Bayesian spatio-temporal exposure modelling of air pollution in Rome, Italy, and short-term effects on cause-specific mortality

Assessing fine-scale spatio-temporal air pollution contrasts in urban contexts is a major challenge for environmental epidemiology. We propose a Bayesian spatio-temporal model to predict daily concentrations of NO2, PM10, and PM2.5 in Rome (Italy), over 2011-2022, on a fine grid scale (1 km), using data from 8 to 13 monitoring stations, depending on pollutant. The model includes meteorological and temporal (working day/weekend) fixed effects together with a lag-1 autoregressive spatio-temporal random effect aimed at capturing spatial and daily dependence. Predictive performance was assessed by leave-one-site-out cross-validation. Estimated exposures were then linked to geolocated cause-specific mortality data for Rome (2012-2019), and a case-crossover time-stratified approach was adopted to investigate acute effects. Cross-validation showed good overall predictive performance, with larger errors at high-traffic sites. Estimated exposure (mean lag 0-5) was positively associated with natural-cause mortality, with percent increase in risk of 1.3 (95% CI: 0.6-2.0) for PM10, 2.1 for NO2 (1.4-2.9), and 2.4 for PM2.5 (1.4-3.3) per 10 μ g/m^3 increase. Corresponding estimates when using the city-specific daily average exposure, instead of our 1 km resolution model, were of comparable magnitude. The proposed Bayesian spatio-temporal framework provides reliable fine-scale exposure estimates for epidemiological use, with results consistent across independent exposure estimates, supporting its application in urban air-pollution health studies.