JMIR Publications

ABSTRACT

Background:

In the United States, Black sexual minority men (BSMM) are more affected by human immunodeficiency virus (HIV) than any other group. Although epidemiological and behavioral surveillance are integral to identifying BSMM at risk for HIV, over-reliance on self-reported data, inability to observe social contexts, and neglect of populations that have limited engagement with the healthcare system can limit their effectiveness. Digital epidemiological approaches that draw on social media data offer an opportunity to overcome these limitations by passively observing in organic settings activities, beliefs, behaviors, and moods related to HIV vulnerability that are otherwise challenging to capture.

Objective:

This study aimed to demonstrate the potential of using features of BSMM’s social media communication and networks to predict HIV-related health and behavioral outcomes.

Methods:

We draw on Facebook and survey data collected (2016-2018) from BSMM aged 18-35 living in Chicago (n = 316). Using natural language processing, we characterized an individual’s Facebook posts using four novel HIV-related topic dictionaries (sexual health, substance use, sex behavior, and ballroom culture, a salient subculture in Black/Latinx LGBTQ communities), and captured affective tone using the psycholinguistic analysis software LIWC. We used social network methods to capture structural features of BSMM’s friendships (centrality, brokerage, local clustering) and their group affiliations. Adjusted multivariable regressions were performed to examine relationships between Facebook communication and network features and six HIV-relevant health and behavioral outcomes (HIV status, STI incidence, depression, linkage to status neutral care, condomless sex, and sex drug use).

Results:

Among the Facebook communication features, sexual health content was positively associated with linkage to care and condomless sex, substance use content was positively associated with sex drug use, sex behavior content was positively associated with HIV status, and content about ballroom culture was positively associated with depression and negatively associated with linkage to care. With respect to Facebook network features, an individual’s connectedness to other well-connected BSMM was the most telling predictor for its negative associations with depression, condomless sex, and sex drug use. In all, STI incidence was the only outcome for which there were no significant social media features. Adding Facebook predictors to models with self-reported risk factors alone yielded significant model improvements.

Conclusions:

HIV infection disproportionately impacts BSMM in the United States. Finding innovative strategies to detect high risk individuals in this population is critical to eliminating these disparities. Our findings suggest that social media data enable passive observance of social and communicative contexts that evade detection using traditional epidemiological and behavioral surveillance methods. As such, we consider social media data to be promising complements to these more traditional data sources.