Leveraging Internet Search Activity to Identify Marijuana Use at Population Scale

Background: Marijuana use is associated with significant morbidity and mortality. Recent changes in the regulatory landscape around marijuana have been accompanied by historic declines in the percentage of persons who consider marijuana use to be harmful and sharp increases in use. Harnessing big data from population engagement in Internet search activity about marijuana may afford high resolution views into marijuana-related health behaviors and problems, to complement traditional survey reports, a novel approach that builds on successful efforts to harness Internet search activity about myriad health phenomena (infectious disease surveillance, sentiment analysis) to advance population health monitoring efforts.

Methods: We sought to ascertain associations among state-level measures of Internet search activity for marijuana and individual risk for marijuana use as reported in a population-based health survey. Data sources: Past 30-day marijuana use was obtained from the 2016 Behavioral Risk Factor Surveillance System. Internet search activity metrics for marijuana at the state-level were derived from Google Trends – an application that produces relative search volumes (RSVs) for specified search terms. Data Analyses: Multivariate logistic regression was used to estimate odds of marijuana use as predicted by state-level RSVs, adjusting for sociodemographic factors.

Results:Among 106,820 BRFSS respondents (51.9% female, mean age 48.9 years), 7.42% were past 30-day marijuana users.RSVs for “marijuana,” “weed,” “pot,” “joint,” “reefer,” and “cannabis,” were positively associated with individual marijuana use (p<0.01) in bivariate and multivariable analyses, with “cannabis,” “marijuana,” and “weed” showing the strongest association with use (AOR:1.49, 95%CI:1.43-1.56; AOR:1.42, 95%CI:1.37-1.48; AOR:1.37, 95%CI:1.31-1.43, respectively).

Conclusion:Strongpositive associations were found between measures of state-level marijuana-related Internet search activity and individual reports of marijuana use from a structured denominator-based survey, an important proof of concept. Future studies that build on this effort might harness Internet search activity to evaluate trends over time including in relation to changing policies governing marijuana use. Such activities could provide much needed insight into the impact of policy changes on behaviors and harms, to guide prevention and policy sciences.