Smartphone-delivered cognitive behavioral therapy for insomnia: A preliminary randomized controlled trial

Abstract

Study objectives

Cognitive behavioral therapy for insomnia (CBT—I) is the first-line treatment for insomnia, but many individuals lack access. This study evaluated the preliminary efficacy of Hvil®, a 9-week, fully automated, smartphone delivered eCBT-I program, in reducing insomnia severity.

Methods

Adults with insomnia (Insomnia Severity Index ≥10) were randomly assigned to an eCBT-I intervention via Hvil® or to a waitlist control group. The primary outcome of insomnia severity was assessed at baseline, weekly during the intervention, and post-intervention. Secondary outcomes included fatigue at baseline and post-intervention and sleep diary-based outcomes assessed daily during the study period. Mixed linear models (MLMs) based on the intent-to-treat sample were used to assess group differences over time. Moderating effects of demographic characteristics and adherence were explored.

Results

A total of 165 participants were randomized to the intervention (n = 83) or control group (n = 82). A significant group × time interaction was observed for insomnia severity (p < .001) favoring the intervention group, yielding a clinically meaningful effect size (Cohen's d = 0.66). Significant improvements were also noted in fatigue, sleep quality, time in bed, and nocturnal awakenings. Effect sizes for secondary outcomes ranged from 0.15 (total sleep time) to 1.06 (time in bed). Greater engagement with the Hvil® app was associated with larger reductions in insomnia severity (p < .001).

Conclusions

Hvil®, an automated eCBT-I intervention, demonstrated significant efficacy in reducing insomnia severity among adults in the general population. While attrition was substantial, the results suggest its potential as a scalable, accessible intervention.

Highlights

• •Hvil® offers fully automated smartphone cognitive behavioral therapy for insomnia.
• •The intervention provided in Hvil® significantly reduced insomnia severity.
• •Hvil® also improved fatigue, time in bed, nocturnal awakenings, and sleep quality.
• •Increased engagement was associated with greater improvements.
• •Fully automated, smartphone-based digital CBT-I is efficacious and scalable.

1. Introduction

Insomnia is a common and distressing disorder characterized by persistent difficulties initiating or maintaining sleep, despite adequate opportunities for sleep, resulting in significant daytime impairment (World Health Organization, 2020). A recently published meta-analysis of 47 studies established that 12.4% (95% CI 9.0% to 16.8%) in general population samples met the diagnostic criteria for an insomnia disorder (van Straten et al., 2025). The associations between insomnia and various adverse health outcomes are well-documented in the literature, showing chronic insomnia to be associated with increased risk of physical and psychological morbidity, including cardiovascular disease (Laugsand et al., 2011; Palagini et al., 2013), mental disorders (Hertenstein et al., 2019; Staner, 2010), and dementia (de Almondes et al., 2016). At the societal level, insomnia has been found associated with increased healthcare utilization, higher levels of work absenteeism, reduced work-related productivity, diminished learning capacity, and poorer academic performance (Curcio et al., 2006; Daley et al., 2009a; Daley et al., 2009b). Thus, beyond the individual health implications, insomnia, when left untreated, represents considerable public health concern due to the substantial societal costs.

Pharmacotherapy remains widely used for managing insomnia (Parsaik et al., 2016; Trauer et al., 2015). While certain sleep medications may be effective in treating acute insomnia, they have generally been associated with low tolerability due to a range of associated adverse effects such as dizziness, fatigue, and increased risk of accidents (Sateia et al., 2017), e.g., fall accidents in the elderly (Glass et al., 2005). Furthermore, evidence of long-term efficacy is generally lacking (De Crescenzo et al., 2022). National and international guidelines on the management of insomnia advise against long-term use of sleep medication, typically beyond two to four weeks, and instead recommend cognitive behavioral therapy for insomnia (CBT—I), as the first-line treatment for chronic insomnia (Edinger et al., 2021a; Morin et al., 2023; Riemann et al., 2023). CBT-I is a multicomponent intervention aiming at changing maladaptive sleep-related behaviors (e.g., daytime napping and irregular sleep schedules) and cognitions (e.g., unrealistic expectations and catastrophizing thoughts about sleep) that contribute to maintaining insomnia over time (Morin and Benca, 2012). CBT-I usually involves two or more of the following components; sleep restriction, stimulus control, cognitive therapy, relaxation, and sleep hygiene education (Morin and Benca, 2012). Meta-analyses of a large and growing number of trials support CBT-I as a highly efficacious treatment of insomnia associated with considerable improvements in insomnia severity, sleep quality, and a number of other sleep-related parameters, e.g., sleep onset latency (SOL) and sleep efficiency (SE), with minimal adverse events (Trauer et al., 2015; van der Zweerde et al., 2019; van Straten et al., 2018). Beyond being effective short-term, these effects seem to be sustained at longer follow-ups (Trauer et al., 2015; van der Zweerde et al., 2019), with CBT-I demonstrating more durable effects than pharmacotherapy (Zhang et al., 2022). Additional indirect effects of CBT-I include improvements in mood (Trauer et al., 2015), daytime functioning (Trauer et al., 2015; Van Houdenhove et al., 2011), and well-being (Ashworth et al., 2015; Espie et al., 2019).

Despite growing recognition of CBT-I as first-line treatment, substantial barriers limit its general implementation. These include a shortage of trained CBT-I providers, low referral rates from primary care, and low help-seeking and awareness of non-pharmacological treatment options among individuals with insomnia (Conroy and Ebben, 2015; Koffel et al., 2018; Roach et al., 2021). In response, alternative delivery formats have been developed and evaluated to expand access, reduce patient burden, and minimize healthcare utilization (Koffel et al., 2018). These include remote formats such as telephone and videoconferencing, as well as fully automated digital formats such as web-based platforms and mobile applications (eCBT-I).

While in-person delivered CBT-I generally appears to yield larger improvements in self-reported insomnia symptoms compared to eCBT-I, the differences between delivery formats tend to be smaller for diary-based sleep outcomes such as SOL, WASO, total sleep time (TST), and SE (Gao et al., 2022; Hasan et al., 2022; Knutzen et al., 2024; Simon et al., 2023; Soh et al., 2020). Furthermore, when examined in head-to-head comparisons, differences are generally within the suggested minimal important differences (MIDs) (Knutzen et al., 2024). Meta-analyses further affirm the clinical efficacy of eCBT-I in improving insomnia severity, sleep quality, and sleep diary-based outcomes such as SOL, WASO, and SE (Cheng and Dizon, 2012; Seyffert et al., 2016; Ye et al., 2016; Zachariae et al., 2016), not only in individuals with insomnia as their primary problem, but also in those with comorbid insomnia, e.g., cancer patients and survivors (Nissen et al., 2024).

Recent studies highlight both the efficacy and cost-effectiveness of fully automated eCBT-I (Darden et al., 2021; Hasan et al., 2022; Soh et al., 2020), including in the long term (Amidi et al., 2022), with meta-analytic evidence indicating medium to large effects on insomnia severity across diverse control conditions (Hwang et al., 2025). Given that over 80% of U.S. and Western European adults now own smartphones and use continues to rise (GSMA Intelligence, 2025; Pew Research Center, 2024), mobile eCBT-I has the potential to overcome major access barriers while enabling personalized delivery and real-time assessment, thereby enhancing ecological validity (Horsch et al., 2017; Konrath, 2015).

In collaboration with the health-tech company Enversion A/S (https://www.enversion.com/en/), our team developed Hvil®, a fully automated, smartphone-based CBT-I intervention incorporating the five core components of CBT—I: sleep restriction, stimulus control therapy, relaxation, cognitive therapy, and sleep hygiene education.

The present study evaluated the preliminary efficacy of Hvil® in reducing insomnia severity (primary outcome) as well as improving sleep diary metrics, and alleviating fatigue, relative to waitlist controls. The study was undertaken to inform a larger factorial optimization study and a potential subsequent confirmatory RCT by providing an initial estimate of the magnitude and direction of the effect of Hvil®.

2. Materials and methods

2.1. Study design and participants

The efficacy of Hvil® (Version 1.1.0–1.1.3) was tested with a randomized controlled trial (RCT) with participants enrolled from Sept. 6 to Sept. 9, 2022. Participants were recruited via a newsletter e-mailed to approximately 145,000 members of TryghedsGruppen, the largest shareholder in Tryg Insurance, a Danish private insurance company, with approximately 1.4 million members corresponding to almost a fourth of the Danish population. The newsletter included links to an online information pamphlet about the study and a brief secure online screening questionnaire.

Eligible participants were Danish-speaking adults (≥18 years) with access to a smartphone, adequate technological proficiency, and who experienced symptoms of clinical insomnia, indicated by a score of ≥10 on the Insomnia Severity Index (ISI), which has been suggested as the optimal cut-off score to detect insomnia cases in community samples (Bastien et al., 2001; Morin et al., 2011). Exclusion criteria were psychological or physical comorbidities directly affecting sleep, other diagnosed sleep disorders (e.g., sleep apnea, restless legs syndrome), a shift-work schedule, current maternity/paternity leave, and limited Danish reading proficiency, all assessed using a self-report screening questionnaire. Use of sleep medication was permitted and not considered grounds for exclusion. The study was approved by the Regional Science Ethical Committees (Registration No. 1–10–72-9-21) and the Danish Data Protection Agency and was preregistered at ClinicalTrials.gov (NCT05561790). As this was a preliminary RCT, an a priori statistical power analysis was not conducted to determine the sample size.

2.2. Randomization and blinding

Eligible participants who signed the informed consent form were randomly assigned to the intervention or waitlist control and received an e-mail with instructions on how to download the app. Randomization was based on a 1:1 allocation ratio and was conducted in REDCap (Research Electronic Data Capture), a secure web-based software platform designed to support data capture for research studies, hosted at Aarhus University (Harris et al., 2019; Harris et al., 2009). Participants were informed that they would either receive the intervention immediately or be placed on a waitlist. All research staff were blinded to the allocation sequence until interventions were assigned. One staff member was responsible for handling technical support upon randomization and was thus not blinded to the allocation status of participants throughout the remaining study period.

2.3. Intervention and control conditions

The intervention was based on the current consensus on non-pharmacological treatment of insomnia and builds on previous validated CBT—Is (Edinger et al., 2021a; Morin et al., 2006; Wilson et al., 2019). It was delivered as a 9-week treatment program consisting of the five widely used CBT-I components (Espie, 2022): sleep restriction, stimulus control therapy, cognitive therapy, relaxation techniques, and sleep hygiene. The sleep restriction component aimed to restrict the time spent in bed to increase the sleep drive and improve sleep efficiency (SE), defined as total sleep time (TST) relative to time spent in bed (TIB) (i.e., TST/TIB × 100%). TIB was set to each participant's average TST (minimum 5 h) based on at least 8 days of pre-treatment sleep diary data and then adjusted weekly to achieve an SE ≥85% (Kyle et al., 2015; Trauer et al., 2015). The stimulus control component involved behavioral instructions aimed at strengthening the association between the sleep environment and sleep with the purpose of eliminating conditioned associations between non-sleep behavior and arousal and the sleep environment (Bootzin and Perlis, 2011; Trauer et al., 2015). The cognitive component aimed to identify, challenge, and modify dysfunctional beliefs about sleep and sleep loss using cognitive restructuring techniques (Belanger et al., 2006). The relaxation component involved a daily winding-down routine before bedtime along with guided imagery and relaxation exercises aimed at reducing physical and cognitive arousal and limiting intrusive thoughts that interfere with sleep (Espie, 2022; Morgenthaler et al., 2006). Sleep hygiene education provided specific information relating to lifestyle (e.g., diet, exercise, substance use) and environmental factors (e.g., light, noise, room temperature) that may interfere with or promote sound sleep (Posner and Gehrman, 2011; Trauer et al., 2015). Each component included four modules that were individualized based on each participant's responses in the app: information (content and purpose), assessment (severity of the targeted issue), application (recommended activities/exercises), and evaluation (treatment outcomes).

Participants in the control condition were encouraged to track their sleep using the integrated sleep diary in Hvil® but did not receive access to the treatment components. Upon completing the posttreatment questionnaire, participants in the control condition were granted access to the full treatment program.

2.4. Assessment

The primary outcome was insomnia severity measured with the Insomnia Severity Index (ISI) (Bastien et al., 2001). The ISI consists of seven 5-point Likert scale items (0 = no problem; 4 = very severe problem) providing a total score range of 0–28, with higher scores indicating more severe insomnia (Morin et al., 2011). Internal consistency (Cronbach's α) in the present sample (0.77) was considered acceptable (Nunnally and Bernstein, 1994). Baseline and posttreatment (11 weeks after randomization) data were collected and managed using REDCap. Additional baseline information was collected in the Hvil® app, including sociodemographic data, educational information, work status and type, insomnia duration, use of sleep medication, and sleep-related health behaviors. Participants were also prompted weekly to complete the ISI in the app throughout treatment.

Every day throughout the study period, participants were prompted via notifications to complete the consensus sleep diary (Carney et al., 2012) integrated into Hvil®. Diary-derived sleep characteristics included TIB (min), TST (min), SOL (min), WASO (min), nocturnal awakenings (NA, number), SE (%), and self-rated sleep quality on a scale of zero to 10 (SQ). Number and length of daytime naps and daytime sleepiness/tiredness were also assessed. For sleep diary variables, baseline outcomes were calculated as the mean of the sleep diary outcomes registered during the baseline assessment period.

Daytime fatigue was assessed at baseline and posttreatment with the 13-item Functional Assessment of Chronic Illness Therapy for Fatigue (FACIT-Fatigue) (Yellen et al., 1997). Total scores range from 0 to 52, with higher scores indicating less fatigue. Internal consistency in the present sample was 0.89. Application usability and satisfaction were assessed posttreatment with the mHealth App Usability Questionnaire (MAUQ) (Zhou et al., 2019). The MAUQ includes 18 items scored on a 7-point scale assessing the ease of use and user satisfaction, the experience of system information arrangement, and the usefulness of mobile health applications. Usability is determined as the average of responses to all items (range 1–7), with higher scores indicating higher usability. Internal consistency in the present sample was 0.97.

We intended to assess cognitions about sleep with the Dysfunctional Beliefs and Attitudes about Sleep Scale (DBAS-16) (Morin et al., 2007) and psychological distress assessed with the Depression, Anxiety, and Stress Scales-21 (DASS-21) (Sinclair et al., 2012). These outcomes were omitted from the analyses due to a technical error in the digital distribution of these questionnaires.

2.5. Statistical analyses

Possible baseline differences between treatment groups were explored with t-tests or χ² tests as appropriate. In case of missing data at the item level within a scale, mean substitution based on the respondent's mean score across completed items was applied to scales with internal consistencies of at least 0.70 for which respondents had completed at least 50% of the items (Schafer and Graham, 2002). Mixed linear models (MLMs) based on the intent-to-treat (ITT) sample and maximum likelihood estimation were used to compare groups over time on the primary and all secondary continuous outcome variables. MLM accounts for the dependence structure introduced by repeated measurements within subjects and allows for the inclusion of participants with missing data at one or more measurement time points (Heck et al., 2022). For all outcomes, a two-level model with time nested within individuals was specified. A linear time function was fitted from baseline through posttreatment to compare endpoint effects between groups. Time was defined as time in treatment for intervention participants and time in study after baseline period for control group participants (weeks). MLMs were fitted using random intercepts and random slopes for time as this provided the best fit, as assessed by a change in the −2 log likelihood fit statistics. A variance/covariance structure modeled as identity at level 1 and variance components at level 2 was chosen as the best fit across models. Bootstrapping based on 1000 bootstrap samples was performed for sleep diary outcomes due to high degrees of skewness, kurtosis, and heteroscedasticity of level 1 residuals. P-values were calculated for two-sided tests. The false discovery rate for multiple baseline comparisons was controlled using the Benjamini-Hochberg procedure (Benjamini and Hochberg, 1995) and the family-wise error rate for multiple secondary outcomes was controlled using Holm's procedure (Holm, 1979). Effect sizes were expressed as Cohen's d computed as the absolute between-group difference in posttreatment means divided by the pooled standard deviation at posttreatment, with effect magnitudes of 0.2, 0.5, and 0.8 regarded as small, medium, and large effects, respectively (Cohen, 1988).

The moderating effects of age, gender, educational level, marital status, and baseline ISI score on the primary outcome were explored as three-way interaction terms (group×time×moderator). Effects on the primary outcome of adherence, conceptualized as the number of completed activities and the number of commenced treatment components, were examined in the intervention group as the time × adherence interaction. Predictors of study and treatment dropout were examined using logistic regression, and predictors with sparse data were followed up with Firth's penalized logistic regression to assess the robustness of results (Firth, 1993; Peduzzi et al., 1996). Analyses were conducted in IBM SPSS Statistics 29 and R (version 4.5.1).

3. Results

3.1. Participants

Between Sept. 6 and 9, 2022, a total of 345 subjects responded and completed screening for eligibility (see Fig. 1). Of these, 178 either did not satisfy the inclusion and exclusion criteria or did not provide informed consent. Thus, 167 participants were randomized (intervention n = 84, control n = 83). Two participants were excluded post-randomization: one participant in the intervention group due to reporting sleep apnea after inclusion, and one participant in the control group due to participation in another study using Hvil®. Of the included 165 (48%) participants, 138 (84%) participants created a Hvil®-user account during the study period (intervention n = 69, control n = 69), and 109 (66%) participants responded to the posttreatment questionnaire (intervention n = 54, control n = 55). Demographic characteristics of intervention and control participants are shown in Table 1. No statistically significant between-group differences were found for any of the demographic variables or insomnia severity at baseline (see Table 1).

Fig. 1.

Study flow diagram.

Note. CBT-I = cognitive behavioral therapy for insomnia.

Table 1.

Participant characteristics.

	Intervention (n = 83)	Control (n = 82)	Total sample (n = 165)	P⁎
Gender, n (%) women	53/83 (63.9)	43/82 (52.4)	96/165 (58.2)	.137
Missing data, n (%)	0	0	0
Age at inclusion, mean (SD)	57.0 (11.3)	57.5 (11.3)	57.2 (11.3)	.750
Missing data, n (%)	0	0	0
Married/partnered, n (%)	42/56 (75.0)	36/52 (69.2)	78/108 (72.2)	.504
Missing data, n (%)	27 (32.5)	30 (36.6)	57 (34.5)
Cohabiting, n (%)	40/56 (71.4)	35/52 (67.3)	75/108 (69.4)	.642
Missing data, n (%)	27 (32.5)	30 (36.6)	57 (34.5)
Education, n (%)				.751
Short (< 2 y of higher ed.)	14/56 (25.0)	10/52 (19.2)	24/108 (22.2)
Medium (2–4 y of higher ed.)	28/56 (50.0)	27/52 (51.9)	55/108 (50.9)
Long (≥ 5 y of higher ed.)	14/56 (25.0)	15/52 (28.8)	29/108 (26.9)
Missing data, n (%)	27 (32.5)	30 (36.6)	57 (34.5)
Working, n (%)	35/56 (62.5)	38/52 (73.1)	73/108 (67.6)	.241
Missing data, n (%)	27 (32.5)	30 (36.6)	57 (34.5)
Work hours, n (%) day only	35/35 (100)	35/38 (92.1%)	70/73 (95.9.)	.241‡
Missing data, n (%)	27 (32.5)	30 (36.6)	57 (34.5)
Insomnia duration (yrs), mean (SD)	7.5 (9.4)	7.1 (7.1)	7.3 (8.4)	.843
Missing data, n (%)	27 (32.5)	38 (46.3)	65 (39.4)
Sleep medication last 30 days, n (%)	10/56 (17.9)	5/46 (10.9)	15/102 (14.7)	.321
Missing data, n (%)	27 (32.5)	36 (43.9)	63 (38.2)
Non-prescription pharmacological sleep aids last 30 days, n (%)†	16/56 (28.6)	8/46 (17.4)	24/102 (23.5)	.185
Missing data, n (%)	27 (32.5)	36 (43.9)	63 (38.2)
Non-pharmacological sleep aids, n (%)§	34/56 (60.7)	25/46 (55.3)	59/102 (57.8)	.517
Missing data, n (%)	27 (32.5)	36 (43.9)	63 (38.2)
Insomnia severity, mean (SD)¶	18.2 (3.8)	18.1 (3.8)	18.2 (3.8)	.889
Missing data, n (%)	0	0	0

^⁎t-test for independent samples or chi square test (two-sided), as appropriate. As no significant differences were found, results were not controlled for multiple comparisons with the Benjamini-Hochberg procedure (1).

^†E.g., dietary supplements, herbal medicine.

^‡P based on Fischer’s Exact Test due to >20% of cells having an expected count of less than 5.

^§Relaxation, acupuncture, music, breathing exercises, or other

^¶based on the Insomnia severity Index, higher scores = more severe insomnia (2).

3.2. Sleep and fatigue outcomes

As shown in Table 2, statistically significant group × time interaction effects were found at post-intervention based on the ITT sample for the primary outcome of insomnia severity (ISI), with a significantly larger reduction in ISI-score from pre to post in the intervention group compared to the control group (see Fig. 2). Significant group × time interaction effects were also found for the secondary outcomes of fatigue (FACIT-Fatigue), TIB, NA, and SQ. Effect sizes (Cohen's d) based on absolute differences at post-intervention ranged from 0.25 (95% CI = −0.35 to 0.85) for reduced NAs to 1.06 (95% CI = 0.43 to 1.70) for reduced TIB (Table 2). The group × time interaction for NA did not remain significant after controlling for the family-wise error rate using Holm's method (Holm, 1979). For the rest of the sleep diary outcomes, the group × time interaction effects were non-significant. Large effect sizes were found for reduced TIB and improvement in SQ, favoring the intervention group; medium effect sizes were detected for improvements in insomnia severity and SE, favoring the intervention group; and small effect sizes were found for improvements in fatigue, and decreased SOL, WASO, and NA, favoring the intervention group.

Table 2.

Insomnia severity, fatigue, and sleep variables at baseline and postintervention.

Variable	Intervention				Control				Group × time⁎			Effect size
	Pre		Post		Pre		Post
	N	Mean (SD)	N	Mean (SD)	N	Mean (SD)	N	Mean (SD)	B	SE	P	d†	95% CI
Insomnia severity, ISI	83	18.20 (3.83)	54	11.07 (6.00)	82	18.12 (3.81)	55	14.51 (4.27)	−0.35	0.10	<.001	0.66	0.27; 1.05
Fatigue, FACIT-F‡	56	34.73 (7.53)	54	39.04 (8.53)	43	34.81 (8.26)	55	36.31 (9.26)	0.40	0.14	.004	0.31	−0.07; 0.69
Time in bed (TIB) (min)	56	508.55 (55.18)	27	425.48 (60.67)	43	502.17 (46.11)	18	481.08 (37.32)	−3.55	1.47	.002	1.06	0.43; 1.70
Total sleep time (TST) (min)	56	364.86 (62.84)	27	367.26 (80.01)	43	348.27 (69.52)	18	378.92 (70.44)	−1.56	1.58	.184	0.15	−0.45; 0.75
Sleep onset latency (SOL) (min)§	56	36.70 (29.63)	27	13.55 (9.38)	43	39.30 (31.18)	18	21.26 (22.51)	0.45	0.55	.440	0.48	−0.13; 1.09
Wake after sleep onset (WASO) (min)	56	58.84 (46.77)	27	26.62 (35.32)	43	64.15 (46.45)	18	43.69 (37.78)	−0.55	0.82	.317	0.47	−0.13; 1.07
Nocturnal awakenings (NA), no.	56	2.55 (1.61)	27	1.68 (1.74)	43	2.54 (1.56)	18	2.09 (1.39)	−0.05	0.03	.010¶	0.25	−0.35; 0.85
Sleep efficiency (SE) (%)#	56	72.36 (13.32)	27	85.82 (12.83)	43	69.41 (13.09)	18	78.31 (12.09)	0.26	0.28	.166	0.60	−0.01; 1.21
Sleep quality (SQ)⁎⁎	56	5.08 (1.09)	27	7.30 (1.88)	43	4.94 (1.45)	18	5.66 (1.75)	0.09	0.04	.005	0.90	0.28; 1.53

Note. ISI = Insomnia Severity Index (higher score = more severe insomnia) (2), FACIT-F = Functional Assessment of Chronic Illness Therapy for Fatigue Scale (higher scores = less fatigue) (3). Means of raw scores at the pre and post timepoint with standard deviations in parentheses.

^⁎Mixed linear models (MLMs), two-sided significance testing. For sleep diary outcomes (time in bed, total sleep time, sleep onset latency, wake after sleep onset, no. of awakenings, sleep efficiency, sleep quality) MLM results are based on 1000 bootstrap samples.

^†Cohen’s d based on absolute between-group differences at post-intervention.

^‡MLM specified without random slope as the outcome was only assessed twice.

^§MLM specified without random slope due to non-convergence with random slope.

^¶Result did not remain significant after controlling for the familywise error rate using Holm’s method (4).

^#Sleep efficiency (total sleep time/time in bed × 100) (5).

^⁎⁎Diary assessed sleep quality (11-point scale, higher scores = better sleep quality).

Fig. 2.

Change in insomnia severity from pre- (baseline) to post by group.

Note. ISI = Insomnia Severity Index. Higher ISI-score means more severe insomnia (2). Error bars indicate 95% confidence intervals (Intervention (Hvil®): Red, waitlist control: blue).

None of the potential demographic moderators of the effect on insomnia severity (age, gender, marital status, educational level, baseline ISI) reached statistical significance.

3.3. Attrition and adherence

3.3.1. Study dropout

Fifty-four participants in the intervention group and 55 participants in the control group completed the posttreatment questionnaire, resulting in study dropout rates of 35% and 33%, respectively. Neither group (B = −0.09, SE = 0.33, p = .785) nor baseline ISI score (B = 0.02, SE = 0.04, p = .692) predicted study dropout status at post-intervention. Of the investigated demographic variables marital status significantly predicted study dropout, with participants not in a relationship being three times more likely to drop out compared to participants in a relationship (B = 1.16, SE = 0.56, p = .038, odds ratio (OR) = 3.18, 95% CI 1.07 to 9.47). Furthermore, participants with a short education were more likely to drop out compared to participants with a medium-long education (B = 2.52, SE = 0.72, p ≤0.001, OR = 12.38, 95% CI 3.00 to 51.16) and long education (B = 1.82, SE = 0.74, p = .013, OR = 6.19, 95% CI 1.46 to 26.25). None of the remaining demographic variables (age, gender, work status) predicted study dropout (data not shown). The results for marital status, educational level, and work status were followed up with Firth's penalized logistic regression due to sparse data in some categories, which demonstrated similar results (data not shown).

3.3.2. Treatment drop-out, usability, and adherence

Forty-five (27%) participants completed sleep diaries in Hvil® at week 9 (intervention n = 27, control n = 18). On average, intervention participants commenced three of the five treatment components in Hvil® (M = 2.9, SD = 2.3). Forty-three (52%) out of the 83 participants in the intervention began all five treatment components in Hvil® and were considered treatment completers. The remaining 40 participants in the intervention, who did not start all five treatment components, corresponded to a treatment dropout rate of 48%. The mean usability rating of Hvil® as assessed by MAUQ across intervention participants was 5.4 (SD = 1.4, range = 1.9–7.0). On average, item 14, “The app improved my access to healthcare services,” had the lowest score across participants, with a mean of 2.9 (SD = 2.4). Neither perceived usability (MAUQ) (B = −0.09, SE = 0.24, p = .693) nor baseline ISI score (B = −0.02, SE = 0.06, p = .765) predicted treatment dropout. None of the investigated demographic variables (age, gender, marital status, educational level, work status) significantly predicted treatment dropout (data not shown). The results for MAUQ, marital status, educational level, and work status were followed up with Firth's penalized logistic regression due to sparse data in some categories, which demonstrated similar results (data not shown).

Number of completed activities in Hvil® significantly moderated the effect of the intervention over time within the intervention group, with more completed activities being associated with larger reductions in insomnia severity (ISI) over time (F(1, 48.74) = 13.18, B = -0.01, SE = 0.002, p < .001). This was also the case for commenced treatment components, with participants who started more components reporting larger reductions in ISI scores over time (F(1, 71.10) = 8.33, B = −0.11, SE = 0.04, p = .005).

Adverse effects, such as increased daytime sleepiness, were reported by two participants. On average, participants in the intervention group reported having used sleep medication on 9.8% of days (SD = 25.2), while control participants reported using sleep medication on 9.3% of days (SD = 23.4).

4. Discussion

In this two-arm randomized controlled trial, we evaluated the preliminary efficacy of Hvil®, a fully automated eCBT-I program, in reducing insomnia severity, improving sleep diary metrics, and alleviating fatigue, relative to waitlist controls. Compared to waitlist controls, intervention participants significantly reduced their insomnia severity, corresponding to a medium effect size (d = 0.66) at post-intervention. While this effect is somewhat smaller than the pooled effect of 0.87 previously reported for fully automated eCBT-I (Takano et al., 2025), it is still well above the suggested clinical significance threshold for ISI of 0.5 standardized mean difference (Edinger et al., 2021b). Our preliminary findings thus extend previous research demonstrating the efficacy of fully automated eCBT-Is.

A statistically significant, small effect of Hvil® was also found for the secondary outcome of fatigue, favoring the intervention group (d = 0.31). This effect was slightly smaller than the pooled effect of non-supported self-help internet CBT-I reported by a network meta-analysis (d = 0.49) (Benz et al., 2020).

With respect to the secondary sleep diary outcomes, we found large and statistically significant effects of the intervention on TIB and SQ (d = 1.06 and 0.90, respectively), and a small effect on NA (d = 0.25). This indicates that participants in the intervention group reduced their time in bed in line with the sleep restriction component and experienced improved SQ compared to participants in the control group. When adjusting for multiple comparisons, the effect on NA, however, failed to reach statistical significance. While the effects on the sleep diary outcomes of WASO, SOL, and SE also failed to reach statistical significance, the effect sizes ranged from small (d = 0.47) to medium (d = 0.60), which are comparable to the pooled effects of eCBT-I on sleep diary outcomes reported in previous meta-analyses (Hwang et al., 2025; Soh et al., 2020; Zachariae et al., 2016). Considering the magnitude of the effect size and the statistically significant effect found for TIB, on which SE is dependent, the non-significant finding for SE in our study could seem surprising. While the mean SE in the intervention group increased by 14.45 percentage points from baseline to post-intervention, the mean SE also increased in the control group by 8.90 percentage points from baseline to post-intervention. We have no ready explanation, but it could be that having access to a sleep diary in the control group had a beneficial effect on their bedtime patterns. Another issue to consider is the reduced sleep diary completion over the course of the study.

In the intervention and control groups, 35% and 33% of participants, respectively, did not complete the posttreatment questionnaire, resulting in an overall study dropout rate of 34%. These estimates are somewhat higher than the pooled dropout rates of 21% for intervention arms and 13% for control arms reported for previous studies on fully automated eCBT (Takano et al., 2025). This difference could be due to varying degrees of support received by participants between studies. In the present study, all study procedures were fully automated, and participants had no contact with the research team at any time point before, during, or after the intervention, except when technical assistance was required. Neither group nor insomnia severity at baseline significantly predicted dropout status at posttreatment. Among the investigated demographic predictors, marital and educational level significantly predicted study dropout, with participants not in a relationship being three times more likely to drop out than participants in a relationship, and participants with a short education being 12 times more likely to drop out than participants with a medium-long education and six times more likely to drop out than participants with a long education. It should be noted that confidence intervals were wide, especially for educational level, indicating reduced precision of the estimate.

The treatment completion rate in the intervention group was 52%, which is only slightly lower than the pooled completion rates of 56–58% reported in previous studies on fully automated eCBT-I (Hwang et al., 2025; Takano et al., 2025). Neither baseline insomnia severity, perceived usability, nor any of the investigated demographic variables predicted treatment dropout. In contrast, both the number of completed activities in Hvil® and the number of initiated treatment components predicted insomnia severity over time, with more completed activities and initiated components being associated with larger reductions in insomnia severity. This highlights that not only treatment completion as such but also treatment adherence in the form of engagement with the treatment program may influence the effects of fully automated eCBT-I. Together with the relatively low completion rates in the present and previous studies on fully automated eCBT-I, this underscores the importance of addressing factors related to retention within the field. However, as we did not assess the reasons for discontinuation of treatment in the present study, we cannot know whether some participants may have discontinued treatment due to improvement of insomnia symptoms.

A potential challenge in completing the treatment of fully automated CBT-I could be the demands and complexity of the intervention program. Although multicomponent interventions such as CBT-I provide a comprehensive treatment framework, which may explain their efficacy, the time demands and complexity of such treatments may also pose challenges to engagement and adherence, particularly in unguided formats. A scoping review found that a lack of time and content perceived as repetitive, lengthy, complicated, or boring are reported as barriers to the continued engagement with mobile health apps across studies (Giebel et al., 2023). Furthermore, a recent meta-analysis demonstrated that of the five CBT-I components applied in this study, only sleep restriction, stimulus control, and cognitive therapy contributed to remission, and only sleep restriction and stimulus control contributed to improved self-reported sleep continuity in the form of improved SE, SOL, and WASO (Furukawa et al., 2024). Sleep hygiene was non-essential, and relaxation seemed potentially harmful to remission (Furukawa et al., 2024). To streamline the intervention, increase cost-effectiveness, and reduce patient burden, there is a need to identify the active, inactive, and possibly harmful components of CBT—I.

The preliminary evaluation indicates that Hvil® is potentially efficacious in reducing insomnia severity when tested in an ecologically valid setting using a fully automated treatment format without any synchronous or asynchronous contact with a therapist. Nevertheless, some limitations need to be taken into consideration. First, the relatively high study and treatment dropout rates are a limitation. While study dropout was predicted by marital status and educational level, we were unable to identify any other characteristics of participants who dropped out of the treatment. To minimize the effect of missing data, we used MLMs, including the full intention-to-treat sample. Nonetheless, the relatively high drop-out rates may have affected the results and their generalizability. We also had a high rate of missingness on some of the baseline and demographic measures distributed via the Hvil® application, as a substantial proportion of the participants failed to complete all prompted activities in the app. Second, while previous studies of fully automated eCBT-I have demonstrated sustained effects on insomnia over time (Hwang et al., 2025), we did not include a follow-up assessment in our preliminary RCT, and thus cannot yet determine whether effects are sustained over time. Third, we did not include any objective sleep measures. While objective measures, e.g., actigraphy-based, may provide important information on various sleep characteristics, insomnia itself is defined by self-reported symptoms, and not by a specific sleep quantity (World Health Organization, 2020). Also, objective sleep measures do not necessarily correlate well with self-reported insomnia (Rosa and Bonnet, 2000). Fourth, as we had to omit the outcomes of cognitions about sleep (DBAS-16) and psychological distress (DASS-21) from the analyses due to a technical error in the distribution of these scales in the app, the current study provides limited information on the effect of Hvil® on non-sleep outcomes. Lastly, it remains unclear which specific treatment components of Hvil® contributed to the observed effect, and whether any of the components may have had a negative impact on the treatment outcome.

Thus, to maximize treatment outcome and reduce patient burden, future studies on fully automated eCBT-I should focus on enhancing adherence and retention, on investigating longer-term effects, and on clarifying the relative efficacy of the active treatment components in CBT—I.

5. Conclusion

In summary, fully automated eCBT-I delivered via the Hvil® smartphone-based application appears to be an efficacious treatment option for people with insomnia in the general population. It demonstrated a statistically significant effect on insomnia severity, corresponding to a medium and clinically relevant effect, although the effect was somewhat smaller than the reported pooled effects based on previous studies of fully automated eCBT-I. Statistically significant effects were also found for the secondary outcomes of fatigue, TIB, NA, and SQ, with effect sizes ranging from small to large at post-intervention. The remaining secondary effects did not reach statistical significance but were generally of similar magnitude to those found in previous studies. Although study dropout was somewhat higher than in previous studies and treatment dropout was relatively high but comparable to previous studies, the limited availability of CBT-I at a scale that meets population demand makes a fully automated treatment format an affordable and accessible option. The findings underscore the need for future studies of fully automated eCBT-I to focus on strategies to enhance study and treatment retention, as well as on optimizing the intervention to improve its overall efficacy.

Declaration of Generative AI and AI-assisted technologies in the writing process

During the preparation of this work SMK used ChatGPT to assist with preliminary SPSS syntaxes and R scripts for data preparation and analysis. After using this tool, the author reviewed and edited the content as needed and takes full responsibility for the content of the published article.

Funding

This work was supported by TrygFonden [grant number 12741].

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Robert Zachariae reports financial support was provided by TrygFonden. The authors have been involved in developing the intervention program, Hvil®, but they hold no financial interest in it. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.