The objective of this study was to evaluate the effectiveness of the Fun for Wellness (FFW) intervention to increase physical activity in adults with obesity in the United States in a relatively uncontrolled (ie, real world) setting. The study described in this paper was conceptualized as an effectiveness trial (ie, participants were recruited via a national health care panel recruitment company) that built upon a 2015 FFW efficacy trial completed in a relatively controlled setting (ie, participants were recruited at a major research university in the United States) [1-3]. This study is important from a general scientific perspective because the potential utility of interventions should be evaluated under both more controlled (eg, scientifically ideal: an efficacy trial) and less controlled (eg, real-world ideal: an effectiveness trial) conditions [4,5]. Before describing the FFW intervention, we begin with a summary of the 2015 FFW efficacy trial and then introduce key components in this study: target population (ie, adults with obesity), proposed outcome (ie, physical activity) and mediator (ie, self-efficacy), and the theoretical framework (ie, self-efficacy theory).

A randomized controlled trial completed in 2015 provided the initial test of the efficacy of the FFW intervention to promote well-being [1-3]. The FFW intervention was conceptualized as exerting both a positive direct effect and a positive indirect effect through self-efficacy on well-being. Data collection occurred within a relatively controlled environment (ie, adult employees at a major research university in the United States). Results provided some initial evidence for the efficacy of FFW to promote well-being self-efficacy [3]; interpersonal, community, psychological, and economic subjective well-being [1]; and interpersonal and physical well-being actions [2]. The effectiveness trial described in this paper sought to follow up on the initial evidence provided in the 2015 FFW efficacy trial.

Approximately 2 billion adults are overweight per the World Health Organization (WHO) [6]. Moreover, approximately one-third of adults who are overweight can more precisely be classified as adults with obesity and the size of this subgroup has tripled over the past few decades [6]. In the United States, more than 40% of women and 35% of men are obese [7]. From a public health perspective, this trend toward an increasing number of adults with obesity is problematic because obesity is a risk factor for major noncommunicable chronic diseases such as cardiovascular disease, type II diabetes, musculoskeletal disorders, and some cancers [8]. To reduce the prevalence of adults with obesity, the WHO recommends that individuals increase energy intake from high-quality food sources (eg, raw vegetables), limit energy intake from low-quality food sources (eg, highly processed foods high in fat), and engage in a recommended amount of physical activity for health [6]. Examples of a recommended amount of physical activity for health in adults include at least 150 min per week of moderate-intensity physical activity or at least 75 min per week of vigorous-intensity physical activity, or an equivalent combination of the two recommendations listed above [9,10]. However, there is evidence that a very small percentage (eg, <5%) of adults with obesity meet the public health guidelines for physical activity [11]. Fortunately, there is also evidence that cognitive behavioral interventions can successfully promote physical activity in adults with obesity [12,13] and in the more general adult population [9].

Physical activity has been defined as bodily movement produced by skeletal muscles that requires energy expenditure [14]. Insufficient physical activity in the adult population is a global pandemic [15,16]. Successfully addressing this pandemic will require ongoing and wide implementation of a variety of intervention strategies (eg, community-wide, informational, behavioral, social, policy, and built environment) at multiple levels of society (eg, individual, neighborhood, municipality, and country) across the globe [17,18]. At the individual level, there is evidence that behavioral interventions designed to promote physical activity by focusing on personal psychological attributes (eg, self-efficacy) can be effective [19-21]. Delivering a physical activity intervention online has been shown to be an effective mode of delivery [22,23] that also may allow for efficient scaling up of an intervention [18]. Thus, a readily scalable, Web-based behavioral intervention that effectively promotes physical activity in adults with obesity may be useful in responding to a global pandemic (ie, physical inactivity) in an at-risk population (ie, adults with obesity).

The social cognitive theory [24] has provided the theoretical framework for many effective cognitive behavioral physical activity–promoting interventions for adults with obesity [12,13]. Self-efficacy theory [25] resides within social cognitive theory and views an individual as a proactive agent in the regulation of their emotions, cognitions, and behaviors. Self-efficacy beliefs play a primary role in the self-efficacy theory and are defined as domain-specific judgments held by an individual about their ability to successfully execute differing levels of performance given certain situational demands. Self-efficacy beliefs rely upon the cognitive processing of several potential sources of efficacy information: enactive mastery experiences, vicarious experiences, verbal persuasion, and physiological and emotional states. Furthermore, two proposed omnibus outcomes of self-efficacy beliefs are an individual’s thought patterns (eg, goal setting, worry, and attributions) and behaviors (eg, challenges undertaken, effort expended on challenges undertaken, and persistence in the face of difficulties that arise during challenges undertaken). A necessary condition for valid testing of self-efficacy theory is concordance between the domain-specific self-efficacy beliefs and the proposed outcome of interest. There is a rich literature on the potential importance of targeting self-efficacy as a potentially modifiable mediating variable in physical activity–promoting interventions [19-21].

The self-efficacy theory posits that a self-efficacy–level construct may play a central role in the initiation of a behavior (eg, engaging in a recommended amount of weekly physical activity), whereas self-efficacy to regulate a behavior construct may play a central role in the maintenance of a behavior (eg, engaging in a recommended amount of weekly physical activity over time) [25]. A self-efficacy–level construct can be defined as an individual’s beliefs in his or her ability to accomplish levels of a task (eg, engage in at least 150 min of moderate-intensity physical activity in the next week). Self-efficacy to regulate a behavior construct can be defined as an individual’s beliefs to overcome possible barriers to accomplishing a task that he or she already knows how to do (eg, engage in at least 150 min of moderate-intensity physical activity in the next week if you are under personal stress). The importance of both a self-efficacy level construct and self-efficacy to regulate a behavior construct has been demonstrated in exercise contexts [26,27]. However, there still exists a pressing need to systematically test self-efficacy theory–based interventions to promote physical activity in real-world settings [5,9,21].

FFW is a self-efficacy theory–based, online (ie, Web-based and not an app) behavioral intervention developed to promote growth in well-being and physical activity by providing capability-enhancing opportunities to participants [28]. The full conceptual model for the FFW intervention is broader than this study and specifies that FFW exerts both a positive direct effect and a positive indirect effect through self-efficacy (ie, well-being self-efficacy, well-being action self-efficacy, physical activity self-efficacy, self-efficacy to regulate physical activity) on well-being (ie, subjective well-being, well-being actions, and physical activity). The narrower focus of this study was on the FFW conceptual model for the promotion of physical activity (see Figure 1). Consistent with the self-efficacy theory [24,25], the behaviors, emotions, thoughts, interactions, context, awareness, and next steps (BET I CAN) challenges provided in the FFW intervention (described in the next section) are specified as positive sources of self-efficacy information that exert a positive direct effect on self-efficacy beliefs, which are then specified to exert a positive direct effect on physical activity (ie, a behavior) [28]. Thus, self-efficacy is specified as a mediating variable in the FFW conceptual model for the promotion of physical activity.

In all, three construct-level a priori hypotheses were investigated in this study based on the conceptual model depicted in Figure 1. Hypothesis 1 was that the FFW intervention would exert a positive direct effect on self-efficacy. Hypothesis 2 was that self-efficacy would exert a positive direct effect on physical activity. Hypothesis 3 was that the FFW intervention would exert a positive direct effect on physical activity. An additional construct-level exploratory hypothesis (ie, hypothesis 4) was also investigated based on the conceptual model depicted in Figure 1: the FFW intervention would exert a positive indirect effect on physical activity through self-efficacy. Dimension-specific hypotheses were not made because of a lack of previous research on the effectiveness of the FFW intervention to promote physical activity.

The data described in this paper were collected within a more broadly focused trial, the Well-Being and Physical Activity Study (ClinicalTrials.gov, identifier: NCT03194854). Within this section, we provide an overview of the relevant methods used in the Well-Being and Physical Activity Study to provide a context for the specific focus of this paper [32]. The readers are referred to the relevant protocol paper [28] for a fuller description of the protocol for the Well-Being and Physical Activity Study. A populated Consolidated Standards of Reporting Trials-EHEALTH checklist is provided in Multimedia Appendix 1.

All procedures in this study involving human participants were in accordance with the ethical standards of the institutional and national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. The institutional review board (IRB) at the University of Miami provided necessary permission to conduct this study on July 11, 2017, IRB number 20170541. The University of Miami and Michigan State University (STUDY00000979) established an Institutional Authorization Agreement on June 26, 2018, that provided permission for the University of Miami to serve as the designated IRB for this study.

The study design was a large-scale, prospective, double-blind (ie, investigators and outcome assessor were masked), parallel-group randomized controlled trial. Recruiting, screening, random assignment, and collection of data were conducted online from August 2018 through November 2018. Data collection was Web based, fully automated, and occurred at three time points: baseline (T1), 30 days after baseline (T2), and 60 days after baseline (T3). The timeline for this study was similar to timelines used in other physical activity interventions in adults with obesity [12,13].

A sample size of approximately 900 participants was targeted for enrollment in the study. Participants were recruited through the general population panel of the SurveyHealth recruitment company. Partnering with a panel recruitment company is consistent with recruitment in preliminary research on FFW [33,34] and with a movement toward larger and smarter physical activity promotion interventions [18]. Eligibility criteria were (a) the ability to access the Web-based intervention, (b) living in the United States, (c) aged 18 to 64 years, (d) BMI of 25.00 kg/m² or more, and (e) absence of simultaneous enrollment in another intervention program promoting either well-being or physical activity. The BMI criterion included both the overweight (ie, 25.00-29.99 kg/m²) category and the obese category (ie, ≥30.00 kg/m²) consistent with many physical activity–promoting interventions for adults with obesity [12,35].

Informed consent was obtained from each participant included in the study. More specifically, immediately after being determined to be eligible for this study, each eligible individual was directed to a Web-based, IRB-approved informed consent form. Each individual who clicked Consent to Participate was enrolled as a participant in the study. Each individual who clicked Decline to Consent was denied access to the study.

Random assignment of each eligible participant occurred after a unique and secure login credential was created, informed consent was obtained, a medical disclaimer was agreed to, and the T1 survey battery was completed. Eligible participants were randomly assigned to the intervention (ie, FFW) or the usual care (ie, UC) group via software code that was written to accomplish equal allocations to the FFW and UC groups. Participants assigned to the FFW group were given immediate access to the intervention. Participants assigned to the UC group were put on a waitlist for access to the intervention. Both the FFW group and the UC group were provided with modest financial incentives to provide data consistent with a general approach taken in many theory-based physical activity–promoting interventions [9]. The authors of this study are unaware of any previous research that would support casting unique doubt on the results of this study (as compared with other theory-based physical activity–promoting interventions that used modest financial incentives in a study) attributable to the particular financial incentives approach taken in this study.

Instruments designed to measure demographic information, self-efficacy, and physical activity were included in the survey battery. Proposed demographic and biological correlates of physical activity were collected via self-reporting at T1 and included participant age, gender, race/ethnicity, highest level of education completed, marital status, employment status, and household annual income [19]. This set of demographic and biological variables is collectively referred to as the demographic covariates from this point forward.

Statistical models were fit in Mplus 8.3 with maximum-likelihood (ML) estimation with robust SEs [39]. Type I error rate was set equal to 0.05. Missing data were addressed with full information ML estimation using the observed information matrix under the assumption of missing at random [40]. Reliability was assessed using Cronbach alpha [41,42]. Indexes of effect size considered for direct effects were Cohen d [43] and percentage of variance explained. Commonly used heuristics were used to assist in the interpretation of an absolute value of Cohen d: 0.20 (small), 0.50 (medium), and 0.80 (large). For each indirect effect, a bias-corrected bootstrapped estimate of the 95% confidence interval was obtained with the number of draws set equal to 2000 [44]. An index of effect size was not considered for indirect effects because an effect size index for complex mediation models has not yet been firmly established [45].

A single saturated (degrees of freedom=0) path model was fit consistent with the conceptual model depicted in Figure 1 under an intention-to-treat approach [46]. Each of the five domains of self-efficacy at T2 were regressed on FFW (ie, 0=UC, 1=FFW), physical activity at T1, and the demographic covariates. Physical activity at T3 was regressed on FFW, the five domains of self-efficacy at T2, physical activity at T1, and the demographic covariates. The expression adjusted mean difference is used from this point forward to acknowledge the statistical adjustment made by including covariates in the model.

There were four sets of focal parameters in the path model. The first set of focal parameters was the direct effect of FFW on each of the five domains of self-efficacy at T2 (ie, beta₁). Each of these five parameters was interpreted as the adjusted mean difference on a particular domain of self-efficacy at T2 for the FFW group as compared with the UC group. The second set of focal parameters was the direct effect of the five domains of self-efficacy at T2 on physical activity at T3 (ie, beta₂). Each of these five parameters was interpreted as the path coefficient from a particular domain of self-efficacy at T2 to physical activity at T3. The third set of focal parameters was a single parameter: the direct effect of FFW on physical activity at T3 (ie, beta₃). This parameter was interpreted as the adjusted mean difference on physical activity at T3 for the FFW group as compared with the UC group. The fourth set of focal parameters was the indirect effect of FFW on physical activity at T3 through each of the five domains of self-efficacy at T2 (ie, beta₄, where beta₄=beta₁*beta₂'). Each of these five parameters was interpreted as the product of path coefficients from FFW to physical activity at T3 through a particular domain of self-efficacy at T2. Each set of focal parameters tested the numerically corresponding hypothesis (eg, beta₁ tested hypothesis 1).

Figure 2 depicts participant flow from eligibility screening to randomization to retention over the three measurement occasions. A total of 821 consenting participants were randomly assigned to FFW (n=410) or UC (n=411). Forensic analysis by a computer scientist performed before data analysis identified 154 cases as fraudulent, and these cases were excluded from analysis. The researchers initiated the forensic analysis after consulting with the designated IRB, legal counsel, and the office of research compliance and quality assurance about the computer scientist’s report of suspicious activity on the website (eg, participants logging in very close temporal proximity and sending identical emails to the computer scientist in broken English). The forensic analysis revealed that all of these 154 accounts were made by 1 user or group through 2 virtual private server (VPS) services. The analysis was reported as a reportable new information (RNI#00003760) incident to the designated IRB in December 2018. Unlike the 154 fraudulent cases (ie, 154/821, 18.8%), no groupings of the 667 nonfraudulent cases (667/821, 81.2%) appeared to have been made by 1 user or group through VPS services.

An additional 206 cases were outlying cases on the physical activity score and were excluded from the analysis, leaving 461 analyzed cases (ie, participants), FFW (n=219) and UC (n=242). A majority of the participants identified as female (302/461, 65.5%), white, non-Hispanic (342/461, 74.2%), having completed at least a 4-year college degree (307/461, 66.5%), married (314/461, 68.2%), being a full-time employee (309/461, 67.0%), being at least 40 years old (254/461, 55.1%), and as residing in a household with an annual income of at least US $75,000 (238/461, 51.6%). The difference in the proportion of missing data observed at T2 for the FFW group (ie, 0.12) as compared with the UC group (ie, 0.07) was not statistically significant (P=.08.)

Table 1 provides a comparison of demographic characteristics, BMI values, self-efficacy domain scores, and physical activity scores at T1 for participants by randomization group. There were no statistically significant differences in the proportions (for binary variables tested via logistic regression) or means (for continuous variables tested via linear regression) of demographic characteristics, the mean BMI value, the mean self-efficacy domain scores, or the mean physical activity scores at T1 by randomization group. The minimum value of BMI observed across the sample was 25.06 kg/m². The median values of physical activity in hours per week (ie, 10.61 and 9.18) were similar to IPAQ-based values in some other relevant research [48-50]. No important harms or unintended effects were observed in either group. Cronbach alpha ranged from .86 (physical activity) to .97 (work-related physical activity self-efficacy). A majority (201/219, 91.7%) of the participants who were assigned to the FFW group were engaged with the FFW intervention.

The percentage of variance accounted for ranged from 16.8% (work-related physical activity self-efficacy) to 25.3% (self-efficacy to regulate physical activity) across the five domains of self-efficacy at T2 and equaled 37.4% for physical activity at T3. The correlations among the residuals of the four self-efficacy–level constructs ranged from 0.74 (work-related physical activity self-efficacy with leisure-related physical activity self-efficacy) to 0.76 (transport-related physical activity self-efficacy with domestic-related physical activity self-efficacy). The correlations between the residuals of the four self-efficacy–level constructs with self-efficacy to regulate physical activity ranged from 0.04 (transport-related physical activity self-efficacy with self-efficacy to regulate physical activity) to 0.13 (work-related physical activity self-efficacy with self-efficacy to regulate physical activity). The unstandardized estimates of the covariates are available in Table 2, but these estimates are not discussed because of spatial limitations. Table 3 provides the unstandardized estimate of each focal parameter from the path model by hypothesis. Figure 3 provides key focal unstandardized parameter estimates for hypothesis 1 through hypothesis 3. Estimates for hypothesis 4 are not directly provided in Figure 3 because they are not parameter estimates per SE but rather a function of existing parameter estimates. However, they are listed at the bottom of Table 3. The paragraphs below briefly interpret these estimates with regard to the corresponding hypothesis tested.

The objective of this study was to evaluate the effectiveness of the FFW Web-based behavioral intervention to increase physical activity in adults with obesity in the United States in a relatively uncontrolled setting. In general, results from this study provide both some supportive and some unsupportive initial evidence with regard to the objective of this study. Specific findings, both supportive and unsupportive, will be discussed with respect to the four construct-level hypotheses tested within the FFW conceptual model for the promotion of physical activity (see Figure 1) and to the relevant results from the 2015 FFW efficacy trial.

Partial supportive evidence was observed in this study for three of the four hypotheses tested. Supportive evidence for hypothesis 1 includes positive direct effects from the FFW intervention to transport- and domestic-related physical activity self-efficacy and self-efficacy to regulate physical activity at T2. This set of findings provides some support for the conceptualization of the BET I CAN challenges as capability-enhancing opportunities and extends the literature on the ability of FFW to promote self-efficacy beliefs [3]—a potentially modifiable mediating variable targeted by the intervention. Supportive evidence for hypothesis 2 includes positive direct effects from both leisure-related physical activity self-efficacy and self-efficacy to regulate physical activity at T2 to physical activity at T3. This pair of findings provides some support for a central contention of the self-efficacy theory—behaviors are an omnibus outcome of self-efficacy beliefs [25]—and addresses a limitation of the 2015 FFW efficacy trial: not evaluating proposed relationships between self-efficacy and physical activity [2]. Supportive evidence for hypothesis 4 includes a positive indirect effect of the FFW intervention on physical activity at T3 through self-efficacy to regulate physical activity at T2. This finding addresses a limitation of the 2015 FFW efficacy trial: not evaluating the proposed positive indirect effect of the FFW intervention on physical activity through self-efficacy. Beyond the four hypotheses tested, this study has the potential to be important because it provides initial evidence for the effectiveness of the FFW intervention to increase physical activity (indirectly through self-efficacy to regulate physical activity) in an at-risk population [8,13]. Beyond the FFW intervention, findings from this study also contribute to a practical research need identified in the 2018 Physical Activity Guidelines Advisory Committee Scientific Report: to systematically test theory-based interventions in real-world settings [9].

At least partial unsupportive evidence was observed in this study for each of the four hypotheses tested. Unsupportive evidence for hypothesis 1 includes null direct effects from the FFW intervention to both work- and leisure-related physical activity self-efficacy at T2. Thus, it may be that the BET I CAN challenges in the FFW intervention would benefit from being further optimized for providing more meaningful exposure to relevant sources of efficacy-enhancing information with regard to these 2 domains of self-efficacy beliefs [51]. More specifically, future studies that estimate the individual effect of each BET I CAN component, and how BET I CAN components may operate synergistically with each other, may help identify active and inactive intervention components within FFW with regard to promoting self-efficacy and physical activity in adults with obesity. Unsupportive evidence for hypothesis 2 includes null direct effects from work-, transport-, and domestic-related physical activity self-efficacy at T2 to physical activity at T3. This set of null findings may be because of the relatively strong correlations among the four self-efficacy–level constructs (ie, difficult to identify unique relationships with physical activity). Unsupportive evidence for hypothesis 3 includes a null direct effect from the FFW intervention to physical activity at T3. This null finding is inconsistent with relevant results from the 2015 FFW efficacy trial [2] and may be because of differences in either model specification (ie, evaluating the direct effect of FFW on physical activity while controlling for self-efficacy beliefs in this study) or measurement of physical activity (ie, more thoroughly measuring physical activity in this study). Unsupportive evidence for hypothesis 4 includes null indirect effects from the FFW intervention to physical activity at T3 through each of the four self-efficacy–level constructs: work-, transport-, leisure-, and domestic-related physical activity self-efficacy at T2. This set of null findings may be attributable to the idea that, on average, an individual’s self-efficacy beliefs regarding their capability to engage in a recommended amount of physical activity for health may be less important than an individual’s self-efficacy beliefs in their capability to overcome possible barriers to their engagement in a recommended amount of weekly physical activity for health with regard to the promotion of an individual’s physical activity behavior [26,27].

Results from this study provide some initial evidence for both the effectiveness and the ineffectiveness of the FFW Web-based behavioral intervention to increase physical activity in adults with obesity in the United States. Specifically, there is evidence that FFW may be ineffective in directly promoting physical activity in adults with obesity. Similarly, there is evidence that FFW may be ineffective in indirectly promoting physical activity through the four (ie, work-, transport-, domestic-, and leisure time–related) self-efficacy–level constructs (ie, the degree to which an individual perceives that they have the capability to engage in a recommended amount of weekly physical activity for health). However, there is evidence that FFW may be effective in indirectly promoting physical activity in adults with obesity by increasing an individual’s self-efficacy to regulate their physical activity (ie, the degree to which an individual perceives that they have the capability to overcome possible barriers to engagement in a recommended amount of weekly physical activity for health). For this reason, we believe that the FFW Web-based behavioral intervention may have the potential to eventually become useful, in some small but important way, given the magnitude of the problem, in responding to the global pandemic of insufficient physical activity in adults with obesity by increasing an individual’s self-efficacy to regulate their physical activity.

Realizing the potential for the FFW intervention to have practical implications at a local level will require future community-based studies that align with recent recommendations put forth by the Community Preventive Services Task Force [52]. More specifically, the Community Preventive Services Task Force suggests that physical activity interventions for adults with obesity should include activity monitors and promote physical activity within a more broadly focused weight management program where there is access to a health care provider. An implication from the results of this study is that a feasibility study is now underway to implement accelerometer-based assessment of physical activity within the FFW intervention in partnership with a local bariatric service center within a major health care organization in the Midwest of the United States [53]. Gaining necessary approvals for accessing medical records from participants in this ongoing feasibility study may provide important information on certain patient characteristics (eg, comorbidities) that may influence the effectiveness of the FFW intervention.

We are aware of at least four noteworthy limitations for this study that temper the relevant conclusions that can be made. First, we recognize that our hypotheses assume additivity of FFW effects for all covariates (ie, no a priori moderators for the proposed effects of FFW). We encourage future secondary analyses that explore the prospect of heterogeneous FFW effects for subgroups of individuals (eg, comorbidities) on physical activity. Second, we note that another limitation is that all the data collected, except for engagement with the FFW intervention, were collected via self-reporting. Field-based studies that collect physical activity data from objective instrumentation [54-58] in adults with obesity are encouraged [35,52] and are underway in the FFW context [53]. This underway study is employing both self-reported and accelerometer-measured physical activity in adults with obesity, which is consistent with recommendations in previous research [59] that found the physical activity of overweight or obese individuals to be ranked higher by self-reporting than by accelerometer as compared with normal-weight individuals. That said, it is important to note that the aforementioned published study did not provide evidence for randomized group assignment (eg, control vs experimental) as a moderator for the observed mismatch between self-reported and accelerometer-measured physical activity in overweight or obese individuals engaged in physical activity–promoting interventions. Thus, although the aforementioned study provides support for suspecting that a mismatch between self-reported and accelerometer-measured physical activity may have been observed in this study (if accelerometer-measured physical activity had been collected), it does not provide direct support for suspecting that the magnitude of the suspected mismatch may have varied as a function of randomized group assignment in this study (ie, UC group vs FFW group). The third limitation is that 360 of 820 cases (eg, 43.9%) needed to be excluded from the analyses because of either fraud (n=154) or outlying physical activity scores (n=206). Future efforts to better guard against fraud (eg, working more closely with the panel recruitment company) and possible overreporting of physical activity (eg, objective assessment of physical activity) is encouraged and may increase confidence in subsequent findings (eg, in reference to physical activity guidelines). A final limitation is that engagement data were not collected from UC participants who were given 1 month of 24-hour access to the FFW intervention (but were not provided with financial incentives to complete BET I CAN challenges) after data collection for this study was closed. Collecting these data would have provided some insight into the degree to which the very high level of engagement observed in the FFW group (ie, 201/219, 91.7%) may have been because of the inclusion of financial incentives to complete BET I CAN challenges.