Association of interferon regulatory factor 4 gene polymorphisms rs12203592 and rs872071 with skin cancer and haematological malignancies susceptibility: a meta-analysis of 19 case–control studies

We searched the PubMed and Web of Science databases for all relevant articles regarding IRF4 SNPs associated with cancer risk (search last updated on April 10, 2014). The following keywords were used: “IRF4” or “interferon regulatory factor 4”, “polymorphisms or variant or SNP or mutation” and “cancer or tumor or neoplasm or carcinoma”. The search was conducted exclusively on human subjects. The reference lists of reviews and retrieved articles were simultaneously hand-searched. We did not consider abstracts or unpublished reports.

All abstracts of citations and retrieved studies were reviewed. The following criteria were used to identify eligible published studies: (i) the study evaluated the association between the IRF4 polymorphisms (rs12203592 and rs872071) and cancer risk; (ii) the publication was a case–control or case-cohort study; (iii) the paper provided sample size, distribution of alleles, genotypes or other information that can help us estimate an OR with 95% confidence interval (95% CI); (iv) the genotype distribution of the control population is consistent with Hardy-Weinberg Equilibrium (HWE). Accordingly, publications were excluded using the following criteria: (i) articles that were not about cancer research; (ii) the publication contained duplicated previous research; (iii) the study did not include usable genotype data were excluded.

Two investigators independently extracted information from all eligible publications according to the inclusion criteria listed above. The results were compared, and disagreements were resolved by discussion until a consensus was reached. Data extracted from each study included the following characteristics: the first author’s name, the year of publication, the country of participants, ethnicity, cancer type, source of control group (population- or hospital-based controls), and the genotype frequency of the rs12203592 and rs872071 polymorphism in the cases and controls.

All statistical analysis was performed using STATA software (version 11.0; STATA Corporation, College Station, TX). Two sided P-values < 0.05 were considered statistically significant. Our meta-analysis recalculated HWE in the controls for each study. The goodness of fit test (chi-square or Fisher’s exact test) was used to assess deviation from HWE (significant at the 0.05 level). Studies that deviated from HWE were removed.

ORs with 95% CIs were used to estimate the strength of the association between the IRF4 rs12203592 and rs872071 polymorphisms and skin cancer and haematopoietic malignancies risk. In addition, the Z-test was also used, and a P value < 0.05 indicated statistical significance for the association. We examined the association between the rs12203592 IRF4 polymorphism and skin cancer and haematopoietic malignancies using the homozygote comparison (TT versus CC), heterozygote comparison (CT versus CC), dominant genetic model (TT + CT versus CC), recessive genetic model (TT versus CT + CC) and additive genetic model (T versus C). The same methods were applied to the analysis of the rs872071 IRF4 polymorphism. In addition, a stratified analysis was also performed based on cancer type, ethnicity and the source of controls.

The assumption of heterogeneity was ascertained using a chi-based Q-test. A P-value less than 0.05 for the Q test indicated significant heterogeneity among the studies. The pooled OR was estimated using a fixed- or random-effects model, where appropriate. If the P value was less than 0.05 indicative of heterogeneity across studies, a random-effects model (DerSimonian and Laird) was utilised for the meta-analysis [29]. Otherwise, a fixed-effect model (Mantel-Haenszel) was used [30]. We also quantified the effect of heterogeneity using the I ² test. I ² values of 25, 50, and 75% were indicative of low, moderate, and high heterogeneity, respectively.

The sensitivity analysis was conducted by removing one study at a time to evaluate the quality and consistency of the meta-analysis results. Publication bias was qualitatively and quantitatively assessed using the Begg’s funnel plots and Egger’s test, respectively [31]. To ensure the reliability and the accuracy of the results, two reviewers independently assessed the data using the statistical software programmes and obtained the same results.

A total of 179 potential individual publications were initially identified after a systematic literature search of the PubMed, Embase and Web of Science databases. The titles, abstracts and full texts of the retrieved articles were reviewed based on the inclusion criteria shown in Figure 1. Finally, we identified 11 eligible articles comprised of 19 studies for the current meta-analysis [24, 27, 32‐40]. In all the eligible articles, studies by Han et al., Wang et al., Broderick et al., Di Bernardo et al. include different sets of data, and each set of data was treated as a separate case–control study in this meta-analysis (Table 1). Our meta-analysis included 11 of the rs12203592 polymorphism with 7,992 cases and 8,849 controls and 8 studies of the rs872071 polymorphism with 3108 cases and 8300 controls. All studies were published in English. The control genotype distributions of all studies were in accordance with HWE. Detailed study characteristics included in the current meta-analysis are presented in Table 1.

A summary of the meta-analysis findings on the association between the rs12203592 and rs872071 IRF4 polymorphisms and skin cancer and haematopoietic malignancies risk is presented in Table 2. With respect to the rs12203592 polymorphism, a total of 11 studies from 5 articles were included in this meta-analysis. Among the 11 studies, 8 focused on skin cancer and 3 on NHL. Eight studies used Caucasian populations, and 3 were from USA with mixed ethnicity. Four studies used population-based controls, and 7 used hospital-based controls. The overall analyses suggested a significant association between the rs12203592 polymorphism and skin cancer and haematopoietic malignancies susceptibility in the homozygote comparison model (OR = 1.566, 95% CI 1.087–2.256) and recessive model (OR = 1.526, 95% CI 1.107–2.104, Figure 2a) using the random-effects model. The results from the other genetic models were not significant. Specific data for the rs12203592 polymorphism were stratified by cancer type into the NHL subgroup or the skin cancer subgroup (melanoma, BBC and SCC). For skin cancer, a significantly increased risk was observed using the homozygote comparison model (OR = 1.728, 95% CI 1.145-2.608) and recessive model (OR = 1.808, 95% CI 1.127-2.900). However, no significant association was observed in the NHL subgroup. In the subgroup analysis stratified by ethnicity, a significantly elevated cancer risk was found among Caucasians with the homozygote comparison model (OR = 1.566, 95% CI 1.087–2.256) and recessive model (OR = 1.526, 95% CI 1.107–2.104) but not among Asian populations with all genetic models. When stratified by the source of controls, a significantly increased risk was observed in the hospital-based studies under four genetic models (homozygote comparison: OR = 2.094, 95% CI 1.314-3.336; dominant genetic model: OR = 1.314, 95% CI 1.002-1.723; recessive genetic model: OR = 1.959, 95% CI 1.310-2.931; additive model OR = 1.35, 95%CI 1.058 -1.725), whereas no significant association was observed in the population-based studies for all five genetic models.

With respect to the rs872071 polymorphism, a total of 8 studies from 6 articles were included. Of the 8 eligible studies, 4 focused exclusively on CLL, 3 on lymphoma and 1 on MM. Caucasian populations were used in 6 studies, and Asians were assessed in 2. Five of the studies were population-based controls, and 3 used hospital-based controls. Overall, a significantly elevated haematological malignancies risk was associated with the rs872071 polymorphism in all genetic models (homozygote comparison: OR = 1.805, 95% CI 1.402-2.323; heterozygote comparison: OR = 1.427, 95% CI 1.203-1.692; dominant: OR = 1.556, 95% CI 1.281-1.89; recessive: OR = 1.432, 95% CI 1.293-1.587, Figure 2b; additive: OR = 1.349, 95% CI 1.201-1.515). After stratification by cancer type (HL and NHL were merged as lymphoma) and source of controls, a significant association was also observed with all of the genetic models. When we stratified the studies by ethnicity, a significant association was also observed in Caucasians under all genetic models. In Asians, the association was significant in all genetic models except for the recessive genetic model (OR = 1.233, 95% CI 0.924-1.646). The meta-analysis results for the subgroups are listed in Table 2.

For the rs12203592 polymorphism, all genetic models showed significant heterogeneity. After subgroup analysis by cancer type, the heterogeneity was effectively removed in the NHL subgroup. In the analysis of ethnicity, the heterogeneity significantly disappeared in the mixed subgroup. When stratified based on the source of controls, the heterogeneity also disappeared in the population-based control subgroups. The heterogeneity values are presented in Table 2.

Significant heterogeneities were observed in the overall analysis of the association between the rs872071 polymorphism and haematological malignancies risk in four genetic models (homozygote comparison, heterozygote comparison, dominant genetic model and additive model). After subgroup analysis by cancer type, the heterogeneity effectively disappeared in the CLL subgroup and lymphoma subgroup. In the analysis of ethnicity, the heterogeneity was significantly removed in the Asian population but remained in the Caucasians. When stratified based on the source of controls, heterogeneity was not observed in the hospital-based control subgroups. The heterogeneity values are presented in Table 2.

Publication biases of the literature were investigated using the Begg’s funnel plot and Egger’s test. With respect to the rs12203592 polymorphism, the results of the Begg’s funnel plot suggested no publication bias. The Egger’s test did not show statistical evidence for publication bias (homozygote comparison model: t = 1.38, P = 0.201). The shapes of the funnel plots did not reveal any evidence of obvious asymmetry for the rs872071 polymorphism with all genetic models (Figure not shown). Similarly, the results of the Egger’s test indicated a lack of publication bias (homozygote comparison model: t =0.61, P = 0.556).Sensitivity analyses were also performed in the current meta-analysis to assess the influence of each individual study on the pooled ORs by sequential removal of individual studies. For both the rs12203592 and rs872071 polymorphisms in the IRF4 gene, the results suggested that no individual study significantly altered the pooled results, thereby suggesting that the results of this meta-analysis are robust and reliable (Figure 3a and 3b).