Strategies for combining results for NO2 and PM2.5 in health impact assessment

Abstract

OPS 34: Health impact assessments of policies, Room 210, Floor 2, August 28, 2019, 1:30 PM - 3:00 PM Health impact assessment primarily uses single pollutant model results for concentration-response functions (CRFs), partly due to larger numbers of studies available and partly due to effect transfer from the pollutant measured with more error to the pollutant measured with less error in multi-pollutant models, complicating their interpretation. Results using single pollutant models may be presented separately, added ignoring the possibility of overlap in the results, given as a range from no overlap to full overlap or one pollutant may be chosen as an indicator, potentially underestimating the result. The Committee on the Medical Effects of Air Pollutants (COMEAP) compared the combined mutually adjusted hazard ratios per interquartile range (IQR) for NO2 and PM2.5 (range 1.024 to 1.09) with the respective unadjusted hazard ratios within each study – these were similar or a bit greater than the largest single pollutant hazard ratio (range 1.023 to 1.15 per IQR). This bypasses the problem of effect transfer by considering the pollutants together. Application of this to health impact assessment was explored, although the bypassing of effect transfer may not apply to the same extent if applied in contexts where the relative concentration ranges of the two pollutants are different. Effect transfer in multi-pollutant models in time-series studies is influenced not only by the commonly discussed correlations between the pollutants but also by the magnitude and variance of the errors and the correlation between them. Evangelopoulos et al (submitted) estimated typical values for these parameters from a review of the literature and used these to correct multi-pollutant model CRFs. Using European error parameters, this demonstrated a 4% underestimation of the adjusted CRF for PM2.5 and a 31 % underestimation of the adjusted CRF for NO2, as a result of measurement error. Future directions to rationalise approaches in this area will be discussed.