Correlation between circulating interleukin-18 level and systemic lupus erythematosus: a meta-analysis

This study is a meta-analysis aimed at pooling reported data and clarifying the association between circulating level of interleukin-18 and systemic lupus erythematosus (SLE). We searched medical databases including Medline/Pubmed, Embase, Scopus, The Cochrane Library, and Web of Science thoroughly to obtain all related articles published before July 15th, 2020. We pooled computed standardized mean difference (SMD) and its 95% confidence interval using STATA 13.0 and exhibited in the form of forest graph. Meta-regression and subgroup analysis were also performed to explore the source of heterogeneity. Publication bias was first evaluated by the symmetry of the funnel plot and then Egger’s linear regression test. Thirty eligible studies from eighteen regions were finally included and the relevant data from these studies were pooled. The analysis results displayed that SLE patients showed a significantly higher level of circulating IL-18 level in comparison with healthy controls (SMD = 1.56, 95% CI [1.20–1.93]; I2 = 94.9%, p < 0.01). The conclusion was equally applicable in subgroups divided based on sample type, mean age, disease duration, and testing method. Patients with SLEDAI score higher than five, or who were Asian, White, Arab, or mixed ethnicity had an elevated level of IL-18, while the others didn’t. This meta-analysis has elucidated that compared with healthy people, the circulating level of IL-18 is considerably higher in SLE patients, which indicates the underlying role of IL-18 in SLE pathogenesis.

Inclusion and exclusion criteria. Articles which met following standards were going to be contained for further data processing: (1) patients diagnosed as SLE under a definite standard and matched healthy controls had been recruited; (2) circulating IL-18 level from SLE patients and healthy controls were measured and reported either in plasma or serum; (3) study types including the cohort, case control and cross-section; (4) the content was written in English. There were not any restrictions on ethnicity.
We had removed reviews, conference abstracts, editorials, commentaries, case reports and studies performed on non-humans for lack of required information. Studies had also been exempted if SLE patients were pregnant, infected or with other autoimmune diseases. As for repeated or duplicated studies, the more comprehensive and updated one would be selected.
Data extraction and quality assessment. We reviewed all potentially eligible articles again in order to extract useful data and information, including author, region, year of publication, ethnicity, study design, number, mean age, gender, sample type, IL-18 circulating level of SLE patients and normal controls, measurement method, diagnostic standard of SLE, disease duration and SLEDAI score. Data was extracted directly if the study reported the mean and standard deviation of circulating IL-18 level. If values were presented in median, standard error, range and inter quartile range, they were converted into mean and standard deviation for analysis using previously reported formulae 14 . In the meantime, for the sake of evaluating study quality of each article, Newcastle-Ottawa-Scale (NOS) criteria were exploited for case-control studies and cohort studies 15 . The NOS criteria are composed of three assessing aspects, namely selection, comparability and outcome assessment with a highest score of nine. Agency for Health-care Research and Quality (AHRQ) cross-sectional study quality assessment which contained eleven elements was used for cross-sectional studies. If the aggregated NOS score or AHRQ assessment score is above six, the study quality can be regarded as relatively reliable. Two reviewers (YiLun Wang and Jie Wang) completed the extraction independently and the final consent was met in case of any disagreements with the intervention of another reviewer (Jun Liang).

Statistical analysis.
We obtained the mean and standard deviation of circulating IL-18 level from the included studies and then calculated the standardized mean difference (SMD) and its 95% confidence interval. All SMD values were pooled and presented in the form of forest plot. As for research heterogeneity, the Cochran's Q statistic and Higgins I-squared statistical analysis (I 2 = [(Q − df)/Q] × 100%) was undertaken 16 . If the p value is less than 0.05 or calculated I 2 is over 50%, significant heterogeneity may exist and hereby a random-effect model would be used, or else the fixed-effect model ought to be utilized. Meta-regression and subgroup analysis were also performed to explore the source of heterogeneity. Publication bias was first evaluated by the symmetry of funnel plot and then Egger's linear regression test was tried to assess more precisely 17 . If a significant publication bias was detected, the trim and fill method was conducted to yield an unbiased effect size through re-computing the probable missing studies 18 . What's more, the sensitivity test was simultaneously made to ensure the stability of this meta-analysis. The entire data analysis process was completed using STATA 13.0 (STATA, College Station, TX). If the p value is under 0.05, it would be recognized as statistical significance.

Results
Basic features and quality assessment. Utilizing the search strategies listed above, a total of 914 studies were retrieved. 546 of them were then browsed by the title and abstract after removing 368 duplicates. Only 105 were scanned further by full text and eventually 30 of them were confirmed the eligibility for inclusion 11,12, . The flow chart of the detailed inclusion and exclusion process was displayed in Fig. 1 All relevant information was listed in Table 1 including disease duration and SLEDAI score of SLE patients. As for study quality assessment, most of the included studies can be recognized as reliable quality. The score of each research for NOS scale or AHRQ assessment was listed in Table 1.

SLE patients had high levels of circulating IL-18 compared with healthy controls.
We pooled data from altogether 30 studies after computing SMD and its 95% CI for each of them. Results displayed that SLE patients showed significantly higher levels of circulating IL-18 in comparison with healthy controls (SMD = 1.56, 95% CI [1.20-1.93]). Heterogeneity was obvious (I 2 = 94.9%, p < 0.01) and therefore, a random-effects model was applied (Fig. 2). Subgroup analyses based on age, ethnicity, sample type, disease duration, SLEDAI score, testing method and main meta-analysis results were summarized in Table 2

Sensitivity analysis and meta-regression.
Since the heterogeneity of the association between circulating IL-18 and SLE was pronounced, thus additional sensitivity test was done (Fig. 3). There were no conspicuous alternations detected when removing one of the included studies and pooling the rest. On the whole, it indicated that the primary results of this meta-analysis were relatively robust.
To further explore the source of heterogeneity, meta-regression was utilized and continuous variables including publication year, mean age of SLE groups, group size, quality score, disease duration and SLEDAI score were seen as co-variables at a time, yet none of them had shown statistical significance (p > 0.05). Also, we transformed categorical variables into dummy variables and performed meta-regressions with sample type, study design and ethnicity. With these elements contained in the regression model separately, the main meta results hadn't changed with p value all above 0.05.  Trim and fill method was conducted since asymmetry was still presented in the funnel plot (Fig. 4). After filling in the three hypothetical missing studies, the adjusted SMD remained significant (SMD = 1.22, 95% CI [0.68-1.56], p < 0.01) which was in accordance with previous outcomes (Supplementary Fig. S3).

Discussion
This meta-analysis clarified the association for circulating IL-18 and SLE by containing 30 published correlated studies of 1968 SLE patients and 1439 healthy controls from 18 regions. After processing all extracted relevant data, we found that compared with healthy people, the circulating level of IL-18 was much higher in SLE patients, which indicated the underlying role of IL-18 in SLE pathogenesis. What's more, such an association was likewise significant when included studies were subdivided based on sample type, disease duration, patient age, and testing method. In patients whose SLEDAI score was less than five or from Latin America, the IL-18 didn't elevate significantly. Sensitivity analysis presented that the primary result remained unchanged when any of these included studies were exempted. Publication bias was not detected in Egger's linear regression test with p value above 0.05. In all, the conclusion that circulating IL-18 is closely connected with SLE can be drawn and it can be utilized as a new immune marker to identify SLE patients. Conclusions drawn from our meta-analysis were in an agreement with most of the included individual studies which strengthened the reliability. In addition, several recent studies have reported correlations between IL-18 and disease severity, organ involvement and a list of classic testing biomarkers like anti-dsDNA antibody, C3, C4 and etc 32,37 . In lupus nephritis, IL-18 is also related with proteinuria and renal activity and elevates earlier than the occurrence of proteinuria and high disease severity score, indicating that IL-18 has the chance to be used as a predictive marker to distinguish diverse stages of LN, even in subclinical stage 41 . These updated findings can help broaden the clinical applications of IL-18 and enrich its comprehensive significance in SLE.
In our study, the increased circulating IL-18 was detected in both Asian, White, mixed, and Arab group but not in Latin American population. But since only one study was from Brazil of Latin American and the size of this study was relatively small, only 16 SLE patients and 14 healthy controls, the conclusion was still in need of confirmation by a well-designed large-scale study 27 . In the Asian and White group, conclusions were more www.nature.com/scientificreports/ reliable to some extent since over ten studies were contained in each subgroup. What's more, subgroup analysis had revealed that the correlation with IL-18 and SLE was significant in patients who had > 5 SLEDAI score, but not in the group where the SLEDAI score was below five. The results indicated that IL-18 has certain connections with SLE disease activity and may serve as a biomarker to identify the severity of SLE. Another study had proven that serum IL-18 differed in SLEDAI-2k ≤ 4 group versus the SLEDAI-2k > 4 group. The longitudinal analysis had confirmed that variation of serum IL-18 level from baseline was associated with SLEDAI-2K change after following up 11 . In lupus nephritis, a significant correlation had been detected between serum IL-18 and SLEDAI, renal activity score and activity index 38 . This analysis also has several limitations that need to be carefully considered. First, the study population of several included studies is relatively small or from local medical clinics which would reduce the quality and explanatory power in our meta-analysis. Second, heterogeneity is a general problem that most of the metaanalyses may encounter. In our analysis, we had chosen a random-effect model to pool the data and detailed subgroup analyses and sensitivity analyses were performed. The still existing heterogeneity may generate bias and unreliability to some extent. Besides, due to the shortcomings of meta-analysis, our study was unable to answer the causative relationship between SLE and IL-18 and longitude data were lacked. Still, there are abundant advantages that assure the meaning of this meta-analysis. First, we searched altogether five databases to retrieve published studies as much as we can. Second, thirty studies from 18 separate regions including 1986 SLE patients and 1439 healthy controls were gathered, which was, by and large, geographically and numerically comprehensive. Besides, to better illustrate our results, subgroup analyses based on consequential clinical dimensions were done and sensitivity analysis guaranteed the reliability. www.nature.com/scientificreports/

Conclusion
To sum up, this meta-analysis found that elevated circulating IL-18 was observed in SLE patients compared with normal healthy individuals which was influenced by SLEDAI score and ethnicity, but not sample type, age group, disease duration and testing method. In the future, more well-designed or large-scale studies ought to be arranged to strengthen and explore the functions of IL-18 not only in the mechanism but also in clinical application of SLE.

Data availability
All data generated or analyzed during this study are included in this published article (and its Supplementary  Information files).