An additional layer of stigma that individuals with serious mental illnesses may experience is the assumption that they cannot or do not use digital devices for their mental health. A digital divide exists in mental health, wherein individuals experiencing serious mental illnesses are excluded from receiving support through digital means because of a lack of access, skills, or confidence [16]. However, evidence suggests that this exclusion is declining [17], particularly within psychotic disorders and BD [18,19]. However, developing digital interventions for individuals experiencing serious mental illnesses can introduce additional challenges because of illness-related factors, such as cognitive impairments or mistrust, thus requiring specific design considerations [12,16].

The findings of 6 recent systematic reviews of digital interventions for serious mental illnesses demonstrate that these are feasible and acceptable, with preliminary indications that these may be effective for symptoms and cognitive and social outcomes [24-29]. However, in each review, insufficient data were available to draw firm conclusions. Specifically, in a systematic review including individuals with psychosis, results provided preliminary evidence that these may have benefits for positive psychotic symptoms and depression, hospital admissions, medication adherence, socialization, and social connectedness [30]. For example, the recent Actissist proof-of-concept trial, in a small sample of individuals with early psychosis, demonstrated feasibility and acceptability through high use and adherence as well as symptom improvements in favor of the intervention group [31]. The handful of digital interventions for BD evaluated to date have produced inconsistent results. A systematic review of these studies showed that while these appear to be feasible and acceptable intervention options, inconsistent evidence was found for their effectiveness for symptoms or recovery outcomes such as quality of life [32]. The current Canadian Network for Mood and Anxiety Treatments and International Society for Bipolar Disorders treatment guidelines conclude that, to date, there is insufficient evidence to support the use of digital interventions as adjunctive therapies for BD [33]. However, multiple trials are underway, which will advance this literature [34-38].

A number of meta-analyses point to a modest but reliable relationship between the quality of therapeutic alliance and outcomes of face-to-face therapy, with effect sizes typically in the moderate range [42-44]. Some research also supports the notion of a causal relationship between therapeutic alliance and outcomes in serious mental illnesses, for example, increasing the number of sessions attended was only beneficial in the presence of a strong therapeutic alliance for individuals experiencing psychosis [45]. Two systematic reviews focusing on psychosis found some evidence that the therapeutic alliance was related to reductions in psychotic symptoms, hospitalizations, and self-esteem outcomes [46] as well as functioning and treatment adherence [47]. Similarly, in BD, therapeutic alliance have been linked with decreased stigma, more positive attitudes toward medication, and, less conclusively, a reduction in symptoms [48].

Given that the therapeutic relationship is a necessary (and some argue, sufficient) component of change in traditional face-to-face psychotherapies, a significant reservation toward digital interventions is the loss or, at a minimum, the modification of the therapeutic relationship. Some commenters have started to consider whether and, if so, how the therapeutic alliance plays a role in digital interventions [49]. Indeed, the James Lind Alliance identification of key priorities in advancing digital interventions included concerns relating to the therapeutic alliance in the top 10 issues raised [50]. The definition by Bordin has been transplanted to web-based environments, most commonly, digital interventions with human support components. However, it cannot be presumed that traditional conceptualizations of therapeutic alliance transfer to digital environments nor that different digital environments foster therapeutic alliance in equivalent ways.

The dimensions of the therapeutic alliance may differ in digital environments. A systematic review identified additional themes of availability, indicating how freely and conveniently accessible the digital intervention is, and interactivity, indicating the degree to which personalization and feedback based on user input is provided and to which the user feels in control [51]. These themes support the possibility of a perceived bidirectional relationship between a user and a digital system, in which automated aspects of the experience could emulate a reciprocal, trusted relationship. A qualitative analysis mirrored the importance of these domains within the therapeutic alliance for participants. For example, one participant identified that automated personalization helped them understood and conversely, another user would have perceived a relationship if the site responded intelligently [51]. Furthermore, following an examination of a supported digital intervention for carers, users perceived automated feedback, designed to emulate human communication, as supportive and helpful [52]. More work is needed to explore therapeutic alliance features and domains within digital contexts. Nonetheless, these investigations shed some light on the nature of the alliance in digital contexts, providing a provisional characterization of factors relevant to the alliance in digital contexts.

In parallel, the measurement of the therapeutic alliance requires specific consideration for digital environments. Simply replacing therapist with program or app in existing measures may fail to account for the complexity of the therapeutic alliance in digital interventions and parcel out the relative contributions of human-human and human-technology relationships. Researchers are beginning to acknowledge this; for example, WAI-Tech, a digital adaptation of the frequently used alliance instrument, the Working Alliance Inventory (WAI), includes reworded items in the bond subscale, omitting the human element [53]. Another adaptation applies to supported contexts, anchoring the task and goals subscales to the intervention and the bond subscale to the therapist [54]. Similarly, Berry et al [55] adapted the ARM for digital contexts by consulting end users and mental health professionals. In addition, the recently developed Enlight measure was specifically designed to assess the features of digital interventions, including those related to therapeutic alliance [56]. These measures require further validation but represent a starting point for capturing the unique qualities of the digital therapeutic alliance.

Although it has been established that the therapeutic alliance can be experienced in digital interventions, links to outcomes are less clear than in face-to-face therapies.

Although evidence relating to a direct relationship between ratings of therapeutic alliance and treatment outcomes is inconclusive for digital interventions, one possibility is that the therapeutic alliance is associated with increased engagement and adherence, in turn, influencing outcomes [102]. It should be noted that engagement and adherence are often used interchangeably or poorly defined, limiting conclusions. In this review, engagement will refer to measures of the amount, frequency, or depth of program usage, whereas adherence will refer to whether participants met an a priori metric of intended usage or dose; however, we are limited to the information presented. For example, a real-world analysis of a range of apps and sites available to the public, with varying support, demonstrated that ratings of the therapeutic alliance (measured using the Enlight measure) were associated with increased user engagement [76].

The above findings are especially compelling given that adherence and engagement are themselves associated with improved outcomes in digital interventions. For example, a meta-analysis demonstrated that adherence to both the intervention and the study was associated with a range of behavioral and health outcomes in digital interventions [103]. Furthermore, a systematic review showed that a broad range of adherence measures (many of which align with our definition of engagement) was associated with physical and psychological outcomes in digital interventions [104]. Associations with adherence have been demonstrated for a range of outcomes, including depression [105,106] and anxiety [107,108].

Engagement with digital interventions has also been linked to outcomes. Across outcomes such as depression and anxiety [109-111], greater changes in symptoms have been associated with better engagement with the site. Similarly, in a program for preventing eating disorders, the duration and extent of program use predicted some outcomes [112].

These findings suggest that a therapeutic alliance is associated with increased engagement with and adherence to digital interventions, in turn, influencing outcomes. However, this explanation represents one possible interpretation of these limited data and requires direct examination. Given to the inconsistencies within this literature [113], more work is needed to identify the engagement and adherence metrics with the most importance. In addition, no identified studies investigating links between outcomes, adherence, and engagement included samples with serious mental illnesses.

Although it appears that human support is beneficial for therapeutic alliance, in many cases, human support is not practical or desirable. Some users, including those with serious mental illnesses, report a preference for unsupported interventions, with benefits such as the reduced potential for judgment and increased honesty [25,51]. How the definition by Bordin [39] on therapeutic alliance can apply to unsupported interventions is not yet clear, but it has been suggested that therapeutic goals and therapeutic task components (or analogs thereof) may be cultivated within interactions between users and the intervention, while the bond element may be either entirely absent or modified [114]. For example, while equivalent ratings on the remaining alliance subscales were reported, the bond subscale of the WAI was significantly lower in a digital intervention for cocaine dependence than in a face-to-face group [53].

In the absence of human support, it may be more important to ensure that unsupported digital interventions incorporate automated features that resemble a bidirectional therapeutic relationship, such as the communication of empathy, responsiveness, and supporting the user’s motivation. Barazzone et al [115] examined the content of 3 widely used CBT programs to investigate the extent to which they incorporated key features for the establishment, development, and maintenance of a therapeutic alliance between the user and the program. They concluded that the programs exhibited substantial evidence of therapeutic alliance features, such as empathy and acceptance, and the negotiation of goals, providing feedback and building confidence in the program’s effectiveness, and rupture prevention and repair by encouraging users to return to the program after a break. Similarly, Holter et al [116] describe their attempts to cultivate a therapeutic alliance in a smoking cessation digital intervention, simulating a therapeutic alliance by allowing users to negotiate goals and using a conversational agent to implement human strategies such as empathy, interactivity (via remembering previous communications and tailoring), and humor. These authors did not assess the therapeutic alliance, its relationship with program features, or outcomes. However, Bickmore et al [66] evaluated the impact of introducing a relational agent that simulated alliance-promoting behaviors such as social dialogue, empathetic feedback, process comments, humor, and nonverbal communication, alongside a fitness digital intervention. The bond subscale was rated significantly higher within the group that had contact with the agent, whereas the overall WAI scores did not differ significantly.

Furthermore, additional provisional characteristics of the digital therapeutic alliance, availability, and interactivity, are not grounded in human support [51] and speak to the unique strengths of digital interventions. Technology is ideally positioned to be accessible when and wherever users require it, and a number of technological features can be optimized to cultivate interactivity, that is, a bidirectional relationship might be effectively emulated between a user and a digital intervention, with automated features. In addition, people may form attachments to smartphones, similar to an alliance [117,118], although users appear hesitant to call this a relationship [51]. Although human presence may be an important predictor of the therapeutic alliance in digital interventions, the experience of human presence may not require direct human contact. For example, the object-relations theory explains how devices may act as proxies for relationships with caregivers (the humans behind the intervention [119]).

Examining which technological features are best positioned to foster the therapeutic alliance in both supported and unsupported digital interventions is therefore an important next step, and we propose that exploring the additional provisional characteristics of availability and interactivity may represent a significant advancement.

Tailoring systems and personalizing messages to users are important components of system design [121]. An example of tailoring is to suggest specific content based on what is known about users’ preferences or background, whereas personalization might reflect the person’s name or information about their medications, functioning, or previous responses [121].

Empirical research findings have indicated that when intervention content is matched with users’ psychosocial and behavioral characteristics, it is perceived as personally relevant, which enhances engagement [129] in line with the elaboration likelihood model [130]. Conversely, a 2016 review [85] indicated that the most commonly cited reason for low adherence within the digital intervention was a perception of impersonal or irrelevant content.

Therefore, tailoring and personalization may help foster a digital therapeutic alliance via congruency between the users’ personal needs and goals and interactions with the intervention. In addition, tailoring and personalization speak directly to the provisional digital therapeutic alliance’s characteristic of interactivity. For example, tailoring the intervention to a person’s stage of recovery in face-to-face therapy and not assuming motivation to change is predictive of a strong therapeutic alliance and outcomes [131]. In this vein, it has been argued that individually tailored content or responses within digital interventions enhance the sense of relationship with the platform, whether supported or unsupported [55]. Research findings provide some support for this claim, with qualitative accounts indicating that when users perceived an app as not tailored to their individual needs (eg, generic), the development of a relationship was hindered [55]. This is an important consideration when developing digital interventions for individuals with serious mental illnesses, where concerted efforts to build trust that the treatment will address the clients’ own unique goals may provide a buffer against feelings of mistrust or coercion [132].

Dialogue support is a PSD strategy that supports user-digital interactions deemed by users to be interpersonal or social interactions [121]. Specific examples of dialogue support in digital interventions include offering positive feedback (praise), reflecting users’ goals and tasks (reminders), linking users with peers via forums (social role), or providing recommendations for appropriate content (suggestions) [94,121]. Sustained engagement with a digital intervention is likely when the intervention offers ongoing interaction that is relevant, motivating, and tailored to users’ needs [113]. Review findings have demonstrated that adherence to digital interventions, in general, is contingent on the extent to which dialogue support elements are utilized [94]. Similarly, a recent review showed that a greater number of dialogue support elements was associated with beneficial outcomes [125].

Research findings also demonstrate that the quality of the therapeutic alliance is contingent on the perceived quality of the interaction and, therefore, implementing responsive dialogue support features within digital interventions should be a consideration in fostering a digital therapeutic alliance [133]. Improved dialogue between users and digital interventions could not only assist users to achieve their goals and foster a sense of interactivity but also enhance self-efficacy [121]. This is an important consideration for individuals with serious mental illnesses because research findings indicate that individuals who perceive greater control over their lives (ie, internal locus of control and greater self-efficacy) are more positive about the therapeutic alliance [134]. Furthermore, although it has not been studied in the context of digital interventions, an internal locus of control is associated with increased treatment motivation, compliance and treatment adherence, and better treatment outcomes in individuals with serious mental illnesses [135].

System credibility or credibility support refers to how digital interventions incorporate technological features that convince the user that the system is credible [94,121]. Examples of how credibility support can be used in digital interventions include providing evidence-based content, expert moderation (eg, clinician support), and explicit, credible third-party endorsements [136]. Credibility support is associated with perceived authority, expertise, and trustworthiness among users of digital interventions, and findings show that its deliberate implementation is also associated with engagement [137].

Implementing credibility support features to increase users’ perceptions of expertise and trustworthiness is likely to help strengthen therapeutic alliance. For example, the bond between therapists and clients in face-to-face settings is contingent on the client’s confidence in the therapist’s competence [138]. Scant literature has examined this phenomenon within a digital context; however, Mackie et al [139] examined the effectiveness of a mobile-based intervention to treat harmful substance use in veterans presenting with self-harm and found that trust was associated with the quality of the digital therapeutic alliance, whereas damaged trust can lead to disengagement. This is a particularly important clinical consideration for individuals with serious mental illnesses who are often difficult to engage in ongoing treatment and have high dropout rates [132].

Persuasive design strategies in the social support category refer to technological strategies that motivate users by leveraging social influence (eg, cooperation, social facilitation, and social comparison) [121]. Theoretically, individuals will be more motivated to perform a target behavior if they can use a system to observe (eg, social facilitation) or compare themselves (eg, social comparison) with others performing the behavior and if they are provided with a means to cooperate or connect [94,121]. An example of how digital interventions can include technological features to facilitate social support include incorporating newsfeeds and/or social forums. There is some support to suggest that the inclusion of social networking facilities in digital interventions for individuals with serious mental illnesses can assist in fostering a sense of social connectedness [35,140,141], a sense of cooperation (shared goals), and increased accountability toward treatment [137].

Developing digital interventions that utilize social support technological features may also help foster a digital therapeutic alliance. For example, study findings relating to face-to-face therapy have shown an association between clients’ levels of perceived social support and their ratings of the therapeutic alliance [142]. Research findings demonstrate that individuals with serious mental illnesses report benefits from interacting with peers on the web, including greater feelings of group belonging and social connectedness [143]. In addition, as discussed previously, individuals with serious mental illnesses who may be uncomfortable with or avoidant of conventional social contact may perceive digital interactions as less threatening [144]. Therefore, the facilitation of social support should be a design feature of digital interventions for individuals with serious mental illnesses.