- Top of page
- Supporting Information
Background and objective: Patients with OSA manifest different patterns of disease. However, this heterogeneity is more evident in patients with mild-moderate OSA than in those with severe disease and a high total AHI. We hypothesized that mild-moderate OSA can be categorized into discreet disease phenotypes, and the aim of this study was to comprehensively describe the pattern of OSA phenotypes through the use of cluster analysis techniques.
Methods: The data for 1184 consecutive patients, collected over 24 months, was analysed. Patients with a total AHI of 5–30/h were categorized according to the sleep stage and position in which they were predominantly affected. This categorization was compared with one in which patients were grouped using a K-means clustering technique with log linear modelling and cross-tabulation.
Results: Patients with mild-moderate OSA can be categorized according to polysomnographic parameters. This clinical categorization was validated by comparison with a categorization in which patients were grouped by unsupervised K-means cluster analysis. The clinical groups identified were: (i) rapid eye movement (REM) predominant OSA, 44.6%; (ii) non-REM predominant OSA, 18.9%; (iii) supine predominant OSA, 61.9%; and (iv) intermittent OSA, 12.4%. Patients categorized as having both REM and supine predominant OSA showed characteristics of both the REM predominant and supine predominant OSA groups.
Conclusions: Patients with mild-moderate OSA show different polysomnographic phenotypes. This approach to categorization more appropriately reflects disease heterogeneity and the likely multiple pathophysiological processes involved in OSA.
- Top of page
- Supporting Information
OSA (OSA) affects 2–4% of the middle-aged population.1 It is associated with neurocognitive deficits and an increased risk of vascular disease.2,3 Severe OSA appears to be homogeneous, with continuous, repetitive upper airway obstruction across sleep stages and body positions. However, the vast majority of patients with OSA have mild-moderate disease, with an AHI of 5–30 events/h.4 Although the long-term consequences of untreated mild-moderate OSA remain unclear,5 patients are at higher risk of motor vehicle accidents,6 have higher blood pressures7 and experience greater cardiovascular morbidity.8
Patients with severe OSA generally respond well to CPAP.9–11 However, patients with mild-moderate OSA are typically less responsive to this approach,12,13 and although other interventions have been trialled, the effect size is smaller than that observed in patients with severe OSA.12,14,15 Studies addressing the prevalence, pathophysiology and outcomes of mild-moderate OSA in groups of patients have been based on overall AHI. However, the total AHI does not reflect the heterogeneity of mild-moderate OSA, in which clinical experience indicates that there is usually a clustering of obstructive events when the patient is in the supine position or in a particular sleep stage.
Identification of the phenotype that characterizes OSA in individual patients could be diagnostically and therapeutically useful, as causative mechanisms, long-term health outcomes and treatment response may vary across OSA subtypes. For example, patients who experience OSA while supine respond to oral appliances better than those whose OSA is not dependent on position,16 and transnasal insufflation is more effective in patients with OSA that predominantly occurs during rapid eye movement (REM) sleep.17
There is limited data in the literature describing subtypes of OSA, with most studies focusing in isolation on OSA in either the supine position or during REM sleep.18,19 These studies were confounded by comparisons between groups of patients with mild-moderate or severe OSA. No study has systematically investigated patients with mild-moderate OSA and the relative proportions of these patients showing different polysomnographic phenotypes. We therefore hypothesized that patients with mild-moderate OSA are largely comprised of those with (i) supine predominant OSA, which occurs more frequently in those who are minimally obese; (ii) REM predominant OSA, which occurs more frequently in obese individuals and women; (iii) non-REM (NREM) predominant OSA; and (iv) OSA that overlaps between body position predominant OSA and sleep stage predominant OSA. Accordingly, the aim of this study was to devise a comprehensive phenotypic framework that could be validated using a cluster analysis technique and could be used to more accurately recognize the heterogeneity and likely multiple contributory pathophysiological mechanisms among patients with mild-moderate OSA.
- Top of page
- Supporting Information
The classification of OSA is currently based on disease severity as determined by the overall AHI. This works well for identifying patients with severe repetitive upper airway obstruction, who respond to CPAP therapy. It does not, however, recognize the heterogeneity in polysomnographic parameters and the response to treatment of patients with mild-moderate disease. The aim of this study was to improve the way OSA is classified by incorporating comprehensive phenotypic data that is likely to reflect disease heterogeneity and the pathophysiological processes involved in OSA.
This study is the first to report on the relative proportions of patients with different patterns of mild-moderate OSA, who present to a tertiary sleep centre. Six separate phenotypes of mild-moderate OSA were identified; this categorization was supported by an unsupervised K-means cluster analysis, and the groups showed significant differences in patient characteristics and polysomnographic parameters. These results confirm the heterogeneity of OSA and suggest that a structured classification of disease subtypes will facilitate the understanding and treatment of OSA.
Supine predominant OSA
In previous studies, AHI has been reported to vary widely in patients with supine predominant OSA, who have a lower BMI, are younger and more likely to be males.9,24 This is the most commonly described, and in this study, the most frequently encountered polysomnographic phenotype. Contrary to previous studies,18 patients with supine predominant OSA were found to be subjectively sleepier, albeit based on very limited and questionable clinical evidence. The greater sleep efficiency of these patients has been noted previously, but in this group, those with non-positional OSA had more severe disease and were subjectively more sleepy.18 The finding that sleep efficiency was higher but subjective sleepiness was equivalent or marginally greater in patients with supine predominant OSA raises the possibility that respiratory events occurring while in a supine position may be more detrimental to subjective sleepiness.
Supine isolated OSA
Patients with supine isolated OSA demonstrated an exaggerated form of supine predominant OSA. The finding that patients with supine isolated OSA were younger and weighed less reinforces the observation that lower BMI and younger age are intrinsic features of supine OSA. The gradation in the supine position dependence of OSA according to age and body weight implicates these factors as being important in the pathogenesis of supine predominant OSA. Conversely, the present results suggest that excess body weight mediates a risk of OSA in all body positions and raises the concept that the severity of OSA may potentially evolve over time in some groups of patients. It is possible that a proportion of young, thin patients who are predisposed to upper airway obstruction initially have supine isolated OSA, but with increasing age and/or weight gain, develop obstruction in the lateral position as well. Support for this concept comes from the clinical impression that weight loss improves OSA that occurs in the lateral rather than in the supine position. Unfortunately studies of weight loss in patients with OSA have generally only reported changes in total AHI;25 thus data confirming this observation is limited. The application of the present phenotypic framework to such studies is an example of how this approach could facilitate investigations of the pathogenesis of OSA or the effects of treatment.
REM predominant OSA
Patients with REM predominant OSA had a higher BMI, were older and were more likely to be females. The female preponderance of patients experiencing respiratory events during REM sleep is well known.19,26 Although in this study, females comprised a greater proportion of this group, only 44% of all females had REM predominant OSA, and males still constituted the majority of such patients (55%). This contrasts with a previous study in which women comprised the majority of patients with REM predominant OSA.27 These findings suggest that the factors leading to OSA during REM sleep are not gender specific, but nevertheless are more likely to occur in women than in men.
REM isolated OSA
Patients with REM isolated OSA were younger than the other patients in the REM predominant OSA group (mean age 49.7 vs 54.5 years, P < 0.001) and experienced significantly fewer events. Importantly, there was no difference in gender ratio or BMI between these two phenotypes. It was previously noted that sleep disordered breathing is more common and more severe in post-menopausal compared with pre-menopausal women.28 This finding has relevance to the present study. Although the menopausal status of the cohort was unknown, the group with REM predominant OSA, who experienced some NREM OSA, was older and likely to comprise more women who were not post-menopausal, compared with the group with REM isolated OSA, who did not experience NREM OSA. Therefore, the known worsening in the severity OSA following menopause may be due primarily to a worsening of obstruction during NREM sleep, mediated either by increasing age or menopausal status per se. This possibility warrants further investigation.
REM-supine overlapping OSA
In this study, 264 (37%) of the 714 patients with supine predominant OSA were also categorized as having REM predominant OSA. That is, the degree of obstruction experienced by these patients was much less during lateral NREM sleep than during either supine NREM sleep, or REM sleep in any body position. The patients with REM-supine overlapping OSA showed significant characteristics from both ‘parent’ groups, so that they were older, had a higher BMI and were more likely to be females, as compared with patients with supine predominant OSA alone. Conversely, these patients were younger, had a lower BMI and were more likely to be males, as compared with patients with REM predominant OSA alone. This suggests that separate factors that contribute to events occurring during supine and REM sleep interact to cause sleep disordered breathing in these patients.
The existence of a group of patients with REM-supine overlapping OSA has not been recognized in previous studies, but is likely to be relevant as it was also identified by cluster analysis. We believe it is an important distinction, as the mechanisms resulting in REM-related OSA and those resulting in OSA related to the supine position may be distinct. In any study of the mechanistic processes leading to OSA in the supine position, it would be important to perform sub-group analyses based on whether patients also experienced concomitant REM-related OSA, independent of body position. The recognition of this overlapping group is also relevant to the assessment of the effects of interventions such as weight loss, where improvement in OSA may preferentially occur either while sleeping in a particular position or during a particular sleep stage.
NREM predominant OSA
Patients with NREM predominant OSA experienced most respiratory events while in the supine position, and logistic regression analysis showed that the ratio of events in the supine position to those in the NSupine position was most predictive of NREM predominant OSA. Furthermore, in the cluster analysis, patients with NREM predominant OSA were not identified as a separate group. Greater severity of OSA during NREM compared with REM sleep suggests that respiratory control is unstable, with elevated loop gain, during NREM sleep.29 The finding that NREM predominant OSA occurs almost exclusively during supine sleep suggests that sleeping in the supine position plays a critical role in the factors leading to unstable respiratory control in these patients.30,31
The characteristics of the patients with intermittent OSA did not differ significantly from those of the cohort as a whole. This group constituted 13% of the entire cohort, emphasizing the fact that the vast majority of patients with mild-moderate OSA experience obstruction during a particular sleep stage or while sleeping in a particular position.
The phenotypes discussed were derived from a priori definitions based on well-recognized clinical patterns. Cluster analysis avoids the potentially artificial nature of disease sub-classification and allows recognition of spontaneous aggregations of similar patient phenotypes, or clusters. In this study, the K-means cluster analysis identified six clusters that were statistically similar and related to the clinical phenotypes described. In fact, several of the clusters closely resembled specific clinical phenotypes (Fig. 2).
The cluster analysis emphasized a number of themes identified in the analysis of polysomnographic phenotypes. Significant numbers of patients experienced both supine predominant and REM predominant OSA. This group of patients is represented by clusters 5 and 6. In addition, cluster 5 resembled the REM predominant clinical phenotype of OSA, with the patients being older, having a high BMI, being predominantly female and experiencing frequent events during REM sleep. Cluster 6 more closely resembled the supine predominant clinical phenotype and comprised younger, predominantly male, patients with a lower BMI and a tendency towards stable breathing during NREM sleep in a lateral position. The identification of these clusters confirmed the relationships between overlapping phenotypes, as identified in the preceding analysis.
The possibility that younger male patients with a lower BMI are predisposed to OSA while sleeping in the supine position was supported by the cluster analysis. Indeed, cluster 3 reflected the supine isolated clinical phenotype and this trend was also observed for clusters 1 and 2. This indicates the importance of considering patients with pure supine sleep-related obstruction independently of those with merely a predominance of obstruction while sleeping in the supine position.
Limitations of the study
These findings were based on patients attending a single sleep clinic and were therefore susceptible to referral bias. Whether the differences in sleep and demographic parameters that were identified would be replicated in unselected community populations is uncertain. In addition, access to more detailed clinical information such as ethnicity, medications, medical co-morbidities and cranio-facial features, was not available. Patterns of clinical presentation and co-morbidity may vary among the OSA phenotypes, and detailed clinical phenotyping should be an integral component of future work in this area.
It is well recognized that the total AHI can vary from night to night.32 Given that the data was collected from a single sleep study, it is possible that these polysomnographic phenotypes are not stable, and that the categorization of patients may change over subsequent nights. This would not be surprising given the mounting evidence that even within a given night, upper airway collapsibility and arousal threshold may change significantly.33 It has also been hypothesized on the basis of cross-sectional data that patients may progress from positional to non-positional OSA over a period of time, particularly if they gain weight.18 It is possible that OSA may progress to other phenotypes in a similar fashion to the evolution of severity of OSA over time. Confirmation of the categorization or progression from one phenotype to another by repeat sleep studies was beyond the scope of this study. In this context, clinicians also generally make treatment decisions based on the results of a single sleep study, and thus the categorization system that was devised reflects the real world clinical situation. Nevertheless, an important future direction for this work is exploration of the stability of these polysomnographic phenotypes over time, both in the short and long term.
It is important to acknowledge that the ‘Chicago criteria’ were used to score respiratory events, as these criteria were in use in the laboratory at the time of the study. It has been shown previously that hypopnoea scoring criteria can impact on the overall AHI,34 and would therefore be expected to have a similar impact on the determination of AHI in different sleep stages and body positions. This underscores the importance of considering the scoring of respiratory events when comparing studies performed in different laboratories. Future work in this area will require standardized scoring criteria, such as those proposed by the American Academy of Sleep Medicine in 2007.35