Digital cognitive behavioral therapy for insomnia on depression and anxiety: a systematic review and meta-analysis

“Insomnia is one of the most common sleep disorders, posing a significant public health concern, with an estimated prevalence of 10–30% among adults in the general population. These numbers are greater among patients, with reports estimating 69% prevalence among primary care patients. Insomnia disorder is defined by the Diagnostic and Statistical Manual of Mental Disorders – IV (DSM-IV) as the complaint for difficulty in initiating or maintaining sleep, or restorative sleep for at least 1 month. [..]

Depression and anxiety are the most common comorbid mental disorders associated with insomnia which can also exacerbate the sleep disorder. Recently, epidemiologic studies have reported that insomnia predicts the development of major depression, anxiety, and suicide. Various cross-sectional and longitudinal research have presented insomnia to be associated with an increased risk of mood and anxiety disorders as well as suicide. [..]

Cognitive behavioral therapy for insomnia (CBT-I) has been an effective non-pharmacological treatment for insomnia. It is a multi-component, evidence-based treatment and is considered the first-line approach including cognitive restructuring, sleep restriction, stimulus control, sleep hygiene education, and relaxation. Due to the association between insomnia and depression, CBT-I has been viewed an effective approach for managing depression. [..]


[..] The meta-analysis included a total of 10,486 participants, of whom 5494 were randomized to the dCBT-I [digital CBT-I] group, with a median study size of 111 participants (range 21–3755 participants). [..] Participants in control conditions received active interventions including sleep education or general health education (not specifically targeting sleep), or passive controls including treatment as usual and wait-lists. [..]

Twenty-one out of 22 studies reported the severity of depressive symptoms. [..] At the post-treatment assessment, we found a small to moderate effect favoring dCBT-I (Standardized Mean differences (SMD) = −0.42; 95% confidence interval (CI): −0.56, −0.28; p < 0.001; k = 21). The statistical heterogeneity in effect sizes among studies was high (I2 = 81.79; Q = 109.85; df = 20; p < 0.001). [..]

For anxiety symptoms at the post-treatment assessment, we found a small to moderate effect favoring dCBT-I (SMD = −0.29; 95% CI: −0.40, −0.19; p < 0.001; k = 18). The statistical heterogeneity in effect sizes among studies was high (I2 = 57.75; Q = 40.24; df = 17; p < 0.001). [..]

For the severity of insomnia post-treatment, we found a large effect favoring dCBT-I (SMD = −0.76; 95% CI: −0.95, −0.57; p < 0.001; k = 22). The statistical heterogeneity in effect sizes among studies was high (I2 = 90.59; Q = 223.04; df = 21; p < 0.001). In studies including only ISI, we found a large effect favoring dCBT-I (SMD = −0.81; 95% CI: −0.97, −0.65; p < 0.001; I2 = 79.51; k = 19). [..]

To compare the effect of treatment adherence, we divided 12 studies into two groups: (1) high adherent group with >65% of dCBT-I completers; (2) low adherent group with <65% of dCBT-I completers. The treatment effects of the high adherent group were significant for depression (SMD = −0.60; 95% CI: −0.72, −0.47; p < 0.001; I2 = 0.00; k = 5), anxiety (SMD = −0.32; 95% CI: −0.61, −0.02; p = 0.03; I2 = 38.58; k = 4) and sleep outcomes (SMD = −1.12; 95% CI: −1.30, −0.95; p < 0.001; I2 = 15.17; k = 5). [..] For the low adherent group, the treatment effects were also significant but effect sizes were smaller than those in adherent groups for depression (SMD = −0.35; 95% CI: −0.57, −0.14; p = 0.001; I2 = 88.71; k = 7), anxiety (SMD = −0.28; 95% CI: −0.45, −0.11; p = 0.001; I2 = 82.34; k = 6), and sleep outcomes (SMD = −0.69; 95% CI: −1.05, −0.34; p < 0.001; I2 = 95.82; k = 7). [..]

The treatment effects of the fully automated dCBT-I were significant for depression (SMD = −0.43; 95% CI: −0.61, −0.26; p < 0.001; I2 = 88.14; k = 13), anxiety (SMD = −0.29; 95% CI: −0.41, −0.17; p = 0.001; I2 = 68.46; k = 12), and sleep outcomes (SMD = −0.81; 95% CI: −1.04, −0.59; p < 0.001; I2 = 92.69; k = 14).


Although the pooled effect of dCBT-I on depressive and anxiety symptoms is small to moderate, there was considerable heterogeneity in the magnitude of the effects observed. This heterogeneity is comparable to previous research and expected given the diversity of participants recruited, outcome measures, the delivery format of CBT-I, and baseline severity levels of depression and anxiety in the included studies. The effects of dCBT-I interventions on depression and anxiety symptoms were relatively robust after removing the three studies that included participants with mental or medical comorbidities. Considering that the majority of the studies included in this meta-analysis had subclinical depression and anxiety samples, this suggests that dCBT-I interventions are beneficial in reducing subclinical depression and anxiety symptoms. Whilst dCBT-I is developed for insomnia treatment, current findings suggest that dCBT-I has the capability for an effective supplementary therapy beyond its current potential.

[..] previous research has identified that even the most effective apps have minimal effect if these lack user engagement, resulting in a high attrition rate. The attrition-efficacy gap needs to be settled especially for those requiring sustained mental health treatment. The problem of high dropout rates is especially true for fully automated dCBT-I intervention without any support of human therapists. Therefore, adherence-promoting features such as ease of use, rewards, ability to personalize app, tailored interventions, social or peer support in app, personalized feedback, and integration with clinical services should be considered. Although there’s lacking evidence in research comparing the differences between automated support and with or without human support, automated reminders have increased enhanced adherence to treatment. The fears around security and privacy inherent to digital interventions might be an additional factor in adherence and attrition for some participants, therefore user safety should be considered upfront. Furthermore, most studies showed various methods to assess adherence, which make it difficult to compare outcomes meaningfully, though adherence was most often assessed by the degree of program completion. Therefore, a standardized method for assessing adherence is required to reliably predict the impacts of adherence on treatment outcomes.

Given that few of the studies included in the current review involve participants with clinically significant level of depression and anxiety symptoms, our result of significant effects favoring dCBT-I could be seen as pertaining to patients with subthreshold level of depression and anxiety symptoms. In a previous study of internet-delivered CBT-I, when comparing the differences between severe and low to mildly depressed patients, those with severe symptoms more likely to benefit from human support of reminding and encouraging patients by e-mail, while those with low level of depressive symptoms were demonstrated to benefit adequately regardless of the support. This indicates that the addition of some guidance could be preferred depending on the baseline severity of depression although fully automated intervention increases scalability. Thus, further research is needed to determine the role of symptom severity of depression and anxiety for the effect of digital intervention.”

Full article, S Lee, JW Oh, KM Park et al. NPJ Digital Medicine, 2023.3.25