Impact of short-term exposure to extreme temperatures on diabetes mellitus morbidity and mortality? A systematic review and meta-analysis

The relationship between diabetes mellitus and short-term exposure to extreme temperatures remains controversial. A systematic review and meta-analysis were performed to assess the association between extreme temperatures and diabetes mellitus morbidity and mortality. PubMed, Embase, the Cochrane Library, Web of Science, and the Cumulative Index to Nursing and Allied Health Literature (CINAHL) were searched since inception to January 1, 2019, and updated on November 17, 2020. The results were combined using random effects model and reported as relative risk (RR) with 95% confidence interval (CI). In total, 32 studies met the inclusion criteria. (1) Both heat and cold exposures have impact on diabetes. (2) For heat exposure, the subgroup analysis revealed that the effect on diabetes mortality (RR=1.139, 95% CI: 1.089–1.192) was higher than morbidity (RR=1.012, 95% CI: 1.004–1.019). (3) With the increase of definition threshold, the impact of heat exposure on diabetes rose. (4) A stronger association between heat exposure and diabetes was observed in the elderly (≥ 60 years old) (RR=1.040, 95% CI: 1.017–1.064). In conclusion, short-term exposure to both heat and cold temperatures has impact on diabetes. The elderly is the vulnerable population of diabetes exposure to heat temperature. Developing definitions of heatwaves at the regional level are suggested.


Background
The Intergovernmental Panel on Climate Change indicated that the impact of climate change on public health attracted extensive attention in the twenty-first century (Allen et al. 2014). Climate change is already causing tens of thousands of deaths every year and has attracted more and more attention (World Health Organization 2016a). Extreme temperatures, such as heat effect and cold effect, is an important respect of climate change and a significant risk factor for human health throughout the world; associations of temperatures with morbidity and mortality have been supported in numerous scientific literatures (Bayram 2017;Gronlund et al. 2018;Sun et al. 2018). In China, a country with one-fifth of the world's population, the number of deaths related to heatwaves increased fourfold from 1990 to 2019, reaching 26,800 in 2019 (Cai et al. 2020).
Diabetes mellitus is a serious, chronic metabolic disease and important public health problem. Diabetes could lead to numerous health problems, without reasonable control (World Health Organization 2016b), such as serious damage to the heart, blood vessels, eyes, kidneys, and nerve (Peng et al. 2018;Raffaele et al. 2016). Evidence showed that the number of patients with diabetes was increasing. Global report on diabetes indicated that in 2012 alone, diabetes led to 1.5 million deaths, and the number of adults living with diabetes has almost quadrupled since 1980 to 422 million adults (World Health Organization 2016b). The International Diabetes Federation (IDF) mentioned that by 2045, the global number of cases with diabetes will increase from 463 million (2019) to an estimated 700 million (International Diabetes Federation 2019). The burden of diabetes is not only related to the matter of therapy aimed at glycemic control but also significantly influences the patient's survival, quality of life, and development of organ system degeneration.
A series of systematic reviews have been conducted to estimate the effects of extreme temperature on cardiovascular and respiratory diseases (Bhaskaran et al. 2009;Carolan-Olah and Frankowska 2014;Costello et al. 2009;Sun et al. 2018;Cheng et al. 2019). Though a series of published studies had viewed the relationship of extreme temperatures and diabetes, the results were inconsistent. No convincing evidence exists that systematically evaluate the associations between extreme temperatures and diabetes. The aim of this study was to evaluate the impact of heat exposures, cold exposures, heatwaves, and cold spells on diabetes morbidity and mortality, and identify the vulnerable population affected by extreme temperatures.

Methods
The protocol of this systematic review was published online at the PROSPERO (registration number: CRD42019132296).

Search strategy
We conducted a systematic search in PubMed, Embase, the Cochrane Library, Web of Science, and the Cumulative Index to Nursing and Allied Health Literature (CINAHL) since inception to January 1, 2019, and updated to November 17, 2020. The following US National Library of Medicine's Medical Subject Headings (MeSH) terms and key words were used in search: (climate change* OR temperature*) AND (diabet* OR insulin*) AND (time series OR case crossover). The complete search strategy is shown in Appendix S1.

Inclusion and exclusion criteria
The following criteria for eligibility were used: (i) studies that evaluated the effect of extreme temperatures on diabetes morbidity or mortality (ICD-9:250 or ICD-10:E10-E14); (ii) study design was time series or case crossover; (iii) extreme temperatures accord with at least one of the following four temperature exposures: cold, heat, cold spells, and heatwaves; (iv) studies provided at least one of the following estimates: percentage change, relative risk (RR), or odds ratio (OR) with 95% confidence interval (CI) or enough information for calculation; and (v) the publication language was restricted to English. The following criteria for exclusion were used: (i) studies that evaluated indoor, body, and workplace or focused on seasonality temperatures as exposure variables; (ii) studies only providing qualitative evaluation; and (iii) conference abstracts. The definition of temperature exposure was as follows: Heat exposures: Definition 1: effect estimates per unit/other increment increase above threshold or no threshold for linear effects; Definition 2: comparison between heat threshold with the reference value (e.g., 99th vs. 75th percentiles); Definition 3: Temperatures above the heat threshold compared with temperatures below the threshold (e.g., above 95th vs. below 95th percentiles); Cold exposures: Definition 1: effect estimates per unit/other increment increase below threshold or no threshold for linear effects; Definition 2: comparison between cold threshold with the reference value (e.g., 1st vs. 25th percentiles); Definition 3: Temperatures below the cold threshold compared with temperatures above the threshold (e.g., below 25th vs. above 25th percentiles); Heatwaves: temperatures above heat threshold and lasting for 2 or more days.
Cold spells: temperatures below cold threshold and lasting for 2 or more days.

Study selection and data extraction
The screening of potentially eligible studies was carried out using a two-step process. First, four researchers independently screened studies by titles and abstracts. Second, all relevant studies (articles that cannot be excluded by title and abstract) got through full-text screening on the basis of inclusion and exclusion criteria. Two reviewers independently extracted the relevant data from the included studies. The extracted data from the included studies comprise but was not limited to the following items: author, year of publication, study design, temperature exposure type and definition, outcome, and statistical methods. Disagreement was resolved by consensus and the opinion of a third reviewer.

Data synthesis
We summarized the characteristics and determined the number of assessing heat exposures, heatwaves, cold exposures, and cold spells in all included studies. Temperature indicators of the included studies comprise air temperature (express as°C ), apparent temperature, and humidex. Air temperature was considered in the meta-analysis because the majority of included studies used the indicator.
We conducted qualitative and quantitative analyses. Metaanalysis was performed to integrate the overall risk on diabetes. For studies with multiple temperature thresholds reported, the maximum/minimum threshold of temperature was selected to analyze the overall risk of heat and cold effects. Subgroup meta-analysis was conducted for outcomes (morbidity and mortality), definitions and thresholds of heat temperatures, and ages (≥60 years old). Moreover, for diabetes mortality and morbidity, we conducted a meta-analysis of heat temperatures grouped by different definitions and thresholds. The meta-analysis was performed separately where more than three estimates were available.
For definition 1 of heat and cold exposures, risk estimates (percent changes, RR, OR,) were converted into RR per 1°C change in temperature, which allowed us to quantitatively pool estimates from different studies. The standardized risk estimates are calculated by following formulas (1) and (2): One study (Li et al. 2014a) did not show the overall estimate, which reported the stratified estimate of analyzed cities independently. Therefore, we pooled the stratified estimates (Harbin, Nanjing, Shenzhen, and Chongqing in China) of this study in meta-analysis. In addition, the locations and time of two studies Li et al. 2014b) overlapped with the other two studies (Sherbakov et al. 2018;Li et al. 2014a); therefore, we only included the study of Sherbakov et al. 2018 andLi et al. 2014a in the systematic review and meta-analysis.

Sensitivity analyses
Sensitivity analyses were performed to examine the robustness of our outcomes by excluding studies in which the location and study time were partially overlapping from the included studies.

Statistical analysis
A random effects model was performed. Heterogeneity between the included studies was assessed by the chi-square test and the extent of inconsistency was evaluated by the I 2 test. p < 0.05 was considered statistically significant. All analyses were conducted using Stata software (16.0).

Risk of bias assessment
We used a domain-based risk of bias (RoB) assessment tool (World Health Organization 2020) developed by the WHO Global Air Quality Guidelines Working Group on Risk of Bias Assessment to assess the risk of bias on all included articles. There were six domains (confounding, selection bias, exposure assessment, outcome measurement, missing data, and selective reporting) and each domain had 1 to 4 subdomains respectively in the RoB assessment tool. The instrument was designed specifically to assess risk of bias of air pollution studies included in systematic reviews and of studies on short-and long-term exposure to air pollutants (World Health Organization 2020). However, we focused on extreme temperatures rather than the health effects of pollutants on diabetes, so we replaced the key confounding factor temperature in the confounding domain of RoB assessment tool with pollutants. According to the criteria of the RoB assessment tool, each subdomain got a "low," "moderate," or "high" judgment of bias risk. To make an overall judgment on a domain, the following approach was adopted: if any subdomain has a high risk of bias, the whole domain will be regarded as a high risk of bias; if all subdomains have a low risk of bias, the whole domain will be rated as low risk of bias; when at least one subdomain has a moderate risk of bias and other subdomains have no high risk of bias, the whole domain will be rated as moderate risk of bias. All of the included studies were evaluated independently by two reviewers (Xinyi Wang and Liangzhen Jiang) and resolved by a third reviewer (Xuping Song) in case of disagreement.

Results
A total of 2104 articles were identified. After excluding duplicated articles, 974 records were excluded based on title or abstract, with 63 studies for full-text review. Ultimately, 32 studies accorded with our criteria and 18 of these 32 included studies were combined in the meta-analysis (Fig. 1).

Study characteristics
The detailed information of the included studies is presented in Table 1. Among the eligible studies, 11 studies were conducted in the USA, 7 in China, 4 in Australia, 2 in Italy, 2 in Thailand, and the other 7 studies were distributed in Spain, Sweden, South Korea, Brazil, UK, Philippines, and Canada (Oudin Åström et al. 2015 conducted their study in Italy and Sweden). In these 32 studies, 20 used a time series design and 12 used a case-crossover design. Of the included studies, 17 reported the effects of heat effect only, 5 reported heatwaves only, 2 reported both heat effect and heatwaves, and 8 reported both heat effect and cold effect. For exposure variable, 22 studies used temperature with one study ) using both air temperature and apparent temperature, 5 used apparent temperature, and 2 used humidex. For outcomes, 18 studies explored the effects of extreme temperatures on diabetes mortality, 16 studies evaluated morbidity, with 2 studies (Liu and Hoppe 2018;Wilson et al. 2013) exploring both mortality and morbidity. For definition of heat/cold exposures, 13 studies adopted temperature exposure definition 1, 10 studies used definition 2, and 5 studies applied definition 3.   Fig. 2 and Fig. 3).
There is no regionally recognized definition of heatwaves so far. Generally, heatwaves were defined by the duration and intensity. Due to the difference in heatwave definitions, meta-analysis was not performed in our study. There were 5 studies that assessed the association between heatwaves and diabetes (Fig. 4). Wang et al. (2012) indicated a significant association between heatwaves and diabetes mortality of the elderly aged over 75 years. No significant association between heatwaves and diabetes was found in other four studies (Campbell et al. 2019;Sherbakov et al. 2018;Wilson et al. 2013;Chen et al. 2017).

Impact of heat exposures on diabetes mortality and morbidity
The result suggested a higher risk of heat effect for mortality on diabetes as compared to morbidity, with an increased risk of 13.9% (95% CI: 8.9-19.2%) and 1.2% (95% CI: 0.4-1.9%) respectively (Fig. 5).

Impact of heat exposures on diabetes in different ages
The meta-analysis result of heat effect on diabetes in different ages showed that heat exposures increased the risk on diabetes of the elderly aged 60 and over by 4% (95% CI: 1.7-6.4%), which was higher than all populations (RR=1.032, 95% CI: 1.021-1.044) ( Figure S2).  Figure S3 and Figure S4).

Risk of bias
The risk of bias rating for the included studies is shown in Table 2 and more analytically in Table S2. In general, most of the included studies were rated as "low" risk in domains such as selection bias, outcome measurement, missing data, and selective reporting. In the confounding domain, only 5 studies (Stafoggia et al. 2006;Bai et al. 2016;Pudpong and Hajat 2011;Wang and Lin 2014;Wilson et al. 2013) were evaluated for "low" risk, 13 studies were evaluated for "moderate" risk, and 14 studies were evaluated for "high" risk, for the reason that not all of the key potential confounders (e.g., pollutant and influenza) were adjusted in the analysis. Fourteen studies were judged as "moderate" risk in exposure assessment domain, because exposure data were obtained from only one monitoring site over a large geographic area. Four studies were evaluated for "moderate" risk in selection bias, as the result of the study populations only included age-specific people like the elderly or urban residents but not rural residents. Eighteen of 32 (56%) studies were adjusted for air pollutants in the model to control the confounding effects of pollutants. O 3 and PM were the pollutants of greatest concern. Fourteen studies were adjusted for multiple pollutants simultaneously. Three studies were adjusted for O 3 , PM, NO 2 , SO 2 , and CO.

Discussion
In this systematic review, we evaluated the published articles on the relationship between extreme temperatures and   Sun et al. (2018) included studies in the heat/cold exposure category that described the number of degrees above/below threshold or average value (definition 1) or a comparison between extreme hot/cold temperatures and the reference value (definition 2). Definition 3 is relatively small and is rarely used and often ignored in previous systematic evaluation studies. The addition of definition 3 in this paper is more reasonable and comprehensive. We found that both heat and cold temperatures had significant impacts on diabetes, and the impact of heat exposures on diabetes mortality was higher than morbidity. We also found that the higher the threshold for heat temperatures, the greater the impact on diabetes mortality and morbidity. Besides, a higher percentage of the elderly aged 60 years and over with diabetes due to high temperatures compared to the general populations was found. The sensitivity analysis results were consistent with the overall analysis; hence, the results were reliable in this study.

Extreme temperatures and diabetes
We found that both heat and cold exposures have impact on diabetes. A case-only study found that patients with diabetes had a higher risk of dying on hot days than other subjects (Schwartz 2005). Previous studies suggested a complex association. While there were some studies that support our results (Lavigne et al. 2014;Lu et al. 2016;Lam et al. 2018; Kim et al., and Gasparrini et al. found no relationship between heat exposures and diabetes, while Su et al. 2020. andWang andLin (2014) found a significant association between heat exposures and diabetes but no significant association of cold. Acute myocardial infarction is one of the main causes of death in patients with diabetes. Evidence showed that adults with diabetes had an increased risk of cardiovascular disease by 2-4 times compared with adults without diabetes, and the risk increased as blood sugar control deteriorated Dal Canto et al. (2019). Exposure to low temperatures is associated with increased incidence of acute myocardial infarction as well as poor glycemic control (Vallianou et al. 2020;Tsujimoto et al. 2014;Lai et al. 2020). The risk of acute myocardial infarction in diabetic patients rose sharply at low temperatures (Lam et al. 2018). The reason for the discrepancy may be that the study areas were located in different climate zones. In addition, the adaptability of the study populations to specific climatic zones and regional economic conditions may also contribute to the differences in results.
Diabetes is becoming a global public health problem. The International Diabetes Federation has predicted that the prevalence of diabetes will reach 592 million cases with an additional 175 million undiagnosed by 2035 (Aguiree et al. 2013). Extreme temperatures are a health threat to people with diabetes. And with global climate change especially rising temperatures, it is essential to reduce the exposure of diabetics to extreme temperatures.

Vulnerable populations
The elderly with diabetes aged 60 years and over showed to be more vulnerable to heat effect that was observed in our study. Xu et al. 2019, Wilson et al. 2013, and Li et al. (2017 also found the same results that extreme heat was associated with a greater risk of diabetes in the elderly compared to all populations or other age groups. Similarly, Lam et al. found patients with diabetes aged 75 years and over were more vulnerable to heat and cold effect (Lam et al. 2018). Older people, already in poorer health than younger people, have a harder time adjusting in extreme heat and cold exposures with diabetes. Our finding suggests that people with diabetes, especially the elderly, should try to reduce exposures to extreme temperatures.

Potential biological mechanisms
The biological mechanisms of the effects of extreme temperatures on diabetes are not fully understood as yet. Potential biological mechanisms include, but are not limited to, neurological pathways, hemodynamic effects (Ely et al. 2018;Kenny et al. 2016;Vallianou et al. 2020), and brown adipose tissue (Symonds et al. 2019). During heat and cold exposure, cardiovascular regulation is essential for temperature control, so blood must be redistributed to the periphery and to the core respectively to maintain a stable core temperature and thus maintain the body's thermal balance (Kenny et al. 2016). According to reports, diabetic patients have lower skin blood flow and sweat response when exposed to heat, which has a great adverse effect on cardiovascular regulation and blood sugar control. Diabetics have lower skin blood flow and significantly impaired vascular response to cold, which may make it more difficult to prevent core temperature drops associated during cold exposures (Kenny et al. 2016). Greater degrees of neuropathy may lead to lower skin blood flow levels in diabetes patients, so that diabetics cannot keep body temperature stable under heat and cold exposures. Brown adipose tissue has a unique mitochondrial protein uncoupling protein (UCP)1 which can be activated during cold exposures, leading to increased sympathetic nervous system activity, oxidation of large amounts of lipids and glucose, and heat consumption. While with the continued rise in global temperatures, and the increased duration of "summer," the protective effect of brown adipose tissue on diabetes is limited (Symonds et al. 2019).

Heterogeneity
Heterogeneity may be related to study design types and statistical analysis models between studies. Secondly, regional and demographic differences may be an important reason. Studies were conducted in different climate zones, and the average temperature, relative humidity, air pressure, and other meteorological factors are different from each other. The adaptability of residents living in different climatic zones to cold and heat temperatures is also different, and the economic level and social development between regions will have an important impact; for example, there are differences in the popularity of cooling and heating tools, such as air conditioners, between developed and developing countries. Although most studies were adjusted for relative humidity and air pollutants, there was still a part of the studies that did not consider the influence of atmospheric pollutants or other important confounding factors. In addition, some studies only obtained the temperature exposure data from one monitoring site to represent the whole exposure levels of residents in a big area. Combined with the discrepancies of temperature index and lag days, all these factors led to the heterogeneity of the results.

Strengths and limitations
This systematic review and meta-analysis provided pioneering evidence between extreme temperatures and diabetes. We analyzed the impacts of cold and heat exposures on diabetes morbidity and mortality according to different definition methods of extreme temperatures. Studies were selected in strict accordance to the inclusion and exclusion criteria, and the risk of bias in all included studies was assessed using the RoB quality assessment tool. In addition, sensitivity analysis was conducted to ensure the robustness of results.
Our study also has some limitations. First of all, different extreme temperature definition methods used in the included studies limited the combination of effect sizes. Subgroup analysis on temperature definitions was performed to explore the difference effects of heat exposures on diabetes. Secondly, most of the included studies were conducted in developed and a few developing countries among the temperate zone, and some used the same regional data. Sensitivity analysis on heat and cold impacts was conducted to assess the robustness of our results. Thirdly, due to the lack of data, we failed to analyze the lagged effects of extreme temperatures on health. The delayed effects of extreme temperatures on health are not clear at present, so it is necessary to study the effects of extreme temperatures on diabetes under different lagged effects.

Conclusions
Both short-term exposure to heat and cold temperatures have impact on diabetes. The elderly is the vulnerable population of diabetes exposure to heat temperatures. Developing definitions of heatwaves at the regional level are suggested.
Acknowledgements Funding from the Science and Technology Program of Gansu Province, Natural Science Foundation of Shanghai, and Costeffectiveness analysis of electrical stimulation combined with bladder function training in the treatment of neurogenic bladder with spinal cord injury are gratefully acknowledged.
Author contribution XS made the search strategy and wrote the article. LJ analyzed the data, assessed the risk of bias, and wrote the article. DZ wrote the article. XW searched articles and assessed the risk of bias. YM and YH selected studies and extracted data. XL and JT selected studies. YM and DZ wrote the protocol and registered at the PROSPERO. WH, AS, and YF checked data and results of RoB assessment. YZ made some suggestions for improvement of this article.
Funding This study was funded by the Science and Technology Program of Gansu Province (20CX4ZA027), Natural Science Foundation of Shanghai (20JR5RA262), and Cost-effectiveness analysis of electrical stimulation combined with bladder function training in the treatment of neurogenic bladder with spinal cord injury (2020-RC-63).
Data availability All data generated or analyzed during this study are included in this published article and its supplementary information files.

Declarations
Ethics approval and consent to participate Not applicable.

Consent for publication Not applicable.
Competing interests The authors declare no competing interests.