Utilizing big data without domain knowledge impacts public health decision-making

Area of Science:

• Environmental health
• Urban planning
• Data science

Background:

• Emerging data sources like Google Street View (GSV) images offer potential for informing urban interventions.
• However, biases from data robustness issues and spurious correlations can impact decision-making.
• Accurate assessment of built environment features is crucial for effective community health strategies.

Purpose of the Study:

• To investigate the robustness of GSV data for inferring built environment characteristics.
• To analyze the mediating role of physical inactivity in the relationship between the built environment and health outcomes.
• To compare the effectiveness of interventions targeting physical inactivity versus built environment features.

Main Methods:

• Analysis of 2.02 million Google Street View images in New York City.
• Integration with health, demographic, and socioeconomic data at the census tract level.
• Application of a causal framework to account for mediation by physical inactivity.

Main Results:

• Inferred built environment characteristics from GSV labels showed poor alignment with ground truth at the intracity level.
• Physical inactivity was identified as a significant mediator of built environment impacts on health.
• Interventions targeting physical inactivity demonstrated substantially greater potential for reducing obesity and diabetes prevalence compared to crosswalk improvements.

Conclusions:

• GSV data presents robustness challenges, potentially misrepresenting built environment features.
• Ignoring mediators like physical inactivity can lead to biased intervention effect estimates.
• Public health interventions should consider mediating behaviors for more accurate and effective outcomes.