What question did this study set out to answer?

This research aims to analyze the variation in the concentration–response function between PM2.5 exposure and all-cause mortality, considering different exposure assessment methods.

March 10, 2026Open Access

The effect of long-term PM2.5 exposure on all-cause mortality estimated using meta-regression: a meta-analysis of concentration–response by air pollution exposure assessment method

Key Points

This research aims to analyze the variation in the concentration–response function between PM2.5 exposure and all-cause mortality, considering different exposure assessment methods.
Conducted a meta-regression using 68 studies on PM2.5 and mortality.
Categorized exposure assessment methods into monitoring, land-use regression, and dispersion modeling.
Evaluated different parametric forms for concentration-response function fitting using Akaike Information Criterion (AIC).
Employed spline-based modeling to identify deviations from selected parametric forms.
Identified a relative risk of 1.22 for each 10 µg/m3 increase in PM2.5 at a mean exposure of 10 µg/m3.
Found various best-fitting functional forms for concentration-response functions based on assessment methods: linear for monitoring, logarithmic for LUR, and inverse for dispersion and hybrid approaches.
Determined logarithmic fit for low-resolution models and inverse square root for high-resolution models.

Abstract

Fine particulate matter (PM2.5) is a well-known risk factor for premature mortality; however, the relative effect size varies between different studies. The current study aims to estimate the concentration–response function (CFR) between all-cause mortality and PM2.5 while adjusting for differences in population characteristics and exposure assessment method. Separate functions by exposure assessment method by dispersion, land use regression (LUR), and hybrid models as well as monitoring station measurements and by model resolution will be provided. We used the same keywords as those in the review by Chen and Hoek (2020), which included studies up to October 2018. Our updated search covered the period from July 1, 2018, to May 15, 2023, to include more recent publications. A random-effects meta-regression model was developed using the metafor package in R to account for variations in exposure estimates. Exposure assessment methods were categorized into Monitoring, Land-Use Regression (LUR), Dispersion Modeling, and Hybrid approaches, and spatial resolution was classified as high (≤ 1 km) or low (> 1 km). We evaluated linear, logarithmic, inverse, and inverse square root transformations as potential parametric forms of the CRF and selected the one that provided the best fit according to the Akaike Information Criterion (AIC). Additionally, spline-based modeling was employed to test for deviations from this parametric forms in the CFR. The analysis identified 68 eligible studies, incorporating diverse exposure assessment methodologies and geographic regions. In the fully adjusted model, the relative risk (RR) per 10 µg/m3 higher PM2.5 concentration at a mean exposure of 10 µg/m3 was 1.22 (95% CI: 1.02–1.47) and at a mean exposure of 17.85 µg/m3 1.15, (95% CI: 0.97–1.36). In the subgroup analysis, the best-fitting functional form for the CRF varied by exposure assessment method: linear for monitoring, logarithmic for LUR, and inverse for dispersion and hybrid models. For resolution, logarithmic fit best for low-resolution models, and inverse square root for high-resolution models. This study underscores the role of modeling choices in quantifying PM2.5-related health risks. Our analysis offers an updated increased CRF, including the recent evidence, for use in global health risk assessments of particulate air pollution.

اسأل الذكاء الاصطناعي

Bookmark

View Full Paper