The frequency of weight entries was analyzed. In most health care app research programs, user adherence to a certain intervention has often been measured by the frequency of self-reporting [4,24,26,27]. Therefore, the number of body weight entries per week served as a trustworthy variable in measuring the degree of adherence to the weight loss intervention in this research [4,24,26,27].

The study participants reported their weight through the app. After reporting, weight-related menu-usage logs were automatically stored on a server. The collected usage log data were summed at weekly intervals to generate data indicating adherence to weight-reporting activity over 16 weeks, for each subject.

BMI is often used to determine whether a person is within a healthy weight range based on their height [4,14,28]. Since this is calculated by dividing weight by height (kg/m²), the BMI value represents relative weight with respect to height [29,30]. In other words, BMI values indicate whether a person is “underweight,” “healthy,” “overweight,” or “obese” [4,14,28]. Because of its ease of interpretation and satisfactory explanatory power, BMI has been used instead of basic weight measurements for many diet-related studies on web and mobile weight tracking environments [4,14,25,31-33].

Participants reported their body weight daily through Noom. The daily BMI values for each participant were calculated by dividing the daily reported weights by their initially entered height values, and the daily results were averaged on a weekly basis.

BMI delta values were used to identify weekly weight fluctuation trends [26,34,35]. For this analysis, BMI delta was defined as the difference between the largest and smallest BMI values in a given week. As such, for each participant, the average weekly BMI values representing their overall obesity level and the weekly BMI deltas indicating their degree of weight fluctuation over 16 weeks were used for the statistical analysis.

As with weight entries, the frequency of self-reported meal history entries has frequently been used in prior studies to address the level of adherence to weight loss intervention programs [4,32]. In this study, participants entered not only their main meals but also snacks in between meals. As participants recorded food items by searching for them within Noom’s food database, their eating habits and selections could be grouped in detail by food types.

These food types were categorized as green, yellow, and red groups based on their caloric density, that is, calories per gram or milliliter [29,30,36]. In general, foods in the red group have very high caloric density and include foods high in fat and oil [29,30,36]. Foods categorized as yellow have moderate caloric density; they are less fatty and oily than the red group but still not optimal [29,30,36]. Foods in the green group have very low caloric density and are comprised mostly of natural fruits and vegetables [29,30,36]. The applied food type criteria according to caloric density are described in Multimedia Appendix 1.

Variables representing food intake by food type were generated based on the aforementioned criteria. Furthermore, food intake was reported 6 times per day (ie, breakfast, morning snacks, lunch, afternoon snacks, dinner, and evening snacks). This provided a substantial amount of data regarding the changes in food intake for the 3 food groups of each participant over 16 weeks.

Figure 2 shows the number of weight entries reported by user each week for the 16 weeks. The number of weight entries decreased over time in all 3 groups even though the decreases in the slopes were different (ie, −0.16 for the group who lost >10% of their starting weight, −0.13 for the group who lost 5%-10%, and −0.12 for the group who lost <5%). The members of the group with >10% weight loss recorded their weight more often in Noom than the other groups (ie, the groups who lost ≤10%). This trend at the aggregated level was statistically significant throughout the entire study period (Table 2). Moreover, the statistical significance of the trend was noticeable in women (Table 2).

Figure 3 shows the trend in the BMI values for the 3 groups over time at the aggregated level. The differences in the BMI values between the 3 groups was statistically significant after week 5 (Table 2). Furthermore, the statistical difference was more noticeable in women than in men (Table 2).

In all 3 groups, the BMI delta values — representing the degree of weight fluctuation — were only high during the initial 3 weeks (Figure 4). The group that lost >10% weight had larger BMI delta values during the entire study period. The difference in BMI delta values among the 3 groups was the most statistically significant during the first several weeks. However, most of the differences in values became insignificant over time (Table 2). Moreover, the BMI delta was more statistically significant in women (Table 2).

As seen in Figure 5, the number of meal history entries decreased over time in all 3 groups (slope values of −0.67, −0.77, and −0.79 for the groups with >10% weight loss, 5%-10% weight loss, and <5% weight loss, respectively). The 2 groups that reduced their weight by ≥5% had a higher total number of meal history entries than the less successful group, and the differences in total number of meal history entries among the 3 groups were statistically significant during most of the study period (Table 2). No significant differences in the total number of meal history entries were found for men (Table 2). On the other hand, for women, significant differences were found during most of the study period (Table 2).

Table 3 shows food intakes divided into the green, yellow, and red groups for 16 weeks for women. Women had many statistical differences between the yellow and red food groups, especially at lunch and dinner. In particular, when the intake of foods in the red category was averaged over all study periods, statistical differences between the 3 groups were significant for all meal types (Table 3). However, for men, there were no statistical differences between weight loss outcome groups and food groups over the 16 weeks.

Previous studies showed that the degree of adherence to self-reporting in mobile weight loss intervention programs decreases over time [4,6,23]. Going one step further, this study found that adherence decreases over time regardless of weight loss outcomes. Specifically, both the total weight and meal history entries decreased gradually over 16 weeks (Figures 2 and 5). Prior studies had assumed that mobile technology utilization could overcome the low adherence to self-reporting that occurs when utilizing rudimentary paper tools during weight loss interventions [6,13]. Although mobile technology clearly makes self-reporting easier, the results of this study revealed that there is a natural tendency to lower one’s adherence to self-reporting over time regardless of weight loss outcomes.

Among the many factors that hinder long-term adherence, the most relevant to this study may be the lack of serious awareness of current health conditions [37,40,41]. In general, most people are not likely to feel any ominous threats even if they do not lose their weight right away, and they are prone to stop dieting. Therefore, to promote long-term adherence, it is necessary to provide an environment in which diet participants recognize the critical importance of weight management [42,43]. Given the lessons learned from previous diet studies that have shown that digital push notifications have no effect on adherence, reminding participants regularly about the importance of dieting may be ineffective [44,45]. Instead, there seems to be a need for further research to provide divergent strategies for people with obesity.

Conversely, it may be entirely reasonable to simply admit that humans are less committed to interventions over time, especially for slow and arduous tasks like trying to lose weight. In other words, it may be wise to avoid long-term adherence issues that are unlikely to be solved. Rather, the best practice may be to focus on understanding the characteristics of adherence with latent patterns. Recent data science algorithms may be employed to understand specific patterns of adherence. Specifically, the use of time series decomposition algorithms that eliminate the inevitable long-term declines in adherence may be used for exploring repeated patterns of adherence [46,47]. Then, demonstrating the association between the pattern and weight loss outcomes can provide new insights.

According to the results of this empirical analysis, BMI changes can be characterized in 3 time series periods (Figures 3 and 4). The first interval is from week 1 to week 3, where the overall BMI delta value started high and fell rapidly over time and the overall BMI value was high in the initial stages. High values for both BMI and BMI delta may indicate substantial weight fluctuations with low weight losses within a week. This phenomenon may be due to the participants' biological responses to homeostasis (ie, steady weight) [27]. It takes time for the human body to adapt to a diet (ie, limiting caloric intake and increasing energy expenditure through exercise) [27,48,49]. During this adaptation process, human weight can fluctuate based on a variety of factors, including hormones, psychology, and physiology [21-23]. Although the period of adaptation may vary depending on certain factors (eg, race, diet style), several empirical studies have found that it takes human bodies approximately 2 to 4 weeks to adapt to a diet [48,49]. Therefore, participants in this study might have experienced weight fluctuations during the dietary adaptation process period.

The second interval is between weeks 3 and 13. This interval was characterized by differences in BMI that increased over time among the outcome groups. Further, most BMI deltas were statistically different among the groups (Figures 3 and 4, Table 2). In this period, the differences in weight loss between the successful and unsuccessful diet groups were noticeable. Specifically, the successful group had high BMI deltas and low BMI values, which may indicate that the successful participants lost weight. On the other hand, the unsuccessful group during the same period had low BMI deltas and high BMI values, thereby suggesting that the unsuccessful participants rarely lost any weight at all.

The last interval is between weeks 13 and 16. This interval was characterized by no differences in BMI deltas between the groups (Figure 4). Another feature during this period was the manifestation of weight increases in the group that lost the most weight (>10%). This may indicate that the group members stopped trying to lose weight, because people do not lose weight beyond their ideal weight. From a physiological point of view, the feature may be attributed to the widely known “yo-yo” effect that commonly occurs after rapid weight loss [45-47]. The yo-yo effect is defined as a rapid weight gain after a diet. Indeed, many previous studies have shown that the majority of diet participants regain their weight over the long term [45-47]. In this study, the yo-yo effect may have occurred in participants who had dietary laxity as the mobile weight loss intervention program was nearing its end in week 16.

Gender effects on the differences in self-reporting adherence have often been discussed in previous studies of weight loss interventions [4,6,22]. Going one step further, this study explored the time-dependent relationships among adherence to self-reporting, gender effects, and weight loss outcomes.

Unlike female participants, male participants had little differences in the number of weight entries among the outcome groups during the 16 weeks (Table 2). Furthermore, in terms of the number of meal history entries, there were no statistical differences in the values among the outcome groups for male participants (Table 2). However, statistical differences in BMI values among the 3 groups were found around the last several weeks (Table 2).

These results suggest that there was no statistical association between adherence to self-reporting and weight loss outcomes in the male population. This contradicts some previous studies that showed positive relationships between the two [6,7,50]. Inconsistencies in the results can be addressed as follows. Considering that the tendency to self-report is influenced by various factors, such as gender and race [6,22,51], Korean adult men may not be as inclined to self-report. In other words, they may have avoided timely self-reporting regardless of the success of the diet. As self-reporting itself has no direct effect on weight loss, male participants might have been able to lose weight without consistent self-reporting if they managed their diet well and consistently exercised.

Another interesting finding in this analysis is that the male participants, unlike the female participants, demonstrated no statistical difference in the intake of food types among the 3 weight loss groups. Previous studies found that men were insensitive to food intake during their diets, while women were very sensitive [52,53]. Therefore, the results may have captured the food intake characteristics of dieting men during mobile interventions. However, it should be recognized that these characteristics may also depend on other factors such as age or ethnicity [17–19]. Therefore, further research is needed in different settings to identify the factors that may contribute to gender differences in self-reporting and weight loss outcomes.