Cardiometabolic & kidneyCross-sectional study
Deep Learning–Based Assessment of Built Environment From Satellite Images and Cardiometabolic Disease Prevalence
- Cross-sectional study
- machine learning (convolutional neural networks, light gradient-boosting machines, activation maps)
- census tract-level
- Analyses February to December 2022
Authors' conclusion
In this cross-sectional study, a significant portion of cardiometabolic disease prevalence was associated with GSI-based built environment using convolutional neural networks.
Study context
- Population
- Census tracts in 7 US cities; CDC 500 Cities project data (BRFSS, adults ≥18 years)
- Participants
- 789 census tracts
- Outcome
- Census tract-level estimated prevalence of CHD, stroke and CKD based on image-based built environment features
Main finding
As reported in the abstract
Association of GSI-based built environment features with prevalence
Coronary heart disease (CHD), R2
0.60
Stroke, R2
0.65
Chronic kidney disease (CKD), R2
0.64
Observational data: associations, not effects.
Further findings
CKD: within-city vs between-city performance
- Highest within-city R2
- 0.39
- Between-city R2
- 0.64
CHD: model with Google satellite images (GSI) features vs model with age, sex, race, income, education and social determinants of health indices only
- With GSI features, R2
- 0.83
- Without GSI features, R2
- 0.76
- P
- P < 0.001
Activation maps
- Revealed roads, highways, railroads and recreational facilities (amusement parks, arenas, baseball parks)
Original publication
Deep Learning–Based Assessment of Built Environment From Satellite Images and Cardiometabolic Disease Prevalence. JAMA Cardiol. Published 1 June 2024. doi:10.1001/jamacardio.2024.0749
Summarised from the published abstract; every number is checked against it