Despite the importance of knowledge acquisition and skill enactment in psychological interventions, the digital health literature seldom mentions a guiding framework that positions the user as a learner. One potential reason for this could be the field's limited efforts to integrate findings from seemingly unrelated disciplines (11, 33). A useful starting point for bridging this gap may involve drawing insights from a different line of research, as there exists an array of learner-centric models in the education literature that could potentially be applied to explore user engagement in the digital health context. Self-regulated learning (SRL) is a key conceptual framework in educational psychology developed to understand how individuals engage with learning tasks in a self-directed way. The term SRL refers to the self-driven process whereby learners actively control their emotions, cognitions, and behaviors, using strategies oriented towards learning and, ultimately, goal attainment (45). An important assumption in SRL theory is learner agency; the recognition that individuals have the potential to actively control their cognitions, motivation, and behaviors throughout the process of learning. However, not everyone can or will control these to the same degree (46). This is supported by a large empirical base suggesting that people significantly differ in their SRL profiles, even among those who self-select into programs (e.g., online university courses, massive open online courses) that require considerable independent learning skills (47–50). Given their self-guided nature and focus on knowledge transfer within these self-directed learning opportunities, it is plausible that this pattern of findings may also apply to DHIs. More specifically, it may be that variability in engagement behaviors with DHIs, at least in part, reflect individual differences in SRL ability across users. This is because not all users will have the same proficiency to cope with the information imparted in a self-guided intervention program.

Interestingly, there is a growing body of research showing that aspects of SRL (e.g., self-observation, and goal-setting) have been considered in the design of commercial smartwatches and smart home systems with respect to their role in supporting increased awareness and self-management of health-related issues (51–53). Smartwatch features that enable the collection of individual health data (e.g., heart rate variability, daily step count, sleep data) indirectly supports principles of SRL, as access to this information can empower people to set personal goals, track their progress, and regulate their behaviors to improve their ongoing health. Similarly, smart home technology has drawn considerable interest in recent years for its potential to facilitate self-management behaviors in individuals with cardiovascular diseases (53, 54). The emerging research shows that smart home ecosystems can support people with monitoring their symptoms, adhering to treatment regimens, and generating feedback for self-regulation. These features enable patients to play a more active role in their recovery, which has been reported to enhance treatment compliance (55). This literature further highlights the likely relevance of the SRL framework to guide the exploration of potential facilitators of engagement in both web and app-based interventions.

Zimmerman's SRL model (56) is one of the most prominent frameworks in the education literature (57). Heavily influenced by Bandura's social cognitive theory (58), Zimmerman's model describes how metacognitive and motivational processes cyclically influence SRL across three interdependent phases: (a) forethought, (b) performance, and (c) self-reflection. In the current version of the model, learners in the forethought phase analyze learning tasks using skills like goal setting and strategic planning, which is facilitated and fueled by personal motivational beliefs. During performance, learners use self-control strategies such as time management and self-instruction to maintain the cognitive engagement required to complete the learning activity. During this phase, self-regulated learners also track their progress using self-observation strategies like self-monitoring and self-recording (59). In the self-reflection phase, learners appraise their recent performance and find reasons for personal successes and failures. The model posits that future task performance is shaped by a learner's self-reaction and judgments, as this can positively or negatively impact future self-efficacy and learning efforts (57, 60).

Pintrich's model (61) of SRL is another leading conceptual framework in the education literature embedded within the socio-cognitive theory (46). The model classifies SRL into four interactive phases: (a) forethought/planning/activation, (b) monitoring, (c) control, and (d) reaction/reflection. Each phase has four areas for regulation (cognition, motivation, behavior, context), thus giving rise to a vast number of possible SRL processes (60, 62). While the different SRL phases appear to reflect a linearly ordered sequence, it is recognized that earlier phases may not always occur before later phases. Instead, planning, monitoring, controlling, and self-reflection are viewed as ongoing dynamic processes that can simultaneously occur as a learner engages and progresses with a learning task (61). The overarching theme forming the focus of Pintrich's model is the strong integration of motivational processes (e.g., goal orientations and self-efficacy) across each phase, despite its position as an independent area for regulation (62, 63).

Boekaerts' (64) Dual Processing Model (previously called the Model of Adaptive Learning) highlights emotion regulation as an essential aspect of SRL, as emotions can influence when, how, and why students self-regulate their learning (65, 66). The model discerns three purposes for self-regulation during learning: (a) to develop one's skills and knowledge, (b) protecting the ego from harm and threat, and (c) protecting one's devotion to the learning task (60). In the model, Boekaerts explains that learners activate the growth/mastery pathway when a learning task aligns with personal goals and positive emotions are triggered. This encourages learners to commit and metacognitively sustain their engagement to the learning task without needing much extrinsic assistance because the learner is intrinsically driven to develop their competence (64). Conversely, learners initiate the well-being pathway when tasks are perceived as difficult or threatening to the ego and negative emotions are triggered. When the well-being pathway is activated, students are motivated to protect the ego from harm by either avoiding or giving up the task to prevent further unpleasant emotions (64, 65). Changes in appraisal may also happen, whereby students initially activate the growth pathway but redirect themselves to the well-being pathway after experiencing negative triggers or cues for failure, likely due to previous failed attempts. However, students may also revert back to the growth pathway by implementing volitional strategies to re-focus on the learning task at hand (65).

Shaped by the Information Processing Theory (IPT), Winne and Hadwin's model offers a metacognitive conceptualization of SRL (67). The authors postulate that SRL consists of four recursive and loosely sequenced phases: (a) task definition, where the learner generates their perceptions of the task; (b) goal setting and planning, where the learner forges goals and formulates a plan to achieve them; (c) enacting study tactics and strategies, where the learner executes the plan; and (d) metacognitive adapting, where the learner adjusts their strategies and future approach based on their recent task execution (68). Five IPT-influenced processes (referred to using the COPES acronym) are proposed to underpin each phase: conditions, operations, products, evaluations, and standards (67). Conditions are the resources accessible by the learner and constraints specific to the task or environment (e.g., instructions, time). Operations are the various SMART cognitive strategies learners use to manipulate and process information, this includes searching, monitoring, assembling, rehearsing, and translating (60, 69, 70). Products are the outcomes (i.e., information) generated by the aforementioned SMART operations. Evaluations refer to the internal feedback generated from self-monitoring concerning the discrepancy between the products and the Standards; the set of criteria the learner considers to be the optimal outcome (63, 67).

Building on the ideas proposed in earlier SRL models (e.g., Zimmerman, Pintrich, and Winne and Hadwin), Efklides' MASRL model focuses on the regulation of motivation, affect, and cognition, and explains their interrelated yet distinct roles throughout the SRL process (71, 72). The model proposes two levels of SRL functioning. First, the Person-level (i.e., macro-level) depicts the top-down processes in SRL and shows how a learner's goals and personal traits (e.g., self-concept, motivation, affect etc.) can interact with each other to impact decisions relating to engagement with a learning task (72). Conversely, the Task x Person-level (i.e., micro-level) represents the bottom-up regulation of learning where the processes are less characterized by the learner's traits and more by the specific demands of the learning task. This level is more data-driven in nature, as metacognitive monitoring of task processing and performance becomes the driving force behind the learner's actions. The two levels of SRL functioning are theorized to interact and reciprocally inform each other to influence the learner's affect, motivation, and effort regulation in a specific task.

SRL models offer a cohesive overview of the key factors underpinning the process of engagement in the learning context (broadly conceived). We use the term “broadly” to reflect the variation in nomenclature used by different SRL researchers to describe conceptually similar constructs (e.g., self-efficacy and confidence). Through appraising the five SRL models, we deduced a list of recurring factors or skills that have been theorized to influence people's engagement with learning tasks (see Table 1). Whereas some SRL factors are common across all models (e.g., motivation), others are more unique to specific frameworks, due to the slight differences in factor emphasized by each author (63, 72). Nevertheless, there is a general consensus that higher levels across these SRL variables can facilitate and sustain effortful engagement throughout one's learning journey, and in turn, promote positive learning outcomes (56, 61, 64, 68, 72). Thus, we henceforth refer to these variables collectively as “drivers” of engagement. Conversely, lower levels across these drivers are viewed as potential threats to engagement and can hinder the learning process.

As summarized in Table 1, several key SRL drivers have already been considered in digital health research, both in empirical studies looking at predictors of user engagement, and in existing theories of digital health. For instance, five of these drivers including motivation, self-efficacy, outcome (treatment) expectations, social support, and goal setting have been associated with positive engagement outcomes in a number of DHI studies (4, 6, 13, 17, 27). Moreover, drivers like task interest, (positive) emotions, and task value have been hypothesized as potential facilitators or attributes of engagement in various conceptual frameworks of DHIs (11, 31, 32), albeit lack the evidence base to confirm their influence on DHI usage. Despite the potential overlap in variables explored across these disciplines, a key point of difference is how some of these drivers are conceptualized and studied with respect to their potential role in influencing engagement behaviors. This is evident in drivers like metacognition and self-control, which have seldom been addressed or sufficiently considered in digital health research, though their closest comparators are concepts such as self-monitoring, self-evaluation, and goal setting.

SRL researchers describe learners as metacognitively active when they effortfully observe, reflect, and modify their behaviors to improve learning performance and, through this process, may be expected to observe positive progress towards their goals (68). Superficially, this may seem comparable to self-monitoring and goal setting as understood within the digital health literature, however there are important—though complementary—points of differences. In the DHI context, users are typically encouraged to learn about the antecedents and consequences of problematic behaviors in order to intervene earlier in the onset of these problematic behaviors (73). The learning emphasized in this DHI context is about monitoring symptoms, whereas in SRL, the focus is on monitoring one's learning, a higher order cognitive task which may inform whether the user has appropriately understood the treatment content and can apply it effectively within their daily lives. More specifically, the SRL focus is about being or becoming metacognitively aware of one's progress in acquiring and enacting new knowledge and skills from a learning task.

Adopting the SRL perspective would allow us to apply a metacognitive lens to how key skills are utilized in DHIs, by also asking individuals to reflect on their comprehension and attempted implementation of the content received. Quizzes and self-assessment may be deployed to evaluate learning in DHIs (22), yet SRL approaches would go a step further in offering learning strategies to enhance understanding in cases where knowledge and/or enactment gaps are evident. More specifically, digital health users would be encouraged to reflect on whether they (a) understood the intervention module they have just engaged with, (b) felt confident with putting the newly acquired skills into practice, and (c) getting them to reflect on their attempts at applying these skills. This focus on self-awareness is central to understanding one's strengths and weaknesses with learning intervention skills and could help users identify which modules require further engagement to facilitate their primary goal of symptom improvement.

Similarly, self-control is another variable mentioned in both bodies of research, although with notably different foci. In the SRL literature, self-control has been viewed as an individual's discipline to engage in and sustain goal-directed behavior (74, 75). This is an effortful process that requires a set of self-management strategies (e.g., time-management, self-instruction, and self-consequences) to help the learner maintain concentration and interest with their learning tasks (56, 57). On the other hand, self-control in DHIs is typically about minimizing harmful behaviors, often in-the-moment. For example, DHIs may focus on identifying when an urge is present and learning how to control and override temptations. Integrating SRL theory to how common drivers are understood in the digital health literature would encourage the field to explore the role of self-control in a new angle. In particular, investigating its impact on successful learning and completion of intervention content. Though seldom acknowledged from this perspective, exercising self-control (and effective time management) to learn the information received is relevant, and likely essential, for treatment engagement. This is because individuals with greater self-discipline are theoretically more inclined to use the necessary strategies that will help sustain their motivation to learn successive intervention content.

In addition to these nuances in how specific concepts are understood, the interrelations between certain constructs are more strongly articulated in one discipline than the other. For instance, motivation in digital health is often referred to as an individual's readiness to change, or intention to engage in and complete treatment (76, 77). However, SRL models suggest that this concept is more complex, and characterize motivation as a dimension of self-regulation influenced by multiple other drivers (e.g., self-efficacy beliefs, goal-directed behaviors, and task interest) throughout the learning process (35, 78). Thus, adopting the user-as-learner perspective may bring more attention to the inter-relations between multiple drivers and highlight possible mediating or moderating pathways for increased engagement with DHIs (further discussed in Theme 2).

Hence, the two lessons from Theme 1 are as follows. First, there are a number of potentially relevant SRL drivers that have been insufficiently considered or directly tested in the digital health context. This, therefore, suggests the need for broadening the range of predictors that we measure in studies investigating engagement with DHIs. Second, we should also think about appraising and conceptualizing these drivers in a new light. That is, rather than limiting our understanding of these constructs to the DHI context, we should consider how they are defined in different disciplines, particularly in bodies of work with theoretical models that can more comprehensively explain the interrelations between key drivers. A greater understanding of the potential interactions and mediating pathways can suggest new opportunities for intervention to enhance user engagement and guide the prioritization of what to intervene upon.

The second key theme across the SRL models is the notion that learning is an evolving and dynamic process (79). That is, the constructs that drive engagement are not constant, and instead can wax and wane over time. In turn, increases or decreases in these drivers can also affect levels of learner engagement. Using Boekaert's model (64) as an example; a student's appraisal for a learning task can change when there is a shift in emotional experience during task performance. For instance, challenging tasks initially perceived as easy may elicit feelings of hopelessness after several failed attempts. However, the student can also re-appraise the situation using emotion regulation strategies to deescalate those negative emotions that hinder effort investment for the learning task (64, 65). Two of the five SRL models (56, 68) also suggest that engagement itself may provide a feedback loop for the drivers by enhancing motivation when one feels growing competence from successful task completion, or diminishing motivation in response to difficulties or failures. In broader terms, the outcome of engagement may also have an effect on its drivers, which in turn, can reinforce more or less of this positive outcome. This indicates that there are complex interactions occurring between different SRL variables over time (72), which can determine the degree to which an individual seeks to initiate successive cycles of learning (59).

This theme offers three important lessons with notable implications for the digital health context. The first, is that if these drivers of engagement do fluctuate over time as SRL research suggests, then when we measure them in the user journey matters. To date, much of the empirical work in the digital health field investigating predictors of engagement have been largely limited to baseline data, particularly in studies measuring drivers like motivation and self-efficacy (6, 13, 27, 76, 77). This is despite the large evidence base from related disciplines showing that motivation for treatment and behavior change can vary considerably over time (80–84). Relying on baseline predictors alone may be a critical oversight, as new users can often present with an inflated attitude during the onboarding phase. Their initial motivation, however, may diminish when faced with unexpected challenges (e.g., difficulties understanding psychoeducational content), and thus, may not reliably predict engagement patterns later in the user journey. Therefore, it is essential to recognize that those factors driving initial uptake may differ to those factors sustaining engagement and eventual symptom improvement. An important takeaway for digital health researchers is to consider measuring these drivers not only at baseline, but also during the intervention. Exploring the relative contributors of engagement at different milestones in the user journey would serve to test this possibility if certain drivers are more important at particular times, and if time of measurement impact predictive ability.

The second point of relevance for digital health research is that we need to better understand the complex relationships between the different drivers and engagement itself. As briefly discussed in Theme 1, there are potential moderating and mediating pathways that need to be considered. Ignoring how the different variables inter-relate more broadly (e.g., looking at one variable alone without the interaction of the moderator) can result in an inaccurate estimate of the true effect of each driver on engagement. To articulate the complex interplay between different variables, we describe the potential impact of self-efficacy and its proposed correlates on digital health engagement. A user's self-efficacy beliefs about their ability to learn new intervention skills can determine not only their motivation, but also their goal setting behaviors, both of which can influence one's efforts to self-regulate their engagement with the intervention task (78). Furthermore, this engagement or SRL of content received (dependent on the user's initial experiences) can form a positive or negative feedback loop that sustains or diminishes their ongoing self-confidence and motivation to persist with successive treatment content. Thus, to advance our understanding of people's likelihood of engagement, it is imperative to build upon existing models of digital health, to better define these complex—and potentially bi-directional—relationships between different variables.

Finally, if we can understand the temporality of engagement and its drivers, as well as how they relate to each other throughout the user journey, we can identify more opportunities to intervene. Importantly, this could also highlight the need for different intervention approaches at different stages to impact specific levers and bolster treatment engagement. To put it simply, digital health developers should consider implementing targeted interventions at specific points throughout the self-guided program to enhance key drivers of engagement where necessary. Efforts to increase engagement across the user journey could support users, particularly those with lower SRL skills, to complete the recommended psychoeducational modules and facilitate symptom improvement.

The third key theme emerging from these common models of SRL is the recognition that people significantly differ in their capability for SRL, but also emphasizing that this capability for SRL skills is malleable and teachable. Learners are viewed as active participants in their learning journey (46, 56, 61, 65, 68, 72), meaning that with effortful attention and action, SRL skills can be developed and improved over time. To ground this discussion, there is a rich evidence base in the education literature showing that people substantially differ in their SRL profiles, particularly in online learning environments (47–50). Despite being cognizant of the general academic skills (e.g., time management, planning, motivation) required to successfully complete a self-guided syllabus, the consistent pattern across many online courses is low completion and high dropout rates (50). This may, at least in part, reflect the wide range of SRL capacities of people who self-select into digital learning environments, showing that not all learners have the ability or skillset to successfully complete an online course with minimal guidance.

Consideration of these individual differences in people's SRL capacity is potentially relevant in the digital health context because not all users have the required abilities or skillset at time of commencement. Nonetheless, there is an ongoing and misguided assumption that those who sign up for DHIs will know how to effectively manage and direct their learning of the content received. As a result, there is typically no direct attempt to (a) assess people's current SRL capacity, and (b) enhance these skills where necessary. For these reasons, many DHIs are not well-equipped or designed to support users who start out with low levels of SRL. This raises serious practical implications for those relying on DHIs as their primary mode of treatment (due to geographical, financial, or time restraints), as many first-time users may find intervention tasks difficult or overwhelming without the support of a clinician. Difficulties engaging with the treatment content may in turn, diminish people's motivation to persist in their user journey and complete the full program as recommended. Using SRL as a guiding framework, we believe it may be essential for those who enroll into DHIs to (a) be aware of their current learning abilities, and (b) effortfully enhance their SRL skills to drive ongoing engagement with the intervention material. Integrating an SRL perspective into the design phase may open up new avenues to overcome issues with sustained engagement. For instance, the development of support material to teach users the necessary skills to complete their DHIs and relieve unwanted symptoms.