This study did not involve human participants and therefore did not require ethical approval, in accordance with institutional and national guidelines.

This overview of systematic reviews adhered to the PRIOR (Preferred Reporting Items for Overviews of Reviews) guidelines (Checklist 1) and was prospectively registered in the PROSPERO (International Prospective Register of Systematic Reviews; CRD42024604784) [21].

Three deviations were made to the registered protocol (PROSPERO CRD42024604784) prior to data extraction and synthesis to improve coherence and practical use:

A comprehensive search was conducted in MEDLINE, Web of Science, Epistemonikos, and gray literature sources (Social Science Research Network, WorldCat, Health Management Information Consortium [Ovid]) from inception to June 9, 2025. Search terms combined keywords and MeSH related to “chronic conditions” and “mobile.” No date or language limits were applied initially. The strategy was developed with domain experts and 2 senior librarians (Multimedia Appendix 1).

Systematic reviews were eligible if they

Reviews were excluded if they did not distinguish app-based interventions from other digital tools or focused solely on passive monitoring (eg, wearables). Reviews were therefore excluded when the population was defined solely by hypertension, obesity, isolated symptoms, exposures, risk factors, or health indicators, rather than by a diagnosed long-term physical condition within the scope of the review. This was based on World Health Organization (WHO) and Global Burden of Disease risk-factor frameworks reporting raised blood pressure (hypertension) and high BMI (overweight and obesity) as metabolic risk factors or health indicators rather than eligible long-term physical condition populations in their own right [22,23]. Search terms were intentionally broad to maximize sensitivity and capture related risk-factor literature; however, final inclusion was determined during screening based on how each review defined and reported the condition.

Search results were imported into EndNote X7 for deduplication [24]. Two reviewers independently screened titles and abstracts using Covidence, an online software tool to manage screening, extraction, and quality assessment for systematic reviews [25]. During screening, inclusion was limited to English, German, Spanish, or Turkish (the languages fluently spoken by the research team) to ensure accurate appraisal. This allowed us to note how many potentially relevant reviews were excluded for language while maintaining screening reliability. Full texts were retrieved for all records deemed potentially relevant by at least 1 reviewer. Ten percent of the full-text reviews were randomly selected for double screening to assess interrater reliability, yielding a Cohen κ of 0.62, which indicates substantial agreement between authors [26]. Discrepancies were resolved through discussion, and the remaining full-text screening was completed by the first author. Reference lists of included reviews were hand searched.

Data extraction used a standardized form piloted by the research team. The first author extracted all data, with a random 10% independently checked by a second reviewer (GHB), yielding full agreement. The remaining data extraction was completed by the first author.

Quantitative systematic reviews were appraised using the AMSTAR 2 (A Measurement Tool to Assess Systematic Reviews 2) tool and categorized as high, moderate, low, or critically low [27]. Qualitative and mixed methods reviews were appraised using the CASP (Critical Appraisal Skills Programme) qualitative checklist [28]. CASP does not provide a numeric scoring system and instead uses descriptive summaries with justifications. We considered the breadth of “yes” responses, with particular emphasis on core methodological areas, including clarity of aims, appropriateness of qualitative methodology and research design, adequacy of data collection, rigor of data analysis, clarity of findings, and the stated value of the research. We did not apply a prespecified numerical cutoff for most CASP items. Instead, reviews were judged as higher quality when “yes” responses were consistent across these core domains and when any limitations were minor or related to less applicable domains for systematic reviews, such as recruitment strategy. A random 10% were double checked, with full agreement. We considered alternative tools for quality appraisal, including the Joanna Briggs Institute (JBI) checklist for umbrella reviews. However, because our overview included qualitative and mixed methods reviews, we judged CASP to provide more explicit prompts for assessing the interpretive and methodological features of qualitative synthesis. CASP was therefore used as a structured framework to appraise the qualitative synthesis components of the included reviews.

A narrative synthesis was undertaken. Reviews were grouped according to the engagement-related data they reported:

Where definitions were absent, engagement-related terms (eg, usage, adherence, and compliance) were interpreted using the context from primary studies. For consistency in characterizing core mHealth app functionalities, we applied the WHO classification of digital health interventions [29].

Synthesis of modifiable barriers and facilitators offered insight into users’ cognitive and affective responses to mHealth apps, reflecting how perceptions (eg, relevance and trust) and emotions (eg, interest and burden) influenced decisions to engage or disengage. This enabled exploration of subjective influences on engagement.

Modifiable barriers and facilitators of engagement were categorized as intrinsic or extrinsic motives. The intrinsic motives were then systematically mapped onto the 3 basic psychological needs proposed by SDT—autonomy, competence, and relatedness—using established definitions of SDT psychological needs [17].

Mapping was completed independently by 2 reviewers (JB and MMI). Decisions about which barriers and facilitators aligned with each SDT need were informed by existing theoretical and empirical literature [30,31] and refined through iterative discussions within the research team.

The frequency with which themes were mapped to each psychological need was descriptively recorded across reviews and their included primary studies to enhance the transparency of the synthesis rather than to infer effect size or relative importance.

Although the Capability, Opportunity, Motivation-Behavior (COM-B) model was initially considered as an organizing framework [32], preliminary synthesis showed that most reviews described engagement processes in terms of the quality of motivation rather than broader behavioral determinants, such as physical capability or environmental opportunity. Common descriptors included perceived personal relevance and feelings of being supported, which align more closely with SDT’s distinction between intrinsic and extrinsic motivation. Given that sustaining engagement with mHealth apps typically relies on intrinsic, self-endorsed motivation rather than externally regulated behavior [17], SDT provided a more coherent and mechanistic framework for interpreting engagement processes.

Primary study overlap across the included systematic reviews was assessed using a citation matrix. Rows represented unique primary studies, and columns represented included systematic reviews. Overlap was quantified descriptively by identifying primary studies that appeared in more than 1 review and by calculating the corrected covered area (CCA) [33]. As this overview of systematic reviews synthesized data at the systematic review level, primary study overlap was used to inform transparency and interpretation only; no primary study-level outcome data were extracted, reanalyzed, or pooled.

The search identified 5888 records from databases and registers and 744 through citation searching. After removing duplicates, 4684 records including 3940 (84.1%) from databases and registers and 744 (15.9%) from other methods. Overall, 799 full-text articles were assessed for eligibility by 2 independent reviewers, of which 333 (41.7%) were identified through databases and registers and 466 (58.3%) through other methods. The primary reason for exclusion (n=478, 59.9%) was the absence of any mention or reporting of engagement-related data. Ultimately, 19 systematic reviews were included in this overview of systematic reviews [34-52]. A PRIOR (Preferred Reporting Items for Overviews of Reviews) flow diagram is presented in Figure 1.

The 19 systematic reviews comprised 12 (63%) quantitative, 5 (26%) mixed methods, and 2 (11%) qualitative syntheses published between 2015 and 2025. Of the 17 reviews with quantitative data, 3 (18%) conducted a meta-analysis. Of the 7 reviews with qualitative data, 1 (14%) undertook a formal meta-synthesis, while the remainder used thematic or narrative approaches. LTC populations included people living with cancer (n=6, 32%), diabetes (n=4, 21%), and chronic pain (n=2, 11%). Others focused on heart failure (n=1, 5%), stroke (n=1, 5%), chronic kidney disease (n=1, 5%), osteoarthritis (n=1, 5%), Parkinson disease (n=1, 5%), traumatic brain injury (n=1, 5%), and a mixed group of LTCs (n=1, 5%).

The number of primary studies per review ranged from 3 to 30 (total n=262), with sample sizes ranging from 1 to 3606 and mean participant ages typically between 50 and 63 years. Across the 262 primary studies, the three most common study designs were randomized controlled trials (RCTs; n=111, 42.4%), qualitative studies (n=62, 24%), and observational studies (n=33, 13%). Of the 195 primary studies that reported geographic origin, 83 (43%) were conducted in North America, 68 (35%) in Europe, 24 (12%) in Asia, 12 (6%) in Oceania, 4 (2%) in South America, and 4 (2%) in Africa. Table 1 summarizes these characteristics, and Multimedia Appendix 2 provides a more detailed table.

Qualitative and mixed methods reviews were strong overall, with 3 rated “high quality” [37,38,50], 2 rated “moderate-to-high” [44,45], and 2 rated “moderate” [36,46], though reflexivity and analytic transparency were often limited (Table 1).

Quantitative reviews were generally low quality: 6 were rated “critically low” [39,40,42,43,48,49], 5 were rated “low” [34,47,51,52], and 1 was rated “moderate” [41]. None met the criteria for high quality, mainly due to missing protocol registration, limited risk-of-bias assessment, and insufficient reporting of heterogeneity or funding (Table 1). Multimedia Appendices 3 and 4 include the quality appraisal scores for AMSTAR 2 and CASP, respectively.

Across the 19 included systematic reviews, 262 unique primary studies were identified. Primary study overlap was slight [33]. Most primary studies appeared in only one review (n=243, 92.7%), while 19 studies appeared in more than one review (19/262, 7.3%). Of these overlapping studies, 14 appeared in 2 reviews, 4 appeared in 3 reviews, and 1 appeared in 4 reviews. The total number of primary study occurrences across reviews was 287. The CCA was 0.53%, indicating slight overlap. Given the low level of overlap and because synthesis was conducted at the systematic review level, all 19 reviews were retained in the narrative synthesis.

Across the 19 reviews, 98 distinct mHealth apps were reported. The most common app functionality, using the WHO classifications of digital health interventions, was symptom self-monitoring, which was present in 17 (90%) systematic reviews. Targeted delivery of health information, often as educational content tailored to users’ LTCs, was identified in 16 (84%) reviews. Alerts and reminders were reported in 12 (63%) reviews. Active user data capture, primarily for symptom tracking and remote consultations with health care providers, appeared in 8 (42%) reviews. Peer group support features were reported in 5 (26%), while user-to-system feedback mechanisms were present in 2 (11%) reviews. Only 1 (5%) review included apps where users could access their personal health records. Full characteristics are provided in Multimedia Appendix 5.

For clarity, we describe behavioral indicators as microlevel usage metrics (eg, amount, duration, depth, and frequency of use) based on previous literature [10-12].

Of the 19 included systematic reviews, only 5 (26%) explicitly operationalized engagement a priori [36-38,46,50]. Fourteen (74%) reviews reported engagement metrics, but predominantly as microlevel indicators (eg, frequency of logins, duration of use, and number of tasks completed). Terms such as “usage,” “adherence,” and “engagement” were often used interchangeably, with no review operationalizing macrolevel engagement (eg, integration into daily life and alignment with goals). Definitions and thresholds varied widely, precluding meta-analysis. Multimedia Appendix 6 provides a detailed summary of the engagement indicators reported in each review.

Eight of the 19 (42%) reviews assessed links between engagement and health outcomes. Seven of these reviews (moderate-to-high [n=1], moderate [n=1], low [n=4], and critically low [n=1]) reported positive associations between microlevel app use and outcomes such as improved symptom knowledge, health behavior change, or reduced distress [39,41,45,47,48,51,52]. One meta-analysis (critically low quality) found no significant association [40].

Importantly, no review linked macrolevel engagement with health outcomes or if engagement was sufficient to achieve intended health outcomes (ie, effective engagement) [11]. Qualitative reviews provided insight into subjective engagement indicators, such as users’ sense of motivation or identification with the intervention, but no review integrated these with quantitative outcome data. Behavioral and subjective indicators of engagement were reported descriptively in separate strands in the included reviews, with no further analysis to link them.

Seven of the 19 (37%) systematic reviews highlighted nonmodifiable user-specific barriers and facilitators of engagement [36,41,42,45,46,48,52] (Figure 2). A total of 57% (4/7) of reviews were quantitative (moderate [n=1], low [n=1], and critically low quality [n=2]), and 43% (3/7) of reviews were mixed methods (moderate-to-high [n=1] and moderate quality [n=2]).

Findings were mixed, as 2 reviews reported higher engagement among older adults (≥63 years in 1 review; age not specified in the other) [42,46], whereas another reported greater engagement among younger users [36]. Similarly, disease-related characteristics such as duration of diagnosis had mixed results, with both newly diagnosed individuals and those living with a condition for over 5 years demonstrating active engagement across different studies [36,42]. Findings related to disease severity or multimorbidity were also mixed, with one review reporting lower engagement [41] and another reporting higher engagement, potentially due to increased pain burden [40].

Thirteen of the 19 (68%) reviews reported modifiable barriers and facilitators of engagement with mHealth apps for LTCs [34,36-38,42-46,48-50,52]. Seven reviews were quantitative (moderate [n=1], low [n=2], and critically low quality [n=4]), 4 were mixed methods reviews (moderate-to-high [n=2] and moderate quality [n=2]), and 2 were qualitative reviews (high quality [n=2]).

Reviews were categorized into 9 overarching themes (Figure 3). Six themes mapped onto intrinsic motives: app design, app functionality, trust, time burden, baseline technological readiness, and baseline readiness for health behavior change; and 3 themes mapped onto extrinsic motives: awareness of the app, availability of the digital tool, and cost. The reported extrinsic motives function primarily as structural preconditions for engagement because they enable initial uptake but do not, by themselves, sustain engagement. A full thematic breakdown with illustrative quotations and frequency of reporting is provided in Multimedia Appendix 7.

For intrinsic motivation, barriers and facilitators supporting competence were the most frequently reported, appearing across 63% (12/19) of systematic reviews and 89.3% (234/262) of distinct primary studies. These included app design and functionality features that enabled users to feel capable of using the app effectively, such as ease of use, clear and accessible language, visual aids, real-time feedback, and symptom monitoring. Engagement was undermined when competence was challenged by technical problems, high data entry burden, information overload, unmet expectations, low digital literacy, or concerns that app use could exacerbate psychological distress related to the LTC.

Barriers and facilitators mapping onto autonomy were reported across 58% (11/19) of systematic reviews and 59.5% (156/262) of distinct primary studies. Autonomy-related barriers and facilitators reflected users’ appraisal of relevance, control, and fit with daily life, including readiness for behavior change, perceived need for the app, ability to integrate app use into daily routines, goal-setting functionality, personalization, and privacy considerations.

Factors supporting relatedness were less frequently reported overall, with 53% (10/19) of systematic reviews and 40.5% (106/262) of primary studies, and focused on interpersonal connection and trust. Engagement facilitators included interaction with health care professionals and other users, endorsement by clinicians, secure data sharing with caregivers during emergencies, and integration with electronic medical records. Barriers to relatedness included lack of emotional connection to the app and stigmatizing or overly intrusive reminders.

The mapping of modifiable barriers and facilitators to intrinsic and extrinsic motivation informed the development of design recommendations for future evidence-based mHealth apps for people with LTCs (Table 2). Design implications were generated for all themes identified in the synthesis, irrespective of how frequently they were reported across primary studies. This approach reflects the conceptual aim of this overview of systematic reviews to understand mechanisms of engagement rather than to rank predictors by prevalence. Less frequently reported themes may still represent decisive barriers or facilitators for different user groups, LTCs, or contexts of use, and excluding them would risk overlooking design considerations essential for equitable, scalable, and effective engagement.

To our knowledge, this overview of systematic reviews is the first to offer theory-informed, mechanism-oriented guidance for understanding and designing mHealth app–based interventions that support engagement among adults with LTCs. By mapping modifiable barriers and facilitators of engagement onto SDT, we move beyond descriptive lists of factors to explain why these influences matter and how they may support or undermine intrinsic motivation.

This approach aligns closely with the Medical Research Council framework for developing and evaluating complex interventions, which emphasizes the value of theory for specifying mechanisms of action, strengthening intervention design, and enhancing reproducibility. Applying SDT as an overarching motivational theory provides a coherent structure for identifying modifiable targets, linking app design and features to psychological processes that shape engagement, and treating engagement as an intentional, theorized component of an intervention rather than a by-product. The resulting recommendations (Table 2) therefore offer a workable, mechanism-based framework for integrating psychological needs into both the development and evaluation of mHealth interventions.

Across the 19 included systematic reviews, three consistent findings emerged: (1) conceptual ambiguity in engagement definitions, (2) limited and low-quality evidence on engagement-outcome associations, and (3) converging evidence on patterns in modifiable factors influencing engagement.

First, engagement was rarely defined explicitly. Reviews often used terms including “engagement,” “usage,” and “adherence” interchangeably and reported only microlevel behavioral indicators (eg, number of logins or tasks completed). No review operationalized macrolevel engagement (eg, sustained integration into daily life or alignment with goals). Consequently, the current evidence base largely reflects observed usage, limiting insight into whether engagement was clinically relevant to the person using the app.

Second, although 7 of the 8 reviews assessing engagement-outcome links reported positive associations, these findings were drawn from heterogeneous operationalizations of engagement and were often based on critically low-quality quantitative syntheses. Qualitative reviews contributed insight into subjective indicators of engagement, but no review integrated them with behavioral indicators. This methodological divide reflects broader measurement immaturity in engagement research rather than limitations of individual reviews.

Third, when modifiable barriers and facilitators to engagement were mapped to motivational processes, the vast majority were intrinsic—related to autonomy, competence, or relatedness—rather than extrinsic or structural. Most barriers and facilitators reflected whether users experienced the app as personally relevant, manageable, intuitive, trustworthy, and aligned with their own goals and capabilities. Only a small subset of factors was purely extrinsic (eg, awareness, connectivity, and cost), which is still prominent as this is the minimum needed to enable engagement. However, once extrinsic preconditions are met (eg, users can download the app), engagement largely depends on intrinsic motivational processes. Hence, mHealth apps that aim to replace or reduce clinician input must intentionally support these psychological needs.

SDT provided a clear mechanism through which to interpret these patterns, enabling us to identify specific motivational targets and derive design recommendations to guide the development of interventions that are more capable of sustaining engagement without ongoing clinician reinforcement.

Our findings align with longstanding critiques that digital health research often emphasizes “how much” people engage through behavioral indicators while neglecting the quality of engagement and/or the “why” (ie, subjective [cognitive and affective] influences on engagement) [53-55]. Persistent heterogeneity in engagement terminology and measurement constrains the development of cumulative evidence and makes it difficult to establish whether disengagement reflects unmet needs of the user, symptom improvement, or success in achieving the intervention’s goal [10,56-58]. Without distinguishing between these possibilities, establishing cutoffs such as the “minimum effective dose” of engagement also remains elusive.

The positive engagement-outcome associations, such as improved symptom knowledge, health behavior change, or reduced distress, mirror earlier evidence, including Donkin et al [7] findings, but these associations remain difficult to interpret given inconsistent operationalizations and low review quality. Similarly, although persuasive design features (eg, tailoring and feedback) have been shown to improve engagement [8], none of the included reviews in this overview of systematic reviews analyzed user-level engagement or linked design features to specific mechanisms. Collectively, these findings highlight a need for standardized behavioral metrics and improved reporting of how specific app features are linked to engagement.

The implications include the evaluation of existing mHealth apps and the development of new mHealth app interventions.

Definitions and measures of behavioral engagement indicators must be standardized. Future research should adopt a validated multidimensional operationalization of engagement that moves beyond single indicators such as logins. We recommend reporting the amount (eg, number of sessions), duration time spent (eg, minutes spent in sessions), depth (eg, number of tasks completed in the same session), and frequency (eg, number of logins) [10,11]. Complementary use of validated instruments such as the Twente Engagement With eHealth Technologies Scale (TWEETS) [59] can capture both behavioral and subjective dimensions of engagement, including interest, attention, and enjoyment. This combined approach is essential for distinguishing mere usage from outcome-relevant engagement.

As outlined in Table 2, the motivational mechanisms identified, specifically those supporting autonomy, competence, and relatedness, translate into clear, actionable design recommendations. These recommendations connect psychological needs directly to design choices, supporting mechanism-informed intervention development.

Importantly, many autonomy- and competence-related barriers, such as digital exclusion, cost, and low digital literacy, are structural challenges requiring systemic responses. The features summarized in Table 2 include practical solutions such as offering offline or low-bandwidth app versions, SMS text message–based alternatives for users without smartphones, and device-loan schemes to expand access. To support competence among digitally excluded users, guided onboarding, clear tutorials, and accessible interface design (eg, large fonts, high-contrast modes, and multilingual options) are recommended. Moreover, partnerships with community health workers, primary care providers, and advocacy organizations (relatedness-supportive features) can improve trust and contextual relevance for the LTC populations that the mHealth apps aim to serve [60,61]. These measures directly address engagement gaps and help prevent mHealth innovations from exacerbating health inequalities [62-64]. Further work is needed to quantitatively confirm if some of these recommendations do, in fact, make a difference.

A key strength of this overview of systematic reviews is its theory-informed synthesis across quantitative, qualitative, and mixed methods reviews, enabling a multidimensional understanding of engagement with mHealth apps. Unlike prior syntheses focused narrowly on behavioral indicators, this review synthesized the barriers and facilitators of engagement within a motivational framework, which offers insight into subjective indicators. Applying SDT provided a coherent framework for mapping barriers and facilitators onto psychological needs and for generating design-oriented recommendations. Methodological rigor was supported through prospective PROSPERO (International Prospective Register of Systematic Reviews) registration and adherence to PRIOR guidelines.

Several limitations should be acknowledged. Review quality varied; quantitative reviews were often low quality, while qualitative and mixed methods reviews were stronger. In addition, the CASP qualitative checklist was developed for the appraisal of primary qualitative studies rather than review-level evidence syntheses. However, we considered it appropriate as a structured framework for assessing the qualitative and mixed methods components of included reviews because it provides methodological depth in domains such as reflexivity, rigor of analysis, credibility of findings, and the value of the research. This approach is also consistent with a previous overview of systematic reviews that used CASP as part of the quality appraisal of systematic reviews of qualitative studies [65]. Key populations, including older adults, people with low digital literacy, and those in rural or underserved settings, were underrepresented, limiting generalizability. Publication bias may also have excluded studies reporting no engagement-outcome associations. Applying SDT introduced some interpretive subjectivity, as most reviews were not theoretically framed; however, mappings were independently cross-checked to enhance consistency. These limitations reduce the certainty of the conclusions but highlight clear priorities for future research.

mHealth apps are a promising approach to supporting self-management of LTCs, but their effectiveness partly relies on users’ engagement with the intervention and their ability to apply its therapeutic content in daily life. This overview of systematic reviews highlights conceptual and methodological gaps in engagement research and demonstrates the value of applying SDT to understand why engagement occurs and how it can be supported. The SDT-informed design guidance provides a foundation for developing more engaging, equitable, and user-centered mHealth app interventions. Embedding these principles in future evaluation and development can strengthen both user experience and clinical impact and contribute to more consistent, theory-driven progress in digital health.

The graphical abstract for the article is provided in Multimedia Appendix 8.