The hypnogram obtained during a sleep study in a patient with untreated obstructive sleep apnea (OSA) has certain distinct features. The majority of them have fragmented sleep architecture with limited amounts of Rapid Eye Movement (REM) and slow wave sleep. Historically, the sleep fragmentation has been attributed to the preponderance of arousals that are temporally associated with respiratory events in patients with untreated OSA (Remmers et al. 1978). Early experimental studies of selective sleep deprivation have demonstrated that stage REM sleep is highly susceptible to eradication by repetitive arousals, more so than stage slow wave sleep (Agnew et al. 1967). This might suggest that REM sleep is more frequently abolished by untreated OSA, as compared to slow wave sleep.

When first exposed to continuous positive airway pressure (CPAP), certain favorable changes occur in the sleep architecture. CPAP therapy leads to a reduction in the arousal index, sleep stage shifts and non-rapid eye movement (NREM) stage 1 (Loredo et al. 2006). Consequently, sleep becomes more consolidated with an increase in the duration of REM sleep compared to the diagnostic study. REM rebound (Brillante et al. 2012; Koo et al. 2012; Kushida et al. 2011; Yaegashi et al. 2009; Osuna et al. 2008; Drake et al. 2003; Verma et al. 2001; Randerath et al. 2001; Parrino et al. 2000; Yamashiro and Kryger 1995; Lamphere et al. 1989; Aldrich et al. 1989; Issa and Sullivan 1986; Collard et al. 1996) is the polysomnographic phenomenon of substantial increase in duration of REM sleep in a patient with untreated OSA when first undergoing CPAP titration. Similarly, some patients may exhibit an increase in the duration of slow wave sleep called slow wave sleep rebound (Brillante et al. 2012; Osuna et al. 2008; Verma et al. 2001).

REM rebound on the first night of CPAP use has multiple clinical implications. Clinically, patients exhibiting REM rebound have reported better sleep quality on CPAP titration night than on nights with untreated OSA (Osuna et al. 2008). In new CPAP users early CPAP adherence was found to be higher in patients exhibiting significant REM rebound during CPAP titration night (Koo et al. 2012). Currently, there are no consensus guidelines as to how much increase in duration of REM sleep in the titration night over the baseline sleep study (or baseline portion of a split night study) qualifies as REM rebound. The primary objective of this study was to determine if there is a statistically significant increase in percentage of REM sleep during CPAP titration as compared to baseline polysomnogram (PSG), in patients undergoing CPAP titration after being diagnosed with OSA. Secondary objectives were to quantify the relative increase in REM sleep duration and to predict polysomnographic factors associated with REM rebound. In order to meet the objectives of this review, a systematic review of the literature was performed to identify studies reporting REM rebound on the CPAP titration night as compared to the baseline sleep study night, and the quantitative data was used to determine the percent increase in REM sleep duration using a meta-analysis with random effects modeling.

The search included Medical Subject Headings (MeSH) terms, key words, and phrases in combinations to obviate missing articles due to the use of select terminology in the different databases. To make the search thorough, hand searches of the reference lists of relevant articles were performed in order to identify other pertinent articles. Also, meticulous grey literature and Google Scholar searches were performed to identify relevant publications that could have been missed during the electronic database search. Search was restricted to English language articles only. An example of a PubMed search is: ((("Continuous Positive Airway Pressure"[Mesh] AND "Sleep, REM"[Mesh] OR ("Continuous Positive Airway Pressure"[Mesh] AND "Sleep, REM"[Mesh]) AND "Sleep Apnea, Obstructive"[Mesh], Increase* in REM sleep AND CPAP*, rapid eye movement sleep [tiab] AND CPAP [tiab]))) All articles were reviewed which discussed REM rebound in patients with OSA. Articles meeting the inclusion criteria were included in the systematic review.

Inclusion criteria using PICOS were: 1) Patients: those diagnosed with OSA, 2) Intervention: CPAP therapy, 3) Comparison: a) studies assessing amount of increase in REM sleep during CPAP titration as compared to amount of REM sleep in baseline PSG, b) and/or studies discussing slow-wave sleep rebound in addition to REM rebound, and c) studies looking for correlation between sleep quality or CPAP adherence and REM rebound, 4) Outcomes: the REM sleep duration differences on CPAP therapy during the titration night compared to the baseline PSG, and 5) Study design: randomized controlled trials, prospective and retrospective cohort studies, and case series. Exclusion criteria included: 1) Studies on sleep-disordered breathing (SDB) which exclusively discussed slow-wave sleep rebound, 2) Home sleep apnea studies with no electrooculography (EEG) and electromyography (EMG) monitoring capability, and 3) Individual case reports, editorials, review articles, and meeting abstracts.

Statistical evaluation was performed using Review Manager (REVMAN) Software version 5.3. The pre- and post-CPAP REM percentage, standard deviations (SD), mean differences (MD), 95% confidence intervals (CI) and p-values were calculated using the IBM Statistical Package for Social Sciences (SPSS) software. Combined mean differences and 95% CI were calculated only for studies reporting means and SD. The null hypothesis for the study was that there is no difference in percentages of REM sleep rebound between baseline PSG versus titration study and in order to test this hypothesis the data was analyzed using post minus pre CPAP therapy outcome data. A random effects modeling was utilized and the overall effect size estimation was performed by calculating the standardized mean difference (SMD). Cohen’s guidelines were used to determine the magnitude of the effect size, and the SMD cutoff values were: small = 0.2, medium = 0.5 and large = 0.8 (Cohen 1988). Heterogeneity was defined as a REVMAN Q-statistic value of ≤0.10 (Lau et al. 1997), and the REVMAN I² value cutoffs for inconsistency were 25% = low inconsistency, 50% = moderate inconsistency and 75% = high inconsistency (Higgins et al. 2003). The risk of bias was assessed (as recommended by the Cochrane Collaboration) by assessment of the funnel plots if there are at least ten studies in the variable of interest.

Of the 14 studies meeting inclusion criteria, only three studies (Brillante, Koo and Osuna et al. studies) defined what they considered as REM rebound. The remaining studies provided some measure of increase in REM sleep during titration night compared to baseline sleep study. Most studies reported REM rebound in percentage increase in REM sleep duration when comparing titration to baseline sleep study except two studies (Aldrich et al. and Parrino et al.) where the relative increase in REM sleep was reported only in minutes. On an average, percentage of REM sleep in diagnostic PSG varied from 6.7 ± 9.3% in Koo et al. study to 18.4 ± 2.0% in Issa et al. study. During the titration study, 11 of the 14 studies noted a statistically significant increase in amount of REM sleep compared to baseline sleep study, with two studies (Koo and Kushida et al.) not mentioning the statistically significant status. One study (Randerath et al.) reported that the increase was not statistically significant. REM sleep percentages during titration study varied between 17.1 ± 7.4 in Randerath et al. study to 30.6 ± 2.0 in Issa et al. study.

In studies reporting statistically significant increase in the percentage of REM sleep, the percentage of increase varied on average from 1.5% in Brillante et al. study to 14.1% in Yamashiro et al. study (Fig. 2). Although Koo et al. study reported an even higher increase in the average percentage of REM sleep at 16.6% (compared to Yamashiro study reporting an increase of 14.1%) it was not stated whether this increase was statistically significant, when considering all patients. Incidentally, the two studies with highest percentage increase in REM sleep duration during CPAP titration also represented the only split night studies of this review.

Mean differences and standardized mean differences

Based on polysomnography data, the CPAP REM percentage means ± standard deviations (M ± SD) in 1119 patients increased from 13.8 ± 8.2% to 20.0 ± 10.1% (a mean difference of 6.2%) which corresponds to a 44.6% relative percentage increase of REM sleep during the CPAP treatment compared to baseline (Table 1). Analysis of studies using random effects modeling was performed for REM rebound percentage on titration with mean difference (MD) 7.86% [95% CI 5.01, 10.70], overall effect Z = 5.41, p-value <0.00001, Q statistic p-value <0.00001 (significant heterogeneity), I² = 94% (high inconsistency) (Fig. 3). The mean difference of 7.9% corresponds to a 57% relative percentage increase of REM sleep during the CPAP treatment compared to baseline. The risk of bias was high as the funnel plot for REM rebound percentage mean difference was significantly asymmetrically shaped. The sensitivity analysis demonstrated two sub-groups of studies that were non-heterogenous when combined, the first non-heterogenous sub-group comprised the studies by Yaegashi et al., Drake et al., Verma et al., Randerath et al. and Collard et al. (Q-statistic p = 0.62 and I² = 0%); while the second sub-group comprised the studies by Yamashiro et al., Lampere et al. and Issa et al. (Q-statistic p = 0.54 and I² = 0%). The SMD is 0.90 [95% CI 0.59, 1.22], representing a large magnitude of effect (Cohen’s guidelines) with an overall effect Z = 5.60, p-value <0.00001, Q statistic p-value <0.00001 (significant heterogeneity), I² = 90% (high inconsistency).

Predictive factors for REM rebound

Of the 14 studies, five studies (Brillante et al., Koo et al., Osuna et al., Verma et al. and Aldrich et al.) discussed predictive factors for REM rebound on CPAP titration (Table 3). High AHI on baseline PSG was found to be a predictive factor in all five studies, however a direct linear correlation could not be conclusively established between baseline AHI and “Delta REM” (denoting difference between percentage of REM sleep in post CPAP study and percentage of REM sleep in pre CPAP study), when data from all studies reporting REM sleep percentage were included (Fig. 4). Inability to establish direct correlation could be related to variability in rapidity of up-titration of individual studies as well as exclusion of Verma et al. and Aldrich et al. studies from scatter dot diagram data due to computing limitations; Aldrich et al. reported REM sleep in minutes rather than as percentages and in Verma et al. study, only patients who reported improvement on CPAP were included for combined means. Brillante et al. study also noted male gender and high arousal index on baseline PSG to be associated with REM rebound. A low percentage of REM sleep on baseline PSG was likely to result in greater “Delta REM” or REM rebound as found in three studies (Brillante et al., Koo et al. and Verma et al.). It was further corroborated when this inverse relationship was depicted on the scatter dot diagram when data from all studies reporting REM sleep percentage were included (Fig. 5). In the study by Osuna et al. (average BMI: 35.5 kg/m²), obesity was associated with REM rebound; however, the study by Brillante et al. (average BMI: 33.3 kg/m²) failed to find a statistically significant association between body weight and REM rebound. Low oxygen saturation on baseline PSG was associated with REM rebound as noted by Aldrich et al.

Study	High AHI on baseline PSG	Low % of REM sleep on baseline PSG	High arousal index on baseline PSG	Male gender	Low oxygen saturation on baseline PSG	High BMI
Brillante et al.	✓	✓	✓	✓	-	✖
Koo et al.	✓	-	-	-	-	-
Osuna et al.	✓	✓	-	-	-	✓
Verma et al.	✓	-	-	-	-	-
Aldrich et al.	✓	✓	-	-	✓	-

✓: Denotes that the parameter was studied and was found to be statistically significant factor for REM rebound

✖: Denotes that the parameter was studied but was not found to be statistically significant factor for REM rebound

- : Denotes that the parameter was not studied

This systematic review and meta-analysis has five main findings. First, eleven of fourteen studies reported a statistically significant increase in REM sleep duration during titration night compared the baseline sleep study. While all studies report REM rebound, only a few studies define what REM rebound entails based on their statistical prediction models. Studies adopted cut-off criteria of 6%-20% increase in percentage REM sleep on titration night when compared to baseline PSG to define REM rebound. Accepting a lower REM percentage increase as the cut -off criteria for REM rebound (when comparing REM percentage in titration study to REM percentage in baseline PSG study) leads to reporting a higher prevalence of REM rebound on CPAP titration. Accordingly, based on cut-off criteria used to define REM rebound, the prevalence of REM rebound varies between 23 and 46% in all patients undergoing CPAP titration study.

Second, when the specific cut-off criteria (in terms of duration of increase in REM sleep percentage) defining REM rebound are overlooked, the majority of patients undergoing a CPAP titration experience increased duration of REM sleep compared to their baseline sleep study. The percentage increase in REM sleep duration varies between 1.5 and 14% and may depend on certain polysomnographic and demographic factors. The mean difference of 7.86% corresponds to a 57% relative percentage increase of REM sleep during the CPAP treatment night compared to the baseline study. When evaluating the effect size based on Cohen’s guidelines, the SMD is 0.90 [95% CI 0.59, 1.22], which is defined as a large effect. Interestingly, the sensitivity analysis demonstrated two sub-groups of studies that were non-heterogenous when combined, and the subgroups tended to be grouped by the year of the study publication; with the first non-heterogenous sub-group consisted of studies from 1996 to 2009 (Yaegashi et al., Drake et al., Verma et al., Randerath et al. and Collard et al.) with a Q-statistic p = 0.62 and I² = 0%. The second non-heterogenous sub-group consisted of studies from 1986 to 1995 (Yamashiro et al., Lampere et al. and Issa et al.) with a Q-statistic p = 0.54 and I² = 0%. It is unclear as to why these two sub-groups were non-heterogenous when grouped alone, but did cause heterogeneity and inconsistency when grouped as part of the fourteen studies.

Third, a high percentage increase in REM sleep was observed specifically during split night studies. A relatively high proportion of REM sleep during split night studies (compared to full night titration studies) could be multifactorial, associated with at least three factors: (A) the natural timing of REM sleep, (B) the rapidity of up-titrations and (C) the baseline severity of OSA. First, the PAP titration portion of split night study temporally coincides with timing for natural preponderance of REM sleep in the latter half of sleep period (Ciftci et al. 2008). Second, in contrast to dedicated, full night titration studies, during split-night studies, sleep technicians might be up-titrating the pressures relatively rapidly due to limited titration window leading to early, brief but repetitive bursts of saw tooth waves or low voltage mixed frequency REM spikes. Lastly, patients who undergo split night study typically have severe OSA. In such patients both high baseline AHI and high arousal index from sleep fragmentation, are being simultaneously resolved with PAP therapy, which may be promoting additional periods of restorative REM sleep. Immediate increased REM duration from CPAP use as early as after one night have been observed (Loredo et al. 2006; Issa and Sullivan 1986). So, it is plausible that the rebound REM increase occurs even within a few hours of initiation of CPAP, given that untreated OSA represents a state of several years of chronic partial sleep deprivation as well as sleep fragmentation.

We speculate that the quantity of REM rebound could depend upon rapidity of up- titrations as well how quickly the patient adapted to these up-titrations. In some studies, rapid CPAP up-titration may have led to early attainment of therapeutic pressure providing more time during sleep period to exhibit REM. In other studies, a more conservative and cautious approach may have lead to at least some REM sleep attained chronologically prior to attainment of therapeutic CPAP pressure leading to less REM rebound. Of note, about 5-20% patients may have experienced treatment emergent central sleep apnea or TECSA on first exposure to CPAP during these studies (Nigam et al. 2016). TECSA may have lead to increased arousal index and sleep fragmentation thereby limiting REM rebound which would be subsequently seen after several weeks of CPAP use in this cohort of CPAP users.

Fourth, certain polysomnographic and demographic factors could enhance REM rebound. Two such factors predicting enhanced Delta REM (and thereby predicting substantial REM rebound) as suggested in several studies include: 1) a lower amount of REM sleep on diagnostic PSG (strong association) and 2) a higher severity of OSA as noted by the AHI during diagnostic PSG (weak association, if any). Other possible associations include higher arousal index on baseline PSG, low oxygen saturation on baseline PSG, male gender, higher BMI and utilization of split-night PSG to study REM rebound. Additional research is needed to further evaluate the variables that predict which patients are more likely to experience REM rebound.

Fifth, almost half of the currently published studies were retrospective observational studies and prone to limitations of any retrospective study. There were differences amongst studies as to the definition of REM rebound. The sleep stage and respiratory events scoring rules have undergone considerable changes over the last few decades. Given the review includes studies conducted over last 30 years, the majority of the included studies were scored using Rechtschaffen and Kales manual for the scoring of sleep stages, while others were scored based on the American Academy of Sleep Medicine scoring manual introduced in 2007. The rules for scoring respiratory events have undergone considerable changes as well over the years, making it difficult to directly compare baseline AHIs reported by different authors; however, given that REM is defined based on the electroencephalogram, we believe that despite the differences in the AHI scoring criteria, the percentage of sleep scored for each of the sleep stages, including REM, should not change.

Significant increase in REM sleep duration can occur in the first CPAP titration night compared to REM sleep duration noted on the baseline sleep study. In studies reporting REM rebound, the REM sleep duration increased by 57% during the first CPAP titration night compared to the baseline sleep study. Although no consensus guidelines exist, various retrospective studies have outlined that 6-20% increase in percentage of REM sleep on titration night compared to baseline PSG could be indicative of REM rebound. The most crucial polysomnographic factor that predicts and promotes REM rebound is low amount of REM sleep attained on the diagnostic PSG. In order to increase the level of evidence, additional high quality, prospective studies will be required in the future.