We searched titles, abstracts, and keywords of peer-reviewed articles published from January 1, 2010 to June 1, 2015. A comprehensive bibliographic search was conducted through Medline, PreMedline, PsycINFO, Embase, Health Technology Assessment, Education Resource Information Center (ERIC), and Cumulative Index to Nursing and Allied Health Literature (CINAHL) by using key search terms, such as mobile application, mobile app, smartphone, and information technology, and using the qualifier “behavior change” (see Multimedia Appendix 1 for the full search strategy). In addition, the Journal of Medical Internet Research (JMIR) was hand-searched for the same period on the journal’s website.

We included articles if they were published in English, in a peer-reviewed journal, after 2010, targeted at an adult population, and presented results from the analysis of primary or secondary outcomes. We only included randomized controlled trials (RCTs), case-control studies, and cohort studies that were designed for app-based interventions to improve any health-related behaviors. The exclusion criteria were quasi-experimental studies or qualitative studies; text message, Web, email, Twitter, social network services, or personal digital assistant-based health interventions; absence of behavior change indicators or outcomes; an app was not the primary intervention tool; and articles focused mostly on app design and development. Conference abstracts, protocol papers, reviews, editorials, and commentary were also excluded.

The initial search returned 3353 articles: 1405 in Medline, 356 in Embase, 791 in CINAHL, 344 in PsycINFO, 296 in ERIC, 71 in PreMedline, 37 in Health Technology Assessment, and 53 in JMIR. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Figure 1), we eliminated duplicates and screened the titles and abstracts, which narrowed the results to 868 articles. A full-text review reduced the sample to 88 articles; after applying the exclusion criteria, we further narrowed that to 55 articles, of which 32 were quasi-experiment studies or an app was not the primary intervention tool and they were subsequently excluded. This left a final sample of 23 articles to be included for the review. Studies excluded during the full-text review stage and their reasons for exclusion are listed in Multimedia Appendix 2. Data extraction from identified articles was completed by authors JZ and ML with disagreements resolved through discussion with author BF.

The following information was extracted and analyzed from each of the 23 articles: authors, research location and year of publication, study type, sample size, intervention duration, intervention tools with behavior change mechanisms, target behavior change, control group variables, measurement of behavior change indicators, and reported outcomes and significance levels. The search was kept wide with no specific target health behaviors in the search strategy. Based on the health behaviors identified, the articles were organized into 11 themes: mental health improvement or alcohol addiction, physical activity, weight control and diet control, medication management, lifestyle improvement, diabetes management, sun protection, hypertension management, cardiac rehabilitation, smoking cessation, family planning, and pain management. Apps were deemed effective if they reported quantitative measures of successful behavior change [21]. The characteristics of the studies meeting inclusion criteria are summarized in Multimedia Appendix 3. For trial sample size, large samples usually meant at least 100 participants in each randomized group, moderate sample size was between 60 and 100 participants in each group, and small sample size was less than 60 participants in each group [22,23]. According to the Cochrane Handbook for Systematic Reviews of Interventions [22], studies with retention over 80% are classified as having low attrition and studies with retention between 60% and 79% are classified as having moderate attrition. Influencing factors of completing app trials were evaluated to understand determinants of retention rates; features of effective apps were also examined.

Behavior change mechanisms, including the use of theory, techniques, and therapies, were extracted from each study. Behavior change theories applied by the included studies were noted [24]. Behavior change techniques used in the interventions were coded according to Abraham and Michie’s taxonomy of behavior change techniques (BCTs) [25]. Mental health or alcohol addiction apps were most likely to be based on a specific behavior therapy (see Multimedia Appendix 3).

All included studies were appraised using the Cochrane Risk of Bias Assessment Tool [22]. This requires assessing each study against a set of seven criteria: random sequence generation, allocation concealment, blinding of participants, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias. Low risk of bias for completeness of follow-up was defined by a cut-off of 80% complete follow-up [22] (see Multimedia Appendix 4).

The 23 articles analyzed in this review were organized under 11 themes according to target behaviors. Of these, 7 targeted mental health or alcohol addiction; 4 targeted increasing physical activity, weight control, and diet control; 3 aimed to improve medication management; 2 involved an intervention for lifestyle improvement; and 1 study was identified in each of the following themes: diabetes management, sun protection, hypertension management, cardiac rehabilitation, smoking cessation, family planning, and pain management. All studies were conducted in high-income countries, 10 in the United States, 3 in Australia, 2 in the United Kingdom and Sweden, respectively, and 1 each in South Korea, Italy, New Zealand, Spain, Switzerland, and the Netherlands. As defined by the inclusion criteria, all included studies used RCT design, except one case-control study [26]. There were 6 large sample studies [10,11,27-30]. A three-arm RCT study had the largest sample size (N=1932) [28], whereas 14 studies had a small sample size (ie, <60 participants per group) [9,26,31-42]. Others had moderate sample sizes. The intervention duration ranged between 3 weeks [36] and 8 months [27]. Of all the apps, only 6 studies evaluated commercially available apps [10,11,29,30,40,41] and 1 study tested a publicly downloadable app developed by the Swedish government [28]; other apps were not publicly available. Only one app, from Switzerland, was designed for people older than age 65 years [40]. All apps were designed in the English language, with the exception of one Spanish app [38]. In total,19 included in this analysis had more than 65% retention in the intervention group with a high of 100% [31,35,36] and a low of 60% [32]. Three studies did not report retention rate [26,34,37] (see Multimedia Appendix 3).

Across the 23 studies, 3 mechanisms were employed to promote behavior change: behavior change theories, BCTs, and specific behavioral therapies. In total, 6 studies reported using behavior change theories to underpin their app interventions [9,10,27-29,36]. The most commonly used theory was the theory of planned behavior [9,10,28], followed by social cognitive theory [29,36]. The top 3 most commonly used BCTs were self-monitoring (12 interventions) [10,27-29,38- 45], feedback provided on performance (8 interventions) [11,28,29,36,37,41- 43], and tailoring messages (8 interventions) [10,26,30,36,38,41- 43]. Apps related to mental health or alcohol addiction were usually based on a specific behavioral therapy, such as motivational enhancement therapy [35], behavioral activation therapy [33], and cognitive behavior therapy [34] (see Multimedia Appendix 3).

The quality of reviewed studies is summarized in Multimedia Appendix 4. All 23 studies had some kind of risk of bias according to the Cochrane Risk of Bias Assessment Tool. Only 9 articles adequately reported random sequence generation. A computer random number generator was used in 2 studies [9,27]. The process of minimization, used to make small groups similar, was described in 3 studies [30,43,45]. A total of 11 studies explicitly stated that allocation was concealed (eg, using sequentially numbered opaque, sealed envelopes, central allocation) [9,27-29,31-33,41-44]. Participants were blinded in 1 study, but the assessors had full knowledge of the assignments [36]. Only 1 RCT study of a smoking concession app was double-blinded to the 196 participants and assessors [45]. Assessors were blinded in another 4 studies [9,28,35,38]. Due to the nature of using apps, subject blinding was often not possible across the interventions. The remaining studies were either not blinded or information was not explicitly provided in the reporting. We used a cut-off of 80% completion for low risk of bias for completeness of follow-up [22]. A total of 10 studies were at low risk of attrition bias [9-11,31,35,36,38,42,43,45]. Only 3 studies did not outline the statistical analyses or dropout rate [26,34,37]. With regard to bias of selective outcome reporting, insufficient information was present in 1 study [36] and a high risk of bias was present in 5 studies [30,37,38,40,44]. The quality assessment of the reviewed studies is presented in Multimedia Appendix 4. The Cochrane risk of bias summary is reported in Figure 2.

Mobile phone apps are seen as a potential low-cost way to deliver health interventions to both general and at-risk populations. Many such apps exist; however, rigorous research to test their effectiveness and acceptability is lacking. There were 7 publicly available apps that were used in the reviewed studies [10,11,28-30,40,41]. Despite their apparent popularity, public and commercial apps have not been comprehensively evaluated to date; they are currently being used without a thorough understanding of their associated risks and benefits [53]. There is a gap between app concept, delivery, and translation into health behavior change.

The Cochrane Risk of Bias Assessment is a good tool to assess the quality of intervention trials. However, in our findings, the “blinding of participants and personnel” was poor; only one study [45] was double-blinded due to the unique nature of app interventions. The quality of mHealth evidence reporting could be improved through the use of recently published guidelines to aid better understanding and synthesizing findings. The Consolidated Standards for Reporting Trials (CONSORT) provides a 22-item checklist for reporting Web-based and mHealth RCTs [54]. The mHealth Evidence Reporting and Assessment (mERA) checklist could also aid quality improvement of mHealth intervention reporting [55]. Additionally, the Transparent Reporting of Evaluations with Nonrandomized Designs (TREND) statement could assist to improve the reporting quality of nonrandomized evaluations of public health interventions [56]. In this review, only 4 studies described “blinding of outcome assessment” [9,28,35,38]. It might be possible to blind outcome assessors, those doing data analysis, or those administering co-interventions, which is one of the 22 essential items recommended in the CONSORT checklist [54]. It is important for researchers to adopt these guidelines vigilantly for better reporting and communication of research results.

One of the primary benefits of apps is their potential for incredibly high reach. With mobile phone use reaching near saturation among some populations, particularly young adults, and the high rates of consumer acceptability, app effectiveness research must also consider total app reach. This aspect of health behavior change apps has not been assessed, with most studies being exceptionally small in scale. Apps that offer even a small health benefit could still be a valuable public health intervention if the population-level reach is high enough. But, encouragingly, we identified some registered large-scale clinical trial protocols of app-based interventions, suggesting that the current limited scientific evidence may be eased in coming years [57-60].

All identified studies were conducted in high-income countries, which could be partly due to our search criteria limiting publications in English only. However, it is also possible that a significant demand for app research on health behavior change in lower- and middle-income countries is being neglected. The burden of NCDs, such as heart disease, diabetes, cancer, and mental disorders, is high in low- and middle-income countries and is predicted to grow [4]. Mobile phones have great potential to reach populations that previously had restricted access to interventions or health care information [61]. Apps have also created new opportunities and possibilities to reach populations who were largely unreachable via traditional health care channels [62]. mHealth interventions have a positive impact on some chronic diseases in developing countries [63] and text messaging has been recognized as a successful tool to improve behavior change outcomes [13,15]. In comparison with text messaging only, mobile phone apps offer more active engagement in health care and improved convenience at substantially lower cost. However, the current evidence base for the use of app-based interventions in developing countries remains small [64]. The widespread adoption of mobile phones highlights a significant opportunity to impact health behaviors globally, particularly in low- and middle-income countries.

Limitations of this review are worth noting. The search terms are restricted to health behavior change, and we focused mostly on medicine- and health science-related databases, which may have excluded publications in other areas. Although iPhone and Android app stores debuted in June 2007 [65], they have experienced exponential growth in popularity since 2010; some relevant articles published before January 2010 could have been missed. The included studies were all conducted in high-income countries where the health care systems are different from many low- and middle-income countries, which limits the ability to draw generalizable conclusions [66]. The inclusion of studies targeted at the adult population could also confine interpretations about whether app-based interventions can influence behavior change among younger users.

To our knowledge, no previous study has completed a comprehensive thematic literature review of mobile phone apps for health behavior change. Although a majority of the studies reviewed reported statistically significant effects in targeted behavior change, adequately powered and relatively longer duration RCTs are still required to determine the effectiveness of app-based interventions. Further research should focus on conducting evaluation research in low- and middle-income countries. Moreover, these results highlight the need for better reporting of health-related app interventions. Collaborations between researchers, HCPs, app developers, and policy makers could enhance the process of delivering and testing evidence-based apps to improve health outcomes.