The clinical mobility study, which began in July 2018, was designed to identify factors that impact balance and fall risk in people poststroke and determine how manipulating these factors influences self-reported perceptions of walking quality and the ability to recover from experimentally imposed loss of balance. The accrual goal was 40 participants. The study required a single visit to a research lab based on a university medical campus in Los Angeles and involved video-based recording of the participants’ walking patterns. After providing informed consent and completing a screening questionnaire to determine eligibility, participants completed a set of clinical assessments. These included the lower-extremity portion of the Fugl-Meyer Assessment, which quantifies motor impairment; the Berg Balance Scale (BBS) [29]; the Functional Gait Assessment [30]; a 10-m walking test; the Activity-Based Confidence Scale [31]; and a fall efficacy and fall history questionnaire if they had experienced at least 1 fall within the past year.

The participants then completed 4 walking trials on a dual-belt treadmill. They first walked on the treadmill for 2 minutes at their self-selected speed. The participants then completed 3 subsequent tests, where they were asked to modify their step lengths to match the lengths of visual targets displayed on a screen in front of the treadmill. In each test, the participants responded to rapid accelerations of the treadmill belts, which acted as perturbations of the participants’ walking patterns. The participants were compensated at a rate of $20 per hour for the time that they spent in the lab. The trial was approved by the IRB of the University of Southern California (USC; HS-18-00417-AM002).

To support research participant recruitment, researchers in the lab where the research was conducted relied on a preexisting participant registry and word-of-mouth referrals from local physical therapists to reach out to stroke survivors who might like to participate in new studies. However, this strategy often had a relatively low rate of return. As of March 2019, with 4 months left in the award period, this study was behind its enrollment target of 40 participants, having recruited only 15 participants. Missing this enrollment target would have resulted in an underpowered study and wasted resources. Therefore, the team decided to include digital recruitment methods on Facebook and Google during the latter portion of the study in 2019.

Eligible study participants included individuals 18 years and older who had a single stroke and were currently ambulatory in the community. Inclusion criteria included (1) a unilateral brain lesion from a single stroke; (2) paresis confined to one side; (3) ability to walk on a treadmill for 2 minutes; (4) absence of cognitive impairment, as demonstrated by a Mini-Mental State Examination score greater than 24; and (5) ability to provide informed consent.

The primary outcomes of interest for both conventional and digital methods included recruitment speed (enrollment rate) and sample characteristics. We captured the number of participants enrolled, the time required for enrollment, and participant demographic and clinical characteristics. These measures were used to compute the enrollment rate for each method and to determine differences in the sample representativeness. Secondary outcome measures for our digital recruiting methods included the number of post impressions (ie, the number of times a post was seen without a user necessarily interacting with it), the number of clicks on the link in the message, the number of resulting study page contacts, the number of people screened, the cost per click, and the cost per enrollee.

Recruitment-related data for each ad were collected in three different ways: (1) SM-based engagement data, such as clicks on the ads and impressions (the number of times an ad was displayed), were collected from Facebook and Google; (2) study website engagement data, including contact requests, were collected using Google Analytics; and (3) enrollment data were provided by the study team. Multimedia Appendix 5 details the technical techniques we used to collect the data from the digital platforms.

Information on recruitment sources was collapsed into two categories: (1) general practice and (2) digital (ie, Facebook and Google). Digital recruitment was further broken down to compare the recruitment results by digital platform. All statistical analyses were performed in MATLAB 2019b (The Mathworks, Natick, USA). We analyzed the normality of participant characteristics, including age, the Fugl-Meyer Assessment score, and the BBS score, using the Lilliefors test. Normally distributed data were expressed as the mean (SD) of the corresponding mean, and we compared the participant characteristics between groups (ie, recruited via general practice or digital approaches) using the Welch two-sample t test. Nonnormally distributed data were expressed as the median (IQR). We used the Mann-Whitney test to compare between groups. Significance was set at the P=.05 level.

Overall, the participants (N=40) included 15 (37.5%) women and 25 (62.5%) men, and were aged 29-78 years (mean 59, SD 12 years) (Table 1). This age range is consistent with the current literature, which indicates that stroke is most prevalent in adults aged 60 years and older [32]. Most of the recruited participants were non-Hispanic, that is, Asian (12/40, 30%), followed by one-quarter Hispanic (10/40, 25%), white (9/40, 22.5%), more than one race (9/40, 22.5%), and African American/black (8/40, 20%).

Throughout 2018, general practice recruitment proceeded slowly as most of the members of the stroke registry who were contacted either did not return the phone calls or were unable to participate in the study. Between July 1, 2018, and May 20, 2019 (46 weeks), we recruited 26 participants through our standard recruiting procedures. Between May 21, 2019, and July 16, 2019 (8 weeks), we recruited an additional 5 participants through general practice recruiting methods. Of these 31 participants, we recruited 20 participants from the existing registry and 11 through word-of-mouth referrals from local physical therapists and participants. The mean age of those recruited through general practice was 58.7 (SD 11.9) years.

A total of 8 advertised messages were posted (6 on Facebook and 2 on Google) over 5 weeks (between May 21, 2019, and June 26, 2019) (Table 2). Figure 2 shows the recruitment flow diagrams for Facebook and Google. The combined digital recruitment efforts on Facebook and Google resulted in a total of 85 valid referred potential study participants, of which 25 (29.4%) contacted the study team using the web-based contact form. Of these 25, 9 participants met the inclusion criteria, enrolled, and completed the study. The mean age of the participants recruited through digital recruitment was 59.0 (SD 10.7) years, comparable to the mean age of the participants recruited through general practice approaches.

The digital recruitment ads used different types of images, including infographics, images of older couples, and older individuals after a fall. Figure 3 shows the ad message that ran on Facebook, with the lowest cost per enrollee ($221.60) compared to a higher cost per enrollee on Google ($964.42) (Table 2). Multimedia Appendix 2 shows the recruitment ad messages used in this study. We only ran two simultaneous Google ads for about 3 weeks because of the cost of the ads and the available budget.

We found that the rate of enrollment was higher when using digital media (1.8 people/week) than when using general practice approaches (0.57 people/week). We did not find any significant differences in age, motor impairment level measured by Fugl-Meyer scores, and functional balance measured by the BBS (Table 3). Because of the small sample size, we were not able to statistically compare the outcomes related to gender and race/ethnicity for general practice and digital recruitment methods.

Our results indicate that digital recruitment methods can address recruitment challenges regarding stroke survivors. The use of digital approaches allowed us to enroll study participants and close the recruitment gap at a faster rate (1.8 participants/week) compared to using general practice approaches (0.57 participants/week). The effectiveness of SM-based recruitment methods, for example, in comparison to more general (traditional) approaches has been shown by a range of studies [33]. However, comparable data on the effectiveness of recruitment efforts for stroke survivors are currently limited. The characteristics of the enrolled stroke survivors did not differ significantly by age (P=.95) or clinical scores (P=.22; P=.82) between digital and general practice approaches. However, we observed a trend toward recruiting a higher percentage of female participants through digital methods as compared to general practice approaches. Our findings demonstrate that digital and general recruitment practices can achieve similar levels of sample representativeness, with the possibility that digital methods may even result in samples with distributions of sex that are more representative of the general population. There is limited research on the representativeness of participants recruited from digital media and SM. However, although some studies have reported issues related to representativeness and selection bias [34,35], our data support previous findings from other studies that reported the successful use of SM to recruit comparable and representative samples, as described in the systematic review by Whitaker et al [36]. Additionally, Yu et al [37], for example, were able to recruit geographically representative samples of individuals with myeloproliferative neoplasms in the United States for a survey study using multiple recruitment strategies, including Google and Facebook.

Furthermore, comparing the cost-effectiveness of Facebook and Google, we found that the use of Facebook resulted in a lower cost per click and cost per enrollee per ad. The effectiveness (with varying costs) of Facebook as a recruitment mechanism has been reported in a number of studies across disease or health topics and study types [36], such as mental health [38], mobile health studies in psycho-oncology [39], a clinical trial involving healthy elderly [40], and suicide prevention research [35]. Google ads have also been reported as recruitment methods but less frequently [41,42].

The external validity of our findings is limited. Study participants were targeted and recruited in LA County. Therefore, it is unclear to what extent the same recruitment and targeting approach will lead to comparable results in other US regions or internationally. It is also important to note that certain populations are less likely to respond to social media ads. These populations include people who are older, live in rural areas, have little or no access to the internet, have a low level of internet literacy, or do not use SM. Additionally, we were not able to conduct a complete cost comparison between digital and general recruitment approaches. At the time of study setup, the study team did not plan to compare the cost-effectiveness of different recruitment methods and did not collect related data, for example, the percentage effort spent by the study coordinator and the principal investigator on recruitment efforts and the number of potential participants contacted and screened with general practice methods. Lastly, it should be noted that the sample size recruited using SM was small in this study, and thus the statistical comparison of differences in age and clinical scores between recruiting methods should be confirmed with larger samples. Future studies should also determine whether there are meaningful differences in sex or race between participants recruited via conventional and digital methods, as these methods may differ in their ability to identify a diverse pool of potential study participants.

Digital recruitment can be used to expedite participant recruitment of stroke survivors compared to general, more traditional recruitment practices, while also achieving equivalent sample representativeness. Both general practice and digital media recruitment methods will be important for the successful recruitment of stroke survivors. The data we provide here demonstrate the potential of digital recruitment methods to aid in meeting the accrual goal without delay. Future studies could focus on testing the effectiveness of different digital platforms and include robust cost-effectiveness analyses. Additionally, examining the effectiveness of different messaging and visual approaches tailored to culturally diverse and underrepresented target subgroups could provide further data to develop evidence-based recruitment strategies.