Measurement error correction methods for the effects of ambient air pollution on mortality and morbidity using the UK...

Epidemiological cohort studies associating long-term exposure to ambient air pollution with health outcomes most often do not account for individually assigned exposure measurement error. Here, we implemented Cox proportional hazards models to explore the relationships between NO₂, PM_2.5 and ozone exposures with the incidence of natural-cause mortality and several morbidity outcomes in 61,797 London-dwelling respondents of the UK Biobank cohort. Data from an existing personal monitoring campaign was used as an external validation dataset to estimate measurement error structures between "true" personal exposure and several surrogate (measured and modelled) estimates of assigned exposure, allowing for the application of two health effect estimate correction methodologies: regression calibration (RCAL) and simulation extrapolation (SIMEX). Uncorrected hazard ratios (HRs) suggested an increase in the risk of natural-cause mortality for modelled NO₂ estimates (HR: 1.028 [0.983, 1.074] per IQR increment of 14.54 μg/m³) and no statistically significant association was observed for PM_2.5 surrogate exposure measures. Measurement error corrected HRs were generally larger in magnitude, although exhibited wider confidence intervals than uncorrected effect estimates. Chronic obstructive pulmonary disease (COPD) was associated with increased exposure to modelled NO₂ (1.087 [1.022, 1.155]). Both RCAL and SIMEX correction resulted in increased HRs (1.254 [1.061, 1.482] and 1.192 [1.093, 1.301], respectively). SIMEX correction of modelled PM_2.5 (IQR: 1.72 μg/m³) associations with COPD increased the HR (1.079 [1.001, 1.164]) in comparison to uncorrected (1.042 [0.988, 1.099]). These findings suggest that health effect estimates not corrected for exposure measurement error may lead to underestimation in the magnitude of effects.