Combining PM2.5 Component Data from Multiple Sources: Data Consistency and Characteristics Relevant to Epidemiological Analyses of Predicted Long-Term Exposures

BACKGROUND Regulatory monitoring data have been the exposure data resource most commonly applied to studies of the association between long-term PM2.5 components and health. However, data collected for regulatory purposes may not be compatible with epidemiological studies. OBJECTIVES We studied three important features of the PM2.5 component monitoring data to determine whether it would be appropriate to combine all available data from multiple sources for developing spatiotemporal prediction models in the National Particle Component and Toxicity (NPACT) study. METHODS The NPACT monitoring data were collected in an extensive monitoring campaign targeting cohort participant residences. The regulatory monitoring data were obtained from the Chemical Speciation Network (CSN) and the Interagency Monitoring of Protected Visual Environments (IMPROVE). We performed exploratory analyses to examine features that could affect our approach to combining data: comprehensiveness of spatial coverage, comparability of analysis methods, and consistency in sampling protocols. In addition, we considered the viability of developing spatiotemporal prediction models given a) all available data, b) NPACT data only, and c) NPACT data with temporal trends estimated from other pollutants. RESULTS The number of CSN/IMPROVE monitors was limited in all study areas. The different laboratory analysis methods and sampling protocols resulted in incompatible measurements between networks. Given these features we determined that it was preferable to develop our spatiotemporal models using only the NPACT data and under simplifying assumptions. CONCLUSIONS Investigators conducting epidemiological studies of long-term PM2.5 components need to be mindful of the features of the monitoring data and incorporate this understanding into the design of their monitoring campaigns and the development of their exposure prediction models.