Machine learning approaches to characterize the obesogenic urban exposome

Area of Science:

• Environmental Health
• Epidemiology
• Urban Planning

Background:

• Urban environments present complex drivers of obesity.
• Simultaneous analysis of multiple environmental factors influencing obesity is under-researched.

Purpose of the Study:

• To investigate the relationship between urban exposome factors and body mass index (BMI).
• To assess the consistency of findings across various statistical methodologies.

Main Methods:

• Cross-sectional analysis of 14,829 participants from the Occupational and Environmental Health Cohort.
• Estimation of 86 geocoded environmental exposures including air pollution, noise, green space, and built environment characteristics.
• Application of six statistical approaches (e.g., sparse group Partial Least Squares, Bayesian Model Averaging, Extreme Gradient Boosting) to identify exposure-obesity associations.

Main Results:

• Consistent associations found between BMI and neighborhood socioeconomic factors (income, home value), oxidative potential of particulate matter air pollution (OP), availability of healthy food outlets, and household composition.
• Higher BMI was associated with low-income neighborhoods, lower average home values, fewer one-person households, fewer healthy food retailers, and higher OP levels.
• Model performance varied based on the ability to capture linear or nonlinear associations.

Conclusions:

• The study reinforces the impact of neighborhood socioeconomic status, urbanicity, and air pollution on obesity.
• Pluralistic analysis of environmental obesogens provides robust evidence for these associations.