Characterisation of social media conversations on syphilis: an unobtrusive observational study

Introduction

Syphilis is a sexually transmitted disease caused by Treponema pallidum.¹T. pallidum does not have an animal reservoir, and penicillin continues to be an effective treatment for syphilis.² These two factors have made syphilis a perfect candidate for disease elimination, and in 1999, the Centers for Disease Control and Prevention proposed the Syphilis Elimination Effort to reduce the number of individuals with primary and secondary syphilis in the USA to <1000.³ In 2013, however, the initiative was suspended due to a significant increase in syphilis case numbers.² The incidence of syphilis is highest in low-income countries, but has been up-trending across multiple high- and medium-income countries.⁴ A global prevalence estimate published in 2019 suggests that 0.5% of women and 0.5% of men worldwide have the disease.⁵

In recent years, researchers have looked to conversations on social media platforms, such as Twitter (https://twitter.com/?lang=en), as a means of capturing the public perception of different diseases.^6–10 One group of investigators reported an association between the level of syphilis-focused Twitter activity and numbers of reported cases across USA counties.¹¹ However, Twitter-based discussions about syphilis have not been analysed qualitatively. The content of engagement with user posts on social media could provide information about public understanding of syphilis. Measuring this knowledge may suggest strategies to address gaps that reduce the prevalence of syphilis. Our study characterises the posts – or tweets – written by Twitter users in 2019 related to syphilis.

Methods

The dataset was sourced on 8 September 2020 from Twitter (https://twitter.com/?lang=en) by TrackMyHashtag, a social media analytics company that uses proprietary software (https://www.trackmyhashtag.com). Every tweet posted in English between 1 January and 31 December 2019 that contained the word ‘syphilis’ was identified, including tweets containing the term ‘#syphilis’. Retweets were excluded. The tweet identification number and URL, the date and time of posting, and the number of retweets and likes were collected. The number of replies was not collected. To provide context to the tweets, the author’s screen name, username and biographical statement, plus a link to their Twitter page, were also obtained.

This unobtrusive observational study involved analysis of publicly available data, and review by a human research ethics committee was not indicated.

Numbers of tweets for each month were compared by Poisson regression analysis, and considered to lie outside predicted if the P-value was <0.05. A 10% sample of the dataset was taken for qualitative analysis. This sample size was chosen for feasibility of analysis, and was comparable to similar published studies.^9,12–15 The sample was taken by systematic sampling, including every 10th tweet in the chronological list of all tweets, to avoid selection bias.

Sampled tweets were classified into the following categories based on content: medical resource, non-medical resource, personal experience, public conversation, joke, medical question, non-medical question, spam, testing advertisement and foreign language. This straight-forward classification system was modified from Chew et al.,⁹ who undertook a content analysis of tweets related to the 2009 H1N1 influenza pandemic (Supplementary Table S1). One researcher (AD) classified all tweets taking a two-round approach to ensure precision. If needed for categorisation, the tweet was accessed online via hyperlink and read in the context of any available thread. A small number of tweets in a language other than English were collected; these tweets were removed from the content analysis.

All tweets assigned to the category of medical resource were reviewed by an infectious disease registrar (AF) to determine whether they were medically correct or incorrect. The content of the text, any associated still images or videos and links provided within each tweet were considered in the assessment. Engagement was calculated for each content category, as a ratio of the sum of retweets plus likes to the number of tweets within the category.

Results

In 2019, a total of 111,388 English-language tweets containing the term ‘syphilis’ were posted by 69,921 different authors. The number of tweets in each month was consistent overall, with a mean of 9282 and a median of 9079. A surge in tweets occurred during October, with an increase of 3096 from the mean number of monthly tweets, and there was a drop in tweets during the months of May and June, with a decrease of 1243 and 1239 from the mean, respectively (P < 0.05 for each difference).

Analysis of the content of the 10% sample of tweets (n = 11,113 after removal of 25 foreign language tweets) revealed that the most frequent category was joke (n = 5370, 48%). Other categories of tweets were less frequent, including: public conversation (n = 2515, 23%), medical resource (n = 1762, 16%), non-medical resource (n = 622, 6%), spam (n = 462, 4%), testing advertisement (n = 161, 1%), personal experience (n = 112, 1%), medical question (n = 62, 0.6%) and non-medical question (n = 47, 0.4%; Fig. 1a).

Fig. 1.

Content of tweets containing the term ‘syphilis’ posted in 2019, determined by analysis of a 10% subset (n = 11,113 written in English). (a) Graph presenting number of tweets by content category, showed from most to least frequent. (b) Pie chart showing percentages of medically correct and incorrect tweets in the medical resource category (n = 1762). (c) Graph presenting engagement ratio, representing the ratio of retweets plus likes to all tweets for each category.

Of the 1762 tweets categorised as medical resource, 1484 tweets (84%) were judged to be medically correct, whereas 240 tweets (14%) were deemed to be medically incorrect; for 38 tweets (2%), it was not possible to determine medical accuracy (Fig. 1b). Although this shows the majority of tweets categorised as medical resource are often medically correct, in the context of the sample analysed, only 13% of all tweets presented medically accurate information. Ranking the authors according to the number of tweets posted, ‘HIV Insight’ posted the most tweets categorised as medical resource (n = 34; Table S2).

A calculation of the engagement ratio (retweets + likes:all tweets) for tweets in different content categories revealed the highest engagement for those categorised as personal experience (2109:112, approximately 19:1), whereas the category with the least engagement was non-medical question (39:47, approximately 1:1) (Fig. 1c). Medical resource tweets had an engagement ratio of approximately 7:1 (12,141:1762). The medical resources tweets attracting the highest engagement are presented in Table S3.

Discussion

A total of 111,388 tweets posted on Twitter in 2019 referenced syphilis. By analysing a 10% sample, we found that approximately 15% of tweets provided medical information, and the majority of these medical resource tweets presented correct information. We found that tweets on syphilis that presented a personal experience generated the most engagement, although medical resource tweets also generated some engagement.

Review of the literature identifies no prior studies of the prevalence of medical information on syphilis on social media. However, several groups have published studies analysing Twitter content related to other infectious diseases: Meadows et al.,¹⁵ who focused on the 2015 Californian measles outbreak; Chew et al.,⁹ who reported on the 2009 H1N1 influenza A pandemic; and Gabarron et al.,¹⁶ who considered chlamydial and HIV infections. The annual number of tweets on syphilis in our study far outweighed the time-corrected number on chlamydial and HIV sexually transmitted infections reported by Gabarron et al.¹⁶: 111,388 versus 15,340.

The other author groups found 41–80% of tweets were an information source,^9,15,16 up to five-times more than the percentage we measured. We observed considerably more tweets categorised as a joke (48%) than Chew et al.⁹ and Gabarron et al.,¹⁶ who reported percentages of 8% and 10%.^9,16 Oh et al.¹⁷ studied engagement of HIV-related Twitter messaging, and found that tweets from personal accounts generated more engagement than those from institutional accounts. Consistently, our analysis of tweet engagement suggests those that convey a personal experience with the infection disease are of most interest to Twitter users.

Health misinformation is more common on Twitter than other social media platforms.¹⁸ Incorrect medical information poses risks to the health of persons with syphilis, which may cause serious neurological and cardiovascular complications, and TORCH infection (toxoplasmosis, rubella, cytomegalovirus, herpes simplex and other organisms, including syphilis, parvovirus and Varicella zoster).¹⁹ A total of 84% of tweets that represented a medical resource on syphilis were found to be medically correct. In a study of medical information posted on Twitter around the onset of the coronavirus 2019 (COVID-19) pandemic, albeit on a smaller scale than our work, Swetland et al.²⁰ described a similar percentage of correct medical resource tweets; that is, 74.4% of 358 tweets.

One limitation of our investigation into the syphilis content on social media is the focus on Twitter alone. Although Twitter has been identified as the most popular social media forum for sharing health care-related information,²¹ future studies could expand on our findings by exploring conversations occurring on other social media platforms. As our research was unobtrusive, we cannot comment on author motivations for content type and correctness. Additionally, as only tweets written in English were analysed, our findings are representative of conversations about syphilis led by English-speaking authors. We observed temporal changes in tweet volume, suggesting value of research into correlations with public health events.

Research in the area of syphilis and social media may have value for guiding public health strategies that address the increasing rates of this sexually transmitted disease. Through understanding the type of information online and the types of messages that generate the most user engagement, social media could be used for effective public education. Our study has demonstrated that there is limited medical information about syphilis on Twitter. However, our results also suggest a potential solution to this health education gap could be framing resources with statements of personal experiences for maximum engagement.

Supplementary material

Supplementary material is available online.