Greenspace and depression incidence in the US-based nationwide Nurses' Health Study II: A deep learning analysis of street-view imagery
Li Yi, Jaime E. Hart, Charlotte Roscoe, Unnati Mehta, Marcia Pescador Jimenez, Pi‐I D. Lin, Esra Süel, Perry Hystad, Steve Hankey, Wenwen Zhang, Olivia I. Okereke, Francine Laden, Peter James
Abstract
BACKGROUND: Greenspace exposure is associated with lower depression risk. However, most studies have measured greenspace exposure using satellite-based vegetation indices, leading to potential exposure misclassification and limited policy relevance. We examined the association of street-view greenspace measures with incident depression in a prospective cohort of US women. METHODS: We applied deep learning segmentation models to 350 million US street-view images nationwide (2007-2020) to derive ground-level greenspace metrics, including percentage of trees, grass, and other greenspace (plants/flowers/fields), and linked metrics to Nurses' Health Study II participants' residences (N = 33,490) within 500 m each year. Cox proportional hazards models estimated the relationship between street-view greenspace metrics and incident depression, assessed through self-report of clinician-diagnosed depression or regular antidepressant use and adjusted for individual- and area-level factors. FINDINGS: ) levels (HR per IQR, 0.79; 95%CI: 0.71-0.86). Results were consistent after adjusting for additional spatial and behavioral factors, and persisted after adjusting for traditional satellite-based vegetation indices. CONCLUSION AND RELEVANCE: We observed participants who lived in areas with more trees visible in street-view images had a lower risk of depression. Our findings suggest tree-planting interventions may reduce depression risk.