PubMed, SCOPUS, EBSCO and LILACS databases were searched for all studies that investigated for any IL6 polymorphisms and RA. The following keywords/index terms and any of their derivations were used: “Interleukin 6 or IL-6 or IFNB2”, “rheumatoid”, and “variant or SNP or polymorphism or genotype”. The search was performed without any language restrictions for publications published until December 15, 2018. Afterwards, the complied publications references were hand searched.

Two authors independently determined if each study was to be included. If a disagreement occurred about a publication, a third author analyzed the publication in question. Initially, the titles and abstract were examined to determine if the article focused on RA and IL-6. Afterwards, the publications were thoroughly examined for the IL6 polymorphisms and the genotype distribution. For inclusion, the studies must have met the following criteria: (1) case-controls studies; (2) examined for any IL6 polymorphisms; (3) focused on human subjects; (4) RA-confirmed patients using either the American College of Rheumatology or the European League Against Rheumatism24 criteria; and (5) contained information about genotype frequencies. Studies were excluded if they were: (1) not a case-control study; (2) information was used in a previous publication; or (3) meta-analysis, reviews, or editorial articles.

Two authors independently assessed the quality of the studies using the Newcastle–Ottawa Quality Assessment Scale.25 The following aspects of each study were appraised: selection of cases and controls, comparability, and outcome or exposure. For the analysis, the possible quality scores ranged from 0 to 9 (see Supplement information). Studies that scored≥6 were considered as a high-quality study. The following data was collected from each study: first author's name, year of publication, geographical location, diagnosis criteria of RA, technique used to detect the polymorphism, source of controls, and genotype distribution for cases and controls. Before a study was to be excluded for missing information, we attempted to contact the corresponding author by email at least three times.

For each study, the Hardy–Weinberg Equilibrium (HWE) was determined by the ψ2-test for the controls and was considered in agreement when the p-value was>0.05. The crude odds ratios (ORs) and 95% Confidence Intervals (95% CI) were used to assess the strength of the association between the IL6 polymorphisms and the risk of RA. The pooled crude ORs were calculated for allelic (2 vs. 1), dominant (12+22 vs. 11), recessive (22 vs. 12+11), heterozygous (12 vs. 11), and homozygous (22 vs. 11) genetic models, where for the −174 polymorphism 1=G (Wild-type) and 2=C (mutant), for the −572 polymorphism 1=G (Wild-type) and 2=C (mutant), and for the −597 polymorphism 1=G (Wild-type) and 2=A (mutant). Heterogeneity was determined using the ψ2-based Q-test and its degree was assessed by the Inconsistency Index (I2). If there was not significant heterogeneity (ψ2-based Q-test p-value≥0.10 and I2<50%), the fixed-effects model was used (Mantel–Haenszel method), or if there was significant heterogeneity (ψ2-based Q-test p-value<0.10 and I2≥50%), the random-effects model was used (DerSimonian and Laird method) to calculate the pooled OR and 95%CI. Sensitivity analysis, removing one study and recalculation of the pooled OR, was conducted to verify the stability of the results. Begg's funnel plot, Begg–Mazumdar's test, and Egger's linear regression test were used to assess for publication bias. For geographic sub-analysis, initially, the studies were categories based on their country into Northern Africa, Sub-Saharan Africa, Latin America and the Caribbean, Northern America, Central Asia, Eastern Asia, South-eastern Asia, Southern Asia, Western Asia, Eastern Europe, Northern Europe, Southern Europe, Western Europe, or Oceania, according to United Nations M49 standard.26 Afterwards, the regions were group into Eastern Europe, Western Europe (Northern, Southern, and Western Europe), Latin America, Asian (Eastern and South-eastern Asia), and Middle East (Northern Africa, Southern and Western Asia). The Middle East category was based on the criteria that most countries identify as Arab and/or Muslim.27 All statistical analyses were conducted by using Comprehensive Meta-analysis v2 (Biostat, Inc., Englewood, New Jersey, USA). Unless noted otherwise, p-values <0.05 (two-sided) were considered statistically significant.

After duplicate removal, our literature search resulted in the recovery of 708 publications that focused on the association between the IL6 polymorphisms and patients with arthritis (Fig. 1). To note, 7 of the hand search articles were identified by the reference list, but the titles were originally in Chinese. After reviewing titles and abstract, 79 publications were considered for full-article examination. Of these publications, 24 were not case-control studies, 10 lacked sufficient data for analysis, 5 focused on the wrong pathology, and 7 were duplicate studies. Therefore, this meta-analysis contains 33 publications, which were divided between the −174 polymorphism (29 studies: cases=5920 and controls=6246), the −572 polymorphism (13 studies: cases=1973 and controls=1963), and the −597 polymorphism (6 studies: cases=1278 and controls=1524). The characteristics of the studies are presented in Table 1. 27.3% of the studies are from Asians countries: China,10,11,17,28–31 Taiwan,32 Japan.33 18.2% are from Eastern European countries: Russia,20,23 Czech Republic,34 Macedonia,35 Poland.18,36 24.2% are from Western European countries: Germany,37 Netherlands,12 Spain,38–40 Sweden41 and UK.42,43 21.2% are from the Middle East: Egypt,9,14,19 Iraq,22 India,15,44 and Turkey.8 Lastly, 3 studies are from Latin America.21,45,46 All the studies used the American College of Rheumatology criteria for RA diagnosis. For only the Pavkova et al. and the Lo et al. studies, the data was represented as allelic frequencies.32,34 Using the Newcastle-Ottawa scale, none of the studies were determined to contain significant study bias, but the Guseva et al. study9 and the Marinou et al. study9 could possible contain potential study bias. For the controls with respect to HWE, 5 studies for the −174 polymorphism,8,17,29,37,45 1 study for the −572 polymorphism,9 and 4 studies for the −597 polymorphism8,17,37,38 were determined to be not in agreement.

Fig. 1.

Flow chart for literature review of studies to be included in the meta-analysis.

(0,4MB).

The ORs and 95%CIs were calculated for each study for all 5 genetic models (all Forest plots are available as supplement information). Using the selected studies, for the −174 polymorphism, there was significant heterogeneity for all the genetic models; therefore, the random-effects model was utilized (Table 2). When the studies were pooled together, the heterozygous (OR: 1.47, 95%CI: 1.13–1.91, p=0.004), dominant (OR: 1.50, 95%CI: 1.15–1.96, p=0.003), and allelic (OR: 1.38, 95%CI: 1.13–1.68, p=0.001) genetic models demonstrated a significant association. Removing one study did not change the association for any of the genetic models.

Publication bias was assessed by examining the funnel plot. The funnel plots demonstrated no significant asymmetry and the shape of the funnel plots suggested no evidence of publication bias (see Supplement information), even though some over-dispersion can be seen. No correlation was determined by the Begg-Mazumdar's test or bias by Egger's test for the homozygous and recessive genetic models (Table 2); however, there was the presence of publication bias for the heterozygous, dominant, and allelic genetic models (Begg–Mazumdar's test: p<0.01, Egger's test: p<0.01).

When the cohort was stratified by geographic region, for the Asian population, the −174 polymorphism did demonstrate a strong association for all the genetic models (Table 3). For Middle East countries, the −174 polymorphism showed an association for the heterozygous (OR: 2.30, 95%CI: 1.10–4.81, p=0.028), dominant (OR: 2.27, 95%CI: 1.22–4.22, p=0.010), and allelic (OR: 2.29, 95%CI: 1.24–4.23, p=0.008) genetic models. Interestingly, for Latin America, the −174 polymorphism demonstrated a protective benefit for the homozygous (OR: 0.26, 95%CI: 0.08–0.82, p=0.022) and recessive (OR: 0.25, 95%CI: 0.08–0.80, p=0.019) genetic models.

For the −572 polymorphism, the homozygous and heterozygous genetic models did not present with a significant level of heterogeneity and were analyzed using fixed-effects, whereas the rest of the genetic models were analyzed with random-effects. Only the homozygous (OR: 1.60, 95%CI: 1.21–2.12, p=0.001) and recessive (OR: 1.61, 95%CI: 1.11–2.33, p=0.012) genetic models demonstrated a positive association between the polymorphism and the risk of developing RA (Table 2). Removing the Li et al. study9 resulted in a significant positive association for the dominant genetic model (OR: 1.32, 95%CI: 1.08–1.63, p=0.008), none of the other studies had any effect. For the remaining genetic model, the results were resilient to any change from removing one study. No publication bias was determined by examining the Funnel plot, Egger's bias test, or by Begg–Mazumdar's correlation test.

When stratified by geographic region, it appears that the association was due to the inclusion of studies from the Asian population (Table 4). For the Asian population alone, there was an association between the −572 polymorphism and the development of RA for the homozygous (OR: 1.56, 95%CI: 1.16–2.09, p=0.004) and recessive genetic models (OR: 1.63, 95%CI: 1.08–2.45, p=0.020). No association was observed for the Middle East region.

For the −597 polymorphism, all the genetic models were analyzed with the fixed-effects model, except the allelic genetic model. When the studies were pooled together, there was no effect for any of the genetic models (Table 2). However, removing the Arman et al. study,8 a positive association for the heterozygous (OR: 1.33, 95%CI: 1.01–1.74, p=0.040) and dominant (OR: 1.30, 95%CI: 1.01–1.68, p=0.040) genetic models was observed. No publication bias was determined by examining the Funnel plot, Egger's bias test, or by Begg–Mazumdar's correlation test. No sub-analysis could be performed for the −597 polymorphism, due to the few studies.