Two separate systematic searches were performed: one to identify cytokines associated with H. pylori infection, and one to identify cytokines associated with GC. For the purpose of this systematic review and meta-analysis, we conducted a systematic search of all publicly available material found in Embase (1971-2022), Medline Ovid (1946-2022), Web of Science Core Collection (1975-2022), and the Cochrane Database (1992-2022). The following search terms were used: (cytokine OR chemokine OR interferon) AND (Helicobacter pylori OR H. pylori) AND (blood OR serum OR plasma OR biopsy) for the H. pylori meta-analysis, and (cytokine OR chemokine OR interferon) and (stomach cancer OR gastric cancer) and (blood OR serum OR plasma OR biopsy) for the GC meta-analysis. The full search strategies can be found in Supplementary File 1 . The meta-analysis followed the PRISMA guidelines (16).

Prior to the literature search, inclusion and exclusion criteria were specified. For studies to be included in this meta-analysis, they had to satisfy the following criteria: 1) written in English, Chinese or Dutch in peer-reviewed journals; 2) cytokine level assessment based on ex vivo sample measurement; 3) control group negative for H. pylori/free of stomach cancer described, 4) cytokine levels reported based on protein measurement. Studies were excluded if 1) they consisted of review articles, published abstracts, meta-analyses, editorials or case reports; 2) the method of H. pylori infection status determination was not described or gastric cancer status was not individually distinguished from other cancers; 3) cytokine levels were measured after in vitro stimulation of cells or in animal models; 4) quantitative cytokine levels were not reported for both cases and controls; 5) data was reported for ≤10 cases or controls.

Two authors independently screened each article based on the title and abstract, to assess whether or not the research was based on clinical data, whether or not levels of cytokines were reported, whether or not the H. pylori infection was current, and whether or not the full text was accessible. Any disagreements were resolved through further discussion. Data were collected from selected studies from article text and tables. Studies without written description of cytokine levels were excluded (i.e. data was not extrapolated from figures). Main data extracted for the meta-analysis was sample size, mean cytokine concentrations, standard deviation and P-value. Additional data included were year of study, country or region of patient inclusion, study setting (e.g. alternative underlying diseases), method of analysis of inflammatory markers and cytokines levels, method of H. pylori detection. For all selected publications, bias was assessed using Newcastle–Ottawa scale, following the recommendation of the Musculoskeletal Group (17).

Mean and standard deviation (SD) for cytokines levels were used when reported. When median and interquartile range (IQR) were reported, SD was computed from these data as described by Wan et al. (18, 19), while means were derived using the method described by Luo et al. (18, 20), in order to maximize the number of studies eligible for inclusion. When different subgroups were presented, which for the purpose of this review should be viewed as one group, means and SD were combined as described by Altman et al. (21, 22). The META pack in R (version 4.2.0) was used for meta-analysis. Study heterogeneity was described using I², which shows the proportion of variance attributable to factors between studies as opposed to sampling mistakes (23). We deemed I² values of more than fifty percent to be indicative of substantial heterogeneity (24). If a study reported on several groups (e.g. various diseases), we compared H. pylori-positive and -negative groups that had the same disease status individually. We expected a significant heterogeneity since reported quantitative levels of inflammatory cytokines may be strongly influenced by patients’ status and different ELISA kits. A random effects model was used to construct a weighted mean difference and 95 percent confidence intervals (CI) for continuous outcomes data. In the calculation of the uncertainty (confidence interval) around results, this meta-analytic method incorporates both within-study and between-study variation. A random effects model, unlike a fixed effects model, assumes that the genuine effects change from one study to the next. Significant heterogeneity in the results of the included studies yields broader confidence ranges in random effects models as compared to fixed effect models (25). To describe quantitative differences in cytokine levels between H. pylori-positive and -negative cases in meta-analysis, we followed the suggestion from Cochrane Handbook for Systematic Reviews of Interventions, using the standardized mean difference (SMD) to a uniform scale. The SMD quantifies the magnitude of the intervention effect in each study in relation to the variability seen in that study. Consequently, regardless of the exact scales used to perform the measurements, studies where the difference in means is the same fraction of the standard deviation will have the same SMD (26).

We performed sub-analyses for the H. pylori meta-analysis by grouping the individual studies per global region, divided as Europe, Russia, Africa, America, Middle East, East Asia, and Southeast Asia ( Supplementary Table S1 ). Only cytokines described in ≥10 studies were included in this analysis.

We analyzed the Pearson’s correlation between Age-Standardized Rate of Incidence of GC and the SMD of cytokines which our meta-analyses indicated were significant increased in both H. pylori infected individuals and GC patients. Data on the incidence of GC was retrieved from the International Agency for Research on Cancer, 2020 (Cancer today).

After removal of duplicates, 2250 unique papers were identified, of which 274 full text manuscripts were assessed for eligibility. A total of 54 publications investigating the association between the pro-inflammatory cytokines and H. pylori infection were considered for inclusion in this systematic review, of which 40 studies were suitable for meta-analysis. A flow chart is presented in Figure 1 . Included for analysis were 40 studies in which systemic cytokines were measured from serum. Nine studies measuring cytokines from biopsies through immune-based quantitative single or multiplex assays and 6 studies reporting on immunohistochemical analysis of formalin-fixed paraffin embedded (FFPE) sections, were not pooled in the analysis due to low numbers and different methods of reporting cytokine levels. Studies and patient data are summarized in Supplementary Table S2 . The results of risk of bias analysis for each study are presented in Supplementary Table S3 .

The following cytokines and chemokines were measured in relation to H. pylori infection in one or more studies: TNF-a, TNF-β, IFN-γ, CCL-1, CCL-2, CCL-3, CCL-7, CCL-8, CCL-11, CCL-13, CCL-15,CCL-17, CCL-19, CCL-20, CXCL-10, CXCL12, IL-1a, IL-1β, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-12, IL-13, IL-15, IL-16, IL-17A, IL-18,IL-27, IL-33, IL-35 and IL-37 ( Supplementary Table S4 ). Only cytokines reported in ≥3 studies were pooled, leaving eight cytokines available for further analysis: IL-1β, IL-2, IL-4, IL-6, IL-8, IL-10, IFN-γ and TNFα (see Table 1 ).

By far the most studied cytokine was IL-6, which was reported in 22 studies. In total, 3711 subjects were included in these studies. The overall SMD between H. pylori-infected and non-infected individuals was 1.93, 95%CI [0.63;3.23], indicating a large stimulatory effect of H. pylori infection on circulating IL-6 levels (P=0.0036). The second most commonly studied cytokine was TNF-α, which was investigated in 15 studies, including a total of 1224 individuals. On pooled analysis, serum TNF-α levels were found to be significantly increased upon H. pylori infection (SMD 0.88, 95%CI [0.46; 1.29], P ≤ 0.0001). In contrast, no effect of H. pylori infection on IL-8 levels was seen in 12 studies including 1469 subjects (SMD 21.33, 95%CI [-5.94; 48.60], P=0.1253). Similarly, no effect of H. pylori infection was observed in overall analysis of 11 studies investigating IL-10 levels in 811 subjects (SMD 0.78 95%CI [-0.93; 2.48], P=0.373). In addition, H. pylori infection did not modulate serum IL-1β (6 studies including 463 subjects, SMD 0.10, 95%CI [-0.39;0.59] P=0.682), IL-2 (4 studies including 452 subjects, SMD -3.99, 95%CI [-8.86; 0.87] P=0.1073), or IL-4 levels (5 studies investigating 452 subjects, SMD 1.84, 95%CI [-0.63;4.31] (P=0.144). IFN-γ, which was investigated in 459 individuals over 6 studies, did show significantly increased levels in serum upon infection with H. pylori (SMD 0.89, 95%CI [0.02;1.89] (P=0.0451).

Significant heterogeneity was present in the SMD for the majority of the comparisons, justifying the use of a random effects model. This is consistent with other meta-analyses investigating cytokine levels (25, 27–29), and may be caused by unavoidable variation in, for example, different kit usage and different geographical locations. Another possible factor influencing heterogeneity in the H. pylori meta-analysis is the method by which H. pylori-infected individuals are identified, as some assays (e.g. pathology reports of H. pylori presence in gastric mucosa, urea breath test) signify current infection, while other methods (stool antigen, anti-H. pylori IgG levels) may also show positivity for some time after eradication of the bacterium. Thus, we performed a sensitivity analysis, removing 21 studies which used anti-H. pylori IgG levels to identity infection status. However, while significant differences for IL-6 and TNF-α serum cytokine levels upon H. pylori infection were still observed, heterogeneity was only marginally reduced ( Supplementary Table S5 ). To further analyse heterogeneity, we investigated the impact of individual studies. Only two studies affected the I² by more than 5%. Upon exclusion of these studies, IL-1β serum levels were no longer found to be significantly enhanced upon H. pylori infection (exclusion of Raman Mishra et al. (30) IL-1β SMD -0.06 [-0.49; 0.36], P=0.7678, exclusion of C. Roubaud-Baudron et al. (13) IL-1β SMD 0.27 [-0.15; 0.69], P=0.2039). None of the other cytokines were affected.

Next, we analysed whether cytokine modulation upon H. pylori infection was dependent on the region of study. Only cytokines reported in ≥10 papers were included in this sub-analysis ( Table 2 ). IL-6 levels were increased upon H. pylori infection in East Asia (4.60, 95%CI [1.48; 7.72]), the Middle East (2.01, 95%CI [0.21; 3.80]) and Southeast Asia (although based on only one study 0.63, 95%CI [0.18; 1.09]), but not North America, Europe, Russia and Africa, suggesting that regional differences exist for some inflammatory responses induced by H. pylori (visualized in Figure 2A ). For TNF-α, sub-analysis of diverse global regions indicates that serum levels are significantly increased upon H. pylori infection in all global regions studied (East Asia: SMD 1.07, 95%CI [0.30; 1.85], Europe 0.36, 95%CI [0.11; 0.61], except the Middle East ( Figure 2B ). In contrast, IL-10 levels were not significantly affected in any individual region ( Figure 2C ), while sub-analysis of global regions indicates that IL-8 levels may be induced by H. pylori in Africa, although this only pertained to one study (SMD 1.99 95%CI [1.42;2.56], P ≤ 0.0001) ( Figure 2D ).

For meta-analysis of cytokine levels associated with GC, 4034 abstracts were screened, resulting in 229 full length manuscripts considered for eligibility. A total of 70 publications investigated the association between pro-inflammatory cytokines and GC, of which 61 studies investigating cytokine serum levels were included for meta-analysis ( Figure 3 ). The details of all included studies are presented in Supplementary Table S6 . Nine studies investigating cytokine levels in gastric biopsies through either immune-based quantitative analysis or immunohistochemistry were excluded due to low numbers. The risk of bias is presented in Supplementary Table S7 .

The 61 studies included compared TNF-α, TNF-β, IFN-r, CCL-2, CCL-5, CCL-18, CCL-20, CXCL-8, CXCL11, CXCL12, CXCL13, IL-1a, IL-1β, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-11, IL-12, IL-13, IL-15, IL-16, IL-17, IL-18, IL-23, IL-24, IL-26, IL-29, IL-32, IL-35, IL-37, TGF-β, TGF-β1, TGF-β2, IFN-λ1 and IFN-λ2/3 serum cytokine levels between GC patients and controls ( Supplementary Table S8 ). Cytokines investigated in three or more studies were IL-1β (7 studies; 1308 individuals included), IL-2 (5 studies, 1127 individuals), IL-4 (6 studies, 1148 subjects), IL-6 (19 studies; 2804 subjects), IL-7 (3 studies; 466 individuals included), IL-8 (12 studies; 2455 individuals), IL-10 (13 studies, 2512 individuals), IL-12 (7 studies; 1016 individuals), IL-17 (7 studies; 783 individuals), IL-18 (3 studies; 234 individuals), IL-33 (3 studies; 316 individuals), TNF-α (12 studies, 1642 individuals), TNF-β (3 studies, 506 individuals), and IFN-γ (12 studies, 1896 individuals) ( Table 3 ).

A significant upregulation of IL-6 serum levels was seen in GC patients (SMD: 1.64, 95%CI [0.88; 2.39], P<0.0001). Other cytokines showing a significant increase in serum from GC patients compared to controls were IL-7 (SMD: 1.04, 95% CI [0.45; 1.63], P=0.0006), IL-10 (SMD: 2.17, 95%CI [0.63; 3.70], P=0.0056), IL-12 (SMD: 2.1, 95%CI [0.91; 3.29], P=0.0056), and TNF-α (SMD: 2.34, 95%CI [0.29; 4.38], P=0.0251). While heterogeneity was high for all cytokines, none of the studies individually affected the I² with more than 5%.

We next set out to explore the relationship between serum cytokine changes upon H. pylori infection and the risk of GC development. IL-6 and TNF-α were the only cytokines presenting a significant increase upon both H. pylori infection and gastric cancer development. Therefore, serum levels of IL-6 and TNF-α after H. pylori infection were calculated individually for each country/region ( Figure 4 ) and correlated to the GC incidence in this same region. SMD of IL-6 levels presented a significant correlation with the relative incidence of GC (R = 0.85, p= 0.000035, see also Supplementary Figure S1A ). Visualization of the relationship between IL-6 levels (SMD) and the risk of GC using a bubble plot shows that in countries and regions with an age standardized rate (ASR) greater than the global average, higher IL-6 levels are induced upon H. pylori infection than in countries with a relative GC risk below the global average ( Supplementary Figure S1C ). However, such a correlation was not observed for TNF-α (R = 0.29, p= 0.41, Figure 4B ; Supplementary Figures S1B, S1D ).