Identification of neighborhood clusters on data balanced by a poset-based approach

The identification of territorial clusters where the population suffers from worse health conditions is an important topic in social epidemiology, in order to identify health inequalities in cities and provide health policy interventions. This objective is particularly challenging because of the mechanism of self-selection of individuals into neighborhoods, which causes selection bias. The aim of this paper consists in the identification of neighborhood clusters where elderly people living in Turin, a city in north-western Italy, are exposed to an increased risk of hospitalized fractures. The study is based on administrative data and is a retrospective, observational cohort study. It is composed by a first phase, in which the individual confounding variables are balanced across neighborhoods in order to make them comparable, and a second phase in which the neighborhoods are aggregated into clusters characterized by significantly higher health risk. In the first phase we exploited a balancing technique based on partially ordered sets (poset), called Matching on poset based Average Rank for Multiple Treatments (MARMoT). On the balanced dataset, we used a spatial scan to identify the presence of clusters and we checked whether the risk of fracture is significantly higher in some contiguous areas. The combination of both MARMoT procedure and spatial scan makes it possible to highlight two clusters of neighborhoods in Turin where the risk of incurring hospitalized fractures for elderly people is significantly higher than the mean. These results could have important implications for the implementation of health policies.