Social media posts published from May 27, 2014 (the date of the first PHN) to December 31, 2022 that explicitly mentioned multi-jurisdictional enteric illness outbreaks for which a PHN had been published were included. Posts were collected from official Canadian federal public health accounts on Facebook (Healthy Canadians, Canadiens en santé) and X (Health Canada and PHAC, Santé Canada et l’ASPC) in English and French.

Social media posts were manually reviewed and collected by reviewing all account posts to identify those that met inclusion criteria. No specific keywords or search terms were used in the collection process. A unique identifier was created for each eligible social media post and the following metadata was recorded: link to the social media post, pathogen, publication date, and date of collection. Data was collected from May 10–18, 2023 by the first author and managed using Microsoft Excel (Microsoft Corporation, 2018).

An engagement rate was calculated for each social media post using published platform-specific formulas (Sehl and Mikolajczyk, 2023). Engagement metrics included number of reactions, comments, and shares on Facebook, and number of retweets and comments on X. For all calculations, the total number of followers at the time of data analysis was used (November 2023). The engagement rate per post (ER post) was calculated by adding the number of engagements divided by the number of followers. The overall average engagement rate per platform and language was subsequently calculated by dividing the total number of engagements by the number of social media posts. Descriptive time trends in engagement rates over the study period (2015–2022) were assessed by examining differences in annual engagement rates.

A deductive, iterative content analysis of all social media posts was conducted following established methodology (Hsieh and Shannon, 2005). French social media posts that were direct translations of English social media posts were excluded to avoid redundancy in the content analysis. A codebook was developed by VP to guide the analysis, and included engagement metrics, CCI criteria, HBM constructs, other health communication best practices identified in the literature, and measures of consistency (Supplementary Table A1).

The modified CCI for short form communication materials includes a total of 13 items in four parts (Supplementary Table A2). Part A applies to all communication materials and includes the main message and call to action and language. Parts B, C, and D selectively apply and include behavioural recommendations, numerical information, and risk, respectively. For Facebook, 11 of the 13 CCI items were used, as written. Two criteria were excluded because no Facebook posts required mathematical equations or used numeric probability to describe risk. For X, 7 of 13 items were used, as written, while six items were excluded due to the lack of mathematical equations and the non-applicability of certain criteria in the context of short messages.

Similar to other studies, HBM constructs were operationalized to identify explicit features of the text that aligned with these constructs (Meadows et al., 2019; MacKay et al., 2022b; Bajouk and Ferré-Pavia, 2023) (Supplementary Table A3). For example, perceived susceptibility was coded as present if the post mentioned the number of cases.

Other health communication best practices were also included, specifically the use of hashtags (Lin et al., 2016), mentions (Zappavigna and Martin, 2018), visuals (Chou et al., 2021), tone (Maal and Wilson-North, 2019), intended audience information (Korda and Itani, 2013), interaction (Lin et al., 2016), and combatting misinformation (Lin et al., 2016) (Supplementary Table A1).

Consistency was measured at a high level through the inclusion and repetition of key information from the PHN (e.g., source, pathogen, and geographical distribution) and prevention information (e.g., link to additional information, behavioural recommendations) (Supplementary Table A1).

Two researchers (VP, HS) independently coded a ~ 10% simple random sample of 50 Facebook and X posts using the codebook. Inter-rater reliability was calculated in R using the ‘irr’ package (Gamer et al., 2010; Posit team, 2023) and met the a priori cut-off of 0.8. A kappa value (κ) of 0.92 was achieved indicating almost perfect agreement (McHugh, 2012). Minor disagreements were resolved through discussion, leading to no major codebook revisions. The codebook was applied to the remaining 432 posts by one researcher (VP). All coding was done using Microsoft Excel (Microsoft Corporation, 2018).

Descriptive statistics, including mean, standard deviation, and minimum and maximum values, were computed for continuous variables. Frequencies and proportional frequencies were computed for categorical variables. Data were tabulated and visualised using RStudio version 2023. 6. 1. 524, using base packages (Posit team, 2023), dplyr (Wickham et al., 2023), ggplot2 (Wickham, 2016), and stringr (Wickham, 2023). Analyses were stratified by language, platform, and both language and platform.

As per the University of Guelph’s Research Ethics Board, ethics approval was not required as this study used publicly available data.

English social media posts (n = 250) had an average engagement rate of 0.36 ± 1.32% (0.01–12.45%) on Facebook and 0.01 ± 0.02% (0.00–0.21%) on X (Table 1). French social media posts (n = 232) had similar engagement with an average engagement rate of 0.37 ± 0.89% (0.00–7.02%) on Facebook and 0.01 ± 0.01% (0.00–0.05%) on X (Table 1).

Facebook and X average engagement rates varied across time and by language (Figure 1). On Facebook, engagement rates peaked in 2016 for French-language posts (ER post = 2.37%) and 2021 for English-language posts (ER post = 1.21%). On X, engagement rates were highest in 2016 and 2020 for English-language posts. Contrastingly, French-language posts demonstrated a consistent engagement rate of ~0.01% over the study period.

Facebook posts frequently used cues to action (95.9%) and benefits constructs (91.8%). However, posts less often employed the severity (14.4%) and barrier constructs (1.0%). On average, Facebook posts most often incorporated three of the six HBM constructs (40.2%) per post. No Facebook posts employed all six constructs in a single post. Likewise, X posts also used cues to action (95.4%) and benefits constructs (87.6%) with infrequent use of severity (2.0%) and barriers constructs (0.0%). X posts typically included two HBM constructs (45.1%) per post, and like Facebook, no X posts contained all six constructs in one post (Table 3).

Facebook posts frequently incorporated visuals in 94.8% of posts, primarily using still photographs (43.5%). Hashtags were rarely used on Facebook (10.3%), but when they were, posts adhered to the best practice of including a maximum of two or less hashtags per post (100%). Mentions of other organisations were infrequent (2.1%). Posts containing messaging to combat misinformation were never used (0.0%) (Table 4).

Likewise, X posts included visuals in 97.4% of posts, with still photographs most often used (53.0%). Hashtags were commonly used on X (70.6%), and like Facebook, a maximum of two or less hashtags per post was usually adhered to (94.4%). Mentions of other organisations were slightly more common on X (7.2%), aimed at increasing engagement with other governmental bodies like InspectionCan (54.5%), CFIA_Food (36.4%), and CDCgov (9.1%). X posts did not contain messaging to combat misinformation (0.0%) (Table 5).

Posts on both platforms consistently provided details on outbreak investigations, including the source (99.0% on Facebook, 100% on X) and the pathogen (97.9% on Facebook, 100% on X). All Facebook posts and most X posts (98.7%) included a link to additional information for preventing illness. However, providing geographical distribution information was less consistent, with 52.6% of Facebook posts and 17.6% of X posts including this information (Table 6).

Engagement rates on social media platforms can be used to gauge public awareness of posts. Rates between 1 and 5% are indicative of good engagement (Sehl and Mikolajczyk, 2023). At the end of 2023, average engagement rates on Government of Canada accounts were 1.62% on Facebook and 1.36% on X (Mikolajczyk, 2023). However, our study revealed average engagement rates below 1% across all accounts and platforms for multi-jurisdictional enteric illness outbreaks with relatively higher engagement on Facebook. Facebook is more commonly adopted across all demographics (Krallman et al., 2016; Pew Research Center, 2021), as it traditionally serves as a platform for social interactions (Voorveld et al., 2018). In contrast, X is considered more news-oriented (Kwak et al., 2010) and tends to garner less engagement than Facebook (Reuter et al., 2021). This trend, where Facebook demonstrates higher engagement rates than X, aligns with our findings (Bhattacharya et al., 2017; Haro-de-Rosario et al., 2018; Reuter et al., 2021; DePaula et al., 2022).

Lack of engagement, as evidenced by rates below industry standards, may reflect an overall inattentive and unresponsive audience (Guidry et al., 2017), hindering the awareness and amplification of posts (Bajouk and Ferré-Pavia, 2023). Engagement metrics, such as share and like functions on social media posts, affect the perception of the post and other user’s behavioural intentions (Kim, 2018). These functions can serve as a cue, triggering the bandwagon heuristic which leads people to accept others’ beliefs, attitudes, or decisions, as their own (Sundar, 2008). Negative engagement can cue others to view the content in the same light. Therefore, the lower engagement rates observed on PHAC’s social media posts may cause others to view PHAC content as uninteresting or irrelevant. Additionally, lower engagement rates may impact how platform algorithms rank and display these posts in followers’ feeds (Meta, 2024; Twitter, 2023), potentially reducing visibility and further affecting engagement.

Social and mixed media features, almost always employed on both platforms, can increase engagement (Rus and Cameron, 2016; Moran et al., 2019; Li and Xie, 2020; Cao et al., 2021). The use of pictures in communication materials can enhance comprehension and adherence, especially among individuals with low literacy skills (Houts et al., 2006). Moreover, the use of a dialogue loop, initiated by hashtags and questions, can promote two-way communication (Men et al., 2018). These features can increase engagement by encouraging the public to voice their opinions through social media functions, such as sharing, liking, and commenting (Watkins, 2017). To complete the dialogue loop, organisations should take the opportunity to respond to questions, comments, and concerns (Kent and Taylor, 1998). Utilising dialogue loops enhances engagement and supports a public-centred approach, fostering trust between the public and the posting organisation (Kent and Taylor, 2002). In the future, social media posts should consider the use of social and mixed media features to improve engagement, comprehension, and adherence.

An analysis of the posts, focusing on their adherence to the CCI, revealed that posts effectively communicated the main message, behavioural recommendations, and numerical information. However, the use of plain language and the communication of risk could be improved. A similar study examining government health messaging in Canada found inconsistent application of the CCI, possibly impacting the clarity and comprehension of communication materials (Kelly et al., 2023).

Recognising the importance of clear health communication language, often termed plain language, the Government of Canada introduced the Directive on the Management of Communications. This directive emphasises the need for plain language in all public communications, defining it as writing that is easily understandable upon first reading (Government of Canada, 2021). Our analysis revealed that posts contained few familiar words that the public could easily comprehend. Unclear language can hinder public understanding and ability to adopt health behaviours (Zarcadoolas, 2010). Communicating about enteric illnesses, which often requires complex scientific terminology, is further complicated by the character limits imposed by social media platforms. These constraints make it challenging to streamline the content effectively, which can further impede understanding (Zarcadoolas, 2010). While plain language is crucial for improving health information comprehension (Grene et al., 2017), it is also important to recognise that some scientific terminology may be unavoidable, for example, pathogen names. Therefore, it is recommended to minimise the use of complex terms as much as possible while preserving accuracy, rather than solely relying on plain language. Further, social media is a common source of health information for older adults (Dooley et al., 2012; Tennant et al., 2015; Zhao et al., 2022). However, older adults, who have lower health literacy on average (Rootman and Gordon-El-Bihbety, 2008), are of particular concern as they are also at higher risk for enteric illnesses (Barkley et al., 2016). Therefore, ensuring information shared on social media is written in plain language is important to help older adults both understand and act on the content.

Effectively crafting posts during outbreak responses entails curating content that resonates with affected populations, including clear identification and explanation of public health risks (Tumpey et al., 2019). While social media posts often identified the cause of harm, crucial information enabling individuals to assess the risk’s relevance, such as associated symptoms, was seldom provided. Without this information, affected populations may not understand the significance of the risk and how it relates to them, potentially hindering adherence to recommended health behaviours (Tumpey et al., 2019).

According to the extended parallel processing model (EPPM), a communication theory, the susceptibility and severity constructs constitute a threat that needs to be elicited for individuals to respond to fear and engage in preventive behaviours (Witte, 1992). The posts inconsistently included the susceptibility construct and seldom the severity construct, hindering the ability of individuals to properly appraise the threat and take preventive actions. A prior meta-analysis of the HBM constructs’ effectiveness indicated that benefit and barrier constructs are more predictive of behaviour change compared to severity and susceptibility constructs (Carpenter, 2010). While benefit constructs were almost always present, barrier constructs were rarely present in posts. This is important given that barrier constructs are influential in motivating behaviour change.

Nonetheless, the presence of all constructs is important: perceived susceptibility, perceived severity, perceived barriers, and perceived benefits to maximally influence health behaviours (Sulat et al., 2018). As such, social media posts should aim to incorporate all relevant HBM constructs to optimise the potential to influence behaviour change. For instance, posts could include the number of cases (susceptibility) and, if applicable, hospitalisations and deaths (severity) to influence risk perception and support accurate threat appraisal. Additionally, when providing behavioural recommendations, posts should offer stepwise and detailed instructions to enhance self-efficacy, helping individuals take preventive actions to avoid or reduce the threat. Briefly outlining the benefits (prevention of illness) and overcoming barriers (ease of task) can enhance an individual’s confidence that completing the action will have a desirable effect and encourage individuals to adopt preventive actions to avoid or reduce the threat (Witte, 1992). Due to character limits imposed on social media, integrating the full HBM in a single post may not be feasible. Instead, multiple posts, such as a thread on X, can be used to convey information. Additionally, using visuals with text overlays or videos in combination with the captions can convey more information, while adhering to character limits.

During public health threats like enteric illness outbreaks, individuals intensify their reliance on trusted sources and perceived expertise for emotional release during information seeking (Ball-Rokeach, 1985). People seek information from the media and turn to organisations they trust (Holroyd et al., 2020). Public health organisations, such as PHAC, were historically viewed as trusted sources of health information (Lachlan et al., 2014). However, the COVID-19 pandemic has been associated with growing concerns about trust in the federal government (Herati et al., 2023). This growing scepticism may affect how people perceive and value information from PHAC, potentially impacting PHAC’s communication effectiveness during enteric illness outbreaks.

For government communication, consistent posts are crucial to establishing trust (MacKay et al., 2022a,b; Seeger et al., 2018; Wray et al., 2008). The consistent messaging demonstrated by PHAC, exemplified by including outbreak investigation details and prevention information in each post, helps individuals sift through information, distinguishing between authentic and misleading content (Wajeeha et al., 2020), and builds trust in the information provided (Vereen et al., 2020). Conversely, inconsistent messaging creates confusion and hampers the public’s ability to take preventive actions (Seeger, 2006). Trust in government organisations as key information providers is important for risk perception and subsequent adoption of preventive behaviours (Dohle et al., 2020; Lim et al., 2021; Jeong and Kim, 2023). Maintaining consistency in communication aids in building public trust and ensuring effective information dissemination during enteric illness outbreaks.

Our study has several limitations. First, the study assumed that the follower count recorded in November 2023 was constant over the study period when calculating engagement rate metrics. This assumption may potentially impact the accuracy of engagement rate calculations. If followership increased over time, the engagement rate in earlier time periods would have been underestimated, potentially leading to an inaccurate assessment of engagement trends. Second, the application of CCI criteria was scored as absent or present but did not measure the effectiveness of each scored item. This means that recommendations coded as present might not be actionable, which could have overestimated how comprehensible the post was by the public. Third, the HBM was operationalized to identify explicit features that may motivate an individual to undergo a behaviour change; however, this study could not ascertain whether an individual would change their behaviours. Fourth, consistency was measured broadly to assess post patterns for key information surrounding outbreak and prevention information. However, this approach did not account for variation in the types of information present, which can mask inconsistencies.

Future research should explore the underlying causes of engagement rate variability over time, including assessing post characteristics to understand which features result in higher engagement to inform effective communication strategies. Further, the effectiveness to which the CCI was applied, beyond presence/absence, should be explored. While our study employed tools to assess the potential impact of posts on comprehension, other data collection methods, such as focus groups or surveys, are needed to directly assess comprehension. These insights can offer valuable perspectives on public comprehension of social media posts during multi-jurisdictional enteric illness outbreaks. Focus groups or surveys could also include open-ended questions to explore reasons for low engagement. Additionally, exploring consistency of content within and between posts, beyond merely assessing for the presence of key types of information, can help provide insights into content variations which may impact public trust. Furthermore, future studies could benefit from international comparisons to understand how communication strategies vary across different public health contexts.