ≥ 110 (1)].

In the comparison of log PB%Device by type of Af, no significant difference was observed between permanent (average PB% = 6.81, n = 25) and persistent (average PB % = 2.37, n = 10) types (Fig. 3a), while there were significant differences between paroxysmal (average PB % = 0.69, n = 25) and persistent types (p = 0.032) and between paroxysmal and permanent types (p < 0.001). For log AHI, a significant difference between paroxysmal and permanent types (p = 0.002; Fig. 3b) was observed.

In comparisons between subgroups based on abnormal QRS duration on ECG, patients with QRS duration ≥ 120 ms group showed significantly higher log PB%Device values (average PB%: 3.16%, n = 16) than those with 120 ms > QRS duration ≥ 110 ms group (average PB%: 0.82%, n = 54; p = 0.002; Fig. 4a). Similarly, log AHI values were also significantly higher in those with QRS duration ≥ 120 ms group (average AHI: 6.64, n = 16) than those with 120 > QRS duration ≥ 110 ms group (average AHI: 3.59, n = 54; p = 0.003; Fig. 4b).

In 61 of 79 patients with PB%Device ≥ 2%, we re-examined the data from diagnostic PSG, calculated PB%PSG, and compared PB%PSG with PB%Device. Table 4 shows clinical factors of the persistence group and the emergence group.

In the persistence group, PB%PSG was 1% or higher at baseline in 37 of 61 patients (60.7%). Of these 37 patients, 33 (89.2%) showed at least one of the following findings: past history of CHF, stroke, and atrial fibrillation; diastolic dysfunction detected by echocardiography between 2019 and 2020; and increased QRS duration detected by electrocardiography between 2019 and 2020 (CHF in 13 patients, cerebral infarction in 5 patients, atrial fibrillation in 19 patients, diastolic dysfunction in 8 patients, and increased QRS duration in 13 patients). In the remaining four patients, elevated PAP device pressure was not associated with the occurrence of PB, and the frequency of PB indicated that TECSA was unlikely. Two of these patients were public transportation drivers, and their PB% increased consistently on their days off; they had a habit of drinking a relatively large amount of alcohol on days off.

In the emergence group, PB%PSG at the time of diagnosis was < 1% in 24 patients (39.3%).

Of these 24 patients, 20 (83.3%) showed at least one of the following findings during the period from diagnostic PSG to May 31, 2020: new onset or relapse of heart failure, cerebral infarction, or atrial fibrillation, diastolic dysfunction detected by echocardiography between 2019 and 2020, and increased QRS duration detected by electrocardiography between 2019 and 2020. (New onset or relapse of heart failure in 6 patients, new onset of cerebral infarction in 1 patient, comorbid atrial fibrillation in 7 patients, diastolic dysfunction in 4 patients, and increased QRS duration in 10 patients). In the remaining four patients, elevated PAP device pressure was not associated with the occurrence of PB. Their general condition was favorable, and their lifestyles did not change. The frequency of PB was approximately 2% in 3 of these 4 patients.