The researchers from Stony Brook University, in collaboration with computer scientists and psychologists from Stanford University and the University of Pennsylvania, conducted a study to evaluate the prevalence of depression and anxiety in almost half of the counties in the United States. 

Depression and anxiety pose substantial mental health obstacles in society, affecting around 10.8 per cent of American adults in 2019, as revealed by the National Center for Health Statistics (NCHS) and the Census Bureau. The correlation between declining mental well-being and the escalating incidence of suicide and mortality connected to opiate use is becoming more evident.

Traditional population health assessments use expensive telephone surveys to measure "sadness" or "worry." However, these polls often need more data to track local community developments accurately.

The study employed a vast dataset on depression and anxiety measurements, using Language-based Mental Health Assessments (LBMHAs). This cutting-edge AI system assesses communities' mental health by analyzing language used on social media. By examining close to one billion tweets from over two million users across 1,418 counties in the United States, researchers discovered that LBMHAs (Location-Based Mental Health Assessments) provide more dependable evaluations than traditional surveys.

Stony Brook University researchers found that LBMHAs were more reliable than major public polls when evaluating public well-being. In addition, they exhibited significant external validity by accurately forecasting other community indicators linked to mental health, such as mortality rates, with more efficacy than surveys. Furthermore, the scores provided by AI showed greater prediction capability for a range of social, economic, and political factors, suggesting that this approach has extensive possibilities beyond just assessing mental health.

The Language-based Mental Health Assessments (LBMHAs) system represents the culmination of nearly a decade of research to develop robust mental health assessment tools. This extensive work involved several key components:

• Geo-Locating Twitter/X Users: Researchers developed techniques to geo-locate Twitter or similar platform users, enabling the identification of users' geographical locations.
• Determining Language Use Patterns: Analysis was conducted to understand users' language use patterns based on their tweets or posts. This step involved extracting linguistic features indicative of mental health states.
• Combining Estimates into Regions: The data from individual users were aggregated and combined into regions, allowing for a comprehensive understanding of mental health trends at various geographical levels.
• Adapting AI Models: AI models capable of analyzing language to estimate mental health were adapted and optimized to function effectively within the context of Twitter or similar platforms.

In 2020, the effectiveness of the LBMHAs system was evaluated by matching rates of depression with expressions of sadness and rates of anxiety with expressions of worry, as collected through representative phone surveys. Surprisingly, the system demonstrated remarkable accuracy in this matching process.

Furthermore, the study found that the proposed LBMHA system outperformed traditional survey methods by a significant margin, improving by ten percentage points in correlation with external factors such as education, housing, income, and socialization. The LBMHAs system provides accurate mental health assessments and offers valuable insights into the socio-economic and environmental factors associated with mental well-being.

Source: https://www.nature.com/articles/s41746-024-01100-0

Image source: Unsplash