Assessing population-level symptoms of anxiety, depression, and suicide risk in real time using NLP applied to social media data

Alex Fine,Joshua Carroll,Jenny Blase,Glen Coppersmith,Patrick Crutchley

doi:10.18653/v1/2020.nlpcss-1.6

Abstract

Prevailing methods for assessing population-level mental health require costly collection of large samples of data through instruments such as surveys, and are thus slow to reflect current, rapidly changing social conditions. This constrains how easily population-level mental health data can be integrated into health and policy decision-making. Here, we demonstrate that natural language processing applied to publicly-available social media data can provide real-time estimates of psychological distress in the population (specifically, English-speaking Twitter users in the US). We examine population-level changes in linguistic correlates of mental health symptoms in response to the COVID-19 pandemic and to the killing of George Floyd. As a case study, we focus on social media data from healthcare providers, compared to a control sample. Our results provide a concrete demonstration of how the tools of computational social science can be applied to provide real-time or near-real-time insight into the impact of public events on mental health.

Highlights

Measurements of the mental health of large populations often become quickly outdated, given traditional techniques for data collection, analysis, and dissemination
Estimates of suicide rates in the United States are often delayed by two years (Hedegaard et al, 2018)
We find no evidence that rescinding stay-at-home orders reversed the deleterious effects of the pandemic on mental health, nor do we find evidence that either healthcare workers or the general population had returned to their respective pre-COVID levels of anxiety, depression, and suicide risk at the time of writing

Summary

Introduction

Measurements of the mental health of large populations often become quickly outdated, given traditional techniques for data collection, analysis, and dissemination. The COVID-19 pandemic, which took root in the United States in February and March of 2020, in addition to threatening the health of a broad swath of the population, placed heavy demands on healthcare providers charged with responding to a highly contagious and deadly novel virus, often under resource-constrained circumstances. The killing of George Floyd on May 25, 2020 elicited nationwide responses of grief and anger, and is widely believed to have surfaced latent psychological trauma in large swaths of the American and international population In both instances, we saw that there was and is no scalable technique for collecting population-scale data to quantify changes in mental health over time, to ask which segments of the population are most severely affected by the situation, or to determine which psychological symptoms are changing in prevalence and what interventions should be prioritized by the community

Methods

Results

Discussion

Conclusion