This meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [27] (the PRISMA checklist is provided in Checklist 1) and was registered on the International Prospective Register of Systematic Reviews (PROSPERO registration number: CRD42024513381).

Six databases (ie, PubMed, Embase, CENTRAL, Web of Science, PsycINFO, and CINAHL) were independently searched by 2 authors (SJL and YZ) for relevant RCTs published in English from inception to May 20, 2024. The search strategy was developed by one author (SJL) and confirmed by another author (YZ) according to the PICOS (Participants, Interventions, Comparisons, Outcomes, and Study Design) framework. The key search terms and MeSH (Medical Subject Headings) included obesity, overweight, body weight, mobile apps, mHealth, and apps. The full search strategy is provided in Multimedia Appendix 1. In addition, we manually searched the references of previously published relevant reviews.

Studies included in this meta-analysis had to meet the following criteria: (1) participants: adults with overweight and obesity aged ≥18 years; (2) intervention: mobile phone apps used as the primary component; (3) comparisons: usual care without mobile phone app interventions or no intervention; (4) outcomes: weight-related outcomes (ie, weight, BMI, WC, fat mass, and fat mass percentage), behavioral outcomes (ie, MVPA and energy intake), and metabolic outcomes (ie, SBP, DBP, triglycerides, and HbA_1c); (5) study design: RCTs published in English.

Studies were excluded if they met the following criteria: (1) participants diagnosed with diseases other than obesity; (2) conference articles, letters, reviews, commentaries, or protocols; and (3) unavailable full texts or incomplete relevant data.

All titles and abstracts of the studies were downloaded and imported into EndNote X9, with duplicates automatically removed. Two investigators (SJL and YZ) independently screened the titles and abstracts based on the eligibility criteria. Subsequently, the full texts of potentially relevant studies were downloaded and reviewed to select the included articles. Reasons for exclusion were recorded. Any disagreements were resolved through consultation with a third author (HMM).

Data regarding paper characteristics (author, year, and country), study design, study population (eg, average age, sex distribution, and baseline BMI), intervention content (eg, intervention duration and brief description of the intervention), comparison content (eg, brief introduction to the comparison), and outcomes were extracted into a predesigned Microsoft Excel by one reviewer (SJL) and checked by a second reviewer (YZ). We did not make any assumptions about missing or unclear information when extracting data to avoid introducing misleading information. Other statistics (eg, 95% CI or SEs) were converted to SD if not available (Cochrane Handbook version 6.4, Chapter 5, 5.7) [28]. If the units of the outcome measures were not uniform, such as triglycerides measured in mg/dL, we converted them to mmol/L to ensure consistency in units. We used Michie BCT Taxonomy (BCTTv1) to identify the BCTs present in the included studies (Multimedia Appendix 2) [24]. Intervention and control descriptors in the original RCTs that aligned with BCT definitions were marked with a “✓.” The initial identification of BCTs was performed by 1 reviewer (SJL) and checked and confirmed by another reviewer (YT). Subsequently, we mapped the BCTs used in each study to the corresponding BCRs.

Cochrane risk-of-bias version 2 (ROB 2) tool was used to assess the methodological quality of the included RCTs by 2 reviewers (SJL and YZ) independently [29]. This tool contains 5 domains: the randomization process, deviations from the intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Each domain was rated as either low, some concerns, or high risk of bias. An overall judgment was then assigned to each study. A third investigator (HMM) was responsible for resolving any discrepancies during the evaluation process.

Two independent authors (SJL and YZ) used the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) framework to assess the overall quality of evidence. The framework evaluates 5 domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Based on these evaluations, the quality of evidence was classified as high, moderate, low, or very low. Any disagreements during this process were resolved by a third reviewer (HMM).

The meta-analysis was performed to evaluate the effectiveness of mobile app interventions on weight loss among adults with overweight and obesity, with forest plots used to visually display the results of individual studies and the overall effect size. Outcomes were pooled in the meta-analysis if change scores from baseline were available, and the number of participants was recorded. The heterogeneity of the data was assessed using the chi-square test and the I² statistic, with 25%, 50%, and 75% representing low, moderate, and high heterogeneity, respectively [30]. Review Manager software (version 5.3; Cochrane Collaboration) was used to analyze data. The mean difference (MD) with a 95% CI was used to express the effectiveness of interventions involving mobile apps. A random-effects model was applied for data analysis to obtain more conservative results.

Subgroup analyses were conducted to explore factors, including intervention contents, presence of theory, intervention duration, number of BCTs, and number of BCRs, which may influence the effects of the intervention on weight and help identify possible sources of heterogeneity. Sensitivity analyses were also performed to assess the robustness of the results in R (version 4.4.2; R Foundation for Statistical Computing). Publication bias was evaluated by visual evaluation of funnel plot asymmetry and quantified by Egger test in Stata software (version 15; StataCorp). Publication bias assessment was performed separately for outcome measures included in meta-analyses that incorporated more than 10 original studies (Cochrane Handbook version 6.4, Chapter 13, 13.3) [28]. In all analyses, P<.05 was considered statistically significant.

The literature search across 6 databases generated 8424 articles, and an additional 2 records were added through hand-searching relevant reviews. After removing duplicates, titles and abstracts of 5427 studies were screened. Subsequently, 120 full texts were reviewed. However, 91 studies were excluded based on the eligibility criteria for the following reasons: wrong populations (n=14), wrong interventions (n=33), wrong control (n=21), wrong outcome (n=8), unavailable data (n=7), wrong study design (n=3), duplicates (n=4), and conference abstract (n=1). A detailed list of exclusions with corresponding reasons can be found in Multimedia Appendix 3. Finally, a total of 29 studies [19-23,31-54] were included in the systematic review and meta-analysis. Details are presented in the PRISMA flowchart in Figure 1.

The details of the study characteristics are listed in Multimedia Appendix 4. The 29 studies were conducted in Spain [21,31,32], France [33], the United Kingdom [34-37], Korea [19,38,39], India [20], Australia [40-42], China [43], Japan [22], the United States of America [23,44-51], Belgium [52], New Zealand [53], and Germany [54]. Sample sizes ranged from 20 to 650 across the included studies, with the average age ranging from 22 to 55 years. Behavioral theories (ie, social cognitive theory, self-regulatory theory, control theory, operant conditioning, ecological theory, social network theory, transtheoretical model, habit formation theory, social support theory, and behavior self-management theory) were used in the design of the mobile app interventions. The intervention duration ranges from 2 months to 24 months. The intervention contents shared some common features, such as goal setting, social support, self-monitoring, and the use of credible sources. Usual care and no intervention were the primary forms of control groups. Outcome measures included weight-related outcomes (weight, BMI, WC, fat mass, and fat mass percentage), behavioral outcomes (MVPA and energy intake), and metabolic outcomes (SBP, DBP, triglycerides, and HbA_1c).

In the 29 included RCTs, 34 BCTs from 13 categories were identified in the intervention group, while 20 BCTs from 9 categories were identified in the control group. The number of BCTs identified per study ranged from 2 to 16 (median 8, mean 8.69, SD 3.79) in the intervention group and from 0 to 14 (median 2) in the control group (see Multimedia Appendix 5). The most frequently identified BCTs in the intervention group were: 2.3 “self-monitoring of behavior” (n=25), 4.1 “instruction on how to perform the behavior” (n=24), 2.2 “feedback on behavior” (n=20), 1.1 “goal setting (behavior)” (n=19), and 1.4 “action planning” (n=15). In the control group, the most frequently used BCT was 4.1 “instruction on how to perform the behavior” (n=18). Details are shown in Table 1.

The identified BCTs were categorized into 3 resource types: facilitating, boosting, and nudging (see Figure 2 and Multimedia Appendix 6). The specifics of BCRs identified in each study are presented in Table 2. Among the 29 included RCTs, 24 studies [19-22,31-34,36-38,40-44,46-52,54] used facilitating resources, 24 [19-22,31-38,40-42,44,47-54] used boosting resources, and 19 [21,22,32-38,40-42,44,45,47-50,54] used nudging resources. Fifty-nine percent of the included studies used all 3 types of resources, while only 4 studies [43,45,46,53] used 1 resource type.

Of the 29 included studies, 10 [20-22,33,37-39,42,44,53] were judged to have low risk of bias, 4 [31,47,48,54] as having some concerns, and 15 [19,23,32,34-36,40,41,43,45,46,49-52] as having high risk of bias (Figure 3). While all studies reported the use of randomization, 13 studies [19,23,35,36,40,41,43,46,49-52,54] failed to explain the allocation concealment process, which was considered a high risk in this domain. Other significant sources of bias included high dropout rates with unreported reasons and the absence of appropriate analysis to assess the impact of missing outcome data.

Of the 11 outcomes included, the overall quality of evidence was assessed to be very low to moderate according to the GRADE assessment (Table 3). The risk of bias was considered serious based on the results of ROB 2 assessments. Indirectness was not considered serious, as the outcomes were directly measured. No publication bias was detected for any of the outcomes. Inconsistency ranged from not serious to very serious based on the heterogeneity values, and imprecision ranged from not serious to serious, as determined by the 95% CI.

The review examined the effects of mobile phone app interventions on weight-related outcomes (weight, BMI, WC, fat mass, and fat mass percentage), behavioral outcomes (MVPA and energy intake), and metabolic outcomes (SBP, DBP, triglycerides, and HbA_1c). A summary of the meta-analyses’ results was shown in Table 4.

The mobile phone app intervention decreased MVPA, but this change was not statistically significant (MD=−0.69, 95% CI −5.67 to 4.28; P=.78; Multimedia Appendix 11). No heterogeneity was observed. When pooling the data on energy intake in the meta-analysis, the results indicated a nonsignificant effect size (MD=−62.72 kcal/day, 95% CI −181.62 to 56.18; P=.30) with moderate heterogeneity (Χ²=13.33, df=5, I²=62%; Multimedia Appendix 12).

SBP was nonsignificantly reduced (MD=−0.14 mm Hg, 95% CI −2.66 to 2.37; P=.91). The mobile phone app intervention significantly decreased DBP (MD=−1.76 mm Hg, 95% CI −3.47 to −0.04; P=.04). It led to a nonsignificant increase in triglyceride levels (MD=0.06 mmol/L, 95% CI −0.19 to 0.31; P=.64). In addition, there was a significant decrease in HbA_1c levels (MD=−0.13%, 95% CI −0.22 to −0.04; P=.005), with high heterogeneity (Χ²=22.79, df=4, I²=82%; Figures S7-S10 in Multimedia Appendices 13-16).

Five subgroup analyses were conducted to explore potential factors influencing the effects of the apps, intervention content (single intervention [diet or PA] versus combined intervention [diet+PA]), theory presence (theory-based versus nontheory-based), intervention duration (short-term [≤3 months] versus medium-term [6 months] versus long-term [>6 months]), number of BCRs used (3 resources vs <3 resources), and number of BCTs used (<median [8] vs ≥median [8]).

Among these, the combined intervention demonstrated a greater effect size (MD=−1.82, 95% CI −2.48 to −1.16 kg) compared with the single intervention group (MD=−0.24, 95% CI −1.00 to 0.53 kg), with statistical significance. Regarding intervention duration, the medium-term (6 mo) group exhibited the largest effects (MD=−2.50, 95% CI −3.65 to −1.35 kg), significantly outperforming the other two groups. In addition, the number of BCTs used ≥8 had a greater effect size (MD=−1.83, 95% CI −2.56 to −1.09 kg vs MD=−0.61, 95% CI −1.28 to −0.07 kg; P=.02; Table 5).

Sensitivity analyses were performed using the leave-one-out and high-risk study exclusion methods. Notably, after excluding one study from the meta-analysis, the weight reduction effect size changed from −1.45 kg (95% CI −2.01 to −0.89) to −1.20 kg (95% CI −1.61 to −0.79), indicating that this study had a relatively significant influence on the overall effect size [49]. The heterogeneity also decreased, from 56% to 23%. In addition, excluding high-risk studies resulted in a weight reduction of −1.24 kg (95% CI −1.84 to −0.83) and an I² of 36%. High-risk studies may overestimate the effect size. However, the sensitivity analyses demonstrate that the results are robust, as no statistically significant changes were observed. This specific study and high-risk studies may account for the heterogeneity to some extent.

Funnel plot and Egger test for publication bias were conducted when a minimum of 10 studies were incorporated into the meta-analysis. No significant publication bias was detected for weight (Egger test, t₂₇=−0.28; P=.78), BMI (Egger test, t₁₇=0.63; P=.54), WC (Egger test, t₁₁=0.10; P=.92), fat mass (Egger test, t₉=0.54; P=.61), or fat mass percentage (Egger test, t₁₀=−0.69; P=.51). The funnel plots were displayed in Figure 5 and Figure S11-S14 in Multimedia Appendices 17-20.

A total of 29 RCTs [19-23,31-54] were included in this systematic review and meta-analysis, examining 11 outcomes. The results of this review indicated that mobile app interventions significantly reduced weight, BMI, WC, fat mass, DBP, and HbA_1c. However, nonsignificant effects were observed for other outcomes, including fat mass percentage, behavioral outcomes (ie, MVPA and energy intake), and metabolic outcomes (ie, SBP and triglycerides). BCTs and BCRs used in the included studies were also identified, with most studies (59%) using 3 resource types (ie, facilitating, boosting, and nudging). In addition, intervention content, duration, and the number of BCTs used were identified as influencing factors through subgroup analyses.

Several strengths can be identified in this review. First, a thorough examination of the effects of mobile phone apps on weight loss among adults with overweight and obesity was conducted. Our meta-analysis includes 11 outcomes based on findings from 29 studies. Weight-related outcomes reflect the direct effects of mobile apps, metabolic outcomes provide evidence of the association between weight reduction and health benefits, and behavioral outcomes inform researchers about the extent to which current interventions target diet and PA, offering guidance for future intervention design. This comprehensive set of outcomes offers substantial insights into the current effectiveness and status of mobile apps in weight loss interventions. Second, only RCTs were included in this review, as they are considered the gold standard in clinical research. This selection criterion ensures that our findings are based on high-quality studies, minimizing bias and enhancing the reliability of the results. Third, the BCTs used in the included RCTs were identified. Systematically reviewing the use of BCTs contributes to the standardization of intervention strategies and enhances comparability across studies, especially given the considerable methodological variability in current weight loss interventions. This provides a reference for researchers to replicate, refine, or design new intervention approaches. Fourth, mapping BCTs to BCRs based on the BCRM helps clarify the resources required to implement specific BCTs during intervention design. This approach also supports more tailored and practical intervention strategies by taking into account the resources available to patients, thereby promoting interventions that are both targeted and feasible.

However, there are also some limitations in this study. First, we only included adults with overweight and obesity without pre-existing medical conditions, which may limit the generalizability of the results to broader populations, such as individuals with diabetes or CVD. However, adults with overweight and obesity without medical conditions represent a critical population from the perspective of preventive medicine, as they are at high risk of developing chronic diseases. Early interventions targeting this population group could potentially reduce the burden of future disease onset. Moreover, there are notable differences in intervention responsiveness and health needs between adults with overweight and obesity with and without medical conditions. By including only participants without comorbidities, this study allows for a more focused analysis of lifestyle-based mobile app interventions, without the influence of complex confounding factors such as medication use or the bidirectional interactions between obesity and chronic diseases. This approach allows for a clearer assessment of behavior change outcomes, especially in the identification of BCTs and the associated BCRs. Second, we identified the BCTs and BCRs used in the RCT interventions included in this review. Although interrater agreement in coding the BCTs and resources was high, the coding process inherently involves a degree of subjectivity, which may influence the results related to BCTs and BCRs. Furthermore, given the relatively limited app of the BCRM in previous research, standardized procedures for mapping BCTs to corresponding resources have not yet been established. The mapping criteria proposed in this study, therefore, require further testing and validation in future research. Finally, the effectiveness of weight loss was influenced not only by the quantity of BCTs and resources used but also by their quality, including factors such as the intensity, frequency, and tailoring of interventions to individual needs. As a result, no meta-regression was conducted to link specific BCTs or BCRs to their effectiveness. We advocate for the conducting of more RCTs to better clarify the specific effects of individual BCT or BCR.

This meta-analysis presents findings that demonstrate that mobile phone app interventions significantly reduce weight, BMI, WC, fat mass, DBP, and HbA_1c among adults with overweight and obesity. These mobile phone app interventions are cost-effective and can be applied to a large population. However, current mobile apps have not achieved clinically significant weight loss. Future studies should focus on optimizing app interventions by incorporating more effective behavior change strategies and resources to enhance their overall effectiveness.