Paradoxical intention has been considered an evidence-based treatment for insomnia since the 1990s; it seems to result in great reductions in sleep-related performance anxiety & marked clinical improvements

Paradoxical intention for insomnia: A systematic review and meta-analysis. Markus Jansson-Fröjmark, Sven Alfonsson, Benjamin Bohman, Alexander Rozental, Annika Norell-Clarke. Journal of Sleep Research, August 17 2021. https://doi.org/10.1111/jsr.13464

Summary: Paradoxical intention (PI) has been considered an evidence-based treatment for insomnia since the 1990s, but it has not been evaluated with modern review techniques such as meta-analysis. The present study aimed to conduct the first systematic review and meta-analysis of studies that explore the effectiveness of PI for insomnia on insomnia symptomatology and theory-derived processes. A systematic review and meta-analysis was conducted by searching for eligible articles or dissertations in six online bibliographic databases. Randomised controlled trials and experimental studies comparing PI for insomnia to active and passive comparators and assessing insomnia symptoms as outcomes were included. A random effects model was estimated to determine the standardised mean difference Hedge’s g at post-treatment. Test for heterogeneity was performed, fail-safe N was calculated, and study quality was assessed. The study was pre-registered at International Prospective Register of Systematic Reviews (PROSPERO, CRD42019137357). A total of 10 trials were identified. Compared to passive comparators, PI led to large improvements in key insomnia symptoms. Relative to active comparators, the improvements were smaller, but still moderate for several central outcomes. Compared to passive comparators, PI resulted in great reductions in sleep-related performance anxiety, one of several proposed mechanisms of change for PI. PI for insomnia resulted in marked clinical improvements, large relative to passive comparators and moderate compared to active comparators. However, methodologically stronger studies are needed before more firm conclusions can be drawn.

4 DISCUSSION

4.1 Summary of main results

The present study is the first comprehensive systematic review and meta-analysis of the effectiveness of PI for insomnia. Relative to passive comparators, PI resulted in large improvements in several central insomnia symptoms. Although the effectiveness of PI was smaller compared to active comparators, the effects were still moderate for several key outcomes. Relative to previous reviews, the present study extends the quantitative assessment of PI as an evidence-based intervention in that it compared PI with passive versus active comparators and included both night-time and daytime symptoms (Jansson-Fröjmark & Norell-Clarke, 2018; Morin et al., ,1999, 2006). A unique finding was support for great reductions in sleep-related performance anxiety by PI. This finding strengthens the notion that decreased performance anxiety is a mechanism through which PI might work.

Cumming and Finch (2001) have recommended that effect sizes should be compared to other relevant estimates in the literature to grasp their significance. In one of the largest and more recent meta-analysis, cognitive and behavioural interventions (e.g. CBT-I, relaxation, stimulus control, psychoeducation, and sleep restriction) were compared with passive comparators (van Straten et al., 2018). Comparing the effect sizes from van Straten et al., (2018) for cognitive and behavioural therapies with the present study’s effect sizes for PI relative to passive comparators, the effects were larger in the present study for PI on SOL (0.57 versus 0.82), NAW (0.28 versus 1.10), and TST (0.16 versus 0.51), and smaller on SE (0.71 versus 0.00). Although inferences from comparisons of this sort are difficult to draw from a methodological viewpoint, a reasonable conclusion would be to state that PI tentatively has a similar effectiveness as other cognitive and behavioural interventions. At the same time, this conclusion is hampered by several limitations in the trials exploring the effectiveness of PI. The relatively few studies, limited number of study participants, and other methodological characteristics of the studies makes an overall conclusion about effectiveness and generalisability of PI uncertain.

4.2 Methodological considerations and quality of evidence

The present review identified 10 studies that evaluated the effectiveness of PI. There were a number of notable methodological limitations of the studies. The study quality assessment showed that the quality of the 10 studies ranged from 15 to 20 points out of 26, implying a moderate study quality. The methodological quality was particularly weak in two areas. First, no studies reported using blinding of subjects, even though it appeared as if this would have been possible. Second, it was uncommon that studies appeared to have sufficient power to detect group differences. While some of these limitations were noted in the study quality assessment, others will be underscored more specifically below.

Across the 10 studies, there was diversity concerning the design. In nine trials, PI was compared with a passive comparator, which means that non-specific factors (e.g. therapist contact) were not controlled for in the estimations comparing PI with passive comparators. Concerning design, it is also worth underscoring that the aggregation of various active comparators into one active comparator category was based on that they provided study participants with active treatment content. This aggregation could, however, have resulted in that comparators with differing effects were combined, so that the comparison between PI and active comparators becomes uncertain.

Another limitation regards the patient characteristics. The total sample size was limited to <400 participants, and none of the trials reported that power calculations were made prior to study start. In all, Type 2 errors are likely, particularly when active treatments were compared. Further, all participants were recruited from the community, which might make the present findings less generalisable to health settings, as patients in clinical settings tend to display elevated symptoms (Davidson et al., 2009). Another observation is that, in almost all of the studies, we categorised the participants as meeting criteria for sleep-onset insomnia or primary insomnia. Therefore, it is uncertain whether PI should be viewed as an effective intervention for other types of insomnia, such as comorbid insomnia. It is also worth noting that there might be specific insomnia profiles that are particularly susceptible to PI. For example, Espie et al., (2006) have proposed that PI might be specifically suited for patients with psychophysiological insomnia, as this profile of patients are believed to be characterised by attentional bias, preoccupation with sleep, and using several strategies to avoid sleeplessness. In future research, the study of PI and the effectiveness for different insomnia profiles might also be based on recent empirical attempts to subtype insomnia (Blanken et al., 2019). On a related note, we observed that comorbidity was not formally assessed in the included studies. Although several studies used certain criteria to assess and/or exclude comorbidity, the lack of validated assessments of psychiatric and somatic conditions limits generalisability. As comorbid problems are more common than “pure” insomnia (Stepanski & Rybarczyk, 2006), the lack of assessing comorbid conditions and exclusion of participants with comorbid problems are problematic.

Another issue of methodological uncertainty concerns the administration of PI. There were slight variations concerning several features of the delivery. The rationale and instructions varied across studies, although the original approach by Ascher and Efran (1978) was most commonly employed. Also, the delivery format was mixed, with individual, self-help, and group formats identified. Further, in several treatment-related parameters, it was rare that sufficient information was provided; this concerned whether a treatment manual was used, who delivered PI, whether the therapists were trained and/or supervised, and whether treatment integrity was assessed. Also, the dose of PI varied across studies. Often, PI was delivered across 2–4 weeks, but longer treatment periods were also identified. Based on the limited number of studies in the present review, we were unable to investigate whether certain formats of delivery of PI was more effective than others. During the review process, we also noted that none of the studies assessed treatment-relevant domains that might have importance for the interpretation of findings, such as acceptability, adherence, credibility and expectancy ratings, and perceived usefulness of PI. It should also be emphasised that worsened sleep after PI has been reported in the research literature (Espie & Lindsay, 1985). As none of the included studies in the present review reported on adverse events or deterioration, more research is warranted to examine whether PI produces negative effects among patients with insomnia in general or in subgroups of patients.

An inclusion criterion for the present review was that trials must report insomnia-related outcomes (i.e. night-time and/or daytime symptoms). Across studies, it was less common to index objective sleep outcomes, daytime symptoms, theory-derived processes, and global insomnia symptoms [e.g. with the Insomnia Severity Index; (Bastien et al., 2001)]. Due to the lack of studies assessing several outcome domains, all meta-analytical estimations were based on sleep diary or questionnaire data assessing sleep performance anxiety. As a result, we can only draw conclusions for PI concerning sleep diary-assessed night-time symptoms and, to a lesser extent, sleep performance anxiety. A related limitation is that estimations of effectiveness for PI was not possible to assess in the longer term, as there were not sufficient data for such calculations.

A further limitation is that sensitivity and moderator analyses were not employed due to the limited number of studies. For example, it would have been interesting to explore the effects of the addition or removal of lower quality studies and, to examine whether insomnia symptomatology at baseline and PI administration might moderate the effectiveness of PI. A final limitation is that it was required that the included studies were published in English, thereby introducing a possible language bias.

4.3 Putative mechanisms

In the present study, we identified three studies that assessed sleep-related performance anxiety as a putative mechanism, and no trial indexing other potential mechanisms (e.g. sleep intention). As a whole, performance anxiety was reduced to a large degree after PI in the included trials. However, it is important to emphasise that this does not imply that performance anxiety has been demonstrated to act as a putative mechanism. As all trials in the present review analysed sleep-related performance anxiety only as pre- to post-treatment changes, future research might design studies so that mediational analyses become possible. In such studies, repeated assessment of mediators is necessary, and then analysing whether change in mediators precede improvements in insomnia symptoms. This would pave the way for evidence-based explanations for how PI produces improvements (Kazdin, 2007).

Another important methodological aspect of the research literature on performance anxiety is that the self-report scales used in the three studies have not been systematically validated in psychometric terms (Broomfield & Espie, 2003; Buchanan, 1988; Fogle & Dyal, 1983). As a result, it is uncertain whether the construct validity of the self-report scales is sufficiently captured, so that conclusions about sleep performance anxiety can be drawn in the present review. Concerning the measurement of sleep performance anxiety, it should be noted that validated self-report scales are available, such as the Glasgow Sleep Effort Scale (Broomfield & Espie, 2005; Meia-Via et al., 2016; Vand et al., 2020), and such instruments are recommended for future research. The use of validated measures in future trials would enable stronger conclusions about the effectiveness of PI on sleep performance anxiety as well as the possibility to examine mediation in a more rigorous way and explore moderation (e.g. whether PI is particularly effective among insomnia patients with elevated sleep performance anxiety).

One should note that sleep-related performance anxiety is not the only candidate as a putative mechanism for PI. First, PI could be viewed as an intervention that exposes patients to learned, feared stimuli in the bed or bedroom (Lundh, 1998), which enables extinction and the formation of new learning (Craske et al., 2014). However, this notion has not yet been articulated in detail in the research literature and not examined empirically. A second putative mechanistic pathway is described in the attention–intention–effort model (Espie et al., 2006). Although the pathway by Espie et al., (2006) appears to have high face validity, the model has not, to our knowledge, been explicitly tested in its full complexity in the realm of PI treatment.

4.4 Future directions

There are several important areas that future research could focus on to enhance the understanding of PI. Following from the limitations and uncertainties described above, we recommend future research to use active comparators, sample sizes based on power calculation, samples from clinical settings, a variety of insomnia types (including insomnia disorder), formal assessments of comorbidity, different delivery formats, broad assessments of insomnia symptoms and correlates as outcomes, and different mediators to examine mechanistic pathways.

One unknown dimension of PI is the optimal dosing and administration. Although PI has commonly been implemented by patients during a 2–4-week period, it could be argued that shorter administration of PI could be beneficial as well. Based on the theoretical rationale; that is, breaking a vicious cycle of sleep intention and associated performance anxiety, PI could potentially also be delivered as a behavioural experiment, during which patients test their predictions (e.g. “If I do not try to fall asleep, I will remain awake all night”), followed by testing PI for a limited number of nights. Another topic for future research is the optimal treatment rationale and instructions for PI. Based on two studies included in the present review (Ascher & Turner, 1980; Ott et al., 1983), it appears likely that PI with a desensitisation rationale or with feedback is less beneficial than the original approach by Ascher and Efran (1978). Beyond that, the ideal rationale and instructions remains unknown when delivering PI.

Based on the findings in the present review, the notion of how PI should be used warrants reflection. On the one hand, we believe that CBT-I should still be regarded as the first-line intervention for insomnia disorder (Riemann et al., 2017). On the other hand, PI might play a role in some cases. For example, if a patient remains unimproved after CBT-I, PI could be one option. Also, if the patient reports high sleep-related performance anxiety, and this appears as the primary maintaining factor, PI could be used in isolation or in combination with other efficacious CBT-I components, such as sleep restriction (Miller et al., 2014). To date, current CBT manuals do not include PI as a treatment component (van Straten et al., 2018). Whether the addition of PI could add efficacy to CBT-I is currently unknown. Future research could explore the notion of combining PI with CBT-I to explore potential additive effects, but also whether there are subgroups of patients who benefit more from PI.