Estimating County-Level Overdose Rates Using Opioid-Related Twitter Data: Interdisciplinary Infodemiology Study

Area of Science:

• Public Health
• Data Science
• Epidemiology

Background:

• Drug overdose deaths increased significantly, with a notable 75% rise between April 2020 and April 2021.
• Official overdose data has a substantial reporting lag and limited geospatial resolution (state-level).
• Public social media data offers near real-time availability and precise location information.

Purpose of the Study:

• To determine if opioid-related Twitter data can estimate county-level overdose mortality burden.
• To explore the utility of social media analytics in public health surveillance.

Main Methods:

• Obtained county-level overdose data (ICD codes) and demographic information.
• Collected tweets containing drug-related terms using the Twitter API.
• Employed unsupervised classification for tweet clustering and generated normalized variables and Getis Ord Gi statistic z-scores.
• Utilized linear regression models to predict county-level overdose mortality.

Main Results:

• Demographic variables alone explained only 7.4% of overdose mortality variability.
• A model incorporating demographic data, geospatial analysis (z-scores), and tweet topic counts achieved the best fit (adjusted R²=0.133).
• Key predictors included median age, female population, web-based drug sales mentions (positive association), and Asian race/Hispanic ethnicity (negative association).

Conclusions:

• Statistically transformed social media data can enhance real-time, county-level overdose mortality estimation.
• This approach can improve prediction of opioid-related outcomes, informing prevention and treatment strategies.
• An interdisciplinary method aids evidence-based funding for substance use disorder programs.