Climate change has augmented heat-related illnesses and deaths. Selecting proper thresholds in a heat-warning system is critical to reducing health risks.
This work evaluates the applicability of a modified generalized additive model (GAM) at different spatial scales, and identifies proper heat-warning thresholds using empirical morbidity and mortality records of 18 years from a population database and taking into account substantial health risk increases of the lag effects of 0–2 days. Reference-adjusted risk ratio (RaRR), i.e., risk ratio of threshold candidates against that of a reference, was respectively evaluated for heat-related emergency and hospital visits and all-cause mortality. The threshold with the highest RaRR among the candidates with infrequent occurrence is potentially the best one. Wet-bulb globe temperature (WBGT) and temperature were both used heat indicator in the model for comparison.
It was found that WBGT is a more sensitive heat-health indicator than temperature. The highest RaRR with WBGT for the whole Taiwan island was observed to shift from lag 0 in emergency visits (1.44) to lags 0–1 in hospital visits (1.18) and also to lag 1 in all-cause mortality (1.04). For different age groups, children had the highest RaRR with WBGT of emergency visits on lag 0 (1.87) while the elderly had the highest RaRR for all-cause mortality on lag 0 (1.04), for hospital visits on lag 1 (1.23), and for emergency visits on lag 2 (1.38), respectively. Emergency visit is the most sensitive heat-related health record and should thus be employed, if available, to select heat-warning threshold. With the highest RaRR in emergency visits and the occurring frequency considered, the best area-specific thresholds can be chosen for various sizes of population at-risk. The novel RaRR allows comparison of health risks across different categories, providing a solid scientific basis for threshold selection.
This work demonstrated the feasibility and flexibility of the proposed approach which considers substantial enhancement of health risks on lags 0 to 2, removes the rare-event interference, and accommodates at-risk populations of different sizes. This methodology can be applied by authorities worldwide for selecting proper heat-health thresholds according to their own morbidity/mortality records.