The study design and protocol (Multimedia Appendix 1 [2,15,25,29,40-75]) were approved by the following supporting agencies: Colorado Multiple Institutional Review Board (COMIRB 17-0920), Rocky Mountain Regional Veterans Affairs Medical Center Research and Development Committee, and the US Army Medical Research and Development Command Human Research Protection Office (HRPO Log Number A-19051.2; ORHO Log Number E03947.2). Additional information on the informed consent process, data privacy and security measures, and compensation are provided below and in Multimedia Appendix 1.

This is a single-blind (participants blinded to treatment allocation) 2-group (SHUTi vs IEW) longitudinal (4 time points) RCT. Assessments were administered at baseline (T1), 9 weeks postintervention (T2), 6 months postintervention (T3), and 1 year postintervention (T4). Data were collected on the web, and participants provided informed consent before study enrollment. The RCT was registered with ClinicalTrials.gov (NCT03366870).

Due to the empirical association between insomnia and suicide risk, randomization was stratified by self-reported suicide attempt history (“yes” or “no”). A statistically generated allocation sequence, using random blocks of 4 and 6, stratified by suicide attempt history, was conducted using SAS PROC PLAN (SAS Institute). Randomization tables were generated by the trial’s statistician and programmed to automatically assign a participant to the next allocation immediately after T1 completion.

OEF, OIF, and OND veterans aged between 18 and 55 years were eligible to participate if they met the following criteria: current Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) [45] insomnia diagnosis; reliable access to the internet; eligibility to receive VHA care; English-speaking; and ability to provide informed consent. Exclusion criteria included the following: being currently enrolled in another intervention research study, the presence of another untreated sleep disorder, currently receiving psychological treatment for insomnia (excluding sleep medications), a change in schedule or dosage of sleep medications in the past 3 months, and the presence of a condition contraindicated with CBT-I. Examples of conditions contraindicated with CBT-I included untreated seizures or seizure disorder, pregnancy or a plan to become pregnant, irregular shift work, and significant cognitive impairment. Veterans with comorbid psychiatric presentations were eligible to participate in the RCT unless they had used any nonalcohol substances (excluding cannabis) more than once within the past 3 months; endorsed a problematic pattern of alcohol use associated with significant impairment or distress (past 3 months); or had ever been diagnosed with bipolar I disorder, schizophrenia, schizoaffective, or a psychotic disorder. Those with comorbid medical conditions were included unless conditions were deemed active, unstable, or degenerative in a manner expected to influence sleep. Research staff received extensive training on conducting semistructured interviews focused on clinical sleep history to inform differential diagnoses (eg, untreated sleep apnea, periodic limb movement disorder, parasomnia, and circadian rhythm disorder) and structured clinical interviews for DSM-5 Research Version [76] modules (eg, insomnia disorder and alcohol use disorder) to determine RCT eligibility; cases were regularly reviewed by the team and principal investigator to ensure accuracy.

Total compensation for completion of all study assessments was US $208, and the participant payment for each study assessment was as follows: US $1 for each preintervention sleep diary (up to US $14 maximum), US $50 for T1 and T2 assessments, US $1 for each postintervention diary (up to US $14 maximum), and US $40 for T3 and T4 assessments.

The primary outcome measure was the Insomnia Severity Index (ISI) [77], one of the most empirically examined and clinically used self-report scales assessing insomnia symptoms. The ISI is comprised of 7 items that assess current (ie, past 2 weeks) insomnia severity. Each item is scored on a 0-4 scale, with total scores ranging from 0-28; each 7-point increment is associated with increasing levels of insomnia severity. The ISI has been shown to have adequate test-retest reliability over 3 months and concurrent validity with sleep diaries and polysomnography [46] and has been validated for web-based use [78]. In the current sample, the ISI demonstrated adequate internal consistency (Cronbach α ranged from .78 to .90 across the 4 time points).

Our secondary outcome measure was the Veterans RAND 36-item Health Survey (VR-36) [79]. The VR-36 assesses perceived general health-related quality of life; respondents indicate the impact of health on physical functioning, role limitations due to physical problems, bodily pain, general health perceptions, vitality, social functioning, role limitations due to emotional problems, and mental health. The VR-36 was formerly called the Short Form Health Survey for Veterans (Veterans SF-36) [80]. The VR-36 has been thoroughly validated and has strong psychometric properties [79,80]. Scores are used to derive 2 summary scales (ie, Physical Component Summary [PCS] and Mental Component Summary [MCS]). Scores range from 0-100, with a mean score of 50 (SD 10). Lower scores indicate poorer health-related quality of life.

Daily sleep diaries based on consensus sleep diary [81] questions were collected for 2 weeks before T1 (preintervention sleep diaries) and immediately following T2 (postintervention sleep diaries). Sleep parameters included sleep onset latency (SOL; time to fall asleep), wake after sleep onset (WASO) + early morning awakenings (EMA; combined time awake after falling asleep and before final awakening), time in bed (TIB), total sleep time (TST), and sleep efficiency (percentage of time asleep while in bed).

A total of 4 psychometrically sound assessments (Multimedia Appendix 2 [47,48,82-85]) were used to characterize baseline clinical characteristics: (1) Adult Suicidal Ideation Questionnaire [47], a 25-item self-report instrument that assesses past-month suicidal ideation frequency; (2) Beck Anxiety Inventory [48], a 21-item self-report rating inventory that measures past-week subjective, somatic, and panic-related symptoms of anxiety; (3) Beck Depression Inventory II [82], a 21-item self-report rating inventory that measures past 2-week characteristic attitudes and symptoms of depression; and (4) Posttraumatic Stress Disorder Checklist for DSM-5 (PCL-5) [83], a 20-item self-report measure that assesses past-month severity of posttraumatic stress disorder symptoms.

Veterans were screened over the phone for eligibility. Of the 844 veterans who expressed interest in the study, a total of 61 were unable to be reached for screening, and 494 were ineligible for participation. The top reasons for ineligibility were identification of another untreated sleep disorder (162/494, 32.8%); not serving in the OEF, OIF, or OND conflict (112/494, 22.7%); multiple exclusion criteria (54/494, 10.9%); and not meeting criteria for insomnia disorder (52/494, 10.5%). After screening, a total of 289 veterans were eligible for the study, and 250 were enrolled.

Most participants (234/250, 93.6%) enrolled were recruited from targeted mailings sent to veterans with diagnosed sleep problems, as identified by medical codes in the VA’s Corporate Data Warehouse. An additional 3.6% (9/250) of enrolled participants were recruited based on their expressed interest in learning more about the study. Participants were enrolled in the RCT between April 2018 and January 2019. The final T4 assessment was completed in June 2020.

After providing informed consent, participants were given instructions on how to complete a web-based, 2-week sleep diary assessment and informed that they had to complete 10 out of 14 sleep diaries to move forward in the study; a total of 6% (15/250) of enrolled veterans did not complete the sleep diary requirement. Baseline assessments (T1) were unlocked once the requirement was fulfilled. A total of 3 participants did not complete T1 assessments, resulting in a total of 231 participants (92.4%) who were randomized.

A total of 84.8% (196/231) of all randomized participants completed at least 1 postintervention assessment, of which 151 participants completed all 3 assessments (77%), another 21 completed 2 assessments (10.7%), and 24 completed 1 assessment (12.2%). A total of 9 participants asked to be withdrawn from the study: 1 during the intervention, 2 at T2, 5 at T3, and 1 at T4. A total of 2 participants randomized to the IEW died during the study (deaths unrelated to research procedures); these deaths occurred between T2 and T3, and T3 and T4. Figure 1 depicts the study enrollment flow.

Participants received reminder calls from study staff to complete initial sleep diaries and the T1 assessment. During the 9-week intervention window, study staff did not initiate any contact with participants, so participants only received automated emails built into the interventions during the intervention window. Participants received reminder calls, automated emails, and reminder postcards to support postintervention (T2-T4) assessment completion.

Participants were compared across demographic variables, ISI, and VR-36 summary scale scores to examine whether there were differences between participants who did not complete a postintervention assessment (n=35) and participants who completed at least 1 postintervention assessment (n=196). The only statistically significant difference observed was that participants who did not complete a postintervention assessment were younger (mean 35.4, SD 6.4 years) than participants who completed at least 1 postintervention assessment (mean 40.0, SD 7.8 years; 2-tailed t₂₂₉=3.33; P=.001).

Of the 35 participants who did not complete a postintervention assessment, a total of 20 randomized to SHUTi demonstrated significantly lower VR-36 MCS scores (mean 31.9, SD 8.2) than the 15 randomized to the IEW (mean 40.1, SD 12.6; 2-tailed t₃₃=2.35; P=.02). No other statistically significant differences were observed between groups.

Because data were collected on the web, steps were taken to verify data quality before conducting analyses, including examining abnormally low survey completion times and surveys with potentially biased or inconsistent responses. No inconsistent, biased, or inaccurate response patterns were identified in the data set. Missing data was minimal across measures and was handled in accordance with manual instructions. Sleep diaries were excluded from analyses if participants provided too few (less than 7) usable sleep diaries during the sleep diary assessment window (T1, n=2; and T2, n=4).

We used any pair power and Holm sequential procedure [86] to control the familywise error rate, considering all 3 primary (ISI: T2-T4) and 6 secondary outcomes (PCS: T2-T4, and MCS: T2-T4). A final sample size of at least 170 completing T2 assessments (85 per group) would provide 80% power to detect an effect size of 0.6 for each outcome. Assuming an attrition rate of 25%, we sought to randomize at least 226 participants to achieve at least 85 per group completing T2 assessments.

The analysis of covariance was used to model the change from T1 to T2, from T1 to T3, and from T1 to T4 for each outcome as a function of group, the baseline value of the respective outcome, and suicide attempt (stratification variable). Generalized η² [87] is reported with all analysis of covariance results, where values of 0.01 are considered small, 0.06 as medium, and 0.14 as large. Fisher exact tests were used for post hoc analyses that examined the clinical significance of ISI changes at each postintervention assessment. All analyses were conducted using SAS (version 9.4; SAS Institute) statistical software [88].

There were no significant differences between groups on demographic or baseline clinical characteristics (Table 1 and Multimedia Appendix 2). At T1, participants on average endorsed moderate insomnia and reported deficits in functioning, with PCS and MCS scores about 1 SD below the normative US population sample but consistent with the VHA normative sample. Table 2 contains the ISI, PCS, and MCS descriptive statistics over time.

SHUTi participants experienced a significantly larger T1 to T2 (–3.47, 95% CI –4.78 to –2.16 points; P<.0001), T1 to T3 (–3.80, 95% CI –5.34 to –2.27 points; P<.0001) and T1 to T4 (–3.42, 95% CI –4.97 to –1.88 points; P<.0001) decreases on the ISI compared with IEW participants. The magnitude of the results was maintained across all time points (generalized η² values of 0.13, 0.12, and 0.10; Table 3).

SHUTi participants experienced a significantly greater T1 to T3 improvement on the VR-36 PCS (3.22, 95% CI 1.07-5.36 points; generalized η²=0.04; P=.004) compared with IEW participants (Table 3). Significance was not observed for the change from T1 to T2 (P=.03) or from T1 to T4 (P=.14). For the VR-36 MCS, SHUTi participants had a significantly greater T1 to T3 (4.33, 95% CI 1.11-7.55 points; generalized η²=0.03; P=.009) and T1 to T4 (4.53, 95% CI 1.39-7.67 points; generalized η²=0.04; P=.005) improvement, but the difference in change from T1 to T2 did not achieve statistical significance (2.58, 95% CI –0.07 to 5.24 points; P=.06; Table 3).

Results for the sleep diary parameters are displayed in Table 4. SHUTi participants experienced a significantly greater T1 to T2 decrease in SOL (P=.0001), WASO+EMA (P=.006), TIB (P<.0001), and TST (P=.04), compared with IEW participants, and a significantly greater improvement in sleep efficiency (P=.02). Specifically, SOL decreased by 16.9 (95% CI –25.6 to –8.3; generalized η²=0.04) more minutes for SHUTi participants, WASO+EMA decreased by 15.3 (95% CI –26.1 to –4.4; generalized η²=0.03) minutes more, TIB decreased by 48 (95% CI –65.4 to –30.0; generalized η²=0.12) minutes more, and TST decreased by 18.6 (95% CI –36.6 to –0.60; generalized η²=0.02) minutes more. Sleep efficiency improved for SHUTi participants 3.74% (95% CI 0.72%-6.75%; generalized η²=0.03) more than for the IEW participants.

A significantly greater proportion of the SHUTi group were deemed treatment responders (reduction of greater than 7 points on the ISI) at all time points compared with IEW participants (31% vs 4% at T2; 43% vs 10% at T3, and 37% vs 11% at T4; all P<.001; Figure 2). Additionally, a greater proportion of the SHUTi group met criteria for insomnia remittance (change from ≥8 at T1 to <8 on the ISI at T2, T3, and T4) at all time points compared with IEW participants (30% vs 4% at T2; 37% vs 7% at T3, and 40% vs 13% at T4; all P<.001; Figure 2).

Despite the empirical support for CBT-I as an efficacious first-line treatment for insomnia and evidence that iCBT-I improves insomnia severity with comparable effect sizes to those found in F2F CBT-I, there is a paucity of RCTs examining iCBT-I in veterans. Data from this RCT confirmed the efficacy of SHUTi for the treatment of insomnia in OEF, OIF, and OND veterans, and this is the first study to demonstrate that improvements in insomnia severity are sustained up to 1 year postintervention. SHUTi participants reported significant changes in insomnia severity across all time points compared with IEW participants. These effects were medium-large and consistent with civilian SHUTi trials, an iCBT-I trial [89] in a military sample, and self-guided iCBT-I interventions in veteran samples [32-34]. SHUTi participants also experienced significant improvements in sleep diary parameters compared with IEW participants. Although SHUTi participants experienced a reduction in total sleep time compared with IEW participants, they spent less time in bed and improved their sleep efficiency, which are consistent with CBT-I desired outcomes.

IEW participants completed pre- and postintervention sleep diaries and had access to high-quality, patient-oriented insomnia materials, constituting a more “active” control compared with waitlist designs. For example, behavioral tracking can facilitate awareness of patterns that foster insight and promote behavior change. Given this, the fact that SHUTi participants significantly outperformed IEW participants in reducing their insomnia severity demonstrates the value of metered, interactive, and tailored iCBT-I content. These findings add to the robust empirical data that supports the superiority of CBT-I interventions over approaches such as sleep hygiene education. This study extends those by demonstrating the efficacy of self-administered iCBT-I in a veteran population.

SHUTi participants were also more likely than IEW participants to report significant improvements in health-related quality of life as assessed by the VR-36 during follow-up, though effect sizes were small. These findings suggest that mental and physical health functioning may require sustained improvements in insomnia to ultimately improve functioning. Furthermore, given the many health factors impacting veterans’ health-related quality of life, it is possible that small effect sizes were due to improved sleep but the continued negative impact of comorbid issues. Our findings were consistent with those observed in a military sample [90] but differed slightly from findings reported in a systematic review on F2F CBT-I and functioning [91], suggesting potential differences in the insomnia-functioning association across populations. Although our findings fall below the change typically indicative of clinical significance at an individual level, they are above the cutoff for population clinical significance [79,92]. Paired with the insomnia severity findings, this suggests that health care implementation of iCBT-I could be an especially promising approach to improving veteran outcomes (eg, systems level).

SHUTi participants were more likely to achieve clinically meaningful changes in insomnia severity compared with IEW participants across all time points, consistent with improvements observed at posttreatment by veterans following use of Insomnia Coach [33]. The proportion of treatment responders (31%) and those who achieved remittance (30%) at posttreatment was greater in our SHUTi sample compared with an implementation-focused study of SHUTi in VA primary care (20% responders, 18.6% remitted) [34]. These differences may have been driven by sample characteristics, as participants in this study were younger (mean age of 39.3, SD 7.8 years) than those in the study of Hermes et al [34] (mean age of 67, SD 16.9 years). Furthermore, participants in this study may have been more motivated to engage in the iCBT-I as enrollment originated from self-interest versus provider referral. Despite SHUTi participants achieving clinically meaningful changes in insomnia, responder and remittance percentages were lower than what had been observed in nonveteran SHUTi trials. IEW participants, however, were also less likely to naturally recover or receive benefit from the IEW when compared with nonveteran IEW participants, suggesting that differences may be attributable to methodology and sample clinical complexity. Given that this study was not designed to specifically examine comorbidity, future studies are necessary to investigate how clinical comorbidity impacts iCBT-I engagement and benefit.

For self-administered approaches to be a viable health care option, individuals must be interested in using technology to receive an intervention and effectively engage. OEF, OIF, and OND veterans’ interest in this study was high, with almost 900 veterans reporting interest in approximately 9 months. Engagement was also high, with 58% of SHUTi participants engaging in the intervention to receive an adequate dose (ie, completion of more than 4 cores) [44]. This finding is notable given there was no clinician or research staff support during the intervention, demonstrating that some veterans can achieve positive outcomes from a self-administered iCBT-I without clinician or coaching support.

Although this trial was adequately powered and included a longitudinal design with a high retention rate in an important clinical population, several limitations are important to note. To mitigate primary outcome confounding, there was no ISI inclusion requirement. As such, about one-third of randomized participants reported insomnia severity below the ISI clinically significant threshold. Although CBT-I has been shown to provide benefit in subthreshold insomnia [93], this has not been widely studied in veteran populations and may have resulted in a floor effect. Most participants in this study identified as male, White, and non-Hispanic. Thus, our results may not generalize to veterans with different identities who face additional barriers to accessing medical care. To ensure safety, participants were ineligible if they had untreated sleep apnea (a commonly cooccurring sleep disorder with insomnia) [30,94]; our findings may not generalize to veterans with insomnia and untreated sleep apnea. During the consent process, participants were told that the purpose of the study was to learn whether participation in computerized insomnia programs may reduce insomnia symptoms and improve functioning in veterans (ie, SHUTi and IEW were not described or compared). Although participants were blinded to their treatment allocation, based on their experiences (eg, access to static IEW information vs access to SHUTi, seeking information on ClinicalTrials.gov), they may have speculated that their assigned condition was better or worse than the other. To better understand how participant perceptions of assigned conditions impact outcomes, asking participants what condition they believe they received would be of benefit to future iCBT-I trials. Given the RCT design (ie, no initiated contact during the intervention), we do not have a reliable way to ascertain the duration or engagement in SHUTi or the IEW. Although this design improves the ecological validity of our findings, it limits our ability to determine whether there were moderating factors associated with trajectories of improvement. Future research examining levels of support will be integral in determining CBT-I treatment options (ie, clinician-delivered CBT-I and clinician-supported and self-administered iCBT-I) [33,95-97]. Although participants must have experienced insomnia symptoms for at least 3 months to meet the inclusion criteria, we did not systematically collect information on insomnia symptom duration, sleep medication use, or medical comorbidity; this information would have further characterized the sample and will be important to include in future iCBT-I research. Finally, although commonly used as an important indicator, sleep outcomes were based on self-report, with no objective indicators. Future research examining potential differences in subjective versus objective outcomes following SHUTi will be critical to understanding the mechanisms of change.

This study is the first known longitudinal RCT comparing an evidence-based iCBT-I to a nonwaitlist control in a veteran population meeting diagnostic criteria for insomnia disorder. In this RCT, we observed significant reductions in the primary outcome of insomnia severity between groups immediately after the intervention and up to 1 year postintervention. Although the secondary outcomes of improved physical and mental health functioning were not detected immediately after the 9-week intervention, SHUTi was efficacious in improving mental health functioning at both follow-up time points and physical health functioning at 6 months. In exploratory analyses, participants randomized to SHUTi experienced significant improvements across sleep parameters. The results of this RCT support the use of a self-administered iCBT-I in OEF, OIF, and OND veterans who meet criteria for insomnia disorder.