Assessing the Quality of Mobile Phone Apps for Weight Management: User-Centered Study With Employees From a Lebanese University

Introduction

Background

Mobile health (mHealth) apps offer cost-efficient and effective strategies to prevent noncommunicable diseases such as obesity or diabetes [1], as these technologies can reach millions of users. According to the 2017 mHealth App Economics report, there are more than 350,000 health apps available in online stores [2], a market worth US $25 billion in 2017 [3] and estimated to reach US $31 billion by 2020 [4]. mHealth apps are generally designed for chronically ill people (56%), fitness enthusiasts (33%), and physicians (32%) [4], with users downloading them with the aim to monitor their fitness and track foods as well as to manage chronic conditions [5]. A recent study specifically evaluating the market of weight management apps in 10 different countries [6] identified 28,905 unique apps that focus on physical activity (34%); diet (31%); and on tracking exercise, calorie intake, and body weight (23%) [6].

Although the mHealth app market is expected to expand in the next 3 years [7], recent market research reports show a decline in app usage [4]. Some qualitative studies show that users stop using apps because of hidden costs, increased data entry burden [8], and low perceived engagement [9]. Engagement with an app is generally associated with sustained app usage [1], but it has also been associated with positive changes in physical activity [10,11] and diet [12], fundamental behaviors to obtain an optimal weight management. Understanding which apps are perceived engaging and of good quality is important to develop effective public health strategies addressing these problems [3]. The more people use the apps they like, the more likely people will perform the desired behaviors.

Are mHealth apps effective? Several recent systematic reviews suggest that mobile phone apps are effective in promoting dietary self-regulation [13] and weight management [14-20]. Despite lacking evidence-based content [6,21], health apps can be used as stand-alone delivery modes in self-directed weight loss interventions [22,23] or as supplemental components of complex interventions. Some studies employing researcher- developed apps [24] or popular calorie counting apps (eg, MyFitnessPal [25,26]) in combination with face-to-face delivery modes showed generally larger effects compared with interventions using the apps as standalone [27-29].

How do these apps work? According to several app audits or reviews, mobile phone apps include features that can trigger cognitive processes underpinning effective behavior change strategies or techniques [30-35], combining principles derived from self-determination theory [22,23] and persuasive technology [36,37]. For example, apps may include messages or notifications that remind users about their weight goals and provide positive feedback or reinforcements for achieving those goals. In a recent review of 23 popular weight management apps [30], researchers found that most apps included several change techniques that are commonly employed in effective behavior change interventions. The most frequently identified change techniques were self-monitoring of behavior (20/23, 87%), self-monitoring and goal setting of outcomes (both 19/23, 83%), feedback on outcomes (17/23, 74%), feedback on behavior (16/23, 70%), and goal setting of behavior (13/23, 57%) [30]. Although research demonstrated the efficacy of these techniques in influencing behavior, available evaluations of app quality cannot demonstrate app efficacy. Assessing app quality has become an important stream of research, with several authors arguing for the need to improve the quality evaluation and the need to use standardized tools and systematic approaches [38]. However, expert app evaluations or reviews do not take into account the point of view of end users. Little is known about how end users perceive the apps and in what terms they judge their quality.

In a recent review on app quality assessment methods [39], the authors emphasized the need to use multidimensional tools to comprehensively determine the quality of mobile phone apps, which should also include end users’ viewpoints. This is because the views of researchers and end users tend to diverge. On one side, researchers focus on aspects related to theoretical and evidence-based content [38,39]. For example, in the aforementioned expert app review [30], the authors judged the 23 apps as highly functional but poor in information quality, lamenting the absence of references to evidence-based content. At the same time, their quality ratings were not significantly associated with the 5-star ratings derived from Google Play and iTunes stores, suggesting a potential gap between the wisdom of the crowds and of the experts [30]. App store ratings cannot be entirely trusted as these ratings can be piloted through reviews and ratings provided by humans or bots paid by the same developer companies [40]. On the other side, developers tend to focus on usability and aesthetic aspects, such as design, ease of use, and customizability, as some qualitative studies demonstrate that these aspects are particularly appreciated by end users [8,9,41].

One of the most comprehensive and multidimensional tools to evaluate app quality is the Mobile App Rating Scale (MARS). Developed by Stoyanov et al for expert reviews [42], the MARS has also been developed and validated for end users [43]. The MARS and the user version of the Mobile App Rating Scale, uMARS are multidimensional as they encompass the domains of engagement, functionality, aesthetics, and information, which are used to estimate an objective app quality dimension (calculated as an average score of the aforementioned domains), based on objective features and characteristics of an app. Each domain consists of a set of items, assessed on 5-point scales. The engagement domain includes 5 items: entertainment, interest, customization, interactivity, and target group. Functionality includes 4 items: performance, ease of use, navigation, and gestural design. Aesthetics includes 3 items: layout,graphics, and visual appeal. Information includes 4 core items: quality, quantity, visual information, as well as credibility of the source of information. The MARS scale includes 2 additional items: accuracy of app description and goals (ie, Does app have specific, measurable, and achievable goals specified in app store description or within the app itself?). The latter items, in fact, require additional information that a lay user might not easily find while using the app. Finally, both scales have also a subjective quality domain, which includes 4 items: Would you recommend this app to people who might benefit from it?; How many times do you think you would use this app in the next 12 months, if it was relevant to you?; Would you pay for this app?; and What is your overall star rating of the app? Due to the third item, it can be assumed that the higher the subjective quality score, the more likely the users would use the app in the future; however, the instrument does not include a measure of actual behavior (eg, “How many times have you used this app in the past day or week”). The MARS and uMARS tools are available from the respective MARS [42] and uMARS [43] development studies.

The MARS tool, generalized to primary prevention apps [44], has been used in several expert reviews of apps for a variety of behaviors such as drink driving [45], sustainable food consumption [46], medication adherence [47], mental health and mindfulness [48], quality of life [49], rheumatoid arthritis [50], weight loss related to smoking cessation [51], and weight management [30]. The user version, originally tested on 2 harm minimization and affect management apps [43], assessed the apps according to the same domains. The only differences between the 2 tools are wording of the questions and the number of items assessing the information domain. The uMARS use has been documented in research protocols of trials addressing type 2 diabetes [52], health-related quality of life [53], pneumococcal disease [54], and breastfeeding [55]. However, to the best of our knowledge, the uMARS tool has not been used to quantitatively evaluate commercially available weight management apps. In addition, little is known about what users believe are important app characteristics, that is, app quality dimensions and how these dimensions relate to the overall app quality. Furthermore, according to the leading author of the scale (Stoyanov S, personal communication, November 2017), the items belonging to each domain were logically grouped, but no MARS or uMARS studies to date appear to have evaluated the relationships among different app quality dimensions.

Aims of the Study

In response to the call for more research on app quality evaluations from end users [39], the overarching goal of this study was to explore how laypersons evaluate the quality of a set of weight management apps, which experts considered of high quality in a recent review [30]. Specifically, this study aimed to (1) test the uMARS within a set of weight management apps; (2) understand which dimensions of app quality contribute the most to the overall app quality and how functionality, aesthetics, engagement, and information dimensions are related to subjective quality (as proxy of future app use); and (3) explore the associations between uMARS scales and users’ characteristics.

Methods

App Selection

A user experience study was used to examine the perceived quality and usability of selected apps and identify which apps achieve the best quality scores, which could be used in further studies with the same target population (employees of an academic institution). The units of analysis of this study were derived from a recent review of mobile phone apps for weight management [30]. In the cited review, only 6 out of the 23 apps reviewed (Table 1) scored above the median point of the MARS scale (3 out of 5), which is the median value of a 5-point scale. This value has been considered the minimum threshold of acceptability in the study by Mani et al [56].

Participants and Procedures

Following recommendations from user experience and usability testing literature [57,58], we aimed to recruit 5 to 6 evaluators per app (30-36 participants). Participants were employees (faculty and staff) of the American University of Beirut, who were recruited through social media postings and email invitations (the research team obtained a list of randomly selected email addresses).

Interested employees submitted an informed consent and completed a Web-based eligibility survey. Inclusion criteria were participants aged 18 to 65 years, employees of the university, and owning either Android or iPhone devices. After enrollment and after signing an informed consent, which included all study schedules and requirements, participants completed a Web-based sociodemographic and behavioral baseline survey. Then, they were randomly assigned to use 1 of the apps for 2 weeks. A member of the research team helped each participant install the assigned app and verified that it was correctly installed and functioning. The same member of the research team encouraged participants to use the app at least daily for the duration of 2 weeks. At the end of this study period, they were invited to complete a final Web-based app evaluation survey. They received US $10 to complete each survey. The study was approved by the local institutional review board (reference number FHS.MB.01) and was conducted between October 2016 and March 2017; analyses were completed in November 2017.

Measures

Background Characteristics

Background characteristics of the users included sociodemographic (age, gender, marital status, education, income, and number of working hours), health-related, and behavioral factors (perceived health status, height and weight, and physical activity assessed through the International Physical Activity Questionnaire-short form) [59]. App usage characteristics included operative system (Android or iOS) and previous experience with mHealth apps (for physical activity, diet, or weight tracking).

Quantitative Outcomes

App quality was evaluated employing the uMARS tool [43], which includes 20 items, as described in the introduction. The items are grouped into 4 objective subdomains: engagement (5 items), functionality (4), aesthetics (3), information (4), and 1 additional domain of subjective quality (4). Subjective quality scale includes 4 items that assess the intention to use the app in the future (ie, “Would you recommend this app to people who might benefit from it?” and “How many times do you think you would use this app in the next 12 months if it was relevant to you?”), propensity to pay for it (“Would you pay for this app?”), and an overall 5-star rating (“What is your overall star rating of the app?”), which reflects the way app stores rate the apps. All uMARS items are assessed through 5-point scales. Subscales are computed by averaging the respective domain items. A total uMARS score is calculated by averaging all subdomains, whereas subjective quality is calculated by averaging its related subitems. In the source study, the uMARS tool showed good internal consistency (Cronbach alpha=.90) and good test-retest reliability [43].

Data Analyses

Survey data were summarized using descriptive statistics. Background characteristics were kept continuous (age), dichotomous (gender), or categorical (height and weight were used to compute body mass index, BMI). Following the International Physical Activity Questionnaire scoring protocol, physical activity was categorized as high, moderate, or low [59]. For uMARS items, answers categorized by users as “don’t know/not applicable” were coded as missing. Missing value analysis was performed to estimate the frequency and level of missingness and determine the best strategy to address the issue (eg, multiple imputation [MI] and listwise deletion). Internal consistency (Cronbach alpha) was interpreted as excellent (≥.90), good (.80-.89), acceptable (.70-.79), questionable (.60-.69), poor (.50-.59), and unacceptable (<.50) [44].

As each app was evaluated by different groups of users, traditional interrater reliability (IRR) indices (ie, intraclass correlation coefficients, ICCs), reported in MARS and uMARS development studies, were not applicable [60,61]. To ensure that ratings could be aggregated, we evaluated interrater agreement (IRA) following literature recommendations [62,63], using 3 families of indices: James et al’s r_WG(J) [64,65] (based on multiple null distributions) [66], Brown et al’s a_WG(J) [67], and the adjusted average deviation index A_DMJ(adj) [68]. IRA was established with pragmatic and theoretical cut-off points such as for the r_WG(J): no agreement (<.29), weak (.30-.49), moderate (.50-.69), strong (.70-.89), and very strong (>.90) [64,65]; a_WG(J): not acceptable (<.59), weak (.60-.69), moderate (.70-.79), and strong agreement (>.80) [67]; and A_DMJ(adj): agreement above .80 [68]. Strong agreement was considered when all indices were consistently indicating an acceptable level of agreement.

In addition to the arithmetic mean of each uMARS score, we calculated a response data–based weighted mean (WDMEAN) [69]. The WDMEAN allows to incorporate individual raters’ disagreements as it is calculated as the sum of each individual score multiplied by its weight, which is a function of the distance of the individual response from the unweighted group mean. This aggregation approach has been employed in organizational and management literature to summarize opinions from key informants who may not share the same knowledge about the object of study [70,71] and have some expected disagreement [69,70,72]. Unweighted and weighted mean scores (range: 1-5) were expressed as percent scores. The scale midpoint (3, converted in percent, assuming that 1=0%, 5=100%, and 3=50%) was considered the minimum level of acceptability, as reported in the study by Mani et al [56]. The WDMEAN, in presence of full agreement, would correspond to the arithmetic mean.

Considering the small sample size and the nature of the scores (which might be prone to non-normal distribution), associations among and with uMARS domain scores were examined by inspecting Spearman rho (ρ) coefficients. Total and uMARS subdomains were associated with subjective quality, as the associations among uMARS subdomains are not considered meaningful [43] or interpretable (Stoyanov S, personal communication, November 2017). Given the multiple tests, P values were corrected for type 1 error [73]. Mann-Whitney and Kruskal-Wallis (K-W) tests examined differences in continuous variables. Due to the exploratory nature of the study, no inferential statistics were attempted. All analyses were performed with IBM SPSS Statistics v.24 for Mac.

Results

Participant Recruitment

Invitations were sent to 600 randomly selected email addresses, and additional 145 employees were recruited through social media postings. Out of 745 potentially interested employees, 44 provided informed consent and 5 were ineligible. The remaining 39 employees successfully enrolled in the study. Moreover, 36 of them completed the app evaluations and were included in the analyses. Their characteristics are reported in Table 2. Employees were on average 36 years old (SD 10.8), mostly female (24/36, 67%), married (19/36, 53%), with a graduate-level university degree (16/36, 44%), earned less than US $2000 per month (17/36, 47%), and worked on average 48 hours per week (SD 11.9). The majority reported being in very good or excellent health status (16/36, 44%), normal weight (17/36, 47%), or overweight (16/36, 44%), and moderately active (28/36, 78%), spending on average 6.6 hours per day (SD 2.4) sitting. Most users owned an iOS device (21/36, 58%), and some had previously used apps for tracking physical activity (22/36, 60%), diet (8/36, 23%), or weight (4/36, 11%). A total of 6 participants had previously used 1 of the reviewed apps (MyFitnessPal). Group allocation was not associated with any background characteristic.

App Quality Evaluation

Of the 36 users, 14 (39%) provided complete data covering 91% of values across the 20 uMARS items. The highest proportion of missingness was in the 3 information items (credibility of source: 39%; visual information: 25%; and quantity of information: 22%) and in 1 engagement item (customization: 19%). As missing was completely at random (Little’s missing completely at random test: χ²₂₆₄=251.8; P=.69), MI was employed. We generated 10 complete datasets [74,75] and ran the analyses with both incomplete and complete datasets to ensure comparability of results. For clarity and accuracy, all uMARS scores presented here are based on pooled means and variance estimates obtained from the MI datasets.

Internal consistency and IRA estimates are reported in Multimedia Appendix 1. Overall, Cronbach alpha values varied across the uMARS subdomains, being acceptable for engagement (alpha=.75) and aesthetics (alpha=.71), questionable for functionality (alpha=.61), poor for information (alpha=.51), and good for subjective quality (alpha=.88). Within each app, alphas were good for subjective quality (median .82, range .74 [My Diet Diary] to .93 [Lark]), acceptable for engagement (median .71, range .46 [My Diet Coach] to .93 [MyFitnessPal]) and aesthetics (median .70, range .42 [Lark] to .86 [SparkPeople]), and unacceptable for information (median .23, range .15 [SparkPeople] to .46 [MyPlate]). Negative alpha values were found among engagement and information items (SparkPeople and MyFitnessPal groups, respectively), indicating negative correlations among those items. IRA indices suggested overall agreement among users in most subdomains and for most apps. Moderate to strong agreement was found in functionality and aesthetics (all apps), whereas low agreement was found in engagement (MyFitnessPal and My Diet Diary), information (My Diet Diary, MyPlate, and SparkPeople), and subjective quality (Lark, MyFitnessPal, and My Diet Diary).

The unweighted, WDMEANs, and percent scores are presented in Table 3. Unweighted and WDMEANs were practically the same, with the former being generally lower than the latter. Information was the domain with the largest difference between unweighted and WDMEAN (1.3%), followed by engagement and functionality (both 1%), aesthetics and total uMARS score (0.1%). Subjective quality scores were also similar, with the highest difference in Lark (−2.4%).

Overall, all apps scored above the minimum threshold for acceptability (50%) in the total uMARS score and its main 4 subdomains. Functionality was the highest rated domain (median 3.9, interquartile range [IQR] 0.3), followed by aesthetics (median 3.7, IQR 0.5), information (median 3.7, IQR 0.1), and engagement (median 3.3, IQR 0.2). The subjective quality score was low (median 2.5, IQR 0.9). The scores are presented in the boxplot below (Figure 1). Only 2 apps (MyPlate and Lark) scored above the median thresholds in both uMARS and subjective quality scores.

After applying the Bonferroni correction for P values (P=.01), subjective quality was strongly and positively related (P<.001) with total uMARS score (ρ=.75), engagement (ρ=.68), information, and aesthetics (ρ=.60) and not significantly related with functionality (ρ=.40; P=.02).

Associations With Users’ Characteristics

Correlations with users’ background characteristics are reported in Table 4. After applying the appropriate P value corrections for multiple correlation tests [73], good health status was associated with engagement, total uMARS, and subjective quality; being obese with total uMARS score; and use of Lark with functionality and information. Very good or excellent health status was negatively related to engagement; use of SparkPeople was negatively related to information. K-W tests revealed significant differences across health status groups in engagement (χ²₂=11.9; P=.003), total uMARS (χ²₂=9.4; P=.009), and subjective quality (χ²₂=8.1; P=.02). Participants in good health status had higher median scores than those of the other 2 groups. Similarly, the 3 BMI categories (normal, overweight, and obese) scored significantly different in engagement (χ²₂=6.8; P=.03), functionality (χ²₂=6.1; P=.05), total uMARS score (χ²₂=6.6; P=.04), and subjective quality (χ²₂=6.11; P=.05). Obese individuals had higher median scores than those of the other 2 groups. Finally, K-W tests showed significasent differences among app groups in information (χ²₅=14.4, P=.01) and total uMARS score (χ²₅=12.4; P=.03). Users of Lark reported larger median information and total uMARS scores than the other apps. In Lark, subjective quality was positively associated with engagement (ρ=.87; P=.007) and total app quality (ρ=.90; P=.006). In SparkPeople, subjective quality was positively related to information (ρ=.97; P<.001).

‎View this figure

Discussion

Principal Findings

This is the first study that explored how laypersons evaluated the quality of free and popular mobile phone apps for weight management, using the uMARS tool [43]. The tool showed acceptable internal consistency levels in most subdomains, except for information (alpha=.51). Heterogeneity in alpha values was found within each app group. In 2 cases (SparkPeople for engagement and MyFitnessPal for information), alphas assumed negative values, which indicate small, negative correlations among the items in those subscales and lack of consistency. The internal consistencies we found are below those reported in the uMARS source study [43] and below the levels commonly recommended by the literature, suggesting large measurement errors [76]. Low alphas might be because of the number of items and sample size [77]. In addition, users might have had different interpretations of the items as some IRA indices pointed to low or no agreement within engagement items (MyFitnessPal and My Diet Diary), information (My Diet Diary, MyPlate, and SparkPeople), and subjective quality (Lark, MyFitnessPal, and My Diet Diary). Although IRA does not imply reliability [62,63], low agreement suggests a large degree of subjectivity in evaluating the apps, which can be expected, as the users are supposed to be free to have their own opinions about the apps, based on their own characteristics and needs.

Furthermore, large item nonresponse rates were registered in the information domain (22%-39%). Some users might have misunderstood these items or might not have known how to answer, thus leaving them blank. The missing information might explain the poor consistency and low agreement estimates in this specific domain. Unfortunately, the uMARS source study does not provide solutions in case of poor internal consistency or low agreement [43], and other studies employing uMARS did not report such issues [54,55]. To account for these limitations, we calculated the WDMEAN [69], an approach that allowed to retain all items. Eventually, the unweighted and weighted means were very similar, suggesting that applying the uMARS scoring protocol can still yield robust results. Nevertheless, the uMARS tool should be generalized to weight management apps, with larger user populations. We also recommend exploring users’ perceptions about the items including qualitative methodologies such as the think aloud method [78].

In this study, we employed the WDMEAN approach to estimate the responses from our key informants who were asked to apply the uMARS tool without previous training. To the best of our knowledge, no uMARS and MARS studies have used this approach, employing users who have undergone some level of training. This is the first study that utilizes the tool for users. By employing the WDMEAN method, it is possible to estimate app quality while accounting for the respondents’ potential disagreements, hence providing a truer average score, which accounts for the response of each individual [69]. On the contrary, the arithmetic mean can be influenced by extreme values (either very low or very high scores), and at the same time, it might reduce the intrinsic variability among raters’ ratings. The WDMEAN approach can be applied to many other studies, with small samples, in which researchers are interested in estimating scores while accounting for the agreement or disagreement among raters.

The second objective was to understand which app quality dimensions (ie, engagement, functionality, aesthetics, and information) contributed the most to the overall app quality score. All apps scored high in functionality, followed by aesthetics, information, and engagement. This is consistent with some qualitative research suggesting that users appreciate functional and aesthetic characteristics [8,9,41]. This is also consistent with the findings reported in the expert review, upon which this study is based, as the apps were deemed highly functional and with limited information quality [30]. However, engagement, aesthetics, and information appeared to be strongly related with subjective quality, which includes questions that indicate the propensity of using the apps in the future (“Would you recommend...,” “Would you pay,” “How many times would you use it...?,” and “What is the overall star rating?”). This might indicate that users might not engage with these apps regardless of their good functional features. This is consistent with findings from qualitative studies, which show that users might stop using an app not because of technical features but rather because of low engagement or hidden costs [8,9]. Another important consideration was that in our study, subjective quality was only weakly correlated with functionality (ρ=.40; P=.02). Conversely, engagement had the strongest correlation with subjective quality (ρ=.68; P<.001). This might indicate that app engagement can play an important role in achieving sustained app usage [10,12]; however, future studies should be conducted to establish whether a causal link between engagement and future app use exists.

The third objective was to explore the associations between uMARS scales and users’ characteristics. In this sample, we found that obese users and those in good health status provided higher app quality ratings in engagement, total uMARS, and subjective quality. In other words, healthy, obese individuals perceived these apps particularly engaging and of high quality. As engagement is related to app usage [1], these individuals might be more likely to use the apps in the future. These findings are particularly suggestive, as these popular weight management apps (in particular, Lark, MyFitnessPal, and MyPlate) may be used in interventions addressing obesity prevention (healthy volunteers) and treatment (obese) [9]. Future research could test whether these apps, which had demonstrated having high behavior change potential [30], can effectively influence behavior and promote weight loss among overweight or obese individuals. This study informed the development of a self-directed weight control intervention, which targets the same population (clinical trial registry: NCT03321331).

Limitations

The results of this study need to be interpreted bearing in mind its limitations. A major limitation is the design (noncrossover). For feasibility reasons (budget and time constraints), we could not ask all users to evaluate each app, hence allowing us to calculate IRR using ICC indices. To overcome this limitation, we employed methodological solutions that have never been employed in similar studies (ie, IRA estimates [60,62] and WDMEAN [69]). These solutions allowed us to ensure the robustness of the responses obtained from the employees recruited in this study. This solution is pragmatic and allows to be applied in real-life scenarios, whereby research study participants might not be willing or able to dedicate more time to the study. Moreover, users evaluated the free version of the apps used for 2 weeks. Ratings might have differed if they had used the pro versions with additional functionalities. App evaluation might also be influenced by actual app use and by the amount of time spent on each app. As the authors of the expert review noted [30], some apps prompt different feedback and unlock features only after repeated use. We instructed participants to use the apps at least daily for 2 weeks, but we did not assess actual app use. Another limitation is the sampling of this study as we had access to a convenience sample of employees from an academic institution in Lebanon, who voluntarily agreed to participate. Although we found correlations with health status and BMI categories, this study might not be generalizable to the entire population and to other cultural contexts and settings, as we recruited mostly female, educated, and healthy individuals. The small sample size is also another limitation; however, the size was based on pragmatic considerations and aligned with recommendations from the heuristic evaluation literature [57,58]. Larger samples should investigate whether these findings hold truth in different segments of the population. It will be practical to focus studies on specific segments of the population to increase the accuracy of the findings. Nevertheless, we believe the results are generalizable to similar academic institutions in Lebanon or in the Middle East region or who have similar employee populations, although the tested apps are available internationally. Another limitation is the use of self-reported data and self-administered Web-based surveys that are prone to missing data. We used Web-based tools because we wanted to avoid interviewer bias and we did not want to interfere with the users’ evaluations of the apps. We wanted the users to test the apps in the wild for 2 weeks, without specialized training, which is usually a prerequisite of expert reviews. We could have used interviewers to reduce data entry mistakes or inconsistencies, but we opted for self-administered Web-based forms to avoid interviewer bias. A related limitation is the presence of large amounts of missing data in some of the subdomains of the uMARS scale (eg information domain), which forced us to apply caution when interpreting the results. Although we employed modern techniques to deal with missing data, we cannot make strong assumptions on the reasons for the missing responses backed on data, as the instrument (Web-based survey) did not capture comments related to the uMARS scale. We recommend that future studies investigate how users respond to the survey and how they apply the answers. We have already suggested that qualitative techniques such as the think aloud method [78] could be applied to understand the thought processes that people use when answering questionnaires. These techniques would allow to identify potential pitfalls in the scale, hence improving its validity across cultures and sample populations.

Conclusions

Across the 6 popular and free weight management apps analyzed in this study, functionality is the quality dimension that laypersons valued the most. However, engagement was strongly associated with subjective quality, a dimension that includes future app use. The higher the subjective quality and engagement, the more likely users might use the app. App developers and public health professionals should ensure that an app is both functional and engaging so that users will be more likely to use it. Future longitudinal studies are needed to ascertain this connection.

The tested apps (in particular Lark and MyPlate) were perceived as more engaging and of higher quality among healthy, obese individuals, making them promising modes of delivery for obesity prevention and treatment interventions.

From a methodological standpoint, the uMARS tool is a practical and feasible tool that can be used to assess app quality by laypersons without specialized training. However, further research is needed to establish its validity in the domain of weight management.