Associations of neighbourhood food retail with disability and death in older adults: Cardiovascular Health Study.

Background: A healthier diet is associated with lower chronic disease burden, but the impact of neighbourhood food environments on disability and death in older adults is not known.

Methods: In the Cardiovascular Health Study, a cohort study of adults aged 65+, we calculated study years until death (years of life (YOL)), study years without activities of daily living (ADL) difficulty (years of able life; YoAL) and percent of study years without ADL difficulty (compression of disability). Linear regression quantified associations of food establishments within 5 km of baseline home address (as a z-score) with each outcome, adjusted for sociodemographic characteristics. Sensitivity analyses considered adjustment for risk factors and comorbidities, multiple imputation, alternate neighbourhood definitions (1-km radial buffer, census tract) and restriction on residential stability.

Results: We included 4298 participants followed for up to 26 years. All food retail establishments were associated with 6 months higher YoAL per SD in the main model (beta, 0.50 years; 95% CI 0.01, 0.98; p=0.046), with similar findings across sensitivity analyses except when restricting on residential stability. Supermarkets and produce markets were associated with compression of disability (beta, 2.31; 95% CI, 0.04, 4.57) and when using 1-km buffers with YOL (beta, 0.23 years; 95% CI 0.03, 0.43) and YoAL (beta, 0.21 years; 95% CI 0.01, 0.41). Non-supermarket food stores were associated with YoAL (beta, 0.67 years; 95% CI, 0.07, 1.27) and compression of disability (beta, 3.03; 95% CI 0.44, 5.62), but significance was not consistent across sensitivity analyses. Fast-food restaurants did not reach statistical significance in any model.

Conclusion: All food retail was associated with YOL without impairment. Neighbourhood food retail access and type may both have roles in extending YOL and years of able life among older adults, but the findings were sensitive to decisions made during measurement and modelling.