The association between interleukin-6 gene -174G/C single nucleotide polymorphism and sepsis: an updated meta-analysis with trial sequential analysis

We searched PubMed, Web of Science and China National Knowledge Infrastructure (CNKI) for available studies published before September 2018. The keywords for searching were a combination of “IL-6, interleukin-6, rs1800795, -174G/C, polymorphism, sepsis, septicemia, septic shock”. Meanwhile, references of the relevant literature reviews were screened to identify potentially relevant publications.

The publications fulfilling the following criteria were included: an original study evaluating the association between IL-6-174 G/C polymorphism and the risk and/or mortality of sepsis; objects in each study were from same epoch; including a case group of sepsis; including a control group; including precise sample size of IL-6-174 G/C polymorphism of case and control groups which could be directly extracted or calculated according to the information available. Excluded were: repetition of the published studies; a meta-analysis or a review; study with insufficient or incorrect data; study with only G or C allele carriers. The screening process is illustrated in a flow chart (Fig. 1). All publications were identified by two reviewers independently. If two reviewers had opposite views, a third reviewer would be consulted for a decision.

The quality of included studies was evaluated according to Newcastle-Ottawa Scale (NOS) by two researchers [15]. Available data were extracted by two authors independently. Disagreements were dealt with by a third reviewer. The following information was extracted from studies: first author, year of publication, country of the study and features of case and control groups such as ethnicity, age group, type of case and controls. Some studies discussed both the risk and mortality of sepsis. In this case, the information was collected respectively.

The role of minor allele C in the risk and mortality of sepsis was targeted. Thus an analysis was made using dominant model (GC + CC versus GG), recessive model (CC versus GC + GG), co-dominant model (GC versus GG and CC versus GG) and allelic model (C versus G). While some studies only yielded a sum of number of genotypes, they were enrolled in analyses just under certain genetic models [16‐19]. Statistical analysis was conducted on Stata 14.1. Each control group underwent goodness-of-fit χ²-test for evaluating Hardy-Weinberg equilibrium (HWE). When P > 0.05, the objects were under Hardy-Weinberg equilibrium and within a same Mendelian population. Homogeneity of the studies was tested. When P < 0.1 or I² > 50%, a high level of heterogeneity between studies was envisaged and random-effect model was adopted; when P > 0.1 and I² < 50%, there was no significant heterogeneity between studies, and fixed-effect model was used. The effect of IL-6-174 G/C on the risk or mortality of sepsis was measured by P value, odd ratio (OR) and 95% confidence interval (CI). P < 0.05 was reviewed as statistical significance. In the analysis on sepsis risk, subgroup analyses based on age group, ethnicity, restricted healthy controls were conducted. Subgroup analysis on age and ethnicity in the mortality analysis was also made. If no information of age group was given in a study, we considered it would be as a study on adult. The majority of the nation of study was taken as the ethnicity of the study population in case that it was not specifically mentioned. The objects in three studies were mainly Caucasians, so they were included in the subgroup analysis of the Caucasians as the number of non-Caucasians were very small [16, 20, 21]. Meanwhile, sensitivity analysis was conducted by repeating analysis after omitting one study each time to estimate the effect of quality of studies on the final result. Publication bias was evaluated by Egger’s test. Once P > 0.05 in regression test, no obvious publication bias would exist.

Meta-analysis might be affected by type I error due to the increased risk of random error and repeated significance testing. Trial sequential analysis (TSA) was a useful tool to verify the reliability of the results from meta-analysis by estimating the required information size (RIS) (sample size of included studies) and calculating the threshold for statistical significance [22]. A type I error of 5%, power of 80%, relative risk reduction of 20% were defined and control event proportion was an average of each included study. If the Z-curve crossed RIS line, the result of meta-analysis would be conclusive. If the Z-curve crossed the O’Brien-Fleming boundary or futility boundary, the conclusions could be made even before it crossed the RIS line that IL-6-174G/C polymorphism have or did not have a correlation with sepsis, respectively. TSA was conducted in the genetic model with the most included studies. If each genetic model had the same number of included studies, TSA would be conducted in allelic model. Meta-analysis which presented a significant result in the pool analysis was also tested under TSA. TSA was conducted on TSA software 0.9.5.10 Beta.

354 studies were initially acquired from databases and 4 studies were from references of publications. After exclusion of irrelevant studies, review and meta-analysis as well as repetitions, 29 studies were identified for full-text review. 5 studies were further excluded after full-text review because of inadequate data for analysis [23‐27]. Eventually, 24 studies fulfilled all criteria for inclusion, including 16 studies on the risk of sepsis, 4 studies on the mortality and 4 studies on both topics [10‐14, 16‐21, 28‐40]. All included studies had a NOS score ≥ 6 (see Additional file 1: Table S1). Totally, there were 3282 cases and 4926 controls for the analysis on sepsis risk together with 610 cases and 1856 controls for the analysis on mortality. Main characteristics of all studies included as well as the genotype distributions of IL-6-174 G/C polymorphism were shown in Table 1.

The results were shown in Table 2. The analysis resulted in no association between IL-6-174 G/C polymorphism and the risk of sepsis concerning the overall population under dominant model, recessive model, codominant model and allelic model (dominant model: P = 0.743, OR = 0.965, 95% CI: 0.782–1.192; recessive model: P = 0.96, OR = 0.992, 95% CI: 0.726–1.356; codominant model GC vs. GG: P = 0.57, OR = 0.950, 95% CI: 0.686–1.435; codominant model CC vs. GG: P = 0.966, OR = 0.992, 95% CI: 0.686–1.435; allelic model: P = 0.894, OR = 1.014, 95% CI: 0.831–1.236) (Fig. 2). When limiting control group to healthy ones or Mendelian population, the results remained unchanged. In the subgroup analysis on non-adults, adults and Caucasians, no association between IL-6-174 G/C polymorphism and the risk of sepsis was found, neither.

The results were shown in Table 2. In the analysis of overall population, IL-6-174 G/C polymorphism did not appear to be associated with the mortality of sepsis in dominant model (P = 0.812, OR = 1.050, 95% CI: 0.705–1.563), recessive model (P = 0.444, OR = 1.299, 95% CI: 0.665–1 = 2.538), codominant model (GC vs. GG: P = 0.825, OR = 1.045, 95% CI: 0.709–1.540; CC vs. GG: P = 0.465, OR = 1.340, 95% CI: 0.611–2.939) and allelic model (P = 0.558, OR = 1.103, 95% CI: 0.795–1.531). Subgroup analysis on adults and Caucasians did not show any association between IL-6-174 G/C polymorphism and mortality. Nevertheless, subgroup analysis on non-adults manifested that IL-6-174 G/C polymorphism was associated with worse survival of septic patients in recessive model (P = 0.003, OR = 2.957, 95% CI: 1.442–6.066), co-dominant model (CC vs. GG: P = 0.005, OR = 3.221, 95% CI: 1.431–7.250) and allelic model (P = 0.015, OR = 1.631, 95% CI: 1.101–2.416).

No obvious publication bias was found in every section of this meta-analysis in Egger’s test. The results and graphs of Egger’s tests were shown in Table 2 and supplementary figures (see Additional file 2). The results of sensitivity analysis were shown in Table S2 (see Additional file 1). The results of meta-analysis on the association between IL-6-174 G/C polymorphism and the mortality of non-adult (CC vs. GG and allelic model) were changed after omitting certain studies. Additionally, some relevant studies might influence the results of study were also tested in our analyses. An analysis after omitting two studies including elder children from non-adults group was also made, the results being unchanged [32, 33] (see Additional file 1: Table S3). Moreover, an study was included in the meta-analysis by Gao et al. but excluded in our study [26]. Thus an analysis was carried out and proved that the exclusion of this study did not influence the results of our meta-analysis (see Additional file 1: Table S4).

In the meta-analysis on the risk of sepsis, TSA was conducted on analysis of overall population (dominant model and recessive model), non-adults (dominant model), adults (recessive model), Caucasians (recessive) and analysis involving healthy controls (allelic model) and Mendelian population (allelic model). Z-curves crossed the futility boundary and reached the required sample size in those analysis. The exception was the analysis on non-adult group, where Z-curve crossed neither conventional test boundary nor the futility boundary (Fig. 3). In the meta-analysis on the mortality of sepsis, TSA was conducted on analysis of overall population (allelic model), non-adults (recessive model, codominant model: CC vs. GG and allelic model), adults (allelic model) and Caucasians (allelic model). Although Z-curves crossed the conventional test boundary in the analyses on non-adults, they did neither cross O’Brien-Fleming boundary nor cross the RIS boundary (see Additional file 2). Z-curves in other TSAs crossed futility boundary and RIS boundary.