Effect of Digital Cognitive Behavioral Therapy for Insomnia on Health, Psychological Well-being, and Sleep-Related Quality of Life: A Randomized Clinical Trial

The researchers compared two interventions for people with self-reported insomnia symptoms:

**Digital Cognitive Behavioral Therapy for Insomnia (dCBT):** A fully automated, interactive online program called Sleepio, delivered via web and mobile app. It included 6 core sessions (approximately 20 minutes each) covering stimulus control, sleep restriction, cognitive restructuring, relaxation, and sleep hygiene. Participants could access additional "booster" content. The program used an animated "virtual therapist" and was designed to be completed within 12 weeks, with sessions released weekly.

**Sleep Hygiene Education (SHE):** A control condition consisting of a static website and a downloadable booklet providing standard sleep hygiene advice (e.g., avoid caffeine, keep a regular schedule, make your bedroom dark and quiet). This is considered a minimal active control — it provides some useful information but lacks the structured, therapeutic components of CBT.

Both groups continued with their usual medical care ("treatment as usual"). The key question was whether dCBT would improve not just sleep itself, but also daytime functioning, mood, and quality of life — the reasons most people actually seek help for insomnia.

The primary outcomes were three measures of daytime function: general physical and mental health (PROMIS Global Health), psychological well-being (Warwick-Edinburgh Mental Well-being Scale), and sleep-related quality of life (Glasgow Sleep Impact Index). Secondary outcomes included depression, anxiety, fatigue, daytime sleepiness, cognitive failures, work productivity, and relationship satisfaction. Insomnia severity was measured as a potential mediator.

**Sample size:** 1,711 participants (1,339 in the dCBT group, 372 in the SHE group — a roughly 4:1 allocation ratio)

**Population:** Adults aged 18+ with self-reported symptoms of insomnia, recruited online through media advertisements and health-related websites in the United Kingdom

**Demographics:** Mean age 48.0 years (SD 13.8); 77.7% female; 91.1% white; 66.4% employed; 56.8% had a university degree

**Inclusion criteria:** Score ≤5 on the Sleep Condition Indicator (indicating significant insomnia symptoms); access to internet; willing to be randomized

**Exclusion criteria:** Currently receiving CBT for insomnia; shift work; pregnancy; self-reported diagnosis of bipolar disorder, epilepsy, or parasomnia; high risk of sleep apnea (based on screening questionnaire); high alcohol consumption (>14 units/week for women, >21 for men); currently using sleep medication more than once per week

**Setting:** Entirely online — participants never met researchers in person. Recruitment and all assessments were conducted via the internet.

This is a relatively well-educated, predominantly female, white British sample. The online recruitment method means participants were likely motivated and comfortable with technology, which may limit generalizability to less tech-savvy populations.

All measurements were self-reported via online questionnaires at four time points: baseline (week 0), mid-treatment (week 4), post-treatment (week 8), and follow-up (week 24).

**Primary outcome measures:**

**Patient-Reported Outcomes Measurement Information System (PROMIS) Global Health Scale:** A 10-item scale measuring general physical and mental health. Scores range from 10 to 50, with higher scores indicating better health. This is a validated, widely used measure of overall functional health.

**Warwick-Edinburgh Mental Well-being Scale (WEMWBS):** A 14-item scale covering positive mental health (e.g., optimism, feeling useful, feeling relaxed). Scores range from 14 to 70, with higher scores indicating greater psychological well-being.

**Glasgow Sleep Impact Index (GSII):** A 10-item scale measuring how much insomnia interferes with daily life (e.g., mood, relationships, work, leisure). Scores range from 1 to 100, with higher scores indicating greater impairment (worse sleep-related quality of life).

**Secondary outcome measures:**

**Sleep Condition Indicator (SCI):** An 8-item measure of insomnia symptoms (range 0–32, higher = better sleep). This was used to assess insomnia severity and as a potential mediator.

**Patient Health Questionnaire-9 (PHQ-9):** Depression severity (0–27, higher = worse)

**Generalized Anxiety Disorder-7 (GAD-7):** Anxiety severity (0–21, higher = worse)

**Fatigue Severity Scale (FSS):** Fatigue (1–7, higher = worse)

**Epworth Sleepiness Scale (ESS):** Daytime sleepiness (0–24, higher = sleepier)

**Cognitive Failures Questionnaire (CFQ):** Everyday cognitive lapses (0–100, higher = worse)

**Work Productivity and Activity Impairment (WPAI):** Work and activity impairment due to health (percentage)

**Relationship Satisfaction Scale (RSS):** Satisfaction with partner relationship (0–30, higher = more satisfied)

**Important note:** No objective sleep measures were used (no actigraphy, no polysomnography). All outcomes are subjective self-reports, which is both a strength (insomnia is defined by subjective distress) and a limitation (potential for reporting bias, especially in an unblinded trial).

**Study design:** Two-arm, parallel-group, superiority randomized controlled trial (RCT) with 4:1 allocation ratio favoring dCBT. The unequal allocation was intentional — the researchers wanted more data on the dCBT group to explore mediators and moderators, while still maintaining a valid control group.

**Randomisation:** Participants were randomized using a computer-generated algorithm, stratified by baseline insomnia severity (SCI score ≤2 vs. 3–5) and whether they were currently taking sleep medication. The allocation sequence was concealed from researchers until assignment.

**Blinding:** This was an open-label trial — participants knew which intervention they received (you cannot realistically blind someone to whether they are using an interactive CBT program vs. reading a sleep hygiene booklet). Outcome assessors (the researchers analyzing data) were blinded to group assignment. The lack of participant blinding is a significant limitation because expectations can influence self-reported outcomes, especially for subjective measures like well-being and quality of life.

**Duration:** The intervention period was 12 weeks (participants had 12 weeks to complete the 6-session dCBT program). Assessments occurred at weeks 0, 4, 8, and 24. The 24-week follow-up is a strength — it shows whether effects persist after treatment ends.

**Statistical approach:** Primary analysis used intention-to-treat (ITT) — all randomized participants were analyzed in their assigned group regardless of how much of the intervention they completed. This is the gold standard for RCTs because it preserves the benefits of randomisation and reflects real-world effectiveness (where not everyone completes treatment). Missing data were handled using multiple imputation. The primary analysis used linear mixed models with fixed effects for group, time, group-by-time interaction, and stratification variables. Mediation analysis used structural equation modeling to test whether improvements in insomnia (SCI scores) statistically explained improvements in the primary outcomes.

**What this design can and cannot prove:**

**Can prove:** That dCBT causes greater improvements in self-reported health, well-being, and sleep-related quality of life compared to sleep hygiene education, in an online sample of adults with insomnia symptoms, over a 6-month period. The randomisation controls for confounding variables (differences between groups at baseline), and the ITT analysis preserves this control.

**Cannot prove:** That dCBT is better than face-to-face CBT (no direct comparison). That effects are due to specific CBT components rather than general factors like attention, expectation, or engagement (the control condition was minimal — a static website vs. an interactive 6-session program). That effects generalize to objective sleep measures (no actigraphy or PSG). That effects generalize to clinical populations with diagnosed insomnia disorder (participants had self-reported symptoms, not necessarily clinical diagnoses). That effects persist beyond 6 months. That the 4:1 allocation ratio didn't introduce bias (unequal allocation can sometimes affect participant behavior or expectations, though statistical adjustments were made).

**Major methodological weaknesses:**

1. **No blinding of participants** — the most serious limitation for subjective outcomes

2. **No active control for engagement** — dCBT involved 6 interactive sessions; SHE was a one-time reading of a booklet. Differences could be due to attention, structure, or expectation rather than CBT content

3. **Self-report only** — no objective sleep or activity measures

4. **High attrition** — at 24 weeks, 73.5% of dCBT and 64.2% of SHE participants completed assessments. While acceptable, differential dropout could bias results

5. **Unequal allocation** — while statistically valid, it reduces the effective sample size of the control group

6. **Single-country, predominantly white, well-educated sample** — limits generalizability

All results are adjusted mean differences (dCBT minus SHE) with 95% confidence intervals, from the intention-to-treat analysis. Positive differences favor dCBT (except for GSII where negative = better).

**Primary outcomes (the main things they wanted to test):**

**Functional health (PROMIS Global Health, scale 10–50):**

- Week 4: +0.90 (0.40 to 1.40), p < 0.01

- Week 8: +1.76 (1.24 to 2.28), p < 0.01

- Week 24: +1.76 (1.22 to 2.30), p < 0.01

- *Interpretation:* Small but statistically significant improvement. A 1.76-point difference on a 40-point scale is modest.

**Psychological well-being (WEMWBS, scale 14–70):**

- Week 4: +1.04 (0.28 to 1.80), p < 0.01

- Week 8: +2.68 (1.89 to 3.47), p < 0.01

- Week 24: +2.95 (2.13 to 3.76), p < 0.01

- *Interpretation:* Small-to-moderate improvement. A 2.95-point difference on a 56-point scale is noticeable but not transformative.

**Sleep-related quality of life (GSII, scale 1–100, higher = worse):**

- Week 4: -8.76 (-11.83 to -5.69), p < 0.01

- Week 8: -17.60 (-20.81 to -14.39), p < 0.01

- Week 24: -18.72 (-22.04 to -15.41), p < 0.01

- *Interpretation:* Large improvement. An 18.72-point reduction on a 100-point scale is substantial — roughly a 19% reduction in insomnia-related life impairment.

**Secondary outcomes (other things they measured):**

**Insomnia severity (SCI, 0–32, higher = better):**

- Week 4: +3.72 (3.09 to 4.35)

- Week 8: +6.83 (6.16 to 7.50)

- Week 24: +6.88 (6.18 to 7.58)

- *All p < 0.01.* Large improvement — a 6.88-point difference on a 32-point scale is clinically meaningful.

**Depression (PHQ-9):** Week 24: -1.73 (-2.22 to -1.24), p < 0.01 — small improvement

**Anxiety (GAD-7):** Week 24: -1.44 (-1.86 to -1.02), p < 0.01 — small improvement

**Fatigue (FSS):** Week 24: -0.37 (-0.48 to -0.26), p < 0.01 — small improvement

**Daytime sleepiness (ESS):** Week 24: -0.58 (-0.88 to -0.28), p < 0.01 — small improvement

**Cognitive failures (CFQ):** Week 24: -3.67 (-5.08 to -2.26), p < 0.01 — small improvement

**Work productivity (WPAI):** Week 24: -3.46% (-5.09% to -1.83%), p < 0.01 — small improvement in productivity

**Relationship satisfaction (RSS):** Week 24: +0.57 (0.09 to 1.05), p = 0.02 — very small improvement

**Mediation analysis:** The researchers tested whether improvements in insomnia (SCI scores) statistically explained the improvements in the primary outcomes. They found that insomnia improvement mediated:

45.5% of the improvement in functional health

54.9% of the improvement in psychological well-being

84.0% of the improvement in sleep-related quality of life

This means that most of the benefit in sleep-related quality of life was directly explained by better sleep, while about half of the benefit in general health and well-being was explained by better sleep (the other half may be due to other factors like reduced anxiety about sleep, improved daytime routines, or increased self-efficacy).

Let's translate these numbers into plain English:

**Functional health:** The 1.76-point improvement on PROMIS Global Health (scale 10–50) is roughly equivalent to moving from "fair" to "good" on one or two of the 10 items. It's noticeable but not life-changing — like going from feeling "somewhat limited" in moderate activities to "not limited at all."

**Psychological well-being:** The 2.95-point improvement on WEMWBS (scale 14–70) is about half a standard deviation. This is roughly equivalent to feeling "more of the time" optimistic, useful, and relaxed — but not a dramatic transformation.

**Sleep-related quality of life:** The 18.72-point reduction on GSII (scale 1–100) is large — about 0.8 standard deviations. This means that, on average, someone who reported that insomnia "moderately" interfered with their daily life would now report it interfering "a little" or "not at all." This is the most clinically meaningful finding.

**Insomnia severity:** The 6.88-point improvement on SCI (scale 0–32) is substantial. The SCI has a clinical cutoff of ≤5 for insomnia; the average dCBT participant moved from below this cutoff to well above it (from about 3.5 to about 10.4), indicating that most no longer met the threshold for clinically significant insomnia.

**Depression and anxiety:** The 1.73-point reduction on PHQ-9 and 1.44-point reduction on GAD-7 are small — about 0.2–0.3 standard deviations. This is roughly equivalent to one less symptom (e.g., "little interest in doing things" goes from "several days" to "not at all").

**Daytime sleepiness:** The 0.58-point reduction on ESS (scale 0–24) is very small — barely noticeable in daily life. This makes sense because insomnia is more about hyperarousal than sleepiness.

**Overall:** The effects are largest for sleep-specific outcomes (insomnia severity, sleep-related quality of life), moderate for general well-being, and small for mood and cognitive function. This pattern suggests that dCBT primarily improves sleep, and these sleep improvements then spill over into other domains, but the spillover is modest.

**What the authors acknowledge:**

Open-label design (no participant blinding) could inflate effects due to expectation

Self-report measures