B1094
Title: A novel approach for spatiotemporal confounding bias reduction
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 epidemiological studies, the association between exposure and outcome is of fundamental interest. It is possible, though, for one or more covariates, associated with both exposure and outcome variables, to be unavailable, leading to the presence of spatiotemporal confounding bias in the estimation procedure and, as a result, making it impossible to recover the desired association. Besides, the unknown functional form of this association may be non-linear and/or time-varying, so the commonly used linear regression models are inadequate. A time-varying coefficients regression model is proposed that is able to capture both potential non-linearities and interaction effects of the exposure with other (measured or unmeasured) variables. A simulation study is built to assess how well the proposed approach performs (in terms of confounding bias reduction) in comparison to a model that does not adjust for confounding. The results indicate that the proposal performs better than the unadjusted model in all of the scenarios considered. Finally, the short-term association between fine particulate matter (PM) concentrations and all-cause mortality counts are evaluated in the 117 health districts of two contiguous Italian regions (Piemonte and Lombardia): there is evidence of a non-linear exposure effect and a possible interaction between PM concentrations and air temperature.
