Poisson regression was fit to daily data, including all-cause mortality counts, apparent temperatures (AT) and heat waves calculated based on temperature and dew point observations from the Detroit metropolitan airport (DTW), and temperature and dew point predictions from six different forecast products, from May 1, 2002 to December 31, 2006. We estimated effects of heat and heat waves on mortality controlling for daily concentrations of O3, PM10 or both.
On average, days with observed AT of 25.3 oC were associated with 3.5% excess mortality (95% confidence interval (CI): -1.6%, 8.8%), compared to days where AT was 8.5 oC, controlling for heat waves lasting for two and more days. Heat associations calculated from forecasts generally were higher than the associations with observed weather, and differences depended on forecast products and forecast timeframe. Heat waves lasting for two days were associated with a 6.2% increase in mortality risk (95% CI: -0.4%, 13.2%) compared to non-heat wave days. Most of the forecast products had lower estimated heat waves effects than those calculated from observations. Heat wave associations were more robust to inclusion of O3, PM10 or both, in the models, compared to associations with continuous AT.
Heat- and heat-wave-mortality associations differ when calculated using weather forecasts versus observations. These differences vary by forecast product and forecast timeframe. Forecasts generally overestimated heat effects but underestimated heat wave effects. A local operational product (FOUS5) appears to be the best choice from the perspective of heat-mortality associations. These findings have important implications for public health officials, local meteorologists and researchers in designing HHWSs and potentially conducting health impact or risk assessment to heat and heat waves when using forecasts or similar data.