We carried out a comprehensive search of articles published up to June 30, 2023 using PubMed, ISI Web of Science, EBSCO, CNKI and Wanfang Data with no restrictions in language or publication date. The following terms were used in our search: {(“stomach cancer”[all fields] OR “stomach neoplasm” [all fields] OR “gastric cancer”[all fields] OR “gastric neoplasm”[all fields] OR “cancer of the stomach”[all fields]) AND (“MedDiet”[all fields] OR “Mediterranean diet” [all fields] OR “Mediterranean”[all fields] OR “Dietary pattern”[all fields] OR “dietary score”[all fields] OR “dietary adherence”[all fields])}. In addition, a manual search in the reference lists from the retrieved articles or reviews or meta-analyses was performed to find the potentially eligible studies. All of these steps were accomplished by two independent reviewers (ZQ and SL), and any disagreements with article selection were resolved through discussion with Y-LF. Our selection criteria was based on the PECOS (e.g., participant, exposure, comparison, outcome, and study design) framework, which is shown in Supplementary Table S1.

Two independent reviewers (QZ and LS) carried out an initial screening of all titles and abstracts from retrieved articles to identify eligible studies that should be included in the analysis. Any disagreements were settled by discussion or in consultation with the third reviewer (Y-LF). When all reviewers agreed, the full-text articles were reviewed against inclusion and exclusion criteria for the present systematic review and updated meta-analysis. To be included in the present meta- analysis, articles had to meet the following criteria: (1) observational studies (i.e., case–control and cohort studies) conducted in adult population (aged ≥18 years); (2)considered adherence to the Mediterranean diet as the exposure; (3) assessed the association between adherence to the Mediterranean diet and risk of gastric cancer; (4) provided estimates of RRs, HRs, ORs with their corresponding 95%CIs; (5) If the data in retrieved article lacked sufficient detail, the corresponding author of the original study was contacted by email; (6) gastric cancer diagnoses were confirmed by clinical interviews, or self-report on a previous physician-made diagnosis of gastric cancer. Moreover, studies were excluded if they fulfilled one of the following criteria: (1) unrelated articles; (2) non-observational studies, e.g., reviews, editorials, case reports and conference letters; (3) animal, cell culture, and in vitro studies; (4) studies not reported as HRs, RRs or ORs with 95%CIs.

After completing selection of all eligible studies, two independent authors (LS and FZ) extracted the following information: first author’s last name, publication year, study design, country, sample size, number of gastric cancer cases, mean age/age range, follow-up time (cohort studies), components of the Mediterranean diet score, methods of dietary assessment, reported risk estimates (HR/OR/RR) and the corresponding 95%CIs and confounding factors that were adjusted for in the multivariate analyses. Any discrepancies and disagreements about data extraction were resolved by consensus or discussion with the third author (Y-LF).

The Newcastle-Ottawa Scale (NOS) was used to evaluate the overall quality of the included studies in the present study (29). We assigned 0 ~ 9 “stars” to each study based on three major domains: the population selection (maximum of 4 stars), comparability of the groups (maximum of 2 stars), and outcome/exposure assessment (maximum of 3 stars). For this analysis, we considered that an NOS score ≥ 7 indicated high methodological quality (10). Differences were resolved by consensus with a third author (ZQ).

The reported HRs in the primary studies were considered as equal as RRs (30). ORs were converted into RRs using the formula:RR = OR/[(1-P₀) + (P₀*OR)], in which P₀ indicates the incidence of the outcome of interest in the non-exposed group (31). Log-transformed RRs with their corresponding standard errors (SEs) were obtained using risk ratios (ORs, HRs, and RRs and corresponding 95% CIs) which were previously extracted for the relationship between adherence to the Mediterranean diet and risk of gastric cancer. Heterogeneity across studies was tested using the Cochran’s Q test and the I² statistic. If p-values of Cochran’s Q-test ≤0.10 or I² ≥ 50% indicated an absence of heterogeneity among studies, and the random-effects model (DerSimonnian and Laird method) was used to pool the RRs and 95%CIs of the highest versus the lowest category of Mediterranean diet in relation to gastric cancer. Otherwise, the fixed effect model is used (32). If significant between-study heterogeneity was observed, sensitivity and subgroup analyses would be performed to further find out the source of heterogeneity. In our analyses, subgroup analyses were performed based on sex (male or female), study design (cohort or case–control studies), anatomical site (gastric cardia or non-cardia), study area [Western countries or other countries (Afghanistan and Jordan)], mean age (≥50 or < 50), and sample size (<5,000 or >5,000). Sensitivity analysis was performed by removing one study at a time, and to clarify whether the results were robust or sensitive to the influence of a single study. If ≥10 comparisons were available, publication bias was evaluated through the visual inspection of the funnel plot and quantified by the Begg’s test and Egger’s test, respectively (33). If there was evidence of publication bias, we further evaluated the number of missing studies in a meta-analysis by the application of the Duval and Tweedie trim- and- fill method and recalculated the pooled estimates with the addition of those missing studies (34). Finally, we also performed a dose–response meta-analysis to estimate the RRs for each 1-score increment in Mediterranean diet adherence. A two-stage GLST model based on generalized least squares method was used to test the potential linear or non-linear dose–response association between adherence to a Mediterranean diet and risk of gastric cancer. We modeled Mediterranean diet scores by using restricted cubic splines with 3 knots at fixed percentiles (10, 50, and 90%) of the distribution. A p-value for curve linearity or non-linearity was computed by testing the null hypothesis that the coefficient of the first spline is equal to the second spline. All the mentioned data analyses were performed using STATA, version 12.0 (StataCorp, College Station, Texas, USA). A p-value less than 0.05 (two-tailed) was considered statistically significant.

Figure 1 indicates the process of study selection. Our initial search yielded 348 potential articles, of which 52 were duplicates. Of the remaining 296 articles, we excluded 225 articles on the basis of the titles and/or abstracts; 42 articles on the basis of irrelevant studies. Then, after reading the full-text versions of the remaining 29 articles, 18 articles were excluded for the following reasons: 7 were systematic review or meta-analyses, 4 did not evaluate gastric cancer risk; 2 were conference abstracts; 1 reported the same participants; 4 did not mention Mediterranean diet score. Finally, 11 studies met the eligibility criteria and were included in this meta-analysis.

Table 1 shows the general characteristics of all included studies. A total of 11 studies, including 5 prospective cohort (18, 21, 23, 25, 26) and 6 case–control (16, 17, 19, 20, 22, 24) studies, met the inclusion criteria and was included in this study. These included studies were published between 2010 and 2023. The age of participants ranged from ages 18 to above. Two of the included studies were carried out in the United States (21, 25), two in Spain (19, 22), two in Italy (17, 24), one in Afghanistan (16), one in Sweden (23), one in Netherlands (18), one in Jordan (20), and one study in European countries (26). Thus, United States, Spain, Italy, Sweden, Netherlands and European countries were regarded as Western countries, while Afghanistan and Jordan were regarded as other countries. In all included studies, dietary intake was measured by using an FFQ (16–26). For outcome assessment, four studies had used cancer registries (18, 21, 23, 25), two studies used medical records (20, 24), and five studies used pathology reports (16, 17, 19, 22, 26). All included studies had an NOS score ≥ 7, which were of high quality (16–26). Among these studies, three studies (18, 21, 25) separately reported the relationship between adherence to Mediterranean diet and gastric cardia adenocarcinoma and non-cardia adenocarcinoma. The number of study participants varied between 270 and 494,968.

Eleven articles comprising 5,708 gastric cancer cases and 1,366,318 participants, were included to assess the link between adherence to the Mediterranean diet and the risk of gastric cancer. Combining 14 effect sizes from 11 studies, Figure 2 showed the evidence of a decreased risk of gastric cancer in the highest compared with the lowest categories of Mediterranean diet (RR = 0.71; 95%CI:0.59–0.84, p < 0.001). There was evidence of substantial heterogeneity between studies (I² = 79.7%, p < 0.001), and therefore the effect was assessed using the random-effects model.

Ten studies (16–22, 24–26) containing 6 case–control studies with Mediterranean diet scores on the same scale (0–9) were included in the dose–response analysis for gastric cancer risk. A linear dose–response analysis showed that each 1-score increment in Mediterranean diet score was associated with a 5% lower risk of gastric cancer (RR = 0.95;95%CI:0.94–0.96, p < 0.001; P_{non-linearity} = 0.330) (Figure 3). The analysis of six case–control studies showed a positive linear relationship between Mediterranean diet and risk of gastric cancer (OR = 0.89; 95%CI:0.80–0.98; p = 0.018; P_{non-linearity} = 0.382) (Figure 4). In addition, the analysis of four cohort studies also showed a positive linear relationship between Mediterranean diet and risk of gastric cancer (HR = 0.98, 95%CI:0.96–0.99; p = 0.002; P_{non-linearity} = 0.538) (Figure 5).

Given the high heterogeneity of this study (I² = 79.7%; p < 0.001), subgroup analyses were performed to explore the potential sources of heterogeneity (Table 2). In this study, subgroup analyses were carried out basing on study design (cohort or case–control studies), study area (Western countries or other countries), mean age (≥50 or <50), sample size (<5,000 or >5,000), anatomical site (gastric cardia or non-cardia) and sex (male or female). When we conducted analyses separately by study design, results showed an inverse relationship between Mediterranean diet and risk of gastric cancer in cohort studies (RR = 0.88; 95%CI: 0.79–0.98, p = 0.024), with less evidence of heterogeneity between studies (p = 0.115; I² = 39.6%). In contrast, there was also significant association between Mediterranean diet and decreased risk of gastric cancer in case–control studies (RR = 0.44; 95%CI: 0.32–0.61, p < 0.001), with evidence of heterogeneity (p = 0.005; I² = 69.8%). For study area, we found a significant inverse association between adherence to Mediterranean diet and gastric cancer risk in other countries (RR = 0.28; 95%CI: 0.16–0.49, p < 0.001). There was no evidence of heterogeneity between studies (p = 0.822; I² = 0.0%). When the results were stratified by mean age, we found a significant inverse association between Mediterranean diet and gastric cancer risk in age ≥ 50 and age < 50 (for age ≥ 50: RR = 0.70; 95%CI: 0.57–0.87, p = 0.001 and for age < 50: RR = 0.71; 95%CI: 0.53–0.95, p = 0.022). The heterogeneity was most apparent in age ≥ 50 (p < 0.001; I² = 82.5%). For sample size, we found a significant inverse relationship between Mediterranean diet and gastric cancer risk in the subgroups of sample size > 5,000 (RR = 0.88; 95%CI: 0.79–0.98, p = 0.024). However, the heterogeneity was less (p = 0.115, I² = 39.6%). In addition, a protective association was found between Mediterranean diet and gastric cancer risk in the studies with sample size<5,000 (RR = 0.44; 95%CI: 0.32–0.61, p < 0.001), and there was significant heterogeneity (p = 0.005, I² = 69.8%). We also performed stratified analysis based on sex, and results showed the inverse association between Mediterranean diet and gastric cancer risk in male and female (0.77 vs. 0.91). however, there was no evidence of heterogeneity in female (p = 0.604; I² = 0.0%).

Funnel plots showed little evidence of asymmetry (Supplementary Figure S1) and therefore no evidence of publication bias existed (highest compared with lowest category of Mediterranean diet: Begg’s test, p = 0.324; Egger’s test, p = 0.161). Due to no evidence of publication bias, we could not perform the trim and fill analysis to adjust the pooled effect estimates. All included studies in our analyses have received a NOS score ≥ 7, and these studies were regarded to be of high quality (16–26) (Table 3).

Based on the results of sensitivity analysis, the pooled results of Mediterranean diet on gastric cancer did not materially change when removing any single studies in the main analysis (RR ranged between 0.65 and 0.76) (Supplementary Figure S2).

The current meta-analysis has several notable strengths. First, this is the first comprehensive systematic review and dose–response meta-analysis so far evaluating the association between adherence to the Mediterranean diet and gastric cancer risk. Our findings add the available evidence and underline the importance of supporting the people in adhering to the Mediterranean diet for prevention of gastric cancer. Second, we used a comprehensive search strategy of five main databases which identified all the observational studies available. Third, the cases of gastric cancer have been diagnosed through view of cancer registry or medical records or pathological records by clinicians, avoiding misdiagnosis. Fourth, no signs of publication bias were evident in the funnel plot, and the Begg’s and Egger’s tests for publication bias were non-significant. Thus, our results were relatively stable. Fifth, the quality assessment showed that all of the included studies in this meta-analysis were of high quality, and the reported ORs/RRs/HRs were multivariate and adjusted for some known confounders. Finally, the adequate number of included studies allowed us to perform subgroup analyses for some important risk factors, e.g., study design, sex and anatomical site. Besides, the dose–response analysis was performed to strengthen the relationship between adherence to the Mediterranean diet and gastric cancer risk. Despite the above-mentioned strengths, some limitations of the current meta-analysis should also be acknowledged in interpretation of our findings. First, 6 out of 11 studies included in current study used the case–control design, which is more prone to recall and selection biases. In addition, owing to the observational nature of included studies, the possibility of residual bias from unmeasured or unknown confounders remains. Hence, further prospective cohort studies or randomized controlled trials are needed to confirm the role of Mediterranean diet in the prevention of gastric cancer. Second, in all of the included studies, dietary intake was measured through FFQs, which carried an inherent recall bias. Meanwhile, the levels of the highest and the lowest categories of Mediterranean diet scores were inconsistent in the included studies, which might have attenuated the true association. Third, although some known confounding factors have been adjusted in all eligible studies, the existence of residual confounding from unmeasured factors cannot be completely excluded. Fourth, significant heterogeneity was observed in our analyses. Even though we performed subgroup and sensitivity analyses to explore the potential sources of heterogeneity, we could not ascertain and explain the sources of inter-study heterogeneity sufficiently. Finally, this meta-analysis had a geographical restriction, as the majority of included studies came from the United States and European countries. Hence, the generalization of our findings to other populations should be taken with caution. Further large prospective studies are needed in other populations with different environmental conditions and dietary preferences.