Full details of the trial protocol, study population, Balance intervention, and recruitment processes have previously been published [14] and are summarized here. Balance participants were adults with overweight or obesity who received care at participating Piedmont CHCs. Participants were randomized to receive either (1) the 12-month Balance weight gain prevention intervention or (2) a healthy living usual care arm. There were no trial-specific follow-up visits conducted following randomization. Participant outcomes were collected from the Piedmont electronic health record (EHR) at 6, 12, 18, and 24 months after randomization.

Approvals for the trial were obtained by the Duke University institutional review board (2017-0738/D0479) and the Piedmont board in 2016. Verbal informed consent was provided from all participants prior to enrolling in the trial. The trial was registered on ClinicalTrials.gov (NCT03003403) on December 16, 2016. All data presented in this manuscript are deidentified. As Balance was a pragmatic trial with outcomes extracted from the Piedmont EHR and thus did not require participant study visits or follow-up surveys for evaluation, monetary compensation was not provided.

Congruent with the trial’s pragmatic design, we sought to limit eligibility criteria to those that were fundamental to intervention implementation or evaluation or represented a significant threat to patient safety. To be included, participants needed to be fluent in English or Spanish and have a BMI between 25 and 40 kg/m² (inclusive) and a weight of 380 pounds or less, as recorded by a provider in the Piedmont EHR within the previous 14 days. The weight eligibility criterion was based on the weight capacity of the connected scales used for intervention delivery. Participants also needed a mobile phone and service plan that could receive weekly trial-related text messages. Exclusion criteria included pregnancy within 12 months, giving birth within 6 months, or breastfeeding within 2 months; having prior weight loss or bariatric surgery; or planning to relocate outside the Piedmont service area during the trial period.

Trial recruitment launched in February 2017. In-person recruitment was conducted through a joint effort by Piedmont staff and a Balance research staff member, with Piedmont staff identifying possibly eligible participants and referring them to research staff for questions and additional information. Due to slower-than-expected study recruitment, 4 months into the trial, a new recruitment method was added: an EHR query was used to identify patients who met basic criteria (eg, age and BMI) for Balance and had an upcoming appointment at Piedmont. These patients identified via the EHR query were mailed materials ahead of their appointments and presented trial information during visits to the CHC, if a research team member was present at their appointment time. Interested patients from either recruitment method were prompted to sign an authorization form to allow research staff to assess their medical record for eligibility—and later for evaluation purposes, if enrolled. Following completion of the authorization form, patients underwent additional phone screening to assess criteria not available in their medical records, such as willingness to receive text messages. After eligibility was established, research staff continued to conduct verbal informed consent, enrollment, randomization, and program orientation procedures. Additional details regarding recruitment and screening procedures were previously described [14,15]. Randomization took place concurrently with enrollment between February 2017 and December 2018 using block randomization with stratification by the patient’s CHC within Piedmont Health, with equal allocation to each treatment group.

Participants randomly assigned to the healthy living usual care arm received enhanced usual care. They received their standard primary care at Piedmont as well as 6 months of automated weekly text messages with healthy living information and printed materials adapted from the National Heart, Lung, and Blood Institute’s Aim for a Healthy Weight [16].

The goal of the Balance intervention was to prevent weight gain by encouraging behavioral changes that result in a slight energy deficit. Intervention participants received (1) tailored behavior change goals, (2) skills training materials and videos, (3) weekly self-monitoring prompts for behavioral goal tracking, (4) connected scales for weight self-monitoring, and (5) responsive health and weight coaching from trained registered dietitians.

Behavioral goals were individually assigned and tailored to participants using the interactive obesity treatment approach. The processes for interactive obesity treatment approach goal assignment for Balance have been described elsewhere [14] and have been used in previous primary care obesity treatment trials [13,17,18]. Briefly, participants were administered a simple survey at the start of the trial with a built-in algorithm that creates a net caloric deficit on the backend. For example, participants were asked to rate the frequency of consuming sugary drinks or walking 10,000 steps per day and their perceived confidence to change this behavior. Based on the results of the frequency of each health behavior and their self-efficacy to change it, participants received a set of 3 goals which changed every 8 weeks. They monitored their goal progress by responding to weekly interactive voice response calls and text messages. Participants then received automated, tailored feedback that described their progress, reinforced successes, and offered motivational strategies or short skills training tips.

Throughout the intervention, participants were also asked to weigh themselves daily on a connected scale, as previously described [14,19,20]. Data were transmitted through cellular networks to a database accessible to the research team. A rolling 7-day average of these weights was used to activate the responsive coaching from a Balance dietitian when a participant reached a designated threshold, or zone, of weight gain. If no weights were received across 1 intervention week, study staff would attempt to contact the participant via text or phone to re-engage or troubleshoot.

Data collection procedures were designed to maximize the use of data collected during routine primary care [21]. Participant sociodemographic characteristics were collected by a brief survey administered as part of the enrollment phone call. All other participant baseline and follow-up data were pulled directly from the EHR (GE Centricity CPS, version 12). Weight measurements in the EHR were recorded in pounds to the nearest 0.1 pound and converted to kilograms for analysis. Height was recorded in inches to the nearest 0.01 inch and converted to centimeters for analysis. BMI was recorded as weight/height (kg/m²). EHR data were obtained on all participants between August 6, 2015, and April 30, 2021, including up to 18 months before enrollment through 30.5 months after enrollment.

For participants who became pregnant during the study, all clinical values were censored as of the beginning of their pregnancy. For pregnancies that occurred before enrollment, we censored clinical values from the beginning of pregnancy to 6 months after the end of pregnancy.

Baseline variables were summarized by the intervention arm as means and SDs for continuous variables and as counts and percentages for categorical variables. We used a 2-stage modeling strategy to answer our primary hypothesis. All analyses were intent-to-treat. For the first stage, we estimated the intervention effect on weight using a constrained longitudinal linear mixed-effects model [22], using all available data from between 18 months before enrollment through 30.5 months after enrollment. We explored nonlinearities for the fixed effects of time and random effects of time in the models, guided by the Bayesian information criterion and likelihood ratio tests. The model included fixed effects for time and interactions between time segments and intervention, with a random intercept and random slopes for time. The covariance between random effects was modeled using an unstructured covariance matrix. Randomization was stratified by CHC; thus, we adjusted for CHC in the analyses but did not adjust for other variables. Using this linear mixed-effects model, we estimated mean weight change by intervention arm, and the difference between intervention arms in mean weight change, at 6, 12, 18, and 24 months.

Second, we compared the percentage of ≤3% weight gain—our primary outcome—in each arm at 24 months after randomization using individual empirical best linear unbiased predictors (EBLUPs) from the mixed model [23]. To ensure good predictions, we only used individual EBLUPs from participants with at least 1 EHR weight documented within a 6-month window centered on the 24-month time point (ie, between 21 and 27 months after enrollment). We compared the percentage of ≤3% weight gain in intervention and usual care arms using both a log-binomial model and linear risk model on the EBLUP output, in order to obtain both risk ratios and risk differences, which provide estimates of relative and absolute efficacy, respectively [24]. The extra variability induced by the EBLUPs being predicted from the model was taken into account using a resampling procedure, explained further in the Multimedia Appendices 1 and 2. Sensitivity to the 21- to 27-month window was assessed by obtaining predictions after expanding the window. We additionally used EBLUPs to compare the percentage of ≤3% weight gain in each arm at 6, 12, and 18 months after randomization.

Additional sensitivity analyses included adjusting the model for baseline variables imbalanced across arms, by weight measured in the primary time window (P<.10) and removing telehealth visits from the analysis. We also explored effect modification by adding interactions to the primary model with age tertiles; gender; BMI class; race and ethnicity; and preferred language (English or Spanish), each in a separate model. Additionally, in post hoc analyses we added the completion date of the study (before or after COVID-19) as an effect modifier to examine the potential impact of COVID-19 on the intervention effect.

We and others have demonstrated the efficacy of weight gain prevention interventions in adults affected by obesity [13,25]. However, in Balance, we did not find long-term evidence to support the dissemination of a digital weight gain prevention intervention to patients receiving primary care at CHCs. Compared with the usual care arm, the intervention arm experienced greater weight loss at 6 months, yet this difference was not clinically meaningful and had attenuated by 24 months. Further, we did not observe a difference between the 2 arms in the proportion of individuals who stayed within 3% of their baseline weight.

Unlike our previous trials using similar approaches resulting in significant weight gain prevention or weight loss [13,18], Balance was designed to use a high level of pragmatism to deliver and evaluate real-world intervention effectiveness. This aligned with the key domains of the Pragmatic-Explanatory Continuum Indicator Summary index, including limited exclusion criteria, nonrestrictive control group, no formal follow-up data collection, and extraction of outcome data directly from the EHR [21]. Pragmatic trial designs vary considerably in the aforementioned domains and these characteristics can assist with contextualizing study findings. For instance, we did not restrict the usual care arm in this trial. As a result, some of these patients may have received medical nutrition therapy or other weight management services from Piedmont registered dietitians or community partners. These efforts may have resulted in the usual care arm experiencing better weight outcomes than expected among a restricted control group, ultimately diminishing trial outcomes.

As described in the methods section and more thoroughly in our protocol paper, the Balance study had a responsive weight and goal coaching protocol. There were 3 zones a participant would be categorized in, depending on their weight change, in which they may have received minimal to no coaching. This differs from our previous weight gain prevention trial that successfully prevented weight gain, in which participants received a standardized coaching protocol that included 20-minute monthly phone calls for 12 months, regardless of weight status [12,13]. It is possible that participants of the Balance intervention would have benefited from this standardized coaching protocol or a higher-intensity intervention.

To our knowledge, no examples exist of other pragmatic weight gain prevention trials in primary care settings among medically vulnerable populations with which to compare our findings. However, there have been several pragmatic weight loss trials implemented in the primary care setting. For example, the PROPEL and REPOWER trials were large, pragmatic randomized weight loss trials implemented in primary care clinics working with medically vulnerable populations and tested a total of 5 weight loss interventions with varying levels of pragmatism [26]. Both trials’ intervention arms observed significant weight loss at 6-months. Yet, not all intervention arms were as successful at producing weight loss, and notably, the arm designed with the highest level of pragmatism observed the least sustained weight loss at 24 months. Along with our study, these findings highlight the challenges of long-term weight management, particularly in pragmatic trials. Future studies might benefit from including additional support for changing motivation, group support, and targeted problem-solving strategies [27,28] to overcome external challenges.

This study had limitations that should be noted. A potential challenge facing pragmatic weight management trials is the quality of clinic-measured weights. We previously assessed concordance between EHR and research-collected weight data in our earlier weight management interventions and did not find differences in baseline weight, changes in weight from baseline, or differences between intervention and controls in weight change [29]. However, estimates based on EHR weights were generally more variable than weights directly collected by the study team. Another more recent trial demonstrated that weight loss differences observed at 24 months were smaller when using EHR data [30]. The authors offered several explanations for this finding, including the larger variability in follow-up time and the less restrictive procedures for in-clinic weight measurement, such as those observed in Balance. Because we did not collect a secondary measure of weight, we are unable to assess if more controlled weight measurement procedures would have improved our ability to detect differences in weight gain between groups.

We expect that the timeframe of implementation posed challenges to identifying the effects of the Balance intervention on weight outcomes. Namely, the COVID-19 pandemic began during the implementation of the Balance intervention and our 24-month follow-up period. This decreased the number of in-person visits being conducted at the CHCs and, ultimately, impacted follow-up data collection on nearly half of our participants. COVID-19 could have also impacted other weight-related behaviors and risk factors among participants, including physical activity [31,32], food access and dietary quality [33-35], and mental health stressors [36,37]. In addition to COVID-19, there were several other events during the intervention and follow-up timeframe that could have impacted data collection, including 2 major North Carolina hurricanes and immigration reform resulting in increased deportation of undocumented individuals. These factors likely impacted patient appointment attendance and thus the availability of baseline and follow-up EHR data. As such, these events may have affected the precision of the intervention effect estimates for outcomes at 24 months. However, it is unlikely the intervention effect estimates themselves were biased by these events, as there was no difference between the number of visits with weights measured by treatment arms.

Additionally, some components of trial implementation did not align with a pragmatic approach, such as using research staff to conduct recruitment at the health centers. There were also institutional requirements for research compliance that would not be necessary if implemented within a real-world setting. For instance, a Health Insurance Portability and Accountability Act (HIPAA) authorization form was required from each participant before screening could occur, which posed challenges to recruitment ease and timing, as highlighted in our previous article [15]. However, if Balance were disseminated further within Piedmont or other similar CHCs without the inclusion of external research partners, these specific requirements would likely have been lessened or eliminated, thereby decreasing the amount of time required to enroll in the program. Finally, as discussed above, several events, such as COVID-19, occurred during study implementation that impacted data collection. However, as described above, we expect this did not bias effect estimates.

Despite these limitations, there are many notable strengths. The Balance trial was designed with a high level of pragmatism, with limited criteria for eligibility and no study-specific data collection or follow-up procedures, and enrolled a diverse sample of participants. For these reasons, the Balance trial also provides results that can inform decisions in real-world care settings for the management of obesity. Although the intervention did not have a significant impact on weight gain prevention, it is possible that participants adopted healthy dietary and physical activity behaviors that positively impact health and chronic disease management [38]; however, these outcomes were not assessed in this study. Future assessment is warranted given long-term healthy lifestyle changes are possible independent of weight change after participation in a weight gain prevention program [39].

To the best of our knowledge, Balance was also the first weight gain prevention trial to be implemented with a pragmatic design in a primary care setting serving a medically vulnerable population, serving as a strong example of successful partnerships between research teams and CHCs. This sentiment was echoed by Piedmont providers and staff in a qualitative evaluation of Balance [40], in which they stated that early buy-in from providers and staff, respect for the patients and health care setting by the research partnership, and filling a treatment gap for patients all contributed to successful trial implementation. Combined with the trial outcomes, these findings can be used to inform future programs and interventions to be tested and implemented in the CHC setting.

Balance did not exhibit a positive impact on long-term weight maintenance, yet it provides important information for future pragmatic trials in the primary care setting. Future trials might benefit from including an assessment of the comparator group’s engagement in weight management behaviors and collecting a secondary weight measurement. Researchers may also consider measuring other outcomes associated with chronic disease prevention that are independent of weight, such as changes in diet quality, physical activity, and psychological health.