Administrative Supplement: Enhancing Opioid Surveillance in RADOR-KY using social media

The Rapid Actionable Data for Opioid Response in Kentucky (RADOR-KY; 1-R01 DA057605-01; MPIs Slavova & Talbert) is a program built with funding from the National Institute on Drug Abuse and the support from many Kentucky state agencies and local stakeholders, to deliver timely data and predictive analytics to guide state and local response to the opioid overdose epidemic in the state. The analytical outputs from the program are intended to support data-driven decisions on planning and evaluation for evidence-based practices such as overdose education and naloxone distribution (OEND), medications for opioid use disorder (MOUD), and safer opioid prescribing and dispensing practices, proven to reduce opioid overdose mortality. In order to reduce the time between data collection and dissemination of analytical output, RADOR- KY is designed as an integrated and highly automated surveillance system that will provide an analytical data hub for timely, comprehensive, and actionable drug overdose-related data to end- users (state, and local agencies and organizations). ML predictive models are being built in the automated system to increase analytical productivity. Automated ML supports automation of algorithm selection, parameter tuning, and predictive modeling results that can be disseminated immediately as automated reports and visualizations easy for interpretation by non-technical end- users. A variety of timely data sources are critical for providing better understanding of risk and protective factors relevant to community resource allocation. One new data source for RADOR is social media data that has the potential for improving the predictive validity of the developed models. We propose to work in collaboration with a research team at Stanford that has developed a methodology for extracting signals of opioid use (via mentions of fentanyl and other opioids) from social media data. Using this methodology, the aims of the supplement is to extract direct mentions of opioids and other drugs of abuse in Kentucky counties and use these signals as a new, timely source of data for RADOR-KY’s predictive models for fatal and non-fatal opioid overdoses The work proposed under the supplement has the potential for a significance and impact, enhancing Kentucky’s and RADOR-KY’s analytical capacity. The proposed work will also advance the methodology for predictive ML algorithms in opioid overdose surveillance and forecasting, utilizing social media data.