Ambient temperature and emergency department visits: Time-series analysis in 12 Chinese cities

The association between ambient temperature and mortality has been well documented worldwide. However, limited data are available on nonfatal health outcomes, such as emergency department visits (EDVs), particularly from China.

To examine the temperature-EDV association in 12 Chinese cities; and to assess the modification effects by region, gender and age.

Daily meteorological data and non-accidental EDVs were collected during 2011-2014. Poisson regression with distributed lag non-linear model was applied to examine the temperature-lag-EDV association in each city. The effect estimates were pooled using multivariate meta-analysis at the national and regional level. Stratified analyses were performed by gender and age-groups. Sensitivity analyses adjusting for air pollution and relative humidity were conducted.

A total of 4,443,127 EDVs were collected from the 12 cities. Both cold and hot temperatures were associated with increased risk of EDVs, with minimum-mortality temperature located at 64th percentile of temperature. The effect of cold temperature appeared on day 2 and persisted until day 30, causing a cumulative relative risk (RR) of 1.80 (1.54, 2.11). The effect of hot temperature appeared immediately and lasted until day 3, with a cumulative RR of 1.15 (1.03, 1.29). The effect of temperature on EDVs was similar in male and female but was attenuated with increasing age. The effect of cold temperature on EDVs was greater in southern areas of the country whereas the hot effect was greater in northern cities. The association was robust to a large range of sensitivity analyses.

In China, there is a U-shaped association between temperature and risk of EDVs that is independent of air pollution and humidity. The temperature-EDV association varies with latitude and age-groups but is not affected by gender. Forecasting models for hospital emergency departments may be improved if temperature is included as an independent predictor.