Obesity has become a major, rising health epidemic worldwide. As a complex multifactorial disease, the management of obesity is challenging because there is no single effective treatment. Of late, there has been great interest in using apps for weight loss. Mobile apps were effective for weight loss [1-6] by using different behavior change techniques to a certain extent [1-6]. These behavior change techniques include intention formation, goal setting, barrier identification, problem solving, planning, general encouragement, self-monitoring of behavior, feedback on performance, social support, and social comparison [4,7].

On the basis of recent research, mobile apps help users to adhere to self-monitoring and weight loss goals better than the traditional pen-and-paper methods and other mobile health interventions (web-based or PDA) [8-12]. In 2015, Mateo et al [13] conducted the first meta-analysis that focused on mobile apps and found a modest weight loss of –1.04 kg (95% CI –1.75 to –0.34 kg; I²=41%) among mobile app users. In 2014 and 2015, similar results were found by Khokhar et al [14], Hutchesson et al [15], and Liu et al [16] after they expanded the inclusion criteria to include email, SMS text messaging, monitoring devices, and smartphones. Cai et al [17] observed similar findings when they measured the effect of mobile apps in patients with diabetes mellitus. In 2020, Islam et al [18] updated the literature review and extended the scope of the previous meta-analysis performed by Mateo et al [13] by including more subgroup analyses.

Most of the interventions are smartphone apps combined with other behavioral nonmobile interventions; yet, it’s unclear whether the app’s effect on weight loss is due solely to its use or to the addition of the behavioral component. If the behavioral component relies on human coaches and personalized feedback by dietitians, this will affect the scalability of the mobile app used for weight loss. Personalized feedback provided by mobile apps has proven to be an essential feature of such apps because the feedback increases users’ logging-in frequency and engagement with the apps [19-23]. Personalized feedback from an interventionist or professional also affected the results positively [24-26]. Thus, the combination of mobile app use with in-person contacts such as coaching or counseling sessions, interventionist feedback, web-based chatting, or telephone calls with professionals was more effective than mobile app use alone [19,24,27-29].

The aim of this meta-analysis is to evaluate the effectiveness of mobile app interventions alone or in combination with other behavioral interventions on weight loss. Although Lyzwinski [30] analyzed the intervention components of mobile devices in a narrative review, the author did not compare the effect of the various components on weight loss. This research extends the scope of the previous systematic reviews by including 2 subgroup analyses based on nonmobile interventions that accompanied smartphone use and human-based versus passive behavioral interventions. The results are organized according to duration because it is inaccurate to compare weight loss at 3 months with weight loss at 6 or 12 months. The secondary outcomes of the study include the impact of different mobile app features on weight loss and mobile app adherence.

The protocol was not registered.

The PubMed, MEDLINE, and EBSCO databases were searched for relevant studies published between the database inception date and January 31, 2022. The search strategy incorporated keywords. The terms used included weight loss, obesity, overweight, smartphone, mobile phone, cell phone, mHealth, eHealth, and adherence. The search was then filtered to studies involving randomized controlled trials. All previous systematic reviews and meta-analyses were researched to find further missing studies. EndNote X9 (Clarivate) was used to remove duplicate publications and for screening purposes (for further information on the search strategy, please see the example provided in Multimedia Appendix 1). Multimedia Appendix 2 includes a summary of the reasons for excluding articles.

To investigate the effectiveness of nonapp interventions combined with smartphone apps, studies were eligible if (1) the design included randomized controlled trials, (2) they included the use of a smartphone app in at least 1 arm, (3) weight loss was an outcome, and (4) the population consisted of adults. There was no restriction on the population regarding overweight versus obesity or being diagnosed with chronic diseases, the language or year of publication, length of interventions, or follow-up duration. Gray literature, such as abstracts and conference proceedings, was included. We also searched the lists of references of the articles that we included.

On the basis of the eligibility criteria, 2 research team members (JA and HI) independently screened all the articles by study title and abstract. If the information listed in the title or abstract was insufficient to determine the study’s relevance, the full text of the study was selected to be reassessed later. Next, each member further screened the selected studies at the full-text level. Any disagreements were resolved by consensus.

Working independently, 2 investigators (JA and HI) extracted the data from the articles, resolving disagreements by consensus. A form developed by using the KoBo toolbox (Kobo Inc) was used to extract data from eligible research papers, including digital object identifier, the title of the study, year of publication, type of article, country of study, population, sample size, trial name, trial period, number of arms, and details of each arm. Mean body weight changes were recorded from baseline to the end of the trial with SDs and adherence-related outcomes. SD or 95% CI was recorded if available. Neither authorship nor publication journal nor study results were blinded for data extraction.

The risk of bias was assessed independently by 2 authors (NA and AE) using the Cochrane Collaboration tool for assessing the risk of bias in randomized trials [31]. The tool covers the following bias domains: selection bias, performance bias, detection bias, attrition bias, and reporting bias. Each author independently judged each domain as having a low, unclear, or high risk of bias. When differences of opinion arose between the 2 evaluators, the item was discussed until a consensus was reached. JA randomly selected a few articles to assess the risk of bias and compared the results with those of the 2 primary evaluators to consolidate the assessments further.

The study’s primary outcome is the mean weight change (measured in kilograms) from baseline to 3, 6, and 12 months. All outcomes were recorded if the study measured the outcome at multiple points. Adherence measures were examined, and a systematic review of the literature was performed.

The meta-analysis evaluated whether smartphone app interventions were effective on their own or whether other behavior interventions were necessary for weight loss. A fixed effect model was used to obtain the overall effect size across included studies and its associated 95% CI. On the basis of the studies chosen, the outcome of weight loss was measured as the weight change from baseline to the 3-, 6-, or 12-month periods. When SD was not mentioned, the variance was calculated from the 95% CI. We examined heterogeneity using the I² test, which describes the percentage of variability in effect size estimates because of heterogeneity rather than sampling error. The statistical analyses were performed using Review Manager software (version 5.4; Cochrane Training).

In studies with more than one arm that included an app, inverse variance meta-analysis was used to produce an overall effect size across all treatment arms, creating a single intervention-versus-control comparison for each study. The exact process was performed when there was more than one control arm.

All the interventions used were reviewed by 2 authors (JA and HI), who then grouped them into the following categories: smartphone apps, trackers (weighing scale, step tracker, or bite counter), behavioral therapy or advice (podcasts, telephone calls, booklets, SMS text messages, and in-person meetings), feedback (SMS text messages, email, oral, or written by a coach), self-monitoring, social support (social media or web-based forum), meal replacement, and financial incentives. Human-based active behavioral coaching or feedback included in-person meetings, interaction with the interventionist through Twitter or chat feature of the app, and tailored feedback from a coach or interventionist using telephone calls, SMS text messages, or group sessions. The passive behavioral component included passive standardized behavioral messages as part of the app, Facebook, or podcasts. The apps were also classified according to the following features: self-monitoring, education, feedback, social support, rewards, and gamification. On the basis of these features, 2 associations were later analyzed. Using an independent 2-tailed t test, the first analysis examined the association between weight loss and each app feature. The second analysis, using 1-way analysis of variance, studied the association between weight loss and the number of features. We aim to measure the association between adherence percentage and app features. Nevertheless, the definition of adherence was not homogeneous, and we ultimately conducted a systematic review of the adherence outcome.

Concerning the intervention period, it included both active treatment and follow-up. If provided, the baseline weight of the intervention group was noted; however, if it was not available, the average weight of participants in the whole cohort was captured from the demographics table.

The search strategy enabled us to compile 1081 articles from different resources, of which 34 (3.14%) were included in this meta-analysis (Figure 1). The studies selected were published between 2011 and January 31, 2022; however, 68% (23/34) of the articles were published in the last 5 years (2017-2021). The studies were conducted in the United States (22/34, 65%), Australia and New Zealand (3/34, 9%), Europe (4/34, 12%; Germany: 1/4, 25%; Spain: 1/4, 25%; and the United Kingdom: 2/4, 50%), and Asia (5/34, 15%; Japan: 1/5, 20%; Singapore 1/5, 20%; and South Korea: 3/5, 60%). The sample size ranged from 16 to 440, with a mean of 113.09 (SD 94.1). The population in the studies ranged from men and women from the general population to adults at risk for diabetes as well as adults with diseases such as cardiology issues, diabetes, or metabolic syndrome. Of the 34 studies, 7 (20%) used a theoretical framework: social cognitive theory (3/7, 43%), social cognitive theory and transtheoretical model (2/7, 29%), and social cognitive theory and self-efficacy theory (2/7, 29%). Tables 1 and 2 present a summary of the characteristics of the included studies.

The studies selected for the meta-analysis used the following two types of analyses for the intervention results: intention-to-treat and per-protocol analyses. The intention-to-treat studies (n=21) were assessed for risk, and the per-protocol studies (n=13) were similarly evaluated. Among the 21 intention-to-treat studies, 28% (6/21) were determined to be low risk, 33% (7/21) had some concerns, and 38% (8/21) were determined to be high risk. The highest risk was related to outcome measurement because of the nature of the app and the consequent lack of blinding. Similarly, among the 13 per-protocol studies, 23% (3/13) were determined to be low risk, 38% (5/13) had some concerns, and 38% (5/13) were determined to be high risk. In the per-protocol studies, high risk was due to deviation from the intended interventions. Figure 2 [12,26-29,32-60] shows a summary of the risk of bias assessment of the included studies.

Of the 34 included studies, 24 (71%) examined the effectiveness of smartphone apps on weight loss at 3 and 6 months, whereas 5 (15%) measured the outcomes at 12 months; of these 5, 2 (40%) did not include SD or 95% CI [49,56] and 1 (20%) included an app in all arms [53]. Consequently, the meta-analysis and subgroup analysis were performed for the 3- and 6-month outcomes. Compared with the control group, smartphone apps resulted in a pooled net estimate weight loss of –1.99 kg (95% CI –2.19 to –1.79 kg; I²=81%) and –2.80 kg (95% CI –3.03 to –2.56 kg; I²=90%) at 3 and 6 months, respectively (Figure 3 [27,28,32,33,36,39-42,44,48,52,58,60] and Figure 4 [12,27-29,35-37,39,43,46-48,51,52,54,57,58]). Subgroup analysis was performed based on the different interventions that accompanied the use of the mobile app. When compared with control, the combination of the mobile app, tracker, and behavioral interventions showed a statistically significant weight loss of –2.09 kg (95% CI –2.32 to –1.86 kg; I²=87%) and –3.77 kg (95% CI –4.05 to –3.49 kg; I²=90%) at 3 and 6 months, respectively (Figures 3 and 4). Another subgroup analysis was performed based on the type of behavioral interventions, human-based versus passive. When compared with control, only the combination of the mobile app with intensive behavior coaching or feedback by a human coach showed a statistically significant weight loss of –2.03 kg (95% CI –2.80 to –1.26 kg; I²=83%) and –2.63 kg (95% CI –2.97 to –2.29 kg; I²=91%) at 3 and 6 months, respectively (Figure 5 [27,28,32,33,36,42,44,48,50,52,55] and Figure 6 [27-29,35,36,43,46,47,51,52,54,57]). The funnel plots (Multimedia Appendix 3) were symmetrical, suggesting that there was no publication bias.

The studies included a total of 50 arms with an app and 35 control arms. Table 3 presents the characteristics of the various components of both the intervention and control arms. The app alone was used in 18% (9/50) of the intervention arms, whereas social and feedback interventions were not common, and financial incentives were the least used. The two most common combinations of interventions included the app, behavior component, and tracker (10/50, 20%) and app and behavior component (8/50, 16%). Among the 35 control arms, the behavior component was present in 21 (60%) and self-monitoring in 10 (29%), whereas no action was present in only 4 (11%). Multimedia Appendix 4 shows the details of the various components by study.

Commercial mobile apps were used in 56% (28/50) of the study arms. Key features of the mobile apps included self-monitoring (50/50, 100%), feedback (20/50, 40%), education (15/50, 30%), social support (12/50, 24%), rewards (7/50, 14%), and gamification (3/50, 6%; Table 4). Two-thirds of the mobile apps included 1-2 features. There was no association between weight loss and any specific feature; neither was there an association between weight loss and the number of app features (data not shown).

In this meta-analysis, of the 34 included articles, 22 (65%) studied adherence to smartphone apps with diverse approaches to assessing adherence (Multimedia Appendix 5), leading to limitations in a direct comparison of their findings; thus, the results are described in a systematic review rather than a meta-analysis. Adherence data were extracted from each study, and a member of the research team (NA) coded the assessments based on the following four themes: (1) self-monitoring of weight, (2) self-monitoring of dietary intake, (3) self-monitoring of physical activity, and (4) interaction with the app. Each of these themes was defined differently in terms of measurement of adherence. Of the 22 articles that studied adherence to smartphone apps, 12 (55%) studies used more than one theme of adherence to weight loss apps; therefore, the total number of studies under each theme is not equal to the combined total number of studies included in this review.

Although dietary self-monitoring was the most commonly used adherence method among the studies (16/22, 73%), its measurement and analysis varied considerably. Studies defined adherence to dietary self-monitoring with a smartphone app either as recording any food or calorie intake [26,27,32,36,43,44,48,49,52,61] or a specific amount of calories [12,28,50] or a particular number of meals or entries [34,54] or both [37]. The frequency of dietary intake was based on the total number of days reported in percentages, ratios, or discrete numbers. Of the 22 studies, only 1 (5%) measured adherence as the percentage of participants logging food or calorie intake at least once per week [32]. The adherence rate to dietary intake ranged from 48% to 79% of the days when using self-monitoring.

App adherence for self-monitoring of weight and exercise was recorded in 36% (8/22) [27-29,32,37,38,49,54] and 27% (6/22) [27,28,36,44,52,54] of the studies, respectively. Common calculations included mean or percentage of daily recordings, recording at least once per week, or an average of days that participants recorded per week.

Of the 22 studies, 9 (41%) reported adherence as the frequency of the interaction with the app. Adherence was defined as wearing the wearable [32,33] or logging in [43,45] or opening the app, irrespective of the participant’s use [39]. The frequency of wearing the wearable ranged from at least 4-5 days per week [32,33] to at least 8 hours per day [33]. In contrast, a group of studies considered counting the daily interaction with specific app components or features as a sign of adherence [29,34,38,61]. For example, some considered completing entries immediately after taking photographs of the meal a sign of adherence [34].

Some studies reported positive associations between adherence and weight change by using combined adherence measurements [28], tracking adherence with dietary intake app recording [12,26,27,48], and measuring adherence with physical activity app recording [44]. In some of the studies [29,33,34,50,52], the adherence percentage showed evident decline throughout the study period.

This meta-analysis aims to measure the effectiveness of smartphone app–based interventions on weight loss, considering the additional components available in the mobile app. Similar to previous meta-analyses [13,18], the use of mobile apps resulted in a small significant weight loss of –2.03 kg (95% CI –2.57 to –1.5 kg; I²=83%) at 3 months. Although the mobile apps included different behavioral strategies, all relied on self-monitoring and only one-third included more than two features. It is important to note that there was no association between weight loss and mobile app features. A review of mobile app features revealed that self-monitoring was most commonly used, whereas social support and personalized feedback were less commonly used [62]. Subgroup analysis integrating additional nonapp intervention components with the mobile app showed that use of the tracker and behavioral components resulted in the most significant weight loss of –3.77 kg (95% CI –4.05 to –3.49 kg) at 6 months. Human-based behavioral interventions were associated with weight loss of –2.63 kg (95% CI –2.97 to –2.29 kg) at 6 months.

The various app features were not associated with weight loss. The meta-analysis has shown significant heterogeneity among the different apps used and the additional nonapp intervention components, reaching 90% in some forest plots. It is difficult to determine the role of a mobile app in weight loss management beyond self-monitoring. Only a few studies (7/36, 19%) based their work on theoretical frameworks such as social cognitive theory, transtheoretical models for health behavior change, and self-efficacy theory. Both the transtheoretical framework and self-efficacy theories rely on the individual, supporting the importance of self-monitoring as the main feature of mobile apps. Only social cognitive theory addresses the importance of support; the study results highlighted the importance of support by human coaches. The behavioral components of the included studies in this subgroup analysis were mainly in-person meetings [28,32,33,40], with additional feedback or telephone calls by a coach [28,36]. There were numerous smartphone app features such as self-monitoring with additional feedback [32,36,48], gamification and awards [33,40], and social support [28]. In the form of an in-person meeting or telephone call, the human component, in combination with the user app interactions through self-monitoring and feedback, is crucial for weight loss. The need for the human component raises the question of whether artificial intelligence would raise the mobile apps to a new level in the management of weight loss [63], such as the use of chatbots [64,65], and whether users would accept such a mode of coaching [66].

Weight maintenance is defined as losing 5%-10% of body weight and maintaining this loss for at least 1 year. In this meta-analysis, the studies analyzed were mainly short term and lasted for a maximum of 1 year. Of the 36 studies included in this systematic review, 2 (6%) had a longer intervention duration, and neither showed any difference in weight from baseline between the app-based intervention and the control at 18 months [37] and 24 months [29]. In a systematic review, Varkevisser et al [67] have provided strong evidence that behavioral determinants such as self-monitoring of weight and eating predict weight loss maintenance. Moreover, web-based interventions were effective for weight maintenance [68]. It would be helpful to examine further how mobile apps can be a form of self-regulation and adherence that can help users maintain weight loss. As obesity is a long-term relapsing disease and mobile apps are cost-effective, further research should address whether mobile apps could play a role in weight loss maintenance.

Some studies in this review and the literature [69-72] suggest that greater adherence to self-monitoring has been associated with greater weight loss. However, many articles do not provide detailed measurements of adherence to self-monitoring in weight loss apps. Furthermore, the results across studies could not be accurately compared because of the numerous variations in measurement methods and the definitions of themes used to assess adherence. Irrespective of the measurement method used, adherence to self-monitoring decreased with time, emphasizing the importance of studying different app features and associated interventions that could have affected the participants’ adherence. Of note, the variation in adherence measurements made it challenging to compare data across various studies. Hence, it is necessary to formulate a well-structured standard definition of adherence measures that can be used across future studies.

To our knowledge, this is the only meta-analysis that has performed a subgroup analysis based on the add-on interventions to mobile apps. This meta-analysis also included gray literature such as conference abstracts, and the funnel plots showed good symmetry, excluding the possibility of publication bias. In contrast, the findings of this meta-analysis should be treated with caution because of the vast heterogeneity in the studies that would limit real-life applicability. Moreover, one-third of the articles had a high risk of bias; however, this bias could not have been avoided because of the nature of the app and its effect on blinding. Although some of the studies used commercial apps, the study team developed most of them. Another limitation is the heterogeneous behavioral component that ranged from simple booklets to in-person meetings and telephone calls. Finally, it is essential to note that the weight loss outcome was measured in kilograms rather than as a percentage of weight loss from baseline.

Mobile phone apps have a role in weight loss management and result in modest weight loss compared with active control. Combining a mobile app, tracker, and human-delivered behavioral component led to the highest degree of weight loss at 6 months. Further research should use artificial intelligence to replace the human-delivered behavioral component for better mobile app use scalability.