Regional variability and sources of error in mortality risk of total and bushfire-specific PM<sub>2.5</sub> — The Association Specialists
  1. Schedule
  2. Oral Presentations - Climate, environment and human health 1
  3. Regional variability and sources of error in mortality risk of total and bushfire-specific PM2.5

Regional variability and sources of error in mortality risk of total and bushfire-specific PM2.5 (23946)

Joseph Van Buskirk 1 , Ivan Hanigan 1 2 3 4 , Edward Jegasothy 1 , Richard Broome 4 , Geoff Morgan 1 5 6
  1. University of Sydney, University Of Sydney, NSW, Australia
  2. The Centre for Air Pollution, Energy and Health Research, Glebe, NSW, Australia
  3. WHO Collaborating Centre for Environmental Health Impact Assessment, School of Population Health, Faculty of Health Sciences, Curtin University, Bentley, WA, Australia
  4. Central Coast Local Health District, NSW Health, Gosford, NSW, Australia
  5. Healthy Environments and Lives (HEAL) National Research Network, HEAL, Australia, NSW
  6. University Centre for Rural Health`, University of Sydney, Lismore, NSW, Australia

Background

Bushfire-specific PM2.5 exposure has been shown to increase the risk of all-cause mortality in Australia. However, potential pitfalls exist when quantifying this relationship that may introduce bias and error. For example,

  1. Modelled PM5 is more valid in areas where observed data (i.e. government monitoring sites) is available, often in more populated urban centres
  2. Aggregation using meta-regression may bias national estimates towards the most populous areas
  3. Areas are subject to seasonal variability, and differences in sources of PM5, which may affect the validity of separating bushfire-specific exposure from total PM2.5
  4. Models are often sensitive to parameterisation.

 The current study aimed to explore the impact of such differences on national, and sub-national, estimates of risk.

Methods

Newly developed high resolution (5km² grid) 20-year Australia-wide daily PM2.5 surfaces were decomposed into daily seasonal, trend and remainder components. Bushfire days were identified using a combination of environmental exposure flags. Regression models utilising an underlying quasi-Poisson distribution were fitted. Various sensitivity analyses were conducted, varying the parameterisation of models, the level of smoothing applied to the exposure surfaces, as well as restricting analysis to areas proximal to monitoring stations.

Results

National estimates showed a positive, exposure-dependent risk relationship between PM2.5 and all-cause mortality. Greater Sydney and Melbourne, the two most populous cities, often reflected national estimates. However, other sub-national estimates were considerably more variable and showed greater sensitivity to model specification and rate of smoothing. Areas with greater temperature fluctuation, differential seasonal patterns and lower density of monitoring stations, showed especially high sensitivity.

Discussion

Sub-national results showed sensitivity to exposure definition and model specification. This persisted after adjusting for population size and estimates of socioeconomic status by area. Caution is advised when applying national PM2.5 and mortality risk estimates to sub-national regions.