Background: This study aimed to perform an Exposome-Wide-Association study of body mass index (BMI) in a multi-cohort setting.
Methods: This cross-sectional study analysed data from 15 cohorts (303,660 adults, 102,240 men/ 201,420 women) individually. Studies were affiliated with the Dutch Geoscience and Health Cohort Consortium (GECCO), had different population sizes (688-141,825) and covered the entire Netherlands. Ten studies contained general population samples, others focused on specific patient groups. Study outcomes were measured or self-reported BMI, depending on the study. Associations with 69 environmental factors (air pollution, noise, temperature, neighborhood socio-economic and demographic factors, food environment, drivability, walkability) were explored. Multivariable random forest (RF) regression addressed potential nonlinear and non-additive associations. In the absence of formal methods for multi-model inference for RF, a rank aggregation based meta-analytic strategy was used to summarize the results across the studies.
Findings: Six exposures were associated with BMI: four indicating neighborhood economic or social environment (home values, percentage of high-income residents, average income, livability score), and two indicating the physical activity environment (walkability and job accessibility by road). Lower neighborhood home values were associated with higher BMI scores, but up to 300K€. The directions of associations of walkability and job accessibility with BMI were less consistent than those of the economic or social environment.
Interpretation: Neighborhood social, economic and physical environments are the strongest predictors of high BMI. Rank aggregation allowed to flexibly combine the results from various studies, however, between-study heterogeneity could not be estimated based on RF models.
Funding: This work was supported by GECCO and Exposome-NL projects.
Keywords: Built environment, Socio-economic factors (non-chemical stressors), Obesity and metabolic disorders, Modeling, Meta-analytical approach for machine learning models