Introduction

Insomnia is a common sleep disorder [1] that can lead to negative health outcomes, such as fatigue, increased daytime irritability, cognitive deficits and poor health status [2, 3]. Insomnia often co-exists with major medical conditions, such as diabetes [4], hypertension [5] and cancer [6], which increase personal suffering and even the risk of all-cause mortality [7]. In addition, insomnia also occurs in up to 40%–50% of patients with psychiatric disorders [8], such as depression [9], bipolar and anxiety disorder [10], and increases the risk of suicide [11, 12].

There is a complex association between insomnia and major medical/ psychiatric disorders. Medical and psychiatric disorders could precipitate the development of insomnia, while insomnia could increase the risk of medical/ psychiatric conditions. Therefore both insomnia and major medical/ psychiatric disorders need to be treated concurrently [1, 13]. Psychotropic medications, such as benzodiazepines and non-benzodiazepine hypnotics (e.g. Zolpidem), are commonly prescribed, although the long-term use is not encouraged due to the risk of dependency [14, 15].

Psychosocial interventions, such as cognitive behavioural therapy for insomnia (CBT-I), have been increasingly used in treating patients with insomnia disorder [16,17,18,19]. The efficacy of CBT-I in primary insomnia has been shown in several meta-analyses [20,21,22,23,24,25,26] and CBT-I is recommended for insomnia by the American Academy of Sleep Medicine [17, 27].

Research findings on the efficacy of CBT-I in patients with major medical conditions (e.g. cancer [28]) or psychiatric disorders (e.g. depression (Manber et al., 2008) and posttraumatic stress disorder (Ger-main, Shear, Hall, & Buysse, 2007; Talbot et al., 2014)), have been mixed. One recent meta-analysis [29] examined the efficacy of CBT-I for patients with medical or psychiatric comorbidities. However, the diagnosis of insomnia was based on either patients’ complaints or standardized diagnostic scales, such as the insomnia severity index (ISI), and only a proportion of participants in the included studies had comorbid medical and/or psychiatric disorders. Due to the heterogeneity in diagnostic criteria and study samples across studies, the findings could be biased to an uncertain extent.

Thus we conducted this meta-analysis of randomized controlled trials (RCTs) comparing CBT-I monotherapy with active control treatment for insomnia in patients with major medical or psychiatric disorders using stringent diagnostic criteria.

Methods

Selection Criteria

Following the PICOS acronym [30], the inclusion criteria were as follows: participants (P): patients with a diagnosis of insomnia and major medical or psychiatric comorbidities; insomnia was diagnosed using standardized diagnostic criteria, such as the Diagnostic Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) [31], the International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10) [32], or the International Classification of Sleep Disorders (ICSD) [33, 34]. Intervention (I): CBT-I monotherapy. Comparison (C): active control group, such as behavioural desensitization and sleep hygiene education. Outcomes (O): the primary outcome measure was the improvement of insomnia symptoms as assessed with standardized scales, such as the Insomnia Severity Index (ISI). Secondary outcome measures included the Pittsburgh Sleep Quality Index (PSQI), dysfunctional attitudes and beliefs about sleep scale (DBAS) and sleep parameters based on polysomnography, actigraphy or sleep diaries, such as sleep efficiency, total sleep time, sleep latency, wake after sleep onset, time in bed, sleep quality and number of awakenings. Study design (S): Randomized controlled trials (RCTs). Studies that did not specifically mention CBT-I were excluded, although certain components of CBT-I were used.

Search Methods

Literature search was independently conducted by three authors (FCZ and YY) in PubMed, EMBASE, PsycINFO, Cochrane Library, WanFang, and CNKI databases from the inception date until November 12, 2017, using the following terms: CBTI, CBT-I, cognitive behavioural therapy for insomnia, cognitive behavioural for insomnia, cognitive behavioural treatment for insomnia, cognitive behavioural treatment for insomnia, cognitive behavioural therapy of insomnia, cognitive behavioural therapy of insomnia, insomnia, sleep, maintenance disorder, dyssomnia, sleepless, early morning awakening, Randomized controlled trials, Randomized controlled trial, RCT, and RCTs. In addition, we manually reviewed the reference list of relevant reviews for additional studies.

Data Extraction

Relevant data were independently extracted by two authors (FCZ and YY). Any discrepancies were resolved by consensus or a discussion with the third author (YTX). If both polysomnography or actigraphy and rating scales were used, polysomnography and actigraphy were preferred as they are more objective than scales. Several studies had several follow-up assessments. In order to reduce heterogeneity caused by study periods, only the data at 3 months follow-up were extracted and analyzed.

Quality Assessment

The study quality was assessed using the Cochrane risk of bias [35] and the Jadad scale [36, 37]. The total score of the Jadad scale ranges from 1 to 5, with a higher score indicating higher quality. Studies with a Jadad total score of <3 was considered as low-quality; otherwise they were considered as high-quality [36]. The system grading of recommendations assessment, development, and evaluation (GRADE) was used in evaluating the evidence level of outcomes [38, 39].

Data Synthesis and Statistical Analyses

Due to the discrepancy in sampling methods, measurements and demographic and clinical characteristics between studies, the random effects model was used for meta-analytic outcomes [40]. Compared to fixed-effects model, the random-effects model is more conservative [40]. Intention-to-treat (ITT) analyses were preferred if available in included studies [41,42,43]. The heterogeneity across included studies was assessed using I2 index [44], with I2 of 25%, 50% and 75% indicating mild, moderate, and high heterogeneity between studies, respectively. The standardized mean difference (SMD) with 95% confidence intervals (CIs) was used for continuous outcome variables. Funnel plots and Egger’s test were performed [45] for publication bias of primary outcome. A significance level of 0.05 was set for all meta-analytic outcomes (two-sided). Review Manager Version 5.3 (http://www.cochrane.org) and Comprehensive Meta-Analysis V2.0 (www.meta-analysis.com) were used for all analyses.

Results

Literature Search and Study Characteristics

A total of 250 relevant articles were initially identified in the literature search. Finally, 13 RCTs with 27 treatment arms were included in the analyses (Fig. 1). One study [46] included three arms (e.g., an internet CBT-I, an in-person CBT-I and a control group), and so data from both CBT-I groups were combined for analyses. Eight hundred and fifty-three participants were included in this meta-analysis, with 441 patients in the CBT-I group and 412 in the active control group. The included studies were carried out in the United States (7 RCTs, n = 481), Sweden (1 RCT, n = 64), Canada (2 RCTs, n = 183), and Spain (3 RCTs, n = 125). The diagnosis of insomnia was established using DSM-IV, DSM-IV-TR, DSM-III-R, International Classification of Sleep Disorders (ICSD), ICSD-2, or American Academy of Sleep Medicine criteria. Treatment frequency varied from weekly to biweekly, and the number of sessions of CBT-I treatment ranged from 3 to 12 (Table 1).

Fig. 1
figure 1

PRISAMA flowchart

Full size image
Table 1 Characteristics of studies included in this meta-analysis
Full size table

Assessment Quality and Quality of Evidence

The risk of bias in the 13 RCTs is shown in Table S1. Five studies were double blinded, 8 studies were single blinded or partial blinded, and the rest did not provide any information of blinding. All the included RCTs described the random sequence generation, and 5 mentioned allocation concealment. All studies were rated as “low risk” in terms of attrition and reporting bias. Jadad total score ranged from 3 to 4 (Table 1). All included RCTs were rated as “high quality”. The quality of evidence of outcome measures were evaluated as “very low” (5.9%, 1/17), “low” (29.4%, 5/17), and “moderate” (64.7%, 11/17) according to the GRADE approach (Table 3).

Publication Bias

As the number of studies was less than 10 for analyses on efficacy measures, publication bias could not be evaluated [47].

Primary Outcome

Insomnia Severity Index (ISI)

Seven studies reported insomnia data assessed by ISI at post-treatment assessment. Compared with the control group, the CBT-I group showed more significant improvement at post-treatment (n = 527; SMD: -0.74, 95% CI: −0.92 to −0.56, I2 = 39%, p < 0.0001, Table 2, Fig. S1.1). Subgroup analyses found superiority of CBT-I over active control group in patients with major medical conditions at post-treatment (N = 3, n = 275, SMD: -0.58, 95% CI: −0.82 to −0.34, I2 = 33%, p < 0.001), as well as in those with psychiatric disorders at post-treatment (N = 4, n = 252, SMD: -0.93, 95% CI: −1.20 to −0.66, I2 = 7%, p < 0.001) (Fig. S1.2). Additional follow-up data were only available in 2 studies, which also found superiority of CBT-I group over active control group at 3-month follow-up (n = 168, SMD: -0.33, 95% CI: −0.64 to −0.01, I2 = 0%, p = 0.04, Table 2, Fig. S1.3).

Table 2 Sleep data at post-CBT-I assessment
Full size table
Table 3 GRADE analyses
Full size table

Secondary Outcomes

Sleep Efficiency

Data on sleep efficiency were available in 7 RCTs (n = 498). Compared with control group, CBT-I was associated with small but significant improvement at post-treatment assessment (n = 498, SMD: 0.18, 95% CI: 0.00 to 0.36, I2 = 36%, p = 0.05, Table 2, Fig. S2.1). Compared with the control group, CBT-I did not show significant benefits in patients with psychiatric disorders at post-treatment assessment (2 RCTs, n = 158, SMD: 0.13, 95% CI: −0.19 to 0.46, I2 = 2%, p = 0.42, Table 2). There were also non-significant findings in patients with medical conditions at post-treatment assessment (5 RCTs, n = 340, SMD: 0.14, 95% CI: −0.19 to 0.47, I2 = 52%, p = 0.40)(Fig. S2.2). Three studies reported additional follow-up assessments and found a significant improvement of sleep efficiency in CBT-I group that persisted at 3 months (n = 234, SMD: 0.31, 95% CI: 0.04 to 0.58, I2 = 0%, p = 0.03, Table 2, Fig. S2.3).

Total Sleep Time

Eight RCTS reported data on total sleep time. There was no significant group difference at post-treatment assessment (n = 531, SMD: -0.13, 95% CI: −0.30 to 0.04, I2 = 45%, p = 0.14, Table 2, Fig. S3.1). In the subgroup analyses, neither patients with major medical conditions nor those with psychiatric disorders showed any significant improvement in CBT-I group at post-treatment assessment (major medical conditions: 5 RCTs, n = 309, SMD: -0.15, 95% CI: −0.49 to 0.18, I2 = 51%, p = 0.37; psychiatric disorders: 3 RCTs, n = 222, SMD: -0.04, 95% CI: −0.44 to 0.35, I2 = 51%, p = 0.83, Fig. S3.2). Three studies reported additional follow-up data at 3 months, but there was no significant group difference (n = 235, SMD: -0.20, 95% CI: −0.47 to 0.07, I2 = 15%, p = 0.14, Table 2, Fig. S3.3).

Sleep Onset Latency

Data were available from 6 RCTs on sleep latency. Compared with control group, the CBT-I group showed significant improvement at post-treatment assessment (n = 505, SMD: -0.36, 95% CI: −0.62 to −0.10, I2 = 51%, p = 0.007, Table 2, Fig. S4.1). No group difference was found in patients with major medical conditions at post-treatment assessment (3 RCTs, n = 283, SMD: -0.27, 95% CI: −0.78 to 0.24, I2 = 78%, p = 0.30). Nevertheless, subgroup analyses revealed a significant group difference in patients with psychiatric disorders at post-treatment assessment (3 RCTs, n = 222, SMD: -0.45, 95% CI: −0.73 to 0.18, I2 = 0%, p = 0.001, Fig. S4.2). Three studies reported data in additional follow-up assessment, but no group difference was found (n = 234, SMD: -0.13, 95% CI: −0.40 to 0.14, I2 = 32%, p = 0.33, Table 2, Fig. S4.3).

Wake after Sleep Onset

Data were available from 7 RCTS on wake after sleep onset at the post-treatment assessment. The CBT-I group showed significant improvement compared with active control group (n = 531, SMD: -0.21, 95% CI: −0.38 to −0.04, I2 = 34%, p = 0.02, Table 2, Fig. S5.1). Significant group difference was found in patients with major medical conditions (4 RCTs, n = 309, SMD: -0.31, 95% CI: −0.54 to −0.09, I2 = 0%, p = 0.006). However, in patients with psychiatric disorders, no significant group difference was found (3 RCTs, n = 222, SMD: -0.08, 95% CI: −0.51 to 0.35, I2 = 59%, p = 0.72, Fig. S5.2). In the three studies which reported additional 3 months follow-up, and the superiority of CBT-I group was found (n = 250, SMD: -0.30, 95% CI: −0.56 to 0.05, I2 = 47%, p = 0.02, Table 2, Fig. S5.3).

Time in Bed

Two RCTs reported data on time in bed between CBT-I group and active control group at post-treatment assessment, but no significant group difference was found (n = 84, SMD: -0.09, 95% CI: −0.72 to 0.90, I2 = 67%, p = 0.83, Table 2, Fig. S6).

Sleep Quality

Data were available in 2 RCTS on sleep quality at the post-treatment assessment, and the CBT-I group showed significant improvement (n = 164, SMD: 0.56, 95% CI: 0.23 to 0.88, I2 = 0%, p < 0.001, Table 2, Fig. S7).

Number of Awakenings

Data were available from 2 RCTs on number of awakenings at the post-treatment assessment, but no significant group difference was found (n = 126, SMD: 0.20, 95% CI: −0.16 to 0.57, I2 = 0%, p = 0.28, Table 2, Fig. S8).

Pittsburgh Sleep Quality Index (PSQI)

Seven RCTs reported the changes of PSQI total score at post-treatment assessments. CBT-I group showed significant improvement (n = 417; SMD: -0.76, 95% CI: −1.09 to −0.42, I2 = 61%, p < 0.001, Table 2, Fig. S9.1). Subgroup analyses revealed a significant advantage of CBT-I for patients with major medical conditions (5 RCTs, n = 329, SMD: -0.76, 95% CI: −1.22 to −0.30, I2 = 74%, p = 0.001) and for those with psychiatric disorders (2 RCTs, n = 88, SMD: -0.76, 95% CI: −1.19 to −0.32, I2 = 0%, p = 0.006, Fig. S9.2). Follow-up assessments were reported in 3 RCTs, and the superiority of CBT-I persisted for three months (n = 215, SMD: -0.56, 95% CI: −1.01 to −0.12, I2 = 55%, p = 0.01, Table 2, Fig. S9.3).

Dysfunctional Attitudes and Beliefs about Sleep Scale (DBAS)

The DBAS total scores were available in 3 RCTs at post-treatment assessment. CBT-I group had significant improvement at post-treatment assessment (n = 283; SMD: -1.09, 95% CI: −1.48 to −0.71, I2 = 60%, p < 0.001, Table 2, Fig. S10.1). Three studies reported data at additional follow-up, but there were no significant group difference (n = 251; SMD: -0.8, 95% CI: −1.35 to −0.25, I2 = 76%, p = 0.004, Table 2, Fig. S10.2).

All Cause Discontinuation

Eleven studies reported discontinuation rates at post-treatment assessments, but no significant group difference was found (n = 773, RR = 0.81, 95% CI: 0.44 to 1.47, I2 = 60%, p = 0.49, Fig. S11).

Discussion

This was the first meta-analysis of RCTs specifically comparing CBT-I monotherapy with active control treatment for insomnia in patients with medical or psychiatric comorbidities.

CBT-I usually contains five core components [27]: stimulus control, sleep restriction, sleep hygiene, relaxation training and cognitive restructuring. Of the included studies, 4 studies used all the five components [46, 48,49,50], 8 studies used four components [51,52,53,54,55,56,57,58] and 1 study used two components [59]. Of the five core components, stimulus control was used in all studies. Four studies also used an additional component (relapse prevention for insomnia) [50, 52, 56, 57], and psycho-education about the association between sleep and fibromyalgia was used in 3 studies [50, 56, 57]. This meta-analysis consistently found superiority of CBT-I over active control group in treating insomnia in patients with major medical or psychiatric comorbidities. The effect size was medium as measured by the primary outcome measure (SMD = −0.74). In addition, the advantage of the CBT-I group remained in most secondary outcome measures, such as sleep onset latency (SMD = -0.36), wake after sleep onset (SMD = -0.21), sleep quality (SMD = 0.56), PSQI total scores (SMD = -0.76) and DBAS total scores (SMD = -1.09). However, no group difference was found in total sleep time, time in bed, number of awakening and sleep quality. This meta-analysis also found that the superiority of CBT-I persisted at 3 months in the following measures: the wake after sleep onset (SMD = -0.30), PSQI (SMD = -0.56) and ISI (SMD = −0.33). The effect sizes in these measures were generally larger than those reported in the Geiger-Brown et al.’s study [29], in which the benefits in the CBT-I group were less pronounced (effect size: −0.17 to 0.10) at 3 months follow-up.

As for the primary outcome measure, the effect size between CBT-I and active control treatment in this meta-analysis (SMD = -0.74) was smaller compared to the Geiger-Brown et al. study (2015) (effect size = 1.22), which could be due to different proportion of comorbidities between study samples. In the 2015 study [29] the majority of the participants suffered from chronic pain syndrome and mixed medical and/or psychiatric conditions. CBT-I was associated with an effect size of 1.00 in patients with medical conditions and with an effect size of 1.51 in those with psychiatric disorders [29]. Subgroup analyses of this meta-analyses revealed similar findings, which suggests that CBT-I appears to be more effective in treating insomnia patients with psychiatric disorders. The present study revealed that CBT-I was associated with a greater improvement in patients with psychiatric comorbidities (SMD: −0.93) than those with medical comorbidities (SMD: −0.58) as measured by the ISI total score.

In this meta-analysis CBT-I showed superiority over control group in several secondary outcome measures, such as sleep onset latency, wake after sleep onset, sleep quality, PSQI and DBAS total scores. The effect sizes were generally smaller than most findings published in previous meta-analyses regarding CBT-I [20, 22, 25, 26]. Moreover, this meta-analysis did not find any advantage of CBT-I in the following domains: total sleep time, time in bed, and number of awakening. These inconsistent findings may be due to different proportion of comorbidities between study samples.

Subgroup analyses revealed that CBT-I had an advantage on sleep onset latency only in patients with psychiatric disorders, while CBT-I had an advantage regarding number of wakeup after sleep onset only in those with medical conditions. This finding was not reported previously. This meta-analysis also found that the advantage of CBT-I persisted at 3-month follow-up in the following measures: wake after sleep onset, PSQI, and ISI. However, the effect sizes appeared to decrease over time.

The strengths of this meta-analysis include the use of stringent diagnostic criteria for insomnia, inclusion of insomnia patients with major medical conditions or psychiatric disorders, and administration of a variety of secondary outcome measures. However, several limitations need to be addressed. First, there was a discrepancy in study designs, participant characteristics, insomnia definition, and outcome measures between studies, although the random effects model and subgroup analyses have been performed. Second, some relevant factors, such as prescriptions of medications and severity of major medical conditions and psychiatric disorders, were not analysed due to insufficient data. Third, the long-term effect of CBT-I could not be examined due to lack of data. Finally, only English and Chinese databases were searched.

Conclusion

This meta-analysis found that CBT-I monotherapy generally had greater efficacy than other active control treatment for insomnia in patients with medical or psychiatric comorbidities. CBT-I was more efficacious in improving sleep onset latency in patients with mental disorders, while it also has an advantage in reducing number of wakeup after sleep onset in patients with medical conditions. However, the advantage of CBT-I was only maintained in some measurements at 3-month assessment, with decreased effect sizes over time.