INLA-SPDE Models to Predict the Spatial Distribution of Beta-Hexachlorocycloexane Haematic Levels in a Heavily Polluted Area

IntroductionThe Sacco River Valley has been characterized over the years by the presence of several industrial plants. In 2006 it was discovered that livestock were contaminated by beta-hexachlorocycloexane (β-HCH) and two biomonitoring campaigns were performed to assess the human burden. The aim of the study is to estimate the spatial distribution of β-HCH hematic levels in the study area as to use it in future studies on the whole resident population.Data and MethodsBetween 2010 and 2015 740 subjects were randomly sampled from the resident population and blood samples were collected. Every subject was georeferenced by residence address and considered as a “monitoring site”. An Integrated Nested Laplace Approximation (INLA) model combined with Stochastic Partial Differential Equation (SPDE) was run to predict the β-HCH levels. This approach allows Bayesian inference when a spatial process defined over a continuous domain is involved. In the model age, sex, latitude, longitude, altitude and distance from the Sacco river (considered as the main driver of contamination) were also fitted as predictors.ResultsAfter excluding subjects spatially isolated, the model was run on 703 observations. A significant relationship was observed between age and the log-transformed β-HCH (posterior mean for 1 year increase: 0.047), higher levels were estimated in women. A negative relationship was found for distance from the river (-0.020 for 100m increase in the distance), while the other predictors had no effect. We validated the model by running a 10-fold Cross-Validation, with a correlation index between observed and predicted equal to 0.77.ConclusionsThis is the first study using blood sample measures to estimate the spatial distribution of β-HCH. Although strong assumptions were made in order to consider subjects as monitoring sites, this kind of analysis would allow to estimate the effect of β-HCH not only on this sampled population, but also on the cohort of all residents.