Predicting the risk of heatstroke: Development of a highly accurate model

Abstract Background High temperatures in urban areas caused by global climate change and urban heat island intensification have led to an increase in the number people experiencing heat related illness, the most serious of which is heatstroke. To help prevent heatstroke, an accurate model should be developed that will predict dangerous conditions so that people can take preventive actions. Method The goal of this study was to compare three methods for predicting heatstroke risk: multiple regression analysis (MR), generalized additive model (GAM), and time-stratified case-crossover analysis (TC). Susceptibility to heatstroke is likely to be dependent on year-wise trends and is sensitive to training data, but most previous models have only tested a limited amount of training data. In this study we investigated the optimal number of years to use as training data. By comparing the errors of each method, the error influencing factors in the training data was identified. Results The TC errors were the smallest (p<0.005) and much less sensitive to the training data than others. The MR and GAM errors were significantly larger when the number of extremely hot days differed between the training and test data (p<0.01, p<0.05). All three methods tended to increase in accuracy as more past data years were added to the training data, but to decrease in accuracy after a certain point. The optimal accuracy was obtained by using data from three or four years. Conclusions As a result, a highly accurate risk model that was robust to training data was developed using the odds ratios produced by TC with low sensitivity to training data, something that has not been possible with previous models. This modeling approach is universally applicable and can be used to make urban areas safer in future.