Introduction

There has been a growing concern over global warming. By the end of the twenty-first century, the global average surface temperature will be very likely to rise by 1.0–5.7 °C (depending on the greenhouse gases emission scenario) compared to the pre-industrial period of 1850–1900 (Intergovernmental Panel on Climate Change 2021 & 2022). The heat-related adverse outcomes are associated with the working population who are exposed to occupational hazard (Gubernot et al., 2015). Meanwhile, the older adults are more vulnerable to the adverse effects of hot weather due to their weakened physiological responses (Meade et al. 2020). These may lead to greater healthcare utilization, including emergency department and inpatient admissions, and mortality (Basu et al. 2012; Bishop-Williams et al. 2015; L i et al. 2015).

Traditionally, the effects of individual meteorological variables on health and healthcare utilization have been investigated. For instance, in a city with subtropical climate in Hong Kong, hospital admissions increased by 4.5% for every 1 °C increase above 29 °C in the hot season (Chan et al. 2013), and a rise in diurnal temperature range by 1 °C was associated with 2.5% increase in emergency asthma hospitalization (Qiu et al., 2015). In addition, a study showed that health-seeking behaviors among the older population started to increase when the maximum ambient temperature reached 30–32 °C (Chan et al. 2011). Positive associations between ambient temperature and morbidity and mortality during hot seasons have also been reported in other climate zones (Dessai 2002; Díaz et al. 2002; Wang et al. 2020; Ye et al. 2012).

As the thermal stress experienced by the human body cannot be explained by ambient temperature alone, there has been an increase in usage of composite indices that incorporate additional meteorological variables, such as relative humidity, wind speed, and radiant temperature, along with ambient temperature, into a single thermal index. In Europe, the associations of maximum apparent temperature with respiratory hospitalization among the older adults and with all-causes mortality have been revealed (Michelozzi et al. 2000; Baccini et al. 2008; Michelozzi et al. 2009) . Various studies in Hong Kong have investigated the associations between health outcomes and these indices during the hot season. Although heat stroke mortality was reported to double when the net effective temperature (NET) increased by one unit above 26 °C (Leung et al. 2008), a significant relationship with all-cause mortality could not be shown with the Weather Stress Index (WSI) (Li & Chan 2000). Subsequently, a Hong Kong specific index, namely, the Hong Kong Heat Index, was developed based on its association with excess hospitalization to better reflect the situation (Lee et al. 2016). Although such an index has practical value in the local context, it may not be comparable to studies in other regions.

The Universal Thermal Climate Index (UTCI) was developed in 2009. This index is intended to assess outdoor thermal stress, with the advantage of being applicable to regions of different climates (Błażejczyk et al. 2010; Jendritzky et al. 2012). The UTCI is defined as “the isothermal air temperature of the reference condition that would elicit the same dynamic response of the physiological model” (Jendritzky et al., 2012). The UTCI incorporates ambient temperature, humidity, wind speed, and radiant temperature into a score based on a multivariate dynamic model, with higher scores indicating greater heat stress (Błażejczyk et al., 2013). The UTCI not only incorporates multiple meteorological variables in a non-linear way, but, more importantly, it also classifies thermal stress based on human physiological responses. Ten thermal stress classifications, ranging from extreme cold stress (UTCI ≤  − 40.0 °C), to thermoneutral conditions (UTCI between 9.1 and 26.0 °C), and extreme heat stress (UTCI > 46.0 °C), were proposed to describe the extreme weather stress experienced by humans (Błażejczyk et al., 2010). However, investigations regarding the association between thermal stress and health outcomes, particularly heat stress, have been mostly conducted in Europe, which generally has a temperate climate. For example, a study in the Czech Republic reported that when the UTCI reached 22 °C, cardiovascular mortality exceeded the mean value by 11.8% (Urban & Kyselý, 2014). A study in Poland reported increased asthma hospitalization among populations younger than 65 years of age for an UTCI between 26.1 and 32.0 °C compared to days with thermoneutral UTCI (Romaszko-Wojtowicz et al., 2020). Another study in Poland reported an increase in mortality by 10–20% for UTCI exceeding 32.0 °C and by 25–30% for UTCI exceeding 38.0 °C (Kuchcik et al., 2021). A study from Athens showed the odds of self-reported heat-related symptoms such as exhaustion, headache, dizziness, and breathing difficulties, increased by 4% per 1 °C increase in UTCI (Krüger, 2021). Among the few studies from Asia, a study conducted in Beijing, a city characterized by a hot summer and continental climate, showed that cerebral infarction emergency room visits among those aged 45 to 65 years increased by 64% on lag 0–2 days when the UTCI reached 38 °C (Ma et al. 2018). Another study in Bangladesh, a country with a tropical monsoon climate, showed that with every 1 °C increase in the UTCI above 34–35 °C, the all-cause mortality increased by 31.3% and the cardiovascular mortality increased by 20%. These relationships were more pronounced among men and those aged > 65 years (Burkart et al., 2014). There is limited amount of research on the health outcomes of heat stress, as defined by the UTCI in other climate zones.

Whether such associations exist in other climate zones, particularly in subtropical climate zones, remains uncertain. Long-term acclimatization could result in differences in the summer thermal comfort thresholds in different climate zones. For example, the population in Hong Kong, a city with a subtropical climate, reported a neutral to warm sensation at a higher UTCI range compared to the dynamic thermal sensation model (Lam & Lau 2018). Hence, there is a need to quantify the adverse health risks in terms of UTCI in subtropical climate zones during hot seasons. This study aimed to investigate the association between UTCI and visits to the accident and emergency department (AED) in a subtropical climate region, with Hong Kong as an illustration.

Method

Study design

This was a retrospective study based on administrative records.

Data sources

This study was performed using data on daily healthcare utilization in the public sector among the Hong Kong population from 2000 to 2016. The years 2003 and 2009 were excluded due to the occurrence of the SARS epidemic and swine flu pandemic, respectively, as the usual healthcare utilization pattern may have been affected. Healthcare utilization data were obtained from the Clinical Management System (CMS), a database of the Hospital Authority that records all hospital admission, and discharge data of patients from the public sector in Hong Kong. The data included the number of daily AED visits in all 18 accident and emergency units operated by the Hospital Authority, stratified by sex (female, male), and age (≤ 18, 19–64, ≥ 65 years). In Hong Kong, those aged ≤ 18 years are generally classified as pediatric patients and those aged > 65 years as geriatric patients. To avoid non-linear relationships arising from whole-year data analyses, only data from the summer season were analyzed. In accordance with the literature, we used May to September to represent the hot season in Hong Kong (Chau et al. 2009; Lee et al. 2016), resulting in 2295 days in the analysis set.

The corresponding population data, stratified by sex and age groups, were obtained from the Census and Statistics Department of Hong Kong. The hourly meteorological data, including ambient temperature, relative humidity, wind speed, global solar radiation, and weather warning signals, including typhoons and rainstorms, were obtained from the Hong Kong Observatory (HKO). The daily general air quality health index (AQHI) (Wong et al. 2013) was obtained from the Environmental Protection Department. Indicator variables on public holidays and Sundays were generated according to the Hong Kong calendar.

Dependent variable

The primary outcome was the all-cause daily AED visit rate (per 100,000 population). The age- and sex-specific rates were calculated by dividing the daily number of corresponding admissions by population, and then multiplying with 100,000.

Independent variables

UTCI was calculated based on ambient temperature, humidity, wind speed, and mean radiant temperature obtained from the HKO. The details of the adjustment to the meteorological variables were presented elsewhere (Lam & Lau 2018). In particular, wind speed was adjusted by the height of the weather station and black globe temperature was used to proximate the mean radiant temperature. The hourly UTCI was calculated according to the Fortran code provided on the UTCI website (ISB Commission 6, 2014). The daily maximum UTCI was selected to reflect heat stress since human beings were hypothesized to be vulnerable to extreme weather conditions. According to the thresholds of the UTCI (Błażejczyk et al., 2010), thermoneutral conditions referred to UTCI between 9.1 and 26.0 °C, moderate heat stress referred to UTCI between 26.1 and 32.0 °C, strong heat stress referred to UTCI between 32.1 and 38.0 °C, and very strong heat stress referred to UTCI between 38.1 and 46.0 °C. In Hong Kong, none of the days in the analysis set was having extreme heat stress (UTCI ≥ 46.1 °C). Thermoneutral condition and moderate heat stress were combined as a group because there were too few summer days in Hong Kong that could be considered thermoneutral. Thus, the daily maximum UTCI between 9.1 and 32.0 °C was categorized as thermoneutral to moderate heat stress condition. To reflect prolonged heat stress, the number of days of strong to very strong heat stress (32.1 °C ≤ UTCI ≤ 46.0 °C) over the past 7 days was computed as a continuous variable. A period of 7 days was considered because the association between hot weather and health outcomes was generally limited to a few days (Wang et al. 2021).

Since there is evidence in the literature supporting the influence of air quality on AED visits (Szyszkowicz, 2019), the AQHI, which is a health risk-based multi-pollutant index, was controlled in the model. The air quality was reflected by AQHI, which was calculated based on “the cumulative health risk attributable to the 3-h moving average concentrations of four air pollutants including ozone (O3), nitrogen dioxide (NO2), sulfur dioxide (SO2), and particulate matter (PM2.5/PM10)” (Environmental Protection Department 2020). According to the thresholds of the AQHI (Environmental Protection Department 2020), daily general AQHI between 4 and 10 was categorized as moderate to very high health risk, and between 1 and 3 was categorized as low health risk. During the study period, none of the days had a general AQHI exceeding 10, and there were 13 days with missing readings. Missing readings were replaced with the average of the readings immediately prior to and after the missing data. Furthermore, the number of days of air quality of moderate to very high health risk over the past 7 days was computed as a continuous variable to reflect prolonged poor air quality.

Typhoons have been associated with AED visits (Lin et al. 2013). Hence, the occurrence of typhoons was controlled in the model. In Hong Kong, there are five levels of typhoon warning signals (signal numbers 1, 3, 8, 9, and 10). When a typhoon is categorized as signal number 8, the wind is expected to reach the gale force. Therefore, in this study, an ordinal variable (none, signal number 3 or below, signal number 8 or above) was applied. There is also a rainstorm warning system in Hong Kong, with three levels of warning, namely Amber, Red, and Black, in ascending order of severity. In this study, the rainstorm warning system was considered an ordinal variable (none, amber, red, and black).

While there were inconsistencies related to a holiday, day of the week, and AED utilization (Castner et al. 2016; Hahn et al. 2018), these were still controlled in the model as independent variables, with the aim of controlling for any possible influence. A binary variable, public holiday, was created according to the public holiday calendar of Hong Kong. The day of the week was included as the nominal variable. Summer vacation was defined as 15 July to 31 August as this was the most common period for summer vacation among primary and secondary school students. As the AED service was free of charge before December 2002 in Hong Kong, a binary variable indicating whether the service was free of charge was included. The year was also controlled as a continuous variable to cater to any potential trends in the population.

Data analysis

Data analyses were conducted for the six population subgroups based on sex and age separately. Descriptive statistics, including frequencies and percentages, were used to summarize the meteorological characteristics of the study days. For each subgroup, the analysis of covariance (ANCOVA) model was used to investigate the associations between the dependent and independent variables. AED visit rate was used as the dependent variable. Categorical independent variables were factors and continuous independent variables were covariates. Heat stress and prolonged heat stress were used as factor and covariate respectively. The models were also adjusted for air quality, prolonged poor air quality, typhoon warning, rainstorm warning, fee charging, public holiday, day of the week, and year by including them as factors or covariates as appropriate. For population aged ≤ 18, summer vacation was also included as a factor. For covariates, the fitted coefficients referred to the change in the mean AED visit rate (per 100,000 population) per unit increase in the corresponding variables. For factors, the fitted coefficients were the comparison of means between different levels of the factors and its reference level, which was “thermoneutral conditions to moderate heat stress” for the factor heat stress, “low health risk” for the factor air quality, “Saturday” for the factor day of the week, “free of charge” for the factor fee of AED visit, and “no/none” for the remaining factors. Partial eta squared (ηp2) was used to quantify the effect size (Cohen 1973), with 0.01 considered a small effect, 0.06 a medium effect and 0.14 a large effect, respectively. The model diagnostics of linearity and normality assumptions were checked. Variance inflation factor (VIF) was used to assess multicollinearity among the independent variables (factors were treated as dummy), with VIF > 4 indicating possible multicollinearity. All statistical analyses were performed using SPSS version 25 (IBM Corp., Armonk, NY, USA). The significance level was set to 5%.

Results

Descriptive statistics

Figure 1 shows the percentage of days with different levels of heat stress in each month over the years 2000 to 2016. The number of days with different levels of heat stress during May to September of 2000 to 2016 is shown in Fig. 2. Comparing to thermoneutral or moderate heat stress, the chance of having a very strong heat stress day and strong heat stress day increased by 15.3% (Odds ratio (OR): 1.153, 95% CI: 1.123, 1.183) and 4.2% each year (OR: 1.042, 95% CI: 1.021, 1.063), respectively. Figure 3 shows the percentage of days with different levels of health risk related to air quality in each month over the years 2000 to 2016. The number of days with different levels of health risk related to air quality in May to September 2000 to 2016 is shown in Fig. 4. Comparing to low health risk, the chance of having a moderate to very high health risk did not show a significant trend (OR: 1.000, 95% CI: 0.983, 1.017).

Fig. 1
figure 1

The percentage of days with different levels of heat stress in each month, 2000 to 2016

Full size image
Fig. 2
figure 2

The number of days with different levels of heat stress in May to September 2000 to 2016

Full size image
Fig. 3
figure 3

The percentage of days with different levels of health risk related to air quality in each month, 2000 to 2016

Full size image
Fig. 4
figure 4

The number of days with different levels of health risk related to air quality in May to September 2000 to 2016

Full size image

The characteristics of the data from 2295 days between May 2000 and September 2016 (excluding the years 2003 and 2009) are presented in Table 1. Most of the studied days had strong heat stress (58.6%) or very strong heat stress (20.7%). Approximately 70.9% of the studied days had air quality associated with low health risk. Only 1.3% of the days had a typhoon warning signal number of eight or more. Only 1.8% and 0.6% of the days had rainstorm warning signals red and black, respectively. Less than 20% of the studied days were public holidays (including Sundays).

Table 1 Characteristics of 2295 days in May to September of the period 2000 to 2016 (excluding 2003 and 2009)
Full size table

During the study period, 13,438,846 AED visits occurred. The population aged ≥ 65 years had the highest daily AED visit rate. Descriptive statistics of the age-and sex-specific AED visit rates are presented in Table 2. For females aged ≥ 65 years, the daily admission rates (per 100,000 population) ranged from 105.1 to 227.9, whereas the male counterparts had a range of 117.7 to 225.7. For females aged ≤ 18 years, the daily rates (per 100,000 population) ranged from 40.4 to 151.1 and from 51.1 to 179.1 for males aged ≤ 18 years. For those aged 19–64 years, the daily rates (per 100,000 population) ranged from 52.6 to 96.4 for females and 56.2 to 103.3 for males.

Table 2 Descriptive statistics of daily AED visit rates (per 100,000) on the 2295 days studied
Full size table

Association between heat stress and AED visit rates

ANCOVA models on AED visit rates were fitted for each age and sex subgroup. The fitted coefficients are listed in Tables 3 and 4, respectively. The model diagnostics did not violate the assumptions of the model. Heat stress was unlikely to have multicollinearity with other variables (VIF ranges from 1.0 to 2.1), although the variable holiday was related to Sunday (VIF > 5).

Table 3 ANCOVA models for daily AED admission rates (per 100,000) among females
Full size table
Table 4 ANCOVA models for daily AED admission rates (per 100,000) among males
Full size table

The results showed that a day with strong to very strong heat stress was significantly associated with higher AED visit rates among all age and sex groups (p ranged from < 0.001 to 0.048). Heat stress attained a small-to-moderate effect size for males aged 19–64 years (ηp2 = 0.037), a small effect size for females aged ≥ 65 years (ηp2 = 0.016), and males aged ≥ 65 years (ηp2 = 0.014), but negligible for those aged ≤ 18 years (ηp2 = 0.003 for females and 0.005 for males) and females aged 19–64 years (ηp2 = 0.008). After adjusting for other possible factors, on a day with strong heat stress, the AED visit rate (per 100,000) increased by 1.7 (95% CI: 1.3, 2.1) among males aged 19–64 years, 4.1 (95% CI: 2.7, 5.4) among females aged ≥ 65 years, and 4.1 (95% CI: 2.6, 5.6) for males aged ≥ 65 years, whereas on a day with very strong heat stress, such rates increased by 2.2 (95% CI: 1.7, 2.7), 4.9 (95% CI: 3.1, 6.7), and 4.7 (95% CI: 2.7, 6.6), respectively. The days with very strong heat stress was associated with more AED admissions as compared to days with strong heat stress, except for females aged 19–64 years.

Prolonged heat stress was associated with lower AED visit rates among those aged ≤ 18 years and those aged ≥ 65 years (p < 0.001 for all groups), but higher AED visit rate among those aged 19–64 years (p = 0.007 for females and p < 0.001 for males). For those aged ≤ 18 years, the AED visit rate (per 100,000) decreased by 0.4 (95% CI: 0.2, 0.7) for females and 0.6 (95% CI: 0.3, 0.9) for males, for an additional day of heat stress following the past 7 days, the effect size (ηp2 = 0.005 for females and 0.007 for males), though negligible, was greater than that from heat stress on the eighth day. Among those aged ≥ 65 years, the effect size from prolonged heat stress (ηp2 = 0.019 for females and 0.012 for males) was comparable to that from heat stress on the eighth day. For an additional day of heat stress following the past 7 days, the AED visit rate (per 100,000) decreased by 1.1 (95% CI: 0.8, 1.4) for females aged ≥ 65 years and 0.9 (95% CI: 0.6, 1.3) for males aged ≥ 65 years. Among those aged 19–64, an additional day of heat stress following the past 7 days was associated with an increase in the AED visit rate (per 100,000) (females: 0.1 (95% CI: 0.04, 0.2); males: 0.3 (95% CI: 0.2, 0.4)), with an effect size (ηp2 = 0.003 for females and ηp2 = 0.021 for males) slightly smaller than that for the eighth day.

Association between air quality and AED visit rates

The results showed that a day with air quality of moderate to very high health risk was significantly associated with a higher AED visit rate among those aged ≤ 18 years only (p = 0.002 for females and p < 0.001 for males), but not for those aged 19–64 years and ≥ 65 years. In the adjusted model, on a day with air quality of moderate to very high health risk, the AED visit rate (per 100,000) increased by 1.6 (95% CI: 0.6, 2.6) among females aged ≤ 18 years and by 2.8 (95% CI: 1.7, 4.0) among males aged ≤ 18 years. Such effect sizes were small for males (ηp2 = 0.010) and negligible for females (ηp2 = 0.004).

Prolonged poor air quality was associated with a lower AED visit rate among males aged 19–64 years (p = 0.020), the effect size was negligible (ηp2 = 0.002). Insignificant associations were observed for other age and sex subgroups.

Association between other factors and AED visit rates

Among all other controlled factors, the AED visit fees had large impact on AED visits for all groups (p < 0.001 for all groups) except males aged ≥ 65 years. The AED visit rate was substantially lower in the charging period compared to the free of charge period. For those aged ≤ 18 years, during the charging period, AED visit rate (per 100,000) decreased by 21.7 (95% CI: 20.3, 23.2; ηp2 = 0.282) for females and 26.8 (95% CI: 25.2, 28.5; ηp2 = 0.307) for males. For those aged 19–64 years, such rates decreased by 16.5 (95% CI: 15.9, 17.0; ηp2 = 0.604) for females and 16.7 (95% CI: 16.2, 17.2; ηp2 = 0.643) for males. For females aged ≥ 65 years, such rate decreased by 15.8 (95% CI: 14.0, 17.6; ηp2 = 0.115). For males aged ≥ 65 years, such rated increased by 4.8 (95% CI: 2.9, 6.8; ηp2 = 0.011).

For those aged ≤ 18 years, summer vacation was associated with lower AED visits (p < 0.001 for all, ηp2 = 0.298 for females and 0.344 for males), with AED visit rate (per 100,000) decreased by 14.3 (95% CI: 13.4, 15.2) for females and 18.4 (95% CI: 17.4, 19.5) for males during summer vacation.

There was a decreasing trend in the AED visit rate for those aged ≥ 65 years, but an increasing trend for those aged ≤ 18 years and 19–64 years. The day of the week was significant in all models, with the “Monday effect” observed in all subgroups (ηp2 = 0.005 and 0.007 for females and males ≤ 18 years, respectively, and 0.105 to 0.336 for other subgroups). Although public holidays were significant in all models, the direction of the effect was different. All those aged ≤ 18 years and females aged 19–64 years had increased AED visits on public holidays, whereas males aged 19–64 years and all those aged ≥ 65 years had decreased visits on those days. However, it should be noted that the public holiday variable and Sunday had some extent of dependency (VIF > 5), fitted coefficients had to be interpreted with care. The impact of typhoons and rainstorms also varied across the different subgroups.

Discussions

Based on hospital administration records, this study demonstrated significant associations between the daily AED visit rate and heat stress as defined by UTCI, using Hong Kong as an example of a subtropical climate. This study not only contributed to expand our knowledge by providing evidence that such associations exist in subtropical climates, but also highlighted that the strength of these associations varied across different age- and sex-specific population subgroups. Furthermore, the association with prolonged heat stress correlated differently in both direction and strength with the AED visit rate among the different subgroups.

Heat stress can be defined by a number of indicators, ranging from ambient temperature to composite indices. In this study, the UTCI was used to define heat stress. A previous local study suggested that the UTCI is capable of determining the thermal stress experienced by the human body (Cheung & Hart 2014). In the current study, we found that in the period studied, almost 80% of the days were considered to have strong or very strong heat stress. Despite the common occurrence of heat stress, significant associations with AED visits have been reported.

Our study showed that heat stress was more influential in adult males and the older population (small-to-moderate effect size (ηp2 = 0.037) for males aged 19–64 years and small effect size (ηp2 = 0.016 and 0.014) for females and males aged ≥ 65 years), but negligible effect size in children and adult females (for females and males aged ≤ 18 years (ηp2 = 0.003 and 0.005) and females aged 19–64 years (ηp2 = 0.008)). Our findings that AED visits among adult males showed the strongest association with heat stress was consistent with the review that adults showed stronger association than the older adults, particularly AED visits related to dysrhythmia, ischemic heart disease, and hypotension (Basu et al., 2012). In Spain, it was reported that the heat-related mortality was observed for middle-aged males but not females (Díaz et al., 2006). It was consistent with our results in the way that males aged 19–64 exhibited larger effect size than their female counterparts, although, in our study, both gender showed significant associations with heat stress.

The working population, aged 19–64 years, may have been working outdoors, for instance, those engaged in the construction and logistics industries. Heat stress in workplaces would cause heat exhaustion and heat stroke if proper precautions are not implemented (Kjellstrom et al., 2020). The excessive heat stress faced by the construction workers has been reviewed by Rowlinson et al. (2014) . In Hong Kong, nearly 10% of the working population is in the construction industry, along with several other industries requiring outdoor workers. A study reported that about 17.5% of the work in the hot summer months of the Hong Kong construction workers were at high risk and 34.2% at moderate risk, and the heart rate during the work period mostly belonged to 55–69% maximum heart rate (Yi & Chan 2017). The physical work capacity showed significant reduction by 78% when wet bulb globe temperature reached 40 °C (Foster et al. 2021). Moreover, a study conducted in Beijing reported that emergency admissions related to cerebral infarction had the strongest association with heat stress among those aged 45–64 years, when UTCI reached 38 °C, such occurrence increased by 64% (Ma et al. 2018). In the case of young adults, an increase in their AED visits may be related to their higher frequency of engagement in outdoor activities, which exposes them to a greater risk of emergency conditions. According to a behavioral risk factor survey in Hong Kong, 3.6% of young adults aged 18 to 24 years experienced a heat stroke in 2015 (Bacon-Shone 2016).

In older adults, the adverse health outcomes from heat stress could be explained by their weakened ability to detect heat and trigger appropriate behavioral and physiological reactions for body temperature regulation. Similarly, those with comorbidities and under medication may also have a reduced ability to detect and respond to heat (Kenny et al. 2010). In a subtropical climate region, thermoregulation becomes more challenging during hot and humid summers, wherein heat is less likely to escape from the human body through radiation and evaporation. Failure in thermoregulation induces various cardiovascular responses and changes in fluid regulation, thus resulting in a higher risk of adverse health events, such as heat stroke, acute myocardial infarction, thrombosis, ischemic stroke, electrolyte imbalances, and acute kidney injury (Meade et al. 2020). These acute conditions require emergency health services.

In children, their underdeveloped thermoregulatory system and higher metabolic rate may render them vulnerable to heat stress; an association between heat waves and several cause-specific morbidities, such as renal disease, respiratory disease, electrolyte imbalance, and fever, has been reported in the literature (Xu et al. 2014). However, our study did not show a significant association between AED visits and heat stress among children. Possible reasons may include the use of all-cause AED visits instead of cause-specific ones, and the control of air pollution and other factors as variables. Further research is warranted to investigate these inconsistencies.

Literature showed the prolonged heat wave in Poland, with midday UTCI means at three standard deviations above that of the period of 1951–2018, increased the total mortality and strong heat-related mortality by 10% and 5-folds in June 2019, respectively, as compared to 2010–2018 (Błażejczyk et al. 2022). We found that prolonged heat, defined by the total number of days with strong to very strong heat stress over 7 days, showed varied correlations with AED visits for different age groups. Males aged 19–64 years had small-to-moderate effect size (ηp2 = 0.021), females and males aged ≥ 65 years had small effect size (ηp2 = 0.019 and 0.012), but negligible effect size for females and males aged ≤ 18 years (ηp2 = 0.005 and 0.007) and females aged 19–64 years (ηp2 = 0.003). While prolonged heat was associated with fewer AED visits among older adults and children, it was associated with more AED visits among adults aged 18–64 years. The negative correlation may be explained by the survival of the fittest concept (i.e., those who were weaker may have already become sick earlier during the 7 days), which is likely for the case of the older adults and children. Due to their lower capacity in thermoregulation, they may be hospitalized once facing heat stress. Those who could survive prolonged heat stress are likely to be healthier. For adults, while they could stand longer heat stress, each day of heat stress did post additional risks to them. Therefore, the longer the period for which they were exposed to heat stress, the higher the risk of acute health events. Previous studies also reported a relatively smaller effect of prolonged stress-induced mortality compared to heat stress on a particular day. For instance, Gasparrini and Armstrong (2011) reported that after four consecutive heat stress day, the added risk of mortality was only 0.2–0.8%, whereas the main effect was 4.9–8.0%. Another local study reported that at least five non-consecutive Hot Days and Hot Nights within a week was associated with a 15.6% increase in all-cause mortality at lag 0–1 but a 2% decrease at lag 2–3 (Ho et al. 2017). In Brisbane, a city with a subtropical climate, a 50.6% increase in ambulance attendances was shown to be associated with a 9.5 °C increase above the ambient temperature of 29 °C, but prolonged heat stress from heat waves had a significant but smaller effect (18.8% increase) compared to acute heat stress (Turner et al. 2013). Focusing on mortality among older adults in Europe, long heat wave had larger effect than short heat wave (D’Ippoliti et al., 2010). On the other hand, it was reported that in Europe, heat-related mortality rose rapidly with heat stress and the lag effect was only about one (Błażejczyk & McGregor, 2008; Laschewski & Jendritzky 2002). Nevertheless, prolonged heat stress could be confounded by lagged effects, as reported in the literature (Ye et al. 2012). Further research is warranted to examine the associations and, more importantly, to explore the underlying mechanisms.

The potential risk factors were controlled in the models. While the AED visit fees, day of the week, public holidays, and years had an influential impact on the rate of AED visits, heat stress and/or prolonged heat stress remained significant after controlling for these factors. The impact of air quality on AED visits was minimal for adults and older adults. Although a multi-city study in China found that the association between mortality and air pollution was amplified by high temperatures, a 10 μg/m3 increase in PM2.5 was found to associate with 0.33–0.52%, 0.39–0.98%, and 1.32–2.25% increase in cardiovascular mortality when the daily ambient temperature as at < 10th, 10th–90th, and > 90th percentiles (Zhang et al. 2020). On the other hand, a study conducted in Beijing suggested that the association between air pollutants and respiratory emergency department visits was significantly stronger on moderately cold days (relative risk being 1.006 for 10 μg/m3 increase in PM2.5) than that on cold days (relative risk being 1.004), but insignificant differences on moderately hot days and hot days (Song et al. 2021). As our study period only included summers, the effects of air quality may not have been so obvious. In contrast, among children, air quality had an effect size (ηp2 = 0.004 for females and 0.010 for males) comparable to that from heat stress (ηp2 = 0.003 for females and 0.005 for males). This may be because over 70% of AED visits among pediatric patients were influenza related (Kwong et al., 2009), and thus, air pollution is more likely to be a trigger.

Our findings were consistent with local evidence supporting the association between hot ambient temperatures. For instance, the relative risk for respiratory-related hospitalization on a day with ambient temperature > 90th percentile (29.5 °C) and > 99th percentile (30.6 °C) was estimated to be 1.06 and 1.14 respectively (Sun et al. 2019); the cumulative relative risk of asthma hospitalization for lags 0–3 days was 1.2 for a day with ambient temperature of 30 °C as compared to 27 °C (Lam et al. 2016); increase of 10 hot degree days (sum of daily ambient temperature in degree above 29.3 °C) in a year was associated with around 2% increase in the annual age-standardized mortality (Goggins et al., 2015). However, perceived thermal comfort can vary depending on the humidity, wind speed, and radiation. Hence, the effect size of hot weather based on ambient temperature may be more suitable in the local context. On the other hand, UTCI has the advantage of considering thermal stress in the development of thresholds. Hence, it is supposed to be applicable to different climate zones, and the effect size of hot weather based on the UTCI could be generalizable to other climate zones.

According to the Hospital Authority statistics, there were 2,048,039 AED admissions in year 2020, giving a daily average of 5596. Based on the 2020 population and our estimated AED visit rate, it is expected that on a day with strong heat stress, we may have an additional 137 AED admissions, compared to days with thermoneutral to moderate heat stress conditions; whereas on a day with very strong heat stress, we may have an additional 156 AED admissions. This means a 2.4 to 2.8% surge in the daily AED usage. In terms of service provision, emergency departments may be generally well prepared for days when heat stress is expected to occur. A local study predicted that there would be more days with strong heat stress in the future (Cheung & Hart 2014). It was projected that the proportion of days with strong heat stress would increase from 20.6% in 1971–2000 to 26.2–31.8% in 2046–2065, and the proportion of days with very strong heat stress would increase from 0.1% in 1971–2000 to 0.3–2.1% in 2046–2065. The urban heat island effect worsen the situation. The urban heat island duration was found increasing during 1990–2015 from 13.59 to 17.47 h in built-up areas in Hong Kong, as a result of urbanization and human activities associated with population growth (Chen & Jeong 2018). Local urbanization, in particular the reduction of greenery coverage, development of new towns, and extensive reclamation in coastal urban areas accounts for the increasing trend of urban temperature (Lau & Ng 2013). It was also reported that the temperature of the impervious land surface in megacities of Southeast Asia was 3 °C hotter than green space, on average (Estoque et al., 2017). Thus, the adverse effects of heat stress cannot be neglected. Nevertheless, as the UTCI was not readily available from the HKO or most of the other observatories, the release of this information during the weather forecast or its incorporation into the weather warning system is recommended.

From a public health perspective, awareness of the influence of heat stress should be raised among the population. In Hong Kong, a hot weather warning system was launched in year 2000. When the extreme hot weather reached the specific threshold, the HKO would issue the Very Hot Weather Warning. Related precautions to prevent heat-related adverse outcomes were provided in the HKO website (Hong Kong Observatory 2021a). It was found that days with absence of such warning was associated with an increase of about 1.2 deaths from ischemic heart disease and 1.0 death from stroke among the older adults per day (Chau et al. 2009). In September 2021, the HKO added more precautionary measures to the website to address the special needs of older adults and those with chronic illnesses (Hong Kong Observatory 2021b). Besides alerting the public, the warning also served to alert relevant government departments and non-governmental organizations to take suitable actions. Moreover, if the threshold for issuing the Very Hot Weather Warning was not met, yet the situation could already trigger adverse health outcomes, the HKO would issue Hot Weather Special Advisory to alert the public and the stakeholders to take early precautions. In the USA, with improved heat warning systems and increased public awareness, decline in heat vulnerability had once been reported; however, a reverse trend was observed recently, particularly among the vulnerable groups (Sheridan et al. 2021). It is important to re-explore effective strategies to ameliorate the adverse health effects from heat stress, in addition to reviewing the thresholds that triggered the warnings to make sure that it could well reflect the latest situations in terms of global warning and heat island effect.

While older people are less likely to engage in outdoor activities during hot days, keeping their home cool at a low cost is important. Proper housing design with passive cooling strategies, such as direction of windows and ventilation, cool materials used in buildings, can prevent heat from entering and allowing heat dissipating with minimal energy load is particularly important for the South Asian regions (Sharma et al. 2022). In Japan, usage of sun-shade (in form of a white polyester tarpaulin) was reported to reduce at least 1 h of unsafe heat exposure during the day (Otani & Lee, 2022). In Hong Kong, open spaces purposefully designed with more breeze and shades were found to reduce outdoor thermal stress by 2 °C (in terms of UTCI) and attract more social activities, as well as more frequent and longer duration of stay in these spaces by older people in Hong Kong (Huang et al. 2020). Free public areas equipped with meaningful activities, such as libraries and town halls, would also be ideal for the older population, both to stay away from heat stress and boost social participation. These measures often involve administrative and policy stakeholders to support the infrastructure and to raise public awareness and promote education though effective information dissemination channels.

The adult population may be engaged in outdoor activities such as outdoor work. Their health risk and physical work capacity have also been underscored (Kjellstrom et al., 2020). Therefore, awareness of occupational safety in terms of preventing heat-related adverse events is important. Developing early warning and forecast system for heat should form part of the heat action plans as an administrative tool to increase people’s preparedness against extreme heat events (Sharma et al. 2022). Preventive measures such as reducing work intensity, taking frequent breaks, improving ventilation, cooling clothing, cold water immersion, heat acclimation, cold fluid intake are reported to be effective to mitigate occupational heat stress (Morris et al. 2020). Teaching public strategies to be resilient during heat stress is of utmost importance in the era of global warming. For outdoor leisure activities, a reduction in risky behaviors should be advocated among young adults, such as avoiding hiking during times of significant heat stress. Furthermore, proper health-seeking behaviors, such as seeking emergency care after a heat stroke, should be included as an additional public education topic.

For all population subgroups, they would be benefited from better urban design. Increasing the percentage of community greenery and blue-green spaces have been widely adopted and promoted measures to regulate heat, noise, and pollution (Filho et al., 2021; Pascal et al. 2021; Krüger 2021). Tall urban vegetation showed positive impact on thermal comfort through shading and evapotranspiration, bringing down thermal exposure and UTCI by up to 10.5 °C (Lehnert et al. 2021). Increasing urban water bodies and blue infrastructures promotes evaporative cooling and reduces urban heat island effects (Krüger, 2021; Sharma et al. 2022). Yet, these infrastructures would better be carefully designed and fitted into city planning to maximum the effectiveness (Kolokotsa et al. 2022). Infrastructural modifications including building materials to increase surface albedo and decrease heat absorption are also mitigation measures (Krüger, 2021; Sharma et al. 2022).

From a research perspective, more in-depth investigations on the association between heat stress and health while controlling for individual-level socio-demographic and health conditions are needed. Other threshold values for defining heat stress, say by using the 90th-percentile or 95th-percentile of UTCI, may be explored. Furthermore, as individuals may be exposed to different heat stresses depending on the surrounding environment, studies based on microclimate information would better reflect the specific association of a particular environment with heat stress. While all-cause AED visits could cover both direct and indirect health outcomes from heat, cause-specific AED visits should be explored in the future to determine differential impacts.

The strength of this study is the use of longitudinal administrative data across 15 years in the analyses, covering the majority of AED visits in Hong Kong. However, as one of the variables in the UTCI formula, the black globe temperature was not available in Hong Kong until 2014; thus, a proxy of global solar radiation was used. The use of hourly maximum UTCI might also differ from that based on 1 min or 10-min basis. Future studies should repeat the same analysis with different calculations of the UTCI when more data of healthcare utilization are available. As aggregated data were obtained, individual-level data were not available. Thus, age-standardized rates could not be used, and individual-level risk factors could not be controlled. For example, as the UTCI describes outdoor weather conditions, the results might not reflect genuine associations if the individual stayed indoors. Moreover, individual perceptions of thermal comfort could not be investigated because of the nature of the aggregated data. The confounding risks of prolonged and delayed effects may not be distinguishable. A sensitivity analysis was conducted by replacing the cumulative number of days of heat stress and poor air quality over 7 days with a lag of 1–7 for heat stress and poor air quality indicators. The results showed that the model fit, as revealed by the adjusted R square, did not improve. While previous studies have shown that the association between hot weather and AED visits varies by age and causes of visits (Basu et al., 2012), our aggregated data did not allow us to analyze cause-specific relationships. Although there could be autocorrelation in the AED rates, we did not include the autoregressive terms in the models as the past rates were also related to the weather. Future studies could use time series approach such as transfer function (Chau et al. 2020). Finally, this study was based on observational data; therefore, only associations could be investigated, rather than causations.

In conclusion, biothermal condition from heat stress was associated with the health of the citizens in a city with a subtropical climate and reflected in the increase of daily AED visit. The associations varied across different age and sex subgroups. Based on these findings, public health recommendations have been made for the prevention of heat-related AED visits.