Conversational topics of social media messages associated with state-level mental distress rates.

Upstream public health indicators of poor mental health in the United States (U.S.) are currently measured by national telephone-based surveys; however, results are delayed by 1-2 years, limiting real-time assessment of trends. The aim of this study was to evaluate associations between conversational topics on Twitter from 2018 to 2019 and mental distress rates from 2017 to 2018 for the 50 U.S. states and capital. We used a novel lexicon, Empath, to examine conversational topics from aggregate social media messages from Twitter that correlated most strongly with official U.S. state-level rates of mental distress from the Behavioral Risk Factor Surveillance System. The ten lexical categories most positively correlated with rates of frequent mental distress at the state-level included categories about death, illness, or injury. Lexical categories most inversely correlated with mental distress included categories that serve as proxies for economic prosperity and industry. Using the prevalence of the 10 most positively and 10 most negatively correlated lexical categories to predict state-level rates of mental distress via a linear regression model on an independent sample of data yielded estimates that were moderately similar to actual rates (mean difference = 0.52%; Pearson correlation = 0.45,  < 0.001). This work informs efforts to use social media to measure population-level trends in mental health.