The 2015-2016 Zika virus (ZIKV) epidemic drew international attention.Reference Hennessey, Fischer and Staples 1 ZIKV infection may trigger Guillain-Barré syndrome and during pregnancy can cause microcephaly.Reference Rasmussen, Jamieson and Honein 2 The World Health Organization declared a Public Health Emergency of International Concern on February 1, 2016. 3
The use of social media as a major form of communication about public health events (eg, outbreaks, natural disasters) is becoming more and more popular. For example, in Latin America in 2016, there were 384 million active Internet users, among whom 260 million used social media. In Brazil alone, there were 70 million Internet users. 4 With this trend, it is imperative that public health practitioners understand not only how people are communicating on social media but also the potential differences between different platforms. Photo-sharing social media platforms in particular have become popular worldwide. We defined photo-sharing social media platforms as platforms that are primarily used to share photos, in contrast to video-based platforms (eg, YouTube) and text-based platforms that allow photo sharing (eg, Twitter). Pinterest (San Francisco, CA) and Instagram (Menlo Park, CA) are 2 of the most popular social media platforms in the United States.Reference Duggan 5 As of 2015, Pinterest had 100 million active monthly users and Instagram had 400 million users. 6 , Reference Beck 7 Pinterest is popular among women. In 2015, 44% and 31% of female American Internet users used Pinterest and Instagram, respectively, vis-à-vis 16% and 24% of male users.Reference Duggan 5 Instagram is popular internationally: 75% of Instagram users reside outside the United States vis-à-vis 45% of Pinterest users. 6 , Reference Beck 7 Both platforms are popular among young adults; 55% and 37% of American Internet users aged 18 to 29 years used Instagram and Pinterest, respectively.Reference Duggan 5 Therefore, Pinterest and Instagram may be used by public health professionals to target young individuals with health promotion messages and to monitor reactions of social media users to health issues through the use of visual images.Reference Seltzer, Jean and Kramer-Golinkoff 8 , Reference Guidry, Carlyle, Messner and Jin 9
We analyzed a cross-sectional sample of Pinterest and Instagram users’ photos pertaining to ZIKV. Given the international demographics of Instagram and the onset of the recent ZIKV outbreak in South America, it was anticipated that more Spanish or Portuguese speakers would post ZIKV-related photos on Instagram than on Pinterest. Likewise, it was anticipated that Latin Americans would be more concerned about the potential impact of ZIKV on pregnant women and how they could prevent ZIKV infection than would the English-speaking world.
Since Pinterest and Instagram have different user demographics and their concerns for ZIKV may be different, we postulated that (H1) the percentage of ZIKV-related photos with Spanish or Portuguese texts embedded therein was higher for Instagram than Pinterest, and that (H2) the contents of ZIKV-related photos shared on Pinterest were different from those shared on Instagram.
METHODS
Data Collection
Publicly available ZIKV-related photos were retrieved from Pinterest and Instagram on April 3, 2016, at 2 PM (US Eastern). Pinterest photos were searched by using the key words “zika” and “virus”; Open Screenshot (a browser add-on) was used to screen-capture photos. 10 We reached Open Screenshot’s maximum capacity and captured a maximum of 616 photos. Application programming interface (API) was used to download 9370 Instagram photos with the hashtag “#zikavirus.” We randomly selected 616 of the 9370 Instagram photos for manual coding. We did not collect any meta-data (eg, user ID or posting time). This study was exempt from full Institutional Review Board (IRB) review under B4 exemption, according to Georgia Southern University’s IRB.
Codebook Development and Coding Process
We developed a manual coding codebook (Appendix A in the online data supplement) by examining ZIKV-related Pinterest photos that were screen-captured in the pilot phase of this study. For the purpose of this study, we defined an image to have words “embedded” if the words were literally printed on the image.
To ensure concordant codings, 2 coders each received 3 training sessions and performed 2 rounds of pilot coding before they independently coded 616 Pinterest photos and 616 randomly selected Instagram photos (κ for each category, range: 0.681–1; Appendix B in the online data supplement). In case of discrepancy, EBB made the final decision.
Statistical Analysis
Chi-squared test and Fisher’s exact test were used as appropriate. Univariate logistic regression was used to investigate co-occurrence of topics in the same posts (Appendix B in the online data supplement).
RESULTS
Among the manually coded samples, 47% (290/616) of Pinterest photos and 23% (144/616) of Instagram photos were relevant to ZIKV. Words were embedded in 57% (164/290) of relevant Pinterest photos and all 144 relevant Instagram photos. Among the photos with embedded words, photos in Spanish or Portuguese were more prevalent on Instagram (77/144, 53%) than on Pinterest (14/164, 9%; P<0.0001), supporting hypothesis H1 (further details in Appendix B in the online data supplement).
Table 1 presents the photo contents by category. Regarding hypothesis H2, there was a higher percentage of photos on Instagram than on Pinterest in 3 of 13 categories: ZIKV prevention (59/144, 41%, versus 41/290, 14%; P<0.0001), ZIKV effects on pregnant women (27/144, 19%, versus 32/290, 11%; P=0.04), and deaths associated with ZIKV infection (4/144, 2%, versus 0/290, 0%; P=0.01).
Table 1 Zika-Virus-Related Pinterest and Instagram Photos by Content Category
a Fisher’s exact test was used if any cell in the two-by-two table was <5; otherwise, χ2 test was used.
Univariate logistic regression revealed that some topics were more likely to co-occur on the same ZIKV-related photo. For example, it was 10 times more likely for “birth defect” Instagram photos to be co-coded as “treatment/medical” than non-“birth defect” Instagram photos; it was 19 times more likely for “laboratory” Instagram photos to be co-coded as “new cases” than non-“laboratory” Instagram photos (see the full results in Appendix B in the online data supplement).
DISCUSSION
This cross-sectional study analyzed contents of ZIKV-related photos on Pinterest and Instagram. A higher percentage of photos on Instagram than on Pinterest had embedded Spanish or Portuguese words, and a higher percentage of photos on Instagram than on Pinterest pertained to ZIKV prevention, ZIKV’s effects on pregnancy, and ZIKV-associated deaths. As national and international public health agencies communicate health messages on Pinterest and Instagram, our pilot study provides health communicators with baseline information illustrating the potential use of social media for disseminating public health messages regarding ZIKV prevention and control.
Even though we retrieved the photos using ZIKV-specific key words (Pinterest) and a hashtag (Instagram), many were deemed irrelevant to ZIKV by our coders. One reason is that many users tagged photos with trending but irrelevant key words or hashtags to draw attention, a phenomenon known as “hashtag hijacking.”Reference Xanthopoulos, Panagopoulos and Bakamitsos 11
Our studies had some limitations. First, Pinterest does not offer a search API function that allows key word–specific download. This forced us to rely on web-scrapping. Our screen-capture tool had its limits and, therefore, we were restricted to a convenience sample of 616 Pinterest photos that appeared at the top of the screen and are subject to potential bias.
Second, Pinterest photos had a tag line whereas the Instagram photos that we downloaded did not. Without a tag line, we found that all the Instagram photos that did not have embedded words were deemed irrelevant by our coders. For example, a photo with a mosquito can refer to many things, including different mosquito-borne diseases, and may not be ZIKV-related. This also explains why a higher proportion of Instagram photos, as compared to Pinterest, were deemed irrelevant. Our findings therefore highlight the importance of word-image interactions in health communication. Third, geolocation data were not available. Future research that distinguishes social media posts based on their location of origin will help national and international public health agencies to develop targeted health messages.
CONCLUSIONS
In conclusion, our findings suggest that Pinterest and Instagram serve as relatively similar platforms for the purpose of ZIKV prevention communication as we observed a higher percentage of ZIKV-relevant photos on Instagram than on Pinterest in only 3 of 13 content categories. However, there are differences between the platforms regarding user demographics 4 as well as technicalities associated with data retrieval. For example, Instagram may serve better than Pinterest as a health communication tool targeted toward the Hispanic/Latina populations in the United States. Future research studies between different social media platforms, or with longitudinal data, are warranted.
Acknowledgment
We thank Keri Lubell, PhD, for her feedback on our preliminary codebook. ICHF and ZTHT received salary support from the Centers for Disease Control and Prevention (15IPA1509134 and 16IPA1619505). This article is not related to their CDC-funded projects. This article does not represent the official positions of the CDC or the US Government.
Author Contributions
ICHF conceived the research idea and led the project. Under the supervision of ICHF, EBB co-led the project, developed the codebook, trained the coders, and served as the adjudicator who resolved disagreement between the coders. MEG and LAM manually coded the photos. EBB, MEG, and LAM can all read English, Spanish, and Portuguese. KCC wrote the Python code for the Instagram application programming interface. NS used KCC’s code to retrieve the Instagram photos. EBB retrieved the Pinterest photos by web scraping. ICHF and EBB analyzed the data together. ICHF wrote the first draft of this manuscript with EBB’s inputs. MEE, KWF, and ZTHT contributed intellectual inputs on vector-borne diseases, health communication, and social media data retrieval. ICHF, EBB, CHD, MEE, KWF, and ZTHT edited the manuscript. ICHF and MEE mentored EBB, MEG, and LAM. CHD is a second-year MPH epidemiology student who helped mentor EBB, MEG, and LAM. KWF mentored KCC. ZTHT mentored NS. All authors approved the final manuscript for submission.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/dmp.2017.23
