A0642
Title: Confounding adjustment with spatiotemporal data
Authors: Carlo Zaccardi - University of Chieti \& Pescara (Italy) [presenting]
Pasquale Valentini - University G. d Annunzio of Chieti-Pescara (Italy)
Luigi Ippoliti - University G.d'Annunzio Chieti-Pescara (Italy)
Abstract: In the context of epidemiological data with a spatiotemporal structure, the association between exposure and outcome is of interest. Besides, spatiotemporal confounders may exist, i.e. factors that influence the outcome and have spatial and temporal trends similar to those of the exposure. The problem known as spatiotemporal confounding arises when these confounders are not measured. For instance, if the association between exposure to fine particulate matter and mortality is of interest, weather variables and influenza outbreaks are examples of confounding factors, with the difference that the first ones are measured but the second ones are not. However, not accounting for these unknown factors leads to distorted conclusions about the exposure's effect; indeed, confounding is considered a major challenge in epidemiology. Also, the shape of the association is generally not known and may vary with space and time. Henceforth, to recover the association of interest, the statistical model must account for its possible non-linearity and for spatiotemporal confounding as well. To this end, a Bayesian approach is proposed where the regression coefficients are allowed to vary with space and time. The proposed method can capture potential non-linearities and alleviate the spatiotemporal confounding bias.
