Assessing Personal PM2.5 Exposure Prediction Improvement After Addition of Indoor PM2.5 Exposure and Personal Characteristics to Outdoor PM2.5 Exposure Measurements

• Activity Number: 452
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Cole Brokamp*+ and M. B. Rao and Patrick Ryan
• Companies: University of Cincinnati and University of Cincinnati and Cincinnati Children's Hospital Medical Center
• Keywords: RIOPA ; random forest ; PM2.5

Abstract

Most epidemiological studies use outdoor PM2.5 concentrations at central monitoring sites as a surrogate of personal exposure. We wanted to test if the addition of indoor PM2.5 measurements or other personal characteristics concerning potential PM2.5 sources increased the prediction accuracy of personal PM2.5. Using data from the Relationship of Indoor Outdoor and Personal Air (RIOPA) study, three models were quantitatively compared using two different predictive methodologies. The first method was linear regression and models were compared using analysis of variance. The second method was random forest and models were compared using root mean square error. Overall, we determined that the addition of indoor PM2.5 measurements to outdoor PM2.5 measurements significantly enhanced the ability to predict personal PM2.5 measurements while the addition of personal characteristics regarding PM2.5 sources did not. Comparing the two model frameworks showed that the random forest model recapitulated the conclusions of the regression framework for most PM2.5 elements while avoiding common problems associated with linear models.