Abstract
Evidence supports the use of technology supported multicomponent interventions for promoting weight loss. Many such programs include the opportunity to synchronously pair commercially available physical activity trackers (PA-T) with a goal to enhance weight loss outcomes. However, little is known about the reach and effectiveness of allowing participants to pair a PA-T within an existing online program. Purpose: This matched cohort, quasi-experimental study aimed to determine 1) the proportion of participants that pair a PA-T to the existing program, 2) the representativeness of participants who pair a PA-T; 3) the relationship between pairing a PA-T, overall weight loss, and the likelihood of achieving a clinically meaningful weight loss; and 4) if pairing a PA-T with program participation is related to weight loss outcomes independently or collectively when considering other indices of program engagement. Methods: Data collected over a four-year period included demographic self-report, objective weight data uploaded when participants weighed themselves at a self-serve program kiosk available to the community, and program engagement data (e.g., logins to program website). A range of analyses, including descriptive proportions, analyses of variance, and path analyses, were used to address the purposes of the study. Results: Participants who paired a PA-T were more likely to be women (p<.001), African American (p<.001), and have a higher BMI (p<.05). Those who paired a PA-T lost on average an extra 1% body weight and were 1.4 times more likely to lose 5% body weight. Pairing a PA-T was related to other indices of online program engagement and both directly and indirectly contributed to weight loss outcomes (p<.05). Conclusions: Pairing a PA-T within an online weight loss program appeals to groups that experience disparities related to obesity and predicts improvements in weight loss. More translational studies are needed to examine the role of personal psychosocial and environmental factors that may enhance or diminish the benefit of pairing a PA-T to evidence-based, online weight loss programs.
Keywords: Web-based, sync, activity tracker, reach, engagement, eHealth
Introduction
The development of scalable, community-based weight loss programs is a public health priority [1–3]. The Community Guide to Preventive Services [4, 5] recommends technology supported, multicomponent interventions with coaching as an evidence-based approach to weight loss [6–8]. Specifically, evidence to date highlights the importance of implementing program features that promote and enhance self-monitoring, counselor feedback and communication, social support, and provide personalized/tailored feedback in the context of a structured educational/informational program [9–13].
An appealing strategy for eHealth interventions is to allow the integration of personal wearable physical activity trackers (PA-T), like those produced by Fitbit or Garmin, with online program participation. Evidence supports the benefit of adding a PA-T to standard care for weight loss [11], and for promoting physical activity among sedentary adults [14]. Allowing participants to sync their personal PA-T to program participation may be a practical way to enhance program effectiveness for a substantial proportion of program participants. Wearable PA-Ts have been consistently ranked among the top fitness trends worldwide for several years, ranking #1 in three of the past four years [15, 16]. Moreover, consumer data indicate that PA-T ownership is on the rise, climbing from just 21% in 2014 to 45% in 2016 among US respondents [17]. The popularity and widespread adoption of PA-T underscore the potential of PA-T/program pairing to reach broadly into the population of program participants, and to enhance program engagement and effectiveness [10, 18]. Little is known, however, about the implementation of this feature within an ongoing evidence-based community weight loss program. Which participants are more likely to use this feature? Do those who volitionally pair their PA-T have greater success in weight loss? Is the benefit of pairing a PA-T independent from that of other program engagement factors? These preliminary questions are critical to inform progress in program development and implementation within eHealth delivery systems.
Using an integrated research-practice partnership approach [19] we aimed to provide actionable information for an organization implementing an evidence-based online weight loss program (i.e. whether they should consider offering participants a personal PA-T with enrollment) by determining the reach and effectiveness outcomes related to volitionally pairing a PA-T with program participation. Therefore, the purpose of this study was to determine 1) the proportion of participants that pair a PA-T to the existing program, 2) the representativeness of participants who pair a PA-T relative to the population of program participants; 3) the relationship between pairing a PA-T, and overall weight loss, as well as the likelihood of achieving clinically meaningful weight loss; and 4) if pairing a PA-T with program participation is related to weight loss outcomes independently or collectively when considering other indices of program engagement, including the amount of time participants have been enrolled, and several program usage indices reflective of program engagement. Using deidentified data from participants in a widely accessible, evidence-based online weight loss program, we assessed the reach and effectiveness of the PA-T pairing feature. We tested the hypothesis that individuals who paired a PA-T would be more likely to lose clinically meaningful amounts of weight. Path analyses were used to explore the relative contribution of engagement variables to weight-loss outcomes, and to test the independence of the relationship between pairing a PA-T and weight-loss outcomes.
Methods
Weigh and Win (WAW) is an online, community-based weight loss program that has been offered to Colorado residents as a community benefit since 2011. The goal of WAW is to provide scalable, accessible, evidence-based programming, free of charge to Colorado residents. Members of the community self-enroll into WAW by creating a profile at one of the publicly available weigh-in kiosks conveniently located across Colorado, or by going directly to www.weighandwin.com. WAW is a multicomponent behavioral program which includes a website, objective weight assessment, daily email and text support with content based in principles of social cognitive theory, online and telephonic access to health coaches, and modest financial incentives for successful weight loss. Content is focused on improvements in diet (increasing intake of fruits/vegetables, lean protein and complex carbohydrates, and monitoring portion sizes) and physical activity (graduated recommendations from walking to combined strength and cardiovascular training). WAW allows participants to pair a PA-T to their participation, though the pairing option is limited to Fitbit brand PA-Ts. Additional details about the program are available at the program website, www.weighandwin.com/Help/FAQ.aspx
Earlier studies highlight the ability of WAW to reach broadly into the community and support successful weight loss in a cost-effective manner [20, 21]. Practice partners from the delivery organization, incentaHEALTH ™ (supported by Kaiser Permanente of Colorado) provided deidentified WAW participant data for years 2011–2014 to the research partners for analysis and interpretation. Individual data from paired PA-Ts (i.e. steps and physical activity data) were not available for analysis.
Program Usage Data.
Demographic information.
Age, height, gender, race, and ethnicity were self-reported during online registration.
Weight.
Participants in the WAW program are encouraged to weigh in at least once each quarter as part of their regular program participation. Private automated kiosks (n~83) set-up in convenient locations (e.g. businesses, libraries, rec centers, clinics) in and around Denver, CO are used to objectively measure weight (kg) at the participants’ discretion. Kiosks are equipped with medical-grade 600- pound capacity Tanita WB-110A scale (Class III NTEP Certified) calibrated prior to installation using a standard reference 48kg weight stack (tolerance +/− 1%). Participants are asked to be consistent in their clothing and footwear at each weigh in. If they prefer to wear shoes, they are asked that they consistently wear shoes. They are asked that they do not alter their weight by earing coats, purses, or backpacks. Weight measurements are automatically uploaded into the participant’s WAW account, eliminating the need for participants to enter their own weight data when using a WAW kiosk. Photographic audit of the weigh ins ensures protocol fidelity. Body mass index (BMI) was calculated from self-reported height and objectively measured weight as kg/m2.
Weight loss.
Total weight loss was calculated as the difference between starting weight and most recent weight. Percent weight loss was calculated by dividing total weight loss by initial weight, then multiplying by one hundred. Finally, clinically meaningful weight loss was indicated for those whose weight loss was ≥ 5% of their initial body weight [22].
Engagement variables.
Program engagement was conceptualized as participant’s observable interactions with online program features as indicated by system usage data. A binomial variable was provided that indicated whether participants had paired their PA-T with program participation. Data for nine additional variables reflecting program engagement were included in the dataset. Data were provided for total counts of (1) Facebook posts (participants are able to share content from their program dashboard—including weight loss progress and their “healthy selfie” digital photo captured by the kiosk— to their Facebook profile), (2) account logins, (3) journal entries (participants are prompted to complete an online “60 second journal” each week to document how successful they have been at overcoming their barriers and following the recommended nutrition and exercise guidance in the preceding week), (4) referrals to friends through their program dashboard, (5) online quizzes completed and (6) weight measurements. Data were also provided for (7) email-open rate (expressed as a percentage of program emails sent that were opened by each participant), (8) the breadth of content requested for delivery by text message (e.g. exercise prompts, weigh-in reminders, etc.) reflected by a count of the total number of options selected (ranging from 0–5), and (9) enrollment duration (measured as a continuous variable by subtracting the date of enrollment from the date of the most recent weigh-in). Though the program is advertised as a 12-month weight loss program, program participants may remain enrolled indefinitely.
Analysis.
Descriptive statistics, independent samples t-tests, and chi-squared tests were conducted in SPSS version 20 [23]. Multiple logistic regression and path analyses were conducted in Mplus 7.11 [24]. Data from participants who were <18 years old or who enrolled for the program but did not return for a post-enrollment weigh-in (i.e. participants who did not have at least 2 weight measurements) were excluded. In addition, some ages appeared to be entered inaccurately and as a result we excluded any participant with an age over 100 years. Analyses were otherwise inclusive in order to maximize external validity of the observations.
Participants were classified as either having a “paired” PA-T or being “unpaired”. The proportion of participants who paired a PA-T was determined using the total number of included cases from the program participant pool (i.e. all records minus those that were excluded based on the above listed exclusion criteria).
To address the representativeness of participants that paired a PA-T to the larger program population, repeated random sampling was used from the larger population of unpaired program participants to conduct multiple independent group comparisons. Three separate groups of unpaired participants of equal size to the paired group were sampled from the program population (i.e. unpaired groups 1–3). Samples were drawn independently from each other, with replacement (i.e. participants from unpaired group 1 were re-entered into the population before unpaired group 2 was sampled, and so on for unpaired group 3). Demographic differences were assessed between the paired group and each of the three unpaired groups using chi-squared and independent samples t-tests. This approach was used to test the consistency of observed effects within the program population to strengthen confidence and inform subsequent analyses. If significant relationships are not observed repeatedly across independent analyses then the representativeness of the PA-T paired group relative to the larger program population would be supported. However, if significant group differences are reproduced across independent analyses significant demographic differences between those who pair a PA-T and those who do not in the WAW program are strongly supported.
The significance and consistency of the relationship between pairing a PA-T and successful weight loss was assessed similarly; independent samples t-tests were used to test for significant differences in weight loss outcomes between the paired group versus unpaired groups 1–3. In the case of repeated observations of significant group differences in weight loss outcomes, a fourth unpaired group of equal size was randomly sampled from the sampling pool. The independence of consistent demographic predictors was tested using multiple logistic regression. Demographic predictors which retained significant relationships with group were considered independent, and therefore included as covariates in all subsequent analyses.
Path analyses were used to iteratively test the direct relationship between pairing a PA-T and weight loss outcomes, beyond any shared variance with other indices of program engagement. All path analyses were modeled using the fourth unpaired group. First, bivariate associations between engagement variables 1–9 were simultaneously estimated as partial correlations. Path analyses were then specified to identify engagement variables that were independently related with 1) pairing a PA-T and 2) weight loss outcomes. Program engagement variables were allowed to correlate freely in all models in which they were included. Models were initially specified to include all nine engagement variables. Engagement variables with non-significant relationships with the respective outcome (i.e. group or weight-loss outcomes) were excluded from model specification, and model fit was re-evaluated. In the case that more than one engagement variable significantly predicted the respective weight loss outcome at the end of these iterations, the model was re-specified to test for interactions between significant predictors. Finally, mediation was tested for the effect of group on each respective weight loss outcome through the engagement variables with which the predictor and outcome were each independently associated. Models reflecting both full and partial mediation were evaluated.
All path analyses were estimated using robust maximum likelihood estimation. Critical z- scores (parameter estimate/SE) were used to test significance of relations between variables (p<.05). Model fit was evaluated using the χ2 statistic, comparative fit index (CFI), root mean square error of approximation (RMSEA) and its 90% confidence interval (CI), and the standardized root mean square residual (SRMR) [25]. Commonly, values of CFI ≥ 0.90 are judged to be acceptable, while values >0.95 indicate good fit. Values of the RMSEA ≤0.06 and ≤0.08 typically reflect close and acceptable fit, respectively. Concurrent values of >0.95 for CFI and ≤0.08 for SRMR provide optimal protection from Type I and Type II error rates [25]. Significance tests of individual indirect paths were assessed by default in Mplus [24] using the delta method [26]. In the case of multivariate mediation, significance was assessed using the multivariate delta solution and differences between indirect paths were assessed by comparing the ratio of total indirect effect by the standard error of the differences between mediated effects with the critical z-score using an alpha of .05 [26, pg 107–8].
Results
Sample.
Of the total enrolled sample between 2011–2014 (N=56,268), 32 did not meet the age requirement for inclusion, 15,928 did not complete an initial weigh-in, and an additional 18,016 did not return for a post-enrollment weigh-in. Therefore, a total of 22,292 (80% female;73% Caucasian) adults with a mean(sd) age of 44.1(12.9) years were eligible for inclusion in the sampling pool for this study. Of those, 1,255 (6%) participants had paired a PA-T to their WAW account. Each analysis examined this paired group coupled with one of the four randomly sampled unpaired groups for an analysis level sample size of n = 2,510. Demographic descriptions of paired and unpaired groups are displayed in Table 1.
Table 1.
Demographic descriptions of the paired and unpaired groups.
| Paired group | Unpaired | |||||
|---|---|---|---|---|---|---|
| Group 1 | Group 2 | Group 3 | Group 4 | |||
| N | 1,255 | 1,255 | 1,255 | 1,255 | 1,255 | |
| M(sd) | ||||||
| Age | 43.3(11.4) | 44.4(13.0) | 43.7(12.7) | 43.8(13.0) | 44.5(12.9) | |
| Initial Weight (kg) | 92.8(21.5) | 91.4(22.7) | 91.1(22.7) | 90.7(22.0) | 91.2(22.2) | |
| Initial BMI | 33.1(7.3) | 32.5(7.8) | 32.4(7.2) | 32.3(7.6) | 32.5(7.2) | |
| N(%) | ||||||
| Gender | Females | 1103(89.4) | 1003(82.2) | 997(81.3) | 992(81.8) | 1019(81.2) |
| Males | 131(10.6) | 217(17.8) | 229(18.7) | 220(18.2) | 198(15.8) | |
| Ethnicity | Hispanic | 260(21.9) | 211(18.3) | 236(20.7) | 225(19.9) | 888(79.3) |
| Non-Hispanic | 926(78.1) | 941(81.7) | 905(79.3) | 904(80.1) | 234(20.7) | |
| Race | African Am. | 170(14.3) | 89(7.8) | 96(8.4) | 102(8.9) | 109(8.7) |
| Asian | 12(1.0) | 23(2.0) | 16(1.4) | 22(1.9) | 21(1.7) | |
| Caucasian | 959(80.7) | 918(80.0) | 909(79.6) | 907(79.6) | 908(72.4) | |
| Native Am. | 10(0.8) | 17(1.5) | 17(1.5) | 18(1.6) | 12(1.0) | |
| Other | 38(3.2) | 100(8.7) | 104(9.1) | 91(8.0) | 82(6.5) | |
Sub-categories may not sum to equal total group sizes due to data missing at random.
Demographic predictors of PA-T use.
Group comparisons revealed significant differences in gender (χ2(df)≥25.92(1), p<.001), race (χ2(df)≥44.44(4), p<.001), and initial BMI (t(df)≤ −2.11(2508), p<.05) between paired and unpaired groups. Initial weight (t(df)=−2.39, p=.017) significantly differed from the paired group in only one of the unpaired groups and was therefore not considered a consistent predictor.
Independence of these predictors was tested with the fourth unpaired group. Multiple logistic regression predicting group, by gender, race, and initial BMI supported independent effects for each predictor variable, such that the paired group was more likely to be female (β(SE)=0.49(0.11), p<.001) and African American (β(SE)=0.85(0.21), p<.001), and have a higher initial BMI (β(SE)=0.29(0.12), p=.013) than the unpaired groups.
Effect of PA-T use on weight loss outcomes.
Table 2 displays the mean(sd) absolute and relative weight loss, as well as the number and proportion of participants in each cohort achieving 5% loss of body weight in each group. The paired group had significantly greater mean weight-loss (t(df)≥3.08(2508), p≤.002) and percent weight loss (t(df)≥2.89(2508), p≤.004), and a greater proportion of members reaching 5% weight loss (χ2(df)≥13.63(1), p<.001) than unpaired groups 1–3. Odds ratios (95% CI) indicate that the paired group was 1.39(1.17,1.65) to 1.46(1.22,1.73) times more likely to lose 5% body weight than the unpaired groups.
Table 2.
Weight loss outcomes and group comparisons of PA-T paired and unpaired groups 1–3.
| Paired group | Unpaired groups | |||
|---|---|---|---|---|
| Group 1 | Group 2 | Group 3 | ||
| M(sd) | ||||
| Weight loss (kg) | 3.97(7.90) | 2.89(6.19)*** | 2.65(6.02)*** | 3.07(6.62)** |
| Weight loss (%) | 3.97(7.51) | 2.99(6.30)*** | 2.74(6.32)*** | 3.15(6.57** |
| N(%) | ||||
| Lost 5% body weight | 407(32.4) | 321(25.6)*** | 311(24.8)*** | 323(25.7)*** |
Differences in means and proportions are significant at *p<.05; **p<.01; ***p<.001.
Positive means reflect a reduction in weight and BMI.
Independence of these effects was tested with the fourth unpaired group. Group effects on weight loss outcomes remained significant after controlling for gender, race, and initial BMI (.10≥ γ ≥.04,SE≤.02, p≤.001), though this only accounted for a significant amount of the variance in achieving 5% weight loss (R2=0.01, p=.012), not in absolute or relative weight loss (p>.05).
Path analyses.
Model fit statistics for all non-mediational path analyses are listed in Table 3.
Table 3.
Model fit statistics for non-mediated path analyses examining relationships between engagement variables and weight loss outcomes.
| Χ2(df) | RMSEA(90%CI) | CFI | SRMR | |
|---|---|---|---|---|
| Engagement variable partial correlations | 54.69(6) | .057(.044, .071) | .959 | .016 |
| Engagement variables on PA-T group | 49.63(6) | .054(.041, .068) | .970 | .014 |
| Non-sig predictors removed | 45.89(17) | .026(.017, .035) | .977 | .044 |
| WL outcomes on engagement variables | ||||
| Total weight loss | 54.73(6) | .057(.044, .071) | .774 | .016 |
| Non-sig predictors removed | 15.88(6) | .027(.011, .043) | .925 | .019 |
| % weight loss | 28.96(6) | .039(.026, .054) | .851 | .013 |
| Non-sig predictors removed | 3.22(3) | .005(<.001, .035) | .996 | .008 |
| Achieved 5% weight loss | 54.65(6) | .057(.044, .071) | .835 | .015 |
| Non-sig predictors removed | 29.03(6) | .039(.026, .054) | .896 | .019 |
| Journal count also removed | 29.03(6) | .039(.026, .054) | .901 | .021 |
| Testing for interaction between sig predictors | 52.10(12) | .038(.028, .049) | .819 | .029 |
Sig: Significant. WL: weight loss. RMSEA: Root mean squared error of approximation. CFI: Comparative fit index. SRMR: Standardized root mean squared residual.
Group and program engagement variables.
Results of the path analysis specifying partial correlations between all program engagement variables while controlling for gender, race, and initial BMI are displayed in Table 4, along with means (sd) for the final analysis sample (i.e. paired and fourth unpaired groups).
Table 4.
Group level mean(sd) of, and parital correlations between program engagement variables.
| Paired group M(sd) | Unpaired group 4 M(sd) | FB posts | Email rate | Logins | Jrnl count | Referral count | Weigh-ins | Quiz count | Text prompts | |
|---|---|---|---|---|---|---|---|---|---|---|
| Facebook posts | 11.4(55.9) | 1.6(18.9) | ||||||||
| Email-open rate | 45.2(34.7) | 22.9(31.3) | .11*** | |||||||
| Logins | 34.2(85.3) | 17.1(61.5) | .46*** | .18*** | ||||||
| Journal count | 5.7(27.1) | 5.8(53.0) | .14*** | .08** | .60*** | |||||
| Referral count | 1.2(8.9) | 0.9(5.2) | .14* | .02 | .22*** | .09* | ||||
| Weigh-ins | 11.4(19.4) | 7.0(10.0) | .39*** | .12*** | .53*** | .15*** | .14* | |||
| Quiz count | 11.5(28.7) | 6.3(14.7) | .33*** | .15*** | .71*** | .48*** | .20** | .42*** | ||
| Text prompts | 2.5(1.9) | 2.2(1.6) | −.04* | −.01 | −.02 | .03* | .06*** | −.01 | .02 | |
| Enrollment duration (yrs) | 1.0(1.0) | 0.7(.9) | .17*** | .06** | .40*** | .22*** | .14* | .44*** | .47*** | .04* |
FB: facebook. Jrnl: journal.
p<.05
p<.01
p<.001.
Bivariate associations were estimated simultaneously, and are controlled for gender, race, and initial BMI.
In the model estimating the direct effect of group on program engagement variables, group significantly predicted seven engagement outcomes. The paired group had, on average, 9.9(1.71) more Facebook posts, 22.24(1.37)% higher email-open rate, 17.07(3.03) more logins, 5.13(1.06) more quizzes completed, 4.48(0.62) more weigh-ins, requested .30(.07) more text prompt types, and were enrolled for 3.06(0.41) months longer than the unpaired group (p<.001). Model fit was acceptable with all nine exogenous variables and became even more favorable with the removal of journal count and referral count, which were not significantly related to group.
Program engagement variables and weight loss outcomes.
Total weight loss.
In the model specifying direct effects of engagement variables on total weight loss, only email-open rate was found to be a significant predictor (γ(SE)=.05(.02), p=.007). When non-significant predictors were trimmed from the model, model fit improved, and email-open rate remained a significant predictor of total weight loss (γ(SE)=.07(.02), p<.001), though the model did not account for a significant amount of the total variance in weight loss (R2=.01, p=.062).
Percent weight loss.
Percent weight loss was significantly predicted by email-open rate (γ(SE)=.06(.02), p=.006). After trimming non-significant predictors from the model, model fit improved, and email-open rate remained a significant predictor of total weight loss (γ(SE)=.08(.02), p<.001), though the model did not account for a significant amount of the total variance in weight loss (R2=.01, p=.067).
5% weight loss.
Email-open rate (γ(SE)=.05(.02), p=.009), journal entries (γ(SE)= −.06 (.03), p=.025), and enrollment duration (γ(SE)=.20(.03), p<.001) were supported as significant independent predictors of achieving 5% weight loss in the initial model. After removing non- significant predictors, model fit improved, and the effect of journal entries was no longer significant (γ(SE)= −.02 (.02), p=.431). Removal of journal entries from the model improved fit and supported positive effects of email-open rate (γ(SE)=.06(.02), p=.002) and enrollment duration (γ(SE)=.23(.02), p<.001) on achieving 5% weight loss (R2=.06, p<.001). The model testing for an interactive effect of email-open rate and enrollment duration on achieving 5% weight loss did not support significant interaction (β(SE)= −.04(.04), p=.247).
Mediated models.
Model fit statistics for all mediational path analyses are listed in Table 5.
Table 5.
Model fit statistics for mediated models of group and engagement variables on weight loss outcomes.
| Χ2(df) | RMSEA(90%CI) | CFI | SRMR | |
|---|---|---|---|---|
| Total weight loss | ||||
| Partially mediated model | 3.66(3) | .009(<.001, .036) | .999 | .012 |
| Fully mediated model | 10.58(4) | .026(.007, .045) | .985 | .012 |
| Percent weight loss | ||||
| Partially mediated model | 3.70(3) | .010(<.001, .036) | .998 | .008 |
| Fully mediated model | 8.63(4) | .021(<.001, .041) | .987 | .011 |
| Achieved 5% weight loss | ||||
| Partially mediated model | 24.24(3) | .053(.035, .074) | .964 | .017 |
| Fully mediated model | 37.12(8) | .038(.026, .051) | .950 | .020 |
RMSEA: Root mean squared error of approximation. CFI: Comparative fit index. SRMR:
Standardized root mean squared residual.
Total weight loss.
The model specifying direct and indirect effects of group on total weight loss with email-open rate entered as a mediator demonstrated acceptable fit, and supported significant direct (γ(SE)=.06(.02), p=.008) and indirect (β(SE)=.02(.01), p=.005) effects of group on total weight loss (R2=.01, p=.039). Model fit remained acceptable when group was restricted from loading directly onto total weight loss, and a similar amount of the variance in total weight loss was accounted for relative to the partially mediated model (R2=.01, p=.048). The partially mediated model for total weight loss is illustrated in Figure 1, Panel A.
Figure 1.
Mediational models of the effect of group on weight loss (Panel A), percent weight loss (Panel B), and achieving 5% weight loss (Panel C) beyond other indices of program engagement. Values are standardized path coefficients, partial correlations, and their standard errors. e: residual error of exogenous variable. d: disturbance term of endogenous variables. Solid lines are significant (p<.05) and dashed lines are not significant (p>.05).
Percent weight loss.
The model specifying direct and indirect effects of group on percent weight loss with email-open rate entered as a mediator demonstrated good fit. Group significantly predicted percent weight loss both directly (γ(SE)=.05(.02), p=.026) and indirectly (γ(SE)=.02(.01), p=.004). Results are illustrated in Figure 1, Panel B. Though model fit of the fully mediated model was acceptable, model fit statistics indicated that his model fit less well than the partially mediated model.
5% weight loss.
The partially mediated model for predicting 5% weight loss, illustrated in Figure 1, Panel C supported direct (γ(SE)=.05(.02), p=.009) and indirect effects of PA-T use through email-open rate (β(SE)=.02(.01), p=.022) and enrollment duration (β(SE)=.03(.01),p<.001; R2=.06, p<.001). Multivariate mediation was supported (z=3.67, p<.001), and indirect effects were not significantly different (z=.51, p=.62). Model fit was similar for the fully mediated model predicting likelihood of achieving 5% weight loss (R2=.06, p<.001).
Discussion
This investigation aimed to provide a focused assessment of the reach and effectiveness of volitionally pairing a wearable PA-T within an ongoing online community-based weight loss program. Repeated random sampling from the large pool of WAW program participants was used to identify demographic characteristics that consistently and independently predicted use of the PA-T pairing feature, and whether using this feature consistently predicted weight loss outcomes. This observation was extended to test the benefit of this feature to program effectiveness while controlling for relevant demographic variables. Participants pairing a device lost on average an additional 1% body weight and were 1.4 times more likely to achieve ≥5% weight loss than those who did not pair a device. It is not surprising to have observed a significant effect of pairing a PA-T considering the wide body of literature highlighting a central role for self-monitoring in successful online weight loss interventions [9–13]. Path analyses indicated that pairing a PA-T increased the likelihood of losing 5% body weight beyond its relationship with and the effects of other engagement indices—but only by a small amount. Without additional predictor variables (e.g. environmental, psychosocial, personal) the models explained less than 7% of the variance in weight loss outcomes.
Individuals who pair a PA-T to their program participation represented a small proportion of all program participants. This should not be taken to indicate that other program participants are not using commercially available PA-Ts to help them reach their goals. Rather, pairing a PA-T was uncommon for program participants in the years 2011 through 2014. There may be participants who use a PA-T that wasn’t compatible with the WAW platform (since this analysis WAW added a ‘device agnostic gateway’ to allow for over 100 apps and devices for synchronization), or who may simply not want to pair a device to an online account. Indeed, there are likely a wide range of factors contributing to the uptake of this feature that were not examined in this study.
Individuals who paired a PA-T with their program participation had a higher initial BMI and were more likely than those who did not pair a PA-T to be women and/or African American. A previous report of the representativeness of the WAW program indicated that there is a greater proportion of women and African Americans in the WAW program compared to the general population of the surrounding area [20]. Results therefore support the idea that pairing a wearable PA-T within the WAW program is occurring within a proportion of the participants that are representative of the larger community engaging in the program.
Given that the population of program participants is predominantly female [20], the consistent gender differences in our paired vs unpaired comparisons may be a reflection of greater engagement by females in general. Analyses focused on gender differences in program usage data would be needed to test this possibility. The observation that those who naturally paired a PA-T were more likely to be African American and have a higher initial BMI than those who did not is encouraging from a translational perspective. African Americans experience weight-related health disparities [27], and are an important target population for health behavior interventions. We hypothesize that offering a compatible PA-T with enrollment could result in a greater overall proportion of African American participants who pair a PA-T and potentially contribute to closing health disparities. Future randomized controlled trials will be necessary to test this hypothesis.
Partial correlations between program engagement variables demonstrated that the more an individual engages with any given program feature (i.e. program emails, journal entries, Facebook posts, etc.), the more likely they are to engage with other program features. Participants who paired a PA-T were more likely to use program features more often, across the board. A more focused examination of program engagement patterns was required to parse out whether this feature provides any additional benefit to its adopters beyond its association with participants’ engagement with other program features. Using path analysis, we were able to untangle the unique contributions of engaging with each specific program feature to weight-loss outcomes, while controlling for the shared variance between engagement indices, and relevant demographic variables. Frequency of opening email correspondence and (in the case of achieving 5% weight loss) overall duration of participation accounted for the effect of program usage variables on weight loss outcomes. These results indicate that among this sample of program participants, significant bivariate associations between the other seven engagement indices and weight loss are explained by their relationship with the participants’ level of engagement with program email content and overall enrollment duration.
The observed effect of enrollment duration corresponds with more comprehensive evaluations of community weight loss programs [20]. The role of email engagement in explaining variance in weight-loss outcomes is less clear. Emails are sent to participants daily and may act as a primary gateway to engagement with other program features for a large proportion of participants. In this way, program usage data on email-opening rate may act as more of a proxy index for prompting additional program usage. On the other hand, the email content is based on Social Cognitive Theory and provides evidence-based support for behavior change. This may indicate that email content (e.g. tips for overcoming barriers, improving self-efficacy, and skill building for setting intentions) was a primary engagement factor driving program success. More detailed qualitative or mixed methods investigations would be necessary to make conclusions about how the email engagement influenced successful weight loss.
When modeled together, the volitional pairing of a PA-T, frequent email engagement, and enrollment duration accounted for 5–6% of the variance in successfully achieving 5% weight loss, while controlling for age, race, and initial BMI. It seems unlikely that the observed effects are due to pairing a PA-T directly, considering the similarity of fit for models predicting 5% weight loss when group was allowed to, or restricted from, loading directly onto the outcome. Further, when considering the individual path coefficients, it is unclear whether pairing a PA-T imparts a meaningful effect on weight-loss outcomes independent of other engagement indices.
Pairing a PA-T was most strongly related to email-opening rate, which contributed only modestly to weight loss outcomes. Enrollment duration demonstrated a significant relationship with clinically meaningful weight loss. The direct association between group and enrollment duration was only about half of that with email-opening rate, and the unique indirect effects did not significantly differ. This is interesting to consider, as it reflects a uniform benefit of this feature through its differential relationships with other engagement indices. A direct and independent benefit of pairing a PA-T was supported for all weight loss outcomes, such that the direct effect of group was about twice that of its indirect effect(s) through other predictor variables.
These findings suggest that there is a small benefit in weight loss for those who pair a PA-T to their WAW participation. However, we are unable to determine from this study whether the use of this feature exerts a causal effect on outcomes, or if its impact can be attributed to other factors known to influence successful behavior change not considered here (e.g. psychosocial or environmental predictors). Studies are needed that address a wider range of psychosocial (e.g. attitudes, intentions, barriers), environmental (e.g. access to recreational areas and healthful food), and personal (e.g. motivational orientation) factors in addition to program engagement data to more fully understand what strategies benefit whom the most, and why.
Results of this study contrast with a recent trial in which the addition of a wearable PA-T in a behavioral intervention that included an online component did not enhance weight loss outcomes compared to intervention delivery without the wearable PA-T [28]. The Innovative Approaches to Diet, Exercise and Activity (IDEA) trial delivered a multicomponent intervention supported by an online platform to two groups of young adults (18–35 yrs.). One group was provided a wearable PA-T (the FIT Core; BodyMedia) to use with the online component of their intervention, and the other group was not. Significant effects of time were observed for objectively measured weight, body composition, cardiovascular fitness, and PA and dietary behavior, supporting the overall effectiveness of the core intervention. Significant group by time interactions were observed for weight and percent weight change over time, such that the intervention group, which received the FIT Core PA-T, lost significantly less weight over time than the standard intervention control group. Importantly, both groups did lose a significant amount of weight over the course of the intervention [28]. Unlike these findings, our observations support a beneficial, though small, effect of pairing a PA-T to participation in an online weight loss program.
There are some notable methodological differences which might account these discrepant findings. The IDEA trial was able to eliminate the possibility that outcomes related to the PA-T were simply due to participant motivation—in our trial, we cannot rule this out. However, study participants in an RCT when compared to participants engaged in a free weight loss trial (those included in our analysis), may have used the wearable PA-T in obligation to protocol fidelity. There are also differences in device type that could explain differences in results. WAW during the study period (2011–2014) interfaced exclusively with FitBit, whereas the IDEA trial used the FIT Core, by BodyMedia. These two PA-Ts have notable differences. The FitBit is wrist worn and has reputability as a popular brand of commercial activity tracker. It can be worn continuously without drawing unwarranted or unwanted attention. In contrast, the FIT Core is an armband, worn mid-bicep, and is more conspicuous and less familiar than the FitBit—which could also contribute to the differences in findings [29].
We also acknowledge a number of limitations of our pragmatic research trial. First, while our study benefits from examining the use of PA-Ts within the context typical implementation, the quasi- experimental nature of this investigation precludes us from making a causal assumption about the effect of pairing a PA-T. Though we attempted to account for individual differences in program engagement, the observed effects could result from underlying individual differences in motivation, rather than the act of pairing a PA-T itself. Second, using a technology-supported, multi-component coaching platform—while evidence-based—may not address issues related to health literacy or digital literacy [30]. There may be predictable differences in health and/or digital literacy between those who pair a PA-T and those who do not and this should be considered in future research. Third, the operationalization of program engagement as reflected by system usage data is a limited representation of a much more complex construct, which has been inconsistently reported in the literature exploring digital behavior change interventions [31]. Future work should use multiple measures of engagement that utilize mixed methods in line with recommendations by Yardley, Spring [31]. Finally, as a pragmatic study, data on psychosocial and environmental predictors of behavior change were not available for this analysis.
This focused investigation lends foundational scientific support for the inclusion of an option to pair a wearable activity monitor within the context of community- and web-based weight loss programs. Further, it provides practical value to the delivery partners by supporting their decision to consider providing program compatible PA-T to participants when they enroll. Continuing to leverage a research/practice partnership approach is crucial for progress. Individual level adoption of program features in a naturalistic setting may differ drastically from observations of feature adoption at the level of a clinical trial. Further, technological advancement in eHealth applications will continue to speed past the necessarily slow pace of evidence-based program evaluation and scientific exploration. Research/practice partnerships should continue to focus on emerging trends in the industry and emphasize the value of more time relevant reports such as that adopted by the information technology community in which conference proceedings hold more weight in the direction of continued research.
Balancing these limitations, our pragmatic trial, using existing program data, allowed for an examination of the naturally occurring use of a PA-T within a scaled evidence-based program on a large sample of participants. It provides detailed information on the prevalence of use (low), the types of participants more likely to use them (e.g., African American participants), and some preliminary path analysis to provide direction on engagement factors as a mechanism for weight loss success. It is an excellent starting point to understand how evidence-based programs and specific add-on features appeal to a wide range of participants and could contribute to ongoing program improvements [32].
Conclusion
The results reported here provide evidence that a representative subsample of the program population volitionally paired a wearable PA-T within the community-based online weight loss program, and that pairing a tracker was related to greater success with weight loss. This provides direction for future pragmatic research that can test the consistency of these relationships when a compatible PA-T is provided to participants who enroll in this type of online program. Experimental methods embedded within ongoing online programs would be ideal to examine the causal impact of pairing a wearable PA-T with online participation on weight loss. Providing a PA-T to randomly selected and assigned participants with a program like WAW would also allow continued examination of the potential proportion use of PA-Ts when the obstacle of access to a tracker is removed. Our findings support the hypothesis that providing PA-T access and encouraging pairing of the device could attract participants from populations experiencing health disparities. Finally, the use of mixed methods that assess barriers and facilitators to initial program engagement, PA-T use, and retention among demographic subgroups such as age and gender classifications will continue to improve our understanding of maximizing the impact of evidence-based interventions taken to scale in communities.
From a practice perspective, the results of this study can be used to inform decision making for community, clinical, and commercial providers of evidence-based programs such as WAW. We recommend that providers encourage the participants to pair PA-Ts to ongoing programs to support self-monitoring and increase the potential of participants to achieve a clinically meaningful weight loss. This could include a focus on African American participants and heavier participants. We do not recommend the provision of PA-Ts based on our findings. While it can be reasonably hypothesized that providing PA-Ts will increase the proportion of participants that pair a device, it remains unclear if this will lead to improved outcomes or if those improved outcomes would provide value when considering return on investment.
Acknowledgements
Dr. Estabrooks is supported in part by the National Institute for General Medical Sciences at the National Institutes of Health (U54 GM115458-01 Great Plains IDeA CTR). We would like to acknowledge Anna Taggart, MS, for her contributions to the review of literature relevant for this study. The results of the present study do not constitute endorsement by ACSM.
Footnotes
Conflict of Interest Statement
Lia Kleppe is employed by IncentaHEALTH LLC. Todd McGuire is employed by, holds the patent to, and is a part owner of IncentaHEALTH LLC. All other authors declare that they have no conflict of interest.
References
- 1.U.S. Department of Health and Human Services, Physical Activity Guidelines for Americans, U.S. Department of Health and Human Services, Editor. 2018: Washington, DC. [Google Scholar]
- 2.Dombrowski SU, Avenell A, and Sniehott F, Behavioural interventions for obese adults with additional risk factors for morbidity: systematic review of effects on behaviour, weight and disease risk factors. Obesity facts, 2010. 3(6): p. 377–396. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Dombrowski SU, Knittle K, Avenell A, Araujo-Soares V, and Sniehotta F, Long term maintenance of weight loss with non-surgical interventions in obese adults: systematic review and meta-analyses of randomised controlled trials. 2014. [DOI] [PMC free article] [PubMed]
- 4.Truman BI, Smith-Akin C, Hinman A, et al. Developing the Guide to Community Preventive Services—overview and rationale. American journal of preventive medicine, 2000. 18(1): p. 18–26. [DOI] [PubMed] [Google Scholar]
- 5.Department of Health & Human Services, USA. About the Community Guide. 2019. [cited 2019; Available from: www.thecommunityguide.org/about/about-community-guide; www.thecommunityguide.org.
- 6.“Guide to Community Preventive Services.”. Obesity prevention and control: Technology supported multicomponent coaching or counseling interventions to reduce weight and maintain weight loss. . 2010. [cited 2016 July 10]; Available from: www.thecommunityguide.org/obesity/TechnologicalCoaching.html.
- 7.Katz DL, O’Connell M, Yeh M, et al. , Public Health Strategies for Preventing and Controlling Overweight and Obesity in School and Worksite Settings: A Report on Recommendations of the Task Force on Community Preventive Services. Morbidity and Mortality Weekly Report: Recommendations and Reports, 2005. 54(10): p. 1–12. [PubMed] [Google Scholar]
- 8.Kahn EB, Ramsey L, Brownson R, et al. , The Effectiveness of Interventions to Increase Physical Activity; A Systematic Review. American Journal of Preventive Medicine, 2002. 22(4S): p. 73–107. [DOI] [PubMed] [Google Scholar]
- 9.Neve M, Morgan P, Jones P, and Collins C, Effectiveness of web-based interventions in achieving weight loss and weight loss maintenance in overweight and obese adults: a systematic review with meta-analysis. Obesity reviews, 2010. 11(4): p. 306–321. [DOI] [PubMed] [Google Scholar]
- 10.Sherrington A, Newham J, Bell R, Adamson A, McColl E, and Araujo-Soares V, Systematic review and meta-analysis of internet-delivered interventions providing personalized feedback for weight loss in overweight and obese adults. obesity reviews, 2016. 17(6): p. 541–551. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Hutchesson M, Rollo M, Krukowski R, et al. , eHealth interventions for the prevention and treatment of overweight and obesity in adults: a systematic review with meta-analysis. Obesity reviews, 2015. 16(5): p. 376–392. [DOI] [PubMed] [Google Scholar]
- 12.Podina IR and Fodor LA, Critical review and meta-analysis of multicomponent behavioral e-health interventions for weight loss. Health Psychology, 2018. 37(6): p. 501. [DOI] [PubMed] [Google Scholar]
- 13.Khaylis A, Yiaslas T, Bergstrom J, and Gore-Felton C, A review of efficacious technology-based weight-loss interventions: five key components. Telemedicine and e-Health, 2010. 16(9): p. 931–938. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Sullivan AN and Lachman ME, Behavior Change with Fitness Technology in Sedentary Adults: A Review of the Evidence for Increasing Physical Activity. Frontiers in public health, 2017. 4: p. 289–289. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Thompson WR, Worldwide survey of fitness trends for 2019. ACSM’s Health & Fitness Journal, 2018. 22(6): p. 10–17. [Google Scholar]
- 16.Thompson WR, Worldwide Survey of Fitness Trends for 2018: The CREP Edition. ACSM’s Health & Fitness Journal, 2017. 21(6): p. 10–19. [Google Scholar]
- 17.PWC. The Wearable Life 2.0: Connected living in a wearable world. 2016. [cited 2019 8/16/2019]; Available from: https://www.pwc.nl/en/publicaties/the-wearable-life-2-0.html.
- 18.Harkin B, Webb T, Chang B, et al. , Does monitoring goal progress promote goal attainment? A meta-analysis of the experimental evidence. Psychological bulletin, 2016. 142(2): p. 198. [DOI] [PubMed] [Google Scholar]
- 19.Estabrooks PA, Harden S, Almeida F, et al. , Translating evidence-based principles into sustained physical activity programs. . Exercise and Sport Sciences Reviews., in press. [DOI] [PMC free article] [PubMed]
- 20.Estabrooks PA, Wilson K, McGuire T, et al. , A Quasi-Experiment to Assess the Impact of a Scalable, Community-Based Weight Loss Program: Combining Reach, Effectiveness, and Cost. Journal of General Internal Medicine, 2017. 32(Supplement 1): p. 24–31. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Michaud TL, You W, Wilson K, et al. , Cost effectiveness and return on investment of a scalable community weight loss intervention. Preventive medicine, 2017. 105: p. 295–303. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Douketis JD, Macie C, Thabane L, and Williamson D, Systematic review of long-term weight loss studies in obese adults: clinical significance and applicability to clinical practice. International Journal of Obesity, 2005. 29(10): p. 1153–1167. [DOI] [PubMed] [Google Scholar]
- 23.IBM Corp., IBM SPSS Statistics for Windows. 2011, IBM Corp: Armonk, NY. [Google Scholar]
- 24.Muthén L and Muthén B, Mplus, in Statistical analysis with latent variables. Version. 2012.
- 25.Hu L.t. and Bentler PM, Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 1999. 6(1): p. 1–55. [Google Scholar]
- 26.MacKinnon DP, Introduction to Statistical Mediation Analysis. 2008, Taylor and Francis Group, LLC: New York, NY: p. 79–126. [Google Scholar]
- 27.Cossrow N and Falkner B, Race/ethnic issues in obesity and obesity-related comorbidities. J Clin Endocrinol Metab, 2004. 89(6): p. 2590–4. [DOI] [PubMed] [Google Scholar]
- 28.Jakicic JM, Davis K, Rogers R, et al. , Effect of wearable technology combined with a lifestyle intervention on long-term weight loss: The idea randomized clinical trial. JAMA, 2016. 316(11): p. 1161–1171. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Klasnja P and Hekler EB, Wearable Technology and Long-term Weight Loss. JAMA, 2017. 317(3): p. 317–318. [DOI] [PubMed] [Google Scholar]
- 30.Sarkar U, Karter A, Liu J, et al. , The Literacy Divide: Health Literacy and the Use of an Internet-Based Patient Portal in an Integrated Health System—Results from the Diabetes Study of Northern California (DISTANCE). Journal of Health Communication, 2010. 15(sup2): p. 183–196. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Yardley L, Spring B, Riper H, et al. , Understanding and promoting effective engagement with digital behavior change interventions. American journal of preventive medicine, 2016. 51(5): p. 833–842. [DOI] [PubMed] [Google Scholar]
- 32.Short SE and Mollborn S, Social determinants and health behaviors: conceptual frames and empirical advances. Current Opinion in Psychology, 2015. 5: p. 78–84. [DOI] [PMC free article] [PubMed] [Google Scholar]