Selenium Exposure and Cancer Risk: an Updated Meta-analysis and Meta-regression

The objective of this study was to investigate the associations between selenium exposure and cancer risk. We identified 69 studies and applied meta-analysis, meta-regression and dose-response analysis to obtain available evidence. The results indicated that high selenium exposure had a protective effect on cancer risk (pooled OR = 0.78; 95%CI: 0.73–0.83). The results of linear and nonlinear dose-response analysis indicated that high serum/plasma selenium and toenail selenium had the efficacy on cancer prevention. However, we did not find a protective efficacy of selenium supplement. High selenium exposure may have different effects on specific types of cancer. It decreased the risk of breast cancer, lung cancer, esophageal cancer, gastric cancer, and prostate cancer, but it was not associated with colorectal cancer, bladder cancer, and skin cancer.

Selenium (Se) is an essential trace element having considerable and particular functions for human health because it is genetically encoded for which incorporation into proteins, as the constitutive part of selenocysteine, the 21 st amino acid 1 . Most se-proteins have been shown to have a wide range of pleiotropic effects, ranging from antioxidant to anti-inflammatory effects 2 , particularly the families of glutathione peroxidases (GPxs) and thioredoxin reductases (TrxRs) 1 , but their precise mechanism are not understood absolutely currently. Despite the scarce knowledge of mechanism, a large number of laboratory and ecologic researches focused on the associations between selenium and human health have been completed, showing that Se is associated with several human diseases including cardiovascular disease 3-5 , central nervous system disease 6 , diabetes mellitus 7-10 , and cancer, but the results are inconsistent.
We can see worldwide debates on the relation between selenium and cancer risk. Observational studies and randomized controlled trials suggest different effects in human. A new meta-analysis 11 published in Cochrane 2014 described the association between selenium and cancer prevention, and this article tended to analyze the effect of selenium supplement based on random controlled trials. There are other similar meta-analyses have been published, few of them established dose-response or beneficial range of selenium exposure associated with the risk reduction or determined the shape of dose-response curve to find whether it is a linear relation, saturation or U-shaped curve relation between selenium exposure level and cancer risk. On the other hand, numerous new studies have been reported in recent years, and we think it is meaningful to conduct an updated meta-analysis including different types of cancer to provide comprehensive evidence and clarify the shape of dose-response association between selenium status and cancer risk.

Methods
Search strategy. We carried out a systematic search for articles which described the relations between sele- , using a comprehensive list of selenium/ selenium supplement/ serum/plasma selenium/ toenail selenium/ antioxidant/ minerals And cancer/ breast cancer/ lung cancer/ esophageal cancer/ gastric/stomach cancer/ colorectal cancer/ bladder cancer/ prostate cancer/skin cancer). We also searched references of relevant studies and reviews Quality assessment. We applied the Newcastle-Ottawa scale 12,13 to assess the quality of the cohort and case-control studies. In this scale, one article was assessed on three perspectives: selection, comparability, outcome by using a "star system". The maximum score was nine stars. We simply regarded scores of 0-3 stars as low quality, scores of 4-6 stars as moderate quality, and scores of 7-9 stars as high quality. According to RCTs, we used the Cochrane collaboration's tool 14 for assessing risk of bias from six domains: selection bias, performance bias, detection bias, attrition bias, reporting bias and other bias. Results were presented as low risk of bias, unclear risk of bias or high risk of bias.
Statistical analyses. We extracted the multivariate-adjusted RRs, HRs or ORs and 95% confidence interval (CI) from original works. If some studies only provided 2× 2 table data, we calculated the responding ORs. We considered these estimates as ORs when took those studies with different designs into account, for RRs and HRs were assumed to be the accurate estimates of ORs. Meta-regression analysis was conducted to figure out whether the associations between selenium exposure and cancer risk were influenced by some covariates (exposure modes, area and design), and we could recognize the influence factor with a positive meta-regression coefficient(P ≤ 0.05). We used Greenland and Longnecker 15 method to conduct study-specific dose-response analyses based on the estimates of each category of plasma/serum selenium (ug/L), toenail selenium (ug/g) and selenium supplement (ug/d) respectively. We used mean or median of selenium exposure for each category when it was  presented, and used midpoint when selenium exposure ranges were presented. When highest or lowest categories of selenium exposure were unbounded, we assumed the category width to be the same as the adjacent one. Number of subjects or person-time and number of cases for at least three categories of selenium exposure were also needed in dose-response analyses. We used restricted cubic splines method 16 described by Orsini, N et al. to test the possible nonlinear relations, applying three fixed knots at 10%, 50% and 90% of selenium exposure level. The dose-response analyses were produced when there were more than 2 studies with relevant data. Pooled ORs of selenium exposure with 95% confidence intervals (CI) for cancer risk were conducted by using fixed or random effects model. Heterogeneity was examined by using Q 17 and I 2 18 index. When Q-test and I 2 statistics does not presented a notable heterogeneity (P > 0.05 and I 2 ≤ 50%), we used a fixed-effects analysis described by  . Otherwise, a random-effects analysis would be conducted described by DerSimonian and Laird method 20 . Publication bias was tested by Begger's test and a weighted Egger test 21,22 . We also conducted sensitivity analyses by omitting one study at a time to present relative influence of each study on pooled ORs. Statistical calculations and figures were produced with software STATA version 12.0 (StataCorp LP, College Station, TX, USA).

Results
Characteristics of the study. The meta-analysis included 69 studies (26 case-control studies, 14 cohort studies, 19 nested case-control studies, 5 case-cohort studies, 5 randomized controlled trials) reporting 114 independent estimates (as shown in table 1) from Asia (4 studies from China, 2 from Japan, and 1 from Malaysia, Iran, and India, respectively), Europe (8 from Netherlands, 5 studies from Sweden, 5 from Finland, 3 from Poland, 2 from Belgium, 1 from Northern Ireland, Britain, Germany and France, respectively, and 3 studies from European countries) and America (27 studies from the United States, 2 from Canada and 1 from Austria). There were more than 364742 participants with 26138 cancer events. 5 studies used all types of cancer as outcome, 14 studies used breast cancer as outcome, 13 studies used lung cancer as outcome, 5 studies used esophageal cancer as outcome, 6 studies used gastric cancer as outcome, 10 studies used colorectal cancer as outcome, 9 studies used bladder cancer as outcome, 25 studies used prostate cancer as outcome, 4 studies used skin cancer as outcome, 1 study regarded urinary tract cancer, pancreas cancer, leukemia/lymphoma, uterine and ovarian cancer as outcome respectively. 11 studies 23-33 mentioned above reported more than one cancer as an outcome, and several studies reported more than one estimate (as shown in table 1). 56 studies assessed biochemical selenium status: 37 used plasma/serum specimens and 19 used toenail specimens as exposure. 11 studies investigated selenium supplement or intake as exposure, using interviews or validated food frequency questionnaires. One study 34 used breast tissue selenium as exposure, and the last one 35 reported selenium intake, plasma selenium and toenail selenium as exposure respectively.
Selenium exposure and all cancer. The relation between selenium exposure and all cancer risk, represented 114 independent estimates from 69 studies (as shown in Table 1). Meta-regression was done to detect the possible influencing factors, and we found that exposure mode (plasma/serum selenium, toenail selenium or selenium supplement), area (Asia, Europe and America) and design (case-control, cohort or RCT) were not influencing factors (exposure mode: P = 0.388; area: P = 0.523; design: P = 0.715). Therefore, we took the 114 estimates into meta-analysis. The result of the pooled analysis showed that high selenium exposure had a protective efficacy on cancer at the highest compared with the lowest category (pooled OR = 0.78; 95%CI: 0.73-0.83), with obvious heterogeneity (Q = 423.52; P = 0.000; I 2 % = 73.3) and publication bias (Begger's test z c = 2.55, P = 0.011; Egger's test t = − 2.61, P = 0.010). Sensitivity analysis showed that the result was robust (as shown in Supplementary  Table S1). The heterogeneity was due to a large amount of included estimates and different types of cancer. The pooled result from 58 independent estimates showed that high serum/plasma selenium had a effect on cancer prevention at the highest compared with the lowest category (pooled OR = 0.75, 95%CI: 0.69-0.82, Fig. 2), with obvious heterogeneity (Q = 268.57; P = 0.000; I 2 % = 78.8) and publication bias (Begger's test z c = 2.54, P = 0.025; Egger's test t = − 2.43, P = 0.018). But the funnel plot was symmetry (supplementary Fig. S1). The heterogeneity could be due to a large amount of included estimates and publication bias. 17 groups of data were incorporated into dose-response analysis. The pooled OR was 0.95 (95%CI: 0.94-0.98) with 10 ug/L increase of plasma/serum selenium. Otherwise, we found obvious downward trends in the plots between plasma/serum selenium and total cancer risk in nonlinear dose-response analysis (P = 0.67 for non-linearity, Fig. 3).
There were 32 independent estimates describing the relation between toenail selenium and cancer risk. The result showed that high toenail selenium decreased cancer risk (pooled OR = 0.74, 95%CI: 0.62-0.87, as shown in Fig. 4), with obvious heterogeneity (Q = 70.95, P = 0.000; I 2 % = 56.3). There was no publication bias (Begger's test z c = 0.05; P = 0.961; Egger's test t = 0.52, P = 0.605), and the funnel plot did not show asymmetry (Fig. S2). 15 groups of data were incorporated into dose-response analysis. The pooled OR was 0.94 (95%CI: 0.92-0.97) with per 0.1 ug/g increase of toenail selenium. An downward trends was found in the plots of nonlinear dose-response analysis between toenail selenium and cancer risk (P = 0.500 for non-linearity, Fig. 5).
There were 23 independent estimates describing the relation between selenium supplement and cancer risk. The result showed that selenium supplement was not associated with cancer risk (pooled OR = 0.91; 95%CI: 0.80-1.03, Fig. 6  asymmetry (Fig. S3). However, we just extracted two relevant data for selenium supplement and all cancer risk, the linear or nonlinear dose-response analysis was not conducted. Selenium exposure and breast cancer. 18 estimates from 14 studies were incorporated in the pooled analysis. We found that exposure mode, area and design were not influencing factor (exposure mode: P = 0.417;   Table S1). We lacked sufficient data to conduct the linear or nonlinear dose-response analysis.
Selenium exposure and lung cancer. 13 estimates from 13 studies were incorporated into the pooled analysis. We found that exposure mode, area and design were not influencing factor(exposure mode: P = 0.706; area: P = 0.581; design: P = 0.705). Therefore, we took the 13 estimates into meta-analysis. The result showed that high selenium exposure presented a protective effect on lung cancer (pooled OR = 0.60, 95%CI: 0.41-0.88, Fig. 8 Table S1). 5 groups of data were incorporated into dose-response analysis between plasma/serum selenium and lung cancer risk. The result of linear dose-response analysis presented that plasma/serum selenium  was not associated with cancer risk per 10 ug/L increases of plasma/serum selenium (pooled OR, 0.92; 95%CI: 0.83-1.01, P = 0.0001). Otherwise, we did not find a threshold effect in the plot between plasma/serum selenium and lung cancer risk in nonlinear dose-response analysis (P = 0.182 for non-linearity; Fig. S4).
Selenium exposure and esophageal cancer. 7    analysis showed that the result was robust (Supplemental Table S1). We lacked sufficient data to conduct the linear or nonlinear dose-response analysis.
Selenium exposure and gastric cancer. 10 estimates from 6 studies were incorporated into the pooled analysis. We found that exposure mode, area and design were not influencing factor (exposure mode: P = 0.288; area: P = 0.077; design: P = 0.769). Therefore, we took the 10 estimates into meta-analysis. The pooled OR was 0.86 (95%CI: 0.77-0.96, as shown in Fig. 10) with moderate heterogeneity (Q = 22.63, P = 0.007; I 2 % = 60.2). There was no publication bias (Begger's test z c = 0.54; P = 0.592; Egger's test t = − 1.29, P = 0.235). Sensitivity analysis showed that the result was robust (as shown in Supplemental Table S1). We lacked sufficient data to conduct the linear or nonlinear dose-response analysis.
Selenium exposure and bladder cancer. 10 estimates from 9 studies were incorporated in the pooled analysis. We found that exposure mode, area and design were not influencing factor (exposure mode: P = 0.05; area: P = 0.708; design: P = 0.601). Therefore, we took the 10 estimates into meta-analysis. The result showed that high selenium exposure was not associated with bladder cancer (pooled OR = 0.76, 95%CI: 0.58-1.01, as shown in Fig. 12) with moderate heterogeneity (Q = 25.06, P = 0.003; I 2 % = 64.1), but without publication bias (Begger's test z c = 0.72; P = 0.474; Egger's test t = − 0.90, P = 0.395). 3 groups of data were incorporated into dose-response  analysis between toenail selenium and bladder cancer risk. The consequence of linear dose-response analysis presented that toenail selenium was not associated with bladder cancer risk per 0.1 ug/g increase of toenail selenium (pooled OR = 0.95, 95%CI: 0.90-1.01). Otherwise, we did not find a threshold effect in the plot between toenail selenium and bladder cancer risk in nonlinear dose-response analysis (P = 0.413 for non-linearity; Fig. S5) Selenium exposure and prostate cancer. 26 estimates from 25 studies described the association between selenium and risk of prostate cancer. We found that exposure mode, area and design were not influencing factor (exposure mode: P = 0.682; area: P = 0.362; design: P = 0.478). Therefore, we took the 26 estimates into meta-analysis. The result showed that high selenium exposure decreased risk of prostate cancer (pooled OR = 0.72, 95%CI: 0.61-0.86, Fig. 13), with moderate heterogeneity (Q = 81.6, P = 0.000; I 2 % = 69.4). There was no publication bias (Begger's test z c = 1.92; P = 0.055; Egger's test t = − 1.90, P = 0.070). Sensitivity analysis showed that the result was robust (Supplemental Table S1). 7 groups of data were incorporated into dose-response analysis between plasma/serum selenium and prostate cancer and 5 groups of data were included between toenail selenium and prostate cancer. The result of linear dose-response analysis presented that plasma/serum selenium was associated with prostate cancer risk per 10 ug/L increases (pooled OR = 0.97, 95%CI: 0.95-0.99; Q = 19.5, P = 0.003). The result presented that toenail selenium was associated with prostate cancer risk per 0.1 ug/g increases (pooled OR = 0.94, 95%CI: 0.89-0.99; Q = 20.27, P = 0.000). We did not find threshold effects in the plots between plasma/serum and toenail selenium and prostate cancer risk in nonlinear dose-response analyses (P = 0.739, P = 0.886 for non-linearity, respectively; Fig. S6,S7).
Selenium exposure and risk of skin cancer. 6 estimates from 4 studies were incorporated into the pooled analysis. We found that exposure mode and area were not influencing factor (exposure mode: P = 0.395; area: P = 0.454). Therefore, we took the 6 estimates into meta-analysis. The result of the pooled analysis showed that high selenium exposure was not associated with skin cancer (pooled OR = 1.09, 95%CI: 0.98-1.21, Fig. 14), with no heterogeneity (Q = 3.65, P = 0.601; I 2 % = 0.0) and publication bias (Begger's test z c = 0.00; P = 1.000; Egger's test t = 0.42, P = 0.697). Sensitivity analysis showed that the result was robust (Supplemental Table S1).
Other subgroup analysis. The further stratified analysis were conducted by gender and study design. The results indicated that the protective effect of high selenium exposure had no gender difference (as shown in Table 2). When stratified by design, we found the results from observational studies presented the protective effect of selenium on cancer while the results from RCTs indicated null effect (as shown in Table 2).
The results of meta-analysis suggest an inverse relation between selenium exposure and the total cancer risk (including breast cancer, lung cancer, esophageal cancer, gastric cancer, colorectal cancer, bladder cancer, prostate cancer, skin cancer, not site-specific cancer and other cancer). What is more, the results of dose-response analysis show a downward trend between plasma/serum selenium, toenail selenium and total cancer risk. The biomarker of selenium (serum/plasma and toenail selenium) was associated with cancer risk and could be easily measured and monitored to evaluate people health status. However, our results find that selenium supplement is not associated with cancer risk. Selenium supplement contains either inorganic or organic species or a mixture of both. The SELECT trial uses L-selenomethionie as an oral supplement, while the NPCT trial uses selenium yeast tablets. The different types of selenium supplement may present different effects on human health. On the other hand, first-pass elimination and bioavailability of different participants should be considered. Burk et al. 94 , study presents the results that the full expression of selenoprotein P requires more Se intake than that required by the full expression of GPxs, indicating that the Se intakes of the current studies are probably inadequate for optimizing the protective effects. We also cannot exclude the possibility that it is what associated with higher biochemical selenium level presents the shielding effect other than selenium exposure itself. We know that RCTs should research the association between selenium supplement and cancer risk, while observational studies usually research the relation between the biomarker of selenium and cancer risk. These reasons could be used to explain the inconsonant consequences of our stratified analysis by study design. And Vinceti, M 11 et al.' meta-analysis also find the inconsistent results between RCTs and observational studies. Future mechanism studies should pay more attention to the procedure from selenium supplement to biochemical selenium status to figure out the reasons for inconsonant effects of selenium supplement and biochemical selenium for preventing cancer. And future epidemiological studies and intervention trials should try to research selenium supplement, plasma/serum selenium and toenail selenium at the same time to reduce the potential bias.
We also find that selenium has diverse effects on specific types of cancer. According to breast cancer, we find an inverse relation when taking all relevant estimates into account. Nonetheless, we lack sufficient data to conduct dose-response analyses. According to lung cancer, we find that high selenium exposure presents a protective efficacy. Though a downward trend is seen in the nonlinear dose-response analysis, there is no statistical significance between plasma/serum selenium and lung cancer risk in linear dose-response analysis. The association between lung cancer and selenium exposure needs more discussion. According to esophageal cancer and gastric cancer, we find an obvious inverse relation. The quantity of estimates included in meta-analyses is not as many as other types of cancer, and we lack sufficient data to conduct dose-response analyses. According to colorectal cancer, we find no association between selenium exposure and cancer risk. Nevertheless, Ou Y et al. 95 meta-analysis which included seven studies showed a protective effect of selenium on colorectal adenomas (OR = 0.67; 95%CI: 0.55-81). Selenium exposure probably plays a protective role in colorectal benign tumor rather than cancer, and the results need more researches. According to bladder cancer, we find no statistical significance between selenium exposure and bladder cancer. However, Amaral A F et al. 96 meta-analysis which included seven epidemiologic studies presents that plasma/serum selenium and toenail selenium have protective effects on bladder cancer risk. According to prostate cancer, we find a protective effect of high selenium exposure for prostate cancer. The results of linear dose-response analyses between plasma/serum selenium, toenail selenium and prostate cancer support the result, and downward trends are shown in nonlinear dose-response analyses. However, two randomized controlled trials (the NPCT trial 28 and the SELECT trial 27 ) focusing on selenium supplement present the consequence that selenium supplement is not associated with prostate cancer risk. According to skin cancer, we find selenium is not associated with skin cancer risk.
There are numerous hypotheses about the potential anticarcinogenic mechanisms of selenium. The major positive effect may be contributed by the antioxidant function of GPxs and selenoprotein P 94 . Selenium is associated with the regulation of protein folding via the function of the endoplasmic reticulum to influence the process of necrosis and apoptosis of malignant cells 97,98 . Selenium also has the effect on DNA stability 98 . However, different malignant cells have their special biological characteristics and microenvironment for progress and invasion. They probably have disparate abilities of utilizing selenium. Hence, selenium probably has no effect on some types of cancer. The exact mechanism has yet to be investigated. On the other hand, the adverse effects of selenium supplement: mainly diabetes 27,99 , glaucoma 28 , and dermatologic alterations 27 could not be ignored. So we should try to clarify what level of selenium supplement is needed for adequate nutrition and at what level dose is "unsafe".
Our meta-analysis has several limitations clearly. Measurement errors in the assessment of selenium exposure may bias the effect estimates. Even among those studies regarding the same biochemical selenium as exposure, different measurement methods, different facilities and different staffs are all easy to produce measurement errors, and it is hard to make corrections. As showed in our inclusion criteria, we select case-control studies, cohort studies and RCTs into our meta-analysis. Selenium exposure may be linked to other behaviors like age, income, race, smoking status, alcohol consumption, body mass index, physical activity. These controlled confounding factors differ among sixty-nine studies and may influence the association between selenium exposure and cancer risk. Because of the insufficient number of relevant estimates, we have limited power to conduct subgroup analysis of pathological types of different cancer, and other controlled confounding factors.
Our study also has a few strength. We bring in a large number of studies and have largely avoided some main influencing factors by meta-regression analyses. And the robust outcomes of sensitivity analysis suggest that there is no distinct date making particularly contribution to the results. We detect the association between selenium exposure and different types of cancer to find a comprehensive understanding from global effects to local effects. We also conduct linear dose-response analyses which are stricter than high-versus-low analysis and the results of nonlinear dose-response analyses show dose-response trends in plots which are visual and accessible.

Conclusions
High selenium exposure could decrease cancer risk, especially high plasma/serum selenium and toenail selenium. High selenium exposure may have dissimilar effects on specific types of cancer. Future epidemiological studies and intervention trials should try to research selenium supplement, plasma/serum selenium and toenail selenium at the same time to reduce the potential bias. The exact mechanism needs to be further investigated.