Relationship between air pollutants and risk of diabetes:a cohort study based meta-analysis

Background: To investigate the effects of air pollutants on the incidence of diabetes, and to provide reliable evidence for the prevention and control of diabetes. Method: Electronic databases of PubMed, Embase, Web of Science, Wanfang Digital Periodicals (WANFANG), WeiPu, China National Knowledge Infrastructure (CNKI), Springer, and Ovid were searched. RevMan5.3 software was used for meta analysis of the literature included in this study, I2 was used to measure the heterogeneity of the research contents of the included literature, and fixed effect model or random effect model was used for combined analysis of the overall incidence of diabetes and the incidence of diabetes under different air pollutants. Result: A total of 22 articles with high overall quality were included in this study, including 2703882 subjects and 535190 patients with diabetes.Pooled effect of the studies was determined using relative risk(RR). PM10 (RR=1.00; 95%CI:0.99-1.02, P=0.61), PM2.5 (RR=1.02 ; 95%CI:1.01-1.04, P=0.01), NOX (RR=1.00; 95%CI:1.00-1.01, P=0.006), NO2 (RR=1.00; 95%CI:1.00-1.01, P=0.01), O3 (RR=1.00; 95%CI:0.99-1.01, P=0.99). No correlation was found between PM10, NO2, O3 and the risk of type 2 diabetes. The risk of diabetes increased by 1% in people exposed to total air pollutants, increased by 2% in people exposed to PM2.5, and increased by 1% in people exposed to NOX. Conclusion: Results of this paper showed that air pollutants are weakly positively correlated with the risk of diabetes. As to the other influencing factors should be further studied.

and control of diabetes, intervention of environmental factors are easier to achieve than genetic factors. Therefore, identifying environmental risk factors and intervening them may reduce the incidence and mortality of diabetes.
Recent studies have shown that exposure to ambient air pollution is an undesirable stimulus and a risk factor for diabetes [3][4][5]. Air pollution is a ubiquitous exposure that affects a large percentage of the world's population. There are many sources of air pollution, including industrial facilities, cars, ships, and aircraft, as well as home burning devices or forest fires and volcanic eruptions [6].
According to the source of the pollutants, environmental factors (including climatic conditions, time, industrial activity, and traffic density), the size, composition, and concentration of particulate matter vary from one region to another. Gas-state pollutants include nitrogen oxides (NOx), carbon oxides (COx), sulfur oxides (SOx), and ozone (O 3 ). Aerosol-state pollutants mainly include dust, particulates (PM), etc [7]. Exposure to air pollutants is an important adjustable and reversible environmental risk factor for diabetes. Understanding the relationship between exposure to air pollution and the development of diabetes is critical to our ability to control air pollution and its impact on diabetes. To our knowledge, several prior studies have performed meta analysis or systematic reviews related to the association between diabetes and air pollutants, but very few of them were based on cohort studies or only reported type 2 diabetes. Therefore we aim to explore the relationship between air pollutants and the risk of diabetes based on cohort studies.

Inclusion and exclusion criteria
Articles meeting the following criteria were included:subjects included non-diabetic population at baseline;exposure factors included relevant air pollutants;studies outcomes were diabetes (type 1 diabetes, type 2 diabetes, gestational diabetes); cohort studies; the studies provide data information associated with the risk of developing diabetes. The exclusion criteria were as follows: non-cohort studies;the outcome was not suitable for meta analysis; not reporting full information; it was not possible to obtain full text; studies with low reliability and small sample size(less than 1000); repeated articles; reviews.

Data extraction and outcome of interest
The collection of literature data uses a unified data extraction form, the main information extracted from each document: including the first author, publication year, country, study design type, study race, sample size of each group, follow-up time, adjusted confounding factors indicators which are risk ratio(RR) and 95% confidence intervals (CI), NOS score. Heterogeneity was assessed using the I 2 statistics.

Quality assessment and statistical analysis
Quality assessment was carried out for all retrieved studies. Quality in the systematic review referred to the potential of biases during data analysis. The methodological integrity of the study was carried out according to the "Newcastle-Ottawa scale"(NOS) for assessing the quality of non-randomized studies in meta-analysis [8].
The RR and Hazard Ratio (HR) values in the literature are directly regarded as RR values. Metaanalysis was performed with Review Manager(version 5.3). Heterogeneity was assessed using the Cochrane handbook and I 2 statistics, P= 0.05 using the χ 2 test for the Q statistic and I 2 , 50% for the I 2 statistic were interpreted as low-level heterogeneity. If I 2 <50%, a fixed effect model is used; if I 2 ≥ 50%, a random effects model is used for pooled data [9].  Table 1. indicates the summary of 22 cohort studies contributing data to systematic review.
Included basic information: first author, published year, study design, follow-up years, age distribution, number of participants,total number of diabetic patients in the outcome,type of diabetes and quality score. The confounding factors of the included studies were diverse, therefore gender, age, smoking, drinking and comorbidities were adjusted. The NOS scores of included studies were 6-9, indicating high quality literature.

Air pollutants and the risk of diabetes
The air pollutants in the included in studies were SO 2 , PM 10 , PM 2.5 , NO X , NO 2 , O 3 . Heterogeneity analysis showed that there was no statistical difference between each included studies (P=0.52, I 2 =0%), so fixed effect model was used. The results showed that RR=1.01 (95% CI: 1.00-1.01), the difference was statistically significant (P<0.05), indicating that air pollutants were associated with the risk of diabetes,the risk of onset increased by 1% in those who exposed to air pollutants. Subgroup analysis was performed according to different regions: (1) North America: the results showed that RR=1.01 (95% CI: 1.00-1.02), the difference was statistically significant (P<0.05) which indicates that there is a correlation between air pollutants and the risk of diabetes, and the risk of diabetes is increased by 1% in people exposed to air pollutants. (2)Asia:the results showed that RR=1.06 (95% CI: 1.00-1.13), the difference was statistically significant (P<0.05), this indicates that there is a correlation between air pollutants and the risk of diabetes, and the risk of diabetes is increased by 6% in people exposed to air pollutants. (3)Europe: the results showed RR=1.00 (95% CI: 1.00-1.01), the difference was statistically significant (P<0.05), details showed in Fig.2.

PM10 and risk of diabetes
A total of 6 articles [11,[21][22][23][24]27,31] studied PM 10 as a risk factor,five of them had type2 diabetes as the outcome,another one was gestational diabetes.The cumulative follow-up number was 1,662,778, and the number of people with diabetes was 83,584. Heterogeneity analysis showed that there was no statistical difference between the included studies (P=0.68, I 2 =0%), so the data were combined using the fixed effect model, and the meta-analysis of PM 10 and diabetes risk showed that RR=1.00 (95% CI: 0.99-1.02), the difference was not statistically significant (P>0.05), indicating that PM 10 has no correlation with the risk of diabetes, as shown in Fig.3. Removed the literature of gestational diabetes, the meta-analysis of the risk of PM10 and type 2 diabetes showed RR=1.01 (95% CI: 0.99-1.02), the difference was still not statistically significant (P>0.05), indicating there was no correlation between the risk of developing type 2 diabetes and PM 10 ,shown in Fig.4.

PM2.5 and risk of diabetes
A total of 16 articles [10-20, 23-28, 31] studied PM2.5 as a risk factor. Two of these articles had both type 2 and type 1 diabetes as outcomes,eleven were type 2 diabetes and 3 were gestational diabetes.
The cumulative follow-up number was 1198374, and the number of people with diabetes was 459834.
Heterogeneity analysis showed that there was no statistical difference between the included studies (P=0.39, I 2 =5%), so the data were combined using the fixed effect model.The meta-analysis of PM2.5 and diabetes risk showed RR=1.02 (95% CI: 1.01-1.04), the difference was statistically significant (P<0.05), indicating that PM 2.5 was associated with the risk of diabetes, the risk of onset increased by 2%. Subgroup analysis conducted according to the type of diabetes, RR = 1.05 (95% CI: 1.03-1.08), the difference was statistically significant (P < 0.001), indicating PM 2.5 and the risk of developing type 2 diabetes is increased by 5% in those who exposed to air pollutants PM 2.5 ,shown in Fig.5.

NOx and risk of diabetes
Five articles studied NOx as a risk factor [10][11]13,23,27]. The cumulative follw-up number was 2029159,the number of people with diabetes were 91590. Heterogeneity analysis showed that there was no statistical difference between the included studies (P=0.59, I 2 =0%), so the data were combined using the fixed effect model. The meta-analysis of NO X and diabetes risk showed RR=1.01 (95% CI: 1.01-1.04), the difference was statistically significant (P < 0.001), indicating that NO X has a correlation with the risk of diabetes, and the risk of diabetes is increased by 1% in people exposed to air pollutants NO X ,shown in Fig.6. NO 2 and risk of diabetes A total of 9 literature [11,[13][14]17,19,[21][22][23]29] including NO 2 as an exposure factor. Two articles had both type 1 and 2 diabetes as outcomes,7 articles had type 2 diabetes as outcomes. The cumulative number of follow-ups was 1876661, and the number of people with diabetes was 89,417.
Heterogeneity analysis showed that there was no statistical difference between the included studies (P=0.69, I 2 =0%), so the data were combined using the fixed effect model. The meta-analysis of NO 2 and diabetes risk showed RR=1.00 (95% CI: 1.00-1.01), the difference was not statistically significant (P>0.05), indicating that NO 2 has no correlation with the risk of diabetes. See Fig.7 for details.

O 3 and risk of diabetes
A total of 3 literature [11,19,27] studied O 3 as a risk factor. The cumulative number of follow-ups were 1,582,148, and the number of people with diabetes were 77,565. Heterogeneity analysis showed that there was no statistical difference between the included studies (P=0.54, I 2 =0%), so the data were combined using the fixed effect model. The meta-analysis of the risk of O 3 and diabetes showed that RR =1.00 (95% CI: 0.99-1.01), the difference was not statistically significant (P>0.05), indicating that there is no correlation between O 3 and the risk of diabetes, shown in Fig.8.

Sensitivity analysis
In the including studies, the most weighted study was removed, and the remaining studies were meta-analyzed. The RR value and 95% CI of the combined before and after exclusion were not changed much, suggesting that the meta-analysis results are stable and reliable. Shown on Table.2.

Publication bias analysis
As shown in Fig.9-Fig.13, most of the research effect points were concentrated in a narrow area, and the funnel plots were symmetrically distributed, indicating that the included articles had no publication bias.

Discussion
Diabetes is currently one of the most serious public health issues in the world and one of the leading causes of premature death in the global population. In both developed and developing countries, the incidence and prevalence of diabetes is increasing. This phenomenon is related to modern lifestyle characteristics, including obesity and lack of exercise [32]. However, exposure to air pollutants was also considered to be a risk factor in the prevalence of diabetes [33]. The results of this meta-analysis showed that the gaseous pollutants and PM were weakly positively correlated with the risk of diabetes. The results of this study are same with a systematic review published in 2014 [34]. This study found that there was no correlation between the risk of diabetes and PM 10 , NO 2 , O 3 respectively. The risk of diabetes increased by 5% in people exposed to air pollutants PM 2.5 , and 1% in those who exposed to NOx. There were sufficient evidences provide that ambient air pollutants play a role in glucose metabolism and type 2 diabetes etiology [35]. Studies have reported that airborne particulate matter PM 2.5 is rich in persistent semi-steroidal free radicals [36]. A multicountry time-series analysis provides evidence on positive associations between short-term exposure to PM 10 and PM 2.5 and mortality of cardiovascular disease and respiratory disease [37]. Besides,the relaxation of blood vessels and endothelial function decreases with increasing PM 2.5 [38]. After exposure to PM 2.5 , the levels of tumor necrosis factor (TNF-α), interleukin-6, resistin and leptin were elevated, and the expression of interleukin-10 was decreased, resulting in an increase in adipose tissue macrophages and dysfunction of intravascular cell barriers. The expression of pro-inflammatory and insulin resistance changes [39]. Increased tumor necrosis factor TNF-α and macrophage inflammatory proteins lead to hyperglycemia and deteriorating beta cell function [40].
Air pollution is an influential factor in the pathogenesis of diabetes. However, the mechanism by which air pollutants increase the risk of diabetes is not fully understood. The possible mechanism by which air pollutants increase the risk of diabetes is through:(1)Inducing and/or triggering the body's insulin resistance[41]; (2)They may act directly on the pancreatic tissue of the body, causing impaired beta-cell function in the pancreas to produce insulin, causing the body to not produce enough insulin.
They may also act directly on the body's cellular insulin receptor, thereby affecting the receptor's binding to insulin, causing insulin to function normally  Table2. Figure 1 Flow diagram of studies identified, included, and excluded.

Figure 2
Forest plots of air pollutants and diabetes risk. Forest plots of PM10 and risk of diabetes.

Figure 4
Forest plots of the relationship between PM10 and risk of type 2 diabetes.

Figure 5
Forest plot of PM2.5 and risk of diabetes.

Figure 6
Forest plot of NOx and risk of diabetes.

Figure 7
Forest plot of NO2 and risk of diabetes.

Figure 8
Forest plots of O3 and risk of diabetes.

Figure 9
Funnel plots of air pollutants.

Figure 10
Funnel plots of PM10 Figure 11 Funnel plots of PM2.5 Figure 12 Funnel plots of NOx Figure 13 Funnel plots of NO2.