Clarifying causality and information flows between time series: Particulate air pollution, temperature, and elderly mortality.

Abstract

Exposure-response associations between fine particulate matter (PM2.5) and mortality have been extensively studied but potential confounding by daily minimum and maximum temperatures in the weeks preceding death has not been carefully investigated. This paper seeks to close that gap by using lagged partial dependence plots (PDPs), sorted by importance, to quantify how mortality risk depends on lagged values of PM2.5, daily minimum and maximum temperatures and other variables in a dataset from the Los Angeles air basin (SCAQMD). We find that daily minimum and maximum temperatures and daily mortality counts two to three weeks ago are important independent predictors of both current daily elderly mortality and current PM2.5 levels. Thus, it is important to control for these variables over a period of at least several weeks preceding death. Such detailed control for lagged confounders has not been performed in influential past papers on PM2.5-mortality associations, but appears to be essential for isolating the potential causal contributions of specific variables to mortality risk, and, therefore, a worthwhile area for future research and risk assessment modeling.