Source-apportioned PM2.5 incorporating gas-phase partitioned semi-volatile organic compounds and hospitalizations in Denver

Abstract

Introduction: Identifying emission sources underlying associations between PM2.5 and adverse health effects is important for guiding air quality management decisions. The Denver Aerosol Sources and Health study aimed to improve PM2.5 source apportionment by including semi-volatile organic compounds (SVOC). Associations between sources and daily hospitalizations were compared using source estimates from PM2.5 components including measured particle-phase SVOC only and estimated total SVOC. Methods: Daily concentrations of PM2.5 mass, metals, ions, elemental carbon, and speciated organic compounds were measured in Denver, CO in 2003-2005. Gas-phase SVOC were estimated using gas-particle partitioning theory to obtain total SVOC. Two sets of PM2.5 chemical source groups were estimated using positive matrix factorization for particle-phase and total SVOC. Six source groups were identified for each: inorganic ions, n-alkanes, steranes, light SVOC, polycyclic aromatic hydrocarbons, and bulk carbon. Daily counts of cardiovascular and respiratory hospitalizations were obtained for the 5-county Denver area. Using generalized additive models, relative risks (RR) of hospitalizations were estimated for total and particle SVOC based sources for day lags 0 to 14, adjusting for meteorology, day of week and long-term trends. Constrained distributed lag models were used to assess lag patterns of RRs. Results: RRs and lag patterns of cardiovascular hospitalization differed for light SVOC and bulk carbon source groups when using particle-phase or total SVOC-based source estimates, but were similar for other source groups. Lag patterns for respiratory hospitalization RRs were generally similar, but some RRs differed, such as those at early lags for light SVOC and bulk carbon source groups. Conclusions: PM2.5 source apportionment using speciated data incorporating total SVOC may better characterize PM2.5 sources and improve identification of health impacts from specific sources.