Does a lag-structure of temperature confound air pollution-lag-response relation? Simulation and application in 7 major cities, Korea (1998-2013)

Background: Temperature must be controlled when estimating the associations of short-term exposure to air pollution and mortality. Given that multi-country studies have implied temperature has lagged effects, we aim to explore confounding by temperature-lag-response and investigate PM₁₀-lag-mortality relation in 7 cities, Korea.

Methods: In a simulation study, we compared the performance of different methods to control for: the same day temperature, a lagged temperature and distributed lags of temperature. In a real data study, we explored PM₁₀-lag-mortality relation in 7 cities using these different methods.

Results: We confirmed that a model with insufficient control of temperature offers a biased estimate of PM₁₀ risk. The degree of bias was from -82% to 95% in simulation settings. A real data study shows estimates among different models by temperature adjustments and PM₁₀ lag variables ranging from -0.3% to 0.4% increase in the risk of all-cause mortality, with a 10μg/m³ increase in PM₁₀. Controlling for temperature as distributed lags for 21 days provided 0.25% (95% CI: 0.1, 0.4) increase in the risk of all-cause mortality.

Conclusions: A lag structure of temperature can confound the air pollution-lag-response relation. Temperature-lag-response relation should be evaluated when estimating air pollution-lag-response relation. As a corollary, air pollution and temperature risk in mortality can be estimated using the same regression model.