Measurement error correction for ambient PM_2.5 exposure using stratified regression calibration: Effects on all-cause mortality

Background: Previous studies on the impact of measurement error for PM_2.5 were mostly simulation studies, did not control for other pollutants, or used a single regression calibration model to correct for measurement error. However, the relationship between actual and error-prone PM_2.5 concentration may vary by time and region. We aim to correct the measurement error of PM_2.5 predictions using stratified regression calibration and investigate how the measurement error biases the association between PM_2.5 and mortality in the Medicare Cohort. Methods: The “gold-standard” measurements of PM_2.5 were defined as daily monitoring data. We regressed daily monitoring PM_2.5 on modeled PM_2.5 using the simple linear regression by strata of season, elevation, census division and time period. Calibrated PM_2.5 was calculated with stratum-specific calibration parameters β₀ (intercept) and β₁ (slope) for each strata and aggregated to annual level. Associations between calibrated and error-prone annual PM_2.5 and all-cause mortality among Medicare beneficiaries were estimated with Quasi-Poisson regression models. Results: Across 208 strata, the median of β₀ and β₁ were 0.62 (25% 0.0.20, 75% 1.06) and 0.93 (25% 0.87, 75% 0.99). From calibrated and error-prone PM_2.5 data, we estimated that each 10 μg/m³ increase in PM_2.5 was respectively associated with 4.9% (95%CI 4.6–5.2) and 4.6% (95%CI 4.4–4.9) increases in the mortality rate among Medicare beneficiaries, conditional on confounders. Conclusions: Regression calibration parameters of PM_2.5 varied by time and region. Using error-prone measures of PM_2.5 underestimated the association between PM_2.5 and all-cause mortality. Modern exposure models produce relatively small bias.