Evaluating Google, Twitter, and Wikipedia as Tools for Influenza Surveillance Using Bayesian Change Point Analysis: A Comparative Analysis

Abstract

BackgroundTraditional influenza surveillance relies on influenza-like illness (ILI) syndrome that is reported by health care providers. It primarily captures individuals who seek medical care and misses those who do not. Recently, Web-based data sources have been studied for application to public health surveillance, as there is a growing number of people who search, post, and tweet about their illnesses before seeking medical care. Existing research has shown some promise of using data from Google, Twitter, and Wikipedia to complement traditional surveillance for ILI. However, past studies have evaluated these Web-based sources individually or dually without comparing all 3 of them, and it would be beneficial to know which of the Web-based sources performs best in order to be considered to complement traditional methods.ObjectiveThe objective of this study is to comparatively analyze Google, Twitter, and Wikipedia by examining which best corresponds with Centers for Disease Control and Prevention (CDC) ILI data. It was hypothesized that Wikipedia will best correspond with CDC ILI data as previous research found it to be least influenced by high media coverage in comparison with Google and Twitter.MethodsPublicly available, deidentified data were collected from the CDC, Google Flu Trends, HealthTweets, and Wikipedia for the 2012-2015 influenza seasons. Bayesian change point analysis was used to detect seasonal changes, or change points, in each of the data sources. Change points in Google, Twitter, and Wikipedia that occurred during the exact week, 1 preceding week, or 1 week after the CDC’s change points were compared with the CDC data as the gold standard. All analyses were conducted using the R package “bcp” version 4.0.0 in RStudio version 0.99.484 (RStudio Inc). In addition, sensitivity and positive predictive values (PPV) were calculated for Google, Twitter, and Wikipedia.ResultsDuring the 2012-2015 influenza seasons, a high sensitivity of 92% was found for Google, whereas the PPV for Google was 85%. A low sensitivity of 50% was calculated for Twitter; a low PPV of 43% was found for Twitter also. Wikipedia had the lowest sensitivity of 33% and lowest PPV of 40%.ConclusionsOf the 3 Web-based sources, Google had the best combination of sensitivity and PPV in detecting Bayesian change points in influenza-related data streams. Findings demonstrated that change points in Google, Twitter, and Wikipedia data occasionally aligned well with change points captured in CDC ILI data, yet these sources did not detect all changes in CDC data and should be further studied and developed.

Trends of infodemiology studies: a scoping review

“…Some of the demand + supply based studies in this review used Web 1.0, Web 2.0 technologies spontaneously for disease surveillance, but none of them used Web technologies alongside epidemiological data, as used by Sharpe, Hopkins, Cook and Striley () and Woo et al. ().…”

Situating Wikipedia as a health information resource in various contexts: A scoping review

“…There is evidence to support the feasibility of Wikipedia in disease monitoring in relation to seasonal changes in mood [84] and seasonal fluctuations in influenza [85][86][87]. However, while the work on the development of a plausible means to use Wikipedia's access log data to monitor and forecast disease continues [88], one study actually proposes Google data could be more useful [89] and another study struggles to prove the utility of Wikipedia for this purpose [90], although this is likely due to the fact that the researchers used a sample of 1,633 diseases rather than focus on a single disease as other studies have done.…”

Ideas for how informaticians can get involved with COVID-19 research

“…For example, while Twitter data has been utilized for surveillance and content analysis, a significant portion of research using Facebook has focused on communication rather than lexical content processing [41,42]. For health monitoring and surveillance research from social media, the most common topic has been influenza surveillance [43,44]. From the perspective of informatics and NLP, proposed techniques have typically been in the areas of data collection (e.g., keywords and queries) [45,46], text classification [47,48], and information extraction [49].…”