In primary care, approximately 10% of patients experience suicidal ideation [1]. Studies indicate that these individuals often seek contact with their general practitioners (GPs) during suicidal crises [2-5] but rarely report their suicidal thoughts on their own initiative [6,7]. Moreover, GPs often feel overwhelmed when interacting with patients with suicidal tendencies [8,9]. Thus, a new short questionnaire for suicidal assessment, Suicide Prevention in Primary Care (SuPr-X), was designed and validated in various treatment settings, optimized for primary outpatient care, and aimed at guiding health care professionals when interacting with patients with suicidal tendencies. Detailed information about the development [10,11] and validation [12] of the questionnaire can be found elsewhere.

However, recruiting the required number of patients for the SuPr-X validation study proved to be a significant obstacle. In medical and psychological research, generating valid study results by collecting data from large representative samples is essential [13]. For many research teams, recruiting the number of participants needed on time and within budget limitations becomes a major challenge [14]. Over 80% of trials worldwide fail to complete patient recruitment within the initially planned time frame, with insufficient case numbers being the primary reason for discontinuing 55% of all studies [15].

Studies in outpatient care rely on physicians’ practices to recruit patients. Conventionally, the first step involves finding collaborative practices that are interested in the research topic [16]. Participating physicians then screen suitable patients and recruit them for the study using leaflets, posters, or directly approaches to potentially eligible patients. Here, two worlds often collide: the interest and willingness of medical practices to actively recruit and the constant pressure to provide care coupled with a lack of time [16]. In studies specifically within the primary care setting in the United Kingdom, less than a third of the studies managed to meet the recruitment time frame: recruitment challenges mainly stem from the GP having to shoulder the responsibility of handling patient consent and inclusion [17]. To alleviate this bottleneck, researchers need to identify different solutions.

In the last 10 years, the use and popularity of social media recruitment in medical research has gained popularity [18]. Targeted advertisements on commonly used search engines such as Google are also often used to extend visibility and cover a more extensive outreach to potential participants to complement traditional recruitment methods, such as email or phone lists, prints, and leaflets [19-21]. Despite the growing use of social media for patient recruitment in mental health studies, the effectiveness and cost-efficiency of Instagram and TikTok remain underexplored. While Google and Facebook are well-established methods, studies on newer platforms such as Instagram are limited and primarily focus on younger audiences [22,23]. Social media provides the opportunity to reach a broad and diverse target audience and is helpful for highly focused recruitment approaches in a hard-to-reach population [18]. The time and costs involved are generally significantly lower than those associated with traditional recruitment methods [18,24-26]. However, potential limitations in representativeness must be considered [27]. When explicitly focusing on mental health studies (n=176), social media recruitment is indeed a cost-effective approach [28]: It was primarily used in studies including substance abuse (43.8%), mood disorders (15.3%), general mental health (11.9%), trauma and stress-related disorders (8.9%), and suicidal ideation and self-injurious behaviors (4.5%). In around two-thirds of the studies, social media performance was just as good or superior to traditional strategies such as mailing lists or referrals [28]. Moreover, participant characteristics from patients recruited online only differ minimally compared to those recruited from general practices, as a study on urinary incontinence found [29]. A web-based mental health program study showed that social media had the highest reach, lowest cost per participant, and similar adherence rates compared to general practice and pharmacies [30]. Although some studies have explored the advantages and disadvantages [18,24-28], there is still a notable gap in understanding the precise mechanisms and factors by which various social media platforms can be most effectively and cost-efficiently used for patient recruitment in mental health studies.

At the beginning of the SuPr-X study in 2022, we could not motivate GPs to cooperate due to the increasing spread of the highly contagious COVID-19 Omicron variant [31]. Conducting research on other topics became increasingly challenging [32]. It was not until spring 2023 that the situation improved, and study recruitment accelerated, resulting in some patients being included in practices. However, by this time, any buffer for recruitment delays had been exhausted, making the recruitment target of 500 seem no longer achievable by the previously used conventional method. Starting in September 2023, we used an online recruitment strategy to support the project.

This study aims to evaluate the effectiveness of recruitment strategies in validating a suicidality assessment tool for primary care settings.

This study used a cross-sectional observational design. For the SuPr-X validation study, we recruited adult patients with depressive symptoms from outpatient settings (GPs and psychotherapeutic or psychiatric practices) and from partial or full inpatient settings and online (subsequently enrolled on-site). The inclusion criteria required a Patient Health Questionnaire-9 (PHQ-9) score >5 and the ability to complete a 30-minute German-language questionnaire. Suicidal thoughts were not mandatory. People were excluded from the study if they had cognitive impairment, had significant positive psychotic symptoms (eg, hallucinations or disorganized thinking), or were minors.

We used Google Ads Manager, AdEspresso (Facebook and Instagram advertisements management), and TikTok Ad Manager to set up the online campaigns targeted at Munich and nearby areas (a radius of 40 km). We implemented tracking pixels as described by the respective advertisement platforms. The advertisements forwarded interested users to the study website [38]. As per the General Data Protection Regulation, we implemented a cookie consent management platform (Borlabs Cookie). Only if users explicitly consented to advertising cookies did we track and include them for website statistics analysis. On the study website, the online prescreening questionnaire was embedded via HTML iframe (survey provider: Lamapoll [Lamano GmbH & Co KG], all data stored on German servers). Website visitors were screened with the PHQ-8 to ensure study participants showed at least mild depressive symptoms (cutoff >5). Three distinct questionnaire exit pages measured the outcomes of the prescreening questionnaire. They were used for conversion tracking of participants who (1) were eligible and left their contact data (“eligible opt-in”), (2) were eligible but did not leave their contact data (“eligible no opt-in”), and (3) were not eligible (“not eligible”).

We defined the desired conversion action as a user passing the prescreening questionnaire and providing their contact details (tracking done using the web page “eligible opt-in”); hence, a conversion represented an eligible patient lead. Throughout answering the online survey, participants remained entirely anonymous. Only the outcome page was then tracked again. Thus, end-to-end tracking from click to enrollment was not possible without the user’s consent; their data were transferred to Trials24’s system only when the participants passed the prescreening questionnaire, explicitly consented to the data protection declaration, and entered their contact details. We calculated the number of eligible patient leads passing the prescreening questionnaire to fill in their contact details by the respective tracking pixel’s measurement (“tracked” leads) or counting the entries in our database (“DB” leads). For some advertisement experiments, for example, pixel versus no pixel, additional versions of the website and the prescreening questionnaire were created so traffic from online advertisements could be directed to a specific version of the web page and prescreening questionnaire, allowing distinct measurements even without conversion tracking.

The online prescreening questionnaire did not include a special captcha question for participants to prove they were human, but the contact center manually checked every contact in the database.

The association between the recruitment method and gender distribution within the different samples was analyzed using the φ coefficient for nominally scaled items. With a 2-tailed t test, we determined if there were group differences regarding the recruitment method (traditional versus online) and age. Nonparametric tests are widely recommended when the assumptions of parametric tests are violated, such as interval scaling and normal distribution. Therefore, they are particularly suitable for ordinally scaled data, which are often found in psychological or medical questionnaires [39]. Data were not normally distributed for each online recruitment channel (Shapiro-Wilk test, P<.05); therefore, we chose a nonparametric test: the Kruskal-Wallis test was performed to examine whether age differences exist depending on the online recruitment channel. A nonparametric Mann-Whitney U test for ordinally scaled items was conducted to assess possible group or subgroup differences regarding depression symptoms measured by PHQ-9. We also used the Mann-Whitney U test to determine possible differences in reported suicidality between recruitment methods measured by the SuPr-X Risk scale (sum score of items 5-11) and BSS screening (sum score of items 1-5). For pairwise comparisons and post hoc tests, Bonferroni-adjusted P values were reported to account for multiple testing. For nonparametric tests, the rank-based effect size r was used, calculated by dividing the test statistic’s z statistic by the square root of the total number of observations (N), with divisions into small (0.1-0.3), medium (0.3-0.5), and large (>0.5) effects [40]. To compare the effectiveness of recruitment methods, a linear regression analysis was performed to estimate the monthly participant enrollment rates (slopes) for the traditional and online recruitment methods. We quantified the magnitude of differences between slopes using Cohen d, calculated as the difference between group means or slopes divided by the pooled SD, with Cohen d=0.2 (small), Cohen d=0.5 (medium), and Cohen d=0.8 (large) [40]. Any P value <.05 indicated statistical significance. We used SPSS (version 29; IBM Corp) for all calculations.

The study and all recruitment activities received ethics approval from the medical ethics committee of Ludwig-Maximilians-Universität München (Munich) on May 9, 2022 (project number 22-0028) and were performed in accordance with applicable guidelines and regulations. The amendment for additional online recruitment was approved on August 8, 2023. All participants signed the informed consent on-site after an informed prestudy briefing with the respective clinician from the cooperating practice or, in the case of online recruitment, on-site at the Institute of General Medicine with the responsible study physicians or psychologists. All data were securely stored, and access to data storage was restricted to authorized personnel only. All participants received a €25 shopping voucher as compensation. The conversion rate (CR) at the time of the study was €1=US $1.074.

From July 2022 to February 2024, a total of 521 patients with depressive symptoms (inclusion criteria: PHQ-9 >5 “mild depression”) were included in the study: 67.8% (n=353) women, 31.7% (n=165) men and 0.38% (n=2) nonbinary participants (n=1 missing, 0.19%). The mean age was 40.89 (SD 14.32) years, and the average PHQ-9 score was 14.8 (SD 4.97), indicating moderate severity [36].

After presenting the overall sample characteristics, we examined how traditional and online recruitment strategies influenced differences within the sample.

The online recruitment strategy focused on outpatient demographics. For reasonable comparative analysis of sample characteristics using traditional versus online recruitment methods, the inpatient psychiatric cohort was excluded to avoid biases. However, online recruitment included an outpatient psychiatric group, which did not have a corresponding comparative group within the traditionally recruited cohort (only the previously mentioned excluded inpatient psychiatric group). As a result, a feasible comparison of sample characteristics among the subgroups “traditional” versus “online” could only be conducted within the GP and psychotherapist samples, as detailed in Table 1. Moreover, 74 (14.2% of the 521 patients) from the psychosomatic day clinic were considered for the total result but are not separately reported in the table.

Table 1 shows gender, age, depressive symptoms mean scores (PHQ-9), and suicidality mean scores (SuPr-X risk: 5-11; BSS screening: 1-5) of the traditional sample in comparison to the online recruitment sample. Subgroup comparison of participants are attributed to GPs and psychotherapists.

In summary, we recruited participants through traditional methods and online (Table 1), with neither significant difference in gender distribution across recruitment methods (φ=0.051; P=.54) nor age differences (t₄₆₅=–0.1886; P=.06). Depression scores also did not vary significantly between groups (U=26,314.500; Z=–.479; P=.63). We observed an increased suicide risk in the online recruitment group (SuPr-X: U=22,969.500; Z=−3.025; P=.002 and BSS screening: U=23,611.500; Z=–2.323; P=.02) with small effect sizes (SuPr-X: r=0.140; BSS screening: r=0.108), as reflected in significantly higher scores on SuPr-X by 1.04 (95% CI 0.41-1.68) points and on BSS screening by 0.53 (95% CI 0.13-0.94) points. However, no significant difference was found in subgroup comparisons (GP and psychotherapist), indicating that the higher rates of suicide risk in the online recruitment sample were attributable to the psychiatric outpatient sample.

Building on the comparison of traditional and online strategies, we further explored differences in sample characteristics across specific online recruitment channels.

Online recruitment channels were classified as (1) Facebook, (2) Instagram, (3) TikTok, (4) Google, (5) internet, and (6) others. The category “internet” included all responses from participants that could not be attributed to a specific online platform or search engine. “Others” included responses from participants indicating they learned about the study through people in their networks who had seen the study advertised online.

There was no statistically significant association between online recruitment platforms and gender distribution (φ=0.188; P=.21). The group recruited via TikTok (median 23, SD 9.238) was significantly younger than those recruited through Facebook (median 47, SD 11.479; Z=5.410; P<.001); Google (median 46, SD 15.005; Z=–4.670; P<.001); and other internet sources (median 52, SD 14.614; Z=–5.308; P<.001), with medium (Facebook r=0.489) to strong (Google r=-.704; internet r=0.766) effect sizes.

The mean age differences were –18.26 (SD 11.479, 95% CI –22.80 to –13.72) years for Facebook, –20.40 (SD 15.005, 95% CI –27.70 to –13.10) years for Google, and –21.94 (SD 14.614, 95% CI –28.72 to –15.16) years for other internet sources. Age did not differ significantly from Instagram (median 29.5, SD 16.796; Z=2.038; P=.62). These age differences suggested that TikTok may attract a younger demographic for online study recruitment (Figure 1). The severity of depression symptoms did not depend on the online recruitment platform (Kruskal-Wallis H₅=8.773, P=.12).

Box plots illustrate the median age and the lower and upper quartiles, minimum, and maximum values across all online recruitment channels compared to the traditional recruitment sample.

The comparison of recruitment methods revealed a highly significant difference in participant enrollment rates (Figure 2). Linear regression analysis estimated a slope of 17.95 (95% CI 15.44-20.46) for the traditional recruitment method and 41.54 (95% CI 32.74-50.33) for the online campaign. The difference between the slopes was 23.59 (SE 3.62), which was highly significant (P<.001; Z=6.51). The effect size, expressed as Cohen d, was 0.22, indicating a small to moderate practical effect. In addition, the correlation coefficient r=0.79 reflects a strong relationship between the recruitment methods and their differences in performance.

Building on the timeline analysis, we illustrate the process of converting clicks into enrolled participants, noting that comparable data for the traditional setting were not collected.

Figure 3 describes the typical patient journey during online patient recruitment, the “patient journey to enrollment.” The advertisements were displayed 890,145 times (impressions), and 6364 (0.7%) people showed interest by clicking on the study advertisement. In addition, 72.6% (4620/6364; CR from click to website) of website users landed on the study website, and 54.8% (2531/4260; CR from website to survey) individuals started the online prescreening survey. In addition, a 53.6% (1356/2531) CR from survey to eligible respondents was met for the study-specific inclusion criteria, making them potentially eligible and reaching the end of the questionnaire. Of those 1356 individuals, 1026 individuals (75.7% opt-in rate; CR from eligible to lead=opt-in rate) left their contact data, becoming an eligible “lead.” When the patient companion service—a dedicated contact center for interested patients—reached out to them, 65.7% (674/1026) CR from lead to people responded, and an 89.2% (601/674) (CR from response to referral=potential referral rate) were placed on the study waiting list because there were only limited slots available for participation.

Overall, 59.7% (359/601) of individuals (CR from referral to first site visit=potential first site visit rate) received an appointment slot. There were reminders 2 days and 1 day before the appointment so participants could cancel their appointment in time if they could not attend (appointments were rescheduled, and open slots were given to the next person on the waiting list). Only 5.3% (19/359) of individuals were listed as “no-shows.” Of those who attended their appointment, 66.7% (239/359; CR from first site visit to enrolled) of individuals were subsequently enrolled.

The patient journey to enrollment in the online recruitment funnel starts with the first click on the study advertisement and then continues to the website visit, before qualification via the online prescreening questionnaire, making an appointment with the patient companion service, and finally ends with participation in the study by signing the informed consent form and being enrolled. The following conditions apply: clicks=registered click on a study advertisement; website users=people on the study website; survey started=people who have started the online prescreening questionnaire; eligible=people who showed potential eligibility online; leads=people who have provided their contact details after being potentially eligible; responded=people who responded when the patient companion service reached out to them; waiting list=all people who wanted to receive an appointment for the study; appointment received=people who received an appointment slot; enrolled=people who completed the study according to the study protocol.

Beyond CRs, we also presented detailed online advertising efficiency metrics to comprehensively assess cost-effectiveness.

We used Facebook advertisements to compare the gender-specific differences in online campaigns. The advertisement spent on men and women was equal, but CPLs and CPE for women were only half as expensive. Metrics for women were as follows: CPL tracked=€9.17; CPL calculated=€6.20; CPL database=€6.32; and CPE=€26.48. The metrics for men were as follows: CPL tracked=€17.87; CPL calculated=€11.92; CPL database=€14.78; and CPE=€55.53. This matched the overall 2:1 ratio of women and men enrolled in the study (Table 1). The total online numbers for women were 725 leads and 160 enrolled, and the totals for men were 302 leads and 79 enrolled.

As we found gender differences in the recruitment of participants, we also wanted to find out whether there were key factors related to the creative material that influenced recruitment.

This study aimed to evaluate the effectiveness of recruitment strategies while validating a suicidality assessment tool optimized for primary care. We found that online recruitment accounted for 46% (239/521) of participants, and they were recruited within a short time frame, demonstrating its potential for reaching vulnerable populations. The samples recruited online and using traditional methods were comparable regarding key demographic and clinical characteristics. Recruitment costs for online methods were similar to traditional approaches, while enrollment was significantly faster. In addition, impact factors such as advertisement formats, landing page design, and pixel tracking substantially influenced recruitment efficiency.

In our analysis, age, gender, and the inclusion criteria for depressive symptoms did not differ among patients based on different recruitment methods (traditional vs online). For ethical reasons, online advertising for the study only mentioned depression and not suicidal ideation. Both groups showed an overrepresentation of women compared to men in a ratio of 2:1, which, however, corresponds to the general gender-specific distribution of depression [42].

Regarding the representativeness of demographic factors (age and gender) and the inclusion criterion (depressive symptoms), patients recruited online were comparable to traditional strategies. Analysis of online recruitment channels showed that TikTok, in particular, targeted younger audiences and could be used strategically to attract a younger sample.

An argument in favor of using social media recruitment in mental health studies is an observed correlation between time spent on social media (r=0.11), addictive use (r=0.29), and depressive symptoms [43], as well as anxiety, low sleep quality, and reduced self-esteem in adolescents. This correlation appeared to be stronger within the female sample [44]. These findings align with our study, where a significant proportion of participants were recruited through social media, including younger people and women, highlighting the platform’s potential to engage vulnerable populations effectively. Consistent with previous suicide prevention studies, Facebook was excellent in generating a broad and favorable reach to attract participants for a survey [45]. However, similar to the findings by Whitaker et al [27] and Tang et al [46], young women were overrepresented. This aligns with broader research indicating that female participants may be more responsive to social media recruitment campaigns, potentially due to higher engagement with mental health–related content [43,44]. A frequently cited limitation when using social media is the representativeness of the collected data. Because targeting can be adjusted to a certain extent in these campaigns, social media advertising can specifically address underrepresented groups to increase patient diversity. Yet, advertising costs would have become more expensive.

Online-recruited patients appeared especially suitable for the validation study, as they reported higher average levels of (self-reported) suicidality than the traditionally recruited cohort, thus providing valuable data for validating our new suicidality questionnaire. The online cohort also included 10 individuals of the “under-the-radar” population, defined in suicidology as those who are mentally distressed and suicidal but have minimal or sporadic contact with the health care system and hence are often missed by preventive measures [46]. We identified this group as patients who confirmed suicidal ideations or behavior in the questionnaire but also stated they were not receiving medical treatment and had not reached out for help due to mental stress. Clinical services often face challenges in reaching individuals most in need of mental health care. A systematic review [46] found that nonreceipt of services was linked to factors such as being male, younger or older age, rural residency, and minority ethnicity. Those without access were also less likely to have a psychiatric diagnosis, a history of suicidal behavior, or contact with health services but more likely to use violent suicide methods. These findings highlight the need for tailored, multifaceted strategies to overcome barriers and engage diverse at-risk groups effectively [46]. However, no clear associations could be derived from our data in this regard. Nevertheless, the campaign successfully reached out to them, and they were enrolled in the study. Suicidality was not mentioned in social media advertising, yet it appeared to have reached a demographic that traditional methods might have missed. Enrolling such underrepresented patients underlines the role of social media campaigns to increase patient diversity in research studies and clinical trials.

Raising awareness is not an issue when recruiting patients online, as you can use paid advertising to display your messages to a broad audience. The critical part is to find participants who click on the advertisement, complete the survey, and provide their contact information. A meta-analysis by Brøgger-Mikkelsen et al [47] showed that online recruitment costs were generally lower than offline recruitment, but costs varied greatly: the median price per enrollment for online recruitment strategies was US $72 (range US $4-$251), and the median price per enrollment for offline recruitment strategies was US $199 (range US $19-$839). In our study, both recruitment strategies—online and offline—resulted in similar costs of approximately €30 per enrollment, positioning them in the lower range of reported online and offline recruitment costs. This highlights the cost-efficiency of both approaches in our specific context. Even if additional expenses for personnel costs are included (approximately €9 to €15), the overall recruitment costs are still in the lower to middle range. Furthermore, enrollment was completed before we could fully use the optimized advertisements. This means using only the most cost-effective creative material and the advertising platform’s artificial intelligence–based targeting algorithm without our restrictions regarding A/B split tests. If we had recruited for a longer period and only run the most optimized and efficient advertisements, we estimate that it would have been 40% to 60% more cost-effective. Furthermore, our appointment slots were limited. Actual online recruitment costs might have been half the reported or even less if every participant on the waiting list could have been included. Surprisingly, only 5.3% (19/359) of individuals did not show up, indicating that the target population was eager to participate in the study. Given that depression primarily occurs with symptoms such as a lack of energy and motivation, the observed low no-show rate was a pleasant surprise.

When specifically looking at clinical trial recruitment, little is known about which factors impact advertisement efficiencies. These advertisements require ethically correct phrasing and must align with country legislation regarding advertising in health care. In Germany, this is, for example, the Heilmittelwerbegesetz (translating to “Health Services and Products Advertising Act,” a law that prohibits advertising for prescription drugs). You can apply some strategies from online marketing; however, the overall toolbox is restricted.

One factor influencing advertisement efficiency was the social media platform used for advertising, as each platform has its predominant age group [48]. A rule of thumb with social media platforms is that teenagers and young adults mainly use the newest and trendiest platforms. At the same time, older people only switch to these platforms later or do not switch at all, resulting in social media platforms “aging up.” We also saw this effect with the TikTok population being significantly younger, followed by a slightly older group on Instagram, and Facebook having the oldest users.

When comparing the images, it was interesting to see that even in the male audience, the image of the woman performed best. It must be mentioned that the stock photos of the sad woman and man did not display the same posture and color scheme, so comparability was somewhat limited. Yet, the observed effect could be a result of increased self-stigmatization in male depression [49].

The use of negative imagery, such as depictions of sad individuals, in participant recruitment raises ethical and methodological concerns. Research suggests such portrayals may perpetuate stigma, deter help-seeking behaviors, and reinforce harmful stereotypes [50,51]. Bennett [52] highlights the limitations of visual clichés, noting they oversimplify mental health conditions and may influence who participates in studies.

When looking at image and video advertising for clinical trials, we did not find any reference that specifically showed insights into how the creative material was designed and which effects resulted from it. In our case, videos outperformed classical image advertisements, even though we used a relatively simple video setting of a person explaining the content. All our texts and scripts followed established conversion copywriting rules adapted to the advertising platforms and legal and ethical compliance guidelines. Therefore, the information flow was always the same, only presented in a different format. Because clinical trials are a medical topic, we were interested in how well a video featuring a medical physician performs compared to a layperson. A systematic review analyzing the impact of a physician’s attire on patient perceptions discovered that 60% of patients preferred a white coat, especially among older patients [53]. However, in our study, older patients seemed to prefer casual outfits, at least on social media. We assume that young adults may already be used to seeing medical influencers on social media, as this generation is also informing itself more and more about health topics online [54].

A well-researched fact is that the online attention span is extremely short, with the first 10 seconds being crucial as to whether a user continues reading or leaves the website [55]. We were interested in which effect might be stronger: condensing the information to a minimum word count so users could see all relevant information without needing to scroll much on their mobile phones or having a longer version with each paragraph spanning one topic. In the small sample we tested, the longer copy worked slightly better for finding potentially eligible participants for the study, indicating it was not the number of words but the text quality that made a difference.

Finally, we gained a technical but essential insight regarding advertising efficiencies when comparing advertisements with and without a tracking pixel. CPC is often used as an indicator of the advertisement’s cost-efficiency, even though it does not show whether the generated traffic is indeed quality traffic [56] that leads to conversions (which is the actual study leads). The CPC without a tracking pixel was lower, but the CPL—the more critical outcome—was 6 times as high. Ethics committees assessing study recruitment materials sometimes struggle to permit this kind of tracking, even though users in the European Union can opt out of being tracked via cookie consent options. Our data clearly show the need to allow this pixel-based tracking technology; otherwise, researchers must invest significantly more capital to find eligible study participants. This creates an additional financial burden, especially for small research organizations or universities with limited resources and delays the studies. Patients may have to wait longer for potential treatment to be approved, or studies may be prematurely terminated due to not reaching recruitment goals.

Having discussed the key factors influencing recruitment outcomes, we now acknowledge the limitations of our study to provide a balanced interpretation of the findings.

In the traditional recruitment setting, clinicians approached patients during routine practice but did not document the total number of patients approached. Only data for included patients were submitted, making it impossible to calculate precise CRs compared to online recruitment. In the online setting, small sample sizes often constrained the subanalyses of the considered impact factors, precluding generalized recommendations concerning image versus video materials, pixel tracking, and how much informational content should be provided. Variations in the images’ graphical design may also contribute to whether a picture gains user attention (Multimedia Appendix 2). Stereotypical imagery may have introduced selection bias by attracting participants who identified with these portrayals while deterring others, limiting sample diversity.

The metrics CPL tracked and CPL calculated are influenced by each user’s tracking blockers and technical settings, while CPL database relies on the accuracy of the individual’s statement. As a result, both technical measurements and human assessment may be prone to error. The low no-show rate may be attributed to our reminders and a common selection bias of motivated participants by the patient companion service. Furthermore, it remains unclear to what extent the results would change in other settings, for example, due to cultural influences on the perception of a person with or without a physician’s coat in an advertisement.

Online recruitment demonstrated high efficiency, enrolling 46% (239/521) of the study patients quickly and providing a diverse sample representative of the study’s objectives. This approach proved particularly effective in reaching vulnerable, hard-to-access populations and enhancing patient diversity. The comparable costs and faster recruitment rate relative to offline methods highlight the practical advantages of online campaigns. While key insights from tested impact factors—such as the superior cost-efficiency of video advertisements, longer landing pages, and pixel tracking—offer promising strategies for optimizing recruitment efforts, these findings are preliminary and based on limited data. Further research with larger datasets is necessary to generalize these results and refine recommendations for future studies.