The Impact of Source Contribution Uncertainty on the Effect of Source-Specific PM2.5 on Cardiovascular Hospital Admissions

摘要

Background and aims: Numerous studies have linked particulate sources to a range of health outcomes. They, however, do not account for the uncertainty in the prediction of the source contributions in the health models. Our goal was to estimate the effect of particulate sources on cardiovascular (CVD) health, accounting for the uncertainty in source identification. Methods: We examined the effects of PM2.5 sources on emergency CVD hospital admissions among Medicare enrollees in Boston, MA (2003-2010) using case-crossover analyses. We identified these sources using two methods. Positive Matrix Factorization (PMF) and Absolute Principal Components Analysis (APCA). We propagated uncertainty in the prediction of source contributions employing a block bootstrap procedure. We further estimated average effect estimates across-methods for each source type using the bootstrap samples. Results: We identified six factors: regional, mobile and crustal sources, residual oil combustion, road dust and sea salt. Accounting for uncertainty, same day exposures to regional sources were associated with a 2.01% (0.21, 3.69%) and 1.96% (-0.07, 4.19%) increase in the odds of CVD admissions for PMF and APCA, respectively. Weekly exposures to residual oil combustion were associated with a 2.20% (0.25, 4.30%) and 2.04% (0.08, 4.15%) increase in odds of admissions for PMF and APCA, respectively. Same day and 2-d exposures to traffic-related PM2.5 were also associated with increased CVD admissions. Confidence intervals were wider than would have been if the uncertainty in contribution predictions had not been considered. Further, the agreement in results for PMF and APCA was stronger across factors when uncertainty was considered in the health models. Conclusions: Accounting for uncertainty in source contributions leads to more stable effect estimates across methods and can potentially lead to fewer spurious significant associations and more consistent findings across studies.