1. Introduction

Obstructive sleep apnea (OSA) is a widespread disease of sleep-disordered breathing (SDB) [1]. In North America, it is estimated that 15%–30% of men and 10%–15% of women have OSA [2], with a total worldwide prevalence estimated at 936 million people [3]. OSA is caused by a collapse of the soft tissues of the upper airway during sleep, which leads to partial or complete reductions in airflow (known as hypopneas and apneas, respectively) that must last at least 10 s but typically are between 10 and 30 s [4]. The severity of OSA is generally measured using the apnea–hypopnea index (AHI), a representation of the average number of sleep disturbance events in 1 h of sleep. The American Academy of Sleep Medicine defines a person without OSA as having an AHI of < 5, mild OSA of 5–15, moderate OSA of 15–30, and severe OSA > 30 events per hour [4].

There has been extensive debate in the past four decades as to whether there is clinical significance of OSA occurring during the rapid eye movement (REM) phase of sleep versus the nonrapid eye movement (NREM) phase of sleep. Early research showed that respiratory events were longer during REM sleep than NREM sleep in OSA patients [5], leading to the notion that OSA had a stronger relationship to REM. However, subsequent studies have shown that apnea events that occur during NREM sleep are associated with excessive daytime sleepiness and impaired quality of life in OSA patients, while apnea events during REM sleep did not show these associations [6, 7].

As the clinical significance of REM-related OSA has been scrutinized, attempts have been made to standardize its definition. The most widely used definition is a AHI_REM/AHI_NREM ratio of ≥ 2, which indicates that the majority of a patient’s SDB occurs during REM sleep [1]. However, several limitations to this method have been exposed [1], primarily in the binary classification between REM-OSA and NREM-OSA; this ratio-based approach is a binary classification that overlooks the diversity of subtypes that can occur within SDB, such as central, postarousal, arousal-promoting, or desaturation-promoting events. Nevertheless, even the phenotypic classification of REM-OSA versus NREM-OSA may be clinically useful as treatment, outcomes may depend on the patient’s phenotypic subtype of OSA. For example, one study has shown that mandibular advancement splints (MASs) have a significantly lower treatment response rate in patients with REM-predominant OSA over NREM or nonstage dependent OSA, with complete resolution of symptoms occurring in only 12% of REM-OSA patients as opposed to other subtypes (42% in NREM and 32% in nonstage dependent). This implies a need for reliable diagnostics to categorize patients with OSA into phenotypic subtypes to better predict treatment outcomes. Notably, the reproducibility of this finding has been mixed as other research has demonstrated no significant difference in MAS treatment outcomes between REM and nonstage-specific OSA. This could be due to patient heterogeneity, the smaller sample size, and the exclusion of NREM-predominant OSA in the analysis. Therefore, while the REM/NREM ratio is a useful metric, a multidimensional classification approach would provide a more comprehensive and clinically relevant method for classifying OSA subtypes.

Latent profile analysis (LPA) is one method we propose to integrate multiple variables to identify distinct subgroups, allowing for a more nuanced classification of SDB phenotypes than the current standard. In the present study, we used LPA to identify and characterize distinct subtypes of SDB within a large cohort of heterogenous patients based on AHI-REM, AHI-NREM, and the AHI-REM/AHI-NREM ratio. We hypothesized that a data-driven classification of patients who were referred for evaluation of OSA would identify subgroups of increased risk for REM or NREM-AHI predominant OSA which vary by their demographic and clinical features. The aim of this study was to identify unobserved distinct subtypes of SDB and examine their demographic and clinical features in patients who were referred for evaluation of OSA and underwent overnight PSG. This study, the first to apply LPA to the AHI_REM/AHI_NREM ratio, offers a patient-centered alternative to traditional statistical methods by identifying latent subgroups that may inform targeted interventions.

2. Materials and Methods

3. Results

3.2. SDB Profiles

The elbow plot reveals that for both the BIC and AIC indices, a sharp decline stops at the four-class solution (Figure 1). We also used the log-likelihood ratio test to determine the optimal number of classes based on several parameters: (1) The p value was smaller than a significance value of 0.05 for 3 (H₀) versus 4 classes (H₁) (p = 0.0197), and (2) the p value was higher than a significance value of 0.05 for 4 (H₀) versus 5 classes (H₁) (p = 0.1114). Thus, a four-class solution was determined to be most appropriate to fit the data. As hypothesized, the LPA identified a four-class solution based on the probabilities of SDB (Figure 2). Class I (n = 2753, 75.9%) includes patients with minimal probabilities of SDB. Class I can be characterized as patients with the lowest value for AHI_REM (low AHI_REM). Class II (n = 42, 1.2%) includes patients with the lowest AHI_NREM and highest ratio (REM-OSA). Class III (n = 633, 17.4%) includes patients with intermediate values for AHI_NREM (AHI_NREM < 30 events per hour). Class IV (n = 198, 5.5%) characterizes patients with the highest AHI_NREM (AHI_NREM ≥ 30 events per hour).

4. Discussion

The question of whether OSA is more severe during REM or NREM sleep has been extensively debated in the literature. Physiologically, it is expected that REM sleep would increase the severity of OSA due to REM atonia, which produces inhibition of upper airway muscle activation [13] and decreased genioglossus reflex in response to negative airway pressure during REM sleep [14]. However, some studies have shown the opposite: The expected symptoms of OSA (excessive daytime sleepiness and impaired quality of life) are actually worse in patients with elevated AHI_NREM [6, 7]. Thus, the AHI_REM/AHI_NREM ratio is one tool that has been explored as an alternative method to interrogate this question [15].

In the present study, we divided patients undergoing evaluation for OSA into distinct profiles of SDB using LPA and compared their means of AHI_REM, AHI_NREM, and AHI_REM/AHI_NREM ratio (Figure 2). Four profiles were chosen based on the abovedesired characteristics. In Class II (referred to as REM-OSA), we noted high AHI_REM with the lowest AHI_NREM and the highest ratio of AHI_REM/AHI_NREM. This class was unique because it was not predicted by neck circumference (Figure 3) and oxygen desaturation index (ODI) (Figures 4 and 5), suggesting obesity may have a stronger association with NREM-OSA than REM-OSA.

The relationship between age and REM versus NREM SDB is a central consideration of this paper. In general, the literature supports a continuous increase in the risk of OSA with age [16]. However, REM-associated OSA has been shown to be more common among younger people [14], while NREM-associated OSA may be more common in older populations [11]. It is hypothesized that this may be due to NREM sleep being associated with greater respiratory effort than REM sleep [17], as airway tone is lost with age, and SDB becomes more apparent during the NREM stage of sleep. This hypothesis would support our finding that older age was associated with SDB in Classes III and IV, which have intermediate and high values of apnea during NREM sleep, respectively.

OSA has been widely shown to be related to obesity, as excess body tissue contributes to compression of the airway during sleep [18]. This relationship may be bidirectional: Obesity has been shown to increase the risk for SDB (e.g., a 10% weight gain is associated with a 6-fold increase in odds of developing OSA [19]), while OSA may also increase the odds of developing obesity, likely through a multifactorial association involving predisposition to weight gain due to daytime somnolence leading to decreased exercise and endocrine dysfunction [18]. Our SDB profiles supported this relationship as well, as demonstrated by the increase in the probability of SDB Class IV with increasing BMI. The probability of being in Classes III and IV was also positively associated with neck diameter (Figure 3), further reinforcing this relationship.

Another parameter evaluated in the present study was the ODI, a measurement that has been proposed as an alternative to PSG for use in grading OSA severity [19]. Interestingly, we found unique patterns of ODI values between the four SDB profiles (Figure 4). The probability of being in Class I (low AHI_REM) decreased and the probability of being in Class IV increased with ODI, as anticipated [20]. However, the probability of being in Class III followed a similar trend to Class IV before an ODI of 40 and then followed a similar trend to Class I at index values higher than 40, suggesting a mean ODI of 40 among moderate-severity OSA patients. This finding is in agreement with other studies, which demonstrated ODI cutoff values of 25 had 89.7% sensitivity, and 20 had 96.5% sensitivity for severe OSA [21]. Our findings suggest that a cutoff value in this range may be clinically useful for the development of a screening test for identifying possible cases of moderate to severe OSA. This would be especially relevant for developing the ODI into a screening test, as it is more cost-effective and requires less equipment to carry out than the traditional PSG test.

We further investigated SpO₂ measurements as another parameter for OSA risk. We found that the probability of being in Classes III and IV increased with decreasing SpO₂ nadir and the low probability of SDB (Class I) increased with increasing SpO₂ nadir (Figure 5). This pattern is consistent with previous studies which have shown that SpO₂ nadir is a predictor of OSA severity [22]. Interestingly, the mean lowest SpO₂ for mild, moderate, and severe OSA in the comparison study were 95.7%, 95.7%, and 93.6% [19], which were all significantly higher than the lowest SpO₂ values for the present study (83.6%, 87.5%, 74.2%, and 65.8% for Classes I–IV, respectively). This discrepancy indicates that further research is needed to interrogate SpO₂ as a reliable predictor of OSA risk.

Arousal index was also found to be associated with Classes III and IV, whose probability increased with this parameter. Arousal index was highest in Class IV (mean = 61.2), followed by Class III (mean = 29.4). Classes I and II were similar, with arousal indices of (mean = 17.0 and mean = 15.9, respectively). As the arousal index is a polysomnographic indicator of disrupted sleep, these results support the association of poorer sleep quality with REM-OSA. These findings are also in line with literature that supports the arousal index being lower in both less severe OSA and REM-predominant OSA [23, 24].

The ESS is an additional measure evaluated in the present study, which has been widely used for the past two decades to measure daytime sleepiness to aid in the diagnosis of sleep disorders [10]. We found that higher values scored on the ESS were associated with a higher probability of Class IV, further supporting that high NREM-associated OSA with a low AHI_REM/AHI_NREM ratio is indicative of more severe disease. Contrary to our findings, Gabryelska and Białasiewicz [25] have found similar ESS scores between REM-OSA and NREM-OSA groups despite presenting with different AHI levels. However, they did not find any group differences in SpO₂ nadir or arousal index, suggesting that these parameters may explain the effect in ESS that was observed in the present study.

Lastly, we found that total sleep time was negatively associated with the probability of Classes III and IV. This finding provides evidence that the higher the proportion of OSA occurring during NREM sleep, the poorer the sleep duration and, thus, the greater the impact of OSA on sleep health. This finding also contradicts the findings of Gabryelska and Białasiewicz [25], who did not find differences in total sleep time between REM-OSA and NREM-OSA. However, the previously stated differences in other parameters may explain the lack of an effect seen on sleep duration.

The present study has several strengths, such as its relatively large sample size of 3626 participants which lends validity to our results. The use of a multinomial logistic model with LPA enabled risk evaluations to be made for four distinct classes of SDB, each with independent associations with various PSG parameters as discussed above.

We acknowledge that this study presents some limitations as well. The study cohort was comprised of individuals who were referred to a sleep clinic and had several comorbidities (82% obese, 46.4% hypertension, and 39.7% Type 2 diabetes mellitus) that are not representative of the general population, thus conferring a possible selection bias. Thus, our findings are most applicable to patients who are already being evaluated for possible SDB. Additionally, the cross-sectional nature of this study also confers certain limitations, most notably in the inability to infer causality between NREM-OSA and the various polysomnographic parameters discussed above. The sleep time was measured during diagnostic and titration studies which might lower the sleep time due to adjustment of the treatment device. This may be due to microawakenings caused by gradual increases in the air pressure that occur during the titration process of the study, as well as general discomfort from wearing the CPAP mask possibly contributing to a shortened sleep time [11].

5. Conclusion

This study demonstrated a strong relationship between NREM-predominant OSA (AHI_NREM < 30 events per hour and AHI_NREM ≥ 30 events per hour) with various PSG parameters, including older age, increased BMI and neck circumference, decreased SpO₂ nadir, increased arousal index, higher ESS values, and lower total sleep time. Interestingly, ODI for the intermediate NREM group (AHI_NREM < 30 events per hour) was found to follow low-NREM patterns below 40, and high-NREM (AHI_NREM ≥ 30 events per hour) patterns above 40. Further research in this area could focus on developing the desaturation index into a predictor of OSA severity.

Ethics Statement

The study’s ethical approval was obtained from the Institutional Review Board at King Abdullah International Medical Research Center, with approval # IRB project No. RC15/058/R. This study has been completed prior to Dr. Anwar Ahmed joining the Uniformed Services University of the Health Sciences.

Consent

The authors have nothing to report.

Disclosure

The opinions and assertions expressed herein are those of the authors and do not reflect the official policy or position of the Uniformed Services University of the Health Sciences or the Department of Defense. The abstract of this paper was presented at the 2024 American Thoracic Society (ATS) Annual Meeting as a poster presentation with interim findings. The poster’s abstract was published in “Poster Abstracts” in the American Journal of Respiratory and Critical Care Medicine 2024; 209: A4556. https://www.atsjournals.org/doi/abs/10.1164/ajrccm-conference.2024.209.1_MeetingAbstracts.A4556.

Conflicts of Interest

The authors declare no conflicts of interest.

Author Contributions

A.E.A. and H.A.J. conceived the project. M.W. and A.E.A. wrote the manuscript. H.A.J. designed and created the database, and conducted sleep studies. A.E.A. performed the data analysis. A.A., A.A.-H., H.A.A., and J.K.W. interpreted the results of the analysis, with all authors contributing edits and reviews. M.W. and A.E.A. have contributed equally to this work as co-first authors.

Funding

No funding was received for this research.