Is the uptake, engagement, and effectiveness of exclusively mobile interventions for the promotion of weight‐related behaviors equal for all? A systematic review

Mobile health interventions are promising behavior change tools. However, there is a concern that they may benefit some populations less than others and thus widen inequalities in health. This systematic review investigated differences in uptake of, engagement with, and effectiveness of mobile interventions for weight‐related behaviors (i.e., diet, physical activity, and sedentary behavior) based on a range of inequality indicators including age, gender, race/ethnicity, and socioeconomic status. The protocol was registered on PROSPERO (CRD42020192473). Six databases (CINAHL, EMBASE, ProQuest, PsycINFO, Pubmed, and Web of Science) were searched from inception to July 2021. Publications were eligible for inclusion if they reported the results of an exclusively mobile intervention and examined outcomes by at least one inequality indicator. Sixteen publications reporting on 13 studies were included with most reporting on multiple behaviors and inequality indicators. Uptake was investigated in one study with no differences reported by the inequality indicators studied. Studies investigating engagement (n = 7) reported differences by age (n = 1), gender (n = 3), ethnicity (n = 2), and education (n = 2), while those investigating effectiveness (n = 9) reported differences by age (n = 3), gender (n = 5), education (n = 2), occupation (n = 1), and geographical location (n = 1). Given the limited number of studies and their inconsistent findings, evidence of the presence of a digital divide in mobile interventions targeting weight‐related behaviors is inconclusive. Therefore, we recommend that inequality indicators are specifically addressed, analyzed, and reported when evaluating mobile interventions.

education (n = 2), occupation (n = 1), and geographical location (n = 1). Given the limited number of studies and their inconsistent findings, evidence of the presence of a digital divide in mobile interventions targeting weight-related behaviors is inconclusive. Therefore, we recommend that inequality indicators are specifically addressed, analyzed, and reported when evaluating mobile interventions.
K E Y W O R D S body weight, health promotion, mHealth, social inequality

| INTRODUCTION
Mobile health (mHealth) interventions are interventions that use mobile, often Internet-supported, tools such as smartphone applications, tablets, wearables (e.g., smart watches and pedometers), and personal digital assistants (PDAs) to promote health, illness selfmanagement, or remotely support treatment. As Internet use and smartphone ownership are common, mHealth interventions may reach a large number of people 1 and, in doing so, may increase access to health care. 2 They are available to the user at any time and in any place, enable the user to self-monitor and self-manage health, and consequently have the technical affordances to empower users in taking ownership of their health. 3 As such, mHealth interventions carry the added potential of a higher retention rate at population scale compared to non-mHealth interventions. 1,2,4 The commonly used techniques of self-monitoring and feedback in mHealth tools have shown high levels of user satisfaction, 5,6 long-term engagement with the tool, 7 and high levels of self-efficacy in changing health behavior. 8 The World Health Organization has consequently proposed that such innovative digital technologies present a potential to improve health care coverage, health risk protection, and enhance health and well-being for all. 9 Indeed, mHealth interventions are becoming increasingly popular in healthcare. For instance, in 2019, Germany introduced its Digital Healthcare Act. Among others, this policy allows physicians to prescribe digital health applications, which are usually delivered exclusively mobile, to allow for monitoring and provision of care without contact to service providers. 10 To date, mHealth interventions have been applied to a wide range of health domains such as HIV prevention, smoking cessation, diabetes self-management, and depression self-management. [11][12][13][14] Furthermore, a large number of mHealth interventions have been studied in relation to weight-related behaviors, such as sufficient physical activity, low levels of sedentary behavior and a healthy diet. 15,16 These behaviors are important determinants of lower morbidity and mortality [17][18][19] and contribute to both optimal physical and mental health. 20,21 For an intervention to have impact on population health, it should reach the target group, the users should show sufficient objective (e.g., uptake and continued usage) and subjective engagement (e.g., enjoyment and perceived usefulness) with the tool, and it must be effective. 22 mHealth interventions were found to be as effective as face-to-face interventions in increasing physical activity 23,24 and reducing sedentary behavior. 25 Yet, other aspects needed to achieve long-term health behavior change, such as uptake, followed by engagement, are reported less often in studies on mHealth interventions for physical activity 26 and other weightrelated behaviors. Whereas uptake is reported by the majority of studies, the information usually relates to sample size, and not to the representativeness of the sample, that is, the number of people who accepted the invitation to take part compared to all contacted persons. 27 Despite the potential of digital technologies, questions remain about the actual public health impact of mHealth interventions in their ability to reduce health inequalities. 28 Certain groups of the population are known to have a lower adoption of health behaviors and to experience lower access to health care and/or higher morbidity, including people from ethnic minorities, 29 from economically disadvantaged backgrounds, 29 from sexual minorities, 30 people with lower health literacy levels, 31 with a lower educational status, 32 or women in certain patriarchy cultures. 33 Moreover, these groups of the population may also experience more barriers to using mobile health apps.
As technology literacy is related to age, digital experience, overall health literacy, education, cultural background, 34 and urban or rural residency, 4 people might differ in their ability to engage with mHealth interventions as a result of these demographic and personal characteristics. It has thus been proposed that mHealth interventions may actually widen health inequalities. 35,36 This is also commonly referred to as the "digital health divide." 36 Inequalities may arise during different stages of an intervention and may relate to access, uptake, objective and subjective engagement, different types of usage, and efficacy. 4 Several studies support the existence of a digital health divide. A rapid evidence synthesis, including seven reviews and eight individual studies, found that the uptake of digital health interventions in primary care is low and that those who use such tools are more often female, younger, more educated and have a higher income. 37,38 A secondary data analysis of a large study showed that younger individuals, with higher income were more likely to have health apps installed, while those of Hispanic ethnicity background and who are less educated were less likely. 39 Similarly, racial and ethnic minorities are less likely to enroll in clinical trials of digital interventions. 40 A narrative review found that there is lack of diversity in studies in dietary selfmonitoring health apps, with study participants being predominantly white and female. 41 In support of this, a review of 83 studies on mHealth interventions found that the diversity of low socioeconomic position and ethnic minorities were not reflected in the included studies, neglecting factors that may widen a digital divide. 42 This idea finds support in the findings from a recent meta-analysis that showed that digital interventions are not effective in changing physical activity in low socioeconomic position populations. 43 There is also a lack of research and evidence on the effect of mHealth interventions in rural populations. 38 To the best of our knowledge, no systematic review has investigated the digital health divide as related to the uptake, engagement, and effectiveness of exclusively mobile interventions for weightrelated behaviors. Mobile interventions are especially promising in increasing access to health care for all, as they may solve the structural barriers to access to care. 44 Though many behaviors have shown to impact body weight, we decided to focus on physical activity, diet, and sedentary behavior for this review, as these are commonly targeted in interventions for managing body weight in studies using mobile interventions 15,45 ; and body weight has been repeatedly found to be an important predictor of morbidities and premature mortality. 46 We defined "uptake" as the act of downloading and installing a mobile intervention device 47 ; in the context of this review, this also included the act of receiving or buying a mobile intervention. As a lot of terms are used interchangeably in relation to "engagement," and as measures of engagement focused predominantly on certain metrics, a more comprehensive approach to measure in-app engagement is suggested. 48 We defined "engagement" as "(1) the extent (e.g., amount, frequency, duration, and depth) of usage and (2) a subjective experience characterized by attention, interest and affect." 49 Finally, "effectiveness" refers to the extent to which an intervention leads to a desired outcome in the real world compared to a comparison condition (e.g., comparing pre-and post-intervention scores within subjects and comparison with a group receiving no intervention or usual care). 50 Therefore, this review aimed to address the following research questions: Does the (1) uptake, (2) engagement, and (3)

This review followed the Preferred Reporting Items of Systematic
Reviews and Meta-analysis (PRISMA) guidelines. 51 The completed PRISMA checklist can be found in Supporting Information S1. The study protocol was registered on the International Prospective Register of Systematic Reviews (PROSPERO: CRD42020192473). Raw extracted data can be downloaded from the project's Open Science Framework page (https://osf.io/kdmz8/).

| Eligible studies
We included studies conducted among adults aged 18 and over, with no apparent pre-existing medical condition, where the intervention was delivered exclusively in a mobile format (e.g., using a smartphone, PDA, or wearable, without any intervention components being delivered face-to-face or using other digital tools such as computers or websites) and targeted at least one weight-related behavior, that is, diet, physical activity and sedentary behavior, and/or weight loss resulting from changing these behaviors. Studies that additionally contained coaching calls or social media as an optional intervention component that was also delivered fully mobile (i.e., not via access to a computer) were also considered. Studies were eligible if the results reported data on at least one of the following inequality indicators: age, gender, socioeconomic position (including occupation, from unskilled to skilled labor or profession; income; employment, considering employed versus unemployed individuals), level of education, health service accessibility, geographical indicators, sexual orientation, health, or digital literacy. Studies where the primary or secondary outcomes were the uptake of or the engagement with, or the effectiveness of the interventions, were included. Any real-life test of the intervention was included, and no control group was required for the interventions; therefore, single-group designs with pre-post-tests were also considered.
Studies focusing exclusively on a clinical population, that is, a population with a medical condition other than overweight or obesity (e.g., diabetes), were excluded. Yet, interventions that focused on a health risk factor (e.g., overweight) were included. Studies examining infants, children, or adolescents, or where the intervention was part of the treatment of a pre-existing condition, were excluded. Interventions that solely focused on text messaging were excluded since they could also be delivered on mobile phones, which provide fewer opportunities for tailoring and personalization, which is typically seen as an advantage of smartphone-based interventions. 5 Phoning or using video calling were excluded since they can also be delivered without using a smartphone (i.e., landline phones or a computer) and require in-person contact (in comparison to purely digital interventions that can be delivered in a purely automated form without immediate response from a person). Finally, interventions that used a combination with other elements such as information material, a website, or a face-to-face element were excluded. Finally, systematic reviews and meta-analyses were also excluded.

| Citation search
A backward and forward citation search of all initially included studies through the electronic search and gray literature search was conducted using Google Scholar. Screening was performed by two independent authors (AD, CF, ES, HB, or LK).

| Screening
All records identified by the search strategy were exported to Endnote X9, deduplicated and uploaded into Covidence software (Covidence Systematic Review Software, Veritas Health Innovation, Melbourne, Australia). To reduce the likelihood of reviewer selection bias, titles and abstracts of all studies were screened independently by at least two reviewers (CF, HB, DS, or LK). Full texts were screened by at least two reviewers (AD, ES, HB, or LK). Inter-rater reliability based on the number of eligible and ineligible studies was tested at the full text screening phase using Cohen's Kappa statistics. 52 The following cut-offs were used: 0.41-0.60 indicated moderate agreement, 0.61-0.80 substantial agreement and 0.81-0.99 almost perfect agreement. 52 Disagreements were resolved by discussion. During full text screening "moderate agreement" was achieved between the three independent reviewers (Kappa = 0.45).

| Data extraction
A data extraction form was developed following the existing guidelines from the Cochrane Collaboration. 53 Study characteristics (author, date of publication, location of the study, aim of the study, sample size and type, target behavior of the study, methodological characteristics, such as design, recruitment, and participants, intervention development, and type) and the main findings related to the research questions of this systematic review were extracted. The data extraction was performed independently by two reviewers (LK and DS); disagreements were resolved by discussion.

| Quality assessment
The quality of included studies was assessed through assessments of the risk of bias randomized studies, using the Cochrane Risk of Bias 2.0 tool (RoB 2.0), 54 and of study quality for nonrandomized studies with the Effective Public Healthcare Panacea Project (EPHPP) tool 55 respectively. The RoB 2.0 tool evaluates biases in selection, performance, detection, attrition, reporting, and other sources of bias. 54 Based on the individual domains, an overall risk of bias rating is derived. Studies scored with RoB 2.0 can be rated as low risk of bias, some concern of risk of bias, and high risk of bias. A high risk of bias is attributed to studies that show a high risk of bias in at least one domain of the quality appraisal. Equally, a rating of some concern of risk of bias is given if at least one domain is rated to be of some concern for risk of bias. The EPHPP tool assessed bias due to selection of participants into the study, type of study, confounders, blinding, measurement of outcomes, and withdrawals from and drop-out of the study. Based on the individual domains, an overall quality rating is derived. Studies scored with the EPHPP tool can receive a rating of high quality, moderate quality, or weak quality. The overall quality rating of weak quality is given to studies that have a weak rating in at least two domains of the quality appraisal. To stay true to the purpose of each tool, we refer to study quality for nonrandomized studies and risk of bias for randomized studies. However, in the discussion, when looking at overarching patterns, both quality and risk of bias may be mentioned jointly. The risk of bias or quality assessments of the studies were assessed independently by two reviewers per study (LK, DS, AD, or HB).

| Data synthesis
Narrative synthesis was conducted to synthesize the findings of the studies for the identified inequality indicators. Due to design heterogeneity of the included studies, as well as the limited number of studies per inequality indicator, a meta-analysis was not considered possible.

| Quality assessment of included studies
Of the 10 randomized studies, assessed with the RoB 2.0 tool, nine showed some concern for risk of bias [56][57][58][59][61][62][63][64]68 whereas one showed a high risk of bias. 65 Two studies only showed some risk of bias in one domain 58,64 ; four studies showed some risk of bias in two domains 56,57,59,63 ; with the remaining three showing concerns for a risk of bias in more than two domains. Some concerns of risk of bias were mostly detected in a selective reporting of results (for eight out   The six non-randomized studies assessed using the EPHPP tool were judged to be of weak quality rating. 60,67,[70][71][72] One study scored weak in two domains 70 ; one study scored weak in three domains 60 ; two studies scored weak in four domains 67,72 ; and two scored weak in five out of six quality rating domains. 66,71 All six studies scored weak on reducing risk of confounders; five scored weak on reducing risk due to withdrawals and drop-outs 60,66,67,71,72 ; and these also scored weak on reducing risk due to study design.
The quality assessment can be found in Supporting Information S4.

| Uptake
Only one study (of moderate quality/showing some concern of risk of bias) examined differences in age, gender, ethnicity, education, and employment between those who downloaded the app and those who did not, and reported no differences. 64

| Engagement
Seven studies reported on engagement in relation to inequality indicators. 56,59,62,[65][66][67]72 Out of the five studies, four did not find any difference in engagement by age. 59,62,63,66 One study reported that older participants (versus younger participants) engaged more often with features that allowed sharing information in Facebook groups. 67 This one study finding differences showed a high risk of bias/weak quality, whereas the four that showed no difference had a moderate quality rating/some concern of risk of bias.
Six studies investigated gender differences in engagement, 59,62,63,66,67,72 of which three studies did not find any difference. 59,62,63 One study reported that the frequency of nutrition tracking and meditation features tracking was higher for females (mean = 36, SD = 48) than for males (mean = 22, SD = 34). 66 A different study reported that females were more likely than males to participate longer in the intervention. 72 Another study only found gender differences when age was not included in the model; once age was accounted for in the model, gender differences ceased to be significant. 67 The three studies where a difference was found, were of weak quality/high risk of bias, whereas the remaining three were rated as moderate quality/some concern of risk of bias.
In terms of ethnicity, 58

69.6%
Smartphone app and smartphone app with personal coaching differences between completers and non-completers at 3 months, with non-Hispanic white participants (16%) being more likely to complete the study than participants of other race or ethnic minorities (39%). 58 All four studies had received a moderate quality/some concern of risk of bias rating.
Four of the included studies examined differences in education levels of participants, 59,63,65,68 out of which three studies did not find any difference. 59,63,68 One study reported that more educated individuals engaged more with the available features (including video lessons and coach calls), while those with shorter time spent in formal education engaged predominantly in coaching calls (66%). 65 The one study that reported a difference had a weak quality/high risk of bias rating, whereas the other three studies had a moderate quality/some concern of risk of bias rating.
Two studies reporting on the same sample investigated the difference based on occupation and no significant difference was found. 62,63 Both studies were scored as of moderate quality/some concern of risk of bias.
Finally, one study (moderate quality/some concern of risk of bias) examined differences in employment status (i.e., whether participants worked full-time, part-time or were unemployed), income, and health literacy and no differences were found between consistent and inconsistent trackers. 59

| Effectiveness
Nine studies examined differences in effectiveness due to inequality indicators. 56 60 The three studies that reported differences were of weak quality/high risk of bias, whereas the other three studies were of moderate quality/some concern of risk of bias.
Eight studies reported on gender differences. 56,57,60,61,66,[70][71][72] Two studies found no difference. 57,61 One study reported marginal differences at p = 0.05, with females having better results in step counts. 56 Another study reported significant gender differences regarding nutritional scores (mean value of nutritional scores for females of 36, SD = 48; versus males 22, SD = 10), but not regarding minutes spent exercising. 66 In contrast, one study reported better results in weight loss for males 71 and increased step count for males, although no difference was found in the weight between males and females. 70 One study found greater absolute weight loss for males, however, when compared to baseline a greater proportion of females achieved 5% weight loss. 72 Finally, one study reported significant gender differences for weight loss but did not provide information on which gender benefitted more. 60 The five studies reporting differences were of weak quality/high risk of bias, whereas the rest of the studies were of moderate quality/some concern of risk of bias.
Three studies (all moderate quality/some concern of risk of bias) reported on ethnicity and no differences were found. 56,57,61 Out of the two studies reported on education, one found no differences. 57 The other study reported that individuals with a degree or postgraduate degree, compared to having completed primary school (5th grade), reported greater weight loss. 65 The study reporting differences was of weak quality/high risk of bias, as opposed to the one reporting no differences, which was of moderate quality/some concern of risk of bias.
Only one study (weak quality/high risk of bias) examined occupation and found that individuals in professional, managerial, or executive jobs, compared to being self-employed or working in agriculture, achieved greater weight loss. 65 Two studies (both moderate quality/some concern of risk of bias) examined differences in employment status in relation to effectiveness, with no differences found. 56,57 One study (moderate quality/some concern of risk of bias) that investigated income found no difference. 57 The one study (weak quality/high risk of bias) that investigated geographical location found that individuals who lived in the capital as opposed to those living outside of the capital reported greater weight loss. 65 Finally, one study (moderate quality/some concern of risk of bias) investigated differences in health literacy and found no difference. 57 The summary of results based on the inequality indicators investigated can be found in Table 2.

| Inequalities in uptake of exclusively mobile interventions
Given only one study included in this review investigated uptake, no conclusions can be drawn regarding a potential digital divide in the uptake of mHealth interventions. This lack of evidence may partly be attributable to the context of the included studies. The present review focused on intervention studies, which included experimental tests of interventions in RCTs and similar study designs. In this context, information on potential participants who decided not enroll in the study is presumably more difficult to obtain and not typically reported. However, recent systematic reviews suggest that, in the context of digital health studies, participants are typically younger and comprise fewer males and ethnic minority participants than the general population, suggesting potential inequalities in uptake. 40 indicators of gender, age, and education were studied most frequently in relation to engagement. Where gender differences were reported, females were found to engage more with the intervention compared to males. This is in line with research on other types of digital health interventions that found women to be more active engagers in health research. 77 Understanding the possible mechanisms and other factors that include but are not limited to individual characteristics (e.g., interest in health-related topics, motivation, and design of the app) could help gain insight into why females, compared to males, seem to engage more with mHealth interventions. Given that higher engagement does not automatically lead to higher effectiveness, however, further research on how to increase effective forms of engagement among all users is needed. 78 Only one study reported significant differences in engagement for age. In this study, older participants were more likely to share their data on Facebook compared to younger participants. 67 These findings contrast with the assumption that younger participants are more likely to engage with digital interventions because of higher digital literacy 76 and self-efficacy. 79 Furthermore, one out of four studies reported differences in education, indicating that individuals with a higher level of formal education (e.g., university degree) engaged more with the intervention. 65 This mirrors the results of previous studies indicating that higher levels of education are associated, for instance, with more frequent use of digital services to access health-related information 80 or treatment 81 and better health in general. 82 However, since the majority of studies included in this review did not report differences in engagement based on education, results could indicate that mobile interventions might reduce the digital divide induced by education, although this assumption requires further testing.

| Inequalities in the effectiveness of exclusively mobile interventions
The findings of this review on the inequality indicators on effectiveness are also inconclusive. The majority of studies focused on gender, age, and ethnicity. Most studies reported gender differences; however, results were mixed regarding which gender benefited more from the intervention. Since the included studies were diverse in terms of target behaviors and intervention components, the present review cannot provide indicators on when an intervention may be more beneficial for one gender than for another. Among the studies that explored ethnicity as a potential influence on intervention effectiveness, no study found differences in effectiveness for the different ethnicities investigated. This perhaps suggests that mHealth interventions may be a tool to reduce or to limit the increase of existing health disparities related to ethnic background. 83 It may also be that mHealth interventions help reduce structural barriers related to socioeconomic position (e.g., cost of in-person interventions and difficulty attending an appointment-all these could be overcome through mHealth interventions) that are more prevalent among ethnic minorities, however, these types of inequality indicators have been largely neglected by Nevertheless, the few studies included in this review might need to be supported by evidence from future research in this area to drawn firm conclusions.

| Implications and future research
Although evidence for a digital divide in exclusively mobile interven- health apps, which also indicated that evaluation frameworks for mental health apps rarely include indicators for diversity and inclusion, such as cost or using simple language. 86 Including inequality indicators such as the PROGRESS-plus criteria 87 in evaluation frameworks would be an important step toward making mHealth interventions accessible to marginalized groups, which may be in greater need of support. [29][30][31][32] Another potential reason for the divergence in findings reported by individual studies may be the study design: RCTs included in this review did not report significant differences in engagement and effectiveness based on age or gender, while studies using other designs (e.g., single arm trials with pre-post comparisons) consistently reported significant differences. Due to randomization, the inclusion of control groups, and the possibility to draw causal conclusions, RCTs are usually considered to be of higher methodological quality than cohort studies, single arm trials or case-control studies. 88  More research is needed to disentangle age and generational effects to determine whether differential preferences of younger and older adults, for example, regarding behavior change techniques included in the intervention, are actually due to age-related changes such as changes in physical abilities, 94 or rather due to generational differences.
Certain inequality indicators such as socioeconomic position, health literacy, or geographical location have hardly been studied in the context of exclusively mobile interventions. This limited our ability to draw meaningful conclusions from the presented data. A recent meta-analysis on digital interventions for physical activity more broadly reports that digital interventions are ineffective in people with lower socioeconomic position. 43 However, Western and colleagues were unable to test differences in uptake and engagement, which are considered a necessary and important prerequisite for intervention effectiveness 43 : if an intervention is not taken up and used as intended, it will most likely not induce any effects. 78 The divergent findings may thus underline the importance of studying a potential digital divide not only regarding effectiveness but also regarding uptake and engagement (see also the work of Birch and colleagues, for weight loss interventions more broadly 27  Although in its early stages, research has already identified starting points for making mHealth interventions more equitable. For instance, mHealth interventions should use plain language and simple labels to benefit users with low health and digital literacy. 99 Also, uptake and engagement may be improved by tailoring apps to the users' cultural backgrounds. 5,[100][101][102] Tailoring the use of different behavior change techniques to age group-specific needs may increase the effectiveness of mHealth interventions across age groups, especially when age group-specific motivational barriers are addressed. 103,104 Additionally, to make mHealth interventions more equitable, required time and financial costs may need to be reduced, since these are resources of which socioeconomically advantaged individuals tend to have more of. 105 Finally, the present review focused on exclusively mobile interventions only, which is why only a small number of studies were identified. While mobile interventions often include non-mobile components (e.g., an accompanying website, counseling via telephone or in face-to-face consultations 69,106,107 ), exclusively mobile interventions become increasingly common. For instance, the German Digital Healthcare act allows medical professionals to prescribe digital interventions as they would prescribe medication. 10 These interventions are often delivered exclusively mobile. However, research suggests that exclusively digital interventions may be less effective than interventions including face-to-face components. 108 This difference may be elevated in deprived populations who might need more support to sustain engagement and thus benefit from the intervention. However, this assumption needs to be investigated in future research.

| Strength and limitations
This review is the first to investigate a wide range of inequality indicators in the context of uptake of, engagement with and effectiveness of exclusively mobile interventions, which are becoming increasingly popular in health promotion and care. However, the PROGRESS-Plus criteria indicate further potential sources of inequality such as culture, language, or disability, 87 which should be taken into account in future empirical research and systematic reviews. Another strength is the focus not only on effectiveness of the digital interventions, but also on uptake and engagement as two further important dimensions of a digital divide. Finally, the comprehensive search strategy was complemented with backward and forward citation tracking to ensure completeness. However, it is important to acknowledge that due to the heterogeneity of the studies included in terms of which inequality indicators were assessed, we were unable to conduct a meta-analysis. The aim of this review focused on health promotion and primary prevention, therefore, excluded participants with pre-existing medical conditions. However, these populations are more likely to be part of a disadvantaged group, which may have led to further indicators of a potential digital divide. Future reviews could focus on groups with existing conditions to explore this. Furthermore, the studies included in this review provided only limited quantitative data on differences based on the inequality indicators, and the indicators were often limited to common demographic variables such as gender, age, and ethnicity. Lastly, included studies were not specifically designed to test for a potential digital divide; accordingly, they might not have been adequately powered to detect these interaction effects, and relevant tests were often not preregistered. These caveats limit the conclusions that can currently be drawn regarding a potential digital divide in exclusively mobile interventions.

| CONCLUSION
There is limited to mixed evidence of a digital divide in exclusively mobile interventions targeting weight-related behaviors. This review highlights the need for researchers to address, analyze, and report on a broad range of potential sources of inequality when evaluating mHealth interventions to test whether they reduce or widen existing health disparities and to identify starting points for developing more inclusive mHealth interventions.