P5 Exploring contextual predictors and modifiers of associations between the neighbourhood built environment and obesity across the UK

Background Studies of neighbourhood built environments and obesity-related outcomes have produced inconsistent findings in different settings. One explanation for this may be that built environment effects on health are context dependent, and therefore vary geographically. Understanding broader contextual factors that might modify or influence health effects of neighbourhood built environments could help identify conditions under which neighbourhood interventions are more likely to succeed. Using the large, geographically diverse UK Biobank sample and linking it to other area-level data, we examine whether various contextual factors at multiple scales modify our previously observed associations between the neighbourhood physical activity (PA) environment and adiposity, and/or predict geographical heterogeneity of those associations. Methods The UK Biobank cohort comprises approximately 4 00 000 adults aged 40–69, recruited from across the UK between 2006 and 2010 using a clustered sampling design. Linked to each individual is detailed information about their neighbourhood environment, derived from national spatial databases. First, we examine whether cross-sectional associations between the number of formal PA facilities within 1 km of people’s homes, and BMI, are modified by other neighbourhood characteristics (e.g. greenspace) operating at the same scale, by fitting interaction terms between the PA environment and potential modifiers and examining stratum-specific associations. Second, we describe how the main association varies geographically across UK nations, regions and local authorities, then explore how contextual factors at various scales might explain this variation. We apply single and multilevel regression modelling techniques to a dataset we have constructed by mapping the UK Biobank sample and linking it to publicly available data on a range of geodemographic and environmental characteristics of areas. Results While the overall association between the PA environment and BMI is negative, models stratified by other neighbourhood characteristics showed some evidence of effect modification at this scale. The main association also varied geographically at various scales, even after comprehensive adjustment for sociodemographic and other characteristics of individuals. For example, we observed an association of above-average magnitude in Scotland, but below-average in Wales, and strong in Bristol and Glasgow, but null in parts of Yorkshire and the North East. Preliminary results suggest characteristics of the broader areas in which neighbourhoods are located may explain some of this observed variation. Conclusion Associations between neighbourhood PA environments and BMI appear to vary across the UK, at multiple geographical scales. Understanding this heterogeneity may help identify where built environment interventions might be expected to succeed or fail, and what contextual factors might support such interventions.