You are here

Investigating the consequences of Measurement Error of gradually more sophisticated long-term personal exposure models in assessing health effects: the LOndon Study (MELONS)

Principal Investigator: 

King's College London, United Kingdom

This study will investigate the consequences of measurement error on estimates of health effects of long-term exposure to outdoor air pollution in London by developing increasingly sophisticated exposure models.The investigators plan to compare exposure models that account for mobility, are based on exposure estimates at the residential address, and are based on concentrations measured at the nearest air pollution monitor.

Funded under

Poster abstract for HEI Annual Conference 2023

Investigating the consequences of Measurement Error of gradually more sophisticated long-term personal exposure models in assessing health effects: The LONdon Study (MELONS)

Klea Katsouyanni1, Dimitris Evangelopoulos1, Dylan Wood1, Hanbin Zhang1, Benjamin Barratt1, Barbara Butland2, Evangelia Samoli3, Sean Beevers1, Heather Walton1, Audrey de Nazelle1, Evangelos Evangelou1, Joel Schwartz4

1Imperial College London, UK; 2St George's University of London, UK; 3National and Kapodistrian University of Athens, Greece; 4Harvard School of Public Health, Boston, USA

Background: The aim of MELONS is to evaluate whether increasingly detailed estimates of long-term individual exposure to PM2.5, NO2, O3 and black carbon (BC), in large-scale studies are useful and effective in yielding better estimates of the health effects of exposure to outdoor air pollution.

Methods: We used personal monitoring from four exposure measurement campaigns in 360 London residents, including COPD patients (COPE), professional drivers (DEMiSt), schoolchildren (BLW) and healthy adults (PASTA). We compiled air pollution and GPS data, performed location tagging and calculated monthly home infiltration efficiency to separate participants’ personal exposure into indoor- and outdoor-generated pollution and extrapolate to annual averages. We collected information from “surrogate” exposures, such as modelled residential outdoor concentrations, measurements from the nearest monitor with and without adjustment for typical time-activity patterns of the London population. By comparing these surrogates with outdoor-generated personal exposure (“true”), we quantified exposure measurement error (ME), its type (classical or Berkson) and potential determinants. These were used as inputs in simulations to assess the consequences of ME on health effect estimation under a Cox model for natural mortality. We also explore the performance of ME correction methods, such as regression calibration and simulation extrapolation. Our findings will be applied to real data from the UK Biobank.

Results: COPD patients spent on average most of their time at home (93%) and only 6% outdoors, while drivers, children and healthy adults spent approximately 55% on average at home. Annually extrapolated personal exposure from outdoor sources (Mean ± SD) were: 5.1±1.6μg/m3 (COPE) and 5.6±3.1μg/m3 (BLW) for NO2; 6.0±2.6μg/m3 (COPE) and 5.0±1.7μg/m3 (BLW) for PM2.5; 5.2±1.6μg/m3 (COPE) for O3; and 1.3±0.9μg/m3 (PASTA) and 1.3±0.7μg/m3 (DEMiSt) for BC. Estimated correlations between “true” and “surrogate” exposure estimates were low (<0.3) for several pollutants and campaigns. Gaseous pollutant exposures were more prone to classical error than particles. Age, time-spent-outdoors and socioeconomic status explain part of the ME variability. Test simulations suggest that the epidemiological estimates can be highly biased towards the null due to ME (up to 27% average bias for only moderate additive ME), but ME correction methods reduce this bias substantially (to within 6% over- or under-estimation of the true association).

Conclusions: Our findings show that separating indoor- and outdoor-generated pollution and estimating long-term averages from short-term campaigns introduces uncertainties in the assessment of a “gold-standard” exposure. The low correlation between “true” and “surrogate” exposures illustrates that ME bias can be substantial in air pollution epidemiological analyses. It is important to understand ME and correct for it to estimate the true impact of air pollution on health.