Background The incidence of hospital admission with an alcohol specific diagnosis is used as an indicator of alcohol-related harm in young people. Rates vary greatly and are of particular local concern in the research setting. The study was set in the County of Worcestershire, a Local Authority in central England with a mixed urban and rural population of approximately 566,000 people.

Methods We measured retail outlet proximity for 5000 randomly selected addresses in 361 neighbourhoods using licensing data and geographical information systems (GIS). For each neighbourhood, five-year alcohol specific admission counts in 12–17 year olds were obtained as a response variable. Other predictor variables were: the at-risk population, distance to the closest Emergency Department and socio-economic deprivation measured as a neighbourhood level. A generalised linear model was used to measure the association between the response variable and the predictor variables. We used a variety of software including ArcGIS 10.0 for the spatial analysis and R for the modelling and other visualisations.

The unit of analysis was the neighbourhood. The neighbourhoods were highly heterogeneous in terms of their demographic structure, rurality, retail environment and socio-economic deprivation. We excluded supermarket retailing from this analysis to concentrate on small neighbourhood retailing only.

Results The number of admissions was positively correlated to deprivation and positively associated with retail outlet proximity. The association with outlet proximity is strongly positive (and statistically significant), even after adjusting for deprivation and other demographic and topographical variables. Our model estimated that with all other variables (including deprivation) held at their median, populations with an outlet within 400 m will have a five-year admission rate of 11.02 per 1000, but this rate falls by over 50% (5.33 per 1000) where the nearest outlet is five kilometres away. The gradient was similar at different levels of deprivation.

Discussion The findings are of interest because in England, the Public Health function is co-located in Local Authorities which regulate aspects of the built environment including alcohol retailing. We sought to eliminate sources of confounding and bias in the measurement and modelling, but cannot rule out unmodelled variation or a latent confounder that may be driving the associations that we measured. We present this work as hypothesis generating and would seek to finesse the methodological approaches in a future, larger iteration in a different setting. We also present it as an example of how the new Public Health environment can facilitate research using novel data linkages.