Abstract
Study Objectives:
Obstructive sleep apnea (OSA) is a heterogeneous disease with varying phenotype. A cluster analysis based on multidimensional disease characteristics, including symptoms, anthropometry, polysomnography, and craniofacial morphology, in combination with auto-continuous positive airway pressure titration response and comorbidity profiles, was conducted within a well-characterized cohort of patients with OSA, with the aim to refine the current phenotypic expressions of OSA with clinical implications.
Methods:
Two hundred ninety-one patients with a new diagnosis of moderate to severe OSA referred for auto-continuous positive airway pressure titration to the sleep center were included for analysis. In-laboratory polysomnography and craniofacial computed tomography scanning were performed, followed by an auto-continuous positive airway pressure titration. The symptom of excessive daytime sleepiness was assessed using the Epworth Sleepiness Scale.
Results:
Three patient phenotypes—normal weight, nonsleepy, moderate OSA; obese, nonsleepy, severe OSA; and obese, sleepy, very severe OSA with craniofacial limitation—were identified. Among the polysomnography parameters, only percentage of N3 time of total sleep time (N3%) and mean pulse oxygen saturation were found to be associated with the Epworth Sleepiness Scale score, and they only explained a small fraction of the variation (R2 = .136). Neck circumference and craniofacial limitation were associated with the more severe phenotype, which had a higher prevalence of hypertension and metabolic syndrome, greater diurnal blood gas abnormalities, and worse positive airway pressure titration response.
Conclusions:
Three OSA phenotypes were identified according to multiple aspects of clinical features in patients with moderate to severe OSA, who differed in their prevalence of hypertension, metabolic syndrome, diurnal blood gas parameters, and continuous positive airway pressure titration response. Self-reported excessive daytime sleepiness was not related with the severity of sleep breathing disturbance, and craniofacial limitation was associated with the more severe phenotype. These findings highlight the necessity of integrating multiple disease characteristics into phenotyping to achieve a better understanding of the clinical features of OSA.
Citation:
Zhang XL, Zhang L, Li YM, et al. Multidimensional assessment and cluster analysis for OSA phenotyping. J Clin Sleep Med. 2022;18(7):1779–1788.
Keywords: obstructive sleep apnea, phenotypes, cluster analysis, multidimensional
BRIEF SUMMARY
Current Knowledge/Study Rationale: Obstructive sleep apnea (OSA) is a heterogeneous disease with varying phenotype. Because a clinical assessment and management approach for OSA should be based on multidimensional disease profiles, the objective of this study was to refine current phenotypic expressions of OSA with multidimensional disease characteristics.
Study Impact: Three OSA phenotypes were identified according to symptoms, anthropometry, polysomnography, and craniofacial morphology and differed in the prevalence of hypertension, metabolic syndrome, diurnal blood gas parameters, and continuous positive airway pressure titration response. Self-reported excessive daytime sleepiness was not related to the severity of sleep breathing disturbance, and craniofacial limitation was associated with the more severe phenotype. These findings highlight the necessity of integrating multiple disease characteristics into phenotyping to achieve a better understanding of the clinical features of OSA.
INTRODUCTION
Obstructive sleep apnea (OSA) is a heterogeneous disease, with variable disease pathways, clinical manifestation, and treatment response.1,2 Increasing evidence has shown that current widely used metrics, such as the overall apnea-hypopnea index (AHI), fail to reflect the complex heterogeneity of OSA, and the positive airway pressure (PAP)–initiated treatment pathway may not be the optimal option for all patients with OSA.3,4 Individual risk assessment and treatment response prediction require a better characterization of the disease’s complexity.
Cluster analysis has been used to identify phenotypes, or the subtypes of patients with unique characteristics, which may enable more personalized approaches to disease management. Similar to research on other chronic respiratory diseases, cluster analysis has recently been applied to OSA, leading to associations between different clusters and main clinical outcomes, although these studies may vary in terms of study population (eg, population or clinical cohorts), clinical features included for analysis (eg, symptoms or polysomnographic features), sample size, and main outcomes (eg, prevalence and incidence of major comorbidities, mortalities, or treatment response).5–9 Just like the management strategy for other chronic respiratory diseases, the clinical assessment and management approach for OSA should be based on multidimensional disease profiles, including self-reporting of symptoms, anatomic defect, and physiological impairment. However, the relatively limited clinical variables available for cluster analysis for most studies may limit the added value to patient phenotyping. Studies using unsupervised learning methods to identify patient subgroups based on multidimensional metrics among patients with OSA are quite few. In addition, studies have shown that Asian patients with OSA may be different from their Western counterparts in craniofacial morphology and the self-description of symptoms,10,11 which indicates that some clinical phenotypes with unique anatomical or symptomatic features may exist among Asian patients with OSA.12 In this study, we conducted a cluster analysis using multidimensional disease characteristics, including symptom, anthropometry, polysomnography (PSG), and craniofacial morphology indices in combination with PAP titration response and comorbidity profiles within a well-characterized cohort of Asian patients with OSA, with the aim of refining the current phenotypic delineation of OSA with clinical implications.
METHODS
Study population
Patients with a new diagnosis of moderate to severe OSA (AHI ≥ 15 events/h), referred for continuous positive airway pressure (CPAP) treatment to the sleep center of the National Respiratory Disease Research Center at the China-Japan Friendship Hospital were prospectively enrolled from January 2018 to December 2020. All patients underwent questionnaire assessment, PSG monitoring, anthropometry and comorbidities assessment, craniofacial morphologic evaluation, and auto-CPAP titration. Patients with other sleep-disordered breathing including central apnea and sleep-related hypoventilation, significant cardiopulmonary diseases such as chronic obstructive pulmonary disease or chronic heart failure with reduced ejection fraction, or a history of upper airway surgery were excluded from the study. Chronic obstructive pulmonary disease was diagnosed according to documented medical history or a postbronchodilator spirometry at enrollment. The study was approved by the Ethic Committee of the China-Japan Friendship Hospital, and written informed consent was obtained from all participants.
Symptoms, anthropometry, and comorbidity
Excessive daytime sleepiness (EDS) was assessed using the Epworth Sleepiness Scale (ESS); an ESS score ≥ 10 suggests EDS in the general population. Anthropometric indices including body mass index (BMI), neck circumference (NC), waist circumference, and hip circumference were collected, and the comorbid conditions of cardiovascular disease, hypertension, type 2 diabetes mellitus, and metabolic syndrome were examined based on medical history, biochemical profiles, and medications. Metabolic syndrome was diagnosed according to the guidelines for the prevention and treatment of type 2 diabetes mellitus in China (2020 edition).13
PSG and PAP titration
Full overnight PSG (Alice 6; Philips Respironics, Murrysville, PA) was conducted, with the patient breathing room air in a hospital setting. All data were scored by experienced PSG technologists based on the updated 2012 American Academy of Sleep Medicine criteria.14 Positional OSA was defined as patients having an AHI in the supine position that was more than twice that of patients in the nonsupine position according to Cartwright’s criteria.15 Arterial blood gas was collected the next morning after the sleep study.
An auto-CPAP titration was performed the following night after the PSG study in the hospital with the technicians to fit the mask for patients and attend to any problems with the PAP treatment. Auto-titrating machines DreamStation or System One devices, (Philips Respironics, Murrysville, PA) with heated humidification, a pressure range of 4–20 cm H2O, and a pressure-relief mode (A-flex 2 or 3) were used. Either a nasal mask (DreamWear or Wisp Nasal mask; Philips Respironics) or an oronasal mask (ComfortFull 2 or Amara; Philips Respironics) were used. The residual respiratory events index, the titration pressure at the minimum of eliminating obstructive respiratory events for 90% of the night, and the leakage were measured using the Encore software (Philips Respironics).
Craniofacial morphologic computed tomography scanning
All patients underwent a craniofacial and upper airway computed tomography scan using a 16-slice scanner (Light Speed; GE Medical Systems IL, USA). Scanning was performed at the end of the quiet tidal inspiration with the patient awake and supine and with a neutral head position as described in previous studies.16 Patients were instructed to refrain from movement or swallowing. The scans were acquired at a 0.5 mm collimation/interval and were reconstructed at a 1 mm thickness/interval, with 120 kV, 100 mA, and a rotation time of 0.5 seconds. Axial and sagittal image reconstructions were performed to allow for linear and area measurements using an Advantage workstation, version 4.5 (GE Health Care Chicago, IL). All measurements were performed by a single investigator who was blinded to the clinical status of the patients and was under the guidance of a specialist in orthodontics.
As illustrated in Figure 1, the following cephalometric variables were measured to map craniofacial skeletal and soft tissue morphology, as described in previous reports: cranial base length (distance between nasion and sella), cranial base angle (angle formed by nasion to the sella to the basion), face width (distance between the left tragion and right tragion), lower anterior face height (distance between anterior nasal spine and menton), total anterior face height (distance between nasion and menton), the angle formed by the sella to the nasion to point A, the angle formed by the sella to the nasion to point B, the angle formed by point A to the nasion to point B, the vertical distance from the hyoid bone to the mandibular plane, tongue length (the length from the anterosuperior point of the hyoid to the tip of the tongue), upper airway length (length from the hard palate to the epiglottis), the corrected airway length (the upper airway length/height), and the tongue area, which was obtained by tracing the contours of the tongue on the axial plane using the area measurement method.17,18
Figure 1. Sagittal computed tomography scan reconstruction of the upper airway.
A = point A, ANS = anterior nasal spine, B = point B, Ba = basion, EP = epiglottis, Gn = gnathion, Go = gonion, H = hyoid bone, Me = menton, MP = mandibular plane, N = nasion, PNS = posterior nasal spine, S = sella, TL = tongue length, TT = tip of the tongue, U = uvula.
Statistical analysis
The K-mean cluster analysis was used to cluster patients into groups based on anthropometry, ESS score, craniofacial morphology, and PSG profiles. The R package NbClust and clinical interpretability were used to select the number of clusters, and 3 homogeneous clusters were identified (Figure S1 (109.5KB, pdf) in the supplemental material).19 After the clusters were identified, differences among clusters regarding anthropometry, ESS, craniofacial morphology, PSG, PAP titration, and comorbidity profiles were examined via an analysis of variance or chi-square test, as appropriate. A binary logistic regression model was employed to determine the predictors of cluster membership between related clusters. Pearson correlation coefficients were used to determine the association of potential predictor variables with ESS score. Predictor variables with P < .10 according to the bivariate analysis were selected for multivariate regression analysis. All data were analyzed using R software (version 4.1.1, R Foundation for Statistical Computing, Vienna, Austria). A value of P < .05 was considered statistically significant.
RESULTS
Sample characteristics
A total of 365 patients were eligible for this study, among whom 41 patients either with a total sleep time of < 3 hours or < 15 minutes of rapid eye movement (REM) sleep or supine or nonsupine sleep on PSG and 33 patients who declined PAP therapy were excluded. Two hundred ninety-one patients were finally included for clustering analyses. The anthropometric features and disease severity of the excluded patients were comparable to those of the included patients (Table S1 (109.5KB, pdf) in the supplemental material). Participants for clustering analyses were predominantly male (82.5%) and overweight with a mean BMI of 28.41 ± 4.23 kg/m2. The mean age was 44.11 ± 11.62 years, the mean AHI was 44.32 ± 25.44 events/h, and the mean ESS score was 9.19 ± 4.85.
Clustering description
Three distinct clusters were finally identified, and their clinical features were described as follows (Table 1, Table 2, Table 3, and Figure 2). Cluster 1 corresponded to patients who were nonsleepy, of normal weight, and with moderate OSA. It was the most prevalent cluster, consisting of 42.6% of the study population, with an average age of 44.15 ± 11.61 years and a BMI of 25.57 ± 2.61 kg/m2. This group reported a low grade of EDS (mean ESS scores of 8.37 ± 4.42), and the anthropometric features, including NC, waist circumference, hip circumference, and waist-to-hip ratio were within the normal range of Chinese adults.20 The sleep structure was preserved, with the percentages of time spent in stage N3 sleep and REM sleep within the normal range. Cluster 1 patients showed moderate OSA with a mean AHI of 22.70 ± 12.12 events/h and mild nocturnal hypoxia with a percentage of total sleep time spent with oxyhemoglobin saturation below 90% (TST90) of 1.11 ± 2.05%. The sleep-disordered breathing in this cluster of patients was characterized by a high prevalence of positional OSA (68%) and a preponderance of hypopnea events (72%).
Table 1.
Demographic and comorbidity characteristics of OSA clusters.
| Parameters | Cluster 1 (n = 124) | Cluster 2 (n = 105) | Cluster 3 (n = 62) | P Value | Bonferroni Correction |
|---|---|---|---|---|---|
| Age (y) | 44.15 ± 11.61 | 44.10 ± 11.25 | 41.08 ± 11.21 | .176 | — |
| Sex (male, %) | 94, 75.81% | 90, 85.71% | 56, 90.32% | .058 | — |
| BMI (kg/m2) | 25.57 ± 2.61 | 30.33 ± 3.99 | 30.96 ± 3.86 | < .001 | 1 < 2, 3 |
| Neck circumference (cm) | 38.05 ± 3.13 | 41.15 ± 2.75 | 42.89 ± 2.55 | < .001 | 1 < 2 < 3 |
| Waist circumference (cm) | 87.67 ± 7.23 | 99.91 ± 9.73 | 102.55 ± 8.97 | < .001 | 1 < 2, 3 |
| Hip circumference (cm) | 96.05 ± 5.46 | 105.57 ± 8.52 | 106.17 ± 8.06 | < .001 | 1 < 2, 3 |
| WHR | 0.91 ± 0.07 | 0.94 ± 0.11 | 0.97 ± 0.04 | .003 | 1 < 3 |
| ESS | 8.37 ± 4.42 | 8.15 ± 4.30 | 12.58 ± 5.10 | < .001 | 1, 2 < 3 |
| SBP (mm Hg) | 128.83 ± 17.134 | 133.86 ± 17.22 | 133.29 ± 14.91 | .056 | — |
| DBP (mm Hg) | 81.62 ± 9.54 | 84.64 ± 13.28 | 86.32 ± 10.70 | .018 | 1 < 3 |
| PaO2 (mm Hg) | 97.40 ± 12.56 | 93.70 ± 11.41 | 90.20 ± 10.83 | .001 | 1 > 3 |
| PaCO2 (mm Hg) | 41.31 ± 5.20 | 41.57 ± 4.63 | 43.54 ± 5.28 | .016 | 1, 2 < 3 |
| CVD (n, %) | 6, 4.84% | 10, 9.52% | 5, 8.06% | .117 | — |
| MS (n, %) | 36, 29.03% | 58, 55.24% | 38, 61.29% | < .001 | — |
| T2DM (n, %) | 8, 6.45% | 9, 8.57% | 6, 9.68% | .715 | — |
| Systolic hypertension (n, %) | 45, 36.29% | 54, 51.43% | 38, 61.29% | .008 | — |
| Diastolic hypertension (n, %) | 36, 29.03% | 48, 45.71% | 27, 43.55% | .017 | — |
BMI = body mass index, CVD = cardiovascular disease, DBP = diastolic blood pressure, ESS = Epworth Sleepiness Scale, MS = metabolic syndrome, OSA = obstructive sleep apnea, PaCO2 = partial pressure of carbon dioxide in arterial blood, PaO2 = partial pressure of oxygen in arterial blood, SBP = systolic blood pressure, T2DM = type 2 diabetes mellitus, WHR = waist-to-hip ratio.
Table 2.
PSG and titration profiles by cluster.
| Parameters | Cluster 1 (n = 124) | Cluster 2 (n = 105) | Cluster 3 (n = 62) | P Value | Bonferroni Correction |
|---|---|---|---|---|---|
| Total sleep time (min) | 398.66 ± 61.89 | 420.83 ± 71.37 | 417.56 ± 77.67 | .037 | 1 < 2 |
| Sleep efficiency (%) | 82.17 ± 10.32 | 84.06 ± 10.36 | 83.63 ± 9.64 | .347 | — |
| Sleep latency (min) | 16.01 ± 26.81 | 13.54 ± 17.22 | 11.34 ± 15.17 | .265 | — |
| N1% (%) | 16.34 ± 10.00 | 23.70 ± 15.33 | 36.15 ± 19.44 | < .001 | 1 < 2 < 3 |
| N2% (%) | 45.57 ± 10.89 | 42.25 ± 12.11 | 40.86 ± 17.34 | .036 | — |
| N3% (%) | 18.54 ± 10.06 | 14.54 ± 8.90 | 4.20 ± 6.30 | < .001 | 1 > 2 > 3 |
| R% (%) | 19.13 ± 6.44 | 19.61 ± 6.88 | 18.59 ± 6.14 | .619 | NA |
| Total arousal index (events/h) | 32.15 ± 11.23 | 47.72 ± 13.80 | 64.86 ± 16.60 | < .001 | 1 < 2 < 3 |
| Arousal index of REM sleep (events/h) | 40.51 ± 15.30 | 49.00 ± 14.60 | 60.01 ± 15.93 | < .001 | 1 < 2 < 3 |
| Arousal index of NREM sleep (events/h) | 29.75 ± 12.24 | 47.00 ± 15.34 | 66.52 ± 17.19 | < .001 | 1 < 2 < 3 |
| AHI (events/h) | 22.70 ± 12.12 | 50.48 ± 16.68 | 76.69 ± 14.83 | < .001 | 1 < 2 < 3 |
| AI (events/h) | 7.38 ± 8.43 | 29.84 ± 18.67 | 66.84 ± 18.11 | < .001 | 1 < 2 < 3 |
| HI (events/h) | 15.27 ± 8.81 | 20.44 ± 13.23 | 9.36 ± 10.69 | < .001 | 3 < 1 < 2 |
| Hypopnea (%) | 72.07 ± 22.03 | 44.73 ± 26.26 | 12.58 ± 15.21 | < .001 | 1 > 2 > 3 |
| AHIREM (events/h) | 28.40 ± 16.56 | 52.42 ± 21.05 | 66.32 ± 18.38 | < .001 | 1 < 2 < 3 |
| AHINREM (events/h) | 21.25 ± 12.58 | 49.59 ± 18.28 | 78.41 ± 15.02 | < .001 | 1 < 2 < 3 |
| AHIREM/AHINREM | 1.62 ± 1.16 | 1.18 ± 0.60 | 0.84 ± 0.18 | < .001 | 1 > 2 > 3 |
| AHIsupine/AHInonsupine | 4.47 ± 4.93 | 2.68 ± 2.57 | 1.16 ± 0.39 | < .001 | 1 > 2, 3 |
| Positional OSA (n, %) | 84, 67.74% | 49, 46.67% | 3, 4.84% | < .001 | — |
| Mean duration of apnea (s) | 20.83 ± 7.19 | 24.86 ± 7.51 | 32.33 ± 7.17 | < .001 | 1 < 2 < 3 |
| Mean duration of hypopnea (s) | 28.73 ± 9.39 | 28.66 ± 7.57 | 26.59 ± 11.20 | .281 | — |
| ODI (events/h) | 15.29 ± 11.59 | 45.73 ± 18.28 | 78.78 ± 14.98 | < .001 | 1 < 2 < 3 |
| SpO2 base, % | 96.19 ± 1.19 | 95.75 ± 1.26 | 94.53 ± 2.10 | < .001 | 1, 2 > 3 |
| SpO2 mean, % | 95.89 ± 1.23 | 94.32 ± 1.47 | 88.98 ± 3.47 | < .001 | 1 > 2 > 3 |
| SpO2 nadir, % | 85.58 ± 5.44 | 75.41 ± 9.31 | 57.96 ± 10.72 | < .001 | 1 > 2 > 3 |
| TST90 (%) | 1.11 ± 2.05 | 8.13 ± 8.51 | 42.24 ± 17.61 | < .001 | 1 < 2 < 3 |
| P90 (cm H2O) | 6.84 ± 4.74 | 10.20 ± 4.20 | 11.92 ± 4.38 | < .001 | 1 < 2, 3 |
| Mean CPAP pressure (cm H2O) | 5.88 ± 3.71 | 8.03 ± 3.42 | 9.52 ± 3.66 | < .001 | 1 < 2, 3 |
| Residual AHI (events/h) | 4.26 ± 3.68 | 6.69 ± 7.02 | 10.69 ± 11.2 | < .001 | 1, 2 < 3 |
| Mask type (nasal/oronasal) | 108/16 | 90/15 | 48/14 | .209 | — |
| Usage time (min) | 362.23 ± 59.18 | 348.45 ± 71.27 | 327.16 ± 46.67 | .001 | 1, 2 > 3 |
| Large leak (% of night) | 0.52 ± 0.24 | 0.71 ± 0.33 | 4.37 ± 0.80 | < .001 | 1, 2 < 3 |
AHI = apnea-hypopnea index, AHIREM = apnea-hypopnea index of rapid eye movement sleep, AHINREM = apnea-hypopnea index of non-rapid eye movement sleep, AHIsupine = apnea-hypopnea index of supine position, AHInonsupine = apnea-hypopnea index of non-supine position, AI = apnea index, CPAP = continuous positive airway pressure, HI = hypopnea index, N1% = percentage of N1 time of total sleep time, N2% = percentage of N2 time of total sleep time, N3% = percentage of N3 time of total sleep time, NREM = nonrapid eye movement, ODI = oxygen desaturation index, OSA = obstructive sleep apnea, P90 = the titration pressure at the minimum of eliminating obstructive respiratory events for 90% of the night, PSG = polysomnography; R% = percentage of rapid eye movement sleep time of total sleep time, REM = rapid eye movement, SpO2 = blood oxygen saturation, TST90 = percentage of total sleep time spent with oxyhemoglobin saturation below 90%.
Table 3.
Craniofacial morphology by cluster.
| Parameters | Cluster 1 (n = 124) | Cluster 2 (n = 105) | Cluster 3 (n = 62) | P Value | Bonferroni Correction |
|---|---|---|---|---|---|
| Face width (cm) | 15.04 ± 2.57 | 14.35 ± 4.50 | 15.54 ± 2.23 | .071 | — |
| NS (mm) | 66.33 ± 5.78 | 65.21 ± 5.88 | 64.97 ± 6.36 | .224 | — |
| NSBa (°) | 128.82 ± 27.93 | 130.07 ± 15.21 | 129.22 ± 19.13 | .913 | — |
| SNA (°) | 82.28 ± 4.81 | 81.74 ± 5.31 | 82.41 ± 4.89 | .627 | — |
| SNB (°) | 78.14 ± 5.98 | 78.22 ± 6.16 | 78.71 ± 6.19 | .825 | — |
| ANB (°) | 4.11 ± 6.29 | 3.78 ± 2.47 | 3.85 ± 2.64 | .565 | — |
| UAL (mm) | 66.44 ± 10.67 | 66.93 ± 10.79 | 74.05 ± 11.12 | < .001 | 1, 2 < 3 |
| Adjusted UAL (mm/m) | 38.78 ± 5.66 | 38.72 ± 5.85 | 42.75 ± 6.38 | < .001 | 1, 2 < 3 |
| LAFH (mm) | 72.81 ± 7.09 | 74.53 ± 5.56 | 74.01 ± 5.95 | .113 | — |
| TAFH (cm) | 11.56 ± 1.59 | 11.67 ± 1.88 | 11.52 ± 1.83 | .838 | — |
| LAFH/TAFH (%) | 63.84 ± 3.74 | 64.11 ± 4.02 | 64.30 ± 3.82 | .723 | — |
| MPH (mm) | 14.87 ± 7.55 | 18.17 ± 8.04 | 24.50 ± 7.72 | < .001 | 1 < 2 < 3 |
| TL (cm) | 71.51 ± 7.46 | 75.65 ± 7.78 | 77.80 ± 6.88 | < .001 | 1 < 2, 3 |
| TA (cm2) | 32.31 ± 4.75 | 35.33 ± 4.65 | 36.63 ± 5.54 | < .001 | 1 < 2, 3 |
ANB = angle measurement from point A to the nasion to point B, LAFH = lower anterior face height, MPH = distance from the hyoid to the mandibular plane, NS = cranial base length, NSBa = cranial base angle, SNA = angle measurement from the sella to the nasion to point A, SNB = angle measurement from the sella to the nasion to point B, TA = tongue area, TAFH = total anterior face height, TL = tongue length, UAL, upper airway length.
Figure 2. Probability of multidimensional disease characteristics, including symptoms, anthropometry, PSG, and craniofacial morphology within each cluster.
AHI = apnea-hypopnea index, AI = apnea index, ANB = angle measurement from point A to the nasion to point B, BMI = body mass index, ESS = Epworth Sleepiness Scale, HC = hip circumference, HI = hypopnea index, LAFH = lower anterior face height, MPH = distance from the hyoid to the mandibular plane, N1% = percentage of N1 time of total sleep time, N2% = percentage of N2 time of total sleep time, N3% = percentage of N3 time of total sleep time, NC = neck circumference, NREM = non–rapid eye movement, ODI = oxygen desaturation index, PSG = polysomnography, R% = percentage of rapid eye movement sleep time of total sleep time, REM = rapid eye movement, SE = sleep efficiency, SL = sleep latency, SNA = angle measurement from the sella to the nasion to point A, SNB = angle measurement from the sella to the nasion to point B, SpO2 = blood oxygen saturation, TA = tongue area, TAFH = total anterior face height, T-AHI = total apnea-hypopnea index, TL = tongue length, TST = total sleep time, TST90 = percentage of total sleep time spent with oxyhemoglobin saturation below 90%, UAL = upper airway length, WC = waist circumference.
Cluster 2 corresponded to patients who were nonsleepy and obese with severe OSA. There were 95 patients (36.1%) included in this group, with an average age of 44.10 ± 11.25 years and a BMI of 30.33 ± 3.99 kg/m2. Patients had increased NC, waist circumference, and hip circumference and reported a low grade of EDS (ESS scores of 8.15 ± 4.30). Cluster 2 patients showed severe OSA with a mean AHI of 50.48 ± 16.68 events/h and moderate nocturnal hypoxia with a TST90 of 8.13 ± 8.51%. The ratio of positional vs nonpositional OSA was 49/56, with a nearly equal proportion of apnea and hypopnea events.
Cluster 3 corresponded to patients who were sleepy and obese with very severe OSA with craniofacial limitation. There were 62 patients (21.3%) included in this group, with an average age of 41.08 ± 11.21 years and a BMI of 30.96 ± 3.86 kg/m2. The general anthropometric features were comparable to patients in cluster 2 except for NC, which was larger in cluster 3. Patients reported a high grade of EDS (ESS scores of 12.58 ± 5.10). The sleep structure was significantly disturbed, with a decreased proportion of stage N3 sleep and frequent arousal. Cluster 3 patients showed very severe OSA with a mean AHI of 76.69 ± 14.83 events/h and severe nocturnal hypoxia with a TST90 of 42.24 ± 17.61%. The sleep-disordered breathing in this cluster of patients was characterized by a low prevalence of positional OSA (5%) and a preponderance of apnea events (87%). Compared with patients in cluster 1 and cluster 2, patients in cluster 3 showed significant craniofacial limitation, with lower placement of the hyoid bone (larger distance from the hyoid to the mandibular plane), increased length of the upper airway, and larger length and area of the tongue.
The binary logistic regression model was employed to detect predictors of cluster membership between clusters of corresponding BMI and different OSA severity (cluster 2 vs cluster 3). Increased ESS and NC, adjusted upper airway length, and distance from the hyoid to the mandibular plane were associated independently with the more-severe disease phenotypes (Table 4).
Table 4.
Predictors of phenotype membership.
| Parameters | Univariate Analysis | Multivariate Analysis | ||||
|---|---|---|---|---|---|---|
| OR | 95% CI | P Value | OR | 95% CI | P Value | |
| Sex | 5.455 | 1.210–24.596 | .027 | 4.364 | 0.199–19.913 | .350 |
| Age | 0.976 | 0.948–1.005 | .103 | 0.947 | 0.897–0.986 | .078 |
| BMI | 1.041 | 0.962–1.128 | .317 | |||
| Neck circumference | 1.281 | 1.126–1.458 | < .001 | 1.258 | 1.058–1.497 | .010 |
| Waist circumference | 1.029 | 0.996–1.064 | .088 | 1.003 | 0.933–1.077 | .943 |
| Hip circumference | 1.009 | 0.972–1.048 | .641 | |||
| WHR | 6.476 | 1.379–30.419 | .039 | 14.412 | 0.245–85.103 | .087 |
| ESS | 1.227 | 1.132–1.329 | < .001 | 1.302 | 1.176–1.441 | < .001 |
| PaO2 | 0.973 | 0.944–1.003 | .076 | 0.977 | 0.934–1.022 | .319 |
| PaCO2 | 1.084 | 1.013–1.161 | .021 | 1.072 | 0.966–1.190 | .188 |
| NS | 0.969 | 0.919–1.021 | .238 | |||
| NSBa | 0.994 | 0.975–1.012 | .495 | |||
| LAFH | 0.986 | 0.933–1.043 | .627 | |||
| TAFH | 0.809 | 0.547–1.195 | .287 | |||
| LAFH/TAFH | 1.034 | 0.951–1.124 | .432 | |||
| Face width | 1.118 | 0.988–1.265 | .076 | 1.117 | 0.935–1.333 | .223 |
| SNA | 0.988 | 0.929–1.051 | .697 | |||
| SNB | 1.007 | 0.957–1.060 | .788 | |||
| ANB | 0.977 | 0.861–1.108 | .714 | |||
| MPH | 1.109 | 1.058–1.161 | < .001 | 1.094 | 1.030–1.162 | .003 |
| TL | 1.043 | 0.996–1.092 | .072 | 0.990 | 0.916–1.071 | .809 |
| TA | 1.056 | 0.991–1.126 | .095 | 1.101 | 0.972–1.248 | .128 |
| UAL | 1.061 | 1.029–1.094 | < .001 | 0.844 | 0.678–1.051 | .129 |
| UAL/height | 1.115 | 1.054–1.179 | < .001 | 1.108 | 1.023–1.201 | .012 |
ANB = angle measurement from point A to the nasion to point B, BMI = body mass index, CI = confidence interval, ESS = Epworth Sleepiness Scale, Go-Gn-H = angle formed by gonion-gnathion-hyoid bone, LAFH = lower anterior face height, MPH = distance from the hyoid to the mandibular plane, NS = cranial base length, NSBa = cranial base angle, OR = odds ratio, PaCO2 = partial pressure of carbon dioxide in arterial blood, PaO2 = partial pressure of oxygen in arterial blood, SNA = angle measurement from the sella to the nasion to point A, SNB = angle measurement from the sella to the nasion to point B, TA = tongue area, TAFH = total anterior face height, TL = tongue length, UAL = upper airway length, WHR = waist-to-hip ratio.
Correlations of ESS and PSG parameters
The ESS score was associated with the percentage of time spent in stage N1 sleep, percentage of time spent in stage N3 sleep, total arousal index, AHI, oxygen desaturation index, blood oxygen saturation (SpO2) mean, SpO2 nadir, and TST90 according to unadjusted analysis. In multiple regression analysis, with the ESS score as the dependent variable and the percentage of time spent in stage N1 sleep, percentage of time spent in stage N3 sleep, total arousal index, AHI, oxygen desaturation index, SpO2 mean, SpO2 nadir, and TST90 as independent variables, only the percentage of time spent in stage N3 sleep and the SpO2 mean were found to be independently associated with the AHI (F = 23.912; R2 = .136; P < .001; Table 5).
Table 5.
Associations of ESS score and PSG parameters.
| Parameters | Univariate Analysis | Multivariate Analysis | ||
|---|---|---|---|---|
| r | P Value | Coefficient | P Value | |
| Total sleep time | 0.057 | 0.332 | ||
| Sleep efficiency | 0.024 | 0.680 | ||
| N1% | 0.158 | 0.007 | 0.039 | 0.482 |
| N3% | −0.288 | < 0.001 | −0.081 | < 0.001 |
| R% | −0.020 | 0.735 | ||
| Total arousal index | 0.198 | 0.001 | 0.035 | 0.518 |
| AHI | 0.237 | < 0.001 | 0.023 | 0.868 |
| ODI | 0.243 | < 0.001 | 0.028 | 0.775 |
| SpO2mean | −0.343 | < 0.001 | −0.395 | 0.004 |
| SpO2nadir | −0.282 | < 0.001 | −0.002 | 0.982 |
| TST90 | 0.335 | < 0.001 | 0.029 | 0.734 |
AHI = apnea-hypopnea index, ESS = Epworth Sleepiness Scale, N1% = percentage of N1 time of total sleep time, N3% = percentage of N3 time of total sleep time, ODI = oxygen desaturation index, PSG = polysomnography, R% = percentage of rapid eye movement sleep time of total sleep time, SpO2 = blood oxygen saturation, TST90 = percentage of total sleep time spent with oxyhemoglobin saturation below 90%.
Auto-CPAP titration response according to clusters
The level of titration pressure at the minimum of eliminating obstructive respiratory events for 90% of the night and mean CPAP pressure were lower in cluster 1 patients than in cluster 2 and cluster 3 patients. Although the auto-PAP titration pressure was comparable between cluster 2 and cluster 3 patients, the residual AHI and leakage were higher and PAP usage time was shorter in cluster 3 patients than those in cluster 1 and cluster 2 (Table 2).
Diurnal blood gas and comorbidities
The 3 clusters of patients showed differences in diurnal blood gas parameters, with cluster 3 patients showing a lower partial pressure of oxygen in arterial blood and higher partial pressure of carbon dioxide in arterial blood. The prevalence of systolic hypertension and metabolic syndrome was significantly different among patients in the 3 clusters, with the highest rates in cluster 3 patients. The prevalence of cardiovascular disease and diabetes mellitus was similar among patients in the 3 clusters (Table 1).
DISCUSSION
Accumulating evidence indicates that OSA is a heterogeneous and multifaceted disorder, and attempts to phenotype and cluster patients with OSA using symptoms, comorbidities, or PSG profiles have been undertaken. However, most of these studies have been based primarily on 1-dimensional data. In the present study, using multidimensional data and unsupervised analytic methods, we identified 3 patient phenotypes: normal weight, nonsleepy, and moderate OSA; obese, nonsleepy, and severe OSA; and obese, sleepy, and very severe OSA with craniofacial limitation. NC and craniofacial limitation were associated with the more severe phenotype, in which patients had a higher prevalence of hypertension and metabolic syndrome, greater diurnal blood gas abnormalities, and worse PAP titration response. ESS score was not related with the severity of sleep breathing disturbance.
Symptom assessment is critical to disease identification and optimal care strategy development. Frequently reported symptoms, such as EDS, nonrestorative sleep, and insomnia, vary in severity in patients with OSA. Ye et al6 identified 3 distinct subgroups with moderate to severe OSA in the Icelandic Sleep Apnea Cohort: the excessively sleepy subgroup, the disturbed sleep subgroup, and the minimally symptomatic subgroup, using a combination of symptoms and comorbidities. Several studies also reported similar symptom phenotypes.5,11,21 Unlike the above studies using detailed symptom description for cluster classification, the ESS score was adopted for symptom evaluation in the current study because EDS is the cardinal symptom for seeking medical advice for most middle-aged patients with OSA and the ESS is a simple and commonly used assessment tool for EDS. We found that the mean ESS score was 9.19 for the current cohort, which was close to that of several other studies5,11,22,23; however, this symptom burden was lower compared with that of the Icelandic Sleep Apnea Cohort clinical cohort (ESS score approximately 11.7 and 42.3% of the cohort with EDS).6 The discrepancy in symptom scores between cohorts may result from differences in demographic or disease severity characteristics of the cohorts; furthermore, different parameters used to categorize the cohort may be another explanation. In line with previous studies,24 no correlation between ESS score and AHI, oxygen desaturation index, TST90, lowest SpO2, and arousal index was proven, whereas the percentage of time spent in stage N3 sleep and the mean SpO2 were found to be associated with ESS. However, they only explain a small fraction of the variation in the ESS score (R2 = .136). This finding may be because that symptom’s perception, description, and report are influenced by many nonphysiological factors, such as sex, culture, education, and socioeconomic status. Therefore, just like chronic obstructive pulmonary disease assessment tests or the modified Medical Research Council dyspnea scale for symptom grading for chronic obstructive pulmonary disease,25 a concise, stable, and widely validated symptom scale tool is needed for OSA phenotype and therapeutic assessment. In addition, although symptoms are important in defining a patient’s perception of disease and developing clinical treatments, patients can also misclassify or misreport their symptoms; therefore, symptom assessment should be combined with other physiological parameters to better categorize disease phenotype.
Although OSA severity is widely classified using the AHI, this objective index represents a small fraction of the physiological characteristics observed by PSG. In the current study, incorporating sleep architecture disturbance, arousal, breathing disturbance, and hypoxia, we identified 3 patient clusters with distinguishing PSG features. The PSG profile in the “moderate” cluster of patients was characterized by hypopnea event–predominant and more positionally related breathing disturbance with normal sleep architecture and mild nocturnal hypoxia. The PSG profile in the “severe” cluster of patients was characterized by a nearly equal proportion of event type (apnea or hypopnea) and event distribution in sleep stage (REM sleep vs non-REM sleep) or body position (supine or nonsupine), with moderate nocturnal hypoxia. The PSG profile in the “very severe” cluster of patients was characterized by greatly decreased slow-wave sleep, severe sleep fragmentation, and apnea-predominant and nonpositionally related breathing disturbance with very severe nocturnal hypoxia.
Several studies have used unsupervised learning methods to identify patient subgroups based on polysomnographic metrics so far.26–31 Lacedonia et al26 identified 3 clusters among 198 patients with OSA from an Italian sleep center. The breathing disturbance and hypoxia characteristics of the 3 clusters were similar to those in our study; however, because of the limited data provided by polygraphy, their study did not describe detailed sleep profiles of the 3 clusters. Joosten, et al27 identified 4 clinical groups REM-predominant OSA, non-REM predominant OSA, supine predominant OSA and intermittent OSA—among 1,064 patients with mild to moderate OSA mainly based on breathing disturbance metrics. Using comprehensive and well-defined PSG parameters, Zinchuk et al8 described 7 distinguishing polysomnographic features— mild, periodic limb movements of sleep, non-REM sleep and arousal, REM sleep and hypoxia, hypopnea and hypoxia, arousal and poor sleep and combined severe—from a cohort of veterans. incorporating type, duration, and positional effect of respiratory event, minimum oxygen saturation, arousal rate before termination, Joosten, Landry, et al28 identified 3 clusters (high fraction of apnea and severe desaturation, high fraction of apnea and long event duration, low fraction of apnea) in 210 male patients with moderate to severe OSA. The discrepancy among these studies mainly lies in the differences in the characteristics of study population, the metrics used for analysis, and statistical methods. Sleep study, especially PSG, collects massive amounts of physiological information, but the clinical significance of these metrics has not been clarified and most of them have not been utilized in a clinical setting. How to harness sleep information and extract the potentially more clinically relevant metrics for personalized clinical decisions warrants further investigation.
The anatomical relationship of the craniofacial skeletal and soft tissue structures of the upper airway determines its patency. In the current study, we found that the craniofacial and upper airway morphologic measurements were different among the 3 clusters of patients, with patients in the very severe cluster showing significant inferior displacement of the hyoid bone, increased length of the upper airway, and larger tongue size, compared with patients in the other 2 clusters. For patients with similar general anthropometric features (cluster 2 and cluster 3), greater upper airway length and distance from the hyoid to the mandibular plane were associated independently with the more-severe disease phenotypes. A low hyoid bone results in a reduction in pharyngeal airspace, and increased upper airway length is associated with a greater susceptibility to airway narrowing/collapse. Using lateral cephalogram, Kim et al32 identified 3 clusters in 421 Korean patients with OSA, with the clinical characteristics of 2 clusters akin to our analysis, although a 2-dimensional static image analysis may not be matched with real functional disorder in some patients. Physical examination or anatomy assessment is critical to early disease detection and to management and long-term follow-up or adherence to different forms of therapy. For patients with both skeletal discrepancies and obesity as anatomic risks for upper airway collapsibility, craniofacial skeletal intervention combined with body weight reduction can be considered as the preferred treatment, especially for those with poor acceptance or adherence to PAP. For patients with OSA, craniofacial and upper airway assessment may contain valid cues for multidisciplinary disease management strategies and should be listed as a clinical routine.
OSA is a well-known independent risk factor for cardiovascular disease, metabolic disease, and mortality; however, the associations of disease phenotype with important clinical outcomes remain largely uncertain.4 Several attempts to phenotype and cluster patients with OSA using various combinations of symptoms and comorbidities have been undertaken. The Icelandic Sleep Apnea Cohort research reported that the association of OSA and comorbidities was more frequent in the minimally symptomatic group, including cardiovascular and respiratory comorbidities, among patients with a similar AHI or degree of obesity.6 The European Sleep Apnea Database study analyzed similar OSA phenotypes—ie, a combination of diurnal and nocturnal symptoms—and reported similar results, with an additional finding of a high prevalence of psychiatric comorbidities, especially in patients with OSA with self-reported insomnia.5 In the current study, we found that the EDS group was associated with increased comorbidities of systolic hypertension, metabolic syndrome, and greater diurnal blood gas abnormalities, whereas comorbid cardiovascular disease and type 2 diabetes mellitus were comparable among the 3 patient clusters. It is probable that the lack of consistence with previous studies lies in the different goals set for each. The Icelandic Sleep Apnea Cohort and European Sleep Apnea Database studies focused on describing symptomatic phenotypes based on detailed symptomatic questionnaires, whereas our study mapped the clusters according to multiple domain metrics. Different study designs may reveal different underlying phenotypes, and some disease characteristics should be interpreted under specific phenotyping scenarios.
CPAP is recommended as a first-line treatment for moderate to severe OSA, but the low adherence to PAP treatment is a challenge for many patients with OSA. The unintentional air leakage and its related symptoms are reported to be the most troublesome adverse effects of PAP therapy and possible causes of poor adherence. The determining factors of unintentional leakage can be associated with patient-related factors, such as facial or pharyngeal anatomy, age, BMI, and concomitant comorbidities.33 In the current study, the leakage and residual AHI were higher and the PAP usage time was shorter in cluster 3 than cluster 2 patients, whereas BMI, age, and PAP titration pressure were comparable between the 2 groups. Goh et al reported that craniofacial morphology may be associated with adherence to different mask interfaces in Asian patients with OSA.34 Further studies are needed to confirm the influence of craniofacial phenotyping on treatment response and adherence for patients with different PAP interfaces.
Several limitations of this study warrant consideration. First, data from patients of predominantly middle-aged Chinese men with moderate-to-severe OSA referral to a single medical center were collected, which indicates a lack of sex, ethnic, disease severity, and demographic diversity. Validation of our findings in a wider variety of populations is needed. However, this dearth of diversity can also be a strength because the cohort represents a homogenous population with similar craniofacial morphology. Second, the current study was cross-sectional, and some important clinical outcomes were not included in this study. Follow-up studies are needed to examine whether the response to treatment differs among patients with OSA with distinct patterns of clinical presentation, particularly in terms of changes in symptom presentation and cardiovascular and metabolic comorbidities. Third, the ESS was used to assess daytime sleepiness, and some other symptoms that may be common in patients with OSA, such as insomnia and fatigue, were not collected in this study, although we used PSG to measure sleep latency, quality, and architecture. Note that self-reported symptom perception and objective physiologic measurement may sometimes not be consistent.35 Future phenotyping investigations need to identify and quantify symptoms that reflect pathophysiologic abnormalities and are related with disease prognosis. Fourth, 20% of our patients were excluded from clustering analysis because of inadequate total sleep time, sleep duration of a specific stage or position on PSG, or declination to PAP titration, which may increase the risk of bias, although their baseline characteristics and disease severity were comparable to those included for analysis. Finally, as with other cluster analyses, the results are related closely to the type and number of parameters chosen, which means that the clustering results can be different in the number and type of selected variables changed, even in the same population. Thus, our results suggest only a possible method for categorizing patients with OSA.
CONCLUSIONS
We identified 3 OSA phenotypes according to multiple aspects of clinical features, including anthropometry, symptoms, craniofacial morphology, and PSG profiles in patients with moderate to severe OSA. These groups were found to differ in the prevalence of hypertension, metabolic syndrome, diurnal blood gas parameters, and PAP titration response. EDS was not related with the severity of sleep breathing disturbance, and craniofacial limitation was associated with the more severe phenotype. These findings highlight the importance of integrating multidimensional datasets into OSA phenotyping, which may help clinicians and researchers better understand the clinical pictures of OSA and ultimately develop optimal care strategies customized for each subgroup.
DISCLOSURE STATEMENT
All authors have seen and approved the manuscript. Work for this study was performed at the Department of Pulmonary and Critical Care Medicine, Center of Respiratory Medicine, China-Japan Friendship Hospital, National Clinical Research Center for Respiratory Diseases, Beijing, China. This study was funded by research grants from the Clinical and Translational Medicine Research Project (2020-I2M-C&T-B-094) from the Chinese Academy of Medical Sciences. The authors report no conflicts of interest.
ACKNOWLEDGMENTS
The authors thank Dr. L.L. Ma in the Stomatology Department, China-Japan Friendship Hospital, Beijing, China, for her assistance and guidance in craniofacial measurement.
ABBREVIATIONS
- AHI
apnea-hypopnea index
- BMI
body mass index
- CPAP
continuous positive airway pressure
- EDS
excessive daytime sleepiness
- ESS
Epworth Sleepiness Scale
- NC
neck circumference
- OSA
obstructive sleep apnea
- PAP
positive airway pressure
- PSG
polysomnography
- REM
rapid eye movement
- SpO2
blood oxygen saturation
- TST90
percentage of total sleep time spent with oxyhemoglobin saturation below 90%
REFERENCES
- 1. Edwards BA , Redline S , Sands SA , Owens RL . More than the sum of the respiratory events: personalized medicine approaches for obstructive sleep apnea . Am J Respir Crit Care Med. 2019. ; 200 ( 6 ): 691 – 703 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Mansfield DR . Obstructive sleep apnoea phenotypes: the many faces of a public health monolith . Respirology. 2020. ; 26 ( 4 ): 294 – 295 . [DOI] [PubMed] [Google Scholar]
- 3. Malhotra A , Ayappa I , Ayas N , et al . Metrics of sleep apnea severity: beyond the apnea-hypopnea index . Sleep. 2021. ; 44 ( 7 ): zsab030 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Zinchuk A , Yaggi HK . Phenotypic subtypes of OSA: a challenge and opportunity for precision medicine . Chest. 2020. ; 157 ( 2 ): 403 – 420 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Bailly S , Grote L , Hedner J , et al. ; ESADA Study Group . Clusters of sleep apnoea phenotypes: a large pan-European study from the European Sleep Apnoea Database (ESADA) . Respirology. 2021. ; 26 ( 4 ): 378 – 387 . [DOI] [PubMed] [Google Scholar]
- 6. Ye L , Pien GW , Ratcliffe SJ , et al . The different clinical faces of obstructive sleep apnoea: a cluster analysis . Eur Respir J. 2014. ; 44 ( 6 ): 1600 – 1607 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Pien GW , Ye L , Keenan BT , et al . Changing faces of obstructive sleep apnea: treatment effects by cluster designation in the Icelandic Sleep Apnea Cohort . Sleep. 2018. ; 41 ( 3 ): zsx201 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Zinchuk AV , Jeon S , Koo BB , et al . Polysomnographic phenotypes and their cardiovascular implications in obstructive sleep apnoea . Thorax. 2018. ; 73 ( 5 ): 472 – 480 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Pépin JL , Bailly S , Rinder P , et al. ; medXcloud Group . CPAP therapy termination rates by OSA phenotype: a French nationwide database analysis . J Clin Med. 2021. ; 10 ( 5 ): 936 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Schorr F , Kayamori F , Hirata RP , et al . Different craniofacial characteristics predict upper airway collapsibility in Japanese-Brazilian and white men . Chest. 2016. ; 149 ( 3 ): 737 – 746 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Kim J , Keenan BT , Lim DC , Lee SK , Pack AI , Shin C . Symptom-based subgroups of Koreans with obstructive sleep apnea . J Clin Sleep Med. 2018. ; 14 ( 3 ): 437 – 443 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Ida H , Suga T , Nishimura M , et al . Unique clinical phenotypes of patients with obstructive sleep apnea in a Japanese population: a cluster analysis . J Clin Sleep Med. 2022. ; 18 ( 3 ): 895 – 902 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Chinese Diabetes Society . Guideline for the prevention and treatment of type 2 diabetes mellitus in China (2020 edition) . Chin J Diabetes Mellitus. 2021. ; 13 ( 4 ): 315 – 409 . [Google Scholar]
- 14. Berry RB , Budhiraja R , Gottlieb DJ , et al. ; Deliberations of the Sleep Apnea Definitions Task Force of the American Academy of Sleep Medicine . Rules for scoring respiratory events in sleep: update of the 2007 AASM Manual for the Scoring of Sleep and Associated Events . J Clin Sleep Med. 2012. ; 8 ( 5 ): 597 – 619 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Cartwright RD . Effect of sleep position on sleep apnea severity . Sleep. 1984. ; 7 ( 2 ): 110 – 114 . [DOI] [PubMed] [Google Scholar]
- 16. Schwab RJ , Gupta KB , Gefter WB , Metzger LJ , Hoffman EA , Pack AI . Upper airway and soft tissue anatomy in normal subjects and patients with sleep-disordered breathing. Significance of the lateral pharyngeal walls . Am J Respir Crit Care Med. 1995. ; 152 ( 5 Pt 1 ): 1673 – 1689 . [DOI] [PubMed] [Google Scholar]
- 17. Neelapu BC , Kharbanda OP , Sardana HK , et al . Craniofacial and upper airway morphology in adult obstructive sleep apnea patients: a systematic review and meta-analysis of cephalometric studies . Sleep Med Rev. 2017. ; 31 : 79 – 90 . [DOI] [PubMed] [Google Scholar]
- 18. Zhang L , Zhang X , Li YM , et al . Association of craniofacial and upper airway morphology with cardiovascular risk in adults with OSA . Nat Sci Sleep. 2021. ; 13 : 1689 – 1700 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Charrad M, Ghazzali N, Boiteau V, Niknafs A. Nbclust: An r package for determining the relevant number of clusters in a data set. J Stat Softw. 2014;61:1–36.
- 20. General Administration of Sport . National body mass monitoring report. http://sports.china.com.cn/quanminjianshen/quanminjianshenbaogao/detail1_2015_11/18/472339.html . Accessed December 4, 2021. .
- 21. Mazzotti DR , Keenan BT , Lim DC , Gottlieb DJ , Kim J , Pack AI . Symptom subtypes of obstructive sleep apnea predict incidence of cardiovascular outcomes . Am J Respir Crit Care Med. 2019. ; 200 ( 4 ): 493 – 506 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Bailly S , Destors M , Grillet Y , et al . Obstructive sleep apnea: a cluster analysis at time of diagnosis . PLoS One. 2016. ; 11 ( 6 ): e0157318 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Labarca G , Dreyse J , Salas C , et al . A clinic-based cluster analysis in patients with moderate-severe obstructive sleep apnea (OSA) in Chile . Sleep Med. 2020. ; 73 : 16 – 22 . [DOI] [PubMed] [Google Scholar]
- 24. Guilleminault C , Partinen M , Quera-Salva MA , Hayes B , Dement WC , Nino-Murcia G . Determinants of daytime sleepiness in obstructive sleep apnea . Chest. 1988. ; 94 ( 1 ): 32 – 37 . [DOI] [PubMed] [Google Scholar]
- 25. Gupta N , Malhotra N , Ish P . GOLD 2021 guidelines for COPD—what’s new and why . Adv Respir Med. 2021. ; 89 ( 3 ): 344 – 346 . [DOI] [PubMed] [Google Scholar]
- 26. Lacedonia D , Carpagnano GE , Sabato R , et al . Characterization of obstructive sleep apnea-hypopnea syndrome (OSA) population by means of cluster analysis . J Sleep Res. 2016. ; 25 ( 6 ): 724 – 730 . [DOI] [PubMed] [Google Scholar]
- 27. Joosten SA , Hamza K , Sands S , Turton A , Berger P , Hamilton G . Phenotypes of patients with mild to moderate obstructive sleep apnoea as confirmed by cluster analysis . Respirology. 2012. ; 17 ( 1 ): 99 – 107 . [DOI] [PubMed] [Google Scholar]
- 28. Joosten SA , Landry SA , Wong AM , et al . Assessing the physiologic endotypes responsible for REM- and NREM-based OSA . Chest. 2021. ; 159 ( 5 ): 1998 – 2007 . [DOI] [PubMed] [Google Scholar]
- 29. Nakayama H , Kobayashi M , Tsuiki S , Yanagihara M , Inoue Y . Obstructive sleep apnea phenotypes in men based on characteristics of respiratory events during polysomnography . Sleep Breath. 2019. ; 23 ( 4 ): 1087 – 1094 . [DOI] [PubMed] [Google Scholar]
- 30. Ma EY , Kim JW , Lee Y , Cho SW , Kim H , Kim JK . Combined unsupervised-supervised machine learning for phenotyping complex diseases with its application to obstructive sleep apnea . Sci Rep. 2021. ; 11 ( 1 ): 4457 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Ding Q , Qin L , Wojeck B , et al . Polysomnographic phenotypes of obstructive sleep apnea and incident type 2 diabetes: results from the DREAM Study . Ann Am Thorac Soc. 2021. ; 18 ( 12 ): 2067 – 2078 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Kim SJ , Alnakhli WM , Alfaraj AS , Kim KA , Kim SW , Liu SY . Multi-perspective clustering of obstructive sleep apnea towards precision therapeutic decision including craniofacial intervention . Sleep Breath. 2021. ; 25 ( 1 ): 85 – 94 . [DOI] [PubMed] [Google Scholar]
- 33. Lebret M , Martinot JB , Arnol N , et al . Factors contributing to unintentional leak during CPAP treatment: a systematic review . Chest. 2017. ; 151 ( 3 ): 707 – 719 . [DOI] [PubMed] [Google Scholar]
- 34. Goh KJ , Soh RY , Leow LC , et al . Choosing the right mask for your Asian patient with sleep apnoea: a randomized, crossover trial of CPAP interfaces . Respirology. 2019. ; 24 ( 3 ): 278 – 285 . [DOI] [PubMed] [Google Scholar]
- 35. Evangelista E , Rassu AL , Barateau L , et al . Characteristics associated with hypersomnia and excessive daytime sleepiness identified by extended polysomnography recording . Sleep. 2021. ; 44 ( 5 ): zsaa264 . [DOI] [PubMed] [Google Scholar]