Potential associations between behavior change techniques and engagement with mobile health apps: a systematic review

Introduction Lack of engagement is a common challenge for digital health interventions. To achieve their potential, it is necessary to understand how best to support users’ engagement with interventions and target health behaviors. The aim of this systematic review was to identify the behavioral theories and behavior change techniques being incorporated into mobile health apps and how they are associated with the different components of engagement. Methods The review was structured using the PRISMA and PICOS frameworks and searched six databases in July 2022: PubMed, Embase, CINAHL, APA PsycArticles, ScienceDirect, and Web of Science. Risk of bias was evaluated using the Cochrane Collaboration Risk of Bias 2 and the Mixed Methods Appraisal Tools. Analysis A descriptive analysis provided an overview of study and app characteristics and evidence for potential associations between Behavior Change Techniques (BCTs) and engagement was examined. Results The final analysis included 28 studies. Six BCTs were repeatedly associated with user engagement: goal setting, self-monitoring of behavior, feedback on behavior, prompts/cues, rewards, and social support. There was insufficient data reported to examine associations with specific components of engagement, but the analysis indicated that the different components were being captured by various measures. Conclusion This review provides further evidence supporting the use of common BCTs in mobile health apps. To enable developers to leverage BCTs and other app features to optimize engagement in specific contexts and individual characteristics, we need a better understanding of how BCTs are associated with different components of engagement. Systematic review registration https://www.crd.york.ac.uk/prospero/, identifier CRD42022312596.


Introduction
Mobile health applications have the potential to empower people to improve their health behaviors and self-manage health conditions (Forman et al., 2016;Moller et al., 2017;Digital Implementation Investment Guide [DIIG], 2020).They are widely available, with a large global market (Grand View Research, 2021), and provide a means of delivering far-reaching behavioral interventions.To have a significant impact on behavioral and health outcomes, mobile health apps need to be able to support sufficient engagement to achieve the aims of the intervention they are delivering (Yardley et al., 2016;Cole-Lewis et al., 2019).Previous research has established an association between engagement with digital interventions and their impact on intended outcomes (Perski et al., 2017;Grady et al., 2018;Mclaughlin et al., 2021), demonstrating the importance of a certain degree of engagement for the efficacy of an intervention.Maintaining engagement is a common challenge for mobile health apps (Birnbaum et al., 2015;Yeager and Benight, 2018;Baumel et al., 2019;Meyerowitz-Katz et al., 2020;Pratap et al., 2020;Torous et al., 2020a) and is a potential reason why evidence of their effectiveness remains mixed (Dounavi and Tsoumani, 2019;Ng et al., 2019;Milne-Ives et al., 2020;Moshe et al., 2021).
The conceptualization of engagement has recently become a focus in digital health research, resulting in several theories and frameworks (O'Brien, 2016;Yardley et al., 2016;Perski et al., 2017;Cole-Lewis et al., 2019;Kelders et al., 2020).It is generally accepted to be multi-faceted, with components relating to dimension (i.e., affective, cognitive, and behavioral) and scale [i.e., small-scale engagement with specific components of the digital intervention ("micro") and larger-scale engagement with target health behaviors ("macro")] (Milne- Ives et al., 2022).A variety of strategies have been proposed and incorporated into mobile health apps to help improve user engagement, including design features and behavior change techniques (Garnett et al., 2015;Floryan et al., 2020;Iribarren et al., 2021).
Although evidence is still developing, previous research has emphasized the importance of incorporating behavioral theory into digital health intervention design (Gourlan et al., 2016;Moller et al., 2017;Roberts et al., 2017;Klonoff, 2019;Taj et al., 2019).Behavioral theory, and behavior change techniques (a taxonomy of the smallest intervention components that can support behavior change, e.g., goal setting, self-monitoring, rewards), can provide evidence-based predictions to optimize context-specific intervention design and potential impact (Michie et al., 2013;Morrison, 2015;Moller et al., 2017).Impact on engagement, however, is often measured using only behavioral measures of app use (Torous et al., 2020b;Mclaughlin et al., 2021), which cannot capture the full picture of users' multi-faceted engagement.For this reason, our understanding of how different components of engagement are associated with app features, engagement strategies, and outcomes is incomplete.
Despite the growing presence of Behavior Change Techniques (BCTs) (Michie et al., 2013) in mobile health apps, there has been limited investigation of how they can help support engagement specifically.Our preliminary review of the literature (Milne-Ives et al., 2022) identified several systematic reviews that have explored questions around this topic (O'Connor et al., 2016;Szinay et al., 2020;Wei et al., 2020;Borghouts et al., 2021), but none that analyzed how BCTs can best support the various components of engagement.The purpose of this review was to investigate how specific BCTs are associated with the different components of engagement (affective, cognitive, behavioral, micro, and macro) to provide insights for the development of mobile health apps.To examine this, the primary objective was to identify BCTs being incorporated in the design and development of mobile health apps and their hypothesized or evidenced associations with engagement and its various components.Learnings from this review will enable future studies to empirically test causal relationships between specific BCTs and specific components of engagement to further improve our understanding of their associations.

Study design
The Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) framework (Page et al., 2021) was used to structure this review (Supplementary Appendix 1).The protocol was prospectively registered on PROSPERO (CRD42022312596) and published before searches began (Milne-Ives et al., 2022).

Eligibility criteria
The eligibility criteria and search strategy were based on the PICOS framework (Table 1; Richardson et al., 1995;Counsell, 1997).The protocol scope was broadened to include development studies to capture research that hypothesized potential associations.Studies published before 2011 were excluded because we wanted to provide an overview of recent evidence [digital technology evolves rapidly (Steinhubl et al., 2015)] and because the conceptualization of engagement as a multi-faceted construct emerged in digital health relatively recently (Kelders et al., 2020).Behavior change Behavior Control OR Psychological Theory "behavior change techniques" or "behaviour change techniques" or "BCT" or "behavior change technique" or "behaviour change technique" or "behavioral change strategies" or "behavioural change strategies" or "behavior change wheel" or "behaviour change wheel" or "behavioral theory" or "behavioural theory" or "behavior change theory" or "behaviour change theory" or "health behaviour change" or "behavior change" or "behaviour change" or "digital behavior change intervention" or "digital behaviour change intervention" or "DBCI" or "behaviour change intervention"

Data extraction
Data was extracted by one reviewer into a predetermined form (see Box 1).

Data analysis
A descriptive analysis summarized study and intervention characteristics.A quantitative meta-analysis of effect was not possible due to outcome variety, so extracted evidence was analyzed by one author to map potential associations between BCTs and engagement.Evidence for potential associations was divided into three categories: evidenced, hypothesized, or inferred.Associations were inferred if studies did not report app features as BCTs or did not directly examine associations between features and engagement.Features were coded using the Behaviour Change Technique (BCT) Taxonomy v1 (Michie et al., 2013) as generically as possible (e.g., "non-specific" reward) to ensure that they could be analyzed in line with the rest of the included studies.To examine components of engagement, a basic codebook was created using definitions from the literature (Supplementary Appendix 4) to explore how they were assessed and associated with BCTs.

Risk of bias and quality assessment
The quality appraisal was conducted using the Cochrane Collaboration Risk of Bias 2 (RoB 2) tool for randomized controlled trials (RCTs) (Higgins et al., 2011;The Cochrane Collaboration, 2021) and the Mixed-Methods Appraisal Tool (MMAT) (Hong et al., 2018).The MMAT tool was used instead of the ROBINS-I tool (Sterne et al., 2016) because it could be applied to the wide range of study types included in the review.

Included studies
In total, 21,185 articles were retrieved and 28 were determined to be eligible for inclusion.The reasons for exclusion in the full-text review stage are included in the PRISMA flow diagram (Figure 1).One of the included references was a Clinicaltrials.govregistration from which a full text article was identified.

Study characteristics
The characteristics of the 28 included studies are summarized in Supplementary Appendix 5. Qualitative study designs were most common (8/28, 29%).Half of the studies (14/28) described development or early-stage testing.Reported study durations ranged from 2 (Duff et al., 2018) to 83 weeks (Delmas and Kohli, 2020).Sample sizes ranged from 7 (Garnett et al., 2015) to 2,740 (Delmas and Kohli, 2020).Almost three quarters of the studies (18/25) had fewer than 100 participants; the two studies with the largest sample sizes both retrospectively sampled real world app users (Cueto et al., 2019;Delmas and Kohli, 2020).Two thirds of the studies (17/25) focused on adults, a quarter (6/25) on children and adolescents, and two on adolescents and young adults.

Engagement measures
The most common method of measuring engagement (11/28 studies) was through app usage data (e.g., frequency, duration).Other methods included surveys, qualitative methods, appcollected data (active or passive), and data from synced devices (e.g., activity trackers and smart pill bottle caps, Table 5).Many of the studies that used qualitative methods were focused on usercentered design or early-stage usability or feasibility testing.

Associations between behavior change techniques and engagement
Table 6 summarizes the authors' analysis of the potential associations between BCTs and engagement, in the context of the available descriptions and evidence.Methods in the "evidenced" group of studies included comparisons between different mobile conditions (e.g., with and without social support) (Du et al., 2016;Nuijten et al., 2021Nuijten et al., , 2022)), rankings of the effectiveness of particular BCTs for engagement by experts or users (Garnett et al., 2015;Tang et al., 2015;Szinay et al., 2021), and modeling (Delmas and Kohli, 2020;Lin and Mâsse, 2021).Overall, the "evidenced" studies were of moderate quality: of the randomized trials, two had some risk of bias (Du et al., 2016;Nuijten et al., 2022) and one had high risk of bias (Nuijten et al., 2021), but the studies evaluated using the MMAT met most or all of the quality criteria (Garnett et al., 2015;Tang et al., 2015;Delmas and Kohli, 2020;Lin and Mâsse, 2021;Szinay et al., 2021).The six most common BCTs associated with engagement (based on the summed number of studies from the three categories) were social support, goal setting, feedback, prompts/cues, self-monitoring, and rewards (Table 7).

Exploratory analysis of engagement components
Only two studies explicitly discussed multi-faceted conceptualizations of engagement (Nurmi et al., 2020;Szinay et al., 2021).21 studies included measures of engagement with sufficient description to code various components, most of which captured some form of behavioral engagement (16/21), either with the intervention (12/21) or with the target behavior (11/21).All of the studies that examined engagement with the intervention analyzed app use data, sometimes supplemented by self-report (5/12).Health behavior data was captured through self-report via app or questionnaire in all of the studies but one, which used the app to track step count (Höchsmann et al., 2019).
Data related to cognitive engagement also came primarily from qualitative feedback (Tang et al., 2015;Buman et al., 2016;Hales et al., 2016;Duff et al., 2018;Whiteley et al., 2019;Nurmi et al., 2020;Szinay et al., 2021;Thornton et al., 2021), although a few questionnaires included at least one item relating to interest (EQR User Engagement subscale (Baumel and Kane, 2018), uMARS engagement section (Davis and Ellis, 2019), IMI "interest/enjoyment" subscale (Höchsmann et al., 2019), and the Client Satisfaction Questionnaire (Whiteley et al., 2019).Qualitative results identified a dichotomous association between cognitive effort and engagement; some feedback suggested that engagement would be negatively affected by a time-consuming app with a high cognitive load and that a basic, easy-to-use app was desirable (Szinay et al., 2021), while other findings indicated that lots of content could help sustain engagement and disliked apps that were too basic (Tang et al., 2015;Thornton et al., 2021).

Risk of bias and quality assessment
The risk of bias of the six randomized trials was evaluated using the Cochrane Collaboration RoB 2 tool (Higgins et al., 2011;The Cochrane Collaboration, 2021; a The core of the BCW is the COM-B model, but they are reported separately here in line with what the studies reported.b Physical activity included movement behaviors, sedentary behaviors, and sleep.
An occurrence was added each time a behavioral theory was reported in a study for each of the target behaviors supported by that app; therefore, columns and rows do not necessarily add up to the number of studies, as several studies included multiple target behaviors or behavioral theories.
"measurement of the outcome" (because most of the studies used objective measures of engagement) and "missing outcome data" (largely because engagement was the main outcome, so drop-out and attrition were data).Bias in the selection of the reported result could not be determined for the majority of studies, as pre-published protocols or registrations were not found.The Mixed-Methods Appraisal Tool (MMAT) (Hong et al., 2018) was used to assess the quality of the remaining studies (Supplementary Appendix 6).Of the study types, the qualitative studies met the most criteria; however, as the individual quality criteria are different for each study, a direct comparison is inappropriate.Common issues with the mixed-methods studies assessed were a lack of explicit justification for the mixed-methods design and a lack of detailed interpretation of the integration of results.For the six quantitative studies, half did not have participants representative of the target population.Across all of the study types, there was a common issue regarding the item about "clear research questions"; two-thirds of the studies (13/19) reported aims, but not research questions.

Principal findings
In total, 28 studies that developed or examined mobile apps for health behavior change were examined to investigate how behaviour change techniques (BCTs) could influence different components of engagement.Although a variety of methods were used and various components of engagement could be identified in the data captured, most studies did not purposefully examine these different components.Different

Comparisons with existing literature
The BCTs identified as having evidence of associations with engagement are in line with previous findings.The mixed influence of social support on engagement has also been observed by previous research (Tong and Laranjo, 2018;Szinay et al., 2020) and could potentially be explained by a difference between hypothetical preferences and practical experiences (Szinay et al., 2020).The implementation of the social support BCT could also be a factor; the studies that found a positive influence of social support on engagement were team-based (Du et al., 2016;Nuijten et al., 2021), provided practical social support, such as connection with healthcare professionals, or offered optional opportunities to interact with other users to get peer support, compete, or be Evidenced in study • Comparison Team and Solo mobile app conditions found Team condition had significantly higher compliance (M = 0.49, SD = 0.35) than Solo condition (M = 0.30, SD = 0.39; t 50.82 = 1.94, p = 0.05) • Team condition participants were 66% more likely to engage longer with the app than Solo condition participants Some concerns (RoB) a Overall RoB2 assessments are reported as "high, " "low, " or "some concerns" and overall MMAT assessments are reported as * (0-5); three studies (a content analysis and two development papers) could not be assessed and are noted as "N/A."b The authors deemed there was insufficient evidence of a potential association with engagement to justify including these BCTs in Table 7. c These BCTs were not included in the apps evaluated, but were deemed by the authors to have potential associations with engagement based on the papers' reported findings and included in Table 7.

Strengths and limitations
The broad scope is both a strength and a limitation; the inclusive search strategy reduced the likelihood of missing relevant studies, but necessitated automated screening.To mitigate the risk of screening out relevant studies, we used a set of six articles identified in preliminary searches and determined to be eligible to check the screening.After each pass in EndNote, if any of the six studies were missing, the search terms were revised and the pass re-run to ensure we were not incorrectly eliminating relevant papers.Although the search terms were developed with a second author, the review was not independently executed by two reviewers, which would have reduced the risk of bias in The shading of the colours reflects the number of studies in which the BCT was associated with engagement (darker colours = more studies).
the analysis and increased the credibility of the conclusions.It was not possible to have a second reviewer due to time and resource constraints.
The diversity of outcomes examined by the studies prohibited a quantitative meta-analysis to examine effect.Additionally, the studies examined a wide variety of health apps, aimed at various populations and behaviors.In including this variety of studies, the review is limited in its ability to explore potential contextspecific associations.Contextual factors such as behavior, age, culture, app type, and outcomes could influence engagement in ways that it was not possible to capture with the current sample.
The study was also limited to apps that had been evaluated in scientific investigations; this method risks excluding popular publicly-available apps that have not been previously evaluated.
A review of the BCTs included in such apps and how they are associated with engagement could provide additional insights to inform app design.This study instead contributes to the literature by analyzing the body of evidence to hypothesize associations between BCTs and engagement for future testing.The analysis was limited by the available reported data and descriptions of the studies' apps and measures.The lack of explicit analyses of engagement components in the included studies meant the review relied on author coding and inference to identify potential associations.As such, this review represents a starting point for further investigation of engagement and an identification of current limitations in digital health methodology and reporting.

Future research and recommendations
One of the challenges of our analysis was that there were no clear and comprehensive definitions of affective, cognitive, and behavioral engagement in the literature.Constructs commonly associated with affective engagement included motivation, positive and negative feelings, and sometimes aesthetic pleasure, while cognitive engagement was described using terms such as interest, attention, thought, and challenge (O'Brien, 2016;Perski et al., 2017;Kelders et al., 2020).Recent discussions in the literature have highlighted the importance of using mixed-methods designs to enable these elements to be assessed (Yardley et al., 2016;Perski et al., 2017;Short et al., 2018), which is supported by our findings.As more studies intentionally examine the different components of engagement and how they influence each other, our understanding of how to use particular BCTs to provide tailored support for different individuals, in different contexts, and at different times in their process of engagement with the intervention will improve.
There was not sufficient evidence available in the included studies to posit associations between BCTs and specific components of engagement.However, the exploratory analysis enables some hypotheses to be suggested for further investigation.Several of the BCTs with potential associations with engagement -goal setting, self-monitoring, feedback, social support, rewards, and prompts and cues -were identified as being "motivating" and "fun" for users.This evidence could indicate associations between these BCTs and affective and cognitive engagement, although further context about how and why they were motivating it would be needed to determine whether the associations were primarily affective, cognitive, or both.A theoretical examination of those components compared to the Behaviour Change Technique (BCT) taxonomy can provide some additional insight.
The Human Behavior Change Project's Theory and Techniques Tool maps the BCT taxonomy with mechanisms of action (MoA) (Connell Bohlen et al., 2018;Carey et al., 2019;Johnston et al., 2021).A couple of the MoAs associated with the BCTs clearly relate to a component of engagement; for example, the definitions of MoAs with links to goal setting (intention and goals) are theoretically similar to the concept of cognitive engagement as they relate to "conscious decisions" and "mental representations, " as is one of the MoAs linked with prompts/cues (memory, attention, and decision processes).This suggests that, for these two BCTs in particular, their impact on motivation is more likely to be related to cognitive than affective engagement.However, many of the linked MoAs do not provide any additional clarity about how the BCTs relate to different components of engagement.For instance, one of the MoAs for self-monitoring (behavioral regulation) is defined as "behavioral, cognitive and/or emotional skills for managing or changing behavior."Likewise, the MoA environmental context/resources, which is linked with social support and prompts/cues, is defined as "aspects of a person's situation or environment that discourage or encourage the behavior, " which could conceivably include cognitive, affective, or behavioral aspects.In digital health, these aspects can include the availability, accessibility, and interactivity of an app (potentially affecting behavioral engagement), the tone of the app (potentially affecting affective engagement), and the personalization to a users' needs (potentially affecting cognitive engagement) (Szinay et al., 2020(Szinay et al., , 2021)).
Recent studies have identified a lack of justification and guidance for the operationalization of BCTs (Fulton et al., 2018;Lemke and de Vries, 2021).The evidence indicates that the way BCTs such as social support are incorporated into interventions could affect how they are associated with engagement.It will be important to examine if there are certain operationalizations that better support engagement or whether its influence is related primarily to individual characteristics.Understanding when and how they have a positive or negative impact would enable interventions to be tailored to the specific context in which they will be implemented or dynamically to individuals, to emphasize specific BCTs and app features that they are most likely to respond best to.
For app developers, the key takeaways from the evidence gathered in this review is that the inclusion of BCTs such goal setting, feedback, prompts/cues, self-monitoring, and reward could help support users' engagement with mobile health apps.
Social support also has the potential to support engagement, but can have unintended negative effects, depending on how it is incorporated.Based on the studies reviewed, our recommendation for developers considering incorporating social support is to carefully consider how it will be operationalized and to make that feature optional for users.Practical social support, such as linking patients with healthcare providers, is likely to support engagement and team-based competition seems to be less likely than individual competition to generate negative emotions in users.In view of app design, there appears to be a balance needed to obtain the optimal level of cognitive load.Data from participants indicated a preference for a reduced cognitive load -limited complexity, less data input, limited number of features, not time-consuming (Abuhamdeh et al., 2015;Szinay et al., 2021) -while other data and motivation theory suggests that the app needed to be interesting enough to keep users' attention (Tang et al., 2015;Thornton et al., 2021) and provide the right level of challenge to achieve optimal motivation (Abuhamdeh et al., 2015).Therefore, apps should be easy to learn and efficient to navigate (Wei et al., 2020), but contain sufficient or varying content to sustain users' interest.Finally, it is likely to be important for engagement to choose BCTs and other app features carefully to fit the context and the aim of the intervention, rather than incorporating as many BCTs as possible.

Conclusion
The purpose of this systematic review was to incorporate theoretical conceptualizations of engagement and synthesize evidence of how mobile health apps and their BCTs are associated with different components of user engagement.Six BCTs were identified as having the most evidence of a potential association with engagement with digital behavior change interventions: goal setting, self-monitoring of behavior, feedback on behavior, prompts/cues, rewards, and social support.An exploratory analysis of the engagement components identified key areas for future research, including further theoretical clarity of engagement components, how less researched BCTs might be theoretically linked to engagement components, and how BCTs can be operationalized in app design to best support affective, cognitive, and behavioral engagement.Recommendations for app development deriving from the findings are the inclusion and further investigation of the 6 BCTs identified to support engagement and the importance of identifying the right level of cognitive involvement for the target population to keep the app interesting without being a burden.This review contributes to the body of evidence for the influence of specific BCTs on engagement by identifying a gap between theoretical conceptualizations of engagement and its evaluation in mobile health apps and hypothesizing potential associations for future empirical investigation.Greater clarity in the definitions of different components of engagement, and further development and incorporation of means of measuring specific types of engagement, could enable future research to examine how different components of engagement are related to specific BCTs and behavioral and health outcomes.This knowledge could inform the design and tailoring of mobile health apps to improve engagement with the intervention and target behaviors.
InterventionMobile health apps developed using at least one behavior change theory, framework, or technique to target at least one health behavior (including but not limited to: physical activity, diet, sedentary behavior, substance use, etc).

TABLE 2
Search terms.
EngagementTreatment Adherence and Compliance OR Patient Participation OR Patient Compliance Engag* OR "user engagement" OR immersion OR flow OR involvement OR presence OR adherence OR attrition OR compliance OR maintenance OR acceptability OR satisfaction Digital health interventions Telemedicine OR Mobile Applications OR Internet-Based Intervention "mHealth" OR "eHealth" OR "mobile health" OR telehealth OR mobile OR phone OR smartphone OR cell OR digital OR "app" OR "apps" OR application* OR digital OR web OR internet

Table 8 and
Supplementary Appendix 6).The domains with the lowest risk of bias were a SCT, Social Cognitive Theory; COM-B, Capability, Opportunity, Motivation -Behavior model; BCW, Behavior Change Wheel; TPB, Theory of Planned Behavior; BCTs, Behavior Change Techniques; SDT, Self-Determination Theory; TDF, Theoretical Domains Framework; IMB model, Information, Motivation, and Behavioral Skills Model.b Discussed in paper, but unclear if incorporated in app design.

TABLE 4
Behavioral theories used in apps targeting particular health behaviors.
a BCTs to support engagement with digital health interventions: social support, goal setting, feedback, prompts/cues, selfmonitoring, and rewards.A lack of explicit reporting of BCTs and preplanned analyses of engagement components limited our ability to identify potential associations between BCTs and specific components of engagement.
10.3389/fpsyg.2023.1227443TABLE6Analysis of potential associations between BCTs and engagement and the evidence for the association.

TABLE 7
Associations between behaviour change techniques (BCTs) and engagement identified in the studies.

TABLE 8
Risk of Bias 2 assessment of randomized studies.