This study was a pilot study for a local NGO to identify potentially effective strategies for recruitment, implementation, and evaluation of online health communication campaigns using social media in a resource-limited setting. The study was registered retroactively at Clinicaltrials.gov: NCT07636629.

This pilot study examined the influence of peer role model and online influencer approaches in addressing contraceptive knowledge and related behaviors, building off prior research [38-40]. The web-based intervention targeted adolescents and young adults in Burkina Faso, Senegal, and Côte d’Ivoire with the goal of testing its feasibility, efficacy, and potential for future health campaigns by the NGO, RAES. We pre-tested the proposed interventions with selected members of the public who fit the study criteria during the content development phase.

All study procedures of this pilot study were approved by the University of California, Los Angeles Institutional Review Board (23‐000752). Prior to participation, participants were presented with a study information sheet online, which provided details on the study, including its focus on SRH. The sheet also emphasized that participants could choose not to take part in the study or could discontinue participation at any point in the study without repercussion. Participants were paid in the form of phone call credits amounting to about US $3‐$4. Study data were deidentified prior to analysis and were stored securely to ensure participant privacy and confidentiality. In addition, no identification of individual participants in any images in this manuscript and supplementary material is possible.

This study examines the change in SRH outcomes from baseline to end line for various social media user types, using an LCA. In total, 492 participants were enrolled, and 403 completed the pre- and/or postsurveys. Social media use patterns were collected only at baseline so that participants would not conflate their typical social media use at end line with the additional Facebook exposure they gained during the intervention. Therefore, this analysis focused on changes in outcomes from baseline to end line only for participants who filled out the survey that included questions on social media use at baseline (n=267).

The LCA used Full Information Maximum Likelihood, which is a robust method to account for missing data under the missing at random assumption. This approach excludes only those observations that have missing data on all observed latent class indicators [36,47,48]. Thus, the final analytical sample for this analysis included participants who had complete data for the demographic variables that were considered in this study and data for at least 1 of the 5 indicators used to estimate social media use latent class groups (n=262).

Data cleaning and demographic statistics were completed using Stata (version 18; StataCorp LLC) [49], while LCA as well as subsequent analyses was conducted using MPLUS (version 8.11; Muthén & Muthén) [50]. To address the first objective, we conducted an LCA to identify unobserved groups based on participants’ daily use of the following social media platforms: Facebook, Instagram, TikTok, WhatsApp, and YouTube. We estimated models with 2-5 latent classes and identified the best fitting model based on the following indicators: AIC (Akaike Information Criterion), BIC (Bayesian Information Criterion), SABIC (Sample-Size Adjusted Bayesian Information Criterion), Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (VLMRT), Lo-Mendell-Rubin Adjusted Likelihood Ratio Test (LMRT), and the Parametric Bootstrapped Likelihood Ratio Test (PBLRT). These are the most consistently recommended model fit criteria; entropy was also reported as is general practice [34-37]. Following standard practice, the optimal model fit was determined by the lowest values of AIC, BIC, and SABIC, and significant results for the likelihood ratio tests outlined previously (VLMRT, LMRT, and PBLRT). Statistically significant results for the likelihood ratio tests indicate that the current class solution is a better fit compared with the solution with 1 less class [34-37].

For the second objective, we examined whether sociodemographic characteristics predicted latent class group membership (ie, the social media use groups identified in the first objective). For this, we examined the bivariate odds of membership in one class versus the reference class for age, gender, country, and education. These odds were calculated using the 3-step estimation approach that involved identifying the latent class groups before examining their associations with predictors, adjusting for the relevant class estimation uncertainties [51]. We used the 3-step approach to allow for more consistent interpretations of class membership, given that we were examining different predictors. Odds ratios (ORs) and CIs are reported.

To address the third objective, we examined whether the effect of the peer and online influencer interventions, compared with the standard comparison intervention, differed based on one’s social media use latent class group membership using maximum likelihood estimation and Monte Carlo integrations. This involved subgroup analysis to identify whether there was a significant difference in intervention effects, compared with the comparison group, within each latent class group. We also tested whether the outcome of the interventions in one latent class group was significantly different when compared with intervention outcomes in other latent class groups. The intervention outcomes for this analysis were maximum likelihood estimates reflecting changes in knowledge, service provider awareness, and service use from baseline to end line. Missing data in these analyses were accounted for using Full Information Maximum Likelihood, which is a robust method to account for missing data under the missing at random assumption [36,47,48]. The conceptual model (Figure 1) shows how the 3 objectives for this study guide our approach and analysis.

Recruitment occurred from August to October 2023. After recruitment was completed, participants were randomized and then enrolled into the various intervention groups. Study implementation occurred between December 2023 and February 2024. This included enrollment, time for participants to fill out the baseline survey prior to the intervention, and the end line survey after the intervention. The intervention exposure itself was over a 2-week period in early to mid-February. The interventions were delivered as described in the “Methods” section. The intervention delivery followed a content schedule that the team followed so that each Facebook intervention group received respective content every day or every other day. The study did not end or stop unexpectedly and was implemented through completion. Participants did not receive concomitant care with this intervention. There were no harms measured or reported. In total, 353 participants were randomized to each intervention group and 492 participants enrolled to receive the intervention for the comparison (n=125), peer role model (n=122), online influencer (n=125), and the mixed approach (n=120). The final analytical sample (n=262) is based on relevant data availability as discussed in the “Data Analysis” section. The effects of the peer role model (n=70), online influencer (n=63), and the mixed peer and influencer intervention (n=57), compared with the comparison (n=72), were examined by social media user type for contraceptive knowledge, provider awareness, and service use. Figure 2 illustrates the sample sizes at each stage.

This study focused on the sample of participants who had complete data for the demographic variables considered in this analysis and for at least 1 social media platform (N=262; Table 1). Overall, 75.2% (197/262) of the sample were aged between 20 and 24 years while 24.8% (65/262) were in the age range of 15-19 years. Gender distribution for women and men was 52.7% (138/262) and 47.3% (124/262), respectively. Overall, 48.1% (126/262) of this sample resided in Senegal, 24.1% (63/262) lived in Burkina Faso, and 27.9% (73/262) lived in Côte d’Ivoire. In terms of education, close to 15.7% (41/262) of this analytic sample had secondary school or other education, 46.6% (122/262) had attained tertiary technical school (école supérieur) education, and 37.8% (99/262) had attained college or university education or higher. Out of these 262 participants, 89.7% (235/262) of the sample indicated using WhatsApp on a daily basis while 57.6% (151/262) reported using Facebook daily. In addition, 44.3% (116/262) indicated daily Instagram use, 35.1% (92/262) reported daily YouTube use, and 23.3% (61/262) shared that they used TikTok daily. Additional descriptive results, stratified by intervention group, are included in Table S1 in Multimedia Appendix 1.

For the second objective, results in Table 3 showed statistically significant bivariate associations of class membership with age group and with country. Specifically, being between 20 and 24 years, compared with being 15‐19 years, was associated with increased odds (OR 3.61, 95% CI 1.40‐9.36; P=.008) of being multiplatform users, compared with being selective users. In terms of country, residing in Burkina Faso (OR 0.23, 95% CI 0.07‐0.76; P=.02) and Côte d’Ivoire (OR 0.27, 95% CI 0.08‐0.88; P=.03), compared with residing in Senegal, was associated with 77% and 73% decreased odds, respectively, of being multiplatform users, compared with being selective users. There were no significant associations identified for gender and education level.

Findings in Table 4 show that among selective users, exposure to the peer role model approach, compared with the comparison exposure, was significantly associated with an increase in knowledge by 1.84 units (95% CI 0.27-3.41; P=.02), an increase in service provider awareness by 1.81 units (95% CI 1.15-2.47; P<.001), and an increase in service use by 1.49 units (95% CI 0.06-2.93; P=.04).

Results for multiplatform users showed contrary results. Specifically, exposure to the peer role model approach, relative to the comparison exposure, was not significantly associated with a change in knowledge (95% CI −0.33 to 1.10; P=.29) and in service use (95% CI −0.99 to 0.17; P<.17) from baseline to end line. Moreover, this exposure, relative to the comparison exposure, was significantly associated with a decrease in service provider awareness by 0.33 units (95% CI −0.64 to −0.01; P=.04) from baseline to endline.

We also compared whether intervention outcomes were significantly different between the 2 groups (selective users vs multiplatform users). Results indicated that the effect of the peer role model approach was significantly different for selective users versus multiplatform users for service provider awareness (P<.001) and service use (P=.02). There was no significant difference in the intervention effect when comparing change in knowledge outcomes between both social media use groups.

In addition, the online influencer condition as well as the mixed peer role model and online influencer approach, when compared with the comparison exposure, was not associated with significant changes in outcomes for both types of users but with one exception. Specifically, results showed that the mixed peer role model and influencer approach, relative to the comparison, was associated with increased odds of knowledge gain, although these results were marginally significant (OR 1.66, CI −0.01 to 3.32; P=.051). All analyses adjusted for age, gender, country, education, and the baseline values for each respective outcome.

This study sought to identify the underlying social media use patterns in a sample of French-speaking West African adolescent and young adult (aged 15‐24 years) social media users via an LCA. Results suggest that social media users in our sample can be categorized by 2 different types of daily social media engagement. These include selective users (daily use of WhatsApp and moderate Facebook use) and multiplatform users (daily use of all 5 platforms assessed, to varying degrees). We also examined the sociodemographic characteristics associated with social media use patterns and class membership. Our findings indicate that young adults aged 20‐24 years (compared with those aged 15‐19 years) are more likely to be multiplatform users, along with young adults residing in Senegal (compared with those in Côte d’Ivoire and Burkina Faso). Finally, we assessed whether social media use patterns influence the effect of a digital influencer intervention for SRH knowledge, awareness, and practices. Study findings show that the peer role model approach was associated with positive gains in knowledge, service provider awareness, and service use for selective users only. However, among multiplatform users, this exposure was not associated with significant changes for knowledge and service use. Furthermore, it was related to decreased gains for service provider awareness. Online influencer and mixed approaches did not show any effect for both groups in this sample for the most part although the mixed condition yielded some gains in knowledge for selective users. The implications of these findings are highlighted in this Discussion section.

Selective users are characterized by daily use of WhatsApp and moderate Facebook use primarily while multiplatform users reflect daily use of all platforms (Facebook, Instagram, TikTok, WhatsApp, and YouTube) to varying degrees. Drawing from the Uses and Gratifications theory [23], the 2-class solution could be a representation of 2 groups with different needs and motivations for social media use. Research suggests that instant messaging users leverage these platforms for maintaining relationships while Facebook users leverage this platform to stay updated on the activities of people who are generally in their network. Both involve some level of motivation to stay connected to peers and other networks [25]. Additional research reinforces WhatsApp as a desired platform for seeking to build or maintain close relationships but suggests that Facebook seems to be geared more toward information gathering about one’s social connections [58]. Moreover, an exploration of multiple social media platforms found that Facebook (and LinkedIn) were the only platforms where young adults were influenced by peers to join, highlighting that there are peer-related motivations for using Facebook as well [59]. Furthermore, in a study that examined multiple social media platforms, researchers found that Instagram and TikTok were more strongly associated with motivations such as entertainment, escapism, connecting with likeminded people, and content creation opportunities [60].

These findings suggest that instant messaging platforms such as WhatsApp may meet the needs of people seeking to maintain close relationships. While Facebook also leans toward social connections in one’s network, these networks may be more expansive beyond close peers [25,58,59]. Thus, it is possible that the latent class group consisting of selective users in our sample uses social media to meet needs related to maintaining close relationships. This may be why selective users in this analysis have a high probability of daily WhatsApp use and a moderate probability of Facebook use, with negligible probabilities for other platforms. On the other hand, based on findings from Karapanos et al [58], we can infer that multiplatform users may use several platforms as their needs extend beyond social connection to motives such as entertainment and escapism.

These findings have implications for digital health interventions particularly as research suggests that the media use needs and motivations of individuals may interplay with the impact of health messages, depending on the mode of communication and the media channels used [26,27]. Thus, paying attention to distinct social media patterns, and the needs and motivations they address, may ultimately facilitate better tailored health messaging interventions.

Demographic differences (age and country) were associated with class membership in bivariate analysis. Those who were aged 20‐24 years, compared with participants who were aged 15‐19 years, were more likely to be multiplatform users, compared with being selective users. These findings align with prior work suggesting that different age groups have distinct social media use needs and motivations [31,32]. For our sample of young adults, one distinction could be that these groups are in different phases of emerging adulthood. Those in the age range of 15‐19 years tend to be those who may still have heavy parental involvement and whose main social connections are in more controlled settings such as high school. However, for those who are in the age range of 20‐24 years, they may be at a point in adulthood where their social networks are expanding, as they become more responsible for themselves and make job and career choices [61-64]. Thus, for the older group, their needs and motivations may become more expansive beyond close networks, which would amend itself to seeking multiple platforms to meet these growing and expanding needs.

Moreover, in this region, internet use is shaped by access to monetary resources, and WhatsApp is particularly popular, given its amenability to areas with limited internet infrastructure, compared with other platforms [57]. Thus, adolescents (aged 15-19 years) in limited resource settings, who may have less financial autonomy for frequent use of other platforms, may lean more toward using WhatsApp primarily, compared with young adults (aged 20-24 years) who may have more financial autonomy to explore other platforms.

Findings also indicated that residing in Burkina Faso and Côte d’Ivoire, compared with residing in Senegal, was significantly associated with decreased odds of being a multiplatform user, compared with being a selective user. Data show that as of 2023, Senegal had the highest rates of internet use, compared with Côte d’Ivoire and Burkina Faso (17%) [65]. Thus, these results likely reflect differences in internet infrastructure and access, which suggests that a greater population of people in Côte d’Ivoire and Burkina Faso, compared with Senegal, could be more limited to primarily WhatsApp use.

Findings showed that exposure to peer role model approaches was associated with improved contraceptive knowledge, service awareness, and service use for selective users but not for multiplatform users. In line with the Uses and Gratifications theory [23], prior research suggests that preferences for building close relationships may motivate WhatsApp use [25,58]. Thus, selective users may respond more positively to the intervention that includes people they can more easily relate to and connect with on some level (even if they do not know them), particularly as these role models shared their personal stories and experiences.

The difference in the effects of the peer role model intervention for both users could also be related to how different users process the information they are exposed to. The Elaboration Likelihood Model of Persuasion posits that there are 2 paths that shape the persuasion process. The central route includes a more thorough and critical contemplation of messages, which allows for longer-lasting persuasion effects [66]. However, the peripheral route is less likely to elicit long-lasting effects as it involves less critical processing, lower cognitive requirements, and persuasion tends to be shaped by cues (eg, emotions leveraged or number of arguments presented) rather than a thoughtful contemplation of information [66]. This could be relevant to social media use patterns. Studies show that media and digital multitasking is associated with less focus and poorer learning and memory outcomes [67,68]. Perhaps, the distraction of media multitasking could make it harder for multiplatform users to fully process and understand the information they are exposed to, which could make certain health-messaging interventions less impactful for this group.

Thus, additional research should continue to explore strategies that can facilitate message processing for social media multiplatform users who may be less likely to engage in the central route of persuasion. For example, message repetition can allow for additional information processing but with threshold effects [66]. Overall, these findings emphasize that different messaging strategies should be used to target different types of social media users.

Our study did not detect significant changes in outcomes based on the influencer intervention for either type of user. This is contrary to findings that suggest that influencers shape health behavior positively or negatively [38-40]. Prior research has highlighted the importance of building audience trust for effective influencer and celebrity marketing approaches [69-71]. In addition, audiences consider the presence of paid marketing and endorsements in their assessment of influencer credibility [72,73]. Therefore, it is possible that participants did not trust some or all the influencers in this intervention for different reasons, which may have limited the potential of the intervention.

Furthermore, research suggests that people may form parasocial attachments to influencers they follow, which heightens their influence [74]. For this study, however, this was likely not the case since participants may not have felt connected to the group of influencers selected for the study. Perhaps an intervention that intentionally included influencers that participants considered as favorites would have yielded more promising results. Therefore, future research examining the potential of influencer marketing for public health should consider how to gauge participant perceptions of trust and emotional connection to selected influencers before proceeding with interventions.

Generally, the mixed peer role model and influencer approach did not yield significant results for this sample of social media users, although we noted marginally significant improvements in knowledge for selective users only. There are mixed findings on the efficacy of interventions with multiple components [75]. Compared with singular interventions, multifaceted approaches may involve multiple attention demands and cognitive overload, which limit their impact [76]. Thus, it is possible that exposure to both the peer and influencer interventions heightened cognitive burden, which diminished their potential benefits. This may be why the gains from the mixed approach were only marginally significant for knowledge among selective users only and not for any other outcomes and multiplatformusers. It may also be that the distrust of online influencers interfered with the perception of peer role models, given that participants in this group interacted with both strategies at the same time.

This study encountered some limitations. First, a majority of our sample resided in urban areas. Thus, this sample may not be entirely representative of the population of young adult social media users in French-speaking West Africa. Unfortunately, rural-urban internet service gaps remain in sub-Saharan Africa [77]; hence, persons who currently use social media platforms are more likely to be urban dwellers. Thus, these findings give us preliminary insight into how social media behavior can shape the impact of digital interventions among young adult social media users, which can be useful for future research as internet and social media use expand into rural regions. Furthermore, participant blinding is challenging to implement in web-based trials, considering that participants are aware of what they are viewing or interacting with [43]. While participants did not have specific knowledge of which exposures were the treatment versus the comparator, there may also have been social desirability present during survey completion, given their awareness of exposure to reproductive health content. It may also be possible that there are important social media platforms that were not considered in this analysis, as this study was a focused experiment on a specific target population in a more controlled research environment. This may limit the generalizability of findings to the general population of social media users.

Finally, our study was not without its recruitment and retention challenges, given the nature of research in resource-limited settings [41,42]. Therefore, our sample size was 267 at baseline (when social media use data were collected) and 262 for this analysis. Smaller sample sizes present the potential for less precise estimates [78], as is evident by some of the confidence intervals reported in these analyses, particularly for the demographic associations. Moreover, the confidence intervals characterizing the association between demographic characteristics and latent class membership are particularly wide because they reflect uncertainty in both latent class membership and the strength of the association. Thus, we emphasize the need for robust logistical and funding support for additional studies of this kind in resource-limited settings. This will be increasingly pertinent as internet and social media use expand among young adults in low- and middle-income regions [10]. Given our limited sample size, we present these results, interpret them with caution, present them as preliminary findings for future studies, and call for future research with more expansive and representative samples.

Our study presents some notable strengths. First, pre- and postsurvey data collection allowed for a clear temporal order in our analysis of change in outcomes. In addition, using the LCA approach provided a more comprehensive picture of social media use in the sample as opposed to examining independent platforms separately. Finally, working with partners on the ground through survey and intervention development, as well as data collection, allowed for our study to be suitable and relevant for our population of interest.

Overall, this study offers insight into how general social media behavior and purposeful digital interventions interact to determine SRH outcomes in French-speaking West Africa. It moves the literature forward in an understudied yet important area of research in Francophone West Africa. Furthermore, it applies a novel approach to examining how a comprehensive measure of social media use moderates digital health interventions by using an LCA approach, which is underutilized for research in this population. Results show that a one-size-fits-all approach in web-based interventions is not ideal, and online peer-to-peer communication is still an important strategy for certain social media users but not for others. Findings are important for identifying tailored and impactful interventions that consider both demographic contexts and media exposure patterns. The findings also suggest that multiplatform social media use could be detrimental for the impact of health-messaging interventions. This has implications for future public health work as social media platforms continue to maintain popularity and proliferate. It aligns with work that indicates that increasing media saturation may pose a threat to effective health communication as people experience information fatigue and messaging overload [29,30]. This should incentivize health communication researchers to explore what strategies will cut through social media noise, especially for individuals who use multiple social media platforms on a frequent basis.

Given that these results serve as preliminary findings in an understudied research area in French-speaking West Africa, future studies should explore this area of research with larger and more representative samples to ensure consistency of results. Studies of this kind should also consider repeated measurement of outcomes over a longer period after the intervention to examine whether these strategies result in sustained health literacy and behavior changes.