This is an open-access article distributed under the terms of the Creative Commons Attribution License (

With global warming, the number of days with extreme heat is expected to increase and may cause more acute heat illnesses. While decreasing emissions may mitigate the climate impacts, its effectiveness in reducing acute heat illnesses remains uncertain. Taiwan has established a real-time epidemic surveillance and early warning system to monitor acute heat illnesses since January 1, 2011. Predicting the number of acute heat illnesses requires forecasting temperature changes that are influenced by adaptation policies.

The aim of this study was to estimate the changes in the number of acute heat illnesses under different adaptation policies.

We obtained the numbers of acute heat illnesses in Taiwan from January 2011 to July 2023 using emergency department visit data from the real-time epidemic surveillance and early warning system. We used segmented linear regression to identify the join point as a nonoptimal temperature threshold. We projected the temperature distribution and excess acute heat illnesses through the end of the century when Taiwan adopts the “Sustainability (shared socioeconomic pathways 1‐2.6 [SSP1-2.6]),” “Middle of the road (SSP2-4.5),” “Regional rivalry (SSP3-7.0),” and “Fossil-fueled development (SSP5-8.5)” scenarios. Distributed lag nonlinear models were used to analyze the attributable number (AN) and attributable fraction (AF) of acute heat illnesses caused by nonoptimal temperature.

We enrolled a total of 28,661 patients with a mean age of 44.5 (SD 15.3) years up to July 2023, of whom 21,619 (75.4%) were male patients. The nonoptimal temperature was 27 °C. The relative risk of acute heat illnesses with a 1-degree increase in mean temperature was 1.71 (95% CI 1.63-1.79). In the SSP5-8.5 worst-case scenario, the mean temperature was projected to rise by +5.8 °C (SD 0.26), with the AN and AF of acute heat illnesses above nonoptimal temperature being 19,021 (95% CI 2249‐35,792) and 89.9% (95% CI 89.3%‐90.5%) by 2090‐2099. However, if Taiwan adopts the Sustainability SSP1-2.6 scenario, the AN and AF of acute heat illnesses due to nonoptimal temperature will be reduced to 12,468 (95% CI 3233‐21,704) and 62.1% (95% CI 61.2‐63.1).

Adopting sustainable development policies can help mitigate the risk of acute heat illnesses caused by global warming.

With increasing global warming, the health effects of heat have become an important public health issue. Some studies have reported the relationship between hot temperature and cardiorespiratory disease [

Taiwan is an island country in Southeast Asia (23° 58′ N, 120° 58′ E) with a population of 2.3 million. Taiwan’s climate is hot and humid, and the summer begins in June and ends in September. From 1911 to 2005, Taiwan’s temperature warmed by 1.4 °C, indicating that warming in Taiwan is occurring approximately twice as fast as that in the Northern Hemisphere (0.7 °C) [

The Taiwan Earth System Model version 1 (TaiESM1) is a regional climate model specifically designed to simulate and project climate features in Taiwan. Because of the constraint of computing power, the spatial resolution of the earth system models participating in the Coupled Model Intercomparison Project Phase 5 is typically about 100 km [

This study aims to estimate the number of acute heat illnesses in Taiwan for both the past period (2010-2019) and the future period (2090-2099) under different adaptation policies.

A 2-stage approach was applied. First, we used a distributed lag nonlinear model (DLNM) to explore the nonlinear and lag effects of temperatures (mean) on emergency room visits for acute heat illnesses [

We obtained the daily number of acute heat illnesses from the real-time epidemic surveillance and early warning system of Taiwan. The data are publicly provided by the Centers for Disease Control and Prevention of the Ministry of Health and Welfare. We obtained the number of emergency department visits for acute heat illnesses from January 2011 to July 2023 at the first aid responsibility hospitals in Taiwan. The inclusion criteria for acute heat illnesses included (1) heat and light effects (

We obtained the historical daily meteorological data on maximum, mean, and minimum temperatures and relative humidity from the Taiwan meteorological stations of the Central Weather Bureau in Taiwan [

We used segmented linear regression to identify the join point at which there was a significant change in the number of patients [

Where τ* is the number of optimal breakpoints, _{1} is the number of data points in the first segment (_{i}≤τ), α_{1} and β_{1} are the parameters for the first segment, and α_{2} and β_{2} are the parameters for the second segment.

To find the breakpoint τ that minimizes the sum of squared residuals (SSRs) for the piecewise regression model, we used numerical optimization techniques such as grid search, where we tested various candidate values for τ and chose the one that minimizes the SSR. This process requires iteratively fitting the piecewise regression model and calculating the SSR for each candidate breakpoint.

We identified the join point as a nonoptimal temperature threshold and used it as a reference value for estimating the relative risk (RR) of acute heat illnesses [

Visualize the data: begin by creating a scatter plot of data. Look for any patterns, trends, or possible breakpoints that might suggest the existence of distinct segments.

Fit the initial model: start by fitting a simple nonlinear regression model to the data. This provides a baseline understanding of the overall relationship between the variables.

Consider potential join points: based on the scatter plot, we identified potential join points where the scatter plot showed an inflection point.

Fit the segmented model: use a segmented regression model to fit the data. This model consists of multiple linear segments, each with its slope and intercept. The segmented model identifies the join point where the segments shift.

Evaluate model fit: examine the goodness of fit of the segmented model compared to the initial linear regression. Check the Akaike information criterion to see if the segmented model provides a better fit to the data.

Interpret coefficients: analyze the coefficients of the segmented model to understand the slopes and intercepts of the different segments. The join point corresponds to the value of the independent variable where the transition between segments occurs.

Statistical significance: consider the statistical significance of the join point. If it is statistically significant, it suggests that the change in the relationship between the variables is not due to random variation.

Visual confirmation: overlay the segmented regression line on the scatter plot to visually confirm that the join point accurately captures the shift in the relationship (

We applied the TaiESM1 to predict the temperature in Taiwan under different SSPs. The TaiESM1 has been evaluated against observational data and has demonstrated good skill in reproducing historical climate variability and trends in Taiwan. It accurately simulates temperature, precipitation, wind patterns, and other climatic variables, providing valuable insights into past climate conditions and future projections. The TaiESM generally agrees with observations during the period 1979‐2005. It performs better than the median of Coupled Model Intercomparison Project Phase 5 models [

Because the relationship between temperature and its health effects is nonlinear and has lagged effects, we applied DLNM to explore their relationships. Based on the quasi-Akaike information criterion, we selected the natural cubic spline DLNM to model the nonlinear temperature effects and a polynomial function to model the lagged effects [

Where _{t,l}_{t,l}, l_{t}

Where _{x} denotes the coefficient of DLNM when MMT is used as the reference temperature. The AF is the proportion of acute heat illness caused by nonoptimal temperature. AFs can be multiplied by the total number of acute heat illnesses to obtain the number of acute heat illnesses caused by nonoptimal temperatures [

The projected temperatures from 2015 to 2100 in the SSPs scenarios were added to the established model to estimate the AN and AF of acute heat illnesses due to nonoptimal temperature. We compared changes in AN and AF due to nonoptimal temperatures across 4 adaptation scenarios during 2010‐2019 and 2090‐2099.

We used the

The research ethics committee of National Taiwan University approved this study (202305HM147), which uses secondary data analysis and does not require additional consent.

A total of 28,661 cases were included from January 2011 to July 2023. The demographic characteristics of the patients and the distribution of mean temperatures are summarized in

There was a clear seasonal pattern in temperature in Taiwan (

Demographics of patients and weather distribution in Taiwan during 2011‐2023 (N=28,661).

Values | |

21,619 (75.4) | |

44.5 (15.3) | |

23,628 (82.4) | |

2457 (8.6) | |

2576 (9) | |

23.2 (4.9) | |

5.8-30.5 | |

24 (8.3) | |

28.9 | |

29.8 | |

78.9 (5.6) | |

63.5-94.6 | |

79.2 (7) |

Distribution of temperature, daily number of heat injuries, and trend. (

By segmented linear regression (

The threshold of nonoptimal temperature and effect of acute heat injuries. (

Projection of increased temperature in Taiwan under different climate models. The black line is the historical mean temperature of Taiwan from 1960 to 2014. The red, orange, blue, and green lines are the projected mean temperatures of Taiwan from 2015 to 2099 under different scenarios. Notably, the increase in temperature will be attenuated if Taiwan adopts a sustainability approach (SSP1-2.6). SSP: shared socioeconomic pathway.

Projections of temperature distribution and excess acute heat illnesses under 4 adaptation scenarios (A) SSP1-2.6, (B) SSP 2-4.5, (C) SSP3-7.0, and (D) SSP5-8.5. The top of each panel shows the temperature distribution, and the bottom of each panel shows the distribution of excess acute heat illnesses expressed as the fraction of additional cases (%) attributed to nonoptimal temperature compared with the minimal morbidity temperature (15.5 °C). The gray area is the period of 2010‐2019, and the green area is 2090‐2099. The vertical dashed line on the right is the threshold of nonoptimal temperature (27 °C). SSP: shared socioeconomic pathway.

Attributable number (AN) and attributable fraction (AF) of excess acute heat illnesses above the nonoptimal temperature threshold (27 °C) in different scenarios.

Scenario | Days exceeding nonoptimal temperature (95% CI) | AN of acute heat illnesses exceeding nonoptimal temperature (95% CI) | AF of acute heat illnesses above nonoptimal temperature (%) (95% CI) | |||

2010‐2019 | 2090‐2099 | 2010‐2019 | 2090‐2099 | 2010‐2019 | 2090‐2099 | |

SSP1-2.6^{a} | 416 (379-453) | 919 (867-969) | 6463 (1664-11,262) | 12,468 (3233-21,704) | 34.2 (33.3‐35.2) | 62.1 (61.2‐63.1) |

SSP2-4.5 | 371 (336-408) | 1295 (1238-1351) | 6021 (1344-10,699) | 15,995 (3887-28,102) | 31.9 (31‐32.8) | 77.6 (76.8‐78.5) |

SSP3-7.0 | 323 (291-358) | 1509 (1452-1567) | 5198 (904-9492) | 17,548 (3546-31,549) | 27.6 (26.7‐28.5) | 84.3 (83.6‐85) |

SSP5-8.5 | 377 (341-413) | 1783 (1725-1840) | 6200 (150-12,249) | 19,021 (2249-35,792) | 32.9 (32‐33.9) | 89.9 (89.3‐90.5) |

^{a}SSP: shared socioeconomic pathway.

This study shows that temperatures stabilize by the end of the century in the sustainability development scenario, with the lowest burden of acute heat illness above nonoptimal temperatures. In response to the impact of global warming, Taiwan is formulating a climate policy aimed at reducing carbon emissions. Our study contributes valuable insights to climate health research, underscoring the importance of proactive measures to address health risks associated with rising temperatures.

This study is valuable, as it develops a new method for estimating nonoptimal temperature join points, thus providing a more reasonable threshold for acute heat illnesses. Different from the temperature-mortality relationship, which is often U-shaped [

Our study indicates that the Taiwanese population is more vulnerable to acute heat illnesses compared to other countries. In Taiwan, a 1-degree increase in mean temperature increases the risk of acute heat illnesses by 71% (RR 1.71, 95% CI 1.63-1.79), which is higher than the 18% increase reported in previous meta-analyses (RR 1.18, 95% CI 1.16‐1.19) [

In Taiwan, the government announced the Climate Change Response Act in 2023, which requires all ministries and local governments to assess the impacts of climate change and take practical actions to increase the adaptive capacity of vulnerable groups. This study used data on reported acute heat illnesses from the real-time epidemic surveillance and early warning system of the Ministry of Health and Welfare. Using the data, epidemiologists can quickly estimate trends and risks of diseases associated with global warming, providing timely recommendations for government decision-making.

The use of health surveillance for climate health research and policy can provide timely and actionable information for public health interventions. It can help estimate health impacts on populations, assess the effectiveness of interventions, and inform decision-making to mitigate the health impacts of climate change. The generalizability of our findings to other countries or regions depends on multiple factors, including the availability and quality of health surveillance data, local climatic conditions, and the implementation of climate mitigation strategies. As the threat of heat to public health becomes more apparent, the promotion of national heat hazard prevention strategies to reduce public health impacts is important for policy-making. Spain implemented a Heat Health Prevention Plan between 2004 and 2013. As a result of the program, the impact of temperature on mortality declined. The reduction in mortality due to extreme heat was greater in the provinces where more heat health prevention program measures were implemented [

The real-time epidemic surveillance and early warning system did not cover all hospitals in Taiwan, so there may be underreporting of acute heat-related injuries. The actual number of acute heat illnesses would be higher than our estimate. However, our study used a case-crossover design, which involved comparing the numbers of cases in a time-series sequence. The noninclusion of all hospitals would not influence the association between temperature and the number of acute heat illnesses. In a study of reporting systems, reporting rates may change over time, leading to an increase in reporting in later years. We added a time variable to the DLNM to adjust for this effect, which we conservatively assume would not confound our results. Risk factors for acute heat illnesses include urban residence, age, socioeconomic factors, ethnicity, and race [

This study estimated the burden of acute heat illness under different adaptation policies. If Taiwan adopts high emissions and limited efforts to mitigate climate change, acute heat illnesses due to nonoptimal temperatures will increase substantially in the future, while the adoption of sustainable development policies can help reduce the risk.

The authors thank the Centers for Disease Control and Prevention of the Ministry of Health and Welfare for providing real-time heat injury data for Taiwan and the Taiwan Climate Change Projection Information and Adaptation Knowledge Platform for providing climate data for Taiwan. This study was supported by the National Council of Science and Technology in Taiwan (grant NSTC 113-2314-B-002-187-MY3) and the Population Health and Welfare Research Center from Featured Areas Research Center Program within the framework of the Higher Education Sprout Project by the Ministry of Education in Taiwan (grant NTU-113L9004).

The datasets generated and analyzed during this study are available from the corresponding author on reasonable request.

None declared.

attributable fraction

attributable number

distributed lag nonlinear model

minimal morbidity temperature

relative risk

shared socioeconomic pathway

sum of squared residual

Taiwan Earth System Model version 1

_{2}emissions under the shared socio-economic pathways

Estimated coefficient in segmented regression with SE and