The noninferiority complex: abrupt versus gradual restriction of time in bed for insomnia disorder

Contemporary evidence suggests that sleep restriction therapy (SRT) may be the most effective single component cognitive-behavioral treatment for insomnia disorder [1, 2]. While relatively simple to deliver, SRT is challenging for many patients and has been associated with a transient worsening of symptoms and functioning during implementation [3–5]. Importantly, SRT has not been shown to increase the incidence of serious adverse events [6, 7], but the prospect of curtailed time in bed (TIB) and the short-term consequences of sleep loss are key reasons for non-adherence and treatment discontinuation [7–9].

A 2015 review suggested that a research priority should be to compare different types of TIB manipulation to advance understanding of behavioral therapy and guide refinement [10]. Recent work has shown that restriction plus regularization of TIB (“classic SRT”) outperforms regularization on its own, showing that TIB restriction is active [11]. In this issue of SLEEP, Jernelöv, Rosén, and colleagues [12] make a significant and novel contribution to the field by evaluating whether sleep compression (SC) is noninferior to SRT. The trial was designed to test the hypothesis that SC was “not appreciably worse” than the established SRT configuration. The logic was that gradual reduction of TIB may be more palatable and tolerable for patients, translating to improved adherence, treatment satisfaction, daytime functioning, and side-effects while not materially sacrificing the overall treatment effect.

SC was introduced by Dr Kenneth Lichstein in the 1980s [13] as a treatment principally for people who experience a mismatch between sleep opportunity and sleep need (but a general absence of excessive sleepiness or substantial daytime impairment). Thus, the aim of treatment was to gradually reduce TIB over successive weeks but without formal titration based on sleep efficiency. The assumption was that SC would be well-suited for people who may be getting sufficient sleep, albeit in a fragmented way, and that the end goal was a long-term adjustment of TIB to better match sleep requirement rather than prolongation of sleep. It follows, perhaps, that the few published studies on SC have been performed in elderly adults with insomnia [14].

Jernelöv et al. adapted SC for a head-to-head comparison with SRT. They randomized (1:1) 234 patients with insomnia to either 10 weeks of therapist-guided online SRT or SC. Treatments were identical in format and content, differing only on the specific TIB guidelines. Patients worked through online modules, entered daily sleep diary data, and received TIB instructions each week alongside written guidance and encouragement from therapists. One group followed a relatively standard SRT protocol, while the SC group was asked to initially reduce TIB by 20% of total wakefulness followed by weekly adjustment based on sleep efficiency thresholds (see Table 1). In contrast to most of the research and clinical literature, there was no minimum sleep prescription for either arm, which could be relevant when interpreting side-effect findings.

Table 1.

Parameters for setting and adjusting TIB for each treatment arm (extracted from Appendix 1; Jernelov et al. [12])

Parameter	Sleep Restriction Therapy	Sleep Compression
Initial TIB prescription	Mean TST from diary (≥10 nights)	Reduce by 20% of mean total wake-time from diary (≥10 nights)
Minimum TIB	None
Position of window	According to work schedule and patient preference. Goal was to fix rise-time and make adjustments to bedtime.
SE thresholds and TIB adjustment	≥90% over 7 nights increase by 15 minutes 85%–89% over 7 nights keep the same  < 85% over 7 nights reduce by 15 minutes	≥90% over 7 nights increase by 15 minutes 85%–89% over 7 nights keep the same  < 85% over 7 nights reduce by 20% of mean total wake-time
Napping	Recommendation was not to nap. If necessary, nap duration < 15 minutes at the same time of day before 3 pm. Nap time was not included in sleep window calculations.

SE = sleep efficiency; TIB = time in bed; TST = total sleep time.

The primary outcome was the mean difference between groups on the ISI at 10 weeks, while 5 weeks was also reported as a secondary end-point. In a noninferiority trial, a margin is defined a priori, and for one to conclude “noninferiority,” the upper bound of the confidence interval needs to be within this margin. At the outset of the trial, the authors selected a 4-point noninferiority margin, but this was later changed during the trial, though prior to data analysis. While changing the margin in this manner is generally discouraged, the authors transparently reported the change and provided justification (the evolving evidence-base since trial commencement). A margin of 1.6 points on the ISI was decided upon, reflecting 50% of the lower bound of the confidence interval for the pooled meta-analytic effect of SRT versus control [1]. Other outcomes in the trial included patient-attributed side-effects, self-reported and therapist-rated adherence, and actigraphy- and diary-measured sleep throughout the treatment period.

Engagement and satisfaction with treatment were both high and similar across trial arms, as was treatment credibility. Attrition for the primary outcome was relatively low and comparable between arms. Consistent with the instructions for each treatment, SRT participants had a greater and more rapid reduction of TIB (and TST) and reported increased sleepiness at bedtime. Both groups showed large reductions in ISI scores from baseline to weeks 5 and 10. The mean difference was 1.51 (CI:0.75-2.27) points lower for SRT at week 5 and 0.84 (CI: 0.04–1.73) points lower at week 10 relative to SC. Thus, noninferiority of SC was not demonstrated at week 10 (nor week 5) because the upper bound of the confidence interval exceeded the margin—delivering an “inconclusive” result—and there was evidence that SC was statistically worse than SRT at week 5. Interaction tests from the linear mixed model supported this overall pattern of findings with a larger and more rapid rate of ISI change in the SRT arm. In general, a similar pattern of accelerated symptom improvement for SRT was observed for sleep quality and sleep continuity parameters (from diary), though group differences were not readily apparent at the end of treatment (at least judged by descriptive comparison of outcomes on the final night of treatment; see Table S12 in Jernelov et al [12]). The SC group reported less side-effect burden during early treatment (week 2, though not at weeks 4 and 5), and better adherence on two of three metrics examined.

What should one conclude from this trial? The authors emphasize that, in general, SRT should be prioritized over SC, but there may be instances where SC could be more appropriate for individual patients. This seems sensible, and indeed, both treatments showed marked reductions in insomnia. In the introduction, the authors make the point that SC may be particularly well-suited for vulnerable populations, yet the recruited sample had relatively low levels of anxiety and depression (exclusions included depression, bipolar disorder, and sleep-altering medications), and they tended to be well-educated. A head-to-head comparison in a more diverse group of patients, or patients with specific co-morbidities, would be informative and could unmask additional benefit. In such a trial, it would be prudent to decide on whether SC is being tested as an alternative treatment option or as a replacement [15], and if the latter, what would be the acceptable trade-off between (minimal) loss of efficacy and gain in other areas, such as side-effects. A particular challenge here is that assessment of side-effects in trials of behavioral therapies is still in its infancy and there is no agreement on what might constitute an important difference. In future work, it will be important to further develop side-effect measurement [16] and, in prospective studies, include systematic assessments of serious and nonserious adverse events [6] as well as measures that capture change in functioning from baseline. Two participants in the SRT arm (versus none in SC) reported adverse events that they deemed related to treatment, and which could be considered “serious”—a minor car accident and blurred vision requiring medical attention. It should be kept in mind that there was no minimum TIB applied in this study, which would be expected to strengthen the degree of sleep loss in the SRT arm. Another important point is that clinician-administered SRT involves ongoing assessment of and responsiveness to any patient concerns such that “…clinical judgement trumps strict adherence to the SRT method” [17]. While not the focus of the present study, whether online delivery of SRT provides the same degree of responsivity, flexibility, and tailoring is unclear.

This important trial raises several interesting conceptual questions and stimulates future studies. Despite evidence of greater side-effect burden and lower adherence in the SRT arm, there was no detrimental effect on daytime functioning and SRT maintained advantage with respect to speed of sleep improvement. It follows, therefore, that the field should design and test approaches to support patients during SRT implementation aimed at minimizing side-effects and maximizing adherence. While requiring dedicated evaluation, it could be that important cognitive changes brought about by a marked reduction of TIB (and sleep time) contributes to a proportion of the treatment effect [18]. Indeed, this would be consistent with recent preliminary findings from a small trial showing, perhaps surprisingly, that just one night of total sleep deprivation in the laboratory delivered improvement in insomnia severity 3 months later [19]. Ultimately, for a more complete explanation of the mechanisms of SRT (and SC) frequent and long-term assessment of symptoms, arousal, sleep, and circadian physiology, and cognitive-behavioral processes will be required [4, 18, 20]. From a clinical perspective, it will also be relevant to understand how TIB therapies interact and sequence with other cognitive-behavioral therapies. In this context, we look forward to the results of the subsequent extension phase of the trial by Jernelöv, Rosén, and colleagues, which involved randomizing patients within each arm to additional treatment components or no additional treatment (NCT02743338).

Disclosure statements

Nonfinancial disclosure: S.D.K. declares nonfinancial support in the form of no-cost access to use Sleepio (Big Health. Ltd.) in clinical research. C.J.G. was an original inventor of SleepFix and has free access for use in research. Financial disclosure: Nothing to disclose.

Funding

S.D.K. reports current grant support from the Wellcome Trust (refs: 227684/Z/22/Z and 227093/Z/23/Z), the National Institute for Health and Care Research (NIHR) (refs: EME131789 and NIHR203667), and the Oxford Health NIHR Biomedical Research Centre (ref: NIHR203316). The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care. C.J.G. reports current grant funding from the National Health and Medical Research Council (NHMRC, GNT2018668) and the Medical Research Future Fund (MRFF, GNT2035692).