Study recruitment using Facebook

Nearly all veterans in both VHA and non-VHA samples report access to and use of the Internet, with the majority using the Internet daily and over two-thirds reporting routine use for receiving health information or finding services [17–19]. The website Facebook remains a widely used social media networking site among young adult Internet users, with 89% of young adult Internet users reporting use of the site [20]. Daily activity on the site is high worldwide, with 1.09 billion daily users on average; monthly activity is even higher, especially among mobile monthly active users (1.51 billion) [21]. The population of Facebook users is heterogeneous, including similar rates of use within each group among Whites, African-Americans/Blacks, and Hispanic/Latino(a)s.

Researchers have begun using Facebook for recruitment and program delivery, with studies successfully targeting difficult-to-reach populations such as sexual minorities [22], immigrant groups [23], and youth affected by violence [24]. The site has also been used for recruitment of participants into behavioral health intervention studies, such as smoking cessation programs [25], online programs for depressed individuals [26], programs to promote condom use and deliver HIV prevention messages [27–29], and alcohol reduction programs with college students [30].

Facebook’s wide accessibility and popularity among the age group that comprises the majority of OEF/OIF veterans make it a compelling resource through which research can extend its reach to this otherwise hard-to-reach group. Though no population-level data exist on Facebook use among young veterans, groups on Facebook such as veteran service organizations like Iraq and Afghanistan Veterans of America, have over half a million followers (over 536,000 as of November 2016). In addition, pages targeted toward military family and friends are popular (e.g., Military Spouse Central has nearly 126,000 followers as of November 2016). Facebook reported that in 2014 about 4 million of its users in the United States indicated that they were veterans or active duty military, while 12.5 million users indicated they were family members of the population [31]. Thus, a large number of veterans are on Facebook and, perhaps for those that do not have Facebook accounts, their family members are on and can alert them to studies and outreach efforts targeted towards veterans.

Prior work has used Facebook to attract young veterans into research projects. For example, in our own work we recruited 1,023 young adult veterans in about three weeks into an online one-time survey study [32]. Although eligibility criteria were broad (i.e., participants needed to be between 18 and 34 and needed to be United States veterans separated from the Air Force, Army, Marine Corps, or Navy), approximately 70% screened positive for possible PTSD, depression, or hazardous alcohol use [32, 33]. The study suggested that Facebook can be used to reach veterans with behavioral health concerns outside of traditional recruitment settings and veterans that are not receiving care within the VHA. In our study, only about one-half of the veteran participants screening positive for mental health concerns (PTSD, depression, anxiety) reported receiving any mental health in the past year, while about one-fifth screening positive for hazardous drinking reported past year substance use care. Others have used Facebook to recruit veterans interested in online approaches to reduce alcohol misuse and symptoms of PTSD [34]. For this effort, researchers were able to enroll 600 veterans in just 46 days using targeted Facebook advertising (mean age of approximately 32). The majority of these participants (62%) reported use of behavioral health services in the three months prior to the intervention. These services included inpatient or residential treatment, or outpatient counseling as well as medication and/or self-help groups for mental health concerns and/or substance use. Thus, as Facebook ads contained language about receiving help for alcohol use and PTSD, this study appeared to attract either those looking for adjunct treatments to what they were already receiving or stand-alone treatment outside of traditional in-person settings. Yet many young veterans report barriers to seeking care (e.g., concerns about cost, peers or family members would look down on them if they sought care) [9, 13, 35–41] and Facebook has yet to be studied as a mechanism to reach those not specifically looking for help. Research is also needed to learn how Facebook can be used to recruit an in-need sample that can be retained for follow-up, as our prior work included a brief, one-survey of a general sample of young veterans.

The present study

The objective of this study was to describe the feasibility of using Facebook as a platform to recruit and retain young adult veteran drinkers into an online-alcohol intervention study. The intervention was designed to attract heavy drinking veterans that were not necessarily looking for treatment [42, 43]. We expand on the initial study findings from our prior work [32] and that of others [34], which demonstrated promising results with regard to the feasibility of recruiting young veterans via social media for survey and intervention research work. This study builds on previous efforts by exploring three key topics related to reaching young veterans in the community for intervention efforts. First, we discuss the advertisement strategy for the Facebook campaign, including efforts to target heavy drinking veterans not specifically searching for care. In the process, we describe the recruitment and retention rates over one month of follow up, as well as the cost and timeliness of recruitment. Given the risk of non-completion during online surveys (i.e., participants begin a survey but drop off prior to completion), we examine factors associated with attrition prior to completing the baseline survey. Second, we describe participant alcohol use behavior, symptoms of behavioral health problems (i.e., mental health problems such as PTSD, substance use problems such as hazardous alcohol use), and use of services to address these behavioral health problems. In doing so, we explore whether Facebook can be used to attract those in need of services that are not receiving them. Lastly, to better understand how best to reach and retain young veterans for research and programmatic efforts, we examine factors associated with dropout from the longitudinal study one-month after initial survey completion.

Participant eligibility criteria

Participants were enrolled for an online study to pilot test a very brief personalized normative feedback intervention (PNF) for young adult veteran drinkers [42, 43]. In this larger study, participants were randomized to receive the PNF or to receive an attention control feedback condition. Treatment effects were evaluated at one month post-intervention. To be eligible for the study, participants needed to be (1) a United States veteran separated from the Air Force, Army, Marine Corps, or Navy, (2) between the ages of 18 and 34, and (3) have access to a computer/mobile phone, Internet, and personal email address. In order to reach veteran drinkers who could benefit from an intervention, participants also needed (4) a score on the 10-item Alcohol Use Disorders Identification Test [44] of 3 for women or 4 for men [45, 46]. AUDIT criteria values were specified to include participants who drank at both moderate to high levels that were at-risk for hazardous or problem drinking. Participants could not currently be affiliated with the military through active duty service or through service in the reserve components of the United States Armed Forces (i.e., Army Reserve or National Guard, Air Force Reserve or Air National Guard, Marine Corps Reserve, Navy Reserve). Those on active duty Coast Guard or Coast Guard Reserve were also ineligible, as were Coast Guard veterans that had not served in one of the other branches of the military (Air Force, Army, Marine Corps, Navy).

Advertisement and study flow procedures

This study was approved by the RAND Human Subjects Protection Committee (2013–0663). We complied with the terms of service for Facebook for advertising and hosted the online surveys off Facebook on a secure website hosted at the institution. We named the study “Veteran Behaviors Feedback Study” and did not reference treatment for alcohol use in any advertisements to avoid deterring treatment resistant veterans from clicking on ads. The informational page and consent form referred to the intervention only obliquely: “You will also see feedback on-screen after you submit your first survey.”

A series of three Facebook ad campaigns were run to recruit participants into the study. All ads were direct promotions of the survey website, such that an individual who clicked on a Facebook ad would be immediately directed to an external website containing an information page about the study including eligibility criteria, content of the survey questions, and contact information for the Principal Investigator. We also maintained a Facebook page where participants could review this material. This page was reachable from the ads themselves and from a link on the information page. From the information page, participants could click to another page that displayed the consent form. If they consented, then they began a screening survey. If they met eligibility criteria and passed through verification questions (see below), they began the baseline survey. As specified on the information page and in the consent form, only participants who submitted their baseline survey and who met eligibility criteria received the incentive, which was a $20 Amazon gift card delivered via email. Participants who completed a follow-up survey would receive an additional $25 Amazon gift card, for a total of $45. The ad featured the total incentive for study completion.

Verification of veteran participants

Since there is no face-to-face contact with participants, a concern with Internet-based research is that it is difficult to determine if participants are misrepresenting themselves as meeting eligibility in order to receive an offered incentive. To address this, we developed a series of checks to ensure, to the extent possible, that participants were actual veterans. First, ads were targeted such that they were only displayed to Facebook users who expressed an interest in or “liked” specific Facebook content that was related to American veterans. Second, participants completed three items about their military service that needed to be consistent to progress to the baseline survey, provided that they also met eligibility criteria. Items regarding rank at discharge (e.g., specialist), branch of service (e.g., Army), and pay grade at discharge (e.g., E-4) needed to match. If answers were inconsistent, then they were exited from the survey and could not return to it using the same Facebook account. Participants that failed to meet any of the eligibility criteria (e.g., participant endorsed they were on active duty or in the Coast Guard) were also exited from the survey. Participants could only access the survey once per each Facebook account. In addition, after baseline data were collected, we reviewed the data to make sure that responses to an open-ended item about occupation in the military (i.e., military occupational specialty, enlisted classification, or specialty code), matched the rank, branch, and pay grade items. We verified that rank, branch, pay grade, and occupation matched up using sources from the Department of Veterans Affairs and the Department of Defense websites. We reviewed timestamps on survey start and end times and removed participants that completed the baseline survey in less than eight minutes. Such a timeframe was determined to be impossible given the length of the survey in preliminary tests with research assistants. Lastly, participants completing the follow-up survey one month later responded to items regarding gender, branch, and rank at discharge again to verify their responses were consistent over time.

Measures

Participants filled out measures needed to evaluate the outcomes in the larger study and to describe the sample. The screening survey included demographic and military characteristic questions, as well as the AUDIT to determine eligibility. All other items were included in the baseline survey. Participants also completed a follow-up survey that included items of gender, former branch of service, and rank at discharge. Only measures used for the current purposes are described here and other baseline and follow-up measures to achieve the larger study aims are discussed elsewhere [42].

Data analytic plan

Data analyses were primarily descriptive reports of frequencies, means, and standard deviations. We used chi-square tests to compare participants who completed the baseline survey with those who failed to complete the full baseline survey to determine whether demographic factors were associated with drop off prior to baseline survey completion. We also evaluated whether attrition at follow-up was predicted by any of the assessed baseline factors. We ran a logistic regression model predicting dropout from (1) demographic factors (gender, age, ethnicity, race, marital status, education, income, enrollment in college, currently working, manner where survey was completed [mobile phone/tablet vs computer], children [yes/no]), and a population code based on number of veterans per state [state identified from zipcode; states were categorized from 1 to 6 with 6 encompassing the states with the highest veteran population [54]], (2) military characteristics (branch of service, years of service, time since separation, combat severity, and number of months deployed as part of OIF, OEF, OND, or other conflicts), (3) treatment receipt factors (ever used a VHA, past year alcohol care at a VHA or elsewhere, past year substance care at a VHA or elsewhere, past year mental health care at a VHA or elsewhere, and distance from the nearest VHA), and (4) mental health problems and substance use (positive screens for the PCL, PHQ-8, and the AUDIT, nicotine use [including cigarettes, smokeless tobacco, and e-cigarettes], marijuana use, any other substance use).

Findings from the Facebook advertisement strategies

The ads were funded on Facebook from 6/16/15 to 6/22/15 (7 days), at which point we had surpassed an initial target sample of 600 and turned funding off, but ads that contained comments or were tagged by users could still be viewed. We kept the survey live from 6/16/15 to 6/30/15 (15 days) to allow time for some individuals who started the survey to finish it, as well as to allow time for those who learned about the study from others to access and complete the survey (e.g., a spouse “tagged” a veteran in an ad). In total, we recruited a baseline sample of 793 young adult veterans that were verified through our verification procedures. We spent a total of $3,594.78 on Facebook ads, which translated to an advertising cost of $4.53 per obtained verified participant.

Recruitment flow and non-completion of the baseline survey

Table 1 details the participant flow into the study and reasons why individuals were exited from the survey or not retained for analyses. Of the 2,312 individuals that accessed the survey from Facebook, a total of 49.1% (N = 1,135) exited the survey immediately after clicking the ad or because they indicated that they declined to consent. Of the remaining 1,177 that went on to the screening survey for eligibility, 15% were screened out of the survey for multiple reasons. The main reasons were because they did not meet eligibility on AUDIT scores (7% of those screened), they indicated they were in the reserves/guard (4%) or still on active duty (3%), or were over the eligibility age of 34 (2%). Eleven participants were screened out of the study due to our mechanism verifying consistency between branch of service and rank at discharge (1%). At baseline, nine participants could not be determined as legitimate respondents. This was due to post-data collection checks that determined three participants completed the survey within an impossible time frame (less than eight minutes), one endorsed a job in the military that we could not verify against their rank/branch, and five participants (likely the same person) used the same email address to sign up for the study multiple times. For this latter issue, the potential participant must have created five separate Facebook accounts in order to gain access to the survey more than once, since one of the checks for limiting misrepresentation was to deny access to the survey website if a Facebook user had already completed the survey (or been screened out due to ineligibility).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Participant flow.

https://doi.org/10.1371/journal.pone.0172972.t001

Of the 1002 that advanced to the baseline survey, 200 did not complete the baseline survey once they started it. There did not appear to be a section of the survey where participants were more likely to drop off. However, several demographic factors were associated with drop off from the baseline survey. In Table 2, we compare the demographic information of the baseline completers and participants who dropped off during the baseline survey. Men were more likely than women to begin the baseline survey and fail to complete it, χ² (1) = 4.03, p = .045. Those with less education were also more likely to drop off during the baseline survey, χ² (3) = 12.83, p = .005, and those who had ever used a VHA for services were more likely to drop off during baseline, χ² (1) = 9.90, p = .002. In other words, women, those with more education, and those who did not use VHA services were more likely to complete the baseline survey and get randomized to the intervention study. In addition, drop off from the baseline survey appeared to also be a function of the device used to fill out the survey. Specifically, 89.5% of those that failed to complete the survey were on mobile devices or tablets, while 81.9% of the baseline survey completers took the survey using a mobile phone or tablet, χ² (N = 992, df = 2) = 22.22, p < .001.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Demographic Comparison of Baseline Survey Completers and Dropouts.

https://doi.org/10.1371/journal.pone.0172972.t002

Description of the characteristics of the sample

Description of the behavioral health behaviors and service utilization of the sample

Overall use of the VHA.

The majority of the veteran participants (69%) reported use of a VHA for some reason since separating from the military (see Table 3). When asked how far away the nearest VHA was from their homes, 11% reported it was within 5 miles, 25% reported it was between 5 and 10 miles away, 22% reported it was between 11 and 20 miles away, 14% reported it was between 21 and 30 miles away, and 22% reported it was over 30 miles away. Only 6% reported they did not know how far away the nearest VHA was.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Service Utilization for the Full Sample and by Behavioral Health Concern.

https://doi.org/10.1371/journal.pone.0172972.t003

Factors associated with dropout from the longitudinal study

At follow-up, 631 participants completed the survey. As noted in Table 1, nine participants could not be verified as the same participant from baseline based on inconsistent responses to demographic and military characteristic items. Thus, the final follow-up sample was composed of 622 participants. This represented 78.4% of the 793 participants who completed the baseline survey and 79.3% of the 784 participants that had consistent responses to demographic and military characteristic items across surveys.

We conducted logistic regression predicting dropout from the study (as defined as not completing the follow-up survey) where we entered in demographic factors, military characteristics, treatment seeking factors, and mental health problems and substance use, controlling for the intervention condition participants received (i.e., alcohol intervention vs. control). Findings from these analyses revealed that only education predicted dropout (B = 0.65, SE = 0.19, Wald χ² (df = 1) = 11.11, p < .001), such that those with higher education were 1.9 times more likely to complete the follow-up survey. None of the military characteristics, treatment seeking factors, or mental health problems or substance use factors predicted dropout from the study at follow-up. Table 4 contains the model, including coefficients, standard errors, p-values, and odds ratios. As a sensitivity analysis to remove controlling for other factors, we similarly found that only education associated with dropout in bivariate models.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Logistic Regression Model Predicting Dropout at Follow-Up.

https://doi.org/10.1371/journal.pone.0172972.t004