Association between dioxin and cancer incidence and mortality: a meta-analysis

The objective of the present study was to systematically assess the association between dioxin/2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) and cancer incidence and mortality. Systematic literature searches were conducted until July 2015 in Pubmed, Embase and Cochrane library to identify relevant studies. A random-effects model was applied to estimate the pooled odds ratio (OR), risk ratio (RR), standard incidence ratio (SIR) or standard mortality ratio (SMR) for cancer incidence or mortality. In addition, dose-response, meta-regression, subgroup, and publication bias analyses were conducted. Thirty-one studies involving 29,605 cancer cases and 3,478,748 participants were included. Higher external exposure level of TCDD was significantly associated with all cancer mortality (pooled SMR = 1.09, 95% CI: 1.01–1.19, p = 0.04), but not all cancer incidence (pooled RR = 1.01, 95% CI: 0.97–1.06, p = 0.49). Higher blood level of TCDD was both significantly associated with all cancer incidence (pooled RR = 1.57, 95% CI: 1.21–2.04, p = 0.001) and all cancer mortality (pooled SMR = 1.45, 95% CI: 1.25–1.69, p < 0.001). Subgroup analysis suggested that higher external exposure and blood level of TCDD were both significantly associated with the mortality caused by non-Hodgkin’s lymphoma. In conclusion, external exposure and blood level of TCDD were both significantly associated with all cancer mortality, especially for non-Hodgkin’s lymphoma.

Data extraction and quality assessment. Data extraction was carried out independently by two investigators (Drs. Xu JM and Ye Y). Discrepancies were resolved by a third investigator. The endpoints of this analysis were all cancer incidence and mortality as most of the included studies adopted, as well as site/type-specific cancers. The following information was extracted from each study: authors, year of publication, country of each study, study period, population characteristics (sample size, gender and age), and cancer subtypes. ORs (RRs, SIRs or SMRs) reflected the greatest degree of control for potential confounders were adopted in this meta-analysis. The quality of each study was assessed according to NEWCASTLE-OTTAWA quality assessment 46 . The total score ranges from 0 to 9, and a higher score indicates higher quality. Sensitivity analyses are further conducted according to the quality assessment results to explore the source of heterogeneity.
Data synthesis and statistical analysis. The primary meta-analyses were conducted to assess the association between external exposure and blood level of TCDD and all cancer incidence and mortality. Heterogeneity between individual studies was assessed by the chi-square test and I 2 test; P ≤ 0.10 and/or I 2 > 50% indicates significant heterogeneity 47 . Summary ORs (RRs, SIRs or SMRs) and 95% CI were calculated using a random-effects model. The significance of the pooled ORs (RRs, SIRs or SMRs) were determined by Z test (p < 0.05 was considered to be significant). Studies that reported results of a specific type of cancer but no data on all cancer were not pooled for all cancer analysis. Subgroup analyses were applied to explore source of heterogeneity and to evaluate potential effect of modification of variables including cancer subtype, exposure way and TCDD exposure reference category. In order to avoid bias and make the analysis more accurate, subgroup results were shown in pooled form if there were three or more studies for one subtype, otherwise, it was listed in an original form. Funnel plots were constructed and Begg's and Egger's tests were performed to assess the publication bias (p ≤ 0.10 was considered to be significant).
We analyzed the dose-response relationship using first-order, and second-order, and three-order fractional polynomial regression of the inverse variance-weighted data to estimate a curve of best fit. Best-fit curves were selected using decreased deviance compared with the reference model 48 . Comparisons of curves to determine best fit were done using a chi-square distribution. The average values within the blood TCDD categories were specified as the midpoint for bounded ranges, and 0.75 times the higher bound for the lowest (unbounded) range, and 1.25 times the lower bound for the highest (unbounded) range. RRs or SMRs (the ratio of observed to expected cancer deaths multiplied by 100) was the response measure used in these studies. All analyses were conducted using Stata software (version 12.0; StatCorp, College Station, TX, USA).

Results
Study characteristics and data quality. After searching PUBMED, EMBASE and Cochrane library, 6446 articles were identified. 4437 articles were assessed after removing 2009 duplicate papers. Review of titles and abstracts resulted in exclusion of 4206 articles. For the remaining 231 articles, 163 were excluded for the following reasons: insufficient data (n = 60), foreign languages (n = 17), not on the right topic or targeted population (the outcomes of these studies were not cancer incidence or mortality, or the study interests were not dioxin) (n = 56), review articles (n = 14), meeting abstracts (n = 6), letters or comments (n = 10). 68 studies were included for further consideration and then 37 duplicate reports  from the same population were excluded. The detailed study selection methods for the same population are shown in Supplementary Table 1. Finally, a total of 31 studies  were included for the meta-analysis, including 22 cohort studies and 9 case-control studies. There were different TCDD exposure ways as follow: occupational exposure, non-occupational exposure, industrial accidents, and soldiers exposed to herbicides used in Vietnam War. The reference categories also varied among different studies, some adopted the non-exposed population to calculate SIRs or SMRs (external reference), and others adopted the lowest exposure categories (internal reference). We pooled the RRs or SMRs of high-exposed versus non-exposed categories for external reference, and highest versus lowest categories for the internal reference. Of note, all the included case-control studies only provided data on specific cancer types but no combined data on all cancer, and these studies were only pooled for the subgroup analysis but not for the all cancer analysis in order to ensure the accuracy of the results. The selection process is shown in Fig. 1, and the characteristics of the included studies are shown in Table 1. The exposure level and adjustment for confounders of included studies are shown  in Supplementary Table 2. Among the included studies, ten 13,20,22,25,31,32,34,[38][39][40] assessed the association between external exposure level of TCDD and cancer incidence. Eleven [10][11][12][15][16][17][18]20,21,29,30 evaluated the association between external exposure level of TCDD and cancer mortality. For blood and adipose tissue level of TCDD, seven 14,19,26,33,[35][36][37] assessed cancer incidence and seven 14,16,23,24,[27][28][29] evaluated cancer mortality. Ott et al. 14 reported the association between blood level of TCDD and both cancer incidence and mortality. Read et al. 20 reported the association between external exposure of TCDD and both cancer incidence and mortality. Steenland et al. 16 and Manuwald et al. 29 reported the association between both external exposure and blood level of TCDD and cancer mortality. The results of quality assessment were shown in the Supplementary Table 3. The scores of most studies ranged from seven to nine (except for two studies got six points), which indicated the high quality of included studies and enhanced the reliability of the analysis. The PRISM checklist and flow diagram were shown in Supplementary Tables 4 and 5, respectively.
External exposure of TCDD and cancer incidence and mortality. Ten studies involving 18,969 cancer cases and 3,155,159 participants assessed the association between external exposure of TCDD and cancer incidence, including five cohort studies and five case-control studies. The pooled RR of all cancer incidence of TCDD exposure level was 1.01 (95% CI: 0.97-1.06), indicating no significant association (Fig. 2a). There was significant heterogeneity across the included studies (I 2 = 73.5%, p < 0.001), as shown in Fig. 2a. Subgroup analysis was conducted according to cancer subtype, as shown in Table 2. The pooled RRs of different cancer types were all not significant, including breast cancer, Hodgkin's lymphoma, lymphatic leukemia, non-Hodgkin's lymphoma, and soft-tissue sarcoma. The results of subgroup analysis suggested the heterogeneity may be caused by special cancer types. Sensitivity analysis was also conducted to further explain the source of heterogeneity according to quality assessment results. After exclusion of the study 13 of the lowest score (six points), the pooled RR was 1.01(95% CI: 0.97-1.05), while the heterogeneity was not significantly changed (from I 2 = 73.5% to I 2 = 72.7%).
Eleven studies involving 9,122 cancer deaths and 691,326 participants assessed the association between external exposure of TCDD and cancer mortality. The pooled SMR of all cancer mortality of TCDD exposure level was  , indicating a significant positive association (Fig. 2b). There was significant heterogeneity across the included studies (I 2 = 90.8%, p < 0.001), as shown in Fig. 2b. Subgroup analyses for the association between external exposure of TCDD and cancer mortality were conducted according to cancer types and TCDD exposure ways, as shown in , and occupational exposed population (pooled SMR = 1.25, 95% CI: 1.07-1.46). Subgroup analyses suggested that heterogeneity was partly influenced by cancer type and TCDD exposure way (Table 2). To further explore the potential impact of within-study heterogeneity, we also conducted sensitivity analyses according to the quality assessment results. After excluded the study 10 of the lowest score (six points), the pooled SMR was 1.10 (95% CI: 1.01-1.20), while the heterogeneity was not significantly changed (from I 2 = 90.8% to I 2 = 91.2%). The efficiency of the current sensitivity analysis was not able to provide evidence to further explain the source of heterogeneity.

Blood level of TCDD and cancer incidence and mortality.
Seven studies comprising 837 cancer cases and 3,446 participants evaluated the association between blood of TCDD and cancer incidence, including three cohort studies and four case-control studies. The pooled RR of all cancer incidence for the highest versus lowest categories of TCDD exposure level was 1.57 (95% CI: 1.21-2.04), indicating a positive significant association (Fig. 3a). The I 2 and p value for heterogeneity across the included studies were 7.0% and 0.341 respectively, as shown in Fig. 3a. Subgroup analysis was not conducted due to the limited data. Seven studies involving 997 cancer deaths and 13,793 participants assessed the association between blood level of TCDD and cancer mortality. The pooled SMR of all cancer mortality for the highest versus lowest categories of TCDD exposure level was 1.45 (95% CI: 1.25-1.69), indicating a significant positive association (Fig. 3b). There was no significant heterogeneity across the included studies (I 2 = 4.7%, p = 0.394), as shown in Fig. 3b. Subgroup analysis was conducted according to cancer type, exposure way and reference category. Two studies assessed the association between blood level of TCDD and non-Hodgkin's lymphoma, and the SMRs (95% CI) were 4.50 (1.20-11.50) and 1.36 (1.06-1.74), respectively. The results suggested a significant positive association, which was consistent with the results of higher exposure level of TCDD. However, the results should be treated cautiously considering the relatively small sample size (n = 11), and more studies were needed to validate it. The subgroup analyses also indicated that it was all significant for occupational exposed and non-occupational exposed population, and for external and internal reference category, which further verified the stability of the results.
Dose-response analysis was conducted based on five studies 14,16,23,24,29 according to the model of two-order fractional polynomial regression. RRs or SMRs using the low exposure group as the reference group were not appropriate for the dose-response analysis, which needs the RRs or SMRs relative to the normal background  16,77 was obtained by personal communication by the authors 43 , thus we used these data extracted from Crump et al. 43 to improve the validity of our analysis. We adopted Manuwald et al.'s study 29 rather than Flesch-Janys et al.'s 63 for the Hamburg cohort since the former had a longer follow-up time. Cumulative serum lipid concentration (CSLC, ppt-years) was selected as the exposure metric to relate to risk, and the second-order fractional polynomial regression plot indicated a positive correlation between blood TCDD level and all cancer SMR, as shown in Fig. 4a. After log transformation of TCDD dose, the curve showed a non-linear increasing trend (Fig. 4b). The size of the circles in Fig. 4     Publication bias. Begg's funnel plots and Egger's linear regression test indicated no evidence of publication bias in the present study (TCDD external exposure and cancer incidence P Begg = 0.755 and P Egger = 0.245, and mortality P Begg = 0.150 and P Egger = 0.521; blood level of TCDD and cancer incidence P Begg = 1.000 and P Egger = 0.620, and mortality P Begg = 0.711 and P Egger = 0.834). The funnel plots were shown in Supplementary Figures 1 to 4.

Discussion
The current meta-analysis summarized the results of twenty-two cohort studies and nine case-control studies, including ten on external exposure level of TCDD and cancer incidence, eleven on external exposure level and cancer mortality, seven on blood level of TCDD and cancer incidence, and seven on blood level of TCDD and cancer mortality. The results indicated that higher external exposure level of TCDD was significantly associated with all cancer mortality but not all cancer incidence. For external exposure studies, the dioxin exposure ways, exposure quantification methods, reference categories, exposure level and adjustment for potential confounders differed greatly among included studies, which could cause heterogeneity and these results should be taken cautiously. Besides, there was a significantly positive association between higher blood level of TCDD and both all cancer incidence and mortality. The subgroup analysis for TCDD exposure mortality reported significant results for esophagus cancer, larynx cancer, kidney cancer, non-Hodgkin's lymphoma, myeloma, soft-tissue sarcoma and occupational exposed population. However, the IARC's review suggested that the evidence for specific cancers was strongest for lung cancer, soft-tissue sarcoma and non-Hodgkin's lymphoma 3 . The IARC's review listed the related publications, while they didn't distinguish the duplicated studies based on the same population and didn't provided quantitatively pooled results. Thus, the results of the current study may provide relatively more detailed indications on specific cancer types. Interestingly, the subgroup analysis also suggested consistence for increased mortality ratio of non-Hodgkin's lymphoma in both higher external exposure and blood level of TCDD, The present meta-analysis provided epidemiological evidence for the carcinogenic potency of TCDD and the subgroup analysis showed specific cancer sites. Importantly, the consistent results for non-Hodgkin's lymphoma mortality of both external exposure and blood level of TCDD may indicate its specific effect on hematopoietic system. Although the sample size was relative small in the blood level of TCDD and non-Hodgkin's lymphoma mortality subgroup analysis, the results of the included two studies were both significant, independently. The SMRs and sample size of non-Hodgkin's lymphoma by Collins et al. 23 and Boers et al. 27 were 4.50 (1.2-11.5, n = 4) and 1.36 (1.06-1.74, n = 7), respectively, which suggested possibility that the association may be especially significant for non-Hodgkin's lymphoma. It has been reported by Hardell et al. 86 that exposure to phenoxy acids, chlorophenols and organic solvents may be a causative factor in malignant lymphoma as early as 1981. And based on decades of research, it has been realized that, exposure to dioxins, in particular TCDD could induce chloracne 87 , and WHO has also classified it as a human carcinogen 3 . In consideration of the extensive sources, widespread trend and the strong toxicity of TCDD, the present results have considerable epidemiological and public health importance for humans. However its carcinogenic potential to humans and the mechanisms are not clearly demonstrated. It's commonly believed that AhR activation accounted for most biological properties of dioxins, including various physiological and developmental processes, tumor promotion, thymic involution, craniofacial anomalies, skin disorders and alterations in the endocrine, immunological and reproductive systems 50,88 . Furthermore, TCDD may also up-regulate drug-metabolizing enzymes, thus increasing the presence of highly reactive intermediates that form during metabolic activation and/or transformation of several key hormones 3 . Animal experiment also suggested that intraperitoneal injection of TCDD could cause increased incidence of lymphomas in male and female mice 89 .
Determining the sources of heterogeneity is an important goal of meta-analysis. The heterogeneity of our study mainly existed in external exposure level of TCDD and all cancer incidence (I 2 = 73.5%, p < 0.001) and mortality (I 2 = 90.8%, p < 0.001). Subgroup analyses suggested that cancer subtype and dioxin exposure way can partially explain heterogeneity across the studies. Sensitivity analysis was also conducted according to the quality assessment results, while the efficiency was not able to provide evidence to further explain the source of heterogeneity. However, the heterogeneity caused by different TCDD exposure ways, quantification methods, reference categories (internal or external), lag time, background exposure levels and adjustment for confounders couldn't be fully quantified due to the limitation of individual participant data. The future research should pay more attention to the unity of survey methods and the standardization of the exposure reference category to control heterogeneity.
Our study has several strengths. First, we adopted the external exposure and blood level of TCDD to thoroughly assess the association between TCDD and cancer incidence and mortality. Second, subgroup analyses and dose-response analyses were applied, which further strengthened the conclusions and emphasized the TCDD effects on some specific cancer sites. Although the 2012 IARC monographs 3 evaluated the evidence in humans for the carcinogenicity of TCDD and made a list of cohort studies, these issues were not systematically reviewed and quantified by a meta-analysis. Thus, the current meta-analysis fill in gaps in the IARC deficiencies on this issue and it's of considerable interest and public health importance. In addition, no publication bias was observed, indicating that the pooled results should be unbiased.
However, the current analysis is restricted by several limitations. First, the number of studies involved in blood level of TCDD and all cancer incidence was relatively small, and thus some of the subgroup analyses were difficult to conduct. Second, in the dose-response analysis, the normal background uncontaminated by occupational dioxin exposure was different, and only McBride et al. 24 study provided the New Zealand background level of 3.9 ppt. We didn't add the background exposure level to our analysis for the limitation of original data. Third, the Steenland et al. 16 used a 15-year lag time, whereas no lag was used in other cohorts. Although the Crump et al.'s analysis 43 inferred that results based on cumulative exposure lagged 15 years should not differ greatly from those based on unlagged exposure, this could cause inaccuracy and heterogeneity. Thus, the individual participant data meta-analysis is needed to enhance future analysis. Fourth, the subgroup analysis for blood level of TCDD and all cancer mortality was limited in digestive system, respiratory system, lung cancer, prostate cancer and non-Hodgkin's lymphoma. More studies with precise data of different cancer types are warranted to support the effects of TCDD on other cancers.
In conclusion, our findings suggest that external exposure and blood level of TCDD were both significantly associated with all cancer mortality. Higher external exposure of TCDD may significantly increase the mortality rate of esophagus cancer, larynx cancer, kidney cancer, non-Hodgkin's lymphoma, myeloma, soft-tissue sarcoma and occupational exposure population. Of note, such relationship may be especially significant for non-Hodgkin's lymphoma.