The Distributed Lag between Air Pollution and Daily Deaths

TLDR

Air‑pollution studies consistently link PM10 to daily deaths, yet the lag between exposure and mortality is poorly defined, with most effects lasting more than one day. This study aims to establish a systematic method for determining the lag structure between PM10 exposure and mortality to enable cross‑site comparisons and deeper insight into the exposure‑response relationship. The authors fitted unconstrained and constrained distributed‑lag models to daily deaths of persons aged 65+ and PM10 concentrations in ten U.S. cities with continuous monitoring.

Abstract

Many studies have reported associations between air pollution and daily deaths. Those studies have not consistently specified the lag between exposure and response, although most have found associations that persisted for more than 1 day. A systematic approach to specifying the lag association would allow better comparison across sites and give insight into the nature of the relation. To examine this question, I fit unconstrained and constrained distributed lag relations to the association between daily deaths of persons 65 years of age and older with PM10 in 10 U.S. cities (New Haven, Birmingham, Pittsburgh, Canton, Detroit, Chicago, Minneapolis, Colorado Springs, Spokane, and Seattle) that had daily monitoring for PM10. After control for temperature, humidity, barometric pressure, day of the week, and seasonal patterns, I found evidence in each city that the effect of a single day's exposure to PM10 was manifested across several days. Averaging over the 10 cities, the overall effect of an increase in exposure of 10 μg/m3 on a single day was a 1.4% increase in deaths (95% confidence intervals (CI) = 1.15–1.68) using a quadratic distributed lag model, and a 1.3% increase (95% CI = 1.04–1.56) using an unconstrained distributed lag model. In contrast, constraining the model to assume the effect all occurs in one day resulted in an estimate of only 0.65% (95% CI = 0.49–0.81), indicating that this constraint leads to a substantial underestimate of effect. Combining the estimated effect at each day's lag across the 10 cities showed that the effect was spread over several days and did not reach zero until 5 days after the exposure. Given the distribution of sensitivities likely in the general population, this result is biologically plausible. I also found a protective effect of barometric pressure in all 10 locations.