Abstract
Heat warning systems represent a critical health adaptation strategy under global boiling. This study evaluates the appropriateness of the existing criteria of heat warning categories, which were established based on climatology. The objectives were (1) to evaluate the association of the daily maximum temperature (Tmax) and multiple health outcomes (including emergency visits, hospitalization cases, and mortality of heat-related illness (HRI), cardiovascular, respiratory, diabetes, and renal diseases) considering both the lag and cumulative effects, and (2) to identify vulnerable groups (considering their demographic, occupation, geographic, and health status) and areas of heat-health impacts. Weather and health records in Taiwan from May to October from 2000 to 2019 were analyzed using generalized additive models. The results show that HRI was the most sensitive health outcome, and the relative risk (RR) was 1.81 (confidence interval (CI): 1.51 – 2.18) and 2.99 (CI: 1.99 – 4.49) for emergency visits and hospitalizations, respectively, when Tmax was ≥ 34 °C. The corresponding RRs were 2.00 (CI: 1.67 – 2.39) and 2.39 (CI: 1.44 – 3.95) when Tmax was ≥ 32 and ≥ 31 °C for three consecutive days, respectively. The morbidity risks of cardiovascular, respiratory, diabetes, and renal diseases all increased at different temperature thresholds. Significant associations between Tmax and health outcomes occurred at thresholds lower than the current warning thresholds, indicating the need for revision. Both lag and cumulative effects need to be considered in heat-health warning systems. People with hypertension, hyperglycemia, or hyperlipidemia were found to be more vulnerable, as they had higher RRs for pneumonia, COPD, and stroke than the general population. Among different occupations, farmers were found to be most vulnerable. This study demonstrates a methodology considering both lag and cumulative effects that can be applied in other countries to assist in the establishment of evidence-based heat-health warning systems.
Supplementary Information
The online version contains supplementary material available at 10.1038/s41598-025-31832-2.
Keywords: Heat-related health impacts, Sustainable development goal 3, Heat vulnerability, Heat and health, Warning threshold
Subject terms: Environmental impact, Risk factors
Introduction
The World Meteorological Organization (WMO) has officially announced that 2024 was the warmest year on record by a considerable margin, with temperatures approximately 1.55 °C above pre-industrial levels1. In 2024, heatwaves occurring before June in Asia, Africa, Europe, the Middle East, and North America claimed hundreds of lives2. At the 27th United Nations Climate Change Conference in 2022, it was emphasized that a science-based heat-health early warning system is urgently needed3. The 16 heat-health warning systems (HHWSs) currently used in Europe have been reviewed4. Most warning thresholds are mainly defined on the basis of epidemiological evidence of the relationship between temperature and mortality, while others are based on climatological percentiles. Nevertheless, since heat stress is related to numerous health outcomes5, it is preferable to select thresholds based on emergency visits or morbidity records, rather than mortality, for prevention purposes. Increased efforts in this field could contribute to meeting the United Nations’ Sustainable Development Goal 3 (SDG3), which aims to guarantee healthy lives and promote wellbeing for all people at all ages6.
This study is a collaboration between scientists and policy makers aiming to identify suitable thresholds for an HHWS and vulnerable populations/areas for priority setting. HHWSs in other countries were reviewed, and it was found that some countries set warning thresholds for one day, such as US Heat Advisories7, and others, such as France and Spain4, considered a period of three consecutive days. For one-day thresholds, it is important to consider the lag effect; for three-day thresholds, it is crucial to assess the cumulative effect. Thus, we intended to assess both the lagged and cumulative health impacts associated with potential thresholds.
Despite the importance of establishing HHWSs, very few epidemiological studies utilize both the lag and cumulative effects of high temperatures as the basis for selecting HHWS thresholds. In the following, we present a brief review of recent studies on heat-health epidemiological evidence for warning systems in non-European countries. In Canada, the relationships between temperature and 3-day running averages of mortality were evaluated to establish heat alert thresholds; this was conducted by considering only the cumulative effects8. In China, the performance of HHWSs in Shanghai was evaluated using heat-related illness (HRI), considering 3 days of cumulative effects9. In India, the relationship between temperature and mortality in two cities between 2003 and 2017 was assessed to determine thresholds; no lag or cumulative effects were considered10. In the US, the relationships between the heat index and hospitalizations in 1,617 counties over the period 2003–2012 were estimated to inform HHWSs based on cardiovascular (CVD) disorder, respiratory (RD) disorder, diabetes (DM), fluid and electrolyte disorders, and renal failure, considering only a 2-day lag effect7. One paper, focused on Alaska, was one of the few publications that considered both the lag and the accumulative effects; the relationships between the heat index and HRIs plus cardiorespiratory diseases were assessed using a 5-day lag effect and a 2-day cumulative effect11. In the current study, we attempt to fill the aforementioned research gap by evaluating both the lag and the cumulative effects of multiple morbidity outcomes using a Taiwanese population database for demonstration.
Island Taiwan is in the subtropical zone; the temperature increases in urban areas have been twice that of the global increase12. The Central Weather Administration (CWA) in Taiwan does not have an official HHWS yet; however, since June 15, 2018, it announces “High Temperature Information” to the public when the daily maximum temperature (Tmax) reaches above certain thresholds13. Nevertheless, the threshold was selected based on climatological data rather than epidemiological evidence, which would involve investigating the features of the Taiwanese population. Even though Taiwan has a high prevalence of air-conditioning units (93%)14, significant associations between Tmax and heat-related health impacts have been found when the Tmax exceeds certain thresholds15–17. It was found that the elderly (aged ≥ 65) had the highest relative risk (RR = 5.29, 95% confidence interval (CI): 3.73 – 7.51) of heat-related emergency visits when the Tmax reached ≥ 33.5 °C in Taiwan during the period 2000–2014; their RR was higher than those of children aged 0–14 (RR = 3.69, CI: 1.78 – 7.66) and of people aged 15–64 (RR = 2.60, CI: 2.13 – 3.17))16. Moreover, the RR of heat-related hospital visits among the Taiwanese population was 1.31 (CI: 1.29 – 1.33) at Tmax ≥ 33.5 °C during the period 2000–201717. In another study, the emergency room visits between 2000 and 2014 in Taiwan were assessed; extreme high temperatures were associated with an increase in certain health risks, including CVD and chronic kidney diseases; this was especially the case for the subpopulation aged 40–6418. Several studies involving hundreds of cities across more than 20 countries including Taiwan found statistically significant associations between temperature and mortality19–21; for example, temperature variability within or between days has been found to be associated with mortality19. During the years 1990 to 2019, 0.94% (CI: 0.68 – 1.19) of deaths per warm season were estimated to be from heatwaves, accounting for 236 (CI: 170 – 300)) deaths per 10 million residents20. Additionally, extreme hot days contributed to 2.2 (CI: 1.9 – 2.4)) and 0.7 (CI: 0.5 – 0.8)) excess deaths per 1,000 ischemic and hemorrhagic stroke deaths, respectively21.
As heat waves are expected to become increasingly frequent in the coming decades, it is necessary to evaluate whether the current warning criteria thresholds are appropriate for the goal of protecting public health. In addition, identifying vulnerable groups is essential for prioritizing resource allocation. Therefore, the objectives of this study were (1) to evaluate the association of Tmax and multiple health outcomes (including emergency visits, hospitalization cases, and mortality of HRI, CVD, RD, DM, and kidney diseases (KD)) considering both the lag and cumulative effects when Tmax exceeds certain thresholds and (2) to identify vulnerable groups and areas of heat-related health impacts. The findings could serve as a foundation for revising current criteria and providing health authorities with the information needed to confidently plan responsive programs accordingly.
Methods
Data source
Hourly meteorological data from 2000 to 2019 were obtained from 23 non-mountainous (< 500 m above sea level) CWA stations in Taiwan (Additional File Figure S1). Only 0.93% of the 2019 population lived in mountainous areas 500 m above sea level; thus, only non-mountainous stations were selected in this study. Only data from May to October are included in this work since this period corresponds to the warm season in Taiwan22.
Tmax was selected as the indicator based on two rationales: (1) Our earlier work demonstrated that wet-bulb globe temperature (WBGT) had more consistent and stronger associations with the HRI emergency and hospital visits in Taiwan compared to temperature and apparent temperature16,17. However, since temperature is currently used by the CWA in Taiwan in their “High Temperature Information”, the associations of temperature and health outcomes are of interest to the Ministry of Health and Welfare in Taiwan. (2) Tmax is selected rather than mean temperature since most countries using temperature in the HHWS selecting Tmax rather than mean temperature4, such as Belgium, England, France, Greece, Netherlands, Spain, and Sweden.
The daily Tmax was obtained from each CWA station; all of these values were used to calculate the average Tmax for that day. Stations with over 12 h of missing data were excluded. Since these are all standard weather stations with manpower for daily maintenance, only 0.08% of data were excluded from the dataset; daily Tmax were then calculated from the other stations. As a result, there were no missing data for the daily Tmax in our dataset.
Additionally, in order to identify vulnerable areas, Taiwan was divided into six divisions, as shown in Figure S1: North Area 1 (N1), North Area 2 (N2), Central Area, South Area 1 (S1), South Area 2 (S2), and the East Area. Taiwan Island is typically divided into these six geographical divisions, given the similar geographical features and climate patterns within each division. The daily Tmax value for a specific division was obtained by averaging all of the Tmax values reported in that division. Again, stations with over 12 h of missing data were excluded from that day’s calculation. Since there are multiple stations in one division, there were no missing data for the daily Tmax in our dataset for each division. An analysis of health records with finer spatial resolutions was not possible for the studied health outcomes due to privacy considerations. The Taiwan government imposes restrictive regulations on data analysis at the spatial unit that has only a few cases of the specified health outcome.
In the “High Temperature Information”, three categories were established according to the following criteria: yellow—Tmax reaches or exceeds 36 °C; orange—Tmax reaches or exceeds 36 °C for three or more consecutive days or the temperature reaches or exceeds 38 °C; red—Tmax reaches or exceeds 38 °C for three or more consecutive days13. Figure S2 in the Additional File shows the numbers of days with Tmax values in the following categories: between ≥ 34 and < 36, between ≥ 36 and < 38, and ≥ 38 °C in the entire Taiwan and its divisions. Divisions without any days ≥ 36 °C during this 20-year period are not shown in Figures S2–S3. Temperatures above 38 °C were recorded for only 2 days in the N2 region in 2009 and 1 day in the Central region in 2004. Figure S3 in the Additional File shows the numbers of days with Tmax values occurring on three consecutive days in the following categories: between ≥ 32 and < 34, between ≥ 34 and < 36, and ≥ 36 °C in the entire Taiwan and its divisions. Only divisions with days with recorded temperatures ≥ 34 °C are shown between 2000 and 2019. Only three cases in 2010 had 3 consecutive days with temperatures above 36 °C; these occurred in the N1 area.
Daily health records between 2000 and 2019 were obtained from the Health and Welfare Data Science Center database of the Ministry of Health and Welfare, Taiwan. This database contains data on hospital records (including emergency visits, hospital visits, and hospitalization cases) for almost the entire population of Taiwan (23.6 million as of 2019), with variables such as patient age, patient sex, and hospital location. HRI, CVD, RD, DM, and KD cases were selected according to the 9th Revision of the International Classification of Diseases (ICD-9-CM) for the data from the 2000–2015 period and according to ICD-10-CM for the data from the 2016–2019 period, respectively (detailed categories are listed in Table S1(a) in the Additional File). These disease categories were associated with higher temperatures in previous studies7,11. In addition to the main disease categories, certain subcategories with large sample sizes and extensive documentation of their associations with extreme temperature events were also chosen. These included ischemic heart disease (IHD)18 and stroke21 within the CVD category, pneumonia11, asthma11, and chronic obstructive pulmonary disease (COPD)23 within the RD category, and acute kidney failure (AKF)7 and urolithiasis24 within the KD category. Moreover, mortality counts in the above main categories and subcategories between 2008 and 2019 were obtained from the Taiwan National Mortality Registry, with case selection based on ICD-10 (Table S1(a)). Mortality data for 2000–2007 were not used due to incomparable classification for the data of those years. Furthermore, there were very few HRI mortality cases; thus, the association of all-cause mortality (excluding accidents and suicides) and Tmax was assessed instead.
Statistical analysis
The associations between the Tmax and health outcomes were examined using generalized additive models (GAMs) with Poisson distribution. The relationships of health outcomes and temperature over the entire temperature range are non-linear7,25. Nevertheless, the purpose of our work is to examine the effect of temperature above a predefined threshold. Previous work has found that the association between Tmax and HRI is linear above 30 °C in Taiwan16, and our analysis also found that the relationships between the Tmax and studied health outcomes are mostly linear above potential thresholds (Additional file Figure S4). Although effects of a few outcomes may show slight deviations from linearity at the very highest temperatures, applying a non-linear model such as the distributed lag non-linear model (DLNM) could risk overfitting due to the limited number of days with extreme heat. For the purpose of this study, a GAM offers greater interpretability than the more complex DLNM. Besides, we did not set a washout period since multiple hospital visits due to the same cause during a short period of time is not a concern for emergency visits and hospitalization cases. This study was reviewed and approved by the Institution Review Board of Academia Sinica, IRB No. AS-IRB-BM-20085. SAS 9.4 (SAS Institute Inc., Cary, NC, USA) and R 4.1.3 (with mgcv package) software were used. The authors had no access to information that could identify individual participants during or after data collection. The date range during which our analysis was conducted was from February 1, 2021, to October 31, 2025, after IRB approval.
Lag effects of two days are considered for two reasons: Firstly, the main goal of this work is to evaluate the appropriateness of the current thresholds of the “High Temperature Information”. Thus, the impacts of high temperature in the first few days are important. The impact of high temperatures on health is typically most pronounced within 0–1 lag days; for example, a multi-country analysis across 384 locations, including Taiwan, found that the strongest effects occurred within this short lag period26. Secondly, in preliminary testing, RRs with lags of 0–2 for temperature and health outcomes were higher than those with lags of 3–7. Table S2 in the Additional File shows some of our test results. Table S2 presents the RRs of HRI and CVD in models with lags of 0–2 and 0–3. It is evident that RR values decrease from lag 0 to 3, and the RRs at lag 3 are usually not statistically significant. However, RRs are sometimes statistically significant at Lag 2, for example, for HRI and CVD at 33.5 °C. Therefore, only RRs of lags 0–2 are considered. The details of the GAM, model fitting, and parameter selections can be found in previous publications16,17 and are briefly described below.
Considering the two-day lags and avoiding the interference of rare events, the final model used is presented here, with Tmax denoted as
and health outcomes denoted as
.
![]() |
1 |
where
(the same definition for
and
(the same definition for 
where
is a specified threshold candidate of the Tmax;
is the chosen upper limit approximating the 99.5th percentile of the Tmax in the studied area17; and Lt is a dummy variable for removing the interference of rare events when the Tmax exceeds
.
is evaluated starting at 30 °C, with 0.5 °C as an interval. The threshold candidate is selected based on the highest RR with statistical significance, defined as p < 0.01. Since multiple health impacts were assessed here, the best threshold is selected as the one associated with statistically significant RRs of the highest number of the health outcomes among all potential threshold candidates. Additionally,
is set as 34.5, 36.5, 36, 35.5, 35, 34, and 34 °C for the whole of Taiwan and the N1, N2, Central, S1, S2, and East regions, respectively. α is a constant;
is a thin-plate spline function for day t. The potential covariates on day t, denoted by
—such as the daily mean PM2.5 concentration (representing the influence of air pollution, obtained from the Ministry of Environment, Taiwan), the day of the week, whether the day was a holiday, and periods with prevalent severe acute respiratory syndrome—were considered for adjustment. These covariates were selected based on the findings and practices of previous epidemiological studies, which indicated that they would impact the targeted health outcomes23,27. For example, a recent study in Thailand showed that the RR of all-cause hospitalization was affected by both PM2.5 and temperature27. The day of the week, holidays, and periods with prevalent severe acute respiratory syndrome were typically adjusted in epidemiological analyses, as hospitals in Taiwan were not operating as usual during these periods18. Relative humidity was not in the model since it correlates highly with temperature.
The
is calculated as
) for any value
, which is the ratio of the expected health outcome for a k unit increase (k = 1 in this study) over the Tmax threshold compared with the expected health outcome for any value below or equal to the threshold. Since there are three
s, the RR is calculated as their geometric mean, as follows:
![]() |
2 |
For the 3-day cumulative effect, the model is as follows:
![]() |
3 |
where
(the same definition for
.
In order to remove the influence of rare events, the number of cases above a certain Tmax value less than ten in these 20-year periods was not included in the analysis. The same analysis was carried out for the population in Taiwan as a whole and for different classifications: (1) sex and age groups, (2) insurance categories, and (3) areas for the purpose of identifying vulnerable groups. Age groups were defined as age ≤ 14, 15–39, 40–64, and > 65 to differentiate children, young adults, middle-aged adults, and elderly adults, respectively. Additionally, Taiwan’s National Health Insurance included jobless citizens (such as low income group) and those insured through their employees. The insurance category includes broad occupations such as union workers without details. The above three classifications were evaluated in order to identify vulnerable populations; we were unable to evaluate other classifications due to limited information from the database.
In addition, people with pre-existing diseases are also known to be vulnerable to high temperatures23. In Taiwan, people with chronic diseases are mostly associated with one of these three symptoms, namely, hypertension, hyperglycemia, or hyperlipidemia (called the hyper group), and the size of this group is increasing. An analysis was conducted to evaluate whether this group is potentially vulnerable to heat-related health impacts. This hyper group was identified using ICD-9-CM and ICD-10-CM, as listed in Table S1(b) of the Additional File; if a patient had been diagnosed and classified under one of these codes on two hospital visits or one hospitalization event, then the patient was classified under the hyper group for that year. One hospital visit with those codes was not qualified to be included, since the patient may recover soon after diagnosis. Table S3 in the Additional File shows the numbers of daily emergency visits, hospitalization cases, and mortality for the studied outcomes in Taiwan and for different group and area categories for 2000–2019.
Results
This study’s objective was to select an appropriate threshold for an HHWS based on epidemiological evidence considering the features of the Taiwanese population. Thus, the highest RRs among multiple threshold candidates for a specific health outcome are presented, with the following classifications: (1) the general public, (2) hyper group, (3) sex and age groups, (4) insurance categories, and (5) geographical areas. Table 1 shows the highest RR of the lag effect among RRs under different Tmax threshold candidates for the general public and the hyper group for specified health outcomes; most of these RRs occurred at 34 °C for emergency visits, hospitalization cases, and mortality. Among the various health outcomes, the RRs of the HRIs and urolithiasis are the two highest for emergency visits (1.81, CI: 1.51 – 2.18, p < 0.0001 and 1.36, CI: 1.30 – 1.42, p < 0.0001, respectively) and hospitalization cases (2.99, CI: 1.99 – 4.49, p < 0.0001 and 1.20 CI: 1.14 – 1.27, p < 0.0001, respectively) when Tmax is ≥ 34 °C. Regarding mortality, IHD has the highest RR (1.35, CI: 1.12 – 1.63, p = 0.0018) when Tmax is ≥ 34 °C.
Table 1.
The highest RRs and thresholds (°C) for the general public and hyper group (lag effect).
| HRI | CVD | IHD | Stroke | RD | Pneumonia | Asthma | COPD | DM | KD | AKF | Urolithiasis | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| (a) Emergency visits | ||||||||||||
| The general public | ||||||||||||
| Threshold | 34 | 34 | 34 | 34 | 34 | 33.5 | 34 | – | 34 | – | 34 | 34 |
| RR | 1.81 | 1.04 | 1.07 | 1.09 | 1.06 | 1.03 | 1.10 | – | 1.06 | – | 1.26 | 1.36 |
| (95% CI) | (1.51,2.18) | (1.02,1.05) | (1.03,1.11) | (1.05,1.13) | (1.05,1.07) | (1.02,1.04) | (1.05,1.16) | – | (1.02,1.10) | – | (1.08,1.47) | (1.30,1.42) |
| p-value | < 0.0001 | < 0.0001 | 0.0008 | < 0.0001 | < 0.0001 | < 0.0001 | < 0.0001 | – | 0.0009 | – | 0.0031 | < 0.0001 |
| Hyper Group | ||||||||||||
| Threshold | 34 | – | – | 34 | 34 | 33.5 | 34 | 34 | – | – | – | 34 |
| RR | 1.75 | – | – | 1.09 | 1.04 | 1.04 | 1.13 | 1.08 | – | – | – | 1.28 |
| (95% CI) | (1.26,2.42) | – | – | (1.05,1.13) | (1.01,1.06) | (1.03,1.06) | (1.04,1.22) | (1.02,1.14) | – | – | – | (1.18,1.38) |
| p-value | 0.0008 | – | – | < 0.0001 | 0.0068 | < 0.0001 | 0.0043 | 0.0069 | – | – | – | < 0.0001 |
| (b) Hospitalized cases | ||||||||||||
| The general public | ||||||||||||
| Threshold | 34 | 34 | 34 | 34 | 34 | 34 | – | – | 34 | – | 34 | 34 |
| RR | 2.99 | 1.08 | 1.10 | 1.11 | 1.06 | 1.04 | – | – | 1.09 | – | 1.13 | 1.20 |
| (95% CI) | (1.99,4.49) | (1.06,1.10) | (1.06,1.14) | (1.07,1.15) | (1.04,1.08) | (1.01,1.07) | – | – | (1.07,1.12) | – | (1.06,1.21) | (1.14,1.27) |
| p-value | < 0.0001 | < 0.0001 | < 0.0001 | < 0.0001 | < 0.0001 | 0.0049 | – | – | < 0.0001 | – | 0.0003 | < 0.0001 |
| Hyper Group | ||||||||||||
| Threshold | –a | 34 | 34 | 34 | – | 33.5 | – | 33.5 | – | – | 34 | 34 |
| RR | – | 1.08 | 1.10 | 1.12 | – | 1.02 | – | 1.02 | – | – | 1.14 | 1.15 |
| (95% CI) | – | (1.06,1.10) | (1.05,1.14) | (1.08,1.16) | – | (1.01,1.03) | – | (1.01,1.03) | – | – | (1.06,1.23) | (1.06,1.24) |
| p-value | – | < 0.0001 | < 0.0001 | < 0.0001 | – | < 0.0001 | – | 0.0020 | – | – | 0.0007 | 0.0009 |
| (c) Mortality | ||||||||||||
| The general public | ||||||||||||
| Threshold | 34b | 34 | 34 | – | 34 | – | – | – | 34 | 33 | – | – |
| RR | 1.15 | 1.29 | 1.35 | – | 1.20 | – | – | – | 1.19 | 1.02 | – | – |
| (95% CI) | (1.09,1.20) | (1.17,1.42) | (1.12,1.63) | – | (1.05,1.38) | – | – | – | (1.04,1.36) | (1.00,1.03) | – | – |
| p-value | < 0.0001 | < 0.0001 | 0.0018 | – | 0.0076 | – | – | – | 0.0095 | 0.0065 | – | – |
| Hyper Group | ||||||||||||
| Threshold | 34b | 34 | – | – | – | 33.5 | – | – | – | – | – | – |
| RR | 1.16 | 1.27 | – | – | – | 1.07 | – | – | – | – | – | – |
| (95% CI) | (1.09,1.23) | (1.13,1.43) | – | – | – | (1.02,1.12) | – | – | – | – | – | – |
| p-value | < 0.0001 | 0.0001 | – | – | – | 0.0040 | – | – | – | – | – | – |
a: Not available due to small sample size or the result was not statistically significant. b: The studied outcome was all-cause mortality. RR: relative risk, CI: confidence interval, HRI: Heat-related illnesses, CVD: Cardiovascular Diseases, IHD: Ischemic Heart Disease, RD: Respiratory Disease, COPD: Chronic Obstructive Pulmonary Disease, DM: Diabetes Mellitus, KD: Kidney Diseases, AKF: Acute Kidney Failure.
The RRs of the hyper group in most cases are close to those of the general public at almost the same thresholds for emergency visits, hospitalization cases, and mortality. Nevertheless, the hyper group seems to be more vulnerable to pneumonia and COPD. The RRs of the hyper group in the COPD emergency visits, the COPD hospitalized cases, and the pneumonia mortality are 1.08 (CI: 1.02 –1.14, p = 0.0069), 1.02 (CI: 1.01 – 1.03, p = 0.002), and 1.07 (CI: 1.02 – 1.12, p = 0.004), respectively, while those of the general public did not reach statistical significance.
Table 2 shows the highest RRs of the 3-day cumulative effect among RRs at different Tmax threshold candidates for the general public and for the hyper group for the specified health outcomes. These highest RRs are associated with a range of Tmax threshold candidates, with 29.5–33 and 29.5–32.5 °C for emergency visits and hospitalization cases, respectively. HRI and AKF have the top two highest RRs in the emergency visits (2.00, CI: 1.67 – 2.39, p < 0.0001 and 1.20, CI: 1.08 – 1.34, p = 0.0011, respectively) when the Tmax is ≥ 32 and 33 °C for three consecutive days, respectively; they also have the top two highest RRs in the hospitalization cases (2.39, CI: 1.44 – 3.95, p = 0.0007 and 1.11, CI: 1.06 – 1.16, p < 0.0001, respectively) when the Tmax is ≥ 31 and 31.5 °C for three consecutive days, respectively. For the mortality, IHD has the only statistically significant RR (1.24, CI: 1.08 – 1.41, p = 0.0022) when Tmax is ≥ 30 °C for three consecutive days.
Table 2.
The highest RRs and thresholds (°C) for the general public and hyper group (cumulative effect).
| HRI | CVD | IHD | Stroke | RD | Pneumonia | Asthma | COPD | DM | KD | AKF | Urolithiasis | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| (a) Emergency visits | ||||||||||||
| The general public | ||||||||||||
| Threshold | 32 | –a | – | – | 31 | 30.5 | – | 32 | 30.5 | 32 | 33 | 32.5 |
| RR | 2.00 | – | – | – | 1.06 | 1.12 | – | 1.04 | 1.04 | 1.04 | 1.20 | 1.09 |
| (95% CI) | (1.67,2.39) | – | – | – | (1.05,1.06) | (1.10,1.14) | – | (1.01,1.07) | (1.01,1.06) | (1.01,1.06) | (1.08,1.34) | (1.06,1.12) |
| p-value | < 0.0001 | – | – | – | < 0.0001 | < 0.0001 | – | 0.0032 | 0.0010 | 0.0060 | 0.0011 | < 0.0001 |
| Hyper Group | ||||||||||||
| Threshold | 32 | – | – | 31 | 32 | 30.5 | – | 32 | – | 30.5 | 33 | 29.5 |
| RR | 2.45 | – | – | 1.03 | 1.04 | 1.09 | – | 1.07 | – | 1.04 | 1.20 | 1.10 |
| (95% CI) | (1.75,3.44) | – | – | (1.01,1.05) | (1.02,1.05) | (1.06,1.12) | – | (1.04,1.11) | – | (1.01,1.07) | (1.06,1.37) | (1.04,1.17) |
| p-value | < 0.0001 | – | – | 0.0081 | < 0.0001 | < 0.0001 | – | < 0.0001 | – | 0.0028 | 0.0040 | 0.0008 |
| (b) Hospitalized cases | ||||||||||||
| The general public | ||||||||||||
| Threshold | 31 | 30 | 30.5 | – | 30.5 | 30.5 | 29.5 | 31.5 | 32 | 32 | 31.5 | 29.5 |
| RR | 2.39 | 1.03 | 1.05 | – | 1.05 | 1.07 | 1.06 | 1.04 | 1.03 | 1.05 | 1.11 | 1.08 |
| (95% CI) | (1.44,3.95) | (1.02,1.04) | (1.02,1.07) | – | (1.04,1.06) | (1.05,1.09) | (1.02,1.11) | (1.01,1.06) | (1.01,1.04) | (1.02,1.07) | (1.06,1.16) | (1.04,1.12) |
| p-value | 0.0007 | < 0.0001 | < 0.0001 | – | < 0.0001 | < 0.0001 | 0.0047 | 0.0092 | < 0.0001 | 0.0002 | < 0.0001 | < 0.0001 |
| Hyper Group | ||||||||||||
| Threshold | – | 30 | 32 | 32.5 | 29.5 | 32 | 29.5 | 29.5 | – | 31.5 | 31.5 | 29.5 |
| RR | – | 1.03 | 1.04 | 1.03 | 1.04 | 1.06 | 1.09 | 1.05 | – | 1.05 | 1.11 | 1.08 |
| (95% CI) | – | (1.01,1.04) | (1.02,1.07) | (1.01,1.06) | (1.02,1.06) | (1.03,1.09) | (1.03,1.15) | (1.01,1.08) | – | (1.02,1.07) | (1.05,1.17) | (1.03,1.14) |
| p-value | – | < 0.0001 | 0.0006 | 0.0043 | < 0.0001 | < 0.0001 | 0.0044 | 0.0065 | – | 0.0009 | < 0.0001 | 0.0035 |
| (c) Mortality | ||||||||||||
| The general public | ||||||||||||
| Threshold | –b | – | 30 | – | – | – | – | – | – | – | – | – |
| RR | –b | – | 1.24 | – | – | – | – | – | – | – | – | – |
| (95% CI) | –b | – | (1.08,1.41) | – | – | – | – | – | – | – | – | – |
| p-value | –b | – | 0.0022 | – | – | – | – | – | – | – | – | – |
a: Not available due to small sample size or the result was not statistically significant. b: The studied outcome was all-cause mortality. 1. RR: relative risk, CI: confidence interval, HRI: Heat-related illnesses, CVD: Cardiovascular Diseases, IHD: Ischemic Heart Disease, RD: Respiratory Disease, COPD: Chronic Obstructive Pulmonary Disease, DM: Diabetes Mellitus, KD: Kidney Diseases, AKF: Acute Kidney Failure. 2. The mortality of the hyper group is not presented since there was no RR with p < 0.01 for any health outcomes.
The hyper group seems to be more vulnerable to stroke, RD, COPD, KD, and urolithiasis. The hyper group, but not the general public, has statistically significant RRs associated with stroke-related emergency visits (1.03, CI: 1.01 – 1.45, p = 0.0081) and hospitalization (1.03, CI: 1.01 – 1.06, p = 0.0043). In addition, the highest RRs in the hospitalization cases for RD and COPD have lower thresholds than those of the general public, with the difference in threshold candidates being greater than or equal to 1 °C; this was also the case for the emergency visits caused by KD and urolithiasis.
Figure 1 shows the highest RRs of the lag effect with p < 0.01 for different health outcomes for different sex and age groups; the order of presentation is ranked by the corresponding Tmax thresholds. The RRs of the general public are also shown for comparison to identify vulnerable groups for the same threshold candidate. The results of the emergency visits, hospitalization cases, and mortality are shown in Fig. 1a, b and c, respectively. The threshold of 34 °C is associated with the highest RRs for most of the health outcomes for different sex and age groups, considering emergency visits, hospitalization cases, and mortality altogether. For the 3-day cumulative effect for the different sex and age groups, the highest RRs with p < 0.01 of different health outcomes for emergency visits and hospitalization cases are shown in Fig. 2a and b, respectively. The thresholds for different sex and age groups ranged from 29.5 to 33 °C and from 29.5 to 32 °C for the emergency visits and hospitalization cases, respectively. The mortality results are not shown since there was no group associated with significant RRs.
Fig. 1.
RRs of (a) emergency visits, (b) hospitalized cases, and (c) mortality for different demographic groups (lag effect). Note: 1. Relative risks (RR) are presented for the entire Taiwan population and different age and sex groups ranked by thresholds considering 2-day lag effect. 2. Blank cells mean that (1) there was no sufficient case in that category for statistical analysis or (2) the statistical results were with p ≥0.01.
Fig. 2.
RRs of (a) emergency visits and (b) hospitalized cases for different demographic groups (cumulative effect). Note: 1. Relative risks (RRs) are presented for the entire Taiwan population and different age and sex groups ranked by thresholds considering 3-day cumulative effect. 2. Blank cells mean that (1) there was no sufficient case in that category for statistical analysis or (2) the statistical results were with p ≥0.01. 3. The RRs of mortality are not shown since there were no groups associated with RR of p < 0.01.
Vulnerable groups can be identified with these graphs using two criteria. For the same health outcome, the groups with a lower threshold with RRs with p < 0.01 and the groups with the same threshold associated with greater RRs with p < 0.01, compared to that of the general public, are viewed as vulnerable groups. For example, in the lag effect of the emergency visits (Fig. 1a), males aged 15–39 had a lower threshold (33.5 °C) with RRs with p < 0.0001 compared to the general public (34 °C) in HRI; thus, males aged 15–39 years are a group vulnerable to HRI. For RD, females aged 15–39 years have a lower threshold (33 °C) of RRs with p < 0.0001 compared to the general public (34 °C), and males and females aged 0–14, as well as females aged above 65, have higher RRs with p < 0.0001 at the same threshold (34 °C). These four groups are identified as being RD-vulnerable. With the above criteria, the vulnerable groups identified based on both the lag and cumulative effects are summarized in Table 3, and are discussed later.
Table 3.
Vulnerable groups identified based on threshold evaluation (°C) of both lag and cumulative effects.
| HRI | CVD | IHD | Stroke | RD | Pneumonia | Asthma | COPD | DM | KD | AKF | Urolithiasis | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hyper Group | |||||||||||||
| EV | 32a | – | – | – | – | 33.5 | 34 | 34, 32a | – | 30.5a | – | 29.5a | |
| HC | – | – | – | 34, 32.5a | 29.5a | 33.5 | 29.5a | 33.5, 29.5a | – | 31.5a | 34 | – | |
| MA | 34b | – | – | – | – | 33.5 | – | – | – | – | – | – | |
| Sex and age | |||||||||||||
| Female aged 0–14 | EV | – | – | – | – | 34, 31a | – | – | – | –. | – | – | – |
| HC | – | – | – | – | 34 | – | – | – | – | – | – | – | |
| Male aged 0–14 | EV | – | – | – | – | 34, 31a | – | – | – | – | – | – | – |
| HC | – | – | – | – | 30.5a | 30.5a | – | – | – | – | – | – | |
| Female aged 15–39 | EV | – | 29.5a | – | – | 33 | 30.5a | – | – | – | – | – | – |
| HC | – | – | – | – | 33 | – | – | – | 33.5 | – | – | – | |
| Male aged 15–39 | EV | 33.5, 30.5a | – | – | – | – | – | 34, 31.5a | – | 33.5 | – | – | 30a |
| HC | – | 33.5 | – | – | 34, 30.5a | – | – | – | – | 34 | – | 34 | |
| Female aged 40–64 | EV | – | – | – | 30a | – | 30a | – | – | – | – | – | – |
| HC | – | 34 | 34 | – | 34 | – | – | – | 34, 32a | – | – | 34 | |
| MA | 33b | – | – | – | – | – | – | – | – | – | – | – | |
| Male aged 40–64 | EV | 32a | – | 33.5 | 34 | – | 30a | 34 | 30a | – | 33.5 | – | 34, 32.5a |
| HC | – | 30a | – | – | – | – | – | – | 33.5, 30.5a | – | 33.5 | 29.5a | |
| MA | 33.5b | – | – | – | – | – | – | – | – | – | – | – | |
| Female aged above 65 | EV | – | 34 | 34 | – | 34 | 34, 30.5a | 31.5a | 34 | – | 30.5a | – | 34, 29.5a |
| HC | – | 34 | 34 | 34, 32.5a | 34 | – | – | 29.5a | 34 | 33.5 | 31.5a | 34 | |
| MA | 34b | 34 | – | – | – | – | – | – | – | – | – | – | |
| Male aged above 65 | EV | – | – | – | 31a | – | 33.5 | 33.5a | – | 30.5a | 33.5 | – | 34 |
| HC | – | – | 33.5, 30.5a | – | – | 34 | – | 34, 31.5a | 31a | 31.5a | 34, 31.5a | 33.5, 29.5a | |
| MA | – | 33.5 | – | – | – | – | – | – | – | – | – | – | |
| Insurance categories | |||||||||||||
| Manager & Teacher | EV | – | – | – | – | – | – | – | – | – | – | – | – |
| HC | – | 33.5 | – | – | – | – | – | – | – | – | – | – | |
| Service & Sale worker | EV | 31a | – | 31.5a | 34 | 31a | – | – | – | – | – | – | 29.5a |
| HC | – | 30a | 30a | – | – | – | – | – | 33.5 | – | – | 34 | |
| Clerical & Support worker | EV | – | – | – | – | 31a | – | – | – | – | – | – | – |
| HC | – | 33.5 | – | – | – | – | – | – | – | – | – | – | |
| Union workers | EV | – | – | – | – | 33, 31a | 30.5a | 34, 29.5a | – | – | – | – | 34, 32.5a |
| HC | – | 34 | – | – | – | – | – | – | – | – | – | 34 | |
| Farmer | EV | – | 34 | – | 33.5 | 30.5a | 30.5a | 34 | 33.5 | 33.5 | – | – | – |
| HC | – | 34 | 34 | – | 34 | 34 | – | – | 34 | – | 34 | 29.5a | |
| MA | 34b | – | – | – | – | – | – | – | – | – | – | – | |
| Fishermen | EV | – | 31.5a | – | – | 31a | – | – | – | – | – | – | – |
| Low income | EV | – | 32.5a | – | – | – | – | – | – | 34 | – | – | – |
| HC | – | 29.5a | – | – | 33 | – | – | – | – | – | – | – | |
| MA | 30.5ab | – | – | – | – | – | – | – | – | – | – | – | |
| Veteran | EV | – | – | 32a | 33.5 | 34 | – | – | 34 | 34 | 33 | – | – |
| HC | – | – | – | – | 33.5 | – | – | – | 34 | – | 33.5 | – | |
| MA | 33.5b | – | – | – | – | – | – | – | – | – | – | – | |
| Others | EV | – | – | – | – | – | 30a | – | 33 | – | – | – | 34 |
| HC | – | – | – | – | – | – | – | – | – | 32a | – | – | |
| Areas | |||||||||||||
| N1 area | EV | – | – | – | – | – | – | – | – | – | – | – | 32.5a |
| HC | – | – | – | 31.5a | – | 30a | – | – | – | 31.5a | – | 29.5a | |
| N2 area | EV | – | – | – | – | – | – | 29.5a | 30.5a | – | – | – | 31.5a |
| Central area | EV | – | 30.5a | – | 30.5a | 30a | – | – | – | 29.5a | – | – | 31.5a |
| S1 area | EV | – | – | – | – | 31a | – | 33.5, 31a | 34 | – | – | – | 30a |
| HC | – | – | – | – | – | 30a | – | – | 31.5a | – | – | – | |
| S2 area | EV | – | – | 33.5, 31.5a | 33.5 | 33.5 | 33 | 32a | – | 33.5 | – | – | 33.5, 31.5a |
| HC | – | 33.5 | – | 33 | 33.5 | 33.5 | – | 33.5 | 33.5 | – | 33 | 33.5 | |
| MA | 33.5b | – | – | – | – | – | – | – | – | – | – | – | |
| East area | EV | – | – | – | 30a | – | – | 32a | – | – | – | – | 33 |
| HC | – | – | – | – | – | – | – | – | – | – | – | 33 | |
| MA | 33b | – | – | – | – | – | – | – | – | – | – | – | |
a: the one marked with ‘a’ is the threshold identified based on the cumulative effect; the one without ‘a’ is the threshold identified based on the lag effect. b: The studied outcome was all-cause mortality. 1. EV: Emergency visits, HC: Hospitalized cases, MA: Mortality, HRI: Heat-related illnesses, CVD: Cardiovascular Diseases, IHD: Ischemic Heart Disease, RD: Respiratory Disease, COPD: Chronic Obstructive Pulmonary Disease, DM: Diabetes Mellitus, KD: Kidney Diseases, AKF: Acute Kidney Failure. 2. The blank cell means that specific group is not identified as vulnerable by either lag or cumulative effects for the specified health outcome.
In terms of the health outcomes, for the lag effect, RD and urolithiasis are the top two health outcomes associated with increased RRs with p < 0.0001 of more than three sex and age groups in the emergency visits (Fig. 1a). In the hospitalization cases, CVD is the top one associated with increased RRs with p < 0.0001 for the sex and three of the age groups (Fig. 1b). For the cumulative effect, RD and pneumonia are the two health outcomes associated with most of the sex and age groups, with increased RRs in both emergency visits and hospitalization cases (Fig. 2a and b).
Figures 3 and 4 show the highest RRs with p < 0.01 for different health outcomes for different insurance categories with the lag and cumulative effects, respectively. Farmers are the group associated with the highest numbers of different health outcomes, with p < 0.01 in both emergency visits (CVD, stroke, asthma, COPD, DM, and urolithiasis) and hospitalization cases (CVD, IHD, RD, pneumonia, DM, and AKF) with the lag effect (Fig. 3). For example, for CVD, the RRs of the emergency visits and hospitalization cases reach 1.12 (CI: 1.08 – 1.16, p < 0.0001) and 1.10 (CI: 1.07 – 1.14, p < 0.0001) at a threshold of 34 °C, higher than those of the general public and other insurance categories. In the cumulative effect, most of the insurance categories are associated with significant RRs for one or two health outcomes; the union worker group is associated with significant RRs for more than three health outcomes (Fig. 4a and b). Additionally, the farmer and veteran groups are the only two with the lag effect (Fig. 3c), while the low-income group is the only one with the cumulative effect (Fig. 4c), with statistically significant RRs for all-cause mortality.
Fig. 3.
RRs of (a) emergency visits, (b) hospitalized cases, and (c) mortality for different insurance categories (lag effect). Note: 1. Relative risks (RRs) are presented for different insurance categories ranked by different thresholds considering 2-day lag effect. 2. Blank cells mean that (1) there was no sufficient case in that category for statistical analysis or (2) the statistical results were with p ≥0.01.
Fig. 4.
RRs of (a) emergency visits, (b) hospitalized cases, and (c) mortality for different insurance categories (cumulative effect). Note: 1. Relative risks (RRs) are presented for different insurance categories ranked by different thresholds considering 3-day cumulative effect. 2. Blank cells mean that (1) there was no sufficient case in that category for statistical analysis or (2) the statistical results were with p ≥0.01.
Figures 5 and 6 show the highest RRs with p < 0.01 for different health outcomes in different areas with the lag and cumulative effects, respectively. For the lag effect, the S2 area is associated with the highest numbers of health outcomes in emergency visits (IHD, stroke, RD, pneumonia, DM, and urolithiasis) (Fig. 5a) and hospitalization cases (CVD, stroke, RD, pneumonia, COPD, DM, AKF, and urolithiasis) (Fig. 5b). For all-cause mortality, the East Area is associated with statistically significant RR at thresholds lower than those of other areas (Fig. 5c). For the cumulative effect of emergency visits, S2 is still the top area, with RRs associated with most of the health outcomes (Fig. 6a). The N1 area is the top area associated with significant RRs in hospitalization cases for six health outcomes (Fig. 6b). Additionally, S1 is the only area associated with statistically significant RRs for mortality (Fig. 6c).
Fig. 5.
RRs of (a) emergency visits, (b) hospitalized cases, and (c) mortality for different areas (lag effect). Note: 1. Relative risks (RRs) are presented for different areas ranked by different thresholds considering 2-day lag effect. 2. Blank cells mean that (1) there was no sufficient case in that category for statistical analysis or (2) the statistical results were with p ≥0.01.
Fig. 6.
RRs of (a) emergency visits, (b) hospitalized cases, and (c) mortality for different areas (cumulative effect). Note: 1. Relative risks (RRs) are presented for different areas ranked by different thresholds considering 3-day cumulative effect. 2. Blank cells mean that (1) there was no sufficient case in that category for statistical analysis or (2) the statistical results with p ≥0.01.
Based on the aforementioned criteria, Table 3 summarizes the vulnerable groups identified based on both the lag and cumulative effects with the classification of the hyper group, sex and age groups, insurance categories, and geographical areas. These vulnerable groups should be prioritized as targets for heat-health warnings to reduce risks to population health. Targeted warnings for those groups at lower thresholds than the general public could be designed for better public health protection.
It can be seen in Table 3 that females aged above 65 are most vulnerable among all the sex and age groups. They are marked as vulnerable for most of the evaluated health outcomes in emergency visits and hospitalization cases. For the insurance category, all the assessed categories are marked as being vulnerable to at least one health outcome. Farmers, veterans, and service and sales workers are the top three most vulnerable groups associated with increased health risks of at least five different health outcomes on emergency visits. In the hospitalization cases, farmers are the most vulnerable group associated with enhanced RRs for most health outcomes. Regarding the geographical areas, the S2 area is the area marked as most vulnerable to most of the evaluated health outcomes in the emergency visits and hospitalization cases.
Discussion
Our results clearly show that significant associations between Tmax and multiple health outcomes occurred at Tmax values lower than the existing criteria listed in the “High Temperature Information” shown in the Introduction Section, demonstrating the need to revise current thresholds. Notably, the RRs associated with both lag and cumulative effects above certain thresholds are statistically significant. Thus, both lag and cumulative effects should be considered in HHWSs. For the lag effect, since most of the identified thresholds are 34 °C (Table 1a), a warning threshold of 34 °C needs to be added to the current heat warnings in order to reduce the morbidity risks of multiple health outcomes associated with high temperatures. As for the cumulative effect, the identified thresholds varied among different health outcomes for emergency visits, hospitalization cases, and mortality. Since the impacts of high temperature may first manifest most directly as emergency visits for HRI, a warning threshold of 32 °C (the threshold of HRI emergency visits) should be considered in order to reduce the cumulative morbidity risks of multiple heat-health impacts.
Considering the 2 °C difference in the existing thresholds of different categories in the “High Temperature Information”, revised warning thresholds are proposed based on our findings. Four categories were proposed: cautious (yellow), alert (orange), dangerous (red), and highly dangerous (purple). These correspond to the following criteria: (1) yellow—Tmax reaches or exceeds 32 °C; (2) orange—Tmax reaches or exceeds 32 °C for three or more consecutive days, or Tmax reaches or exceeds 34 °C; (3) red—Tmax reaches or exceeds 34 °C for three or more consecutive days, or Tmax reaches or exceeds 36 °C; (4) purple—Tmax reaches or exceeds 36 °C for three or more consecutive days, or Tmax reaches or exceeds 38 °C. Understandably, the CWA needs to consider other factors, such as the frequency of issuing heat warnings, to avoid resource depletion28. The guidance on warning system development highlights different policy concerns, including biometeorology, epidemiology, public health, and risk communication29. The current analysis of morbidity and Tmax for multiple health outcomes provides a solid epidemiological foundation for selecting the proposed thresholds.
Among the evaluated health outcomes, the most notable heat-related health impact is HRI, which is found to have the highest RR in both lag and cumulative effects in both emergency visits and hospitalization cases. The second highest RR is associated with urolithiasis in the lag effect and with AKF in the cumulative effect, respectively, in both emergency visits and hospitalization cases. In addition, most health outcomes are associated with enhanced RR in both emergency visits and hospitalization cases, considering either lag or cumulative effects. Regarding mortality, IHD has the highest RR for the lag effect and is the only health outcome with a statistically significant RR for the cumulative effect.
This study is one of the few epidemiological analyses to consider both the lag and cumulative effects of morbidity risks for the purpose of identifying warning thresholds. The significant cumulative impacts of heatwaves on mortality were reported back in 200930; however, there were only limited studies assessing both the lag and cumulative effects of morbidity risks. In Alaska, both 5-day lag and 2-day cumulative effects were considered to assess the morbidity risks of HRI, CVD, and RD11. In France, various heat wave definitions were compared considering the 1–3-day lag effect and the 3-day cumulative effect of temperature on heat-related mortality31. Our comprehensive study assessed the risks of both morbidity and mortality, considering the 2-day lag and 3-day cumulative effects. Since morbidity precedes mortality, it is advisable to use morbidity results—especially emergency visits if available—as the basis for selecting warning thresholds; this will assist in reducing morbidity risks and subsequent mortality.
It is crucial to design effective responsive actions accompanied by HHWSs to reduce heat-related health risks4. Knowing certain health outcomes with enhanced RRs could prompt hospitals to prepare in advance of heat waves; this includes planning such as ensuring the availability of staffing, medication, and medical equipment to provide proper treatments. Thus, assessing the heat-related impacts of multiple health outcomes is essential, as demonstrated in this study.
Vulnerable groups are the priority for response actions or health adaptation strategies; multiple aspects of vulnerable groups were considered in this work: demographic information, occupation, geographic location, and health status (i.e., hyper group). It was found that the hyper group was indeed more vulnerable to certain diseases (e.g., pneumonia, COPD, and stroke) under the conditions of high temperatures; this was also the case for certain sex and age groups, such as females aged above 65. Among all the insurance categories, farmers are the most vulnerable group, which may be due to the nature of their work, with long hours spent under the sun. As for geographic areas, the S2 area seemed to be the most vulnerable, where may have high percentages of the vulnerable populations. Resources need to be allocated to these vulnerable populations and areas for heat-health risk reduction.
Elderly women have been identified to have a higher heat-related mortality compared to age-matched men in previous epidemiological analyses32–36. The PSU HEAT project (Human Environmental Age Thresholds) confirmed a true sex difference in the heat vulnerability of older adults using a combined biophysical and physiological approach37. The authors defined critical environmental limits as temperature and humidity combinations above which the heat balance cannot be maintained for a given level of metabolic heat production and the core temperature cannot achieve a steady state under prolonged exposures. Their findings showed that elderly women had lower critical environmental limits than age-matched men due to the fact that women typically have lower maximal oxygen consumption, body mass, and sweating rates but greater surface area-to-mass ratios compared to men37. Our findings, which identify elderly women as more vulnerable across all disease categories assessed under high-temperature conditions, are consistent with these previous studies.
Farmers have been identified in previous epidemiological studies as one of the subpopulations most vulnerable to heat-related health risks. For instance, outdoor crop workers in the United States died from heat stroke at a rate nearly 20 times higher than that of all other workers38. In Vietnam, a cross-sectional study conducted in 2021 found that a higher risk of HRI was associated with high-effort farming tasks (odds ratio (OR) = 2.35, CI: 1.34–4.14)39. In Taiwan, female farmers were found to have an RR of 1.48 (CI: 1.06–2.05) for emergency visits due to respiratory diseases, based on national health insurance data from 2009 to 2018, considering lag effects of 0–5 days40. Our results further support these findings, revealing statistically significant increases in RRs for the emergency visits and hospitalizations across most studied health outcomes among farmers when temperatures exceeded 33.5 °C or 34 °C.
Epidemiological findings from regions with similar climate zones were also presented for comparison. A study analyzing data from 18 sites in China during the summer months (June to August) from 2014 to 2017 found that a 1 °C increase in the daily mean temperature was associated with a 1.35% increase in emergency visits (CI: 0.95–1.76%) among populations living in the subtropical monsoon climate zone41. In Vietnam, a 1 °C increase in temperature above 24 °C was significantly associated with a 1.1% increase in all-cause hospital admissions (CI: 0.9–1.4%), a 0.5% increase in cardiovascular disease admissions (CI: 0.1–0.9%), and a 1.3% increase in respiratory disease admissions (CI: 0.9–1.6%)42. Due to differences in the classification of health outcomes across studies, direct comparisons with our results are challenging. Nevertheless, these studies consistently report positive associations between rising temperatures and increases in emergency visits and hospital admissions. Importantly, the present study presents a methodology for identifying temperature thresholds for HHWSs that account for both lag and cumulative effects—an essential consideration for health adaptation that has often been overlooked. These thresholds should be tailored for different regions, as local populations may exhibit varying levels of acclimatization to their typical climate conditions.
This work is a collaboration with policy makers for the purpose of selecting appropriate Tmax thresholds based on their association with multiple health outcomes. Policy makers want to focus on the potential health impacts that occur in the first few days of increased temperatures. If an HHWS is announced, they need to evaluate its effectiveness. Assessing the emergency visits or hospitalized cases in the first few days is much more feasible than assessing those during the long lag period. The coefficient estimates based on the GAMs may be slightly lower than those estimated based on the distributed lag non-linear models, which use long lag periods. Nevertheless, the validity of the association between Tmax and health outcomes will not be affected by the choices of the GAM, which is also a valid model. The obtained coefficient estimates are for lags of 0–2 days rather than long lag periods.
Due to the database’s limitations and the study design’s constraints, certain important personal factors could not be adjusted in the analysis, such as body mass index (BMI). Additionally, since the analysis covered a period of 20 years, certain conditions may vary over the 20-year period. Nevertheless, the most notable intervention strategy, air conditioning, has been prevalent in Taiwan for at least 20 years due to the hot and humid climate. Thus, the prevalence of air conditioning does not significantly impact our results. Moreover, we performed multiple statistical tests, and we did not adjust for multiple comparisons since different statistical significance levels for tests performed for the entire Taiwan population and different groups or areas may cause confusion. To avoid this, we presented p-values for all of the results and purposely plotted data in different colors in Figs. 1, 2, 3, 4, 5 and 6 to show the significance levels so that readers can judge this for themselves. Furthermore, there are many tests with p < 0.0001; we are confident that even with multiple-comparison adjustment, these findings are still statistically significant.
Employing a 20-year population database, this work presents a methodology for identifying warning thresholds and vulnerable populations while considering both the lag and cumulative effects of emergency visits, hospitalization cases, and mortality resulting from multiple health outcomes (including HRI and four prevalent chronic diseases). It was found that these health outcomes had enhanced RRs with 2-day lag and 3-day cumulative effects. Since the RRs associated with lag and cumulative effects above certain thresholds are statistically significant, both lag and cumulative effects should be considered in the HHWS. Four warning categories were proposed—cautious, alert, dangerous, and highly dangerous—with thresholds defined based on the lag and cumulative effects identified in our epidemiological analysis. Vulnerable populations and areas are identified using this methodology. People with hypertension, hyperglycemia, or hyperlipidemia as well as females aged above 65 and farmers are identified as the most vulnerable groups in terms of existing health status, demographic, and occupation, respectively. To establish an evidence-based HHWS, such a comprehensive analysis is a prerequisite. Scientists in countries without proper HHWSs may apply this methodology and contribute to the safety of populations by establishing evidence-based HHWSs; these systems will reduce heat-related health risks under climate change and aid in our collective movement toward sustainable health and wellbeing.
Supplementary Information
Below is the link to the electronic supplementary material.
Supplementary Material 1
One additional file contains Table S1 (ICD-9 and ICD-10 used), Table S2 (the test results of lag-day choices), Table S3 (number of cases used in the epidemiological analysis), Figure S1 (six divisions of Taiwan), Figure S2 (the numbers of days with Tmax above certain thresholds in different divisions), Figure S3 (the numbers of cases with 3 consecutive days of Tmax above certain thresholds in different divisions), Figure S4 (plots of the relationships between Tmax and the studied health outcomes in the emergency visits), Figure S5 (plots of the relationships between Tmax and the studied health outcomes in the hospitalized cases), and Figure S6 (plots of the relationships between Tmax and the studied health outcomes in mortality).
Acknowledgements
We want to express our gratitude to the funding support of Health promotion Administration, Ministry of Health and Welfare from projects numbered N1100101, N1100101-111, and E1111215. We also thank the Health and Welfare Data Science Center of the Ministry of Health and Welfare, Taiwan Central Weather Administration, and Ministry of Environment, Executive Yuan, Taiwan for providing data. The contents of this paper are solely the responsibility of the authors and do not represent the official views of the aforementioned institutes and funding agencies.
Author contributions
S-CC Lung: conceptualization, funding acquisition, methodology, project administration, resources, supervision, visualization, and writing—original draft and editing. J-CJ Yeh: data curation, formal analysis, software, visualization, and writing—review and editing. J-S Hwang: conceptualization, methodology, and writing—review and editing. L-S Chen: conceptualization, resources, and writing—review and editing.
Data availability
The data that support the findings of this study are available from the Ministry of Health and Welfare Data Science Center, but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission from Ministry of Health and Welfare Data Science Center.
Declarations
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.World Meteorological Organization (WMO). WMO confirms 2024 as the warmest year on record at about 1.55°C above pre-industrial level; 2025. Available online: https://wmo.int/news/media-centre/wmo-confirms-2024-warmest-year-record-about-155degc-above-pre-industrial-level.
- 2.World Meteorological Organization (WMO). More extreme heat demands coordinated action. 2024; Available from: https://wmo.int/media/news/more-extreme-heat-demands-coordinated-action.
- 3.United Nations (UN). COP27: $3.1 billion plan to achieve early warning systems for all by 2027. 2022; Available from: https://news.un.org/en/story/2022/11/1130277.
- 4.Casanueva, A. et al. Overview of existing heat-health warning systems in Europe. IJERPH.16(15), 2657. 10.3390/ijerph16152657 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Sorensen, C., Howard, C., Prabhakaran, P., Horton, G. & Basu, R. Heat illnesses in clinical practice. BMJ10.1136/bmj-2022-070762 (2022). [DOI] [PubMed] [Google Scholar]
- 6.United Nations (UN). Sustainable Development Goals (SDGs). 2018; Available from: https://www.un.org/sustainabledevelopment/.
- 7.Vaidyanathan, A. et al. Assessment of extreme heat and hospitalizations to inform early warning systems. PNAS116(12), 5420–5427. 10.1073/pnas.1806393116 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.McLean, K. E. et al. Establishing Heat Alert Thresholds for the Varied Climatic Regions of British Columbia, Canada. IJERPH15(9), 2048. 10.3390/ijerph15092048 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Wu, Y., Wang, X., Wu, J., Wang, R. & Yang, S. Performance of heat-health warning systems in Shanghai evaluated by using local heat-related illness data. Sci. Tot. Environ.715, 136883. 10.1016/j.scitotenv.2020.136883 (2020). [DOI] [PubMed] [Google Scholar]
- 10.Golechha, M., Shah, P. & Mavalankar, D. Threshold determination and temperature trends analysis of Indian cities for effective implementation of an early warning system. Urban CLim.10.1016/j.uclim.2021.100934 (2021). [Google Scholar]
- 11.Hahn, M. B., Kuiper, G. & Magzamen, S. Erratum: “Association of Temperature Thresholds with Heat Illness– and Cardiorespiratory-Related Emergency Visits during Summer Months in Alaska”. EHP10.1289/ehp13433 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Central Weather Administration (CWA). 2022; Available from: https://www.cwa.gov.tw/V8/C/K/Encyclopedia/climate/climate2_list.html.
- 13.Central Weather Administration (CWA). 2023; Available from: https://www.cwa.gov.tw/V8/E/P/Warning/W29.html.
- 14.Taiwan National Development Council (TNDC). Government website open information announcement. 2018; Available from: https://data.gov.tw/en.
- 15.Lin, Y.-K., Chang, C.-K., Li, M.-H., Wu, Y.-C. & Wang, Y.-C. High-temperature indices associated with mortality and outpatient visits: Characterizing the association with elevated temperature. Sci. Tot. Environ.427–428, 41–49. 10.1016/j.scitotenv.2012.04.039 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Cheng, Y.-T., Lung, S.-C.C. & Hwang, J.-S. New approach to identifying proper thresholds for a heat warning system using health risk increments. Environ. Res.170, 282–292. 10.1016/j.envres.2018.12.059 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Lung, S.-C.C., Yeh, J.-C.J. & Hwang, J.-S. Selecting thresholds of heat-warning systems with substantial enhancement of essential population health outcomes for facilitating implementation. IJERPH18(18), 9506. 10.3390/ijerph18189506 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Lin, Y. K., Zafirah, Y., Ke, M. T., Andhikaputra, G. & Wang, Y. C. The effects of extreme temperatures on emergency room visits-a population-based analysis by age, sex, and comorbidity. Int. J. Biometeorol.65(12), 2087–2098 (2021). [DOI] [PubMed] [Google Scholar]
- 19.Guo, Y. et al. Temperature variability and mortality: A multi-country study. Environ. Health Perspect.124(10), 1554–1559. 10.1289/EHP149 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Zhao, Q. et al. Global, regional, and national burden of heatwave-related mortality from 1990 to 2019: A three-stage modelling study. PLoS Med.21(5), e1004364. 10.1371/journal.pmed.1004364 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Alahmad, B. et al. Extreme temperatures and stroke mortality: Evidence from a multi-country analysis. Stroke55, 1847–1856. 10.1161/STROKEAHA.123.045751 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Chou CCK, Lung S-CC, Hsiao T-C, Lee C-T. Regional and Urban Air Quality in East Asia: Taiwan. Springer Nature Singapore. 469–506 (2023).
- 23.Grigorieva, E. & Lukyanets, A. Combined effect of hot weather and outdoor air pollution on respiratory health: Literature review. Atmosphere10.3390/atmos12060790 (2021). [Google Scholar]
- 24.Zhang, Y. C. et al. The impact of ambient temperature on the incidence of urolithiasis: A systematic review and meta-analysis. Scand J. Work Env. Hea.46(2), 117–126. 10.5271/sjweh.3866 (2020). [DOI] [PubMed] [Google Scholar]
- 25.Vicedo-Cabrera, A. M. et al. The burden of heat-related mortality attributable to recent human-induced climate change. Nat. Climate Change11(6), 492–500. 10.1038/s41558-021-01058-x (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Gasparrini, A. et al. Mortality risk attributable to high and low ambient temperature: A multicountry observational study. Lancet386(9991), 369–375. 10.1016/s0140-6736(14)62114-0 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Kliengchuay, W. et al. Effect modification of fine particulate matter (PM2.5) related hospital admissions by temperature in Thailand: A nationwide time-series study. Environ. Res.10.1016/j.envres.2025.121467 (2025). [DOI] [PubMed] [Google Scholar]
- 28.Lung, S.-C.C., Liou, M.-L., Yeh, J.-C.J. & Hwang, J.-S. A pilot heat-health warning system co-designed for a subtropical city. PLoS ONE18(11), e0294281. 10.1371/journal.pone.0294281 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.World Meteorological Organization and World Health Organization (WMO & WHO) Heatwaves and Health: Guidance on Warning-System Development. WMO-No. 1142, 2015; World Meteorological Organization, Geneva, Switzerland
- 30.Anderson, G. B. & Bell, M. L. Heat waves in the United States: mortality risk during heat waves and effect modification by heat wave characteristics in 43 U.S. communities. Environ. Health Perspect.119(2), 210–218. 10.1289/ehp.1002313 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Kanti, F. S., Alari, A., Chaix, B. & Benmarhnia, T. Comparison of various heat waves definitions and the burden of heat-related mortality in France: Implications for existing early warning systems. Environ. Res.215(Pt 2), 114359 (2022). [DOI] [PubMed] [Google Scholar]
- 32.Simón, F., Lopez-Abente, G., Ballester, E. & Martínez, F. Mortality in Spain during the heat waves of summer 2003. Euro. Surveill.10(7), 156–161. 10.1007/s11356-021-16601-8 (2005). [PubMed] [Google Scholar]
- 33.Fouillet, A. et al. Excess mortality related to the August 2003 heat wave in France. Int. Arch. Occup. Environ. Health80(1), 16–24. 10.1007/s00420-006-0089-4 (2006). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Achebak, H., Devolder, D. & Ballester, J. Trends in temperature-related age-specific and sex-specific mortality from cardiovascular diseases in Spain: A national time-series analysis. Lancet Planet Health3(7), e297–e306. 10.1016/S2542-5196(19)30090-7 (2019). [DOI] [PubMed] [Google Scholar]
- 35.van Steen, Y., Ntarladima, A. M., Grobbee, R., Karssenberg, D. & Vaartjes, I. Sex differences in mortality after heat waves: Are elderly women at higher risk?. Int. Arch. Occup. Environ. Health92(1), 37–48. 10.1007/s00420-018-1360-1 (2019). [DOI] [PubMed] [Google Scholar]
- 36.Ballester, J. et al. Heat-related mortality in Europe during the summer of 2022. Nat. Med.29(7), 1857–1866. 10.1038/s41591-023-02419-z (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Leach, O. K., Cottle, R. M., Fisher, K. G., Wolf, S. T. & Kenney, W. L. Sex differences in heat stress vulnerability among middle-aged and older adults (PSU HEAT Project). Am. J. Physiol. Regul. Integr. Comp. Physiol.327(3), R320–R327. 10.1152/ajpregu.00114.2024 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Centers for Disease Control and Prevention-USA, Heat-related deaths among crop workers—United States, 1992-2006, JAMA, 300 (9) :1017, (2008) 10.1001/jama.300.9.1017
- 39.Tran, N. Q. L. et al. The prevalence of heat-related illnesses and associated factors among rice farmers in Vietnam. J Agromedicine28(3), 486–496. 10.1080/1059924X.2022.2154086 (2022). [DOI] [PubMed] [Google Scholar]
- 40.Sharma, A. et al. Emergency room visits (ERVs) among occupational groups associated with ambient conditions in Taiwan. Int. Arch. Occup. Environ. Health97(7), 779–789. 10.1007/s00420-024-02084-w (2024). [DOI] [PubMed] [Google Scholar]
- 41.Wang, Y. et al. High temperatures and emergency department visits in 18 sites with different climatic characteristics in China: Risk assessment and attributable fraction identification. Environ. Int.136, 105486. 10.1016/j.envint.2020.105486 (2020). [DOI] [PubMed] [Google Scholar]
- 42.Talukder, M. R., Chu, C., Rutherford, S., Huang, C. & Phung, D. The effect of high temperatures on risk of hospitalization in northern Vietnam. Environ. Sci. Pollut. Res. Int.29(8), 12128–12135. 10.1007/s11356-021-16601-8 (2022). [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary Material 1
One additional file contains Table S1 (ICD-9 and ICD-10 used), Table S2 (the test results of lag-day choices), Table S3 (number of cases used in the epidemiological analysis), Figure S1 (six divisions of Taiwan), Figure S2 (the numbers of days with Tmax above certain thresholds in different divisions), Figure S3 (the numbers of cases with 3 consecutive days of Tmax above certain thresholds in different divisions), Figure S4 (plots of the relationships between Tmax and the studied health outcomes in the emergency visits), Figure S5 (plots of the relationships between Tmax and the studied health outcomes in the hospitalized cases), and Figure S6 (plots of the relationships between Tmax and the studied health outcomes in mortality).
Data Availability Statement
The data that support the findings of this study are available from the Ministry of Health and Welfare Data Science Center, but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission from Ministry of Health and Welfare Data Science Center.



















