Differential direct effects of PM2.5 constituents on daily all-cause mortality in Delhi, India

Short-term exposure to fine particulate matter (PM2.5) poses a significant environmental risk to human health. The differential direct acute impacts of the chemical constituents of PM2.5 mass remained unexplored in Indian cities due to lack of speciation data at daily scale and a suitable statistical model to account for strong collinearity among the components adjusting nonlinear confounding. We, therefore, developed a semi-parametric quasi- Poisson ridge regression model, to estimate the direct acute effects of strongly collinear PM2.5 constituents on daily all-cause mortality in Delhi, India. The study used four years of data (2013-2016) of daily all-cause mortality, daily masses of PM2.5 components and daily meteorological parameters. A simulation was conducted to determine the benefits of ridge regression over a quasi-likelihood estimate. Differential direct impacts of daily concentration of PM2.5 components for their best lag effects on daily all-cause mortality were estimated for every interquartile range (IQR) increase of individual components. The simulation revealed that ridge regression estimates were more precise when correlation coefficients among the components exceeded by 0.8. The model estimated the highest impact by organic carbon, followed by nitrate and chromium with 3.36% (95% CI:1.99-4.87); 3.29% (95% CI:1.52-4.4) and 2.35% (95% CI:1.12-3.57) increase in mortality per IQR increase in exposure, respectively. Among others, potassium, sodium, lead, sulphate and non-carcinogenic mixture only estimated significant effects. In conclusion, the proposed model offered estimates with high precision and accuracy eliminating indirect effects by highly colinear covariates. As a result, these direct effects of PM2.5 constituents estimated by this study hold potential for understanding the impacts of air pollution on the highly polluted city of Delhi.