Digital health interventions (DHIs) show vast potential in supporting patients and health care systems with the globally increasing prevalence and economic costs of noncommunicable diseases (NCDs), which is the leading causes of death and disability worldwide [1,2]. More specifically, mobile health (mHealth) apps are now considered accessible and scalable solutions to promoting behavior change among patients, improving health outcomes, and reducing health care costs [3-5]. Correspondingly, the number of available mHealth apps has continuously grown to >300.000, with approximately 200 new mHealth apps released each day [2,6].

However, despite increasing evidence and availability, mHealth apps are subject to significant dropout rates, with a substantial proportion of users not adhering to them as intended [7,8]. Recent research has shown that up to 80% of all participants in mHealth interventions only engage at a minimum level, do not log into the mHealth app more than once, and do not consistently use the app in the long term [9]. Another study examining mHealth app use in more extensive real-world settings reported low retention rates, with only 3.9% of participants using mHealth apps for >15 days [10]. The reported low adherence and high attrition levels further highlight the necessity of developing more effective models, best practices, and interventions [8,11].

As nonadherence relative to intended use jeopardizes treatment success and, thus, might lead to an increased number of hospitalizations, it is considered a fundamental concern in the development of mHealth apps [8,12-15]. However, the scientific body of literature lacks concise conceptualizations and measures for the intended use of mHealth apps, whereas intervention components and factors influencing adherence remain to be explored [13,16]. Following previous studies, we define adherence as “the degree to which the user followed the program as it was designed,” which can be paraphrased as “adherence relative to the intended use” [13,17,18].

With smartphone apps being the primary intervention component, adherence relative to the intended use is principally informed by user acceptance and the use of information technology [19]. Previous research underscores the necessity of mHealth apps that must be first accepted and used in an intended way to then achieve a desired health behavior change [3,19]. Correspondingly, previous research has identified factors affecting the uptake of and engagement with health and well-being smartphone apps [11,20-23]. Many of these strategies and factors, such as well-designed reminders, self-monitoring features, and embedded health professional support, have been applied across various health domains [11,20,21]. Some of these factors, such as reminders, can be further applied as retention methods and strategies for cohort studies in general and may thus extend the scope of DHIs [24,25].

Identifying the factors that influence adherence relative to intended use may support and extend these findings. Given previous research and their relation to technology use and acceptance, we can assume that these factors may be not only generalizable across various health behavior domains but also be applicable to DHIs using alternative information technologies. To our knowledge, no systematic review has been conducted on the factors influencing adherence to mHealth apps designed to prevent or manage NCDs. Furthermore, to the best of our knowledge, no review has previously explored the quantifying of adherence to assess qualitatively identified factors.

Preventing intervention dropouts and thus increasing the effectiveness of mHealth apps requires an understanding of the factors that act as barriers to or facilitators of intervention adherence. This review aimed to identify factors influencing adherence relative to the intended use of mHealth apps, which may help stakeholders better plan, develop, and evaluate mHealth apps. To help readers navigate through the identified factors, we further categorized them into intervention-related factors that app developers can potentially improve upon through product changes (eg, the inclusion of certain app features) and patient-related factors that are hardly adjustable (eg, user characteristics). These factors were separated into their potential positive or negative influences on adherence.

In the absence of a universally agreed-upon approach to measuring adherence to mHealth interventions, we exploratively derived an adherence score as the ratio between the intended and actual use of each study to describe adherence quantitatively and consistently. The primary aim of the resulting adherence score was to quantitatively assess the findings from the qualitative extraction of factors influencing adherence. As the intended use varies substantially across different mHealth apps, we extracted the intended use for each included mHealth app individually. We then compared the intended use with the actual use reported in the corresponding study. To the best of our knowledge, this exploratory approach of a quantified adherence score has not been applied previously.

In summary, this review aimed to answer the following research questions:

This review was conducted according to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses; Multimedia Appendix 1). A review protocol was submitted to the Federal Office of Public Health of the Swiss Confederation on October 7, 2020, but was not publicly registered.

The electronic databases Embase (including MEDLINE and PubMed), Web of Science, Scopus, and ACM Digital Library were searched using a predefined search strategy that included search terms related to mHealth apps, app use, and study design (Multimedia Appendix 2). The search terms were customized for each electronic database, and if the respective database allowed it, the corresponding Medical Subject Heading terms or topics were also integrated. Articles published in English between June 2007 (release of the iPhone) and December 2020, which focused on adult populations, were included. Studies that focused on communicable diseases were excluded. The inclusion and exclusion criteria listed in Textbox 1 were used to identify relevant articles.

The selection of publications was conducted in several steps (Figure 1). First, potentially relevant publications were identified by searching the literature databases. After excluding duplicates, titles and abstracts were independently reviewed by 3 researchers (SH, RJ, and AMR) according to the inclusion and exclusion criteria listed in Textbox 1. Disagreements were resolved through discussion. In a second screening step, the full texts of relevant articles were independently reviewed by 4 researchers (SH, RJ, AMR, and JLM) concerning the fulfillment of the inclusion and exclusion criteria. Discrepancies were again resolved through discussions. The web-based program Covidence (Veritas Health Innovation) was used for the entire process. After applying the search strategies outlined in Multimedia Appendix 2, the resulting database reference lists were imported into the Covidence database. The following Covidence features were used in the process: duplicate removal, title and abstract screening, full-text review, and export of PRISMA flowchart.

The following information was extracted for each included study: general study characteristics, study population characteristics, intervention characteristics, factors influencing adherence relative to intended use, and information on app use (Multimedia Appendix 3 [20,26-123]).

General study characteristics included the title, first author, year of publication, journal name, country, study design, and health domain.

Study population characteristics comprised age, gender, type of population (clinical or general population), type of disease, and number of study participants.

Intervention characteristics were app name, smartphone operating system (universal, Android only, or iOS only), type of mHealth app offered (publicly available or research apps), app developer (private company or nonprofit organization), level of personal support (no personal support during the intervention or continuous personal support), external monetary incentives and their value in US dollars, intervention duration in days, and effectiveness of the intervention in terms of health outcomes.

Factors influencing adherence relative to intended use were extracted and characterized as intervention-related factors (factors that developers can potentially improve through product changes) or patient-related factors (factors that can hardly be influenced by app developers, such as user characteristics). These factors were further categorized based on their positive or negative influence on adherence.

Information on app use comprised intended app use, actual app use, the number of intended intervention interactions, and interaction frequency (eg, daily or weekly intended use). The adherence score was defined as adherence relative to the intended use and was derived as the quantified ratio of intended use to actual use.

As a first step, the identified studies were categorized based on the mHealth app they investigated as follows: apps targeting NCD self-management (including the four main NCDs: asthma, diabetes, cardiovascular disease [CVD], and cancer), mental health disorders (anxiety and depression), substance use disorders (alcohol and tobacco), and behavioral risk factors (nutrition, physical activity, and weight loss). The categories were then further refined into the following health domains: NCD self-management (asthma management, diabetes management, CVD management, cancer management, and medication adherence), mental health (anxiety, depression, and multidisciplinary and others), substance use (alcohol, tobacco, and multidisciplinary and others), nutrition, physical activity, weight loss, multicomponent lifestyle interventions, and other NCDs.

In the second step, intervention-related and patient-related factors that were outlined in the studies as barriers to or facilitators of adherence were qualitatively evaluated, summarized within the health domain, and categorized based on their positive or negative influence.

In the third step, the adherence score was derived for each study focusing directly on a specific mHealth app (97/99, 98%) and was calculated as the ratio of intended use to actual use. The mean adherence scores were calculated for each health domain.

In the fourth step, correlations of adherence scores with other extracted variables were examined, and where possible, the qualitative results from step 2 were quantitatively compared for each health domain. Quantitative analyses were performed using SPSS Statistics (version 27; IBM Corp). Correlations of adherence scores with continuous variables (eg, the average age of study participants) were calculated using the Pearson correlation. Correlations with ordinal variables (eg, level of personal support) were calculated using the Spearman correlation.

Finally, a list of universally relevant factors with recommendations for the development and evaluation of mHealth apps was developed.

The search of electronic databases was performed on January 3, 2021, and yielded 2862 articles. After excluding duplicates, 70.72% (2024/2862) of publications remained for the title and abstract screening. Subsequently, the full texts of 17.34% (351/2024) of articles were examined. Of the 351 studies, 99 (28.2%) were finally included in the data synthesis. Figure 1 visualizes the selection process and reasons for exclusion.

In total, 99 studies were included in this review. Of these 99 studies, 2 (2%) were systematic reviews, and 97 (98%) evaluated specific mHealth apps. Randomized controlled trials were the most frequent study type (35/97, 36%), followed by pilot trials (31/97, 32%), pilot randomized controlled trials (12/97, 12%), cohort studies, mixed methods studies, and observational studies (each 5/97, 5%). Most of the studies were conducted in North America (37/97, 38%), followed by Europe (34/97, 35%), Australia (15/97, 15%), and Asia (10/97, 10%). Of the 97 studies, 90 (93%) were published within the past 5 years: 46 (47%) in 2020, 12 (12%) in 2019, 12 (12%) in 2018, 9 (9%) in 2017, 8 (8%) in 2016, 2 (2%) in 2015, 4 (4%) in 2014, and 4 (4%) in 2013. The mean intervention duration was 111.4 (SD 132; range 7-730) days, with 27% (26/97) of studies lasting 1 to 4 weeks, 46% (46/97) of studies lasting between 1 and 3 months, 18% (17/97) of studies lasting between 3 and 12 months, and 9% (7/97) of studies lasting longer than a year. In 32% (31/97) of studies, monetary incentives were provided to the participants as compensation. The mean derived incentive value was US $105.42 (SD US $18.65; range US $7 to US $430).

The total number of participants in the included studies evaluating specific mHealth apps was 72,046. The mean number of study participants was 750.5 (SD 2800.7; range 9-19,233). Of the 96 studies reporting exact participants numbers, 65 (68%) had <100 participants, 21 (22%) had 100 to 1000 participants, and only 10 (10%) studies had >1000 study participants. In several studies, most study participants had a pre-existing condition (82/97, 85%), with mental health conditions being the most prevalent (21/97, 22%), followed by obesity and being overweight (15/97, 15%), substance abuse (9/97, 9%), cancer (7/97, 7%), diabetes (5/97, 5%), CVD (5/97, 5%), and sleep disorder (4/97, 4%). In 15% (15/97) of studies, most participants were healthy. The overall mean age was 44.6 (SD 12.9; range 19.9-86) years, and the mean percentage of women was 62% (SD 22.8%; range 0%-100%).

Of the 97 reviewed apps, 50 (51%) were available for both iOS and Android. The remaining apps were exclusively available on either iOS or Android platforms (both 17/97, 18%). The authors of 13% (13/98) of studies did not clearly outline on which platforms the apps were distributed. Of the 97 reviewed apps, 47 (48%) were publicly available, whereas 52 (54%) were exclusively available to study participants. Approximately 38% (37/97) of apps were developed by private commercial companies (eg, software companies), and 64% (62/97) of apps were developed by nonprofit organizations (eg, academic institutes). Of the studies that clearly outlined their study procedure, 34% (32/93) included personal contact with health personnel during the study as an intervention component. In comparison, 66% (61/93) of the apps provided only personal support in the app onboarding phase. Of the 54 studies evaluating the app’s effectiveness on a primary outcome, the authors of 34 (63%) studies highlighted their app as effective, and the authors of 20 (37%) studies highlighted their app as ineffective. The most common explanation of intended use, according to the authors, or derived from information on intervention design, was daily tracking (eg, daily diary entries; 36/97, 36%), followed by activity completion (eg, completion of a certain amount of coaching modules; 19/97, 20%), daily use (eg, daily log-in; 17/97, 18%), daily activity completion (6/97, 6%), weekly tracking (5/97, 5%), weekly use (4/97, 4%), activity completion+daily tracking (3/97, 3%), weekly use time (eg, using the app 1 hour per day; 2/97, 2%), prolonged use (eg, no inactivity for >2 weeks; 2/97, 2%), and biweekly tracking (1/97, 1%). Approximately 95% (92/96) of studies reported data on actual use based on objective app use data, and 4% (4/96) of studies reported data based on qualitative feedback from users. The mean adherence score across all interventions was 56.0% (SD 24.4%; range 2.6%-96.0%). The mean number of interactions within the study period amounted to 90.1 (SD 145.9; range 1-730) interactions. Of the 97 apps, 14 (14%) apps were intended for ≥2 daily interactions with the app, 53 (55%) apps were intended for 1 daily interaction, and 17 (18%) were intended for weekly interactions; in 13 (13%) apps, users only had to use the app once a month to be considered adherent.

As displayed in Table 1, the included studies evaluating specific mHealth apps were categorized into the following health domains based on the individual app intervention focus: NCD self-management (17/97, 18%); mental health (20/97, 21%); substance use (9/97, 9%); nutrition (7/97, 7%); physical activity (6/97, 6%); weight loss (9/97, 9%); multicomponent lifestyle interventions (8/97, 8%); and mindfulness, including breathing and meditation interventions (9/97, 9%). Studies categorized in the domains of NCD self-management, mental health, and substance use were further subcategorized to report intervention and patient-related factors influencing intervention adherence at a more granular level. The studies (17/97, 18%) targeting NCD self-management were subcategorized into diabetes management (6/17, 35%), cancer management (5/17, 29%), respiratory disease management (3/17, 18%), CVD management (2/17, 12%), and medication adherence (1/17, 6%). Studies categorized in the mental health domain (20/97, 21%) were further divided into apps focusing on anxiety (2/20, 10%), depression (9/20, 45%), and multidisciplinary and other (9/20, 45%). The latter subdomain included other mental health problems such as bipolar disorders or combinations of various mental health problems. Studies in the substance use domain were further separated into apps addressing alcohol (2/9, 22%), tobacco (6/9, 67%), or a mix of various substances (1/9, 11%). Another 12% (12/97) of studies, which was a heterogeneous group targeting NCDs other than diabetes, cancer, CVD, respiratory disease, or medication adherence, were clustered into other NCDs (eg, intestinal and renal disease, insomnia, pain, venous leg ulcers, and dyslipidemia).

As outlined in Table 2, the mean number of participants was highest for studies that focused on substance use (2337.6, SD 6344.8; range 9-19,233) and lowest for other NCDs (54.8, SD 48.6; range 15-189), followed by weight loss interventions (73.2, SD 54.4; range 17-176).

As displayed in Table 3, the mean participant age was highest in apps targeting NCD self-management (57.7, SD 7.3; range 45-70.9 years) and lowest for mental health apps (35.9, SD 5.9; range 19.9-46.5 years).

Female populations were generally overrepresented (Table 4), especially in studies conducted on mindfulness interventions (76.7%, SD 20.2%; range 44.8%-100%). Only studies conducted on apps targeting substance use featured more men than women (percentage of women: mean 49.3%, SD 14.1%; range 27.7%-78%).

As outlined in Table 5, studies conducted on apps for weight loss had the most prolonged mean intervention duration (214, SD 216.3; range 65-730 days), and studies on nutrition had the shortest mean intervention duration (52.5, SD 55.7; range 7-172 days).

Table 6 shows distributions of total intended interactions with the apps over the course of the individual studies. The mean number of total intended interactions was highest for apps targeting weight loss (210.5, SD 213.3; range 52-730) and lowest for nutrition apps (31.8, SD 27.8; range 4-82.5).

The distribution of adherence scores by health domain is summarized in Table 7. The mean adherence scores were highest in the domain of other NCDs (69.9%, SD 18.5%; range 33.3%-90.5%), followed by multicomponent lifestyle interventions aimed at changing multiple behaviors simultaneously (61.3%, SD 22.5%; range 32.4%-96%). Apps from the substance use domain had the lowest adherence scores (46.1%, SD 33%; range 9.1%-84%).

Multimedia Appendix 4 [20,26-123] lists the identified intervention-related and patient-related factors with a positive or negative influence on adherence for each health domain in detail. The results per health domain are summarized in the following sections.

One of the two included systematic reviews featured a multi-domain review focusing on uptake and engagement with mHealth apps in various health domains [20]. First, it showed a positive impact of goal setting, reward offerings, complementary web access, coping games, and self-monitoring. In addition, the low cost of the app helped increase acceptance [20].

The external influence of using an app through health care professionals, friends, and family or by reading user reviews was outlined as having a positive influence. Furthermore, community networking and the connection between the app and health professional support had a positive influence [20]. The study also found the following user characteristics to be positively linked to adherence: female gender, aged <44 years, living in urban areas, good educational level, high income, curiosity, higher health literacy, and app awareness [20]. In addition, interactivity, an established routine to use the app, and customization of the app had a positive impact [20]. In contrast, cognitive overload and unmet expectations negatively influenced adherence [20].

To gain further insights into the universal applicability of the results identified in individual health domains, a quantitative analysis was conducted on the total number of primary studies included. The analysis revealed a positive correlation between adherence score and level of personal support during the study period (r_s91=0.199; P=.06). With respect to various user characteristics, the quantitative analysis did not find significant differences in either average age (r=0.105; 94/97, 97%; P=.32) or gender distribution (r=−0.031; 95/97, 97%; P=.77). Furthermore, no significant quantitative differences were found between healthy participants and participants with chronic diseases (r_s95=−0.049; P=.63). A quantitative comparison of studies with monetary incentives and those without such incentives also revealed no significant effect on adherence scores (r_s92=0.000; P=.99). However, the monetary value of the incentive, measured in US dollars, had a significant effect on the adherence score (r=0.465; 30/97, 31%; P=.01). Apps that were only offered in the context of scientific studies had a significantly higher adherence score than those that were publicly available via app stores (r_s95=0.324; P=.001). The quantitative analysis did not find any significant differences between Android and iOS with regard to the adherence score (r_s83=0.019; P=.87). Furthermore, the quantitative analysis showed a higher adherence score for apps developed by private app development companies than for those developed by public institutions or research groups (r_s95=0.164; P=.11). The correlation between adherence score and intervention duration (r=−0.138; 97/97, 100%; P=.18) was negative but positive in relation to the number of intended app interactions per day (r=0.176; 97/97, 100%; P=.09). The comparison of adherence scores and the total number of intended app interactions (r=0.040, 97/97, 100%; P=.70) yielded no significant results. Studies with a higher number of app users had significantly lower adherence scores (r=−0.228; 96/97, 99%; P=.03).

Regarding the first research question, the intervention-related factors described in the following sections were identified most frequently across all health domains, suggesting universal applicability to increase mHealth app adherence relative to the intended use.

Regarding the second research question, the patient-related factors described in the following section were identified.

Regarding the third research question, this review outlined the differences between health domains in terms of adherence scores. The adherence score of all 97 included mHealth apps averaged 56.0% (SD 24.4%), representing a generally higher adherence level to mHealth apps than previous research suggests [9,10,146]. This could be attributed to the fact that some studies excluded participants who did not perform a certain level of activity (eg, downloading the app). Regarding the short intervention periods with a median of 60.8 days, it is questionable whether the included health apps could reach similar adherence scores in more prolonged studies. Adherence scores by health domain were highest for the other NCDs (69.9%, SD 18.5%), a heterogeneous group of mHealth apps targeting less common NCDs such as intestinal and renal disease, insomnia, pain, venous leg ulcers, and dyslipidemia. Multicomponent lifestyle interventions aimed at changing multiple behaviors simultaneously (61.3%, SD 22.5%) had the second-highest adherence scores, whereas apps targeting substance use had the lowest (46.1%, SD 46.1%). The relatively low adherence levels of mHealth apps treating substance use could be explained by the nature of their intervention design, making it difficult to differentiate between nonadherent users and users who stopped using the app after a successful behavior change. Another explanation could be that substance use disorders are comorbid with depressive disorders. Several qualitative findings indicate that symptoms or a diagnosis of depression negatively affect adherence [27,48,50,78,89]. This also aligns with our findings that apps offering depression support had an even lower average adherence score of 45.3% (SD 28.1%).

Another explanation and potential bias for the difference in adherence scores are asymmetric distributions of trials compared with real-world applications within health domains, represented by the mean number of participants (substance use 2337.6, SD 6344.8; multicomponent lifestyle interventions 284.3, SD 531.9; and other NCDs 54.8, SD 48.6). In general, studies with a higher number of participants had significantly lower adherence scores (r=−0.228; 96/97, 99%; P=.03). Furthermore, apps that were only offered in the context of scientific studies had a significantly higher adherence score than those publicly available via app stores (r_s95=0.324; P=.001). Surprisingly, these differences could not be explained by longer study duration (r=−0.138; 97/97, 100%; P=.18) or the number of intended app interactions (r=0.040; 97/97, 100%; P=.70). Thus, our study supports and provides quantitative evidence for previous findings outlining engagement within trials to significantly differ from real-world applications [147-149]. Further real-world longitudinal studies are necessary to explain these differences.

Although calculating an adherence score as the ratio between intended and actual use has several limitations, this exploratory approach enabled the quantification and comparison of adherence across different mHealth apps to a reasonable extent. The quantitative analysis of adherence scores yielded few significant results but fit qualitative findings regarding positive or negative influences on adherence in most instances, which supports the potential validity of the concept.

The results of explorative analysis based on adherence scores should be considered cautiously. Although we did not find a significant correlation between study duration or the number of intended app interactions and the adherence score, it is possible that other factors, such as patient characteristics, could influence adherence. Further research is needed to establish effective adherence measures. By reporting this novel quantified measure of adherence for individual studies and collectively for defined health domains, we hope researchers will test and challenge our approach, potentially developing more effective measures that help us quantify adherence and make it comparable across heterogeneous groups of mHealth apps.

This study had several limitations. The first limitation is the heterogeneity of the included studies. The studies differ in terms of the characteristics of the target populations (eg, type of pre-existing condition, age, gender, education level, comorbidities, employment status, and experience with mobile technologies) and study duration (a few weeks to over a year). In addition, mHealth apps within the studies and their intended use varied significantly. The problem of the undiversified reach of mHealth interventions, predominantly including female and White participants living in high-income countries [150], also accounts for this study. Approximately 90% (87/97) of these studies were conducted in North America, Australia, or Europe. Similarly, women were overrepresented, with an average proportion of 62% (SD 22.8%). Most included studies had a pilot character, with 80% (77/96) of the studies including <200 participants and 73% (71/97) of the studies having a duration of <100 days. Whether individual study results can be replicated in more controlled and longitudinal studies in the future is questionable. In addition, there were differences in terms of additional monetary incentives and the level of personal support complementary to the use of mHealth interventions. Moreover, the mHealth apps investigated in the studies showed substantial heterogeneity in their goals (eg, increasing physical activity and reducing tobacco consumption). Overall, this limited the generalizability of our results.

This limitation was overcome by categorizing the results into different health domains. However, mHealth apps also exhibited key differentiating characteristics within their health domain, such as their stage of development (prototype vs established app), developer (nonprofit vs private company), the use of peripheral devices (eg, a smartwatch for passive data collection), app features (eg, social features such as leaderboards and elements of gamification), or the level of quality (eg, in terms of the user interface, UX, or technical stability). Another limitation is that few studies considered individual intervention components separately and evaluated their effectiveness, which complicates the identification of intervention-related factors. A further limitation of this study is the inclusion of nonrandomized studies, which, on the one hand, allowed a more extensive consideration of objective app use data but, by contrast, precluded conducting a risk of bias assessment.

The calculation of adherence scores as the ratio between intended and actual use also implies some noteworthy drawbacks. The intended use of the mHealth app, which was derived from the study design or study author comments, was also heterogeneous and differed depending on the mHealth app design (eg, tracking daily symptoms or completing a certain number of coaching sessions), interaction frequency (eg, daily or weekly), and interaction duration (weeks to years). Furthermore, the intended use was rarely justified by applying theory, evidence, or rationale, which has also been addressed in previous studies [13]. In some instances, the intended use could be derived from the intervention design; however, many studies had to be excluded as the app’s intended use was unclear or not stated at all. However, it can be positively highlighted that the actual mHealth app use extracted for the adherence score was based on objective app use data in approximately 96% (92/96) of cases.

This study contributes to the scarce scientific evidence on factors influencing adherence to mHealth apps and is the first to derive quantified adherence scores for various health domains to validate qualitative findings and explore adherence benchmarks.

This paper contains various detailed presentations of the central results. The most detailed presentation of adherence factors extracted from individual studies is outlined in Multimedia Appendix 4. We further classify the factors within the defined health domains and report the results collectively. Finally, we discuss the most common factors influencing adherence across all the health domains. As mHealth apps within health domains remain heterogeneous, we encourage readers to always consider information from the corresponding individual studies outlined in Multimedia Appendix 4 when implementing their interventions according to the factors reported in this review.

Our findings indicate that the following intervention-related factors positively influence mHealth app adherence: user-friendly and technically stable app design, customizable push notifications, personalized app content, passive data tracking, an integrated app tutorial, offering the app free of charge, and the integration of personal support into the mHealth app. Furthermore, gamification and social features show promising effects but may be limited to specific health domains. Time delays within the intervention, long intervention durations, low engagement of other participants, and the competitive effects of other mHealth apps were associated with low mHealth app adherence.

Regarding patient-related factors, the following user characteristics were associated with low mHealth app adherence: lack of technical competence, low health literacy, low self-efficacy, low education level, mental health burden, lack of experience with mHealth apps, privacy concerns, low expectations of the app, low trust in health care professionals conducting the intervention, and lack of time on the users’ side. Age, gender, and pre-existing condition were frequently mentioned factors but differed across and sometimes conflicted within the health domains. Furthermore, personal user recruitment appeared to positively influence adherence as opposed to web-based user recruitment.

The adherence score of the 97 included mHealth apps averaged 56.0% (SD 24.4%). Adherence scores were highest for mHealth apps targeting less common NCDs such as intestinal and renal disease, insomnia, pain, venous leg ulcers, and dyslipidemia. Multicomponent lifestyle interventions had the second-highest adherence scores, whereas apps targeting substance use had the lowest. Exploratory analysis of adherence scores revealed quantitative evidence for higher adherence rates within trials than in real-world applications.

Overall, research on the factors that positively or negatively influence adherence to mHealth apps is still limited. The underlying studies often had a pilot character with a short study duration, and the implementation of techniques was inconsistent. As most mHealth apps contain multiple intervention components, causal statements about individual factors are not possible and require more controlled and longitudinal studies in the future. To facilitate future research on mHealth app adherence; researchers should clearly outline and justify the app’s intended use, report objective data on actual use relative to the intended use; and ideally, provide long-term use and retention data. Further research is needed to establish effective adherence measures.