Digital CBT-I for your patients: what the evidence shows

Major guidelines name CBT-I (cognitive behavioural therapy for insomnia) as the first treatment for chronic insomnia, and the 2023 European guideline explicitly includes digital delivery. Trials and meta-analyses of digital CBT-I show clear benefit over waiting lists and usual care, and face-to-face delivery stays ahead by a modest margin. Dropout is the main weakness. ^[1][3][5]

What the research says

The guidelines agree on the treatment and differ on the format. The American College of Physicians recommends CBT-I as the initial treatment for every adult with chronic insomnia disorder, a strong recommendation on moderate-quality evidence. ^[1] The American Academy of Sleep Medicine gives multicomponent CBT-I a strong recommendation but found insufficient evidence to recommend for or against any delivery method, internet and self-help included. ^[2] Only the 2023 European Insomnia Guideline goes further, naming CBT-I first-line "either applied in-person or digitally" at its top evidence grade. ^[3] The largest meta-analysis, 33 randomised trials, found Insomnia Severity Index (ISI) scores fell 5.00 points more than in control groups after treatment and 3.48 points more at one year; face-to-face CBT-I stayed ahead by 3.07 ISI points, inside the authors' 4-point non-inferiority margin. ^[5] The weak point is completion: between about a third and over two fifths of participants in the trials below did not finish. ^{[9][12][13][14]} In plain terms, digital CBT-I works, trails a therapist slightly, and needs help keeping patients engaged.

The studies behind this

• American College of Physicians (2016): CBT-I as the initial treatment for all adults with chronic insomnia disorder. Strong recommendation, moderate-quality evidence. ^[1]
• American Academy of Sleep Medicine (2021): strong recommendation for multicomponent CBT-I. It found insufficient evidence to recommend any particular delivery method, internet and self-help included. ^[2]
• European Insomnia Guideline (2023): CBT-I is first-line for adults of any age, "either applied in-person or digitally" (the guideline's top evidence grade). ^[3]
• Soh meta-analysis (2020): 33 randomised trials. Insomnia Severity Index (ISI) scores fell 5.00 points more than in control groups after treatment, and 3.48 points more at one year. Face-to-face CBT-I was ahead by 3.07 ISI points, inside the 4-point margin the authors set for non-inferiority. ^[5]
• Dropout: between about a third and over two fifths of people in the trials below did not finish. ^{[9][12][13][14]}

Where the guidelines stand

• ACP (2016): drug treatment is a second step. Add it by shared decision-making when CBT-I alone has not worked. That recommendation is weak, on low-quality evidence. ^[1]
• AASM (2021): the strong recommendation rests mainly on studies where a trained professional delivered CBT-I. For internet and self-help formats the task force found too little evidence to recommend for or against a delivery method. ^[2]
• Europe (2023): digital CBT-I is named alongside in-person CBT-I at the top evidence grade. ^[3]
• Our reading: the guidelines do not conflict. The European text endorses digital delivery. The American text does not rule it out, and says the evidence for specific formats was not enough to grade.

How big is the effect?

Three meta-analyses of randomised trials point the same way.

• Zachariae (2016): 11 trials, 1,460 participants. Effect sizes (Hedges' g) ranged from 0.21 to 1.09 across seven outcomes: insomnia severity, sleep efficiency, subjective sleep quality, wake after sleep onset, sleep onset latency, total sleep time and number of night-time awakenings. Effects were comparable to face-to-face CBT-I and generally held at 4 to 48 weeks. ^[4]
• Seyffert (2016): 15 trials. ISI fell 4.3 points more than in controls, and total sleep time rose by about 20 minutes more. Depression scores fell 2.3 points more. No significant differences from therapist-delivered in-person treatment on sleep efficiency, total sleep time or ISI. ^[6]
• Soh (2020): 33 trials, 4,719 people randomised to digital CBT-I and 4,645 to control. Results were sustained at one year. The results differed a lot between trials (I-squared 79%), and the comparison with face-to-face care favoured face-to-face delivery. ^[5]

Two details matter when you counsel a patient.

• Controls were mostly waiting lists, usual care or education. These meta-analyses mainly show that digital CBT-I beats doing nothing much. ^[5]
• Average gains in total sleep time are modest. In the Seyffert meta-analysis, total sleep time averaged 5.7 hours at baseline and rose by about 20 minutes more than in controls. Our reading: the clearer gains are in insomnia severity and in the diary measures of falling asleep and staying asleep. ^[4][6]

The trials clinicians ask about

• Ritterband (2009). A structured internet program built on the standard CBT-I components, 44 adults analysed. ISI fell from 15.73 to 6.59 in the program group and stayed flat in the wait-list group (16.27 to 15.50). Gains held at six months. A small, early trial. ^[7]
• Espie (2012). 164 adults with insomnia disorder, an automated web program with an animated virtual therapist, against both a credible placebo and usual care. Diary sleep efficiency improved by large margins compared with both the placebo and usual care; daytime outcomes showed only modest superiority over placebo (d 0.23 to 0.37). The placebo arm is what makes this trial useful. Our reading: it speaks to whether simply engaging with a website could explain the result. ^[8]
• Vincent and Lewycky (2009). A Canadian trial: 118 adults, a five-week online program against a wait list. Sleep quality, insomnia severity and daytime fatigue improved. Attrition was 33%. ^[9]
• Espie (2019). 1,711 adults with insomnia symptoms, digital CBT-I against sleep hygiene education. Small gains in functional health and well-being, and a large gain in sleep-related quality of life, mediated largely by the improvement in insomnia. 91% of participants were white and 78% were women. Several authors report ties to the company that makes the program tested. ^[10]
• Freeman, OASIS (2017). 3,755 university students, digital CBT-I against usual practice. Insomnia improved a lot (Cohen's d 1.11 at 10 weeks), and paranoia (d 0.19) and hallucinatory experiences (d 0.24) improved a little. The authors reported no adverse events. They note it is not known whether this holds beyond students. ^[11]
• Van der Zweerde (2020). The trial closest to primary care: 134 patients in 15 general practices, nurse-guided internet CBT-I against care as usual. ISI effect size 1.66 at eight weeks and 1.02 at 26 weeks. Sleep medication use did not change significantly. No adverse events were reported. ^[12]

Who struggles without support?

• How many drop out. In a 207-person trial of a six-week program, 34% did not complete all six sessions. Among those who stopped, 57% did so before the fourth session. ^[13] In an 8-module program, 44% of 233 patients were classed as dropouts, meaning they had not finished at least seven modules within 12 weeks. ^[14] The Canadian trial above lost 33%. ^[9] In the nurse-guided general practice trial, 68% completed the whole program. ^[12]
• Who drops out. In one study, higher baseline sleep time, higher depression scores and milder insomnia predicted not finishing. ^[13] In the Canadian trial, people referred for treatment dropped out more often than people who found the program themselves. ^[9]
• A larger study found no baseline predictor. None of the baseline measures it examined predicted dropout. Two early behaviours did: how long the first module took, and how many diary entries were completed in week one. ^[14]
• Why people say they stopped. Distractions from daily life, a sense the content was not useful, and trouble restarting after a break. ^[14]
• Support helps. In the Zachariae meta-analysis, more personal clinical support and longer treatment were linked to larger effects. Higher dropout was linked to smaller effects. ^[4]
• Our reading: the evidence on who drops out is mixed and drawn from a few studies, so it is more useful for deciding who to check in with early than for deciding who should be excluded.

Who needs assessment first?

CBT-I is recommended for adults with chronic insomnia, including people with comorbid conditions. ^[3] Some patients need a closer look before starting, particularly because sleep restriction is part of most programs.

• High-risk occupations and driving. The AASM guideline notes that sleep restriction may be contraindicated for people in high-risk jobs, such as heavy machinery operators or drivers. ^[2]
• Mania or hypomania risk. The same guideline names people predisposed to mania or hypomania. ^[2]
• Poorly controlled seizures. Also named in the same passage. Where sleep restriction is used on its own, the guideline adds excessive daytime sleepiness to the list. ^[2]
• Suspected sleep apnea, periodic limb movements or another sleep disorder. The European guideline advises polysomnography when other sleep disorders are suspected, and for treatment-resistant insomnia. ^[3] Co-morbid insomnia and sleep apnea is common: a review reports earlier estimates of 30% to 50%. ^[15]
• Early side effects are expected. Daytime fatigue, sleepiness, irritability and concentration problems are mostly confined to the early stages of treatment, when the behavioural parts are introduced, and typically resolve by the end of treatment. ^[2]

Whether a digital program is suitable for these patients, or only with oversight, is a clinical judgement. None of the trials above was designed to answer it.

How you can support a patient using an app

These are practical suggestions. Where a trial finding backs one, it is cited.

• Name the early weeks in advance. Tell the patient that tired, irritable days are common early on and usually pass. Ask about driving and work hazards. ^[2] Our patient-facing explanation of why sleep restriction can feel worse at first covers the same ground in plain words.
• Ask to see the diary at the first visit. Diary completion in the first week was one of the better predictors of who stayed in. ^[14]
• Check in early. Personal support was associated with larger effects, and nurse guidance worked in general practice. ^[4][12]
• Treat medication as a separate decision. The ACP guideline puts drug treatment after CBT-I alone has been tried, by shared decision-making. ^[1]
• Plan a follow-up. The AASM guideline says follow-up at the end of treatment is indicated, and residual symptoms should be evaluated. ^[2]

For a practice-level route that combines these options, see our stepped-care article for primary care.

Where evidence is thin

• Delivery method. The AASM task force found too little evidence to recommend any delivery format, and its main evidence base was professional-delivered CBT-I. ^[2]
• Who was studied. Many trials recruited volunteers online. Where participant characteristics were reported, samples were mostly white and mostly women, or university students. ^[10][11]
• Comparators. Most controls were waiting lists, usual care or education. ^[5][10]
• Conflicts of interest. Some trials tested programs made by companies in which authors have a stake. ^[10]
• Long-term outcomes. Follow-up is mostly under a year. ^[4][5]
• Complex patients. Among the trials we reviewed, none was designed to answer whether a digital program is safe without oversight for people with mania risk, seizure disorders or high-risk jobs.

Where Small Hours fits

Small Hours is a six-week, self-guided CBT-I program for adults that we build. It has no published trial of its own, and it makes no outcome claims. The evidence above is about other digital programs and does not transfer automatically to any specific product.

What it does, as described on our clinicians page:

• A short morning sleep diary, with sleep efficiency worked out for the patient.
• A weekly sleep-window recommendation from fixed, published rules, not computed by an AI model.
• Screening before any restriction, for bipolar disorder, seizure history, chronic migraine, pregnancy and other major conditions. A patient who flags one is offered an alternative with no calculated sleep window (a consistent wake time instead); the offer is never applied automatically. Signs of sleep apnea prompt the patient to raise it with their doctor.
• A coach that answers only from the program's teaching. It never gives a medication dose or tapering schedule, sends those decisions back to the prescriber or pharmacist, and gives no diagnosis.
• A progress report the patient can bring to you, as a PDF made on their own device from their own diary. The app sends nothing about a patient to their clinician; the report reaches you only if the patient gives it to you.

We designed Small Hours so that it can be used alongside a clinician, which is the arrangement we would choose. Small Hours does not require a clinician. That is a design intent, not a research finding.

General education, not medical advice.