The effects of improving sleep on mental health (OASIS): a randomised controlled trial with mediation analysis

The intervention was a fully automated, web-based cognitive behavioural therapy for insomnia (CBT-I) called "Sleepio." The programme consisted of six 20-minute sessions delivered over 6 weeks, plus a digital sleep diary. The comparator was "usual care" — meaning participants in the control group received no active sleep intervention and continued whatever they were already doing (which could include seeking help on their own, but no structured sleep therapy was provided).

The primary outcomes were three separate measures:

**Insomnia severity** (measured by the Sleep Condition Indicator, a 0–10 scale where higher = better sleep)

**Paranoia** (measured by the Green Paranoid Thoughts Scale, Part B, 0–40 scale, higher = more paranoia)

**Hallucinatory experiences** (measured by the Specific Psychotic Experiences Questionnaire – Hallucinations subscale, 0–40 scale, higher = more hallucinations)

Secondary outcomes included depression, anxiety, nightmares, wellbeing, and work/social functioning.

**Sample size:** 3,755 university students (1,891 in the CBT-I group, 1,864 in usual care)

**Age:** Mean age 24.8 years (SD 6.4)

**Gender:** 71% female

**Setting:** 26 UK universities

**Inclusion criteria:** Students aged 18+ with insomnia (scoring ≤6 on the Sleep Condition Indicator, which corresponds to clinically significant insomnia), who had internet access and were willing to be randomised

**Exclusion criteria:** Current or planned psychological therapy for sleep problems, current use of sleep medication prescribed by a doctor, or a self-reported diagnosis of bipolar disorder, epilepsy, or parasomnias (e.g., sleepwalking)

**Key demographic detail:** 78% were undergraduates, 22% postgraduates; 76% white ethnicity; 12% reported a current mental health diagnosis (most commonly depression or anxiety)

**Sleep Condition Indicator (SCI):** 0–10 scale, higher = better sleep. A score ≤6 indicates clinically significant insomnia. This was the primary sleep outcome.

**Green Paranoid Thoughts Scale (GPTS), Part B:** 0–40 scale, higher = more persecutory paranoia. Participants rated how often they had paranoid thoughts over the past month.

**Specific Psychotic Experiences Questionnaire (SPEQ) – Hallucinations subscale:** 0–40 scale, higher = more hallucinatory experiences (e.g., hearing voices, seeing things others don't). Assessed over the past month.

**Insomnia Severity Index (ISI):** 0–28 scale, higher = worse insomnia. Used as a secondary sleep measure.

**Patient Health Questionnaire (PHQ-9):** 0–27 scale, higher = more depression. Used as a secondary outcome.

**Generalised Anxiety Disorder Assessment (GAD-7):** 0–21 scale, higher = more anxiety. Secondary outcome.

**Warwick-Edinburgh Mental Wellbeing Scale (WEMWBS):** 14–70 scale, higher = better wellbeing. Secondary outcome.

**Work and Social Adjustment Scale (WSAS):** 0–40 scale, higher = worse functioning. Secondary outcome.

**Nightmare frequency:** Self-reported number of nightmares in the past week. Secondary outcome.

**Sleep diary:** Participants in the intervention group completed daily online sleep diaries (bedtime, wake time, sleep onset latency, wake after sleep onset, total sleep time, sleep quality rating). Control group did not complete sleep diaries.

All assessments were completed online at weeks 0 (baseline), 3 (mid-treatment), 10 (end of treatment), and 22 (follow-up, 12 weeks after treatment ended).

**Design:** This was a single-blind, parallel-group, randomised controlled trial (RCT). Participants were randomly assigned 1:1 to receive digital CBT-I or usual care. The research team (outcome assessors, data analysts) were blinded to treatment allocation. Participants could not be blinded because they knew whether they were receiving an active sleep programme or not — this is standard for psychological intervention trials.

**Randomisation:** Simple randomisation was used (no stratification). The allocation sequence was computer-generated and concealed from the research team. Participants were randomised immediately after completing baseline assessments.

**Blinding:** The trial was single-blind — the research team was masked to treatment allocation. Participants were not blinded. The authors note that because the primary outcomes were self-reported, the lack of participant blinding could introduce bias (participants who know they're getting treatment might report more improvement). However, they argue that the objective nature of some sleep measures (e.g., sleep efficiency calculated from diaries) partially mitigates this.

**Duration:** The intervention lasted 6 weeks (6 sessions). Primary outcomes were assessed at week 10 (4 weeks after the last session, to capture sustained effects). Follow-up continued to week 22 (12 weeks post-treatment).

**Statistical approach:** Intention-to-treat analysis (all randomised participants included, regardless of how much of the intervention they completed). Missing data were handled using multiple imputation. The primary analysis used linear regression adjusted for baseline scores and university site. Mediation analysis tested whether changes in insomnia at week 3 predicted changes in paranoia and hallucinations at week 10, controlling for baseline levels.

**What this design can prove:**

**Causality:** Because participants were randomly assigned, the design can establish that the sleep intervention *caused* the improvements in paranoia and hallucinations — not the other way around, and not due to some third factor.

**Mediation:** The mediation analysis (testing whether insomnia improvement at week 3 predicted later psychotic symptom reduction at week 10) provides evidence for the causal pathway: fixing sleep → better mental health.

**Generalisability:** Large sample (n=3,755) from 26 universities across the UK, so results are reasonably generalisable to university students.

**What this design cannot prove:**

**Generalisability beyond students:** The authors explicitly note this. University students are younger, more educated, and have different sleep patterns and stress levels than the general population.

**Mechanism:** While mediation analysis suggests insomnia causes psychotic experiences, it doesn't tell us *why* — whether it's through reduced cognitive resources, increased emotional reactivity, or some other pathway.

**Long-term effects:** Follow-up was only 12 weeks post-treatment. We don't know if effects persist for 6 months or a year.

**Objective sleep measurement:** Sleep was measured via self-report and sleep diaries, not polysomnography (lab-based sleep study) or actigraphy (wrist-worn movement trackers). Self-reported sleep can be inaccurate (people often underestimate how much they sleep, especially when anxious).

**Major methodological weaknesses:**

1. **No active control group:** The comparator was "usual care" — not a placebo or attention-control condition. This means the effects could be partly due to the non-specific effects of engaging in a structured programme (expectancy, attention, routine) rather than the specific CBT-I components.

2. **No blinding of participants:** As noted, this is a potential source of bias for self-reported outcomes.

3. **High dropout:** At week 10, 72% of the intervention group and 67% of the control group completed assessments. By week 22, this dropped to 66% and 62% respectively. While multiple imputation was used, differential dropout could bias results.

4. **Self-selection:** Participants were volunteers who responded to advertisements about a sleep study. They may have been more motivated or more sleep-disturbed than the average student.

5. **Single measure of hallucinations:** The SPEQ hallucinations subscale is a brief self-report measure, not a clinical interview. It captures a range of experiences from mild (hearing your name called) to severe (full auditory hallucinations), but doesn't distinguish between them.

**Primary outcomes (week 10, end of treatment):**

**Insomnia (SCI):** The CBT-I group improved by 4.78 points more than the control group (95% CI 4.29 to 5.26, Cohen's d = 1.11, p < 0.0001). This is a very large effect — the largest ever seen in an insomnia treatment trial.

**Paranoia (GPTS Part B):** The CBT-I group improved by 2.22 points more than the control group (95% CI 1.45 to 2.98, Cohen's d = 0.19, p < 0.0001). This is a small effect.

**Hallucinations (SPEQ):** The CBT-I group improved by 1.58 points more than the control group (95% CI 1.18 to 1.98, Cohen's d = 0.24, p < 0.0001). This is a small-to-medium effect.

**Secondary outcomes (week 10):**

**Depression (PHQ-9):** Reduced by 1.82 points more in the CBT-I group (95% CI 1.38 to 2.26, Cohen's d = 0.25, p < 0.0001)

**Anxiety (GAD-7):** Reduced by 1.48 points more (95% CI 1.10 to 1.86, Cohen's d = 0.22, p < 0.0001)

**Nightmares:** Reduced by 0.26 nightmares per week more (95% CI 0.17 to 0.35, Cohen's d = 0.17, p < 0.0001)

**Wellbeing (WEMWBS):** Improved by 2.60 points more (95% CI 1.89 to 3.31, Cohen's d = 0.25, p < 0.0001)

**Work/social functioning (WSAS):** Improved by 1.68 points more (95% CI 1.09 to 2.27, Cohen's d = 0.18, p < 0.0001)

**Mediation analysis:**

Insomnia improvement at week 3 significantly mediated the effect of the intervention on paranoia at week 10 (indirect effect: -0.84, 95% CI -1.17 to -0.54, p < 0.001)

Insomnia improvement at week 3 significantly mediated the effect on hallucinations at week 10 (indirect effect: -0.56, 95% CI -0.75 to -0.39, p < 0.001)

This means that roughly 38% of the reduction in paranoia and 35% of the reduction in hallucinations was explained by the improvement in sleep itself — not by other aspects of the intervention.

**Sustained effects (week 22, 12 weeks post-treatment):**

All primary outcomes remained significantly improved compared to control

Insomnia: d = 0.78 (still a large effect)

Paranoia: d = 0.16 (small but significant)

Hallucinations: d = 0.18 (small but significant)

Depression: d = 0.20

Anxiety: d = 0.17

**Adverse events:** None reported in either group.

To put these numbers in plain English:

**Sleep improvement:** The average student with insomnia (SCI score ~4 out of 10) moved to a score of ~8 out of 10 after CBT-I — essentially moving from "clinically significant insomnia" to "normal sleep." This is a massive effect. For context, the difference between a good sleeper and someone with chronic insomnia is typically about 4–5 points on this scale.

**Paranoia reduction:** A 2.22-point drop on a 40-point scale. The average student with high paranoia (scoring ~12 out of 40) dropped to ~10 out of 40. This is a modest but meaningful reduction — roughly equivalent to going from "I often worry that others are plotting against me" to "I occasionally have these thoughts."

**Hallucination reduction:** A 1.58-point drop on a 40-point scale. The average student scoring ~8 out of 40 dropped to ~6.5 out of 40. This is a small shift — equivalent to going from "I sometimes hear my name called when no one is there" to "this happens rarely."

**Depression reduction:** A 1.82-point drop on the PHQ-9 (27-point scale). This is roughly the difference between "moderate depression" and "mild depression" — clinically meaningful but not transformative for most people.

**Nightmare reduction:** About 1 fewer nightmare every 4 weeks. For someone having 2–3 nightmares per week, this is a noticeable but not dramatic change.

The key insight: fixing sleep produces a *large* effect on sleep itself, and a *small-to-moderate* effect on mental health symptoms. The sleep improvement is the engine; the mental health benefits are the downstream consequence.

**Acknowledged by authors:**

1. **Student population only** — results may not generalise to older adults, non-students, or clinical populations with severe mental illness

2. **No active control group** — cannot rule out non-specific effects of engaging in a structured programme

3. **Self-report outcomes** — no clinician-rated or objective measures of psychotic experiences

4. **No blinding of participants** — potential expectancy effects

5. **High dropout** — 28–38% missing data at follow-up

6. **Single measure of hallucinations** — brief self-report may miss important dimensions

7. **No assessment of sleep architecture** — cannot determine whether changes in specific sleep stages (e.g., REM sleep) drove the effects

**Additional critical notes:**

8. **The effect on paranoia and hallucinations, while statistically significant, was small** — Cohen's d of 0.19–0.24 means the average person in the CBT-I group improved only slightly more than the average person in the control group. For many individuals, the change may not be noticeable in daily life.

9. **The mediation analysis is correlational** — while it shows that early sleep improvement predicts later mental health improvement, it doesn't definitively prove causation. There could be a third variable (e.g., reduced stress) that improves both sleep and mental health simultaneously.

10. **The sample was predominantly female (71%) and white (76%)** — results may not apply equally to men or ethnic minorities.

11. **The intervention was highly structured and automated** — results may not generalise to face-to-face CBT-I, which is more flexible but also more expensive and less scalable.

12. **No objective sleep measurement** — self-reported sleep is known to be unreliable, especially for sleep onset latency (people tend to overestimate how long it takes them to fall asleep) and total sleep time (people underestimate). The actual sleep improvements may be smaller than reported.

13. **The "usual care" control group may have sought other treatments** — 12% of the control group reported starting some form of psychological therapy during the study, which could have reduced the apparent effect of the intervention.

For someone running their own n=1 experiment:

**What to test:**

**Intervention:** A structured, 6-session cognitive behavioural therapy for insomnia (CBT-I) programme. The key components are:

1. **Sleep restriction:** Limiting time in bed to match actual sleep time (e.g., if you sleep 6 hours but spend 8 hours in bed, restrict to 6.5 hours initially)

2. **Stimulus control:** Only go to bed when sleepy; get out of bed if awake for more than 20 minutes; use the bed only for sleep and sex

3. **Cognitive restructuring:** Challenge unhelpful beliefs about sleep (e.g., "I'll never function tomorrow if I don't sleep 8