This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Research Protocols, is properly cited. The complete bibliographic information, a link to the original publication on http://www.researchprotocols.org, as well as this copyright and license information must be included.
Online social networks may be a promising modality to deliver lifestyle interventions by reducing cost and burden. Although online social networks have been integrated as one component of multimodality lifestyle interventions, no randomized trials to date have compared a lifestyle intervention delivered entirely via online social network with a traditional clinic-delivered intervention.
This paper describes the design and methods of a noninferiority randomized controlled trial, testing (1) whether a lifestyle intervention delivered entirely through an online social network would produce weight loss that would not be appreciably worse than that induced by a traditional clinic-based lifestyle intervention among overweight and obese adults and (2) whether the former would do so at a lower cost.
Adults with body mass index (BMI) between 27 and 45 kg/m2 (N=328) will be recruited from the communities in central Massachusetts. These overweight or obese adults will be randomized to two conditions: a lifestyle intervention delivered entirely via the online social network Twitter (Get Social condition) and an in-person group-based lifestyle intervention (Traditional condition) among overweight and obese adults. Measures will be obtained at baseline, 6 months, and 12 months after randomization. The primary noninferiority outcome is percentage weight loss at 12 months. Secondary noninferiority outcomes include dietary intake and moderate intensity physical activity at 12 months. Our secondary aim is to compare the conditions on cost. Exploratory outcomes include treatment retention, acceptability, and burden. Finally, we will explore predictors of weight loss in the online social network condition.
The final wave of data collection is expected to conclude in June 2019. Data analysis will take place in the months following and is expected to be complete in September 2019.
Findings will extend the literature by revealing whether delivering a lifestyle intervention via an online social network is an effective alternative to the traditional modality of clinic visits, given the former might be more scalable and feasible to implement in settings that cannot support clinic-based models.
ClinicalTrials.gov NCT02646618; https://clinicaltrials.gov/ct2/show/NCT02646618 (Archived by WebCite at http://www.webcitation.org/6v20waTFW)
Lifestyle interventions have had established efficacy for over a decade but are not widely disseminated largely due to high cost and burden to patients and providers [
A growing number of studies have explored using online social networks, such as Facebook [
The purpose of this paper is to describe the design and methods of a noninferiority randomized controlled trial (RCT) comparing a lifestyle intervention delivered entirely via an online social network with a traditional in-person clinic-based lifestyle intervention among overweight and obese adults. Intervention strategies in both study conditions are based on the Diabetes Prevention Program (DPP), which is considered the gold standard in evidence-based behavioral weight loss programs [
This study is a noninferiority RCT comparing a lifestyle intervention delivered entirely via the online social network Twitter (Get Social condition) with an in-person group-based lifestyle intervention (Traditional condition) among overweight and obese adults. The use of the noninferiority design provides enough statistical power to establish that one modality is not appreciably worse than another [
A total of 328 overweight or obese adults will be randomized to the two conditions. To be eligible for participation, participants must meet the following inclusion criteria: (1) a body mass index (BMI) between 27.0 and 45.0 kg/m2, (2) aged between 18 and 65 years, (3) written clearance from their primary care providers, (4) possession of a smartphone, (5) an active social media user (ie, currently have an online social network account and log in at least 4 days/week), and (6) an interest in losing weight. Participants with a personal Twitter account will be asked to set up a separate Twitter account to use for participating in the study. Twitter allows users to have multiple accounts, and linked accounts can be toggled back and forth with a click on the Twitter app.
Individuals who meet any of the following criteria will be excluded from study participation: (1) plans to move during study, (2) pregnant, lactating, or plans to become pregnant during the study, (3) taking medication that influences weight, (4) participating in a formal weight loss program (eg, Weight Watchers), (5) lost 5% or more weight in the past 3 months, (6) bariatric surgery or plans to undergo surgery during the study period, (7) a medical condition that precludes dietary or physical activity changes, (8) type 1 diabetes reported by the participant or their primary care provider or uncontrolled type 2 diabetes determined by the participant’s primary care provider in a medical clearance letter, (9) currently smoking more than 3 cigarettes per day, and (10) preference for one condition over the other. Participants who do not complete baseline measures, do not agree to be randomized, and do not participate in a study orientation webinar will also be excluded.
Study participants will be recruited from the local Central Massachusetts area in 9 waves of 36 to 37 participants. Participants will be recruited from the community, including online recruitment sources such as Craigslist, Twitter, postings in local Facebook groups (eg, local parent groups), the University of Massachusetts Medical School and UMass Memorial Health Care employee and student intranet, online newsletters, print newspapers, mass emails, and flyers in local community locations. We will also recruit participants through paid ads on Google, Facebook, and Twitter. Some recruitment efforts will specifically target males and racial/ethnic minorities, given that previous weight loss intervention trials underrepresent these groups [
Before randomization and after completing the baseline assessment, participants will attend a 1-hour orientation webinar, adapted from the model used successfully in two other previous completed weight loss trials [
Each wave of participants will be randomized 1:1 to the two study conditions in randomly permuted blocks of sizes 4 and 6 using the ralloc program in Stata (Stata Corp) [
Twitter was selected as the online social media platform for intervention delivery for several reasons. First, using a commercial platform saves the cost of developing a separate platform. Second, only 24% of the population uses Twitter, so most participants will not be users, unlike Facebook on which 79% of the population has a Facebook account [
Participants randomized to the Get Social condition will attend individual and group orientation visits before the intervention. Given that all participants are locally recruited and could be randomized to either the Get Social or Traditional condition, we opted for in-person orientation visits to ensure that participants understand how to use the technology, though conducting the orientation visit via webinar may be a preferred method for future effectiveness trials.
At the individual orientation visit, participants will learn how to set up and use a private Twitter account. Twitter’s privacy setting protects one’s tweets from being viewed by anyone except those approved by the user. All participants and the dietitian use the “protected” privacy setting so that all interactions are only viewable to each other, thus creating a private group on Twitter. Participants are informed that their followers can see their username, Twitter handle, and profile picture. They will be urged to select aliases and avatars to represent themselves in the biographical section of their account to protect their anonymity. At start-up, participants will be asked to hold off on following nonstudy accounts so that their newsfeed is exclusively for the group. However, as the program posts become less frequent in the later phases of the intervention, they will be given suggestions for evidence-based healthy lifestyle feeds to follow to grow their social network. Participants will be advised not to post anything they feel uncomfortable posting and to remain on topic. Participants who wish to have a private exchange with a dietitian or a fellow participant will be instructed to use the Twitter direct message feature, which allows private one-on-one communication between users. Participants will also be oriented to the study blog and instructed to log in with all devices they use regularly so that their time spent on the blog can be tracked for the cost analyses. The study blog will contain papers that dietitian posts occasionally, with a link to provide more information on a topic.
At the group orientation, participants will follow the dietitian and all other participants in their group on Twitter will receive instruction on ways to participate and practice tweeting. They will receive instruction on how to get the most out of the intervention, including guidelines to read the daily posts by the dietitian, to read and reply to each other’s posts, to post about their progress and challenges, and to ask the dietitian or group any questions that come to mind. They will also learn about the DPP lifestyle intervention, program goals for weight loss and physical activity, and how to use MyFitnessPal for dietary and exercise self-monitoring. A mobile app for tracking is necessary in the Get Social condition because dietitians will not have physical contact with participants to exchange paper records. Instead, they can access participants’ data by logging into their account. Protocols for the individual and orientation visits were developed based on a series of pilot studies [
Participants in the Traditional condition will attend a group orientation visit. At this visit, participants will learn about the DPP lifestyle intervention [
Participants in both study conditions will receive a 12-month lifestyle intervention based on the DPP. The DPP includes instruction in self-monitoring of food intake, nutrition, exercise, and behavioral modification [
Participants randomized to the Get Social condition will receive 12 months of lifestyle counseling via a private Twitter group. For months 1 to 2 of the intervention, the dietitian account will post twice daily, for months 3 to 6 the dietitian account will post once daily, and for months 7 to 12, the dietitian account will post 4 times weekly. The frequency and spacing of these interactions correspond to the frequency and spacing of in-person visits that occur in the DPP.
Each DPP session was distilled into a collection of 14 tweets that met all visit objectives [
Every Friday morning post will ask participants to reply to report their weight change from the previous week. This approximates the “weigh-ins” that occur in the Traditional condition, while protecting their privacy by focusing on change in weight from the previous week (eg, +1 lb and −1 lb) rather than absolute weight. To encourage participants to check their Twitter feed, participants will be informed of the various ways they can receive notifications (eg, emails and pop-ups) and will be encouraged to set notifications of their preference. They will be advised to log in daily to read the counselor’s posts and to engage with the group. Each week, a newsletter will be emailed to participants to encourage participants to engage with the group and to highlight some of the newsfeed from that week in case any participants miss the content when it is posted.
Participants randomized to the Traditional condition will receive 12 months of lifestyle counseling via clinic-based group meetings lasting 90 min per session. Participants will receive the Core of the DPP Lifestyle Intervention Core intervention for 6 months, followed by monthly group meetings for another 6 months (total of 22 sessions) [
At 6 and 12 months, participants will complete an in-person study visit. Follow-up assessments include physical measurements, a follow-up survey, and the first of 3 computer-based 24-hour dietary recalls. Two additional computer-based dietary recalls will be completed randomly at home over the following week. Participants will receive US $40 after completing the 6-month follow-up assessment and US $60 after completing the 12-month assessment.
Trained personnel will measure participants’ height and weight using a digital scale and stadiometer with the participant wearing light clothing and no shoes; measurements will be taken to 2/10th of the nearest inch or pound. Percentage weight loss at 6 and 12 months will be calculated by subtracting follow-up weight from baseline weight divided by baseline weight.
To assess energy intake, participants will complete three 24-hour diet recall interviews over 2 weeks surrounding baseline, 6-, and 12-month in-person study visit using the National Cancer Institute’s (NCI) automated self-administered 24-hour dietary recall (ASA24) [
To assess physical activity, participants will complete the 74-item Arizona Activity Frequency Questionnaire [
We will systematically track costs associated with delivery of both intervention conditions, capturing information on the costs that would be required to implement each intervention in practice (ie, outside the research context), including administrative, interventionists, and participant costs [
We have created an online survey system that will be used to evaluate staff time and money spent on these activities. Staff will record how much time they spent on each task weekly and any monetary costs (eg, purchase of supplies and copying expenses). In the Get Social condition, administrative tasks include scheduling posts, software (Photoshop and Buffer) for modifying and posting tweets, orientation materials, and conducting the orientation visits. In the Traditional condition, administrative tasks include the purchase or printing of participant materials, sending materials to members who missed the group, orientation materials, and conducting the orientation visits. For staff and interventionist time, we will calculate costs based on actual staff salaries [
Interventionists will report time weekly on a spreadsheet and document each task completed and the time taken to perform each task. Get Social condition interventionist tasks include time spent on the following: reading and responding to participant posts, fielding direct messages from participants, answering questions/concerns or researching information to answer questions, emailing participants who do not post for 2 weeks, and reviewing participant diet diaries. In the Traditional condition, interventionist tasks include time spent on the following: travel for group meetings, leading the group meetings, emailing participants who do not attend the group meetings, answering questions/concerns or researching information to answer questions, and reviewing participant diet diaries. Interventionists will record time spent on each task daily. Estimates for the cost of interventionist time spent traveling round trip to group meetings will be calculated as 5 min/mile for a 5-mile radius from the research center and then 2 min/mile beyond 5 miles based on local traffic patterns around campus.
Get Social condition participant tasks include the time spent on Twitter to participate in the intervention and time spent reading the study blog. On the weeks the Traditional condition has group meetings, we will contact participants in the Get Social condition via email requesting them to complete a self-report online survey. In these surveys, we will ask participants to report the time they spent on Twitter to participate in the intervention that week and then use these data to estimate total time across all weeks. Participants with iPhones will be asked to report the time spent on the Twitter app under battery usage settings. To assess the time a participant spent on the Twitter app specifically for intervention participation (rather than using the Twitter app for other reasons), we will ask participants whether they access the intervention feed using another device (such as a desktop or a laptop computer or a tablet), and what percentage of the time they spent using the Twitter app was to participate in the intervention, and adjust time estimates from the iPhone battery statistics on their responses. Participants with Android or Windows smartphones will be asked to self-report their time spent on Twitter. Time spent on the study blog will be tracked through Pardot, a company that provides analytics for customers’ online marketing campaigns. We will download these data from Pardot every 6 months.
Although participants will register their devices on the blog at baseline, it is possible that participants will access the blog from nonregistered devices. Therefore, the weekly surveys will also ask participants to report how many of the articles that the coach tweeted in the past 7 days they read, and we will compare these self-reported data with data downloaded from Pardot. In the Traditional condition, participant tasks include travel to and from the group meetings from their home address (using the same approach as for interventionist travel costs) and time spent attending group meetings. Time at group meetings for each participant will be based on the duration of each meeting recorded by the interventionist and attendance records. Participant time will be converted to costs using information about participant income [
We will track which participants drop out of treatment. We have a protocol for reengaging participants who are not actively participating in treatment. In the Get Social condition, staff tweet participants who have not engaged that week and ask them how they are doing. Following the second consecutive week with no visible online engagement, the coach will email the participant. If another week passes with no engagement on Twitter, the coach will call the participant. In the Traditional condition, staff will email the missed intervention materials to participants who miss the group session. Following the second consecutive missed group session, the coach will email the participant. If another group session is missed, the coach will call the participant. We will consider participants who have not engaged in treatment in 4 consecutive weeks and have not responded to our attempts to reengage, as well as participants who express wanting to withdraw from the intervention to have dropped out of treatment.
At 6 and 12 months, participants in both study conditions will rate the
We will explore predictors of weight loss in the online social network condition, including engagement, age, sociability, neuroticism, openness, and social network use.
Engagement data on tweets by interventionists and participants will be downloaded weekly from Twitonomy, a Twitter analytics and monitoring tool. Information on “likes” for each tweet will be downloaded weekly via a script that captures the data from Twitter. We will calculate the number of original tweets, replies, and likes per participant. We will also sum these metrics for a total summary measure of engagement. At 6 and 12 months, we will also ask participants survey questions about lurking (reading without visibly interacting) [
Participants will report their age at eligibility screening.
Participants will complete the Ten-Item Personality Inventory (TIPI) [
At baseline, participants report which online social networks they have accounts on and frequency of online social network use in the past 4 weeks (study survey). Online social network use is expected to be a predictor of weight loss in the Get Social condition because individuals who use online social networks more regularly may be more likely to lose weight than nonregular users as they are more accustomed to using these platforms and engaging online.
Contamination is defined as the use of other forms of online or in-person weight loss support in either condition during the study. Using the Pew Internet & American Life Project Poll [
We randomly selected one topic within each phase of the study (weeks 1-8, weeks 9-24, and weeks 25-52) for a total of 3 topics (10% of 22 topics). Each topic has the same objectives to meet in both conditions. In the Traditional condition, we have an independent reviewer listen to the recording of the selected group to ensure that each objective was met. In the Get Social condition, the topics span from 1 week to 4 weeks depending on the phase of the study (1 week in phase 1, 2 weeks in phase 2, and 4 weeks in phase 3). The respective weeks of the topic selected will be extracted from Twitter and reviewed by an independent reviewer to ensure each objective was met. Findings will be reported.
Possible risks during the intervention include injury during exercise or breach of confidentiality. Participants who report conditions at baseline that could create a safety concern while receiving the intervention are excluded. To avoid possible risks, participants are instructed to avoid over exercising at intense levels that could lead to discomfort, pain, or injury. Participants reporting discomfort will be referred to their primary care providers. Participant data are stored in network secure data entry programs, and any data on paper are stored in a locked file cabinet. Adverse events that occur during the intervention are assessed, recorded, and followed up until resolved. Serious adverse events are communicated immediately to the data safety monitoring board and the institutional review board (IRB).
We powered the study to be able to detect noninferiority [
For secondary noninferiority outcomes, we used 90% power to calculate noninferiority margins given N=131 per arm, setting alpha=.05 and using observed SDs from the literature. For change in energy (kcal/day) intake at 12 months, the study is powered at 90% to detect that the Get Social condition is not inferior to the Traditional condition with a noninferiority margin of 182 kcal/day (SD=500 kcal/day) [
With N=131 available per arm (N=262 total) and alpha=.05, we have 80% power to detect differences in mean cost per participant of 0.35 SDs. For example, if the SD for cost is US $100, then we have 80% power to detect differences in mean cost per participant of US $35. With N=131 available in the Get Social condition and alpha=0.05, we have 80% power to detect correlations of .243 between continuous predictors and 12-month percentage weight loss. For categorical variables, detectable differences in mean percentage weight loss depend on the proportion of the predictor in the sample. For a predictor with 50% prevalence (ie, n=65 and n=66 with vs without the characteristic), we have 80% power to detect differences of 0.49 SDs. For predictors with 33% prevalence, we can detect differences of 0.52 SDs. For predictors with 20% prevalence, we can detect differences of 0.62 SDs, and for predictors with 10% prevalence (ie, n=13 vs n=118), we can detect differences of 0.82 SDs. For estimated SD of 5.5% for 12-month percentage weight loss, this indicates differences of 2.7%, 2.9%, 3.4%, and 4.5% weight loss.
Power calculations for noninferiority analyses were conducted using PROC POWER in SAS 9.3 (SAS Institute, Cary, NC) [
Reporting and data analyses of this trial will follow the recommendation of the 2012
Baseline participant characteristics will be examined by condition. If groups differ on any characteristics, these variables will be included as covariates in the primary analyses. Other preliminary analyses will include assessing patterns of missing data, dropout rates, distributional properties of dependent measures, and correlations among outcome measures. A series of sensitivity analyses will be performed to examine the extent of potential bias by assuming that the participants who dropped out are missing completely at random (ie, independent of the outcome), are responders to the intervention, or are nonresponders to the intervention. Multiple imputations to impute missing data will be used if more than 5% of data are missing [
We will model percentage weight loss at 12 months using a linear regression model framework, with percentage weight loss as the dependent variable and study condition as the independent variable. Test of the intervention condition indicator will provide a statistical test of the intervention effect and the estimated coefficient, along with the estimated confidence interval. This analytic approach aims to test whether the Get Social condition is not appreciably worse than (ie, not inferior to) the Traditional condition by our a priori inferiority margin of 2%. The effect size estimates will reveal clinical noninferiority of the Get Social condition if the CI lies completely above −Δ or clinical noninferiority of the Traditional condition if the CI lies completely below +Δ, that is, 2%.
Secondary noninferiority outcomes (dietary intake and physical activity at 12 months) will be examined using the same approach as described for the primary outcome. Linear multivariable regression models will be used to estimate change in daily caloric intake at 12 months; such models are reasonable, given the target sample size and that changes in these outcomes are approximately normally distributed [
We will compare total intervention costs per participant and total intervention costs per pound lost by treatment condition. As Ritzwoller and colleagues recommend [
We will compare treatment retention (percentage retained in treatment, ie, percentage who has not dropped out of treatment) and ratings of acceptability and burden (measured on 5-point Likert scales) across treatment conditions using chi-square tests.
Bivariate associations between each potential predictor of weight loss (eg, engagement, age, sociability, neuroticism, openness, and online social network use) and percentage weight loss at 12 months will be examined among participants in the Get Social condition using linear regression models. Multivariable predictors of percentage weight loss at 12 months will be estimated using linear regression models. Variables will be added to the model one at a time, in order of magnitude of the crude effect estimate (largest to smallest;
The first wave of the intervention began in August 2016. Recruitment will continue through May 2018, and we anticipate completing this study by July 2019. Results will be examined at that time.
Online social networks hold great potential for delivering lifestyle weight loss interventions for overweight and obese adults. Delivering such interventions through online social networks overcomes many of the barriers of traditional intervention modalities and provides several distinct advantages. First, online social networks address common barriers to participation such as scheduling, transportation, weather, and childcare [
Limitations of the methods described are as follows. First, it is possible that the time required to deliver intervention via online social networks is the same or more than that via the traditional modality. Being available daily, albeit for short periods of time, may be burdensome for interventionists in different ways than the traditional model. Second, participants in the study must be available to participate in either modality and thus may not represent people who are unable to attend frequent clinic visits, the very people who stand to benefit the most from the social network modality. If hypotheses are confirmed, the next step in this work should be to test the social network–delivered intervention in real-world settings and with patients who are unable to participate in the traditional modality or who strongly prefer an online intervention. Third, use of a social media platform that not all participants are familiar with introduces a learning curve that could be burdensome to some. Difficulties adopting a new technology could impact engagement to the extent that individuals do not prefer the interface, do not already have a habit of using it, or have difficulty understanding the interface (eg, among Twitter users, toggling to a different account might present a barrier to engagement to the extent they forget to switch to the new account). Future research should explore leveraging different commercial social media platforms in the context of behavioral intervention delivery.
Findings from this study may support an intervention delivery modality that is conducive to settings such as worksites, health plans, and clinics that serve large populations but have limited space, staffing, and resources for traditional in-person clinic-based behavioral interventions. If this trial is successful, approaches to dissemination and implementation should be explored as well as models that further reduce interventionist burden to explore how much costs could be reduced while retaining efficacy. Behavior change programming is conducive to delivery via connected technologies such as social media; thus, if proven cost-effective and more convenient relative to traditional models, connected health models could greatly improve the impact of behavioral interventions of all kinds.
Peer-review report from NIH for Grant 1 R01 DK103944-01A1.
body mass index
Diabetes Prevention Program
digital versatile disc
National Cancer Institute
National Heart, Lung, and Blood Institute
National Institute of Diabetes and Digestive and Kidney Diseases
standard deviation
Ten-Item Personality Inventory
The trial is supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) grant 1R01DK103944 (PI: SLP). Additional support for this paper is provided by the National Heart, Lung, and Blood Institute (NHLBI) grant 1K24HL124366 (PI: SLP), NIDDK grant K01DK102447 (PI: MLW), and the National Cancer Institute (NCI) grant R25CA172009 (DJS).
None declared.
SLP conceived of the study and research question. MLW drafted the manuscript. SLP, MEW, DEJ, ZM, JLO, SCL, JMG, and YM provided critical revisions for intellectual content. SLP, JLO, DEJ, JMG, and ZM participated in developing intervention materials and procedures. All authors read and approved the final manuscript.