Ambient particulate matter and health effects: publication bias in studies of short-term associations

Abstract

Time-series studies have indicated short-term associations between ambient particulate air pollution and adverse health effects, but the extent of publication bias in this literature is not well understood. An analysis of effect estimates from studies published up to January 2002, using funnel plot asymmetry and comparison between single-city and multicity studies, investigated this bias. Evidence of publication bias was found among single-city studies for daily mortality, hospital admissions for chronic obstructive lung disease (COPD), and cough symptom incidence, though not for lung function. Statistical correction for this bias reduced summary relative risk estimates for a 10 μg/m³ PM10 increment: for daily mortality from 1.006 to 1.005, for COPD admissions from 1.013 to 1.011, and the odds ratio for cough from 1.025 to 1.015. Furthermore, analysis of a large multicity study suggested that selective reporting of positive estimates from various lags could inflate summary estimates for PM10 and daily mortality by up to 130%. While publication bias is present in single-city time-series studies, statistically corrected associations between particles and adverse health effects remain positive and precisely estimated, though differential selection of positive lags may also inflate estimates. – AI-generated abstract.