Twitter Communication During an Outbreak of Hepatitis A in San Diego, 2016–2018

Abstract

Objectives. To examine how and what information is communicated via social media during an infectious disease outbreak.

Methods. In the context of the 2016 through 2018 hepatitis A outbreak in San Diego County, California, we used a grounded theory–based thematic analysis that employed qualitative and quantitative approaches to uncover themes in a sample of public tweets (n = 744) from Twitter, a primary platform used by key stakeholders to communicate to the public during the outbreak.

Results. Tweets contained both general and hepatitis A–specific information related to the outbreak, restatements of policy and comments critical of government responses to the outbreak, information with the potential to shape risk perceptions, and expressions of concern regarding individuals experiencing homelessness and their role in spreading hepatitis A. We also identified misinformation and common channels of content driving themes that emerged in our sample.

Conclusions. Public health professionals may identify real-time public risk perceptions and concerns via social media during an outbreak and target responses that fulfill the informational needs of those who seek direction and reassurance during times of uncertainty.

The hepatitis A virus (HAV) is a fecal–orally transmitted virus¹ spread primarily through close personal contact with an HAV-infected person and occasionally through consumption of contaminated food and water. Although its incidence has decreased dramatically since the introduction of the hepatitis A vaccine,¹ for reasons that are unclear, since late summer 2016 HAV has had an increased presence across the United States. In particular, a large outbreak of hepatitis A in San Diego, California, was officially declared a local public health emergency by the San Diego County public health officer on September 1, 2017. The outbreak was notable in both its severity and its rapid spread, with almost 600 cases, 20 deaths, and 400 hospitalizations in San Diego County through early 2018. In addition, since the outbreak, numerous other states reported cases of hepatitis A, some of which may be linked to the initial outbreak in San Diego.² An audit by the state later noted that San Diego County’s response to the HAV was greatly delayed.³

During situations with a sudden increase in caseload or transmission, a coordinated public health strategy can facilitate communication about risks and appropriate behavioral responses.⁴ Creating and implementing a coordinated public health strategy requires effective communication deployed in a rapid manner.⁵ Social media offers a way to provide the public with emergency alerts and real-time updates regarding emergencies.⁶ Effective communication also requires that public health professionals deftly handle the rising problem of health-related misinformation on social media.⁷ As questions about how to monitor and assess misinformation during an infectious disease outbreak have emerged among public health professionals, an increased understanding of the context in which health-related misinformation spreads via social media may serve as a starting point for addressing these questions.

At the same time that misinformation on social media has begun to emerge as a possible threat to public health, the United States has experienced several significant infectious outbreaks, including the current COVID-19 pandemic. However, despite the increasing interest in exploring the use of social media message platforms for early disease prediction⁸ and message content during environmental emergencies,⁹ research on content and communication during an outbreak spurred by a vaccine-preventable communicable disease outbreak remains limited.¹⁰ It is also unclear whether and how real-time patterns of health information diffused via Twitter during an outbreak of a vaccine-preventable infectious disease may differ from other outbreaks or how communication varies if the outbreak originated in a stigmatized population.

We examined both how and what was communicated (and miscommunicated) via social media during the 2016 through 2018 hepatitis A outbreak in San Diego. One example of a misinformation effect on public health is the antivaccine misinformation in social media posts describing childhood immunizations as a cause of autism and the subsequent reductions in herd immunity.¹¹ We address the following research questions (RQs):

• RQ1: How did Twitter users communicate about the hepatitis A outbreak?

• RQ2: What were the source types (type of individual or organization) and characteristics of Twitter users contributing to the conversation about the hepatitis A outbreak?

• RQ3: What themes about the hepatitis A outbreak on Twitter were more likely to be shared and by which source types?

• RQ4: Was there misalignment between messaging about the hepatitis A outbreak on Twitter from authorities and members of the public?

• RQ5: Was misinformation shared on Twitter during the hepatitis A outbreak?

• RQ6: In what ways did users seek or share content about HAV on Twitter?

METHODS

We used the hepatitis A outbreak in San Diego (November 2016 to October 2018) and a grounded theory–based thematic analysis to motivate our research. We first examined qualitative data to identify themes of tweets shared during the outbreak. This approach was guided by principles of grounded theory methods.¹² Our primary data were publicly available messages posted to Twitter by official response agencies (e.g., the public health department) and other numerous stakeholders during October 2017 through December 2017—the height of the outbreak.

Data Collection

We used the visualizing information space in ontological networks framework¹³ to examine the interrelationships between online messages, space, and time. The framework consists of an approach our team developed for visualizing and analyzing Web pages and social media content from a spatiotemporal perspective. We focused exclusively on data collected from Twitter because of the real-time and dynamic nature of this platform. We took advantage of the spatial filtering methods provided by the Twitter search application programming interface.

We collected 4401 tweets limited to San Diego County between October and December 2017, coinciding with Governor Jerry Brown’s declaration of a statewide emergency and the downward epidemiological curve of new cases.¹⁴ We used user profile information with place name dictionaries (gazetteers; Appendix A, available as a supplement to the online version of this article at http://www.ajph.org) to determine that all tweets originated in San Diego County. Of these tweets, we randomly selected 1000 tweets for analysis. After identifying the number of unique users, we randomly de-duplicated users so they each had only 1 tweet. We decided to randomly select 1 tweet per user to ensure that we did not violate assumptions of independence in observations for our tests of association. We further constrained the sample by removing tweets that (1) were about flu only, (2) were about vaccines for animals, (3) discussed other forms of hepatitis, or (4) were written in a language other than English. This resulted in a final analytical sample of 744 tweets, which represented 17% of all tweets collected; the sample was composed of tweets from unique users, meaning each user appeared in our sample only once.

Data Analytics

We used retrieved tweets and metadata to tabulate information such as the impressions and engagements of each tweet and to establish the frequencies of hashtag use. We also searched metadata fields (e.g., Twitter @handle, display name, number of followers) for relevant search terms. Two researchers (R. E. H. and T. A.) manually coded the message content of the original tweets from the entire data set to identify relevant themes. Using grounded theory, we performed a thematic analysis to identify themes of tweets in our sample. We additionally coded each targeted account with an included tweet in the sample according to its network size (number of followers) and number of accounts followed by the account.

We first analyzed and coded tweets using the constant comparative method.¹⁵ Using this approach, we focused on patterns of conceptual and exemplar convergence and divergence in the data. The identification of themes occurred through revisits and recodes of tweets to ensure that themes were substantiated by the data. When reasonable, we reconsidered and revised coding categories in cases that departed from identified themes.¹⁶ Over the course of revising thematic categories, we developed a coding scheme for 2 coders (who were not privy to the research questions driving the study) to use to quantitatively examine tweets in our sample. Coders also quantified the number of times a theme was observed (RQ1) and calculated intercoder reliability between both coders for the frequency of themes identified in our sample. Using coding guidelines from previous research,¹⁷ coders also coded tweets for source type (RQ2), which we then used to link to themes (RQ3). Building on patterns of misinformation and misalignment between public health messaging and individuals and organizations who were not public health authorities, we developed coding categories for content inconsistent with known public health messaging about hepatitis A vaccinations and contamination or sanitation issues (RQ4) and misinformation (RQ5).

We developed coding procedures for misinformation related to vaccines with guidance from others.¹⁸ For example, if a tweet promoted a vaccine-related conspiracy theory (e.g., vaccine policymakers are influenced by profit motives) or trivialized vaccine-preventable diseases (e.g., vaccines are worse than the measles), we coded it as containing misinformation. If a tweet contained any misinformation (related or unrelated to the hepatitis A outbreak), we coded it as having misinformation. To qualify as having no misinformation, a tweet could contain only accurate content. Last, given the unique properties of social media, we developed coding categories to capture the frequency with which users engaged in seeking and sharing information (i.e., in the form of news, statistics, known facts) and opinions (RQ6). Our codebook (Appendix B, available as a supplement to the online version of this article at http://www.ajph.org) contains more information about coding approaches. We examined contingency tables using χ² analysis and magnitude of effect with the φ (ɸ) coefficient.

Coders were trained on a training set of tweets and double-coded 20% of the total number of tweets. In instances of coding disagreements, coders discussed discrepancies until resolved. Good intercoder reliability (at least 0.80) was achieved after 3 rounds of double-coding, after which remaining tweets were single-coded. We used Gwet’s agreement coefficient,¹⁹ as it may address limitations observed in the use of other more commonly used coder reliability statistics, including Cohen’s κ and Scott’s π, which are more sensitive to prevalence.^20,21

We conducted the statistical analyses using Microsoft Excel (Microsoft Corp., Redmond, WA) and SAS version 9.4 (SAS Institute, Cary, NC).

RESULTS

We report the findings of tweets in our sample (n = 744) that were used for our thematic analysis. The median number of followers for these users was 480.5, and the median number of users they were following was 453.5.

Research Question 1

RQ1 was, How did Twitter users communicate about the hepatitis A outbreak? To answer this question, we identified major themes in our sample of Twitter users (any publicly viewable accounts that could be organizations, excluding accounts of a single person) or individuals (excluding accounts of an organization or cause). Among the 744 messages, we found that a subset related to policy issues (n = 281) and questions regarding general medical information (n = 177). These were followed by themes related to risk perceptions of hepatitis A (n = 111), concerns regarding individuals experiencing homelessness and their role in spreading hepatitis A (n = 151), specific hepatitis A medical concerns (n = 90), and other (n = 137; Table 1). We did not code themes as mutually exclusive, and intercoder reliability was strong, with an average Gwet’s agreement coefficient of 0.87, ranging from 0.80 to 0.96.

TABLE 1—

Message Examples by Content Theme: San Diego County, CA; October–December 2017

Theme	Frequency (%)	Example Tweets	Gwet’s Agreement Coefficient
Policy/government	281 (37.8)	California Declares State of Emergency Over Hepatitis Crisis. https://t.co/1XgJQHXmjn. https://t.co/LP1UF4koGE	0.84
Medical information specific to hepatitis A	90 (12.1)	5 things you need to know about Hep A from an infectious disease specialist affiliated w/ Sharp Grossmont Hospital: . . . https://t.co/aQclhqK4gZ	0.85
General medical information	177 (23.8)	W.H.O. Approves a Safe, Inexpensive Typhoid Vaccine\ by DONALD G. McNEIL Jr. via NYT. https://t.co/vfJOKVO22g	0.84
Risk perception of hepatitis A outbreak	111 (14.9)	18 Dead In San Diego County Hepatitis A Outbreak. https://t.co/IDHPufh01l. https://t.co/Qg7sdeuDS5	0.96
Homeless	151 (20.3)	Hepatitis A outbreak among homeless a byproduct of California’s housing crunch. https://t.co/UuQpQMgXgx	0.91
Other	137 (18.4)	That’s like asking if you’d like to have herpes or Hep A, there really isn’t any good choice for a normal life	0.80

Note. Population size was n = 744.

Research Question 2

RQ2 was, What were the source types and characteristics of Twitter users contributing to the conversation about the hepatitis A outbreak? Table 2 presents results showing the source types in our sample. The majority of tweets in our sample were shared by individuals (77.2%), and the most common categories for individuals included users who mentioned being a parent (9.1%) and a journalist (11.0%). A portion of tweets shared by individuals expressed firsthand experience of the hepatitis A outbreak (15.7%) and pointed to the spirituality (8.2%) and political persuasion (25.8%) of the user. The most common types of organizations contributing to the conversation about the hepatitis A outbreak were businesses (65.8%), news organizations (37.7%), and nonhealth advocacy groups (18.2%). We noticed, however, that only a small portion of tweets (1%) was shared by government organizations.

TABLE 2—

Source Types for Twitter Content: San Diego County, CA; October–December 2017

Category	Frequency (%)	Gwet’s Agreement Coefficient
Celebrity	40 (5.40)	0.98
Organization	170 (22.80)	0.95
Business	112 (15.10)	0.88
Health	12 (1.60)	0.98
Government	7 (0.90)	0.96
News	64 (8.60)	0.99
Education	5 (0.70)	0.99
Information	11 (1.50)	0.99
Health care	11 (1.50)	0.98
Nonhealth	31 (4.20)	0.98
Individual	574 (77.20)	0.97
Parent	52 (7.00)	0.96
Child	2 (0.30)	0.98
Student	16 (2.20)	0.98
First person	90 (12.10)	0.93
Spiritual	47 (6.30)	0.98
Political	148 (19.90)	0.90
Journalist	63 (8.50)	0.96
Doctor	16 (2.20)	0.99
Practitioner	20 (2.70)	0.98

Note. The population size was n = 744.

Research Question 3

RQ3 was, What themes about the hepatitis A outbreak on Twitter were more likely to be shared and by which source types? Organizations were more likely to tweet themes related to government and policy (ɸ = 0.09), particularly news affiliates (ɸ = 0.24) and content with general medical information, when compared with individuals (ɸ = 0.10; Table 3). By contrast, individuals were more likely than organizations to share tweets with themes centering on the population experiencing homelessness (ɸ = 0.09). Individuals with political affiliations were most likely to tweet about policy (ɸ = 0.15) and homelessness (ɸ = 0.18). Journalists were most likely to discuss risk perceptions (ɸ = 0.14).

TABLE 3—

Associations Between Theme and Source Type: San Diego County, CA; October–December 2017

	Policy/Government (n = 281)			Medical Information Specific to Hepatitis A (n = 90)			General Medical Information (n = 177)			Risk Perception of Hepatitis A Outbreak (n = 111)			Homeless (n = 151)			Other (n = 137)
Source Type	No. (%)	P	φ	No. (%)	P	φ	No. (%)	P	φ	No. (%)	P	φ	No. (%)	P	φ	No. (%)	P	φ
Individual (n = 574)a	203 (27.3)	.02	–0.09	59 (7.9)	.01	–0.10	126 (18.3)	.92	< −0.01	78 (10.5)	.07	–0.07	128 (17.2)	.01	0.09	121 (16.3)	.01	0.13
Celebrity (n = 40)	26 (3.5)	.01	0.13	5 (0.7)	.99	< 0.01	1 (0.1)	.01	–0.12	16 (2.2)	.01	0.17	11 (1.5)	.23	0.04	2 (0.3)	.02	–0.08
Organization
Business (n = 112)	55 (7.4)	.01	0.10	13 (1.8)	.99	< −0.01	27 (3.6)	.99	< 0.01	24 (3.2)	.05	0.08	14 (1.9)	.03	–0.08	9 (1.2)	.01	–0.11
Health (n = 12)	4 (0.5)	.99	–0.01	12 (1.6)	.05	0.08	7 (0.9)	.01	0.10	2 (0.3)	.70	< 0.01	1 (0.1)	.48	–0.04	1 (0.1)	.71	–0.03
Government (n = 7)	5 (0.7)	.11	0.07	4 (0.5)	.01	0.13	0 (0.0)	.21	–0.05	0 (0.0)	.60	–0.04	2 (0.3)	.64	0.02	0 (0.0)	.21	–0.05
News (n = 64)	48 (6.5)	.01	0.24	6 (0.8)	.69	–0.03	2 (0.3)	.01	–0.15	18 (2.4)	.01	0.11	8 (1.1)	.14	–0.06	3 (0.4)	.01	–0.11
Education (n = 5)	0 (0.0)	.16	–0.06	3 (0.4)	.01	0.12	2 (0.3)	.34	0.03	0 (0.0)	.99	–0.03	0 (0.0)	.59	–0.04	0 (0.0)	.59	–0.04
Information (n = 11)	3 (0.4)	.55	–0.03	2 (0.3)	.63	0.02	7 (0.9)	.01	0.11	2 (0.3)	.67	0.01	1 (0.1)	.70	–0.03	1 (0.1)	.70	–0.03
Health care (n = 11)	0 (0.0)	.01	–0.10	4 (0.5)	.03	0.09	5 (0.7)	.14	0.06	1 (0.1)	.99	–0.02	1 (0.1)	.70	–0.03	1 (0.1)	.70	–0.03
Nonhealth (n = 31)	12 (1.6)	.99	< 0.01	9 (1.2)	.01	0.11	5 (0.7)	.39	–0.04	5 (0.7)	.80	0.01	7 (0.9)	.82	0.01	5 (0.7)	.99	–0.01
Individual
Parent (n = 52)	20 (2.7)	.99	< 0.01	7 (0.9)	.67	0.01	14 (1.9)	.61	0.02	5 (0.7)	.32	–0.04	13 (1.8)	.37	0.03	9 (1.2)	.99	< −0.01
Child (n = 2)	0 (0.0)	.53	–0.04	1 (0.1)	.23	0.06	0 (0.0)	.99	–0.03	0 (0.0)	.99	–0.02	0 (0.0)	.99	–0.03	1 (0.1)	.34	0.04
Student (n = 16)	4 (0.5)	.44	–0.04	2 (0.3)	.99	< 0.01	5 (0.7)	.55	0.03	1 (0.1)	.49	–0.04	0 (0.0)	.06	–0.07	4 (0.5)	.34	0.03
First person (n = 90)	34 (4.6)	.99	< 0.01	19 (2.6)	.01	0.10	11 (1.5)	.01	–0.10	5 (0.7)	.01	–0.10	20 (2.7)	.68	0.02	22 (3.0)	.08	0.06
Spiritual (n = 47)	15 (2.0)	.44	–0.03	4 (0.5)	.64	–0.03	11 (1.5)	.99	< −0.01	8 (1.1)	.67	0.02	11 (1.5)	.58	0.02	7 (0.9)	.34	–0.02
Political (n = 148)	78 (10.5)	.01	0.15	18 (2.4)	.99	< 0.01	21 (2.8)	.01	–0.11	20 (2.7)	.70	–0.02	51 (6.9)	.01	0.18	22 (3.0)	.13	–0.05
Journalist (n = 63)	34 (4.6)	.01	0.10	9 (1.2)	.55	0.02	2 (0.3)	.01	–0.15	20 (2.7)	.01	0.14	17 (2.3)	.19	0.05	3 (0.4)	.01	–0.11
Doctor (n = 16)	3 (0.40)	.13	–0.06	3 (0.4)	.43	0.03	12 (1.6)	.01	0.18	0 (0.0)	.15	–0.06	3 (0.4)	.99	< −0.01	0 (0.0)	.06	–0.07
Practitioner (n = 20)	9 (1.2)	.49	0.02	2 (0.3)	.99	–0.01	5 (0.7)	.99	< 0.01	2 (0.3)	.75	–0.02	5 (0.7)	.58	0.02	2 (0.3)	.56	–0.04

Note. The population size was n = 744.

^aOrganization = Ref.

Research Question 4

RQ4 was, Was there misalignment between messaging about the hepatitis A outbreak on Twitter from authorities and members of the public? We explored the frequency with which tweets in our sample contained any content that was inconsistent with known messaging from the Health Department of San Diego County with regard to hepatitis A vaccination and sanitation or contamination concerns (Table 4). Most tweets contained content that was not relevant to messaging on issues related to hepatitis A vaccination (92.2%; n = 686) and sanitation or contamination (87.1%; n = 648). For tweets that were relevant, the majority discussing HAV vaccination (89.7%; n = 52) were consistent with messaging from the health department. However, we observed a different pattern for relevant tweets discussing HAV sanitation or contamination concerns, with the proportion discussing issues in a manner inconsistent with messaging from the health department (49.0%; n = 47) evenly split with the proportion of tweets aligned with the health department (51%; n = 49).

TABLE 4—

Misinformation, (Mis)alignment, and Information Engagement Examples: San Diego County, CA; October–December 2017

Variable	Frequency (%)	Example Tweets	Gwet’s Agreement Coefficient
Alignment with vaccination public health messaging			0.94
Yes	52 (7.0)	California declares state of emergency over Hepatitis A outbreak—leading to vaccine shortage. . . . https://t.co/pi2bMWSo4l
No	6 (0.8)	Sooo I found out I’m vaccinated for hep a, I literally haven’t gone anywhere downtown in a month cuz I was scurred
NA	686 (92.2)	Great news to children of #Ethiopia! Inactivated #Polio Virus vaccine introduced in Ethiopia
Alignment with contamination/sanitation public health messaging			0.87
Yes	49 (6.6)	Can’t stress enough: wash your hands frequently!. https://t.co/ggPxz0FEy1
No	47 (6.3)	Either house the individuals experiencing homelessness or deal with hepatitis outbreaks from urine and shit filling your streets!. https://t.co/kGSAiTngQJ
NA	648 (87.1)	@voiceofsandiego Once again City leaders choose the cheapest poss option. Once the Hep A outbreak is controlled, th. . . . https://t.co/X76I5TPW84
Misinformation			0.82
Yes	55 (7.4)	@RobertGebelhoff Formerly Healthy Triplets All Autistic within Hours of Vaccination (First hand report from parents). https://t.co/evweL07dou
No	689 (92.6)	@yeffsucks Not you too! I can’t afford to be sick during exam season especially with the Hep A outbreak in downtown SD
Information and/or opinion seeking			0.99
Yes	43 (5.8)	@drrachael hiiii. Do you suggest we get our son vaccinated? If so when? I have one boy who is 20months old and has been vaccinated once a month . . .and a new born 3 mo old who has had one shot
No	701 (94.2)	@jholman23 I think I’m gonna be getting one soon. Just got my second dose of Hep A/B since everyone is dying from that down here
Information sharing			0.88
Yes	447 (60.1)	Hep A inquiries flood Utah County Health Department after possible exposures—Daily Herald. https://t.co/UYPowy0zVU
No	297 (39.9)	@Jjbats CA has leprosy cases. \n\nWe have to scour our streets because Hep A. \n\nTB of course . . . oh, and I believe we have plague. \n\nThanks Democrats!
Opinion sharing			0.89
Yes	339 (45.6)	I keep reading these terrible opinions that the policy problem causing San Diego’s #hepA outbreak is because plastic bags are now 10 cents.
No	405 (54.4)

Note. NA = not applicable. The population size was n = 744.

Research Question 5

RQ5 was, Was misinformation shared on Twitter during the hepatitis A outbreak? Table 4 also shows the results of our examination of misinformation. We detected the presence of misinformation shared on Twitter during the outbreak. However, we discovered that only a small proportion of tweets (7.4%) contained any misinformation. Most tweets did not expressly align or misalign with county messaging on selected themes.

Research Question 6

RQ6 was, In what ways did users seek or share content about HAV on Twitter? Our last research question sought to examine different and nonmutually exclusive ways that users engaged with HAV content on Twitter (Table 4). Specifically, we looked at content contained in each tweet that indicated users’ engagement as characterized by seeking or sharing information (in the form of news, statistics, or known facts) or opinions. Overall, we found that a higher proportion of information (60.1%) and opinion (45.6%) sharing than information or opinion seeking (5.8%) occurred among users in our sample.

DISCUSSION

In this descriptive study, we note the similarities of previous research¹⁸ to the themes (RQ1) we uncovered surrounding warnings to the public through updates on the progression of the outbreak. We also note that emergent themes related to risk perceptions of the outbreak (e.g., susceptibility and severity related to infection) and perceptions regarding (in)effectiveness of policy and government responses (e.g., response efficacy of recommendations to the public) aligns with concepts of risk–response theorizing.²² Our study also showed that tweets shared by government sources (RQ2) were rare, suggesting a lack of dialogue between government agencies and the public during the outbreak.

Although previous research emphasizes the public’s engagement with information from government authorities during times of crisis, scholars have also noted that often crisis communication on Twitter takes the form of a 1-way flow of information rather than an interactive dialogue between government authorities and members of the public.²³ This lack of dialogue may have further added to public perceptions of insufficient action on the part of the government during the initial stages of the outbreak and critiques that the county’s response was unsatisfactory. In future outbreak scenarios, using the interactive properties of Twitter may help public health authorities more effectively use this communication channel to promote dialogue with affected communities.

Perhaps unsurprisingly, organizations were more likely to tweet themes sharing broad informational content (RQ3), whereas individuals were more likely to share information on risks and advocacy issues (e.g., homelessness). This difference in focus may suggest that a shift away from top-down communication by organizations in our study toward meaningful and engaging dialogue with the public²⁴ did not take place during this outbreak. Without this dialogue, organizations may struggle to address the concerns and informational needs of public audiences, which may create an unintended informational vacuum in which the public seeks desired information elsewhere.²⁵

We also detected some misalignment between messaging about the hepatitis A outbreak on Twitter from authorities and members of the public (RQ4). Although most tweets discussing hepatitis A vaccination were consistent with messaging from the health department, small fractions of tweets with messages that are inconsistent with official public health messaging may be concerning, as even seemingly small pieces of misinformation can be propagated. For example, as Baker discusses, Andrew Wakefield’s published article on a vaccine–autism connection gained traction in part through the Internet²⁶ despite later retraction, and in the recent global COVID-19 pandemic, we have observed the power that even 1 piece of dangerous misinformation uttered once by a political leader despite an immediate response by health experts to correct the situation.²⁷

For tweets falling under themes providing general medical information and HAV-specific information, we found references to misinformation regarding vaccine safety, all of this in light of an ongoing outbreak of a vaccine-preventable disease. Although only a small proportion of tweets in our study contained misinformation (RQ5) at the individual level, in the aggregate even a small amount of misinformation can pose potential issues for public health at the population level if it leads to vaccine noncompliance among enough individuals and subsequent reductions in herd immunity. However, more research is needed to establish thresholds of misinformation and when they reach problematic proportions requiring intervention.

Additionally, the context of an outbreak of an infectious disease such as hepatitis A suggests that the level, nature, and spread of misinformation is likely to vary depending on public health concerns. This particular hepatitis A outbreak largely affected a marginalized and socially distant population; however, outbreaks affecting populations such as children may evoke a greater level of concern that generates content and levels of misinformation differing from what we observed. Future research may examine content and proliferation of misinformation, as well as outbreaks affecting different populations, to confirm this hypothesis.

Our results also indicate that Twitter users’ activities are likely to be dominated by more information- and opinion-sharing efforts than information- and opinion-seeking ones (RQ6). As individuals are likely to engage in information seeking regarding health issues about which they desire more knowledge,²⁸ it is possible that our coding of tweets captured mostly content from users who had already sought information about the outbreak before posting tweets. Future research may seek to examine how users seek information on Twitter, as it is an important source of public health information, including content about vaccinations.²⁹ Last, findings from our study show evidence of some aspects following a distancing, blame, stigma pattern.³⁰ Some tweets conveyed information that the hepatitis A outbreak was largely confined to individuals experiencing homelessness, and although not true of all tweets in our sample discussing the individuals experiencing homelessness in San Diego, a portion of tweets referred to this population in derogatory ways in the context of the hepatitis A outbreak.

Limitations

This study contains some limitations. Our sample was not representative of the entire population of tweets about the hepatitis A outbreak during the specified time frame of our study; we examined content only about HAV from Twitter, and we could not draw inferences related to content flow on Twitter or elsewhere or of the offline effects of these tweets. We also could not draw inferences about the potential impact of messages offline or attributes of messages that would increase their online reach (e.g., liking, retweeting, mentions), an area ripe for future research. However, we can get a sense of the range of diverse topics, level of public awareness, and early stirrings of misinformation on the issue of HAV. Such findings can help health authorities keep abreast of the information environment during a time-sensitive event and offer a method for detecting unfulfilled informational needs that can serve as a starting point for formative research guiding the design of effective messaging. Last, although the hepatitis A outbreak originated in the population experiencing homelessness, our data do not suggest which Twitter users in our sample may be homeless or whether they were obtaining information from Twitter. However, emerging evidence suggests that the use of these platforms among individuals experiencing homelessness may be especially common among youths.³¹

Public Health Implications

Results of this study may help inform policies or practices that increase services to the population experiencing homelessness and may help reduce the spread of hepatitis A by reconsidering the best way to reach the population most affected by this outbreak. Although our data included sources that were health authorities as well as sources that were individuals acting as advocates and organizations that work with the homeless community, it was not clear whether these sources were attempting to use Twitter to reach individuals experiencing homelessness directly in this context. This may be a missed opportunity, given that a portion of tweets coded under “specific HAV medical concerns” would be of great value if they were read by individuals experiencing homelessness.

By connecting themes and information sources, we saw that the fast-paced spread of a vaccine-preventable disease brought antivaccination argumentation to the forefront. Public health professionals may consider engaging with social media to address the public’s risk perceptions and concerns as they arise in real time during an outbreak and tailoring responses that fulfill the informational needs of those who seek direction and reassurance during times of uncertainty. In addition, given the proportion of sources representing journalists and users articulating political viewpoints, it will be vital to ensure that sources in positions of power outside public health exercise care in the content they share about an epidemic and avoid spreading misinformation for commercial or political gain.

Last, public health agencies can use the results of this study to understand whom their messages are reaching, the concerns of these individuals regarding an outbreak, and their reaction to official public health recommendations. Because the containment of infectious disease outbreaks may depend on human behavior (e.g., getting vaccinated, handwashing), such factors can provide insight for how to best engage such individuals with necessary targeted public health messaging if compliance with public health recommendations begins to weaken among certain population groups.

Conclusions

Previous research has found that exposure to misinformation influences vaccine knowledge³² as well as perceptions of vaccine risk compared with perceived susceptibility to diseases that can be prevented by vaccines.³³ Our findings illustrate the ongoing need to develop and refine approaches for advancing health-related misinformation surveillance. Such methods can help illuminate the nature and extent of health-related misinformation on social media and provide guidance on how and when public professionals may need to intervene and enact corrective action, issues particularly relevant during the current COVID-19 pandemic.