Heat, heatwaves, and ambulance service use: a systematic review and meta-analysis of epidemiological evidence

Ambulance data has been reported to be a sensitive indicator of health service use during hot days, but there is no comprehensive summary of the quantitative association between heat and ambulance dispatches. We conducted a systematic review and meta-analysis to retrieve and synthesise evidence published up to 31 August 2022 about the association between heat, prolonged heat (i.e. heatwaves), and the risk of ambulance dispatches. We initially identified 3628 peer-reviewed papers and included 48 papers which satisfied the inclusion criteria. The meta-analyses showed that, for each 5 °C increase in mean temperature, the risk of ambulance dispatches for all causes and for cardiovascular diseases increased by 7% (95% confidence interval (CI): 5%, 10%) and 2% (95% CI: 1%, 3%), respectively, but not for respiratory diseases. The risk of ambulance dispatches increased by 6% (95% CI: 4%, 7%), 7% (95% CI: 5%, 9%), and 18% (95% CI: 12%, 23%) under low-intensity, severe, and extreme heatwaves, respectively. We observed two potential sources of bias in the existing literature: (1) bias in temperature exposure measurement; and (2) bias in the ascertainment of ambulance dispatch causes. This review suggests that heat exposure is associated with an increased risk of ambulance dispatches, and there is a dose-response relationship between heatwave intensity and the risk of ambulance dispatches. For future studies assessing the heat-ambulance association, we recommend that (1) using data on spatially refined gridded temperature that is either very well interpolated or derived from satellite imaging may be an alternative to reduce exposure measurement bias; and (2) linking ambulance data with hospital admission data can be useful to improve health outcome classification. Supplementary Information The online version contains supplementary material available at 10.1007/s00484-023-02525-0.


Introduction
High ambient temperatures (i.e. heat) can adversely impact human health; including but not limited to the circulatory (Liu et al. 2022), respiratory (Cheng et al. 2019), renal (Liu et al. 2021a), nervous (Liu et al. 2021b), and endocrine systems (Moon 2021). Even in advanced economies, extreme heat is the most deadly natural hazard (Borden and Cutter 2008). Extreme heat, which already causes US $727 billion in economic losses worldwide each year (Callahan and Mankin 2022), is becoming more frequent and intense due to climate change.
Multi-national studies have reported an increased risk of mortality associated with heat Vicedo-Cabrera et al. 2018). The effect of heat on healthcare services (e.g. hospital admissions) has previously been metaanalysed (Phung et al. 2016). Some recent studies have found that dispatch call centres and ambulances can more readily detect the acute effects of heat Xu et al. 2018). Compared with emergency department presentation and hospital admission data, ambulance data allows for a more sensitive indication of total demand for health services allowing for better preparation, management, and deployment during heat events (Bassil 2010). It is worthwhile synthesising published evidence on heat and ambulance service use.
We systematically reviewed and summarised evidence provided in the available literature on heat, prolonged heat (i.e. heatwaves) and ambulance service use, aiming to (1) provide a pooled effect estimate for the impacts of heat and heatwaves on all-cause and cause-specific ambulance dispatches; and (2) identify potential sources of bias in the study methodology.

Materials and methods
This review was conducted in alignment with the most upto-date version of Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) Guidelines (Page et al. 2021), and a protocol was registered through the international prospective register of systematic reviews (PROS-PERO Reference: CRD42022296556).

Eligibility criteria
Empirical studies were considered eligible for inclusion within the review if they met the following criteria: (1) human population; (2) English language; (3) published within the last decade (2011-2022); (4) included a measure of heat or heatwaves as the primary exposure; (5) used routinely collected ambulance records to investigate outcomes; (6) the outcome was public health-related (e.g. not performance assessment of ambulance services, nor occupational based). In this review, we defined heat as temperatures above an optimal temperature range or optimal temperature point which have adverse impacts on human health and well-being (Asseng et al. 2021;Gasparrini et al. 2015). As this review focused on the impact of short-term exposure to heat or heatwaves on the risk of ambulance dispatches, we included studies with the two most common designs which quantify the association between short-term exposure to heat or heatwaves and risk of health outcomes: time-series and case-crossover designs (Wu et al. 2022).
There were three exclusion criteria: (1) any studies relying exclusively on hospital, death registry, or emergency department admission records were excluded; (2) as the primary exposure of interest was heat and heatwaves, if any study investigated the impact of air pollution as the primary exposure, and included temperature as a confounding exposure, then these studies were also excluded. Temperature is a potential confounder of the association between shortterm exposure to air pollution (particularly ozone) and the risk of health outcomes. For instance, the concentrations of ozone tend to increase during hot weather, and the association between short-term exposure to ozone and the risk of health outcomes may be confounded by heat (Alari et al. 2023). Air pollution has been found to be a modifier of the association between short-term heat exposure and the risk of health outcomes ; (3) any study where the performance or evaluation of ambulance and emergency services occurred without evaluation of the public health impacts of heat again were excluded.

Information sources
After consultation with a research librarian at Griffith University, a search strategy was developed and applied to the following six online databases: PubMed, Embase, Cumulative Index of Nursing and Allied Health Literature (CIN-HAL), Scopus, ProQuest, and Web of Science. The search strategy consisted of three major themes: (i) a heat term, (ii) an ambulance term, and (iii) a health term. Terms were adjusted to each database using mesh terms and filters where applicable (Supplementary 1). Pilot searches were conducted to ensure search strategy robustness and inclusion of key literature, the final search was conducted on August 31, 2022. The references of identified articles were screened to make sure all relevant articles were included.

Literature selection and quality assessment
All study results acquired from the six databases searched were imported into Endnote (version X9, 2013), and duplicates were removed before being uploaded into Covidence (v2715, 2021) to complete data extraction and quality assessments. All study titles, abstracts, and full texts were screened by two independent reviewers (JW, and AB, SR, DP, or ZX) using standardised criteria. Any disagreements were resolved via discussion between both reviewers, and if consensus could not be reached, a third reviewer was introduced.
Quality assessment analyses were conducted by two reviewers (JW, and AB, SR, DP, or ZX) using the Newcastle-Ottawa Scale for assessing the quality of nonrandomised studies. The tool was adapted to suit each of the included study designs (time-series and case-crossover). Quality assessment analyses aided in assessing the quality of evidence presented within each study by examining sample representativeness, ascertainment of exposure and outcome measures, inclusion of common confounders within the statistical models used, and the specificity of the outcome presented (Supplementary 2).

Meta-analysis
For studies assessing the impact of heat on ambulance dispatches, if they used the same temperature indicator (e.g. mean temperature) and health outcome (e.g. ambulance dispatches for cardiovascular diseases), they were included in the same meta-analysis. The definitions of different temperature indicators used in the included studies are presented 1 3 in the Supplementary 3. Twenty-three studies were excluded from the meta-analysis mainly due to the temperature indicators used: Six studies used mean temperature as the temperature indicator (Cheng et al. 2016;Hu et al. 2020;Kotani et al. 2018;Onozuka and Hagihara 2015;Prichard et al. 2022;Wu et al. 2021). However, Cheng et al. used warm season temperature to examine the overall effect of heatwaves, Hu et al. used warm season temperature to examine the main and added effects of heatwaves, and Kotani et al., Onozuka and Hagihara, Prichard et al., and Wu et al. did not provide information on the specific values of the temperature cut-offs to define heat. Hence, the effect estimates published in these studies could not be pooled together.
Three studies used maximum temperature (Romani et al. 2020;Turner et al. 2013;Williams et al. 2012 Two studies used excess heat factor (EHF) (Hatvani-Kovacs et al. 2016;Patel et al. 2019). However, the EHF definitions they used were different from each other, and they were different from the EHF definition used in the other two studies included in the meta-analysis (Jegasothy et al. 2017;Williams et al. 2018).
One study used humidex as the temperature indicator (Calkins et al. 2016).  One study used compound heat as the temperature indicator .
Although there were studies using the same temperature indicator and out-of-hospital cardiac arrest (OHCA), we did not conduct a meta-analysis for OHCA because a separate meta-analysis has been published (Wu et al. 2023). We made two assumptions in the meta-analysis: We assumed that odds ratio (OR) is a reasonable approximation of relative risk (RR) in the eligible studies because ambulance dispatches are small probability events (i.e. < 1%). Four eligible studies used RR as the effect estimate indicator (Cui et al. 2020;Sangkharat et al. 2020;Wang et al. 2021;Zhan et al. 2018), and Guo's study used OR (Guo 2017). We assumed OR is a reasonable approximation of RR in Guo's study. The study of Turner et al. used "percentage change in the risk of ambulance dispatches", and we also converted it into RR.
We assumed that the association between heat and risk of ambulance dispatches is linear, acknowledging that the association between temperature (i.e. heat and cold) and health outcomes could be U-, J-, or V-shaped. Hence, we converted the RRs under different temperature increments reported in all six eligible studies (Cui et al. 2020;Guo 2017;Sangkharat et al. 2020;Turner et al. 2012;Wang et al. 2021;Zhan et al. 2018) into RRs per 5 °C increase in temperature. This approach has been used elsewhere (Chersich et al. 2020). The impact of heat on ambulance dispatches may last for more than 1 day (i.e. lagged) but is generally acute. Cui et al. reported RRs across multiple lags (from lag 0-12 h to lag 0-170 h-) (Cui et al. 2020), and Sangkharat et al. reported RRs from 0-2 days to 0-21 days (Sangkharat et al. 2020). We used RRs for lags within 2 days in the meta-analysis because they generally represented the most acute impact of heat (Thomas et al. 2021;Winquist et al. 2016). The acute impact of heat occurred within 2 days was shown in the tables of the three published papers (see Table 2 of Cui et al. 2020, Table 3 of Sangkharat et al. 2020, and Table 3 of Turner et al. 2012).
For studies assessing the impact of heatwaves on ambulance dispatches, if they used the same heatwave definition (e.g. Excess Heat Factor (EHF)), effect estimate indicator (e.g. incidence rate ratio (IRR)), and health outcome (e.g. the daily total number of ambulance dispatches), they were included in the same meta-analysis. Specifically, two Australian studies which used EHF, in the same way, were pooled together (Jegasothy et al. 2017;Williams et al. 2018), and another two studies which used the same heatwave intensity (90th, 95th, or 99th percentiles) and duration indicators (2 or 3 days) were pooled together (Sun et al. 2014;Xu et al. 2018).
Random-effects models were used to pool the effect estimates. For studies which reported RRs across multiple lag periods within 2 days (e.g. 0-12, 0-24, and 0-48 h) (Cui et al. 2020), multiple regions (Williams et al. 2018), or multiple ambulance dispatch indicators (Sangkharat et al. 2020), we used hierarchical random-effects models to pool their RRs with RRs in other studies, accounting for both withinstudy variability (first level) and between-study variability (second level).
As the number of studies included in each meta-analysis was low (≤ 6), we were unable to assess a funnel plot or do more advanced regression-based assessments to evaluate publication bias.

Study selection
The PRISMA diagram in Fig. 1 describes the process of record selection for both the initial database searches and bibliographic screening. The final search yielded 3628 results from the six databases. Following title and abstract screening, this was reduced to 144 studies. After full-text and reference screenings, 48 full texts have been included in this review.

Meta-analysis
Ten studies were eligible for meta-analysis (Table 3), with nine being of moderate quality. The only high-quality study was included in the meta-analysis on the association between heat and risk of ambulance dispatches for all causes (Sangkharat et al. 2020). For each 5 °C increase in mean temperature, the risk of ambulance dispatches for all causes and for cardiovascular diseases increased by 7% (95% confidence interval (CI): 5%, 10%) and 2% (95% CI: 1%, 3%), respectively (Fig. 2). The pooled statistics suggested that the risk of ambulance dispatches for respiratory diseases did not increase when the mean temperature increased.

Potential sources of bias in the included studies
We observed two potential sources of bias in the included studies: (1) bias in temperature exposure measurement; and (2) bias in ascertainment of ambulance dispatch causes.
A wide range of temperature indicators have been used in the included studies, including mean temperature, maximum temperature, minimum temperature, apparent temperature, humidex, and heat index. Mean temperature was the most commonly used temperature indicator (n = 28). Although spatially refined gridded data on temperature is increasingly accessible, 29 of the 48 studies only used temperature data collected from a single weather monitoring station for each study site (Table 3).
Half of the 48 studies included all-cause ambulance dispatches. A limited number of studies reported specific heat-related events (e.g. heatstroke, dehydration, heat exhaustion) (n = 6), cardiovascular (n = 17), and/or respiratory (n = 11) diseases. However, six studies on causespecific ambulance dispatches did not publish or published less explicit information on how the ambulance dispatch causes were ascertained (Cui et al. 2020;Lin et al. 2021;Turner et al. 2012Turner et al. , 2013Wang et al. 2021;Wang et al. 2020).

Discussion
Studies included in this review suggested a higher risk of ambulance dispatches associated with heat or heatwaves, although 44 of the 48 included studies were of moderate quality. The pooled statistics showed that each 5 °C increase in mean temperature was associated with 7% and 2% increases in the risks of ambulance dispatches for all causes and cardiovascular diseases, respectively. There appeared to be a dose-response relationship between EHF-defined heatwave intensity and the risk of ambulance dispatches. Exposure to heat can trigger acute life-threatening cardiovascular diseases, such as acute myocardial infarction (Chen et al. 2019) and stroke (Bai et al. 2018). Heat exposure can also trigger acute kidney injury (Borg et al. 2017). The risk of ambulance dispatches increases when the risk of these acute diseases elevates during hot days at a population level. The dose-response relationship between EHF-defined heatwave intensity and the risk of ambulance dispatches, which we observed in the present review, echoed the findings in a recent Australian study which observed a dose-response relationship between heatwave intensity and mortality risk ).
Although we purposely included studies using the two most ideal study designs in quantifying the association between short-term heat exposure and risk of health Forest plot for the association between heatwaves (defined by threshold and intensity indicators) and risk of ambulance dispatches outcomes (i.e. time-series (Bhaskaran et al. 2013) and casecrossover (Carracedo-Martínez et al. 2010)), the quality of the included studies could have been better if two aspects of the methodology were improved: temperature exposure measurement and ascertainment of ambulance dispatch causes.

Temperature indicator and temperature exposure measurement
One of the main goals of conducting heat and health research is to facilitate health elements to be incorporated into the development and optimisation of heat adaptation strategies. Different temperature indicators have been used in different countries' heat early warning systems (HEWS) (sometimes called heat-health action systems). For instance, Australia's jurisdictional HEWS mainly use the excess heat factor to define heatwaves and trigger heat responses (e.g. the Queensland Heatwave Management Sub-plan (Queensland Health 2019)), and China's national heatwave alert system uses maximum temperature as the temperature indicator for heatwave definition (China Meteorological Administration 2007). In heat and health research, using the temperature indicator that has been adopted in the national, regional, or local HEWS would facilitate the translation of research outputs into practice. For those countries/regions without HEWS, it is worthwhile conducting research to understand which temperature indicator performs better in predicting heat-related ambulance dispatches (Yu et al. 2011).
Heat exposure could vary within cities (e.g. urban heat island effect), and using temperature data collected from a single monitoring station may underestimate the impact of heat on morbidity if the monitoring station is less representative of heat-vulnerable populations' exposure (Thomas et al. 2021). With the advent of publicly accessible and well-interpolated temperature data with high spatial resolution (e.g. the 5 km grided temperature data in Australia: https:// www. longp addock. qld. gov. au/ silo/, or the worldwide ERA5-Land data at 9 km (Muñoz Sabater 2019) that can be further statistically downscaled to 900 m with KrigR (Byers et al. 2022)), it is possible to use these spatially refined gridded temperature data as an alternative to reduce potential measurement bias. However, we acknowledge that if the spatially refined temperature data was poorly interpolated, it could also cause bias. In the case where publicly accessible and well-interpolated temperature data with high spatial resolution is not available, satellite remote sensing temperature data could also be an option to reduce measurement bias (Xu et al. 2014) (e.g. satellite remote sensing data provided by the US National Aeronautics and Space Administration (NASA): https:// ladsw eb. modaps. eosdis. nasa. gov/). Because population density may vary across urban and rural areas within cities, Weinberger et al. have used population-weighted temperature derived from spatially refined gridded temperature data (Weinberger et al. 2019), attempting to reduce the temperature measurement bias. In the available studies comparing the health impacts of heat exposure using temperature data collected from limited number of monitoring stations and using spatially refined temperature data, because the models were fitted to the same health outcome and were non-nested, they were unable to quantify the magnitude of the temperature measurement bias. Simulation studies, such as what Wei et al. did in air pollution epidemiology (Wei et al. 2022), will help characterise the magnitude of temperature measurement bias.

Causes of ambulance dispatches
A main barrier to understanding the potential causal relationship between exposure to heat or heatwaves and the risk of ambulance dispatches is the lack of cause-specific ambulance dispatch data. Some studies included in the present review reported cause-specific ambulance dispatch data. For instance, Campbell et al. used data on ambulance dispatches for cardiovascular, respiratory, renal, diabetic, psychological, direct heat-related, and other heat-related conditions in Tasmania, Australia (Campbell 2021), and they clarified that the records on causes were completed by attending paramedics at the time of, or right after the incident requiring ambulance dispatches.
Unless the patients who used ambulance service have later been hospitalised and gone through diagnostic examinations, it may be hard to accurately ascertain the true causes of ambulance dispatches (particularly in elderly patients who often have multimorbidity (Dobson et al. 2020)). In the present review, the pooled statistics did not suggest an association between heat and the risk of ambulance dispatches for respiratory diseases, but we could not rule out the possibility that respiratory diseases might not have been properly ascertained in the included studies. Data linkage could be an alternative to solve this conundrum. For instance, the information on ICD-codes (international classification of diseases) is generally available in emergency department visit or hospitalisation data. Therefore, linking patients' ambulance dispatch data with their emergency department visit or hospitalisation data could allow researchers to obtain more accurate information on the causes of ambulance dispatches (Vallmuur et al. 2023).

Conclusion
The available epidemiological evidence suggests that each 5 °C increase in mean temperature is associated with 7% and 2% increases in the risks of all-cause and cardiovascular ambulance dispatches, respectively. There is a dose-response relationship between EHF-defined heatwave intensity and the risk of all-cause ambulance dispatches. Future studies using well-interpolated or satellite-derived data on spatially refined gridded temperature and linking ambulance data with hospital data may be useful in reducing biases in exposure measurement and health outcome classification.