Sleep and Mental Health: What a Randomized Trial Showed About Cause and Effect

What the OASIS trial actually showed

The 2017 OASIS trial gave the strongest experimental signal yet that poor sleep is a cause, not only a symptom, of some mental health problems. When researchers at the University of Oxford treated insomnia with an automated digital course of cognitive behavioural therapy in 3,755 university students, the treated group reported less insomnia and, alongside it, fewer episodes of paranoia and fewer hallucinatory experiences. A mediation analysis then estimated that improvement in sleep accounted for a large share of the mental health benefit, while the reverse pathway explained almost none. That asymmetry is what lets careful scientists talk about direction rather than mere association.

The design that made causal talk possible

Association studies had long shown that sleep and mental illness travel together. The problem is that correlation cannot separate three possibilities: bad sleep drives symptoms, symptoms drive bad sleep, or a third factor drives both. A randomised controlled trial breaks that knot by intervening on one variable and leaving everything else to chance.

OASIS (Oxford Access for Students Improving Sleep), led by Daniel Freeman and colleagues and published in The Lancet Psychiatry, did exactly that. Between 2015 and 2016, students with insomnia across 26 UK universities were randomly assigned to a digital CBT program for insomnia (1,891 people) or to usual care (1,864). Because assignment was random, the two groups should have differed only in the sleep treatment, so any downstream gap in mental health can be attributed to that intervention rather than to pre-existing differences.

At 10 weeks the sleep effect was large (Cohen's d around 1.1). The effects on the two primary mental health outcomes were real but small: paranoia and hallucinatory experiences each fell modestly (d near 0.2). Secondary outcomes moved in the same direction, with somewhat larger effects on depression (d about 0.5) and anxiety (d about 0.3), plus improvements in nightmares, psychological wellbeing, and prodromal psychotic symptoms.

What mediation analysis adds, and what it cannot

Showing that a sleep treatment improves mental health is one step. Showing that it works through sleep is another. Mediation analysis tries to open the box, asking how much of the treatment's effect on paranoia travels along the path from treatment, to better sleep, to less paranoia.

Using a mediation approach in the tradition of Baron and Kenny with linear mixed effects models, the authors estimated that change in sleep accounted for roughly 58% of the treatment effect on paranoia and 39% of the effect on hallucinations over 10 weeks. They also used the earlier week-3 sleep change as the mediator, which strengthens the causal reading because the proposed cause is measured before the outcome. Just as important, they checked the reverse: change in paranoia explained only about 3.8% of the change in sleep, and hallucinations about 3.4%. Sleep looked like the driver; the symptoms did not.

Mediation analysis still rests on assumptions. It presumes the treatment altered sleep first and that no unmeasured factor confounds the sleep-to-symptom link. The authors were candid that outcomes measured together at week 10 cannot fully capture the temporal order, which is precisely why the week-3 mediator matters. Estimates like "58% mediated" are model-dependent, not fixed physical constants, and should be read as direction and rough magnitude rather than precise accounting.

Reading bidirectional evidence without overstating direction

Here is the tension worth holding. Observational research shows the sleep and mental health relationship runs both ways: anxiety and depression disrupt sleep, and disrupted sleep worsens mood and threat perception. OASIS does not overturn that. What a trial can do is isolate one arrow and estimate its strength under controlled conditions. The honest reading is that OASIS provides good experimental support for one direction, that treating insomnia improves symptoms, without claiming the other direction is absent.

Two cautions keep the finding in proportion. First, the mental health effects were small even though the sleep effect was large, so improving sleep is a lever, not a cure. Second, the sample was university students whose psychotic experiences sat mostly in the non-clinical range, which limits how far the results extend to people with established psychotic illness.

The broader literature points the same way while widening the base. A 2021 meta-analysis in Sleep Medicine Reviews pooled 65 randomised trials covering more than 8,600 participants and found that interventions improving sleep quality produced reliable gains in composite mental health, depression, anxiety, and stress, with a dose-response pattern in which bigger sleep improvements tracked bigger mental health improvements. A single trial can be a fluke; a converging body of trials with a dose-response signal is harder to dismiss.

What this means for how we reason

OASIS is a useful teaching case in causal inference. Randomisation rules out confounding at baseline, mediation analysis probes mechanism, the timing of the mediator addresses reverse causation, and effect sizes keep enthusiasm honest. Each piece answers a different question, and none alone would justify the causal claim. The reasonable conclusion is that insomnia is one modifiable contributor to certain mental health symptoms, that treating it deserves to be taken seriously, and that a slogan like "sleep causes everything" would overstate what any single study can show.

This article is educational and is not medical advice; decisions about sleep or mental health treatment belong with a qualified clinician.