Bayesian spatio-temporal analysis of exposure to PM2.5 with joint mortality from lung cancer and stroke in California

Activity: SupervisionMaster Dissertation Supervision


Poor air quality may induce a variety of diseases, endangering human health. This study
focuses on investigating the relationship between exposure to serious air pollution and
the mortality of lung cancer and stroke which varies over time and space across California
from 2018 to 2020, taking into account potential meteorological and socioeconomic
confounders. Based on integrated nested Laplace approximation (INLA) method, a combination
of Matern Covariance Function and stochastic partial differential equation (SPDE)
is used to generate PM2.5 concentrations estimations at the county level according to the
concentration recorded from monitoring stations geographically uneven distributed. Then
the Zero-inflated Poisson models with Bayesian approaches considering fixed effect and
the spatio-temporal random effect are proposed and compared. After selecting the model
with the best performance on model fitting, a joint model of two cause-specific mortality
is defined to identify the similar and different effects of interrelated factors. The results
show that accounting for spatial variation between counties and time trend, the counties
with high PM2.5 concentrations have a higher risk of stroke mortality than expected while
the relationship is not statistically significant for lung cancer. High health insurance rates
are significantly associated with low mortality rates for both diseases whereas the effects
of other covariates on mortality are different. Finally, this paper proposes constructive
suggestions concerning air pollution risk warning and public health policy adjustments at
the state and federal level, which will also be helpful to explore how to reduce the risks of
lung cancer and stroke mortality in the U.S.