Patient perspectives on facilitators and barriers to equitable engagement with digital CBT-I

Abstract

Study Objectives:

Digital cognitive behavioral therapy for insomnia (dCBT-I) has significant advantages for dissemination and scalability versus in-person CBT-I and is, therefore, well-positioned to be the first line intervention for insomnia. However, only about half of patients remit following dCBT-I. Evidence suggests that treatment engagement is a critical driver of dCBT-I effectiveness, and barriers to engagement disproportionately impact people from under-resourced communities. For dCBT-I to be effective and scalable, we need to identify facilitators and barriers to dCBT-I engagement.

Methods:

Responses from an exit survey about participant experiences with dCBT-I were analyzed using mixed methods. The survey included quantitative measures of treatment engagement and a free-response item, which was coded and analyzed for themes using both inductive and deductive approaches.

Results:

Analyses revealed five themes that were relevant for engagement: 1) digital person-to-person components, 2) type and extent of information, 3) user’s sense of autonomy, 4) app functionality, and 5) importance of tailored content. Facilitators included enjoyment of dCBT-I elements, particularly those that enhanced a sense of connection (e.g., a digital therapist avatar); content presented clearly and at an appropriate pace; and smooth app functionality. Barriers included desire for additional human support, perception that dCBT-I did not account for clinical complexities, and factors that interfered with implementation of key treatment recommendations.

Conclusion:

Many barriers and facilitators are influenced by health literacy and technological literacy. Those with access to health and technological literacy are better equipped to engage with dCBT-I. Recommendations for adaptations and enhancements are discussed.

Introduction

Insomnia is the most prevalent sleep disorder and is also a robust risk factor for medical comorbidities.¹ Mounting evidence suggests that minoritized individuals (e.g., those with lower socioeconomic positions and racially minoritized individuals) are at increased risk of insomnia ² and experience more severe symptoms.³ Thus, minoritization is a fundamental contributor to health disparities.⁴

While insomnia can be effectively treated with Cognitive Behavioral Therapy for Insomnia (CBT-I), there have been significant barriers to its dissemination, including a shortage of providers certified or trained in CBT-I^5,6, geographic distance to providers that are clustered within metropolitan areas⁶, lack of reliable transportation among those with lower income⁷, and incompatibility between work hours and therapist availability⁸. Fortunately, digital health offers significant advantages: CBT-I can now be delivered digitally (digital CBT-I, or dCBT-I), greatly increasing accessibility—anyone with an internet-capable electronic device (e.g., computer, tablet, phone) can now receive dCBT-I as the first-line treatment without encountering the above-mentioned barriers.

Despite its advantages, about half of those who receive dCBT-I do not achieve remission. One major predictor of remission is treatment engagement, comprising completion of treatment sessions and adherence to the treatment strategies. Treatment non-completion is an endemic problem for Internet-based therapies⁹, and meta-analyses indicate that dCBT-I is no exception; the average non-completion rate of dCBT-I is 35.2%, and roughly half of dCBT-I treatment strategies were not implemented.^10,11 Critically, poor engagement interferes with the dCBT-I effectiveness, thus enabling insomnia to persistent. In our previous work, dCBT-I completers (operationalized as completing at least four sessions¹²) were roughly four times more likely to achieve remission than non-completers.¹³ dCBT-I engagement is even lower in marginalized or under-resourced communities. Our prior research found that those who had lower income and/or education were two to three times less likely to complete treatment than those who were more affluent or educated.¹⁴ Thus, treatment engagement is likely a critical component driving dCBT-I remission, and the social and structural disparities in engagement need to be addressed.

For dCBT-I to be effective and scalable, treatment engagement must increase; however, the facilitators and barriers of engagement are not well-understood. The extant literature has established determinants, such as help-seeking behaviors at the outset (e.g., knowledge of CBT-I as a non-pharmalogical option, beliefs that sleep is not important), and clinician-levels barriers (e.g., knowledge of how to make a CBT-I referral, effectiveness “selling” of CBT-I to patients),^15,16 but limited evidence explicates factors more proximal to the treatment that impact engagement and persistence, particularly in the context of dCBT-I. Although low engagement is prevalent in internet-based therapies,¹⁷ adaptations based on user feedback can substantially improve engagement.^18,19 For example, an app-based CBT study was able to double frequency of usage and increase time-on-app by 80% by refining app implementation based on user feedback.¹⁹ Another study implemented cultural adaptations to dCBT-I for Black women and produced higher treatment completion rates (78.2% compared to 64.8% in the non-adapted control) and greater treatment gains (68% greater insomnia reduction).²⁰ If similar enhancements could be translated for dCBT-I, it could significantly improve remission rates, thereby increasing the real-world effectiveness of dCBT-I and its public health impact; however, the specific facilitators and barriers must first be identified.

To address this gap in knowledge, this study utilized a mixed-methods approach to identify facilitators and barriers of engagement with dCBT-I. We analyzed responses to an exit survey in individuals who received dCBT-I in a treatment research study. To explore potential differences in treatment outcome, responses were tagged by insomnia remission status at post-treatment.

Methods

Participants in this study included individuals who had exposure to dCBT-I through a randomized clinical trial examining the effectiveness of dCBT-I (NTC02988375). All participants were invited to complete an exit survey (appended to the post-treatment survey) that included both quantitative and qualitative items addressing dCBT-I treatment feasibility and acceptability.

dCBT-I Intervention

The dCBT-I platform was Sleepio (Big Health), accessed through an Internet browser or mobile application. The platform comprised six weekly sessions with five core components: 1) sleep restriction, 2) stimulus control, 3) sleep hygiene, 4) cognitive therapy, and 5) relaxation. The content was guided by an animated digital therapist with interactive components. Participants also completed sleep diaries within the platform and had access to adjunctive resources including an online community and a library of educational materials about sleep health. In addition to resources available through Sleepio, participants also interacted with a small team of research staff, who provided support focused predominantly on research-related processes (e.g., signing up for a Sleepio account) and answered participant questions by email and phone. Questions regarding Sleepio beyond the scope of this study were referred to the Big Health technical support team.

Outcome variables

The quantitative questions included five-point Likert-type questions about participants’ experience with the interventions (e.g., “How satisfied are you with the treatment received for your sleep problem?”) The questions measured six domains: satisfaction, acceptability, adherence with treatment strategies, positive outcome expectancy, how much the content made sense to the user, and how appropriate the intervention was for their chief complaint.

The qualitative item solicited free-response feedback regarding the study intervention (i.e., “We would like to hear what you think of the way we conducted this study.”) Three example prompts were presented to aid participants in generating content for the free-response feedback: 1) What did we do well to help you through this study? 2) What could we have done better? 3) Do you have any concerns with the way the study was carried out? Study procedures were approved by the Institutional Review Board, and informed consent was obtained prior to study procedures.

Data Analysis

The exit survey corpus contained a total of 336 responses to the quantitative data, and 287 free-text responses. One response was removed from analysis (participant requested that the study team call them for feedback: “just call”). Individual response data were divided by patient remission status as in remission (final Insomnia Severity Index (ISI) score ≤ 7 ²¹; N = 160) or not in remission (N = 126). A thematic analysis was performed on the free-response items using the method outlined by Braun and Clarke.²²

Quantitative data were compared across remission groups using a multivariate analysis of variance. A thematic analysis was performed using the qualitative data, as it summarized key themes and allowed exploration of similarities and differences.²² All free responses were read multiple times by two coders in order to gain familiarity with the data and identify emerging themes. To ensure trustworthiness of the data, an additional two coders were included in the coding and thematic analysis procedures using the method outlined by Braun and Clarke²² and sample coding exercises. Data were then entered into the QSR NVivo 11 software package (QSR International, 2015).²³

Thematic analysis was conducted using deductive (i.e., a priori codes) and inductive (i.e., emergent codes) approaches to balance a priori areas of interest, while also allowing themes to emerge from the free responses. After preliminary analysis, coders created a thematic framework (i.e., a codebook of a priori codes based on scanning the data and outlined research questions). These initial codes (short, simple, and precise key words to represent one idea or concept)²⁴ were identified for both the remission and non-remission data sets using open coding.²⁵ All coders reviewed the data independently, and then used a negotiated approach to develop themes through several iterations of interaction with the text and codes. In this approach the researchers coded the transcripts, and then discussed their codes with an aim to arrive at a final version in which most, if not all, coded messages have been brought into alignment.⁵ Thus, final themes were created using both deductive (i.e., a priori codes) and inductive (i.e., emergent codes) approaches. After further discussion, the authors (PC and SS) confirmed a set of themes and identified and extracted relevant data to represent each theme. The remaining authors then reviewed the themes for coherence. During this interpretive phase of the data analysis, overarching themes were identified to capture the phenomena described in the raw data.

Results

Participants

Exit survey

Of 336 individuals who were solicited for feedback, 85.4% (n = 287) provided free-response feedback; 55.1% of them (n = 185) achieved insomnia remission post-treatment (ISI ≤ 7). There were no differences in the rate of free-response feedback between remitters and non-remitters. See Table 1 for a more detailed description of the sociodemographic characteristics of participants who provided free-response feedback.

Table 1.

Sociodemographic characteristics of participants (N = 287)

Characteristic	Remission	Non-remission
Age	42.6 (15.2 SD)	47.9 (14.5 SD)
Sex
Female	78.8% (126)	74.0% (94)
Male	21.3% (34)	26.0% (33)
Race
Black or African American	13.8% (22)	2.4% (3)
White	78.1% (125)	70.1% (89)
Other	8.2% (13)	27.5% (35)
Education
High school or less	18.2% (29)	10.3% (13)
Some college	26.3% (42)	25.2% (32)
College	40.0% (64)	45.6% (58)
Graduate school	15.7% (25)	18.9% (24)
Household Income
Poverty (<15k)	11.9% (19)	14.1% (18)
Low (<35k)	26.3% (42)	26.8% (34)
Middle (<75k)	31.2% (50)	33.1% (42)
Higher (75k+)	30.6% (49)	26.0% (33)

Quantitative Results

Individuals from both remission and non-remission groups regarded the dCBT-I treatment positively: in the non-remission group, the ratio of positive to negative free responses was 10:1, but it was much higher in the remission group (194:1). Across the sample, 92% reported that the treatment made sense (i.e. agree or strongly agree; mean = 4.5 ± 0.66 SD), 77% reported that the treatment was acceptable, 74% reported that the treatment was appropriate (mean = 4.0 ± 0.96 SD), 68% reported that they expected success with dCBT-I (mean = 3.9 ± 0.96 SD), 85% reported that they complied with treatment recommendations (mean = 4.2 ± 0.76 SD), and 80% reported that they were satisfied with dCBT-I (mean = 4.1 ± 0.95 SD). As expected, non-remitters reported lower agreement on all items, F(6,329)=15.7, p < .001 (see Table 2).

Table 2.

Treatment ratings by remission status. Means ± standard deviation (Percent of responses that were “Agree” or higher)

Variable	Remission	Non-remission
Treatment made sense	4.63 ± 0.52 (94.6%)	4.30 ± 0.76 (84.1%)
Treatment was acceptable	4.41 ± 0.76 (87.0%)	3.82 ± 0.99 (64.2%)
Treatment adherence	4.36 ± 0.69 (90.3%)	4.10 ± 0.82 (77.5%)
Treatment was appropriate	4.35 ± 0.74 (88.6%)	3.62 ± 1.05 (56.3%)
Expected to have success	4.31 ± 0.75 (87.6%)	3.44 ± 0.97 (44.4%)
Satisfied with treatment	4.49 ± 0.61 (94.1%)	3.68 ± 1.09 (61.1%)

Health literacy may be important to dCBT-I engagement

Because our prior report indicated that those with lower socioeconomic position were more likely to discontinue treatment¹⁴, we also examined whether proxies of health literacy were related to engagement and/or remission. Results indicated that those with less education reported that the treatment made less sense, t(333) = 2.0, p < .05 (Cohen’s d = 0.26). Lower ratings for treatment making sense were also associated with lower adherence with treatment strategies, t(334) = 7.5, p < .01 (Cohen’s d = 0.38), lower treatment outcome expectancy, t(334) = 8.8, p < .01 (Cohen’s d = 0.43), and lower odds of insomnia remission, OR = 0.45, p < .001. This was consistent with user comments (see Type and Extent of Information theme).

Qualitative results: Main themes and insights from exit survey

Using the deductive and inductive approach, content was categorized based on contextual markers into the following five themes: Digital Person-to-Person Component, Type and Extent of Information, User’s Sense of Autonomy, App Functionality, and Importance of Tailored Content. Direct quotes have been taken from each remission group and are indicated by square brackets. Example quotes and the prevalence of each theme are provided in Table 3.

Table 3.

Examples of free-responses by category from the exit survey.

	Prevalence (number of responses coded)		Examples
Theme	Remission (N=160)	Non-remission (N=126)	Remission	Non-remission
Digital person-to-person component	53.1% (n=85)	47.6% (n=60)	“I felt if I had a problem there was somebody there, I could relate to.”	“Unlike many studies, I always felt you were in close contact and willing to help with any problems that would or could arise, and all questions and concerns were met with almost immediate answers.”
Type and extent of information	31.3% (n=50)	34.1% (n=43)	“The way things were broken down was informative without being condescending.” “The program was set up great easy to use follow and understand. The information and skills provided are really helpful and I am so thankful.” “I liked the way the classes were conducted at your own pace plenty of info as needed to find a solution to your problem.”	“I'm still not understanding how much sleep I need each night. I need more time to understand how to read the progress reports. And the graph, not sure how to read it.”
User’s sense of autonomy	43.1% (n=69)	73.0% (n=92)	“Thought the program was well done and in a step by step way - to figure out what things were applicable to you - to put them into practice and review the results of the changes you were able to make.” “I thought the process did make me focus on my sleeping issues more than I otherwise would. The conclusions that I've made about my sleeping issues were the same before and after the survey. I need to get to bed at the same time every night, and I donť.”	“I enjoyed the sleep study because I have learned and applied the techniques to help me with my sleep problem.”
App functionality	58.1% (n=93)	56.3% (n=71)	“If I was not at work I would miss some days and was allowed to catch back up that was very helpful. I like the study and how it identified different areas of sleep such as bedroom, lighting, temperature etc. Some techniques were very helpful. I liked checking my problems and thinking about THE a few were more helpful than others.” “The convenience of doing at my leisure was most beneficial.”	“It was easy, it didnť take long and offered good suggestions. App is user friendly.” “I thought the study was well organized, easily accessible through home computer as well as cell phone.”
Importance of tailored content	23.1% (n=37)	52.4% (n=66)	“My only problem with this study was that it was too limited in sleep patterns...you essentially forced me into a box of monophasic sleep, implying that if I was not sleeping through the night, I had insomnia.” “However, this system does not take into consideration "physical" causes of sleep problems I have severe arthritis and stenosis of the lower lumbar. The arthritis in my neck and the tendency of one leg to cramp up at night contribute to the frequency of waking each night.” “I sometimes felt that I couldnť add enough information (i.e. napping during the day and combining that somehow with my nightly sleeping).”	“It is not appropriate for someone whose sleep problems are primarily related to pain and physical discomfort.” “When I enter my sleep data, the diary only wanted my nighttime numbers. The program doesnť take napping into consideration. When I think of the amount of sleep I need, I include my napping hours as well. My fibromyalgia, anxiety, depression and physical pain drains me, so I often need to nap.” “I do think that the study made some assumptions because I may not be able to afford extra pillows or I may live in a tiny apartment and I'm unable to visit other rooms in my house.”

Digital Person-to-Person Component

In mHealth, a digital person-to-person component is one with any features that generate a sense of interpersonal connection or produce virtual interactions. These features sometimes embody qualities of an in-person or face-to-face component (e.g., feedback regarding performance, guidance, or support).²⁶ Free responses categorized under this theme focused on elements of dCBT-I that cultivated an interpersonal connection.

Many comments in this theme referred to the virtual therapist (“The Professor” or “The Prof”), which is a core digital person-to-person feature in the intervention. Results indicated that across remission groups, most participants felt The Prof was entertaining, provided sufficient guidance, and was a valuable, engaging component of the treatment. Exemplar comments include, “The Prof was entertaining, and it was easy to stay engaged in the program. Loved it.” [remission, F, 28] and “I looked forward to meeting with the Professor to get tips and helpful information.” [non-remission, F, 51]. We were also interested to see if patients would express a desire for demographic matching in the animated therapist given prior findings of demographic differences in engagement, but this did not emerge as a theme. Participants did state that a more neutral avatar may be more accessible for others. Despite positive regard for the virtual therapist, participants across both remission groups still desired additional in-person interactions (e.g., “I would have liked a one-on-one counsellor” [remission, F, 55]), and non-remitters reported that in-person interactions may have increased treatment effectiveness (e.g., “Talking with a real person would be more helpful” [non-remission, F, 44]).

Several free-responses also referred to guidance, feedback, and support provided in dCBT-I. Most participants remarked that social support and connectedness were important elements of treatment. Specifically, many mentioned the platform’s online community or reaching out to the research team, and most appreciated access to a social network. Exemplar comments include, “I liked that there was a community that I could go to discuss sleep issues. It is nice to hear others with the same sleep issues that I have, and how they are coping.” [remission, M, 53] and “It gave me a support group to talk with that help me when I could go on.” [non-remission, F, 63]. Participants also mentioned that these digital social supports helped them feel accountable, motivated, and connected. Moreover, participants who experienced barriers accessing support networks lamented the missed opportunities for increased engagement. For example, “Was not clear how to access Wed. talk by specialist. Still don’t know if it is live or all written responses to questions.” [remission, F, 72] and “I was confused a few times on some things like discussion boards. I kept being told I would get links for it but never did.” [non-remission, F, 44].

Type and Extent of Information

This theme included free responses that focused on the quality of guidance provided through the intervention. This is consistent with prior studies showing that presenting content that is relevant and accessible is central to digital sleep interventions²⁷.

Overall, participants from both remission and non-remission groups felt the pacing and grouping of information into sessions was effective. Most participants perceived that the didactic content was presented in a relevant and meaningful manner. For example, “I gained helpful information to help me with my sleep issues and concerns.” [non- remission, F, 45] and “I thought the info given during the study was clearly stated and understandable. The info was short and to the point. I felt all the topics gave sound advice on how to deal with insomnia. I shared quite a few of the tips because they seem practical and easy to incorporate in everyday life.” [remission, F, 42]. Some non-remitters indicated that they already knew the information or that they were unable to implement the strategies, and thus it was not helpful in improving their insomnia (e.g., “All of the information in this study I already knew as a result of a quick google search. It didn’t help me much.” [non-remission, F, 44], and “I liked it. I was hopeful but sadly I was unable to follow some of the tips like times to wake up.” [non-remission, F, 52]). Though not common, a minority of users did report struggling to understanding the material (e.g., “I need more time to understand how to read the progress reports. And the graph, not sure how to read it.” [non-remission, F, 66])

Users’ Sense of Autonomy

This theme pertains to the extent to which users during treatment felt autonomy and agency, which are important to treatment engagement because they prevent over-reliance on treatment options that patients do not value and thus do not engage in consistently. ^28,29 Comments in this theme included free responses that reinforced users’ ability to have choice and control during treatment. This theme was more salient for non-remitters (73.0% of responses; see Table 3) compared to those who remitted following treatment (43.1% of responses; see Table 3).

For non-remitters, comments centered on opting out of key treatment strategies. Exemplar comments include: “I did not do the sleep restriction aspect of study, difficult to do while on my crazy work schedule” [non-remission, F, 42], and “My fibromyalgia, anxiety, depression and physical pain drains me, so I often need to nap. Attempting to keep my sleep window to that single period even adds to my need to sleep” [non-remission, M, 50]. Additional comments from the non-remission group included a wish for more options that enhanced autonomy and the right to regulate and control their own treatment experience. For example, “the only Problem with the program is that it seems to focus on getting to sleep and sleeping well. However, my main problem / frustration with sleep is waking an hour before my desired time (EVERYDAY) and there is no section that helped me with that.” [non-remission, M, 50].

By comparison, although remitters commented less frequently on sense of autonomy, they appreciated agency in digital treatment (e.g., “I liked the fact that I didn’t have to try all the suggestions, but that I could pick which ones would fit my lifestyle the best.” [remission, F, 19]).

App Functionality

This theme referred to the functions on the dCBT-I platform that allowed the user to engage in certain kinds of behaviors, such as entry of sleep diaries. Comments are sometimes related to the user interface and user experience (UI/UX, i.e., ease of use), but also referred to the infrastructure provided around the digital intervention.

Overall, the majority of participants described the digital platform as flexible, convenient, and “user friendly” Moreover, the user interface and the infrastructure surrounding the digital intervention promoted engagement. Participants who achieved remission commonly commented that, “The quizzes helped me pay more attention. The sessions were fun and interactive. Overall, very interesting and helpful” [remission, F, 25] and “I liked that I received email notifications every day that I needed to fill out my diary and complete weekly sessions. This helped tremendously in holding my accountable and keeping me active in solving my sleeping problems.” [remission, M, 25]. Conversely, non-remitters reported “bugs” that negatively impacted content delivery and functionality or had trouble finding specific functions such as the community forum, all of which detracted from the user experience. For example, “Somehow, I couldn’t see lesson 3 with the professor.” [non- remission, F, 52] and “Tried HTML version during several sessions, need work as it never functioned on either of my computers.” [non-remission, M, 63]. App functionality was the most prevalent theme for participants who achieved insomnia remission (58.1%; see Table 3).

Importance of Tailored Content

This theme referred to free responses pertaining to customization of the content to the individual’s specific circumstances or needs. This theme is particularly relevant to digital therapeutics, as the removal of clinician input as a bottleneck is often a critical trade-off for high scalability; however, the loss of clinician input can result in loss of flexibility to tailor therapeutic content to the individual.

Many of the free responses in this theme indicated the importance of individualized fit and the need for personalized input supplementary to objective data. This theme was more prominent among non-remitters (52.4% of responses) than remitters (23.1% of responses). Non-remitters commonly reported wanting to provide additional contextual information to inform the treatment, e.g., “…I wish there could have been a place to communicate while during the session with the Prof, so that he could understand where I was coming from and why I was having difficulty sleeping in the first place.” [non-remission, F, 48].

For some, the collection of personal or environment challenges could have impacted the effectiveness of the intervention. For example, “The worst part of the study for me was that you avoided the medical issues that I impact my sleep.” [remission, M, 66] and “So many people have extenuating circumstances that may affect their sleep. I have a nursing toddler, a hostile/toxic work environment, and I am caring for elderly parents. I’m one person. I’m exhausted most of the time. Due to all my responsibility, I am often up very late. There is no way to fix that. I don’t think the study considered such extenuating circumstances.” [non-remission, F, 43]. Another said, “I think the study is only effective for a certain cross-section of the population. It does not take into account certain factors, such as parents with small children or parents with special needs children. It also does not take into account people with neurological and autoimmune disorders, who are frequently tired and need more rest, not less. If you could tweak the program for people in those categories, it would be a big help.” [non-remission, F, 42]. Another person commented that “I think daylight savings time should have been addressed when we changed time.” [non-remission, F, 64]

Discussion

Given the advantages of digital therapeutics for wide-scale dissemination, it is likely that dCBT-I will soon be a standard treatment approach for disseminating CBT-I. Indeed, one platform for dCBT-I (Somryst) has recently been FDA approved and is now available as a prescription for insomnia. However, emerging evidence indicates that treatment effectiveness is diminished by disparities in engagement.¹⁴ As such, it is important that we understand the strengths and opportunities for improvement from a patient-centered perspective to ensure equitable implementation and dissemination. This study took a patient-centered approach in understanding the experience of dCBT-I with a focus on understanding potential differences in experience between those who did and did not achieve remission following completion of dCBT-I.

Facilitators of dCBT-I engagement

Overall, patients generally enjoy dCBT-I; even those who did not remit following treatment expressed an extremely high ratio of positive to negative comments (10 positive: 1 negative). In addition to the accessibility and conveniences offered by digital health, results revealed several important components contributing to a positive experience with dCBT-I. One critical element was digital person-to-person components, which include features that enhance a sense of engagement and connectedness. Examples include human support provided through peers via online message groups or by posting or reading bulletins,³⁰ as well as other elements of online systems that create social support, such as chat forums and/or chat rooms.³¹

The dCBT-I platform used in this study (Sleepio) also featured an animated therapist (“The Prof”) as a digital person-to-person component. A large majority of patients expressed affinity towards the animated therapist, with some even stating a desire to be able to share more details about themselves and their insomnia with The Prof. Given prior findings of demographic differences in engagement, it is of note that demographic matching did not emerge as a theme regarding the virtual therapist though participants did state that a more neutral avatar may be more accessible for others. Although demographic matching has relevance in the context of psychotherapy with a provider, its relevance in the context of an animated therapist may be more limited. Clients may have lower expectations for a strong therapeutic alliance with an animated therapist, or they may only expect the animated therapist to make the content more engaging. That said, another recent study indicated that demographic matching in Black women via the addition of videos with actual Black providers did enhance treatment completion ²⁰, indicating that it can improve engagement.

Other facilitators included the perception that the content was relevant and well-paced. Unsurprisingly, a smooth app interface and functionality was important for engagement and was a prominent theme in non-remitters. Notably, both facilitators may also result from higher health literacy and technological literacy. Indeed, higher health literacy likely helps with integrating new sleep health information or strategies, thus creating the perception that the information is presented clearly and at an appropriate pace. Higher technological literacy also likely means more finesse in navigating the dCBT-I platform, thus producing less frustration with the technology. Finally, users also liked the autonomy of picking which strategies to implement; however, free responses in this theme were more prominent in non-remitters who opted out of key treatment strategies.

Barriers to dCBT-I engagement

Despite general enjoyment of dCBT-I, satisfaction with the treatment differed between those who did and did not remit. Though there was general affinity towards the animated therapist, many users still expressed a desire for human interaction, though there was a range in the extent of interaction desired. While some wanted more traditional patient-provider interactions, others just wanted to know that their progress was being monitored by a provider. This reflects the importance of support and accountability in behavioral treatment. Interestingly, the desire for more human interaction was shared among both remitters and non-remitters, suggesting that, at least for a subset of individuals, the human interaction was desirable but not necessary. This suggests that patients with support and accountability independent of dCBT-I (e.g., familial support, prior success with digital health systems, etc.) may be more likely to achieve success with standalone dCBT-I. This also suggests that dCBT-I in its current form may have inadequate support for users without existing support systems, and that enhancing dCBT-I with more opportunities for human interaction (i.e., built-in support and accountability) may be a targeted way to retain engagement and enhance treatment completion.

Another barrier that may be associated with reduced engagement and non-remission is clinical complexity, such as comorbid conditions (e.g., pain disorders), which may make it more difficult to implement strategies as recommended. People who reported clinical complexities also exhibited lower outcome expectancy, in part due to perceived inappropriateness of the treatment due to a lack of individualized accommodations in the treatment strategies. This is important because non-remitters’ lowest scores were on outcome expectancy, followed by perceived appropriateness of treatment.. Non-clinical complexities were also observed, such as long or unpredictable work schedules (e.g., those who work multiple jobs, or do precarious work), parenting, or travel. These complexities are naturally accounted for in face-to-face CBT-I, where the provider tailors strategies to enhance feasibility for the patient; however, this was not available in the digital context.

Although user autonomy emerged as a facilitator for treatment engagement, it may also be a barrier to treatment effectiveness. Increased autonomy is generally desirable because it is associated with treatment engagement and persistence; however, user choices that are underinformed or misguided can also reduce treatment efficacy. For example, those who did not remit were more likely to have opted out of key treatment strategies such as sleep restriction and/or stimulus control, with many users reporting feeling that the strategies lacked feasibility given their circumstances (e.g., work schedule). Given that this theme was more prominent in non-remitters, this may suggest that opting out of key strategies such as sleep restriction could be associated with smaller treatment gains.

Many factors contributing to treatment non-completion (and thus non-remission) could be explained in part by health and technological literacy. For example, those who have greater capacity to acquire and use health information are better equipped to independently adapt the dCBT-I strategies for their specific circumstances, despite clinical or non-clinical complexities. For example, users who understand their comorbid pain condition (e.g., triggers, remedies) and the bidirectional relationship between sleep and pain may be more able to self-identify adaptations of the dCBT-I recommendations or be more motivated to seek out adaptations from the Internet (e.g., Facebook groups and/or online forums about CBT-I). In contrast, without additional support, those with lower health literacy are less likely to implement the prescribed strategies and more likely to disengage from dCBT-I and remain symptomatic, as observed in this study.

Health literacy may also indirectly impact engagement with dCBT-I via outcome expectancy and self-efficacy. In this context, outcome expectancy refers to the belief that a positive outcome will occur as a function of the intervention, whereas self-efficacy refers to the degree of conviction that one can successfully execute the behavior required to produce the desired positive outcome. In accordance with the Expectancy Theory of Motivation ³² and the Social Cognitive Theory ³³, both outcome expectancy and self-efficacy are critical components to attempting and sustaining new health behaviors. Both components are also impacted by health and technological literacy: inadequate comprehension of the recommended strategies reduces motivation to engage in behavioral change, particularly if the behavior is challenging (e.g., sleep restriction); similarly, lower health literacy may thwart efforts to translate new health behaviors into their specific circumstances, reducing the likelihood of engagement and implementation. Indeed, we observed that those with lower comprehension of the intervention reported lower outcome expectancies, and those who did not remit reported low feasibility of the behavioral prescriptions within their individual circumstances.

Potential modifications to enhance engagement

These themes suggest several potential enhancements to dCBT-I that could increase engagement. These primarily include inclusion of patient support as needed, ways to enhance familiarity with the digital platform, and tailoring the type and extent of information. While existing digital person-to-person components (e.g., animated avatar, online communities) have been well-received, these results indicate that there remains a need for additional support. One impactful expansion may be to target digital person-to-person components that enhance technology and health literacy. Examples might include both synchronous (e.g., telehealth appointments) and asynchronous (e.g., web-videos) services to enhance familiarity with the digital platform. These may include additional content that explains how to create an account, what functions are available, why the functions exist and demonstrations of how to use them across multiple contexts. Given recent evidence supporting the efficacy of cultural adaptations to dCBT-I for Black women²⁰, it may be prudent to consider for additional adaptions for other subgroups. For example, enhancing digital person-to-person elements may enhance the accessibility of dCBT-I for individuals with less socioeconomic resources.

Providing as-needed patient support may be an area of priority, as there was consensus that this feature would have promoted persistence with dCBT-I. One approach is to utilize generalist providers (e.g., nurses, physician associates/assistants, social workers) with basic CBT-I training. This would minimize the bottleneck created by limited availability of specialists. Examples of support that generalists could provide include monitoring treatment progression, mid-treatment check-in, or providing feedback and/or encouragement to enhance engagement.

Enhancements to the type and extent of information may also aid engagement. Given the emergent theme that some information was either already known or easily accessed via web search, it may be important to augment the focus on implementation of treatment strategies. Examples may include adjunctive modules that modify existing strategies for common sub-groups of patients (e.g., a module on dysfunctional pain-related beliefs and attitudes about sleep [PBAS]^34,35 for insomnia patients with chronic pain). These should be optional modules to balance engagement across differing levels of health literacy (i.e., those with higher health literacy can avoid feeling bogged down by information that might feel redundant, whereas those with lower health literacy can opt-in for further support). As technology advances, dCBT-I platforms may also consider ways that the content can be tailored for individuals with more complex conditions or circumstances, such as co-morbid pain or unpredictable work hours.

Another enhancement may be integrating dCBT-I within an adaptive stepped-care framework, which begins with the least resource-intensive and burdensome treatment (e.g., dCBT-I), and only steps up to more resource-intensive or specialist-driven treatments when needed (e.g., face-to-face CBT-I). More adaptive approaches to stepped-care (e.g., just-in-time interventions) may increase efficiency by providing only as much care as is needed to promote optimal outcomes. Stepping up to treatments that focus on implementing sleep restriction may also be appropriate (e.g., Sleep Restriction Therapy or Brief Behavioral Therapy for Insomnia).

Conclusions

This was a mixed-methods approach to understanding patient-reported facilitators and barriers to engagement in dCBT-I. Five themes emerged: 1) digital person-to-person components, 2) type and extent of information, 3) user’s sense of autonomy, 4) app functionality, and 5) importance of tailored content. Evidence also suggested that health literacy is an important factor impacting engagement, and thus should be a target in future modifications.