Detecting Youth Drinking and Associations with Alcohol Policies Via Social Media

Introduction: Sizeable populations worldwide generate vast quantities of digital data from communications in online social media. Mining such communications delivers low-cost, high-resolution views into public health phenomena to complement traditional systems. This approach has been used to understand outbreak severity on a daily timescale, understand environmental risk factors for obesity at neighborhood resolution and surveillance of adverse events from populations that otherwise would not be identified. Adolescents and young adults are among the groups most heavily engaged in social media including with regard to posting about their health and social behaviors. Harnessing youth engagement with social media to derive timely information about alcohol use behaviors and problems may accelerate public health awareness and drive targeted response.

Methods: Our aim is to validate social media sourced metrics of underage alcohol use against data from traditional health surveillance systems, and further, to establish the relationship between these social media sourced metrics and surrounding state alcohol policies. The approach includes collection of text reports from Twitter, creating alcohol use metrics from these data by parsing, filtering and classifying them using machine learning techniques and ascertaining associations among extant, validated alcohol policy scores measured at the state level and social media sourced measures of underage drinking and harms.

Results: For this presentation, we report preliminary findings of research to derive and validate metrics of alcohol consumption from information posted on Twitter by underage youth from the US. Preliminary findings suggest alignment between temporal patterns of alcohol related social media posting behavior and known temporal patterns of drinking behavior—a first confirmatory step of using social media sources for surveillance of drinking behaviors.  Approaches to defining metrics and ascertaining temporal patterns are discussed along with implications of preliminary results for advancing surveillance and intervention research. The presentation will close with a brief summary of initial lessons learned concerning implementing the approach on commercial platforms that operate outside of investigator control.

Conclusions: Study findings will enable a new approach to population surveillance of underage drinking that will complement traditional systems and advance public health methods for monitoring underage drinking and harms and generate a paradigm shift for policy evaluation by enabling real-time investigation of impacts at fine temporal- spatial resolution.