Introduction
Sleep problems, particularly insomnia, are highly prevalent among adults with autism spectrum disorder (autism from hereon1), affecting up to 79%–89% [2–9]. Autism occurs in approximately 1% of the global population [10] and is defined by distinctive patterns in social interaction, communication, sensory processing, and focused interests, which can be both strengths and sources of challenge [11, 12]. An estimated 60% of autistic individuals is affected by insomnia symptoms specifically [4], implying difficulties falling asleep, staying asleep, and waking up early without being able to fall back asleep and causing significant daytime impairment.
Sleeping problems such as insomnia reduce quality of life [13, 14], and have been associated with depression [15], suicidality [16], more pronounced characteristics of autism, and poorer psychosocial, executive, and physical functioning [17]. Considering the high prevalence of sleep problems and the substantial impact on wellbeing, it is not surprising that sleep is ranked among the top priorities for research and practice identified by the autistic community [18].
Autism and insomnia may be linked in reinforcing, bidirectional feedback loops, where sleep problems and autistic characteristics exacerbate one another [17, 19], although literature is inconclusive [20]. The widely used 3P Behavioral model by Spielman, Caruso and Glovinsky [21] posits that insomnia arises from an interaction of predisposing (e.g. biological or psychological vulnerabilities), precipitating (e.g. life events, stress), and perpetuating factors (e.g. maladaptive sleep habits or cognitions). For autism, predisposing factors may include sensory processing differences [22, 23], increased anxiety and stress [24–26], female sex at birth, co-occurring attention deficit hyperactivity disorder (ADHD) or other mental health conditions, and physical health problems (e.g. gastrointestinal problems) [27–29].
Precipitating factors may include the more frequent occurrence of stressful life events in autistic individuals compared to non-autistic individuals [25], and disruptions of routines through intolerance of uncertainty and preference for predictability [30, 31]. Perpetuating factors may include hyperarousal [32, 33], and lack of daily structure (e.g. through unemployment, common in autism) [34]. Overall, most predisposing factors for insomnia symptoms in autism appear similar to those in non-autistic populations.
In the general population, The European guidelines on insomnia recommend cognitive behavioral therapy for insomnia (CBT-I) as the first-line treatment [35]. CBT-I is an umbrella term for structured, nonpharmacological treatment employing educational, cognitive and behavioral strategies to improve insomnia, for instance including cognitive restructuring, sleep restriction and stimulus control [36]. A comprehensive body of research has demonstrated the benefits of CBT-I, with large effect sizes and long-term effectiveness [37–39].
CBT-I is also effective in reducing insomnia severity of patients with co-occurring mental disorders, including depression, post-traumatic stress disorder (PTSD), and alcohol dependency, and is shown to simultaneously improve the symptom severity of these conditions [40]. These results underscore the robustness of CBT-I’s effectiveness in improving insomnia, suggesting this may extend to autistic individuals.
However, the effectiveness of CBT-I for autistic adults remains unclear to date. Psychological interventions for insomnia in autism, including CBT-I, have sparsely been part of scientific investigation, with most existing studies focusing on youth populations [41–43]. These interventions have shown promising effects on sleep outcomes and daytime behaviors in autistic youth [41, 44].
A recent meta-analysis (11 studies, 850 participants) found a large effect of behavioral interventions on the subjective Children’s Sleep Habits Questionnaire (d = 1.01), and a small but significant improvement of objective measures total sleep time (TST) (d = 0.31), reduction of sleep onset latency (d = 0.35) [44]. To date, no randomized controlled trials have evaluated CBT-I specifically for autistic adults [42, 43]. A small pilot study (n = 10) into the effectiveness of a 5-week transdiagnostic group treatment of adults with autism and/or ADHD focusing on sleep education, scheduling, hygiene, and relaxation yielded improvement of sleep disturbances [45], but this intervention did not include sleep restriction nor cognitive restructuring.
Another pilot study (n = 8) in adults indicated that a group-based insomnia intervention using Acceptance and Commitment Therapy, a third-wave therapy, was effective in reducing both insomnia and anxiety symptomatology [46].
Generally, given the limited availability of resources to provide such psychological interventions, pharmacological interventions are frequently prescribed for insomnia [47]. However, many licensed sleep medications offer short-term effectiveness but are poorly tolerated and risk negative side effects and addiction. In autism, melatonin is among the most used agents to address sleep difficulties and has shown benefits for some sleep outcomes, including TST, sleep latency and sleep efficiency [2].
Other pharmacological interventions are also frequently prescribed to autistic individuals for sleep problems, despite a lack of research on their effectiveness in this population [48, 49], and despite autistic adults’ preference for non-medicament treatments such as education and advice [49].
Like in non-autistic populations, only a minority of autistic adults with insomnia seek help [49]. Based on their experiences, they have reported limited faith in existing treatments and emphasize the need for participatory research co-creating interventions “sensitive to autistic experiences” [50]. Co-creation may be particularly important given that differences in sensory and information processing, communication styles, and preferences for predictability and routine can influence how interventions are experienced.
For example, core CBT-I components such as sleep restriction may be challenging because they require changes to established routines [51]. Core characteristics of autism may further pose a barrier in (accessing) traditional healthcare. For example, difficulties with eye contact, decoding body language and simultaneously processing different perceptual modalities tend to make face-to-face contact with healthcare professionals strenuous [52].
Together with the limited service availability, these findings highlight the need for more scalable, non-pharmacological sleep interventions, developed in collaboration with those they are intended to serve.
Guided, internet-delivered CBT-I (ICBT-I) may offer a scalable alternative to face-to-face insomnia treatment, requiring fewer specialized resources while reaching more people, without proportionally increasing clinician time and costs [53–56]. ICBT-I has general advantages that may be particularly preferable for autistic individuals: information can be processed at one’s own pace [57], it facilitates autonomy, and it’s predictable, structured format may reduce uncertainty and cognitive load [58, 59]. It may also alleviate some barriers and (sensory) difficulties experienced within face-to-face treatment [60–62].
Being able to receive treatment from home, and freely choose when to work on the intervention, is perceived as pleasant by autistic individuals [60]. Although not appropriate for everyone (e.g. those in high psychological distress or with intellectual disability) [63, 64], digital interventions have shown promise for supporting autistic individuals with social difficulties or other developmental targets [65, 66]. ICBT-I has been found to be feasible and effective in reducing insomnia in adolescents with autism [67].
To date, however, ICBT-I has not been studied in autistic adults.
In this randomized controlled trial, we adapted an existing, internet-delivered, CBT-I intervention (i-Sleep) in collaboration with autistic adults with insomnia. We evaluated the (long-term) effectiveness of “i-Sleep Autism” in guided self-help format compared to waitlist control with psychoeducation and sleep hygiene. Participants following the intervention were supported through asynchronous written feedback after each completed module.
Our primary outcome was insomnia severity. Secondary outcomes included pre-sleep arousal, anxiety & depression, daily functioning and quality of life. At post-test, we expected a significant moderate effect on our primary outcome insomnia severity (d = 0.5) and positive effects on mental health: a decrease in anxiety, depression, and improvements in daily functioning and quality of life, as well as sustained effectiveness for 6 months.
Materials and methods
Study design
We conducted a two-arm, parallel, superiority randomized controlled trial to evaluate the effectiveness of a guided internet-delivered CBT-I intervention compared to a waitlist control group receiving online psychoeducation (PE) about sleep and information about sleep hygiene. Participants were randomized 1:1. After 6 weeks (end of intervention period), control participants were offered access to i-Sleep Autism.
The study was approved by the Medical Ethical Committee of Amsterdam University Medical Centre (METC Amsterdam UMC) and registered in the Overview of Medical Research in the Netherlands (OMON) registry (NL-OMON56692). The protocol is available elsewhere [68].
Eligibility criteria
Inclusion criteria for this study were: an ISI score of ≥10, as this cut-off has been deemed optimal for detecting insomnia cases [69], a formal autism diagnosis as established by an authorized professional, a minimum age of 18 years, and sufficient Dutch reading and writing abilities. Participants required access to a device (preferably laptop/tablet) with internet connectivity. Exclusion criteria included working in night shifts (defined as work between 2 a.m. and 6 a.m.), and current or planned pregnancy or breast feeding.
Other co-occurring psychological or somatic conditions were not considered exclusion criteria, as these are highly prevalent among autistic individuals [70, 71], and CBT-I has demonstrated potential to improve co-occurring mental health symptoms [72, 73]. The use of (sleep) medication and cannabis was allowed and monitored during the study to enhance the external validity of the findings.
Recruitment
Participants were recruited from the Netherlands Autism Register (NAR), a community-based longitudinal cohort of over 3000 autistic individuals in the Netherlands. The NAR collects annual online questionnaires to gain insight into the lives of autistic individuals in the Netherlands and inform autism knowledge, practice, and policy. Upon registration for the NAR, participants report details of their autism diagnosis, including when and where it was established, which professional(s) were involved, and the diagnostic criteria used (DSM-IV-TR or DSM-5).
Random batches of 50–150 adults were invited between May 2024 and August 2024 until the target sample of 160 participants was reached; in total, 1487 autistic adults were invited (Figure 1). Invitation emails included study information and a link to digital informed consent (Qualtrics). After completing this form, potential participants proceeded to an online screener.
Eligible participants were automatically directed to the full baseline assessment. Ineligible participants were informed accordingly and provided with alternative resources for their sleep problems.
Randomization, blinding, and treatment allocation
Immediately after finishing the baseline assessment, a built-in Qualtrics randomizer assigned participants to one of the two conditions. Randomization was implemented using the Qualtrics “Evenly present elements” function, which dynamically adjusts assignment to either condition based on current group sizes to minimize imbalance in group size. No stratification was applied.
Next, after filling out their e-mail address and choosing a preferred (anonymous) username for the intervention platform, participants were blindly allocated to the assigned condition by use of automated e-mailing schemes in Qualtrics. Thus, randomization and allocation procedures were fully blinded and automated. Due to the nature of the intervention, blinding of participants and coaches was not possible.
The researcher responsible for the analysis (KS) was blinded for the study condition of participants.
Procedures
Participant flow is illustrated in Figure 1. Within 24 hours after T0, participants randomized to the intervention group received instructions on how to activate their personal account at Minddistrict, an e-health platform used by health care organizations in the Netherlands to deliver digital therapy to their clients (https://www.minddistrict.com/). The intervention group received the guided i-Sleep Autism intervention, while the control group received online psychoeducation and sleep hygiene.
At 3 weeks (T1), both groups completed a brief questionnaire assessing the primary outcome and adverse events. At post-test (T2; 6 weeks), all baseline outcomes and adverse events were reassessed, with additional questions on intervention use and evaluation for the intervention group. After T2, the control group was offered i-Sleep Autism.
At 6 months post-baseline (T3), follow-up assessments on all outcomes were completed by the intervention group only, as participants in the comparison group were offered access to the intervention after the post-treatment assessment (T2). Consequently, maintaining a non-intervention comparison group at six-month follow-up was not possible.
Intervention
Intervention condition
Intervention set up. i-Sleep Autism is a guided, digital self-help intervention based on CBT-I. The intervention has 5 modules and addresses: (1) psychoeducation on normal sleep, sleep disorders, and sleep hygiene (factors and behaviors influencing sleep), (2) sleep scheduling (using fixed bedtimes) and restriction (limiting time in bed to increase sleep pressure) and stimulus control (strengthening the bed's association with sleep), (3) techniques to manage rumination and promote relaxation (progressive muscle relaxation, breathing exercises, meditation), (4) cognitive restructuring (changing common maladaptive thoughts about sleep), and (5) summary and approaches for relapse prevention.
Participants were advised to complete one module per week, resulting in a 5-week intervention. Each module took approximately 30–60 minutes and encompassed informative texts, audiovisual components (e.g. case examples, summary videos), and exercises. Participants were encouraged to integrate exercises into their daily routines.
A brief daily online sleep diary accompanied the intervention. The program was accessed via the Minddistrict web platform, with sleep diaries completed through the Minddistrict smartphone application.
Guidance by online coaches. Participants were guided by online coaches (psychology students) who supported goal setting, motivation, and adherence through asynchronous written feedback using Motivational Interviewing (MI) techniques. Coaches received approximately 15 hours of training supervised by a licensed psychologist, including studying a coaching protocol, completing a CBT-I e-learning (LeerSlapen, Kempenhaeghe) [74], and attending a live one-day training with lectures on autism (characteristics, challenges, strengths, communication styles, neurodiversity affirming/inclusive language), CBT-I, and online coaching through written feedback.
Two members of the co-creation panel joined the training to share their experiences, thoughts, and preferences.
Coaches provided the written feedback via the Minddistrict platform after each completed module (within three working days) and received bi-weekly supervision. The coaching period could be extended to 8 weeks if needed, and reminder messages were sent after 2 weeks of inactivity. Coaches reported adverse events, including suicidality, to the research team, with a licensed psychologist available for consultation and referral if required.
Adaptation process. i-Sleep has demonstrated effectiveness across various populations suffering from insomnia [75–78]. In collaboration with autistic adults experiencing insomnia and healthcare professionals, i-Sleep was adapted for people with autism while preserving the core components of CBT-I. A summary of overall adaptations and adaptations per module can be found in Supplementary Materials (Table S2).
The intervention was adapted in multiple iterative cycles of feedback. First, we conducted in-depth interviews with three healthcare professionals with experience with adults with autism and insomnia to identify important areas for adaptation. We then set up a panel of seven autistic individuals with insomnia (six females, one male, aged 24–57, varying educational backgrounds).
For the panel, we scheduled an information session to explain the adaptation goals and planning, and to foster a safe, familiar environment for feedback. Panel members were asked to give feedback on content, language, usability, relevance, and acceptability for each module and overall, and to think of possible autism-friendly solutions. The feedback forms were structured in layout but highly open-ended to allow participants to raise issues freely (Supplementary Files “Feedback forms”).
Written feedback was reviewed and synthesized by the research team and then discussed with the panel in two online focus groups. Based on this first round of feedback, the intervention was adapted and reviewed again by the panel through a second round of written feedback and one additional focus group. At this stage, feedback was minimal, so a final version of the intervention was created.
Panel members could review this version and raise objections, but all participants were satisfied with this final version.
Adaptations were implemented when there was broad agreement among panel members and when proposed changes were compatible with the core principles of CBT-I. When panel members expressed conflicting views, decisions were guided by scientific literature and expert input. The adaptation process was pragmatic and informed by co-creation principles but was not based on a specific intervention adaptation framework.
The following adaptations were made: the text was better matched to the situation of autistic individuals (e.g. psychoeducation on insomnia in autism specifically); extra emphasis was put on elements that were deemed important (e.g. light exposure, day time relaxation, bed time rituals); instructions were clarified and made informative rather than directive; more concrete examples were provided to provide better guidance with exercises; and case examples were replaced with three individuals with autism and insomnia sharing their experiences and practical tips (Table S2).
Control condition
Control participants received a digital folder with online psychoeducation and sleep hygiene, based on the first module of i-Sleep Autism (see “Intervention set up”). The information was therefore also adapted to autism, but did not include any exercises, videos or interactive components. Six weeks after their registration, after the T2 assessment, control participants gained access to the full i-Sleep Autism intervention and an online coach, if desired.
Measures
Primary outcome
Insomnia severity. The Insomnia Severity Index (ISI) was used to measure the primary outcome, insomnia severity [79]. This scale consists of 7 items on a 5-point Likert scale ranging from 0 to 4, resulting in total scores between 0 and 28.
Higher scores indicate greater insomnia severity. Scores are categorized as follows: 0–7 (no clinically significant insomnia), 8–14 (subthreshold insomnia), 15-21 (moderate clinical insomnia), and 22–28 (severe clinical insomnia). The ISI is validated for online applications [80], and showed excellent internal consistency (Cronbach’s α of .90–.91) [69].
The ISI has not been specifically validated for autistic adults, but has been used previously in autism research, e.g. by Hohn, de Veld, Mataw, van Someren and Begeer [23]. A 6-point reduction is considered a clinically meaningful improvement [81]. Cronbach’s alpha was .74 across our assessments.
Secondary outcomes
Pre-sleep arousal. The Pre-Sleep Arousal Scale (PSAS) was used to assess symptoms of cognitive and somatic arousal experienced at bedtime [82]. The 16 items are scored on a 5-point Likert scale ranging from 1 to 5, resulting in total scores between 16 and 80, with higher scores reflecting more arousal.
The first 8 items measure somatic pre-sleep arousal (PSAS-somatic, ranging from 8 to 40), and the last 8 items measure cognitive pre-sleep arousal (PSAS-cognitive, ranging from 8 to 40). Scores are considered clinically relevant when they are ≥14 (somatic) and ≥20 (cognitive) [82]. The PSAS has previously shown good internal reliability in an autistic sample for the cognitive (Cronbach’s α = .89) and somatic (α = .81) subscales [83].
Across our assessments, Cronbach’s alpha was .83 for somatic and .88 for cognitive pre-sleep arousal.
Anxiety and depression. The Hospital Anxiety and Depression Scale (HADS) was used to assess anxiety and depression [84]. Seven items are dedicated to both anxiety and depression subscales, comprising a total of 14 items, scored on a 4-point Likert scale (0–3).
Subscale total scores range from 0 to 21, with higher scores indicating greater symptom severity. Scores are categorized as follows: 0–7 (Normal), 8–10 (Mild), 11–15 (Moderate), and 16–21 (Severe). The HADS is a reliable and valid tool for assessing psychological distress, also in autistic individuals, with excellent convergent validity, acceptable divergent validity, and adequate to good internal consistency (HADS-A α .82-.84; HADS-D α .60-.72) [85].
Cronbach’s alpha was .84 for anxiety and .85 for depression across our measurements.
Work and social adjustment. The Work and Social Adjustment Scale (WSAS) was used to assess the extent to which insomnia interfered with daily functioning. It consists of 5 items assessing functioning in work, social and private leisure activities, home management, and the ability to form and maintain close relationships [86].
Each of the five items is rated on a 9-point Likert scale ranging from 0 to 8. Total scores range from 0 to 40, with higher scores indicating greater impairment. The WSAS has good internal consistency (Cronbach’s alpha of 0.70–0.94), and is sensitive to treatment effects [86, 87].
Cronbach’s alpha was .87 in this study.
Quality of life. The Mental Health Quality of Life Questionnaire (MHQoL) was used to assess quality of life [88]. The seven items encompass self-image, independence, mood, relationships, daily activities, physical health, and future.
They are scored on a 4-point Likert-scale (0–3). Total MHQoL scores range from 0 to 21, with higher scores indicating better quality of life. The MHQoL also includes a general visual analogue scale (VAS) for overall psychological wellbeing.
The VAS ranges from 0 to 10, where higher scores represent higher psychological wellbeing. The MHQoL has demonstrated strong internal consistency (Cronbach’s α = .85) and was designed to identify the effects of mental health interventions [88]. The average Cronbach’s alpha was .77 in this study.
Intervention evaluation
In addition to the standardized questionnaires assessing our participants’ satisfaction and alliance with the intervention and the coaches (described below), we asked whether they preferred the online format over face-to-face or vice versa. We also asked what exercises they put most effort into.
Client satisfaction. The Client Satisfaction Questionnaire adapted to Internet-based interventions (CSQ-I) assessed partici-pant’s satisfaction with the intervention [89]. Although the study protocol initially specified the use of the CSQ-8 [68], the CSQ-I was selected instead, as it was considered more appropriate for evaluating satisfaction with our online intervention.
The eight items are scored on a 4-point Likert-scale (1–4). Total scores range from 8 to 32, with higher scores denoting higher satisfaction. The CSQ-I has good internal consistency, factorial structure, and construct validity as an instrument (McDonald omega of .93-.95) [89].
Cronbach’s alpha was .88 in our study.
Working alliance. The Working Alliance Inventory-Short Revised (WAI-SR) assessed therapeutic alliance between participants and their coaches [90]. The scale includes three key aspects of therapeutic alliance: task agreement, goal agreement, and development of an effective bond between client and therapist.
The 12 items are scored on a 5-point Likert scale (1–5). Total scores range from 12 to 60, subdomain total scores range from 4 to 20, with higher scores indicating higher therapeutic alliance. The WAI-SR exhibits good reliability (alpha > .8) and convergent validity (r > .64) [91].
Internal reliability was excellent in this study (Cronbach’s alpha = .97).
Working—and technical alliance. The Working Alliance and Technical Alliance Short Form (WAI-TECH-SF), a modified version of the WAI-SF, assessed therapeutic alliance between clients and our online intervention [92]. The 12 items are scored on a seven-point Likert scale, ranging from 1 to 7.
Total scores range from 12 to 84, subscales range from 4 to 28. The scale maintains the same structural dimensions as the WAI-SR (therapeutic goals, tasks, and bonds). The scale has excellent reliability (Cronbach’s α = .97) [92], similarly, Cronbach’s alpha was .98 in this study.
Adverse events. Adverse events were assessed at T1 and T2 in both groups, through the following questions: "Have you experienced anything unpleasant since the previous questionnaire 3 weeks ago as a result of your sleep problem or as a result of your participation in the study? For example: suicidality, fall accidents, traffic accidents or other incidents due to fatigue and drowsiness?" (No/Yes, namely..) and “Was medical or psychological assistance needed in relation to this incident?” (No/Yes, namely..).
Descriptive variables
Demographic variables, including age, sex, education level, relationship status, and living situation, and co-occurring conditions were obtained from the NAR database and the annual survey conducted prior to data collection for this study (spring 2024). All NAR participants provided informed consent allowing their data to be used for research purposes.
Given the relevance of these variables to sleep outcomes, baseline data on alcohol and drug (ab)use were collected using the Alcohol Use Disorders Identification Test (AUDIT-10) and the Drug Abuse Screening Test (DAST) [93, 94]. Additionally, information on the type and frequency of healthcare use and sleep medication usage in the past 2 weeks (yes/no) was gathered for both study arms throughout the trial.
Power and sample size calculation
Compared to the large effect sizes observed in meta-analyses establishing the effect sizes of CBT-I interventions [38, 95], we powered conservatively for a medium effect size of d = 0.50 on the primary outcome insomnia severity at the post-test measurement. To ensure 80% statistical power with a 2-sided α-value of 0.05, we estimated that a total number of 128 participants (64 per group) was required. Anticipating a 20% risk of dropout [38], we aimed to recruit 160 participants (80 per group).
Statistical analyses
All analyses were performed using SPSS version 27 and R version 4.4.2, with a two-tailed significance level set at α = 0.05. To account for the hierarchical structure of the data (repeated measures nested within participants), Linear Mixed Models (LMMs) were applied throughout.
Model specification
For each outcome, a null model with a random intercept for subject was fitted to assess clustering. For insomnia severity, the ICC was 0.54, indicating substantial between-subject variance; the resulting design effect was 1.84, yielding an effective sample size of n = 226.
Random intercept and slope models and covariance structures were compared using AIC and BIC. An AR(1) heterogeneous structure provided the best fit for insomnia severity, while a standard AR(1) structure was preferred for secondary outcomes. Due to convergence issues, final models included a random intercept for subjects only.
Primary and secondary outcomes
We tested our first hypothesis—whether the intervention led to a greater reduction in insomnia severity (ISI) compared to the control group between baseline and post-test (T0–T2)—using intention-to-treat (ITT) analyses. All participants were analyzed as originally assigned, regardless of adherence. To assess robustness, per-protocol (PP) analyses were also conducted including only intervention participants who completed modules 1 through 4 (considered the core elements, as module 5 provides a summary and focuses on relapse prevention).
In the final two-level LMM (time nested within participants), fixed effects included group (0 = intervention, 1 = control), time (1 = post-test, 2 = mid-intervention, 3 = baseline), and their interaction (group × time). A random intercept for subjects was included. Models were estimated using Restricted Maximum Likelihood (REML).
Baseline-adjusted Cohen’s d effect sizes were calculated and interpreted as small (d < 0.20), medium (d = 0.50), and large (d > 0.80). Chi-square tests compared remission rates (ISI ≤ 9 vs. ≥ 10) and clinically meaningful change (improvement: ≥6-point decrease; deterioration: ≥6-point increase; no change: −5 to +5). Odds ratios were computed for improvement and remission.
The same LMM approach was used to test effects on secondary outcomes: pre-sleep arousal (PSAS), anxiety/depression symptoms (HADS), work and social adjustment (WSAS), and quality of life (MHQoL). Descriptive statistics of the client satisfaction with the intervention (CSQ-I), working alliance with the coach (WAI-SR) and the intervention (WAI-TECH-SF) were obtained to evaluate the intervention.
Moderation and predictors
Potential moderators of treatment effect on insomnia severity were explored using interaction terms in the LMM. Moderator variables included baseline ISI, insomnia duration (years), age, sex, and presence of a co-occurring diagnosis. Continuous moderators were grand-mean centered before inclusion.
Significant moderators were investigated using simple slope analysis. Additionally, we assessed the bivariate correlation between therapeutic alliance (WAI-SR) and ISI change scores.
Long-term follow-up
To explore the long-term robustness of the results, we conducted within-group Repeated Measures ANOVAs for the intervention group only, contrasting post-test (T2) and six-month follow-up (T3).
Results
Sample descriptives
A total of 163 autistic adults participated in this study. The average age was 48.5 years, 59.5% of participants was female, 60.7% highly educated and 55.4% self-reported one or more co-occurring conditions next to autism and insomnia, e.g. ADHD and PTSD (Table 1).
Sleep medication use in the past 2 weeks at post-test was similar to baseline (Table S1). There were no significant baseline differences between the groups on any of the outcome variables.
Intervention adherence
Of all intervention participants, 50% completed the full intervention (5 modules), 81.7% completed one or more, 68.3% completed two or more, 58.5% completed three or more, 56.1% completed 4 modules (PP analyses), and 18.3% never started (Table S3, Supplementary Materials). Post-hoc, exploratory comparisons (t-tests and chi-square tests) of age, sex, co-occurring conditions, and baseline ISI scores between intervention completers and non-completers showed no significant differences. Adherence to the recommended schedule was generally good for participants and coaches.
Those participants who completed all five modules did so in 40.8 days on average, approximately six days longer than the intended 5-week program. Coaches provided feedback within 3.7 working days on average, slightly more than the intended three working days.
Intervention effects on primary and secondary outcomes from baseline to post-test
Compared to control, the i-Sleep Autism intervention reduced our primary outcome insomnia severity at 3 weeks with medium effect size (T1; d = 0.49; SE of d = 0.18; 95% CI of d [0.14, 0.84]; B = -1.33; 95% CI of B [-2.46, -0.21]) and large effect at post-test (T2; d = 1.02; SE of d = 0.18; 95% CI of d [0.65, 1.40]; B = -3.97; 95% CI of B [-5.52, -2.42]) (Tables 2 and 3; Figure 2). For secondary outcomes, the intervention group showed greater reductions in total, cognitive, and somatic pre-sleep arousal and higher quality of life at post-test compared to controls (Tables 2 and 3). No significant effects were found for anxiety, depression, or daily functioning.
All these analyses were performed on ITT principle.
Sensitivity analyses
Per-Protocol analyses comparing those who completed four modules (n = 46) with controls showed larger effects on insomnia severity at T1 (d = 0.5, 95% CI [0.14, 0.86]) and T2 (d = 1.39, 95% CI [0.99, 1.78]), PSAS total (d = 1.1, 95% CI [0.72, 1.48]), PSAS cognitive (d = 1.05, 95% CI [0.67, 1.43]), PSAS somatic (d = 0.86, 95% CI [0.48, 1.23]), and quality of life (d = 0.56, 95% CI [0.19, 0.92]). Depression was significantly reduced with medium effect size in PP but not ITT analyses (d = 0.52, 95% CI [0.16, 0.89]). Additional post hoc sensitivity analyses showed that neither changes in sleep medication use from baseline to post-test nor the frequency of sleep medication use at post-test were significant predictors of post-test ISI scores, and inclusion of these variables did not alter the main treatment effect.
Clinically meaningful change and remission for insomnia severity at post-test
No participants showed clinical deterioration (Table 4). There was a significant difference between the intervention and control group: intervention participants had 7.0 times higher odds of clinical improvement (SE = 0.45, 95% CI [2.9, 16.8]) and 4.5 times higher odds of remission (SE = 0.44, 95% CI [1.94, 10.66]). Confidence intervals for these findings are wide, and only complete cases were analyzed.
Moderators and predictors of the post-test intervention effect
Exploratory analyses showed that only age moderated the intervention effect at 6 weeks (B = 0.11, SE = 0.05, p = .014), with stronger effects for younger than older participants (Figure S1). The intervention effect was stronger for younger participants (age -1SD: difference = 6.57, 95% CI [4.50, 8.65], p < .001) compared to older participants (age + 1SD: difference = 2.70, 95% CI [0.69, 4.71], p = .009) (Figure S1). Stronger therapeutic alliance was associated with greater reductions in insomnia severity (r = −.338, p=.003; Figure S2).
Adverse events
Adverse events, defined as negative events related to sleep problems or study participation, were assessed mid-intervention (3 weeks) and at post-test (6 weeks). Reported events included falls, minor injuries, and social conflicts due to tiredness. At 3 weeks, 8 participants in the control group (9.9%) and 5 in the intervention group (6.1%) reported an adverse event.
At 6 weeks, this was 5 (6.2%) and 6 (7.3%), respectively. In the intervention group, one participant required medical support for a dislocated ankle, and one participant required psychological support due to autistic meltdowns at both assessments.
Long-term effects within the intervention group
Within-group intervention effects were largely maintained at six-month follow-up, with moderate increases only in total and cognitive pre-sleep arousal, reflecting more pre-sleep arousal with moderate effect sizes (Figure 2, Table 5).
Intervention evaluation
Overall satisfaction with the intervention was moderate to high (M = 21.12, range 9–30; Table S4). Therapeutic alliance with the coach was high (M = 39.1, range 12–60) and comparable across Goal, Task, and Bond subscales. Perceived alliance with the online intervention showed substantial variability (M = 49.68, SD = 20.72).
Most participants (70.7%) preferred the online format over face-to-face, mainly due to flexibility and the ability to work at their own pace (Table S5). Preference for face-to-face delivery was primarily related to missing direct interaction and personal contact with the therapist (Table S6).
Participants reported investing the most effort in the following intervention components (from most to least effort): evaluating and altering sleep hygiene, creating a bedtime routine, keeping fixed bedtimes, and consciously dealing with light exposure. The least effort was reportedly put into cognitive exercises aimed at reducing worrying and rumination, and altering dysfunctional thoughts about sleep.
Discussion
In this parallel, superiority randomized controlled trial, i-Sleep Autism significantly reduced insomnia severity compared to a waitlist control receiving online psychoeducation and sleep hygiene with a large effect size. Nearly 40% of participants achieved clinically meaningful improvement, with sevenfold higher odds of improvement than controls. The intervention also reduced pre-sleep arousal with medium to large effect and improved quality of life with small to medium effect, but showed no significant effects on daily functioning, anxiety, or depression (except in per-protocol analyses for depression, where a medium effect was observed).
Effects were stronger among younger participants and those reporting greater therapeutic alliance, and were largely maintained at 6-month follow-up. Overall, the intervention was positively evaluated.
To our knowledge, this is the first ICBT-I intervention specifically tested in autistic adults [43]. The effectiveness of our intervention on insomnia severity (primary outcome) is comparable to effects observed of guided ICBT-I in the general population (umbrella review: d = 1.05 at post-test) [39]. It must be noted, however, that waitlist control designs can inflate effect sizes [96], and the observed effects may partially reflect non-CBT-I-specific factors such as guidance and structured engagement.
Future studies could compare guided ICBT-I with other active comparators such as in-person (group) CBT-I to evaluate its relative effectiveness in autism. Rates for clinically meaningful reduction of insomnia are also comparable (37.9% clinically meaningful change; 35.4% remission) to effects of guided ICBT-I in the general population [39], although these percentages may decline when higher ISI cut-offs (e.g. ≥8 points) are applied. The average within-group reduction in insomnia severity was fully maintained at 6-month follow up, whereas previous studies report attenuation of treatment effects over time [37].
This suggests that, in accordance with our hypotheses, ICBT-I is effective in reducing insomnia severity in autistic individuals, also 6 months later.
Total, cognitive and somatic pre-sleep arousal significantly decreased at T2, which may have mediated the intervention’s effects on insomnia severity [97]. We only found significant moderate to large effects for depression and not anxiety in our Per-Protocol analyses, whilst recent meta-analyses of the effects of ICBT-I on anxiety and depression found small to moderate effects for these outcomes respectively [98, 99]. Functional impairment related to insomnia also did not improve significantly.
This may reflect a true absence of effects on these outcomes in autistic adults, or a lack of power to detect the relatively small effects observed. For quality of life, CBT-I is generally found to have a moderate positive effect [100]. We primarily observed a deterioration in the control group, potentially reflecting either protective effects of the intervention or challenges faced on a waitlist or both [101].
Among participants who evaluated the intervention, overall satisfaction and therapeutic alliance with the coach were moderate to high, and most participants preferred the online format over face-to-face delivery because of its flexibility and self-paced nature, suggesting generally favorable acceptability, although the perspectives of dropouts are not integrated into this evaluation. Post-test dropout in the intervention group (29.3%) and full treatment adherence (completion of 5 modules) (50%) were both less favorable than rates reported for guided ICBT-I in patients with co-occurring conditions (18% study dropout; 69.2% full treatment adherence) [102] and in previous i-Sleep studies in other populations (68–85.7% adherence) [77, 78, 103], indicating lower feasibility within this population. Most intervention dropouts occurred early, with participants either not starting (18%) or discontinuing after modules 1 or 2 (21%).
Non-initiation may have been related to recruitment during the summer holiday period. Reported reasons for later dropout included limited perceived benefit, high assignment burden, and frustration with platform or sleep diary usability, although many reasons remain unknown. These findings highlight the importance of thorough platform testing and user-friendly design, particularly for autistic adults.
Planned platform improvements include enhancing accessibility through the implementation of a more natural text-to-speech function and improving the sleep diary through automatic calculation and clearer presentation of sleep parameters.
Although no participants reliably deteriorated, 32.9% showed no clinically meaningful improvement. While the digital format offers flexibility, its effectiveness depends on personal accountability, technological skills, and confidence [104]. Consistent with findings in the general population, effects were somewhat stronger among younger participants, possibly reflecting greater digital affinity [105, 106].
For some participants, support may have been insufficient or treatment demands too high. Sleep restriction was experienced as particularly challenging, and some exercises were interpreted very rigidly, potentially increasing stress. More interactive formats (e.g. video or face-to-face guidance) might allow greater flexibility in the interpretation and application of exercises.
While findings of drop-out and non-response are common in internet-delivered interventions, certain autism-related characteristics (e.g. rigidness) may have contributed to engagement difficulties. Further analyses of non-response and dropout are needed to refine the intervention.
The intervention may also have been less suitable for individuals with high psychological distress or lower intellectual functioning [63, 64], given its cognitive and behavioral demands. Suitable alternatives should be explored for these groups. There are also indications that circadian sleep–wake disorders are prevalent in autism [19, 107] next to having a sole insomnia disorder.
Even though our study showed improved sleep, targeting both disorders at the same time might lead to even better outcomes. The transdiagnostic approach of Harvey (Transdiagnostic Intervention for Sleep and Circadian Dysfunction, TranS-C) is a promising candidate for that [108]. Adjunctive approaches such as chronotherapy or light therapy may also be beneficial in such cases [109].
Future studies need to examine if these approaches would be even more effective in this target group than treating insomnia alone.
Future research should disentangle which specific content and delivery components drive change in autistic people with insomnia and which require adaptation. Core behavioral elements such as sleep restriction and stimulus control may have contributed most to symptom improvement [110], despite being challenging for autistic individuals due to their impact on predictability and routines [51]. In contrast, sleep hygiene and psychoeducation were used most frequently by our sample but have relatively weak empirical support [36], and may even feel redundant or frustrating for individuals with long-standing insomnia, as reflected in qualitative feedback received through e-mail during and after the trial (e.g.
“I’ve tried these tips for years”). Cognitive restructuring, although generally effective in targeting dysfunctional sleep beliefs [111], was applied least often in our sample, which may relate to its complexity in an online format and/or lower acceptability of cognitive restructuring among autistic individuals [112]. Future research should address these concerns, and alternatives such as ACT should be investigated for autistic individuals who do not respond well to cognitive restructuring [46].
Despite opportunities for refinement, the large effect size and the absence of deterioration seem to indicate that i-Sleep Autism can be offered safely in a broader non-clinical context (e.g. open online platform). Given the high prevalence of insomnia and sleeping problems in autistic individuals [4, 6–8], and the large impact on their lives [13, 14], this intervention could play an important role in improving these problems. The program could be implemented as stand-alone intervention, waiting-list care or as a component of blended care, although these latter applications require formal evaluation, including cost-effectiveness analyses.
Guided internet-based CBT-I may be a promising alternative to face-to-face CBT-I in view of the limited number of trained therapists [113] and the student coaches’ modest time investment per participant (<30 minutes per module) in this study. Further work should examine which coach characteristics (e.g. training in CBT-I, familiarity with autism, communication style) enhance outcomes, and to what extent these preferences from clients align with objective effectiveness, in order to balance personalization with scalability. Finally, insights from this trial may help refine CBT-I protocols for people with neurodevelopmental conditions and insomnia more broadly [43].
The strengths of our study lie in its robust design, large sample size, and our strong focus on collaboration with autistic individuals throughout the research process, including the tailoring of the i-Sleep Autism intervention and the interpretation and writing of this paper. As we aimed to establish the “real-world” effect of assigning the intervention, the study was pragmatic in nature, i.e. the eligibility criteria were broad, and participants were given substantial flexibility in when and how they engaged with the intervention, with minimal reminders, and coaches were not subject to strict protocol monitoring. However, the generalizability of our findings to the broader autistic group may still be limited, as our participants were mainly female, highly educated, and of Dutch nationality, and recruited through an active longitudinal cohort.
It remains unclear how the intervention would be received in more socioeconomically and culturally diverse settings. Furthermore, not all included measures such as the ISI, WSAS and MHQoL have been validated specifically for autistic people, so they may lack sensitivity for this population. Other sleep disorders were not assessed in this sample, limiting our ability to determine whether comorbid sleep disorders such as sleep apnea were present and may have influenced insomnia severity and treatment response.
Lastly, suicidality was not included as a study outcome, despite the known associations between insomnia, depression, and suicide risk [114], and the elevated prevalence of suicidal ideation and behavior among autistic adults [115]. Future research should examine whether improvements in insomnia are associated with reductions in suicidal ideation and behavior in this population.
Conclusion
All in all, i-Sleep Autism appears to be acceptable, and to effectively and sustainably improve insomnia, considering the large effect size, more than one third experiencing clinically meaningful reduction of insomnia severity, and average appreciation. Given its scalability and relatively low resource requirements, guided ICBT-I represents a promising and accessible treatment option for autistic adults, a population with a high prevalence of insomnia. However, the relatively high study- and intervention drop-out rates suggest that the feasibility of the intervention should be enhanced.
Future efforts should aim to support those that dropped out or did not benefit from the intervention.
Supplementary Material