Soon after the outbreak of the COVID-19 pandemic, health-related misinformation flooded the social media space [1,2]. Compelling, but often misleading, content captured the attention of a frightened global community [2]. The rapid and widespread dissemination of such misinformation on social media often overshadowed evidence-based recommendations released through more traditional public health communication channels. As a result, dangerous messages that increased the spread of COVID-19 and led to adverse health outcomes were allowed to spread to the estimated 3.8 billion people worldwide who use social media [3]. Tedros Ghebreyesus, Director-General of the World Health Organization warned, “We’re not just fighting an epidemic; we’re fighting an infodemic” [2].

There is a critical need to rapidly disseminate evidence-based informational videos on social media channels to counteract the epidemic of COVID-19 misinformation. To date, public health efforts have focused on correcting misinformation and debunking myths [4]. As such, these measures have almost exclusively been reactive rather than proactive. The corrective content itself has not been designed to incorporate the very characteristics that support the viral spread of content on social media [5]. For this reason, social media interventions designed to correct misinformation have unfortunately demonstrated far less impact than the content they aim to correct [4]. Researchers studying this emerging global health communication approach have urged health authorities to enter the social media arena more intentionally, with the aim of disseminating valid information, evaluating its impact, and reducing the knowledge translation gap [3]. Social media health messaging interventions need to do more than convey reliable information. They must be as emotionally compelling as they are evidence-based, if public health authorities are to reach broad, global audiences [5]. They also need to be accessible and tailored for cross-cultural acceptability [6].

In March 2020, we designed a short, wordless, and animated video to disseminate information about preventing the spread of COVID-19 [7,8]. The intervention video (the CoVideo) promotes evidence-based messages that focus on a set of preventive behaviors such as hand washing, social distancing, and the sanitation of kitchen surfaces, among others. Importantly, the CoVideo incorporates audience engagement characteristics that motivate widespread sharing on social media [5]. For example, it includes a compelling, familiar narrative and characters that are culturally agnostic; and the soundtrack is designed to evoke high-arousal emotions [9], which reflects the anxiety, altruism, and solidarity [10] of the global community. The CoVideo was released on Stanford Medicine’s YouTube channel on March 21, 2020, and within 10 days reached 332,000 views on YouTube, 220,000 views on Instagram, 294,000 views on Facebook, and 402,000 views on Twitter, with a cumulative count of 1.2 million [6]. It continued to spread organically across social media channels, due to reposting by several global health authorities, including government departments of health, community health organizations, and media channels around the world [6]. Within 4 months, the CoVideo had reached more than 5.8 million people through their social media accounts.

In this study, we evaluate the effectiveness of the CoVideo to improve knowledge and behavioral intent toward COVID-19 prevention. According to the Theory of Planned Behavior, the intention to act is considered to be the immediate determinant of action [11]. Here, we frame behavioral intent as representing the participant’s commitment to undertake COVID-19 prevention behaviors in the next 7 days, which is the second outcome of our study [12,13]. As the primary outcome of our study, we aim to measure changes in knowledge about COVID-19 prevention. Knowledge is often considered to be a necessary but not sufficient condition for motivating a healthy behavior [14]. Specifically, the Theory of Planned Behavior posits that knowledge is more likely to be correlated with behavior if correct answers on the knowledge test support the practice of that behavior [15]. Results from this study, which incorporates several innovations in global health communication, can inform the development of future videos to disseminate evidence-based recommendations related to COVID-19 and other public health emergencies.

Between May 13, 2020, and June 23, 2020, 15,163 participants from the United States, Mexico, the United Kingdom, Germany, and Spain were enrolled in our RCT. Between recruitment and randomization, 171 participants were lost and 14,992 participants were randomly assigned to the CoVideo (n=4940), APC (n=4954), and do-nothing (n=5081) arms (Figure 1). After randomization, another 173 (do-nothing), 177 (APC), and 143 (CoVideo) participants were lost for unknown reasons (possibly due technical issues like lost internet connection; difficulties linking to the video host, YouTube; server complications, etc). A total of 14,482 participants completed the trial and contributed data to the final analysis.

The majority of participants reported their residence in the United Kingdom (n=8519, 58.8%) or the United States (n=3765, 26%), and 84.9% (n=12,288) of participants reported English as their first language. The sample was relatively well educated, with 81.6% (n=11,812) having some college education or higher (bachelor’s, master’s/equivalent, or PhD). Table 1 shows the percentage of participants in each arm and treatment list by age, gender, country of residence, educational status, and primary language.

The knowledge questionnaire had an acceptable reliability correlation coefficient of 0.65 (split-half). Overall, there was extraordinarily high attainment of COVID-19 knowledge. In the do-nothing arm, participants correctly answered 16.86 (95% CI 16.83-16.90) out of 18 items, which is a 93.7% correct response rate (Figure 2). With this high baseline score, the CoVideo could therefore only increase knowledge by a maximum of 1.14 points. Relative to the do-nothing arm, the CoVideo increased knowledge by 0.09 points (mean 16.95, 95% CI 16.91-16.99; P=.002), which represents an increase of 7.6% (0.09/1.14) (Figure 2). The average score for the APC arm was 16.89 (95% CI 16.86-16.93), a correct response rate of 93.8%. When we removed the attention effect of the video format, the CoVideo increased overall knowledge by 0.06 points (P=.06), which represents an increase of 5.3% (0.06/1.11). Figure 3 shows the proportion of correct responses to each of the 18 knowledge items (see also Table S1 in Multimedia Appendix 1). The highest correctly answered item (“An effective way to prevent COVID-19 spread is to wash your hands frequently with soap and water”) had a correct response rate of 99.4%; most items had a >90% correct response rate.

Figure S1 in Multimedia Appendix 1 shows the mean scores for the five list experiments by trial arm and list group. These mean scores were used to calculate the prevalence of behavioral intent for each preventive COVID-19 behavior, including the total and content effects with 95% CIs and P values (Figure 4). Scores for the treatment list are higher because the treatment list has 6 items and the control list has 5 items. For a given trial arm, the difference between the treatment and control means represents the prevalence of intent to undertake the preventive COVID-19 behavior. For example, for the first list experiment in the CoVideo arm (“this week I will go out with friends”), =2.20 is the treatment mean and 2.03 is the control mean. The prevalence is then × 100 = (2.20–2.03) × 100 = 17.2, as shown in Figure 2. Similarly, the prevalence for the APC arm is × 100 = (2.33–2.03) × 100 = 29.4. For our secondary outcome, we report that participants in the CoVideo arm had lower behavioral intent to go out with friends when compared with the APC (content effect=–0.123, P<.001) and do-nothing (total effect=–0.045, P=.24) arms.

In this study, we tested an intervention with several innovations in global health communication that catalyzed a broad, organic global reach on social media [6,8-10]. The intervention, called the CoVideo, packaged critical health messages about COVID-19 prevention within a compelling, familiar narrative, using characters that were free of cultural identifiers and a soundtrack designed to evoke high-arousal emotions. Our results showed that baseline levels of COVID-19 prevention were high, and that the CoVideo intervention increased this prevention knowledge by another 7.6% and 5.3% relative to the do-nothing and APC arms, respectively. It was also found that the CoVideo intervention improved behavioral intent toward COVID-19 prevention when compared with the APC and do-nothing arms.

To evaluate the effectiveness of the CoVideo on knowledge and behavioral intent toward COVID-19 prevention, we used a large, online RCT to enroll 15,163 participants from the United States, Spain, Germany, the United Kingdom, and Mexico. The results for our first endpoint showed high knowledge of COVID-19 prevention behaviors across the five countries. For the three trial arms, the average number of correct answers was nearly 17 out of 18 items, a correct response rate of approximately 94%. Moderate to high knowledge levels about COVID-19 prevention measures among the general public were also observed earlier in Ecuador [29] and the United States [30-32]. On the other hand, a recent systematic review on knowledge, attitude, and practices toward the COVID-19 pandemic on the American continent concluded that many people have insufficient knowledge about the virus, highlighting the need to develop effective educational tools and materials on COVID-19 prevention [33]. The high baseline levels of COVID-19 knowledge in our study could be due to the delay of several weeks that occurred between the original release of the CoVideo and the launch of our online trial, as we awaited ethics approval, and designed and registered the trial. This lag likely facilitated exposure of our participants to COVID-19 prevention messages from other sources. Our results suggest, as we drift deeper into the pandemic, it may be unnecessary to spend more money on public health campaigns to improve COVID-19 prevention knowledge in the five countries from which we enrolled participants.

An important study finding was that the CoVideo improved already high levels of COVID-19 prevention knowledge. In the do-nothing and APC arms, only 1.14 and 1.11 additional correct items were needed to reach a perfect (100%) score, respectively. Our results showed that the CoVideo boosted COVID-19 prevention knowledge by another 7.6% relative to the do-nothing ceiling and by 5.3% relative to the APC ceiling. It seems plausible, therefore, that the CoVideo could significantly improve COVID-19 prevention knowledge in countries where baseline knowledge levels are currently low or moderate.

For our second endpoint, we nested a list experiment in each trial arm to evaluate the effect of the CoVideo on self-reported behavioral intent toward COVID-19 prevention. We used this experimental approach because it is likely that participants (at the time of enrollment) were already primed to give socially desirable responses to questions about COVID-19 prevention. The indirect questions (ie, how many statements do you agree with) provide protection to participants who have no behavioral intent toward COVID-19 prevention, without revealing this intention directly [27]. Our results showed that behavioral intent to go out with friends during stay-at-home recommendations and to stockpile household goods was lower in the CoVideo arm when compared with the APC and do-nothing arms, but not significantly so. We also observed that participants had higher behavioral intent to prevent COVID-19 spread by cleaning dishes after use when compared with the do-nothing arm (significantly different) and APC arm (not significantly different). Several studies have used the list experiment technique in the context of COVID-19 and found that list experiments were less favorable than simpler, traditional measurements, concluding that social desirability had no impact on the reported compliance with COVID-19 regulations [34,35]. On the contrary, other scholars have argued that the list experiment approach counters social desirability and is, therefore, less likely to introduce measurement errors presented by direct questions that measure self-reported compliance with COVID-19 guidance [36,37].

Our study is innovative in its use of both a list experiment and an APC video. Our APC video was selected to account for the possible attention effects elicited by the CoVideo intervention. The APC was designed to mimic the inactive components of the CoVideo intervention (the effect of watching a video of the same length), while not containing the active intervention component (the content of the COVID-19–related health messaging and its delivery design) [22]. The APC, therefore, enabled us to decompose the total intervention effect, which is the difference in knowledge means between the CoVideo and do-nothing groups, into the sum of the content and attention effects. We are not aware of any study that has used this approach to isolate the active component (the content effect) of the intervention video itself. For this purpose, we advise researchers using APCs to choose their APC topic carefully, and to avoid any potential effect of the placebo content on the outcomes being studied.

Our study had several limitations. At the time of our study, no validated scale on COVID-19 knowledge prevention existed. Nevertheless, we used best practices from the survey methods field to inform the design and development of the knowledge questions [38]. Another limitation is that we could not determine if participants watched some or all of the CoVideo or APC video. Once participants were randomized to a video, they could not skip to the end or fast-forward without ending the study. However, it is possible in some cases that the participants could have been engaged in other activities while the video was playing. Because of potential noncompliance, we used an intention-to-treat analysis. One possible limitation is that high baseline knowledge likely reflects the high educational status of our online sample, with 81.6% having some college education or higher (bachelor’s/equivalent, master’s, PhD). Our sample was likely more educated than the general populations of the United States, the United Kingdom, Germany, Spain, and Mexico. A similar educational distribution has been reported in a recent web-based study on COVID-19 knowledge in the United States and United Kingdom [39].

Together, the findings of this study present innovative propositions for content design, dissemination, and evaluation of rapid global health communication interventions. Content designs that emphasize cultural accessibility, convey a compelling narrative, and elicit high-arousal emotions could fuel rapid dissemination across the 3.8 billion global citizens currently using social media. The wordless, animated approach also minimizes barriers traditionally associated with underlying differences in language and literacy levels. Given the massive global penetration of social media, short, animated, wordless video messages, designed to spread organically, may help public health authorities reach people where they are (ie, social media). Evaluating these interventions using online trials, APCs and list experiments can help expedite results and strengthen our efficacy evaluations. The value of such an approach becomes especially apparent during global crises in which lost weeks translate into lost lives. Accessible and compelling video health messages that lean on the shared characteristics of our global community could facilitate the spread of time-sensitive health messages. Public health authorities poised to implement these innovative health communication solutions could better support a global community facing unprecedented, shared challenges.