Retrospective cohort data were extracted directly from Noom’s database in January 2019 and deidentified upon institutional review board approval from Albert Einstein College of Medicine. Noom is an mHealth behavior change lifestyle intervention that provides users with tracking features for food and exercise logging and weighing-in as well as access to a virtual 1:1 behavior change coach, support group, and daily curriculum that includes diet-, exercise-, and psychology-based content [20,21].

Participants were initially recruited by joining the Noom program in the app store (iTunes/Google Play). Informed consent to participate in research is provided by users during the initial sign-up for the program; users can choose to opt out of providing informed consent for research. Individuals in the HW program enrolled based on self-interest in weight loss and purchased the program for $129 for 4 months, on average. Individuals in the CDC-recognized DPP program, however, were encouraged to join following a prediabetes diagnosis from their health care provider and were offered the program for free through a health insurance offer. All Noom users are assigned a virtual health coach who successfully completed a CDC-recognized training course and are placed in a virtual group led by their coach. Users in both programs have access to the same features; the only difference that exists is the program curriculum users receive. While both programs focus on weight, healthy eating, and physical activity, the DPP program includes specific diabetes prevention content stemming from the CDC’s original DPP, which is not emphasized in the HW program.

Inclusion criteria were adults aged 35 years and older who began the HW or DPP program in June 2016 through January 2019 and had at least 1 program action within the first week of the program. The decision for selecting 35 years as the age minimum was made as a qualification of middle-aged adults whose degree of technology interaction was minimal during their youth. Users were considered ineligible and were excluded from the analyses if they self-reported a BMI categorized as underweight (<18 kg/m²) or normal weight (18.5 to 24.9 kg/m²; Figure 1) or were using the free version of Noom, as they do not have access to all program features (ie, no health coach, limited article content and tracking capabilities) and thus they did not receive the full intervention. Users were also excluded from analyses if they had inaccurate self-reported measures (as determined by large fluctuations in weight [ie, ±20 kg in 1 week]), test accounts (used by engineer developers at Noom to test the product), missing gender, and duplicate accounts (caused by errors with data extraction). Our study sample size is based on users who met the inclusion criteria.

The primary outcome was self-reported weight, observed at baseline and weeks 16 and 52. To account for missing data at weeks of interest, 2-week ranges were observed around each time point and the mean of each range was used to calculate the final weight outcomes included in the analyses.

Engagement was observed in two ways. First, a definition was created to observe completion status. With the CDC’s DPP session attendance definition as a frame, it was decided to further adapt this previously used definition, originally created for in-person DPP programs, to improve applicability to mHealth. Therefore, program starters were considered as those who attended at least 1 session, defined as reading 1 article per week over 3 consecutive weeks or more during week 2 to week 6 and weighing in at least once per week for 2 weeks or more during week 2 to week 6. Program completers were considered to be those who read at least 14 articles (60%), a minimum of 1 per week, during the first 24 weeks of the core curriculum.

Engagement was also measured by users’ self-reported and behavioral-based program actions. Engagement variables included the number of self-reported meals logged, exercise logged, minutes of exercise logged, and frequency of weigh-ins, as well as behavioral-based steps recorded, articles completed (articles assigned divided by articles read), group interactions (group posts and comments), and messages to their individual coach, all tracked based on user program activity. The total value within each engagement variable was summed from baseline to week 52 and dichotomized (0 or 1); a score of 1 was given if a user logged at or above the 75th percentile cutoff for the individual variable. Composite scores for each user were calculated for all 9 engagement variables (score range of 0 = low engagement to 9 = high engagement).

Descriptive statistics were calculated for users’ baseline characteristics and expressed in means and standard deviations for continuous variables and frequencies and percentages for categorical variables (Table 1). Differences between demographics at baseline were observed using t tests for independent samples, chi-square analyses, and other nonparametric tests.

Linear mixed effects models evaluated changes in our primary outcome (weight). Linear mixed effects models estimate missing data within the analysis and are robust to data missing at random and not at random [22]. In our dataset, 2030 users of 14,676 recorded their weight at week 16 (±2 weeks) and 431 recorded their weight at week 52 (±2 weeks). Despite missing values and completion statuses, data from all users in the program were analyzed, and weight outcomes were predicted from the linear mixed models conducted.

Three analyses were completed. First, fixed effects were time and curriculum and their interaction to observe potential effects of curriculum. Second, age and time and their interactions were added, in addition to adjusting for curriculum, if found significant. Next, age, time, and total engagement and their interactions were included in the model. Time and the intercept for each participant were included as random effects in all models. Time was conceptualized as a 3-level categorical variable (week 0, 16, and 52). A first-order autoregressive covariance matrix yielded the best fit model for the repeated effect of time, using visual inspection and the Akaike information criteria. Significance tests were 2-sided with α set at .05. SPSS Statistics version 23 (IBM Corp) software was used to analyze the data.

Baseline characteristics are included in Table 1. Of the individuals selected at baseline from Noom’s database, 15.07% (2225/14,767) met criteria for starter in both the HW and DPP programs (Figure 1). Of the those who started both Noom programs, 43.51% (968/2225) of individuals completed the program (577/968 [59.6%] in HW and 391/968 [40.4%] in DPP). In the HW program, 88.27% (10,926/12,378) of participants were women, with a mean BMI of 32.6 (IQR 29.0 to 37.6) kg/m². In the DPP program, 87.90% (2100/2389) of participants were women, with a median BMI of 32.2 (IQR 29.0 to 37.6) kg/m².

DPP users were significantly older (median 51.0 [IQR 44.0 to 58.0] years) than HW users (median 42.0 [IQR 38.0 to 47.0] years; P<.001). Although the omnibus test suggested completion status differed between DPP and HW users (χ²₃=520.93; n=14,767; P<.001), post hoc analyses yielded no significant differences with Bonferroni corrections (P<.006). DPP users were significantly taller (mean 167.2 [SD 9.5] cm) than HW users (mean 165.9 [SD 7.1] cm, t₁₄₇₆₅=–7.61; P<.001). HW users had significantly higher baseline BMI (median 32.6 [IQR 29.0 to 37.6] kg/m²) than DPP users (median 32.2 [IQR 29.0 to 37.0] kg/m², P=.01). No other demographic characteristics significantly differed between curriculum groups (Table 1). The total sum of mean engagement variables for HW and DPP users across the study are found in Table 2.

Prior to running the mixed models, we observed weight loss throughout the program from users who provided data at week 16 and week 52 to better identify the amount of weight lost compared with CDC standards. Results showed that users who completed (as defined by our completer definition) the HW program lost on average 4.74 (SD 4.66) kg or 3.5% of their body weight at week 16 and 6.24 (SD 6.73) kg or 5.2% of their body weight at week 52. Users who completed the DPP program lost on average 5.61 (SD 8.06) kg or 5.7% of their body weight at week 16 and 5.66 (SD 7.16) kg or 8.1% of their body weight at week 52.

Tables 3 to 5 provide estimates and confidence intervals for the linear mixed effects models, with weight as the outcome. Results from the linear mixed model revealed that there was a significant interaction effect between curriculum groups and time (F_2,1401.0=29.44; P<.001; Table 3). From baseline to week 16 and baseline to week 52, individuals in the DPP curriculum showed greater weight loss compared with HW users, losing 3.20 kg more at week 16 and 2.38 kg more at week 52 (baseline vs week 16: β=–3.20, 95% CI –4.02 to –2.37; baseline vs week 52: β=–2.38, 95% CI –4.17 to –0.59; Table 3). Therefore, the remainder of the models were adjusted for curriculum.

When we evaluated the effect of age, we found the interaction effect between age and time was significant (F_2,1398.4=9.20; P<.001; Table 4). From baseline to week 16, adults who were older lost more weight earlier on compared with younger adults, such that for each additional year in age, weight decreased by an additional 0.11 kg (baseline vs week 16: β=–.11, 95% CI –0.16 to –0.06). However, from baseline to week 52, age was not a significant predictor of weight (baseline vs week 52: β=.003, 95% CI –0.11 to 0.11; Figure 2).

The last model evaluated interactions between age, time, and engagement, adjusting for curriculum. The 3-way interaction between age, engagement, and time was not significant and was removed from the model. Two-way interactions of engagement and time, age and time, and age and engagement were left in the model. A significant engagement effect (F_1,15238.5=14.6; P<.001) was modified by the interactions between engagement and time (F_2,1368.7=4.98; P=.01), age and engagement (F_1,14679.5=6.70; P=.01), and age and time (F_2,1351.7=10.37, P<.001; Table 5). In general, higher engagement was associated with lower weight over the entire study. At week 16, engagement was not yet significant as a predictor of weight (baseline vs week 16: β=–.13, 95% CI –0.27 to 0.00); at week 52, engagement was a significant predictor of weight such that as engagement increased by 1 composite score, weight decreased by 0.44 kg (baseline vs week 52: β=–.44, 95% CI –0.74 to –0.14; Figure 3).

The strength of the association between engagement and weight across the study differed by age; engagement was more strongly associated with weight for younger versus older adults (age and engagement interaction β=.02, 95% CI 0.01 to 0.04). Younger adults lost more weight when engaged; however, older adults lost weight over time despite their level of engagement.

As found in the prior model, age was associated with weight loss such that higher age was associated with greater weight loss at week 16 (β=–.12, 95% CI –0.18 to –0.07) but not at week 52 (β=–.02, 95% CI –0.14 to 0.09). Older adults lost more weight earlier on compared with younger adults such that for each additional year of age, weight decreased by an additional 0.12 kg.

This study explored the effect of age and engagement in predicting weight in a mobile intervention. To our knowledge, this is the first quasi-experimental study to consider age effects strictly in an mobile lifestyle intervention.

In support of our main hypothesis, higher age was associated with greater weight loss; older users lost more weight from baseline to week 16. Our second hypothesis that higher engagement would be associated with greater weight loss was supported, while our hypothesis that older users would engage more than younger adults was not found. Higher engagement was predictive of greater weight loss; however, the strength of the association differed by age. Although younger age was associated with engagement in predicting weight, older adults lost more weight from baseline to week 16 despite their level of engagement. These findings demonstrate that not only do older adults lose weight from mobile interventions, but they may benefit more compared with their younger counterparts.

mHealth interventions are used by older adults and appear to be an effective approach to weight loss. A meta-analysis by Valenzuela et al [23] of electronic health exercise programs for older adults (aged 67 to 86 years) yielded promising findings of technology as a well-accepted method, with the mean adherence as 91.3%. This is consistent with Svetkey et al [24], who found adults aged 60 years and older had both greater initial and sustained weight loss over 3 years compared with younger adults (aged younger than 50 years and between ages 51 and 60 years) in both counseling and internet-based intervention groups.

In our study, older age was not associated with engagement in predicting weight, which is not consistent with the DPP findings where adults aged 60 to 85 years attended nearly twice as much as adults aged 25 to 44 years [16]. Honas et al [25] found in a clinic-based weight loss program that younger adults were the only age group with an association with dropout (76% of individuals aged 51 to 60 years completed the program compared with 60% of participants aged 40 years and younger). A meta-analysis showed that in 13 studies, younger age was associated with higher attrition in weight loss interventions [26]. Additionally, adults aged 65 years and older were found to have higher self-monitoring rates and attend more sessions compared with younger adults (aged younger than 65 years) in an adapted DPP intervention [27]. One possible explanation for these findings is that older adults have a lack of work or family responsibilities (ie, fewer work demands). However, because we found program engagement mattered particularly for younger adults in our mobile intervention, these differences in findings compared with previous works may point to unique impacts of the use of technology, which younger adults likely have more experience with. Older adults experienced weight loss despite their total engagement, whereas younger adults benefited more from weight loss when they were engaged more with the program. It is likely that perception or presence of declining health may serve as a motivator for the aging population that extends beyond level of engagement to the mobile program. Further research should explore underlying motivators of engagement across age groups in mHealth interventions.

Our results showed that only 15% of users extracted from Noom met criteria for starters. One reason for this is that while we aimed to incorporate key engagement indicators, it is possible our definition may not capture true engagement within the program; thus, results may change with a different definition. Therefore, better mHealth definitions of engagement are needed. Dropout rates of 6% to 37% are common in mobile weight loss and diabetes interventions [28]; however, our high numbers particularly early in the program are likely related to a 2-week free trial period offered within the HW program at the time of extraction. More users may have joined who were not committed to long-term behavior change.

Throughout the 52 weeks, participants lost on average 6.24 (SD 6.73) kg or 5.2% of their body weight in the HW program and 5.66 (SD 7.16) kg or 8.1% of their body weight in the DPP program. These results meet the CDC standards that state that individuals who lose 5% of body weight or more can benefit from reduced risk for chronic diseases related to obesity [29]. Further research is needed to explore the feasibility of participants’ experience with technology interventions to better understand potential barriers that may exist. Scheibe et al [30] showed that older adults reported difficulty in understanding the functionality of the apps’ touch sensitive areas and that the visual representations were too small to be easily visible as reasons against using mobile diabetes interventions. As findings did not show a strong interaction of age and engagement for older adults, it is likely that barriers exist that affect the overall feasibility of the mobile intervention, requiring adaptations to enhance the users’ experience.

Participants were self-selecting and results may not generalize to populations with less intrinsic interest in weight loss. As our study is observational, the effect of the intervention against a control group is unknown. We decided to use initial weight versus first weigh-in as our baseline weight, given missing data concerns. It is likely the initial weight input at the time of sign-up may not reflect a true weight on a scale, as it is hypothesized many users estimate how much they believe they weigh during the sign-up phase. Third, completion status criteria for never engaged was determined based on overall engagement in week 1. Therefore, participants who were excluded may have engaged in later weeks. Additionally, as some forms of engagement included self-reports, it is hard to distinguish if a lack of exercise logged reflects a lack of exercise versus a lack of reporting. Therefore, behavioral-based engagement steps recorded are more likely to indicate a true level of engagement. Fourth, potential bias in motivational differences likely exists between users in HW and DPP, as users paid for the HW program versus users who received the DPP program for free.

Because of the retrospective design, it was not possible to assess whether users had a prediabetes or diabetes diagnosis in the HW program. The HW program is available to anyone who is able to afford it and owns a smartphone; thus, users may have additional underlying health conditions that were unknown. Finally, as mentioned previously, it is likely the CDC’s definition of attendance does not directly apply to mHealth interventions and may not have optimally captured the true findings of dropout rate or completers of the program.

In conclusion, age and engagement appear to play a significant role in predicting weight while using a mHealth lifestyle intervention at weeks 16 and week 52 in this study. Not only did older adults lose more weight from baseline to week 16, but they may benefit more compared with younger adults. Further analyses are needed to explore potential age differences to better optimize older adults’ experience within a mobile intervention.