Background Individual-level data of health outcomes and their drivers are rarely available at a neighbourhood-level. While national-level surveys may provide insights at a regional scale, these data do not help us understand how complex problems such as poor diet affect communities within cities where policies can be leveraged for better health outcomes. Spatial microsimulation (SMS) can be used to approximate individual-level data for small areas. The aim of our study is to estimate small area fruit and vegetable intake for Liverpool (UK) to explore geographical inequalities in diet.

Methods Spatial microsimulation leverages individual level data and fits it to distinct spatial data. The fitting is based on selected constraint variables contained within both data sets. We perform the SMS using individual-level data extracted from the National Dietary and Nutritional Survey (2014–2018, n=2018) and aggregated data from 2011 UK Census at Lower Super Output Area (LSOA). Our outcome variable is daily servings of fruits and vegetables consumed. Constraint variables include age, sex, highest level of qualifications, self-reported health, and national statistics-socioeconomic classification. Constraints were chosen based on literature of key determinants of dietary predictors existing across both data sets. An iterative proportional fitting approach is executed in R to estimate fruit and vegetable intake for each LSOA. We use internal validation via model fit statistics and external validation through comparing estimates to an Eating Habits Survey (n=1724) with partial representation of Liverpool.

Results We estimated that 25% of adults aged 16 and up in Liverpool meet their daily recommendation of fruit and vegetable intake (5 or more servings), with 7% consuming fewer than 1 serving a day. Fruit and vegetable consumption was unevenly distributed across the city, with consumption of 5 or more servings ranging from 18% to 37%. There were higher levels of fruit and vegetable consumption among populations in the least deprived neighbourhoods with 35% meeting daily recommendations (versus 22% in most deprived). Preliminary model validation appears to suggest estimates are suitable for use.

Conclusion Our study presents a novel framework for estimating fruit and vegetable consumption for small areas and within cities. We find low levels of fruit and vegetable consumption in Liverpool, with wide geographical inequalities in consumption by level of deprivation. These results were largely supported by internal/external validation. Producing small area statistics can support better nuanced decision-making including geographical targeting of interventions, especially for local authorities, as well as provide robust inputs for other modelling methods (e.g. agent-based modelling).