Background: To assess the public health risk of heat waves and to set criteria for alerts for -excessive heat, various meteorologic metrics and models are used in different jurisdictions, generally without systematic comparisons of alternatives. We report such an analysis for New York City that compared maximum heat index with alternative metrics in models to predict daily variation in warm-season natural-cause mortality from 1997 through 2006.
Materials and methods: We used Poisson time-series generalized linear models and generalized additive models to estimate weather-mortality relationships using various metrics, lag and averaging times, and functional forms and compared model fit.
Results: A model that included cubic functions of maximum heat index on the same and each of the previous 3 days provided the best fit, better than models using maximum, minimum, or average temperature, or spatial synoptic classification (SSC) of weather type. We found that goodness of fit and maximum heat index-mortality functions were similar using parametric and nonparametric models. Same-day maximum heat index was linearly related to mortality risk across its range. The slopes at lags of 1, 2, and 3 days were flat across moderate values but increased sharply between maximum heat index of 95 degrees F and 100 degrees F (35-38 degrees C). SSC or other meteorologic variables added to the maximum heat index model moderately improved goodness of fit, with slightly attenuated maximum heat index-mortality functions.
Conclusions: In New York City, maximum heat index performed similarly to alternative and more complex metrics in estimating mortality risk during hot weather. The linear relationship supports issuing heat alerts in New York City when the heat index is forecast to exceed approximately 95-100 degrees F. Periodic city-specific analyses using recent data are recommended to evaluate public health risks from extreme heat.