Weekly heat wave death prediction model using zero-inflated regression approach

Driven partly by the shifting climate and growth of vulnerable populations, excess heat-triggered human fatalities are becoming a serious public health risk concern in Korea. This study develops the zero-inflated regression model for predicting the numbers of weekly heat deaths in South Korea. Defining the heat death as the death caused by "exposure to excessive natural heat," data analyses are performed to examine statistical relationships between the number of heat deaths versus the pertinent temperature-related parameters and the size of the vulnerable population. The weekly mean of daily maximum temperature and the number of consecutive heat wave days with tropical nights are selected as the temperature-related parameters, while the numbers of elderly living alone and those of agricultural workers are selected as the parameters representing the vulnerable population. Using these four regressive parameters, we develop a regression-based model applied for the prediction of the number of heat deaths. Several statistical methods including the Poisson, negative binomial, hurdle, and the zero-inflated models are scrutinized. The results demonstrate that the zero-inflated Poisson regression model is the most appropriate statistical approach, as it addresses the issue of frequent occurrences of zero-valued observations in weekly heat death data. It was evident that a larger number of heat deaths occurred in the period and study region with a correspondingly higher predicted value. In accordance with this statistical performance, we ascertain that the utilized models could be explored by disaster management and public health experts as a scientific contrivance for health risk mitigation and resource allocation more strategically, and for providing the general public with reliable weekly heat wave risk forecasts.