Methods of using real-time social media technologies for detection and remote monitoring of HIV outcomes

Highlights

• We collected 553,186,061 tweets.

• We filtered tweets by whether they suggested HIV-related risk behaviors (N = 9880).

• We presented a visual map of the location of these HIV-related tweets.

• We found a significant positive county-level relationship between HIV tweets and HIV prevalence.

• We established the feasibility of this method to study HIV-related outcomes.

Abstract

Objective

Recent availability of “big data” might be used to study whether and how sexual risk behaviors are communicated on real-time social networking sites and how data might inform HIV prevention and detection. This study seeks to establish methods of using real-time social networking data for HIV prevention by assessing 1) whether geolocated conversations about HIV risk behaviors can be extracted from social networking data, 2) the prevalence and content of these conversations, and 3) the feasibility of using HIV risk-related real-time social media conversations as a method to detect HIV outcomes.

Methods

In 2012, tweets (N = 553,186,061) were collected online and filtered to include those with HIV risk-related keywords (e.g., sexual behaviors and drug use). Data were merged with AIDSVU data on HIV cases. Negative binomial regressions assessed the relationship between HIV risk tweeting and prevalence by county, controlling for socioeconomic status measures.

Results

Over 9800 geolocated tweets were extracted and used to create a map displaying the geographical location of HIV-related tweets. There was a significant positive relationship (p < .01) between HIV-related tweets and HIV cases.

Conclusion

Results suggest the feasibility of using social networking data as a method for evaluating and detecting Human immunodeficiency virus (HIV) risk behaviors and outcomes.

Introduction

Social networking technologies have recently been used for HIV prevention research (Gold et al., 2011, Young, 2012) as tools for recruitment (Sullivan et al., 2011), interventions (Bull et al., 2012, Young et al., 2013a), and mixed-methods research (Young and Jaganath, 2013). Because people sometimes use these technologies to publicly discuss sexual-related attitudes, desires, and behaviors, researchers may be able to use social networking data to understand and detect real-time individual and regional sexual risk behaviors and social norms (Young and Jordan, 2013). An emerging field, known as digital epidemiology, studies how these “big data” can be used to better understand, detect, and address public health problems (Salathe et al., 2012, Aramaki et al., 2011). However, no known research has been conducted on methods for how or whether these data can be used for HIV prevention or detection, making it important to evaluate the feasibility of this approach. Evaluating methods on how to use social media and “big data” in public health and medicine is an important first step in establishing how these data can be used in prevention, detection, and treatment.

For example, millions of social communications from real-time, geographically-linked, social networking sites, such as Twitter, might be used to make inferences about geographical rates of future or recent past engagement in sexual risk behaviors. Twitter, a large and rapidly growing social networking technology, allows participants to send short, public, real-time “tweet” communications (Smith and Brenner, 2012). Twitter provides public access to these data through an Advanced Programming Interface (API) (Twitter, 2013). People who intend to or have just engaged in sexual or drug-related behaviors might tweet to their social networks to inform them of their attitudes and behaviors (Walker, 2013, Young et al., 2013b). Researchers may be able to link these Twitter data to real-time incidence data to better understand and detect public health outbreaks. For instance, influenza researchers have compared flu data with tweets related to influenza symptoms and found tweets have been able to detect influenza outbreaks in regions where the tweets occurred, in advance of traditional surveillance methodologies (Aramaki et al., 2011).

HIV researchers could build on this approach by studying whether engagement in sexual risk behaviors could be inferred from tweet content, for example by filtering for keywords that suggest sexual risk and drug use behaviors (i.e., HIV risk behaviors). Because Twitter provides geographical locations (i.e., geolocated data) for some conversations, HIV risk-related tweets can ultimately be mapped alongside incidence rates to determine whether regional rates of HIV-risk conversations on Twitter could be associated with HIV transmission in those regions. However, these topics have not been studied, making it important to evaluate the feasibility of studying whether and how HIV-risk behaviors are communicated using real-time social media and whether these communications could be linked to allow analysis of data on HIV transmission.

This study is designed to evaluate the feasibility of developing methods of using “big data” to understand whether and how HIV and drug risk behaviors are communicated online in real-time and how these data might be used to inform HIV prevention and detection efforts. Specifically, this study seeks to determine 1) whether geolocated conversations about HIV risk (sexual and drug use) behaviors can be extracted from real-time social networking data, 2) the prevalence and content of these conversations, and 3) the feasibility of using HIV risk-related real-time social media conversations as a method of remote monitoring and detecting HIV transmission.