Table 2.
Characteristics of studies included in the current systematic review.
| Reference | Data source | Studied disease | Study period | Location searched | Purpose of the study | Used keywords | Type of analysis | Main findings |
|---|---|---|---|---|---|---|---|---|
| Roche et al. (14) | Twitter (423 tweets) | Chikungunya | The first 9 months of the 2014 outbreak | Martinique | To determine the predictive power of Chikungunya-related tweets | Chick*, Chik* | Correlational and regression analysis with epidemiological and environmental variables | Models integrating information from Twitter well explain epidemiological dynamics over time |
| Marques-Toledo et al. (15) | Twitter, Wikipedia access logs | Dengue | September, 2012–October, 2016 | Brazil | To explore the predictive power of tweets in forecasting dengue cases | Dengue | Mathematical model | Tweets can be used to predict and forecast dengue cases |
| Nsoesie et al. (16) | Dengue | Not applicable | Brazil | To understand the determinants of sharing tweets related to dengue | Dengue | Machine learning techniques | Sociodemographic variables play a major role in sharing dengue-related tweets | |
| Ghosh et al. (17) | Websites reporting news | Dengue | 2013–2014 | India, China | To explore the predictive power of models incorporating news | Dengue | Mathematical models and time series-regression techniques | News-based models well correlated with epidemiological cases |
| Gomide et al. (18) | Dengue | 2006–April 2011 | Brazil | To explore whether social media can be effectively integrated into disease surveillance practice | Dengue | Content analysis, correlational analysis and spatiotemporal analysis | Excellent correlation between tweets production and epidemiological cases (R2 = 0.9578) | |
| Guo et al. (19) | Baidu | Dengue | January 2011–December 2014 | China | To explore the feasibility of Baidu in real-time monitoring of Dengue | Dengue | Correlational analysis with epidemiological cases | A strong correlation was found |
| Li et al. (20) | Baidu | Dengue | Not applicable | China | To explore the predictive power of Baidu for forecasting Dengue cases | Dengue | Mathematical model | Baidu-based forecasting with one-week lag well correlated with epidemiological cases |
| Nagpal et al. (21) | YouTube | Ebola | Not applicable | Not applicable | To characterize the content of Ebola popular YouTube videos | Ebola | Content analysis | The most relevant YouTube videos were those presenting clinical symptoms |
| Strekalova (22) | Official centers for disease control and prevention (CDC) Facebook page | Ebola | 18 March 2014–31 October 2014 | Not applicable | To characterize the usage of new media from the CDC | Ebola | Content analysis | Audience engagement with Ebola posts was significantly higher compared to other non-Ebola topics, submitted by CDC |
| Odlum and Yoon (23) | Twitter (42,236 tweets—16,499 unique and 25,737 retweets) | Ebola | 24 July 2014–1 August 2014 | Not applicable (tweets in English) | To exploit Twitter as a real-time method of Ebola outbreak surveillance to monitor information spread | Ebola, #Ebola, #EbolaOutbreak, #EbolaVirus, and #EbolaFacts | Content analysis using NLP (Notepad++ and Weka) and correlation with epidemiological cases | Tweets started to rise in Nigeria 3–7 days prior to the official announcement of the first probable Ebola case |
| Pathak et al. (24) | YouTube (118 videos out of 198 videos) | Ebola | From inception–1 November 2014 | Not applicable | To characterize Ebola-related YouTube videos | Ebola outbreak | Content analysis | The majority of the internet videos were characterized as useful, even though some videos were misleading |
| Roberts et al. (25) | English language websites and Twitter | Ebola | 1 July 2014–17 November 2014 | Not applicable | To qualitatively analyze the Ebola-related narrative | Ebola | Content analysis and sentiment analysis | Public engagement was directed toward stories about risks of U.S. domestic infections than toward stories focused infections in West Africa |
| Sastry and Lovari (26) | Official CDC and World Health Organization pages | Ebola | 1 July 2014–15 October 2014 | Not applicable | To understand the development of an ontological Ebola narrative | Ebola | Narrative analysis framework | Three themes: (a) consulting and containment, (b) international concern, (c) possibility of an epidemic in the United States |
| Liu et al. (27) | Baidu, Sina Micro | Ebola | 20 July 2014–4 September in 2014 | China | To understand the public reaction to the Ebola outbreak | Ebola | Mathematical model | Monitoring of social media enables to capture the spreading of fears related to epidemics outbreaks |
| Househ (28) | Twitter (2,592,5152 tweets) and Google News Trend | Ebola | 30 September 2014–29 October 2014 | Not applicable | To understand the role of the media coverage on public reaction to the Ebola outbreak in terms of digital activities | Ebola | Correlational analysis | A significant correlation between media coverage and tweets production was found |
| Jin et al. (29) | Ebola | Late September 2014–late October 2014 | Not applicable | To understand the public reaction to misinformation related to Ebola outbreak | Ebola or #ebola, #EbolaVirus, #EbolaOutbreak, #EbolaWatch, #EbolaEthics, #EbolaChat, #nursesfightebola, #ebolafacts, #StopEbola, #FightingEbola, and #UHCRevolution | Geo-coded analysis, coding, and mathematical model | Some rumors were more popular than others | |
| Lazard et al. (30) | Twitter (2,155 tweets) | Ebola | 2 October 2014 | United States | To understand the public reaction to the Ebola outbreak | Ebola, #CDCcha | Content analysis using SAS Text Miner 12.1 | Public concerned was about symptoms and lifespan of the virus, disease transfer and contraction, safe travel, and protection of one’s body |
| Alicino et al. (31) | Google Trends | Ebola | 29 December 2013–14 June 2015 | Worldwide | Real-time monitoring and tracking of Ebola virus outbreaks | Ebola, virus Ebola, Ebola virus, Ebola 2014, 2014 West Africa Ebola outbreak | Correlational and regression analysis with epidemiological cases | Correlation was stronger at a global level, but weaker at nation/country level |
| Basch et al. (32) | YouTube (100 most viewed videos viewed more than 73 million times) | Ebola | Not applicable | Not applicable | To analyze the most viewed Ebola-related videos | Ebola | Content analysis | YouTube could on the one hand enhance education and on the other hand spread misinformation |
| Fung et al. (33) | Twitter, Google Trends | Ebola | September 2014–November 2014 | Worldwide | To understand the public reaction to the Ebola outbreak and the first US case | Ebola | Qualitative | Worldwide traffic on Twitter and Google increased as news spread about the first US case |
| Fung et al. (34) | Sina Weibo, Twitter | Ebola | 8–9 August 2014 with a follow-up 7 days later | Not applicable | To capture the reaction to misinformation related to Ebola emergency | Ebola | Content analysis (manual coding) | Misinformation about Ebola was circulated at a very low level globally in social media |
| Wong et al. (35) | Twitter (1,648 tweets) | Ebola | September 2014–2 November 2014 | United States | To understand the determinants of tweeting from local health departments | Ebola | Content analysis (manual coding from 2 independent authors) and regression analysis | Approximately 60% of local health departments sent tweets |
| Wong et al. (35) | Twitter via ArcGIS 10.2.2 and Google Trends | Ebola | September 2014–November 2014 | United States | To understand the determinants of tweeting from local health departments | Ebola | Geospatial analysis | Weak, negative, non-significant correlation between online search activity, and per capita number of local health department Ebola tweets by state |
| Towers et al. (36) | Twitter (250,723 tweets), web searches | Ebola | 29 September 2014–31 October 2014 | United States | To understand the impact of the media coverage on the public reaction to Ebola outbreak | Ebola | Mathematical model | 65–76% of the variance in samples was described by the news media contagion model |
| van Lent et al. (37) | Twitter (4,500 tweets from a corpus of 185,253 tweets) | Ebola | 22 March 2014–31 October 2014 | The Netherlands | To understand the predictors of Ebola-related tweet production | Ebola, #Ebola | Content analysis | Significant positive relation between proximity and fear |
| Strekalova (22) | Official CDC Facebook page via a Microsoft Excel add-on, Power Query | Ebola | 25 March 2014–31 October 2014 | Not applicable | To understand the usage of social media by the CDC | Ebola | Content analysis | Differences in audience information behaviors in response to an emerging pandemic, and health promotion posts |
| Fung et al. (38, 39) | Twitter (3,640 tweets on malaria) | Malaria | Not applicable | Not applicable | To characterize malaria-related tweets | #GlobalHealth, #malaria | Content analysis (with unsupervised machine learning techniques) | The main topics were prevention, control, treatment, followed by advocacy, epidemiology, and social impact |
| Ocampo et al. (40) | Google Trends | Malaria | 2005–2009 | Thailand | To exploit the predictive power of Google Trends in forecasting malaria cases | Malaria and malaria-related terms | Correlational analysis | Google Trends-based model well correlated with epidemiological cases |
| Adawi et al. (41) | Google Trends | Mayaro Virus | From inception (1 January 2004 on) | Worldwide | Real-time monitoring and tracking of Mayaro virus outbreaks | Virus Mayaro, Mayaro virus, virus de Mayaro, virus del Mayaro | Correlational and regression analysis | Web searches were driven by media coverage rather than reflecting real epidemiological cases |
| Bragazzi et al. (42) | Google Trends | West Nile virus | From inception (2004 on) | Italy | To exploit the predictive power of Google Trends | West Nile virus | Correlational analysis with epidemiological cases | A positive significant correlation between web searches and cases was found |
| Watad et al. (43) | Google Trends | West Nile virus | From inception (from 2004 on) | United States | To explore the predictive power of Google Trends | West Nile virus | Correlational and regression analyses and mathematical model | Good correlation between web searches and real-world epidemiological figures. Using data 2004–2015 it was possible to predict data for 2016 |
| Basch et al. (44) | YouTube (100 most popular videos) | Zika | Not applicable | Not applicable | To analyze the most viewed Zika-related videos | Zika | Content analysis | Majority of YouTube videos concerned babies, cases in Latin American and in Africa |
| Bragazzi et al. (45) | Google Trends, Google News, Twitter, YouTube, and Wikipedia | Zika | 1 January 2004–31 October 2016 | Not applicable | To capture the public reaction to the Zika outbreak | Zika | Correlational and regression analyses | Public interest was constantly increasing, with public alert on teratogenicity of the Zika virus |
| Dredze et al. (46) | Twitter (138,513 tweets) | Zika | 1 January 2016–29April 2016 | Not applicable | To characterize Zika vaccine-related tweets | Zika vaccine | Content analysis (supervised machine learning techniques) | Most tweets contained misleading information |
| Fu et al. (47) | Twitter (1,076,477,185 tweets collected with Twitris 2.0 via API) | Zika | 1 May 2015–2 April 2016 | Worldwide | Content analysis of Zika-related Twitters data | Zika | Topic modeling was used to group bags of words. The 20-topic model was found to fit the data best, them were grouped in 5 themes | 5 themes: (1) private/public response to the outbreak; (2) transmission routes; (3) societal impacts of the outbreak; (4) case reports; (5) pregnancy and microcephaly |
| Fung et al. (38, 39) | Pinterest (616 posts), Instagram (616 photos) | Zika | Not applicable | Not applicable | To characterize the Zika-related only material shared via Pinterest and Instagram | Zika virus, #zikavirus | Content analysis (manual coding) | Main languages were Spanish or Portuguese. Most popular topics were: prevention, pregnancy, and Zia-related deaths |
| Glowacki et al. (48) | Twitter 1,174 tweets collected | Zika | During an hour-long live CDC Twitter chat on February 12, 2016 | CDC-generated tweets | Content analysis of Zika-related Twitters data | Zika | Text analytics to identify topics and extract meanings, using SAS Text Miner version 12.1 | 10 topics: virology, spread, infants’ sequelaes, how to participate to the chat, prevention, zika test, pregnants’ concerns, sexual transmission, encouraging to engage the chat, symptoms |
| Lehnert et al. (49) | 913 obstetric practice websites randomly selected, Twitter and Facebook | Zika | January 2016–August 2016 | Not applicable | To understand the determinants of social media usage from obstetric community | Zika | Regression analysis | 25–35% of websites reported Zika-related information. Information via social decreased throughout time |
| Majumder et al. (50) | HealthMap and Google Trends | Zika | 31 May 2015–16 April 2016 | Colombia | To develop near real-time estimates for R0 and Robs associated with Zika | Zika | Incidence Decay and Exponential Adjustment (IDEA) model to estimate R0 and the discount factor (d) associated with the ongoing outbreak | Robs estimated with digital data is comparable with the number calculated with the traditional method |
| McGough et al. (51) | Google Trends, Twitter, HealthMap | Zika | May 2015–January 2016 | Colombia, Venezuela, Martinique, Honduras, El Salvador | To explore the predictive power of non conventional surveillance techniques | Zika | Mathematical model | Integrating different non conventional surveillance techniques can improve prediction of Zika cases |
| Miller et al. (52) | Twitter (1,234,605 tweets collected with Twitris 2.0 via API) | Zika | 24 February 2016–27 April 2016 | Not applicable | To determine the relevancy of the tweets regarding: symptoms, transmission, prevention, and treatment | Zika, Zika virus, Zika treatment, Zika virus treatment | Content analysis with a combination of NLP and ML—annotation performed by three microbiologists and immunologists, supervised classification techniques, including J48, MNB, Bayes Net, SMO, SVM, Adaboost, Bagging, and topical analysis with LDA | The majority of the tweets were related to transmission and prevention, and were characterized by a negative polarity |
| Seltzer et al. (53) | Instagram (342 pictures out of 500 tagged images) | Zika | May 2016–August 2016 | Not applicable | To characterize Zika-related images | #zika | Content analysis | Most images conveyed negative feelings (such as fear and concerns) and majority of shared pictures contained misleading information |
| Sharma et al. (54) | Facebook (top 200 posts) | Zika | For a week starting from 21 June 2016 | Not applicable | To characterize the content of Zika-related Facebook posts | Zika | Content analysis | The misleading posts were far more popular than the accurate posts |
| Southwell et al. (55) | Zika | 1 January 2016–29 February 2016 | United States, Guatemala, and Brazil | To determine the role of the media coverage on tweets production | Zika | Correlational analysis | A significant relationship between media coverage and digital behaviors was found | |
| Stefanidis et al. (56) | Twitter (6,249,626 tweets) | Zika | December 2015–March 2016 | Not applicable | To characterize Zika-related tweets in terms of temporal variations of locations, actors, and concepts | Zika | Spatiotemporal analysis | The spatiotemporal analysis of Twitter contributions reflected the spread of interest in Zika from South to North America and then across the globe, with a prominent role played by the CDC and WHO |
| Teng et al. (57) | Google Trends | Zika | 12 February 2016–20 October 2016 | Not applicable | To explore the predictive power of Google Trends | Zika | Mathematical model and correlation with epidemiological cases | The best predictive model was autoregressive integrated moving average (0,1,3) |
| Vijaykumar et al. (58) | Facebook pages of the Ministry of Health and National Environmental Agency (NEA) pages (1057 posts of which 33 were Zika-related) | Zika | 1 March 2015–19 September 2016 | Singapore | To understand the differences in outreach patterns between the preparedness and response stages of an outbreak | Zika | Thematic analysis | Prevention-related posts as garnering the most likes, while update-related posts were most shared and commented upon |