The effect of sleep restriction therapy for insomnia on REM sleep fragmentation: A secondary analysis of a randomised controlled trial

Rapid eye movement sleep fragmentation is hypothesised to be a reliable feature of insomnia, which may contribute to emotion dysregulation. Sleep restriction therapy, an effective intervention for insomnia, has the potential to reduce rapid eye movement sleep fragmentation through its manipulation of basic sleep–wake processes. We performed secondary data analysis of a randomised controlled trial to examine whether sleep restriction therapy reduces rapid eye movement sleep fragmentation in comparison to a matched control arm. Participants (n = 56; 39 female, mean age = 40.78 ± 9.08 years) were randomly allocated to 4 weeks of sleep restriction therapy or 4 weeks of time in bed regularisation. Ambulatory polysomnographic recordings were performed at baseline, week 1 and week 4. Arousals during rapid eye movement and non‐rapid eye movement sleep were scored blind to group allocation. The following rapid eye movement sleep fragmentation index was the primary outcome: index 1 = (rapid eye movement arousals + rapid eye movement awakenings + non‐rapid eye movement intrusions)/rapid eye movement duration in hours. Secondary outcomes were two further indices of rapid eye movement sleep fragmentation: index 2 = (rapid eye movement arousals + rapid eye movement awakenings)/rapid eye movement duration in hours; and index 3 = rapid eye movement arousals/rapid eye movement duration in hours. A non‐rapid eye movement fragmentation index was also calculated (non‐rapid eye movement arousals/non‐rapid eye movement duration in hours). Linear‐mixed models were fitted to assess between‐group differences. There was no significant group difference for the primary rapid eye movement fragmentation index at week 1 (p = 0.097, d = −0.31) or week 4 (p = 0.741, d = −0.06). There was some indication that secondary indices of rapid eye movement fragmentation decreased more in the sleep restriction therapy group relative to control at week 1 (index 2: p = 0.023, d = −0.46; index 3: p = 0.051, d = −0.39), but not at week 4 (d ≤ 0.13). No group effects were found for arousals during non‐rapid eye movement sleep. We did not find clear evidence that sleep restriction therapy modifies rapid eye movement sleep fragmentation. Small‐to‐medium effect sizes in the hypothesised direction, across several indices of rapid eye movement fragmentation during early treatment, demand further investigation in future studies.

+ non-rapid eye movement intrusions)/rapid eye movement duration in hours.Secondary outcomes were two further indices of rapid eye movement sleep fragmentation: index 2 = (rapid eye movement arousals + rapid eye movement awakenings)/ rapid eye movement duration in hours; and index 3 = rapid eye movement arousals/ rapid eye movement duration in hours.A non-rapid eye movement fragmentation index was also calculated (non-rapid eye movement arousals/non-rapid eye movement duration in hours).Linear-mixed models were fitted to assess between-group differences.There was no significant group difference for the primary rapid eye movement fragmentation index at week 1 (p = 0.097, d = À0.31) or week 4 (p = 0.741, d = À0.06).There was some indication that secondary indices of rapid eye movement fragmentation decreased more in the sleep restriction therapy group relative to control at week 1 (index 2: p = 0.023, d = À0.46;index 3: p = 0.051, d = À0.39),but not at week 4 (d ≤ 0.13).No group effects were found for arousals during non-rapid eye movement sleep.We did not find clear evidence that sleep restriction therapy modifies rapid eye movement sleep fragmentation.Small-to-medium effect sizes in the hypothesised direction, across several indices of rapid eye movement fragmentation during early treatment, demand further investigation in future studies.(Maquet et al., 1996).Another feature of REM sleep is the inhibition of the locus coeruleus (LC), a nucleus in the pons that releases norepinephrine during wakefulness and non-REM (NREM) sleep to promote long-term potentiation (Aston-Jones & Bloom, 1981).The silencing of the LC before NREM sleep spindles and during REM sleep is hypothesised to enable synaptic plasticity and aid memory consolidation (Swift et al., 2018).Incomplete LC silencing during REM sleep has been associated with EEG arousals and shifts to N1/N2 sleep stages and wakefulness (Aston-Jones et al., 2007).
Individuals with insomnia relative to controls experience more micro-arousals and awakenings during REM sleep (Feige et al., 2008; although see Feige et al., 2018 andKao et al., 2021 for conflicting findings).Insomnia is therefore suggested to be characterised by REM sleep instability (Riemann et al., 2012) or restless REM sleep (Wassing et al., 2019a).Furthermore, interrupted REM sleep (cortical arousals and bouts of wakefulness/NREM sleep) attenuates the adaptive overnight decrease in amygdala reactivity (Wassing, Lakbila-Kamal, et al., 2019), a process linked to the overnight resolution of emotional distress (van der Helm et al., 2011), which may be dysfunctional in people with insomnia (Wassing et al., 2016).Together, these results suggest that unstable REM sleep may play a role in the pathophysiology of insomnia, interfering with basal processes of emotion regulation (Riemann et al., 2012), and potentially contributing to the development of common mental health problems like anxiety and depression (Hertenstein et al., 2019).
It has been hypothesised that curtailing time in bed (TIB; Van Someren, 2021) and reducing pre-sleep arousal (Vethe et al., 2022) may improve REM sleep consolidation.Sleep restriction therapy (SRT), an effective single component treatment for insomnia (Kyle et al., 2023;Maurer et al., 2021), systematically reduces TIB in order to consolidate and regularise sleep.We previously reported on a randomised controlled trial comparing SRT with a matched control arm (TIB regularisation; TBR).SRT was found to reduce insomnia symptoms and improve sleep continuity (Maurer et al., 2020), change sleep architecture during early intervention (reduced time spent in N1/N2 sleep), enhance sleep pressure and reduce pre-sleep arousal (Maurer et al., 2022).Given these findings, we performed unplanned exploratory analyses to assess for the first time if SRT reduces REM sleep fragmentation.

| Study design and participants
The MARTINI trial (ISRCTN10974094) was a parallel, randomised, controlled evaluation of SRT versus TBR.Fifty-six participants (39 female, mean age = 40.78± 9.08 years) meeting criteria for DSM-5 chronic insomnia disorder were assigned to 4 weeks of SRT (n = 27) or TBR (n = 29) after completing a baseline phase of 2 weeks.Exclusion criteria included: psychiatric diagnoses, symptoms of depression or anxiety (questionnaire-based cut-offs), alcohol misuse, other sleep disorders, CNS medication, shift work, and pregnancy.Both interventions were delivered across weekly 1:1 sessions.Bed-and rise-times were selected according to treatment-specific criteria.In the SRT group, participants were prescribed a sleep window that matched their average self-reported total sleep time (TST) for the previous 2 weeks (minimum TIB = 5 hr).In the TBR group, participants were prescribed a sleep window that matched their average self-reported TIB for the previous 2 weeks.More details can be found in the previous publications of this trial (Maurer et al., 2020(Maurer et al., , 2022)).Polysomnography (PSG), from which REM sleep fragmentation was derived, was assessed at baseline, and at week 1 and week 4 post-randomisation alongside other study outcomes (see Maurer et al., 2020 for details).
The study was conducted in Oxfordshire, UK, and approved by the Central University Research Ethics Committee of the University of Oxford (CUREC R51331/RE005).The objectives for the present report were exploratory, but we registered our intentions with the Open Science Framework (10.17605/OSF.IO/5GR8A) prior to scoring of arousals and data analysis.

| Statistical analyses
Available data from all randomised participants were analysed according to the intention to treat principle.For between-group comparisons, linear mixed-effects regression models were fitted for each of the fragmentation indices, with fixed effects of group and time point.Time point was entered as a repeated variable and participant as a random effect.Baseline values were added as covariate.An interaction between time point and group was included to estimate treatment effects at each time point.The covariance structure was set to unstructured.Cohen's d statistics were calculated as the adjusted treatment effect divided by the baseline standard deviation of the outcome for the combined groups (Cohen, 1988).Summary statistics are presented in the form of means and standard deviations.All analysis was conducted in SPSS.27 (IBM).

| RESULTS
One participant assigned to SRT withdrew from intervention and study procedures before the week 1 assessments.All other participants completed all treatment sessions and all assessments, yielding a total of 166 PSG recordings across groups and time points.Of these, two recordings were not suitable due to data loss (one baseline [TBR] and one week 1 [TBR] recording).Both groups adhered to their prescribed sleep window as indicated by continuous sleep diary reports during the intervention: average bed-and rise-times deviated between 0 and 18 min for SRT, and between 5 and 14 min for TBR, when compared with prescribed bed-and rise-times (Maurer et al., 2020).There were also large between-group differences in sleep opportunity, with both actigraphy and diary data showing lower TIB in the SRT group versus control (Maurer et al., 2020).
To test whether SRT reduces REM sleep fragmentation, we first conducted between-group comparisons on the REM sleep fragmentation index (index 1 = [REM arousals + REM awakenings + NREM intrusions]/REM duration in hours) at week 1 and week 4 (see Table 1).The were prominent during early treatment (Maurer et al., 2020(Maurer et al., , 2022)).
High-resolution studies, assessing nightly change in pre-sleep (cognitive) arousal and REM fragmentation during SRT implementation, would be needed to appraise potential causal pathways and temporal dynamics.
Limitations of this analysis need to be considered.First, this was a secondary analysis of previously collected data and therefore the study was not designed to investigate changes in REM specifically.
While the analysis plan was registered prior to scoring of arousals, K E Y W O R D S insomnia, randomised controlled trial, rapid eye movement sleep, sleep restriction therapy 1 | INTRODUCTION Rapid eye movement (REM) sleep is the most physiologically aroused brain state during sleep, and is characterised by muscle atonia, desynchronised electroencephalographic (EEG) activity, rapid eye movements, and dream mentation by a trained sleep scorer (LFM).Reliability of sleep stage scoring was confirmed by an ESRS-accredited Somnologist, who double-scored 10% of all recordings (agreement = 91.5%).The EEG arousals during NREM and REM sleep were scored according to scoring recommendations by the Atlas task force of the American Sleep Disorder Association (Atlas task force, 1992).Because there is currently no standard for indexing REM sleep fragmentation in insomnia, multiple indices were derived.Of primary interest was the REM sleep fragmentation index (Wassing et al., 2019a), which included the number of cortical arousals and bouts of wakefulness or NREM sleep that interrupted REM episodes, the sum of which is then divided by the duration of all-night REM sleep (index 1 = [REM arousals + REM awakenings + NREM intrusions]/REM duration in hours).If a REM episode was interrupted by multiple shifts (e.g.wake and N1), only the first shift out of the REM epoch was counted.The combined index of REM arousals and awakenings from REM sleep per hour of REM sleep (REM arousal index) has been identified as the most sensitive to differentiate between patients with insomnia and good sleepers SRT group showed a small and non-significant reduction in REM fragmentation index relative to control at week 1 (p = 0.097, d = À0.31).No between-group effect was found at week 4 (p = 0.741, d = À0.06).For the REM arousal index (index 2 = [REM arousals + REM awakenings]/ REM duration in hours), the SRT group showed a statistically significant reduction at week 1 (p = 0.023, d = À0.46)but no effect was observed at week 4 (p = 0.523, d = À0.13).For index 3 (REM arousals/REM duration in hours), there was a medium-sized effect at week 1 (d = À0.39) in support of SRT, but this did not reach statistical significance (p = 0.051).Again, no effect was found at week 4 (p = 0.908, d = À0.02).There were no group effects for number of NREM arousals (index 4) at week 1 (p = 0.395, d = À0.11) or week 4 (p = 0.153, d = À0.30).4 | DISCUSSIONTo our knowledge, this is the first study to examine whether SRT reduces REM sleep fragmentation, a proposed sleep characteristic of insomnia, in comparison to a matched control group.In the absence of a standard REM sleep fragmentation index, multiple indices were considered.We did not find clear evidence of a treatment effect of SRT on REM sleep fragmentation, with a statistically significant reduction being observed for just one of the three REM indices at one time point.However, small-to-medium effect sizes in favour of SRT across all indices suggest a potential reduction of REM sleep fragmentation at week 1 in comparison to TBR (d range = À0.31 to À0.46).The absence of group effects at week 4 suggests that any treatment effect due to SRT may be of a temporary nature and potentially driven by mild sleep deprivation and/or reductions in pre-sleep arousal, both of which and the scorer was blind to treatment allocation and time point, other clinical and mechanistic outcomes, including sleep, were already known by the authors when performing this research.Second, effect sizes and p-values indicate that this study, which was designed to detect large effect sizes, may have been underpowered for appraising group differences in REM (and NREM) fragmentation variables.Third, the comparison group (TBR), while inferior to SRT(Maurer et al., 2020(Maurer et al., , 2022)), may be therapeutically active through regularisation of TIB; and, therefore, our comparison could potentially underestimate the true effect of SRT on REM fragmentation.Future studies with a third arm (e.g.sleep hygiene or no treatment) would be needed to address this issue.Fourth, we analysed three correlated indices without adjustment for multiple testing.To let the reader judge, we provide exact statistical values and confidence intervals around the adjusted means.Fifth, data presented were collected as part of a mechanistic trial, which had strict inclusion and exclusion criteria.Consequently, results may not apply to the broader insomnia population.In summary, results from this exploratory analysis indicate that SRT may have potential to decrease REM sleep fragmentation, a proposed manifestation of hyperarousal, but only during the acute treatment phase.Future, purposely-designed studies are warranted to assess whether SRT reliably (and selectively) modifies REM fragmentation, and whether changes relate to clinical improvement in symptoms and/or overnight emotion processing.To advance research, the field also needs to establish a clear definition of what constitutes marked or clinically relevant REM sleep fragmentation, and agree on a measurement hierarchy for outcome evaluation.
Effects of SRT versus TBR on REM sleep fragmentation indices Note: Cohen's d.M and SD refer to unadjusted means and standard deviations.Abbreviation: 95% CI, 95% confidence interval of the adjusted mean difference; Diff adj , adjusted mean difference derived from linear mixed model; ES, effect size; NREM, non-rapid eye movement; REM, rapid eye movement; SRT, sleep restriction therapy; TBR, time in bed regularisation.a (REM arousals + REM awakenings + NREM intrusions)/REM duration in hours.b (REM arousals + REM awakenings)/REM duration in hours.c REM arousals/REM duration in hours.d NREM arousal/NREM duration in hours.MAURER ET AL.