This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Public Health and Surveillance, is properly cited. The complete bibliographic information, a link to the original publication on https://publichealth.jmir.org, as well as this copyright and license information must be included.
Twitter is becoming an increasingly important avenue for people to seek information about HIV prevention. Tweets about HIV prevention may reflect or influence current norms about the acceptability of different HIV prevention methods. Therefore, it may be useful to empirically investigate trends in the level of attention paid to different HIV prevention topics on Twitter over time.
The primary objective of this study was to investigate temporal trends in the frequency of tweets about different HIV prevention topics on Twitter between 2014 and 2019.
We used the Twitter application programming interface to obtain English-language tweets employing #HIVPrevention between January 1, 2014, and December 31, 2019 (n=69,197, globally). Using iterative qualitative content analysis on samples of tweets, we developed a keyword list to categorize the tweets into 10 prevention topics (eg, condom use, preexposure prophylaxis [PrEP]) and compared the frequency of tweets mentioning each topic over time. We assessed the overall change in the proportions of #HIVPrevention tweets mentioning each prevention topic in 2019 as compared with 2014 using chi-square and Fisher exact tests. We also conducted descriptive analyses to identify the accounts posting the most original tweets, the accounts retweeted most frequently, the most frequently used word pairings, and the spatial distribution of tweets in the United States compared with the number of state-level HIV cases.
PrEP (13,895 tweets; 20.08% of all included tweets) and HIV testing (7688, 11.11%) were the most frequently mentioned topics, whereas condom use (2941, 4.25%) and postexposure prophylaxis (PEP; 823, 1.19%) were mentioned relatively less frequently. The proportions of tweets mentioning PrEP (327/2251, 14.53%, in 2014, 5067/12,971, 39.1%, in 2019;
Twitter may be a useful source for identifying HIV prevention trends. During our evaluation period (2014-2019), the most frequently mentioned prevention topics were PrEP and HIV testing in tweets using #HIVPrevention. Strategic responses to these tweets that provide information about where to get tested or how to obtain PrEP may be potential approaches to reduce HIV incidence.
Globally, 1.5 million (1.1-2.0 million) people became infected with HIV in 2021 [
Social media sites are becoming increasingly important avenues for people of all age groups to seek information about health issues, including HIV [
Previous research on social media and HIV information suggests that social media may be an effective avenue for spreading and consuming HIV information because it allows for anonymity and reduces stigma-related barriers to information seeking [
Some research suggests that social media may have a beneficial effect on the adoption of HIV prevention behaviors. For example, social support provided by social media engagement prevention-specific messages have been associated with improved access to and uptake of HIV prevention and testing [
Despite these positive findings, it has also been shown that Twitter can be used to propagate messages that perpetuate HIV-related stigma and endorse risky sexual behaviors [
Taggart and colleagues [
Importantly, Taggart and colleagues [
This study was designed based on the same reasoning employed by Taggart and colleagues [
We employed a passive, retrospective infodemiology approach in which we collected tweets that included #HIVPrevention (n=69,197) during a 6-year timeframe (2014-2019) corresponding to a critical period related to the uptake of PrEP in the United States and globally. We examined trends in the frequency of mentions of 10 different HIV prevention topics and assessed changes in the proportion of tweets mentioning each topic in 2019 as compared with 2014. We also report descriptive information on the spatial distribution of geotagged #HIVPrevention tweets in relation to the number of state-level HIV cases in the United States, the most frequently used word pairings in the tweets, the accounts posting the most original tweets, and the accounts retweeted most frequently. We conclude by discussing the implications of our findings and suggesting the opportunities for leveraging HIV prevention communication on Twitter to reduce HIV incidence.
We conducted a retrospective infodemiology study using publicly available tweets employing #HIVPrevention between 2014 and 2019.
We utilized the Twitter application programming interface to collect all tweets (including original tweets, retweets, quote tweets, and replies) written in the English language that employed #HIVPrevention between January 1, 2014, and December 31, 2019 (n=69,197). We selected the timeframe 2014-2019 because it corresponds to a period following Food and Drug Administration (FDA) approval of PrEP for HIV prevention in the United States (occurring in 2012) [
We performed several descriptive analyses (eg, tabulation, Pearson correlation) to investigate the characteristics of the data. All analyses were conducted in R statistical software (version 4.0.3; R Foundation for Statistical Computing).
To investigate the change in activity related to tweets using #HIVPrevention over the study period, we tabulated original tweets (including replies and relevant quote tweets) and retweets (including relevant quote tweets) by month and year to identify trends.
To determine the Twitter accounts that generated the highest proportion of original #HIVPrevention tweets during the study period, we tabulated the number of original tweets as a function of unique account usernames.
To determine which Twitter accounts’ #HIVPrevention tweets were retweeted at the highest frequencies, we tabulated the number of retweets associated with #HIVPrevention tweets that each unique account username received.
To identify the most frequently used word pairings, also known as bigrams, we used the tidytext package (version 0.3.1) in R. This method allows for an indication of the context in which words are used. For example, a tweet containing the text “PrEP is an effective tool” will correspond to the following 2 bigrams: (1) PrEP and effective, and (2) effective and tool. Using
To understand the relationship between geotagged tweets and the number of HIV cases, we performed a Pearson correlation to assess the relationship between the number of geotagged tweets at the state level between 2014 and 2019 and the number of HIV cases in 2019 at the state level [
To determine the frequency at which various HIV prevention topics were mentioned in #HIVPrevention tweets and retweets and whether this changed over the study period, we first developed a list of 10 prevention topics and relevant keywords. We selected prevention topics based on the topics identified by the UNAIDS 2016 WAD campaign and our review of the literature. The 10 selected prevention topics were PrEP, postexposure prophylaxis (PEP), condom use, abstinence, VMMC, EMTCT, HIV testing, harm reduction, gender inequity and violence against women, and sex work.
We developed a keyword list for these 10 prevention topics by drawing on the initial stages of summative qualitative content analysis [
Following initial development of our keyword list, we iteratively refined it using a manual inspection process to ensure that our keyword list had a high level of sensitivity and an acceptable level of specificity. That is, we sought to identify all tweets mentioning a particular prevention method (true positives) while minimizing any miscategorization (false positives). An example of a false positive would be a tweet referring to the US President’s Emergency Plan for AIDS Relief (PEPFAR) that was categorized under PEP. As some miscategorization was inevitable, we accepted an error level that was ≤5% (ie, in our manual inspection, ≤25 of the 500 inspected tweets were not related to the respective prevention method). If greater than 5% error was detected, we made appropriate modifications to our keyword list to fix the inaccuracies. We similarly inspected samples of the tweets which were uncategorized to determine if we missed any keywords that were relevant to a particular category (ie, to minimize false negatives). When these were discovered, we refined our keyword list to include the relevant keyword. If a tweet mentioned keywords related to more than 1 prevention topic (eg, “PrEP”, “condom”), then that tweet was categorized in each respective category. If a tweet mentioned multiple keywords related to the same prevention category, that tweet was counted in the respective category only once. The manual inspection process was conducted by the first author (RB) and the final list of keywords (
To evaluate how attention to each topic changed over the study period, we compared the proportion of tweets related to each respective topic in 2019 with the proportion of tweets related to each respective topic in 2014 using chi-square and Fisher exact tests.
Manual inspection process for refining the keyword list.
The study was granted an Ethics Exemption by the Yale University Institutional Review Board (#2000028381).
Our sample consisted of 25,031 original tweets and 44,166 retweets, totaling 69,197 tweets. Geotagged tweets represented 1.81% (n=1253) of the sample and were tweeted from 76 countries.
The 10 accounts that generated the most original #HIVPrevention tweets between 2014 and 2019 are presented in
Annual frequency of #HIVPrevention tweets between 2014 and 2019.
Accounts with the most original and retweeted #HIVPrevention tweets between 2014 and 2019.
Accounts with the most original #HIVPrevention tweets | Number of original tweets per account | Accounts whose #HIVPrevention tweets were retweeted at the highest frequencies | Number of retweets per account |
@HIV_Insight | 3144 | @UNAIDS | 11,239 |
@Sex_Worker_Hlth | 484 | @HIV_Insight | 1880 |
@DrMbere | 465 | @MichelSidibe | 1551 |
@Hlth_Literacy | 444 | @UN | 908 |
@HIVIreland | 396 | @MissUniverse | 705 |
@UNAIDS | 296 | @UNAIDS_AP | 687 |
@EPICBrowardOrg | 262 | @HIVpxresearch | 499 |
@Health_HIV2030 | 240 | @accphivprn | 493 |
@HopeandHelpInc | 221 | @AniShakari | 470 |
@himmoderator | 183 | @HIVIreland | 468 |
The 10 accounts whose #HIVPrevention tweets were retweeted at the highest frequencies between 2014 and 2019 are presented in
A visual word network of the 50 most frequently used bigrams (word-pairings) in #HIVPrevention tweets between 2014 and 2019.
Geographic distribution of geotagged English #HIVPrevention tweets (n=514) in the United States between 2014 and 2019. The numbers in the figure correspond to the number of tweets geotagged to the respective locations indicated on the map. The mapping data presented here is available under the Open Database (CC-BY-SA) License [
Geographic distribution of the number of HIV cases in the United States in 2019, displayed at the state level. The mapping data presented here is available under the Open Database (CC-BY-SA) License [
Relationship between the total number of geotagged #HIVPrevention tweets at the state level between 2014-2019 and the number of 2019 HIV cases by state.
Of the total 69,197 #HIVPrevention tweets in the sample, 28,135 tweets (40.66%) were categorized into 1 or more of the 10 identified prevention topics. The highest proportion of mentions were seen for PrEP (13,895/69,197 tweets, 20.08% of all tweets). This was followed by the proportion of mentions related to HIV testing (7688/69,197, 11.11%), condoms (2941/69,197, 4.25%), harm reduction (2173/69,197, 3.14%), gender equity and violence against women (1695/69,197, 2.45%), VMMC (969/69,197, 1.40%), sex work (872/69,197, 1.26%), PEP (823/69,197, 1.19%), EMTCT (277/69,197, 0.40%), and abstinence (180/69,197, 0.26%). Categorized tweet totals do not add to 28,135, given that some tweets were categorized in more than 1 category.
Annual frequency of mentions of keywords related to abstinence, condom use, elimination of mother-to-child transmission, HIV testing, post-exposure prophylaxis, pre-exposure prophylaxis, and voluntary medical male circumcision as a proportion of total annual #HIVPrevention tweets between 2014 and 2019.
Results of the chi-square and Fisher exact testsa evaluating the overall change in the proportion of tweets related to each topic area in 2014 versus 2019.
Prevention topic | 2014 tweets (n=2251), n (%) | 2019 tweets (n=12,971), n (%) | Direction of change | |
Abstinence | 8 (0.36) | 4 (0.03) | ≤ |
Lower |
Condom use | 138 (6.13) | 421 (3.25) | ≤ |
Lower |
Preexposure prophylaxis | 327 (14.53) | 5067 (39.06) | ≤ |
Higher |
Voluntary medical male circumcision | 78 (3.47) | 295 (2.27) | ≤ |
Lower |
Postexposure prophylaxis | 25 (1.11) | 342 (2.64) | ≤ |
Higher |
Harm reduction | 93 (4.13) | 336 (2.59) | ≤ |
Lower |
Gender inequity and violence against women | 74 (3.29) | 251 (1.94) | ≤ |
Lower |
Elimination of mother-to-child transmission | 9 (0.40) | 51 (0.39) | .96 | N/Ac |
Sex work | 39 (1.73) | 198 (1.53) | .47 | N/A |
HIV testing | 208 (9.24) | 2193 (16.91) | ≤ |
Higher |
aFisher exact test used when expected frequencies are less than 5.
bItalicized values are statistically significant.
cN/A: not applicable.
In this study, we investigated temporal trends in the frequency of mentions of 10 different HIV prevention topics in #HIVPrevention tweets between 2014 and 2019. Our findings describe how attention to different HIV prevention methods on Twitter has changed over time, which may provide insight into changes in the acceptability and uptake of these prevention methods. We also report useful descriptive information about our sample, such as the characteristics of accounts receiving the most retweets of #HIVPrevention tweets. These findings may assist public health professionals in identifying strategic approaches to improving the dissemination of HIV prevention information on Twitter.
Key findings from our analysis include the following: both PrEP and HIV testing were discussed at relatively high frequencies during the study period as compared with other HIV prevention methods such as condom use, VMMC, EMTCT, and PEP. Moreover, there were significantly higher proportions of #HIVPrevention tweets mentioning PrEP, HIV testing, and PEP in 2019 as compared with 2014, although the largest changes are seen for PrEP and testing. There were significantly lower proportions of #HIVPrevention tweets mentioning abstinence, VMMC, condom use, harm reduction, and gender inequity and violence against women in 2019 as compared with 2014. The increases in the proportion of tweets related to PrEP in 2017, 2018, and 2019 likely reflect approvals of PrEP for use in countries around the world between 2016 and 2018, including South Africa, South Korea, and the European Union [
The high proportion of #HIVPrevention tweets related to PrEP and HIV testing is promising given that PrEP is highly effective at preventing HIV transmission [
However, optimal adherence to PrEP can be a challenge for at-risk individuals; barriers include stigma, health system inaccessibility, and competing life stressors [
Although our analysis reveals that individuals are responsible for the majority of accounts that correspond to the highest number of original #HIVPrevention tweets, UN-affiliated institutions and individuals appear to be reaching the most people as indicated by their retweet and follower counts, an unsurprising finding given the 2016 WAD campaign. However, the analysis of content generating and retweeted accounts also reveals the importance of informal advocates (eg, @HIV_Insight) and celebrity endorsements (eg, @MissUniverse); the latter may be particularly effective given the sheer number of users following celebrity accounts and the influence celebrities can have on health promotion [
Finally, our analysis of geotagged tweets suggests that #HIVPrevention tweets at a state level between 2014 and 2019 are positively correlated with the number of state-level HIV cases in the United States in 2019. This finding should be interpreted with caution given the small number of tweets in our sample that were geotagged. However, this finding is aligned with other research that suggests that tweet content such as discussing HIV risk-related behavior (eg, drug use) is associated with the geographic distribution of HIV [
To the best of our knowledge, this study is the first to investigate temporal trends in the relative attention received by different HIV prevention methods on Twitter. We are hopeful that the findings provide useful insight into how attention to HIV prevention methods on Twitter has changed over time, which may reflect or influence changes in the acceptability of these methods.
Some of the findings may be useful in informing strategic approaches to the dissemination of HIV prevention information on Twitter. For example, the findings indicate that a large portion of #HIVPrevention tweets mention PrEP and HIV testing. These tweets could be responded to by providing specific information about where to obtain an HIV test or how to access PrEP, which may empower individuals to engage in these behaviors. Furthermore, public health entities could consider leveraging celebrities as HIV advocates on Twitter given their wide reach and popularity, especially with young people. Finally, public health institutions may consider increasing communication about certain HIV prevention methods on Twitter such as condom use to ensure that populations with diverse needs and resources are aware of the HIV prevention options available to them.
There are some limitations of our study. First, we were limited to results procured from a single Twitter hashtag. Although this was a necessary methodological decision to define a sample of tweets focused on HIV prevention, it omits tweets that discuss HIV prevention but do not employ #HIVPrevention and it is possible that these tweets differ importantly from those that do employ the hashtag. Although we examined tweets corresponding to a critical period in the evolution of the acceptability of PrEP, resource and feasibility constraints limited us from investigating tweets posted immediately after PrEP was approved in the United States in 2012. We point the reader to previous research that yields insights into earlier periods [
Twitter is an important avenue for information seeking about HIV prevention and may be a particularly important platform for disseminating information to young adults who represent a large burden of new infections [
Identified prevention methods, associated keywords and examples of tweets related to each topic.
ending the HIV epidemic in the United States
elimination of mother-to-child transmission
Food and Drug Administration
men who have sex with men
postexposure prophylaxis
US President’s Emergency Plan for AIDS Relief
preexposure prophylaxis
Joint United Nations Programme on HIV/AIDS
voluntary medical male circumcision
World AIDS Day
World Health Organization
The authors thank all in the scholarly community who provided insight or asked critical questions about the study, such as when the findings were presented at the American Public Health Association Annual Meeting in October 2021. Funding to present this work at the American Public Health Association Conference in October 2021 was provided by the Yale School of Public Health and the Yale Graduate Student Assembly. RB’s PhD stipend is funded in part by a Doctoral Foreign Study Award from the Canadian Institutes of Health Research. The article processing fee (APF) for this article was paid by YR’s K01 grant (#K01MH111374).
Data from this study are available upon request. Please email requests to josemari.feliciano@yale.edu. Proof of ethics approval/exemption for proposed use may be required.
RB led the study conceptualization, literature search, the qualitative component of the data analysis, data interpretation, and original draft writing. JTF led the data collection, quantitative data analysis, figure creation, and contributed to data interpretation and original draft writing. LL contributed to the study conceptualization, literature search, data interpretation, and manuscript editing. YR contributed to the study design, data interpretation, manuscript revision, and supervised the project. RB and JTF have directly accessed and verified the underlying data reported in the manuscript.
None declared.