Neighborhood level chronic respiratory disease prevalence estimation using search query data

Area of Science:

• Public Health
• Epidemiology
• Data Science

Background:

• Estimating disease prevalence at a sub-city neighborhood scale is crucial for targeted interventions.
• Challenges include the high cost of localized data collection and sparse, unrepresentative signals.

Purpose of the Study:

• To utilize alternative data sources for estimating localized respiratory illness prevalence.
• To identify neighborhood-scale prevalence of asthma and Chronic Obstructive Pulmonary Disease (COPD) in the United States.

Main Methods:

• Utilized search engine query data for symptoms, medications, and disease information.
• Constructed features from search data to estimate illness rates across over 23,000 census tracts.
• Incorporated additional data sources like census and land cover data.

Main Results:

• Search data alone correlated with CDC estimates at 0.69–0.76, improving to 0.84 with model enhancements.
• Models integrating search data with census and land cover data achieved high correlations: 0.91 for asthma and 0.88 for COPD.

Conclusions:

• Search query data is a valuable, less sparse alternative for estimating localized disease prevalence.
• Combining search data with other sources significantly enhances the accuracy of respiratory illness prevalence estimation.