Using spatio-temporal lagged association pattern to unravel the acute effect of air pollution on mortality

Acute mortality effects of air pollution have been recognized in plenty of environmental epidemiologic studies. However, existing studies usually assume a universal lag association across sites and seasons. Such a strategy ignores the heterogeneity of lag structures and may lead to bias in the estimation of effects.A Bayesian hierarchical model with flexible lag structures was applied to estimate the impact of particulate matter less than 10 μm (PM₁₀) on mortality and determine whether the lag structure varied by season and location. Data from nine US communities, obtained from the National Morbidity, Mortality, and Air Pollution Study (NMMAPS), was used to examine the lagged associations between PM₁₀ and daily mortality. The estimates obtained from the flexible lag approaches were compared with those from the universal lag approach.Of potential varying lag structures, a 10-μg/m³ increase in PM₁₀ was associated with 0.32% (95% credible interval: 0.16, 0.45) and 0.36% (0.18, 0.52) increases in mortality from nonaccidental and cardiovascular-respiratory death. The community-specific estimates of PM₁₀ mortality effects were distinct between the flexible and the universal lag approaches, with relative change of the effects ranged from −7.21% to 9.25% for nonaccidental morality, and from −5.78% to 4.16% for cardiovascular-respiratory morality. Moreover, the lag structure varied by location and season. For instance, the nonaccidental mortality effect of PM₁₀ attributable to the current and previous day was 29.8% in El Paso while 55.0% in Chicago; the overall effect attributable to the previous two to five days were 60.6%, 51.9%, 59.5%, and 59.3% in winter, spring, summer, and fall, respectively.The results indicated that a universal lag association across sites and seasons may bias the mortality effect of air pollution. The varying lag structures should be considered in studies of short-term environmental exposures to get a more precise effect estimate.