Overview of statistical methods for estimating the relative risk of delayed respiratory effect related to ambient air pollution exposure.

Abstract

The occurrence of smog episodes and their significant impact on human health have forced research focused on risk assessment. Over the years, methods of exposure measuring have been improved, as well as statistical models necessary to the biological response estimation including the risk of incidence or death. The aim of presented study is to review and evaluate possibilities of statistical methods of delayed respiratory health effects risk assessment related to ambient air pollution exposure. The review of published data was carried using the PubMed platform from 1994 to the 2020 year. Over 80 references were include in the analysis identifying general characteristics, construction of models estimating the relative risk of respiratory incidents with delayed health effect, and modelling tools available in statistical packages R, SAS, and Statistica. Among various methods of health risk assessment, the Almon model, the Poisson model, and the Distributed Lag Non-Linear Models (DLNM) were most common used. Initially, the Poisson model was used, close to 60% of the cited works apply this method. The interest in the nonlinear modelling implementation has increased (34% of cited papers) in recent years. Mostly researchers used R or SAS statistical software. Usually, was calculated the relative risk of health effect related to short-term exposure (up to a week). About 75% of available papers concern measurements of relative risk in response to the concentration of pollution increase by unit=10 μg/m3. Other describe the risk associated with the exposure increasing by the interquartile range (IQR). Distributed Lag Non-linear Model DLNM is classified as the statistical tool recommended by researchers due to its flexibility in defining, simplicity in interpretation, and increasingly frequent applications to environmental epidemiology.