Rapid eye movement sleep and slow wave sleep rebounded and related factors during positive airway pressure therapy

This study aimed to investigate the clinical characteristics and predictors of increased rapid eye movement (REM) sleep or slow wave sleep (SWS) in patients with obstructive sleep apnea (OSA) following positive airway pressure (PAP) therapy. The study retrospectively analyzed data from patients with OSA who underwent both diagnostic polysomnography (PSG) and pressure titration PSG at the Tangdu Hospital Sleep Medicine Center from 2011–2016. Paired diagnostic PSG and pressure titration studies from 501 patients were included. REM rebound was predicted by a higher oxygen desaturation index, lower REM proportion, higher arousal index, lower mean pulse oxygen saturation (SpO2), higher Epworth sleepiness score and younger age (adjusted R2 = 0.482). The SWS rebound was predicted by a longer total duration of apneas and hypopneas, lower N3 duration, lower SpO2 nadir, lower REM proportion in diagnostic PSG and younger age (adjusted R2 = 0.286). Patients without REM rebound or SWS rebound had a high probability of comorbidities with insomnia and mood complaints. Some parameters (subjective and objective insomnia, excessive daytime sleepiness, age and OSA severity) indicate changes in REM sleep and SWS between diagnostic and titration PSG tests. Treatment of insomnia and mood disorders in patients with OSA may helpful to improve the use PAP.

Polysomnography and pressure titration. All the patients spent two nights in a sleep laboratory: one night for diagnostic PSG, and the other for PSG and PAP titration. Diagnostic PSG was performed using a computerized PSG system (Alice 4 or 5; Respironics, Pittsburgh, PA, USA). The recording montage included an electroencephalogram, electrooculogram, electromyogram, electrocardiogram, breathing effort, airflow, oximetry, and body position. The starting mode was continuous PAP (CPAP) with a pressure of 4 cmH 2 O. CPAP was increased in 1 cm H 2 O increments in response to 2 apneas, 3 min of snoring, 3 hypopneas or 5 respiratory effort-related arousals (RERAs). If patients were unable to tolerate CPAP or if events still presented with CPAP at 15 cm H 2 O, the mode was switched to bilevel PAP. PAP titration was performed by a trained technologist. Sleep was staged according to the American Academy of Sleep Medicine (AASM) Manual for the Scoring of Sleep and Associated Events 18 . Obstructive apneas were defined as cessation of airflow for at least 10 s. Hypopneas were defined as a 30% reduction in nasal airflow from baseline for at least 10 s, associated with desaturation of at least 3% and/or an arousal on the electroencephalogram. Respiratory effort-related arousal was defined as a sequence of breaths lasting ≥ 10 s that were characterized by increasing respiratory effort, a flattening of the inspiratory portion of the nasal pressure (diagnostic study), or a PAP device flow (titration study) waveform leading to arousal from sleep when the sequence of breaths did not meet criteria for apnea or hypopnea. All sleep studies were scored by trained sleep technologists.
Statistical analysis. SPSS version 22.0 software (SPSS Inc., Chicago, IL, USA) was used for all statistical analyses. All figures were carried out using GraphPad Prism 7.0 software (GraphPad Prism, Ver 7.0). Microsoft Word (Microsoft Office 2016, Microsoft Corporation, Redmond, Washington, USA) was used to design the flow chart. GraphPad Prism 7.0 software (GraphPad Software, Inc., La Jolla, California, USA) was used to generate the graphs. The Shapiro-Wilk test was used to assess the normality of all data. Independent-sample t test or the Mann-Whitney U test was used to compare demographic, clinical diagnostic and pressure titration PSG parameters in SWS rebounders versus non-SWS rebounders, and REM rebounders versus non-REM rebounders for normally and non-normally distributed continuous variables, respectively. A paired t test or the www.nature.com/scientificreports/ Wilcoxon signed rank sum test was used to compare PSG variables in the diagnostic PSG and pressure titration studies for normally and non-normally distributed continuous variables, respectively. SWS rebound and REM rebound were defined according to the differences in the REM proportion (%REM) or N3 proportion (%N3) between the pressure titration and diagnostic PSG (change in %SWS = %N3 in pressure titration-%N3 in the diagnostic test; change in %REM = %REM in pressure titration-%REM in the diagnostic test) and the values of the %REM or %N3 in pressure titration PSG. K-means clustering analysis was used to cluster the change in REM sleep characterized by %REM in pressure titration and the change in %REM, and change in SWS characterized by %N3 in pressure titration and the change in %N3. The integers for both the minimum %REM in the pressure titration study and minimum positive change in %REM in the more REM group served as cutoff values for REM sleep with or without rebound. The integers for both the minimum %N3 in the pressure titration study and minimum positive change in %N3 in the more SWS rebound group served as cutoff values for SWS with or without rebound. The χ 2 test or Fisher's test was used to compare proportions. Variables that were statistically significant after comparison between REM rebounders and non-REM rebounders and between SWS rebounders and non-SWS rebounders were entered into multiple linear stepwise regression analyses to determine the best model for predicting changes in REM sleep and SWS. The model with the highest adjusted R 2 value was accepted as the best model predicting changes in REM sleep and SWS. A p value < 0.05 was considered statistically significant.
Ethical approval. All  REM rebound and prediction model. The mean %REM in the pressure titration study was 28.62% (min-max: 16.44-51.88%), and the mean change in %REM between the pressure titration study and diagnostic PSG was 16.42% (min-max: 5.74-41.88%) in one cluster with more REM sleep determined using the K-means clustering analysis. The mean %REM in the pressure titration study was 16.13% (min-max: 0-27.95%), and the mean change in %REM between the pressure titration study and diagnostic PSG was 1.20% (min-max: − 21.84-12.01%) in the other cluster with less REM sleep. Therefore, for clinical communication, considering the integer of the more REM sleep group, REM rebound was defined as at least 16% in %REM in pressure titration PSG and an increase of at least 6% in %REM (Supplement Fig. 1). For the 225 (44.91%) patients that displayed REM rebound, the mean %REM on diagnostic PSG was 12.07% (mean duration of REM: 51.80 min), increasing to 27.27% (mean duration of REM: 113.15 min) on the pressure titration night. Compared with non-REM rebounders, REM rebounders experienced significantly less insomnia (p = 0.004), including early insomnia (p < 0.001) and late insomnia (p = 0.047), and less dizziness (p = 0.011), anxiety (p < 0.001), irritability (p = 0.008), and depression (p = 0.023) ( Table 2).

Discussion
The cutoff values for REM sleep and SWS rebound in patients with OSA treated with PAP were identified using the objective statistical method K-means clustering analysis. In this study, REM sleep rebound was defined by at least 16% REM sleep on the pressure titration night and a 6% increase in %REM, whereas SWS rebound was defined by at least 13% N3 on the pressure titration night and a 3% increase of change in %N3. Two hundred twenty-five (44.91%) patients experienced REM rebound and 164 (32.73%) experienced SWS rebound. In a study by Osuna et al. 11 , 82/179 (46%) patients experienced REM rebound, similar to our study (44.91%). Osuna et al. 11 also reported a 6% increase in %REM as a different point of REM rebound, which may have resulted in the similar prevalence of REM rebounders in our study; however, the authors did not indicate which statistical method was applied to obtain the cutoff value for the 6% change in %REM. In addition, Osuna et al. 11 did not report a difference in the change in SWS between pressure titration and diagnostic PSG studies. In a study by Brillante et al. 8 , 76 of 335 patients (22.68%) experienced REM rebound, and 80 (23.8%) experienced SWS rebound, which was less than the values obtained in our study. The possible explanation for the difference may be that the definitions of REM rebound and SWS differed. Brillante et al. 8 defined REM and SWS rebound according to the differences in REM sleep or SWS duration (in minutes) between pressure titration and diagnostic PSG studies as a percentage of the diagnostic PSG duration of REM sleep or SWS: (duration of REM sleep or SWS in diagnostic PSG-duration of REM sleep or SWS in pressure titration PSG)/duration of REM sleep or SWS in diagnostic PSG*100%. For patients who did not experience REM sleep or SWS during diagnostic PSG, a REM sleep or SWS period of > 15 min on the pressure titration night was considered a > 10% rebound. In patients diagnosed with OSA, SWS rebound was defined by a 40% increase on the pressure titration night compared with the diagnostic PSG night, whereas REM sleep rebound was defined by only a 20% increase. These thresholds were identified objectively using logarithmic equations and a forward sequential regression analysis 8 . Koo et al. 10 found that 35/95 (36.84%) patients experienced REM rebound, and 17/95 (17.89%) experienced SWS rebound in the split night study. More REM sleep was supposed to occur during the last part of sleep when the pressure titration was performed in the split night study. Koo et al. 10 reported a larger change in %REM between diagnostic and pressure titration study than was observed in our study (20% vs. 6%), which may contribute to a lower REM rebound prevalence than in our study. Koo et al. 10 reported a lower change in SWS percentages between diagnostic and pressure titration studies than in our study (10% vs. 13%). The lower SWS rebound prevalence may be due to the lower amount of SWS during the last part of sleep when the pressure titration was performed in the split night study. Koo et al. 10 used a split night study in OSA patients to define REM rebound using 2 criteria, including at least one REM period of ≥ 30 min duration and a ≥ 20% increase in REM sleep during the treatment portion. SWS rebound was defined as a ≥ 10% increase in SWS during the treatment portion; however, the study did not define the REM and SWS rebound threshold. The parameters age, %REM in diagnostic PSG, and the ODI, which is linearly related to AHI, were predictors of changes in REM and SWS in the current study, and BMI and AHI were considered predictors of changes in REM sleep in other studies 9 . Differences in the definitions of REM and SWS rebound, differences in the characteristics of patients with OSA (age: 47.0-58.6 years, BMI: 25.7-39.2 kg/ m 2 , AHI 23.6-72.9 times/h) and %REM in diagnostic PSG (6.7-18.4%) 9 lead to different prevalence rates of REM rebound and SWS rebound on the first night of exposure to PAP therapy.
No difference was observed in the AHI between the REM non-rebound and REM rebound groups during PAP treatment, suggesting that PAP treatment decreases respiratory events; however, differences were observed in the sleep architecture and clinical characteristics between the groups. These phenomena were similar between the SWS non-rebound and SWS rebound groups. According to the definition of REM rebound, compared with patients without REM rebound, patients with REM rebound were younger, had a higher BMI, severe ESS, more TST and SE, a short SL, a higher arousal index, less WASO, more N1 sleep, less REM sleep and SWS, and worse OSA (higher AHI and ODI, longer duration of apneas and hypopneas, and a lower baseline SpO 2 , SpO 2 nadir, and degree of oxygen desaturation) in the diagnostic PSG study. These statistically significant differences were observed in most diagnostic PSG parameters that coincided with clinical symptoms. Compared with non-REM rebounders, REM rebounders experienced less dizziness, insomnia, anxiety, irritability, and depression. Similar to the present study, Koo et al. 10 also found that patients with REM rebound tended to be younger, with a higher AHI and less REM sleep in the diagnostic portion than patients without REM rebound. In contrast to the present study, Koo et al. 10 did not report t differences between the REM rebound and non-rebound groups in the degree of obesity, level of subjective sleepiness, or %SWS. Compared with patients without SWS rebound, patients with SWS rebound had more TST, higher SE, greater N2 and N3 sleep duration time, less N1%, less WASO, a higher arousal index, worse OSA (higher AHI, longer durations of apneas and hypopneas, lower mean oxygen saturation, substantial reduction in oxygen saturation during sleep, and a lower oxygen saturation nadir) in the diagnostic PSG study, a higher BMI and ESS, and were younger. Compared with patients without SWS rebound, patients with SWS rebound experienced less dizziness, insomnia, anxiety, and irritability. The cutoff values of 16% REM sleep in titration PSG and 6% changes in %REM were reasonable to define REM rebound www.nature.com/scientificreports/ www.nature.com/scientificreports/ and non-rebound according differences between the clinical and PSG characteristics. The SWS rebound cutoff value was also reasonable. Insomnia, anxiety, and depression can all lead to changes in sleep architecture 16,17 and are common OSA comorbidities [12][13][14][15] . Thus, changes in the sleep architecture of patients with OSA are not caused by apnea and hypopnea events alone but also by comorbidities. The absence of REM and SWS rebound during the initial treatment of OSA with PAP may be explained by comorbid insomnia, anxiety, and depression. The initial PAP treatment can only decrease respiratory event-related arousal to restore REM sleep and SWS but has no effect on a longer sleep latency, increased WASO, decreased SWS, and REM sleep instability caused by insomnia, anxiety, or depression. Depression can increase REM sleep, explaining the greater incidence of depression in the REM non-rebound group, because patients with depression do not have a greater regulatory range of REM sleep 5 . Moreover, these patients may have increased sleep anxiety due to the use of PAP, resulting in a longer sleep latency, more WASO, and less SWS and REM sleep. The phenomena that patients with OSA but without REM rebound or SWS rebound reported more insomnia, anxiety and depression indicated that treatment of these comorbidities may possibly help restore the sleep architecture of patients with OSA through an initial pressure titration. This study is the first to explore the relationship among OSA comorbidities, REM and SWS rebound during initial pressure titration. According to the clinical data and diagnostic PSG parameters, the present study established a predictive model of REM rebound and SWS rebound for patients with OSA undergoing PAP therapy. The ODI, %REM arousal index, baseline SpO 2 , ESS, and age were entered into the best model for predicting changes in REM sleep (adjusted R 2 = 0.482), and the total duration of apneas and hypopneas, N3 duration, SpO 2 nadir, REM% and age were entered into the best model for predicting changes in SWS (adjusted R 2 = 0.286). According to the adjusted R 2 values, the present prediction models explain a greater proportion of REM sleep variability than changes in    www.nature.com/scientificreports/ index, baseline SpO 2 , ESS and age were analyzed using a multiple linear enter regression model, (R = 0.689, R 2 = 0.475, adjusted R 2 = 0.468). A higher AHI with %REM, lower arousal index, lower baseline SpO 2 , higher ESS and younger age can also be used to predict REM rebound, but the inclusion of the ODI in the prediction model produced better results than the model including the AHI in the prediction model (ODI adjusted R 2 = 0.482, AHI adjusted R 2 = 0.468). Generally, breath event-related parameters, such as a higher AHI, higher ODI and lower baseline SpO 2 , and sleep architecture parameters, such as a higher arousal index, lower %REM and younger age, are common variables for predicting REM rebound. Breath event-related parameters, such as a longer total duration of apnea and hypopnea events and lower SpO 2 nadir, sleep architecture parameters such as a higher arousal index, less REM sleep and N3 duration, and younger, are common variables predicting SWS rebound.
Limitations. The enrolled patients with OSA were derived from a single sleep medicine center in the Department of Neurology, indicating that most patients with OSA had neurological diseases and psychiatric diseases that could cause bias. Further studies are needed to verify whether the cutoff values for the definitions of REM and SWS rebound are suitable for all patients with OSA treated at the other sleep medicine centers. This study did not evaluate changes after-PAP treatment and did not analyze the clinical significance of REM or SWS rebound. All patients underwent only one diagnostic PSG and one pressure titration PSG; therefore, the first night effect of diagnostic and pressure titration were present, which may complicate the results.

Conclusions
The combination of at least a 6% increase in %REM and 16% REM sleep during the pressure titration study reflected a significant REM rebound. The combination of at least a 3% increase in %SWS and 13% SWS during the pressure titration study reflected a significant SWS rebound. REM rebound was predicted by a higher ODI, lower %REM, higher arousal index and lower baseline SpO 2 in the diagnostic PSG, higher ESS, and younger age (adjusted R 2 = 0.482). SWS rebound was predicted by a younger age, longer total duration of apneas and hypopneas, shorter N3 duration, lower SpO 2 nadir, and lower %REM in the diagnostic PSG (adjusted R 2 = 0.286). Compared with SWS rebound, REM rebound was more obvious and prevalent during the pressure titration study. The diagnostic PSG and clinical data predict REM rebound better than SWS rebound. Compared with patients with REM rebound or SWS rebound, patients without REM rebound or SWS rebound had a higher probability of comorbidities with insomnia and mood complaints. The ability to predict which patients will experience REM and SWS rebound during pressure titration may help identify those who will be more likely to respond well to PAP therapy. The treatment of insomnia, anxiety and depression may restore the sleep architecture and improve compliance in patients with OSA who are treated with PAP.