Characterizing Help-Seeking Searches for Substance Use Treatment From Google Trends and Assessing Their Use for Infoveillance: Longitudinal Descriptive and Validation Statistical Analysis

Abstract

Background: There is no recognized gold standard method for estimating the number of individuals with substance use disorders (SUDs) seeking help within a given geographical area. This presents a challenge to policy makers in the effective deployment of resources for the treatment of SUDs. Internet search queries related to help seeking for SUDs using Google Trends may represent a low-cost, real-time, and data-driven infoveillance tool to address this shortfall in information. Objective: This paper assesses the feasibility of using search query data related to help seeking for SUDs as an indicator of unmet treatment needs, demand for treatment, and predictor of the health harms related to unmet treatment needs. We explore a continuum of hypotheses to account for different outcomes that might be expected to occur depending on the demand for treatment relative to the system capacity and the timing of help seeking in relation to trajectories of substance use and behavior change. Methods: We used negative binomial regression models to examine temporal trends in the annual SUD help-seeking internet search queries from Google Trends by US state for cocaine, methamphetamine, opioids, cannabis, and alcohol from 2010 to 2020. To validate the value of these data for surveillance purposes, we then used negative binomial regression models to investigate the relationship between SUD help-seeking searches and state-level outcomes across the continuum of care (including lack of care). We started by looking at associations with self-reported treatment need using data from the National Survey on Drug Use and Health, a national survey of the US general population. Next, we explored associations with treatment admission rates from the Treatment Episode Data Set, a national data system on SUD treatment facilities. Finally, we studied associations with state-level rates of people experiencing and dying from an opioid overdose, using data from the Agency for Healthcare Research and Quality and the CDC WONDER database. Results: Statistically significant differences in help-seeking searches were observed over time between 2010 and 2020 (based on P<.05 for the corresponding Wald tests). We were able to identify outlier states for each drug over time (eg, West Virginia for both opioids and methamphetamine), indicating significantly higher help-seeking behaviors compared to national trends. Results from our validation analyses across different outcomes showed positive, statistically significant associations for the models relating to treatment need for alcohol use, treatment admissions for opioid and methamphetamine use, emergency department visits related to opioid use, and opioid overdose mortality data (based on regression coefficients having P<=.05). Conclusions: This study demonstrates the clear potential for using internet search queries from Google Trends as an infoveillance tool to predict the demand for substance use treatment spatially and temporally, especially for opioid use disorders.