A systematic literature search was performed in MEDLINE (via PubMed), Web of Science and Cochrane Central Register of Controlled Trials (CENTRAL) on 29th September 2022 (17). The following search strategy was performed for MEDLINE:

((“Signet Ring Cell”[tiab] OR “Signet Ring Cells”[tiab] OR “Signet Cell”[tiab]) AND (cancer[tiab] OR carcinoma*[tiab] OR adenocarcinoma*[tiab] OR neoplas*[tiab] OR tumor[tiab] OR tumors[tiab] OR tumour*[tiab] OR malignan*[tiab])) OR “Carcinoma, Signet Ring Cell”[Mesh].

AND

(neoadjuvant*[tiab] OR neo-adjuvant[tiab] OR preoperativ*[tiab] OR pre-operative[tiab] OR perioperativ*[tiab] OR peri-operative[tiab] OR “followed by”[tiab] OR following[tiab] OR “Neoadjuvant Therapy”[tiab] OR “Neoadjuvant Therapy” [MeSH].

AND

radiochemotherap*[tiab] OR chemoradiotherap*[tiab] OR chemoradiation*[tiab] OR chemotherap*[tiab] OR “Chemoradiotherapy”[Mesh]) OR “Induction Chemotherapy”[Mesh].

The full search strategies for the other databases are available in the Supplementary material.

Additionally, a hand search through references of relevant studies was performed.

Randomized and non-randomized, prospective and retrospective two-arm studies, including patients with esophagogastric adenocarcinoma with signet ring cells treated with neoadjuvant or perioperative chemotherapy followed by surgery compared to surgery alone (nCTX + surgery vs. surgery), were eligible for inclusion. All studies investigating survival or histopathological response to nCTX in adenocarcinoma including signet ring cells were included. Studies with exclusively adjuvant chemotherapy, any chemoradiotherapy or radiotherapy were excluded. Animal studies, meeting abstracts, letters, comments, editorials, publications for which the full text is irretrievable and non-English studies were also excluded.

Titles and abstracts were reviewed independently by two reviewers (SS, NC) to select full papers for further evaluation. If there were disagreements additional reviewers (LP, RK) were consulted. Any disagreement was resolved by consensus.

Data was extracted independently by two reviewers using a standardized form composed prior to data extraction. The following items were extracted: title, first author, year of publication, country, study period, sample size, type of chemotherapy, tumor localization, stage, definition of SRCC, survival outcomes and histopathological response data.

If there was a training and a validation set, the data of the validation set was extracted. If hazard ratios (HR) were not explicitly reported the HR was estimated by the formulas proposed by Tierney et al. (18). The original survival curves were extracted using WebPlotDigitizer (19) to calculate HR.

Primary statistical analysis and meta-analysis were performed with RStudio, version 2023.03.1 + 446 using the package meta (20, 21). A random-effects model was used to account for methodological and clinical heterogeneity. Statistical heterogeneity among the effect estimates of the included trials was evaluated using the I² statistic and τ². According to Cochrane heterogeneity was interpreted as follows: 0%–40% low, 30%–60% moderate, 50%–90% high, and 75%–100% considerable. The HR and its standard error were used as effect measure for OS. Data was pooled using the inverse-variance method. The response rate was pooled as odds ratio (OR) with 95% CI using the Mantel–Haenszel (M-H) method (22). The results were graphically illustrated by forest plots.

For randomized studies, the risk of bias and quality was assessed by using the Cochrane Collaboration tool for assessing risk of bias 2.0 (23). The tool includes five standard domains of bias: “bias arising from the randomization process”, “bias due to deviations from intended interventions”, “bias due to missing outcome data”, “bias in measurement of the outcome” and “bias in selecting of the reported result”. These domains were rated as high risk of bias, low risk of bias, some concerns or unclear. Finally, an overall risk of bias was evaluated.

For non-randomized studies, the risk of bias assessment was conducted in accordance to ROBINS-I tool (24). Each of the following domains were evaluated: “bias due to confounding”, “bias in selection of participants into the study”, “bias in classification of intervention”, “bias due to deviations from intended interventions”, “bias due to missing data”, “bias in measurement of outcomes”, “bias in selection of the reported result”. Each potential source of bias was graded as low risk, moderate risk, serious risk, critical risk or no information/unclear. An overall judgement of the presence of bias in each study was made.

Furthermore, for each outcome the certainty of the evidence was rated to be very low, low, moderate, or high for each outcome using the GRADE system (25). This includes limitations in the design from the risk of bias assessment, indirectness of evidence, unexplained heterogeneity or inconsistency of results, imprecision of results, and publication bias.

526 studies were identified by database search, 33 additional sources were identified by screening the references of reviews and meta-analysis of current esophagogastric cancer literature, resulting in a total of 559 articles screened for this review. 129 of these articles meet the inclusion criteria and were evaluated in full-text. From these, 119 trials were excluded because of nSRCC examination or missing histology data (n = 21), wrong study type (n = 25), neoadjuvant chemoradiotherapy (n = 22), adjuvant chemotherapy or no information about treatment (n = 28), missing data (n = 12) and other reasons (n = 11). Ten studies were finally included for meta-analysis. A PRISMA flow chart of study selection is shown in Figure 1. One study was a RCT (26). Nine studies were non-RCTs with one prospective and eight retrospective analyses (12, 13, 27–33). Of ten studies six were from European and four from Asian countries. Three studies (12, 29, 30) analyzed locally advanced cancer patients whereas all other studies analyzed patients at all tumor stages.

A summary of the included studies is shown in Table 1.

The surgical procedure depended on localization of tumor. Five studies (13, 26, 27, 30, 33) analyzed gastric cancer patients. Iwasaki et al. as well as Jiang et al. used total gastrectomy ± lymphadenectomy (D2/D3) for resection. Li et al., Messager et al., Xu et al. used subtotal or total gastrectomy for resection ± lymphadenectomy (D1 or D2). Four studies analyzed patients with gastric cancer and carcinoma of the esophagogastric junction (EGJ) (12, 29, 31, 32). For all gastric cancer patients subtotal or total gastrectomy + D2 lymphadenectomy was used for resection. EGJ carcinoma were treated by transhiatal or abdominothoracic esophagectomy or transhiatal extended gastrectomy + lymphadenectomy (D1, modified D2 or D2).

One study (28) analyzed adenocarcinoma of the esophagus and esophagogastric junction but gave no detailed information about type of resection.

The chemotherapy regimens used in the neoadjuvant setting varied significantly and differed within some studies over time depending on new study results and recommendations. The majority of patients received a combination of platinum- and fluoropyrimidine-based CTX. In some studies triple chemotherapy with additional anthracycline and/or taxane was administered. Only a small fraction of patients (1.9%) received the FLOT regimen. An overview of the used regimens is given in Table 1.

The definition of SRCC varied between the studies. Four studies used the WHO definition (>50% SRC) (13, 27, 29, 33), whereas four studies included patients with tumors containing SRC of any percentage (12, 26, 28, 31). Two studies (30, 32) gave no information for their SRCC definition.

One study (26) was rated using the Revised Cochrane risk-of-bias tool for RCTs. The overall risk was with some concern. All other nine studies (12, 13, 27–33) were rated using the ROBINS-I tool. Overall, the included studies had a low to moderate risk of bias. Due to the limited number of studies included in this meta-analysis assessment of the publication bias via funnel plot was not possible.

The risk of bias of each study for every domain is shown in Supplementary Tables S2, S3. Table 2 gives an overview of the overall risk for each study.

For this meta-analysis three analyses were performed.

The first analysis included four studies analyzing patients with SRCC comparing OS for nCTX + surgery vs. surgery alone (12, 13, 26, 27). The pooled HR indicated no beneficial effects on OS for neither intervention and marked heterogeneity between the studies (HR, 1.01; 95% CI = 0.61–1.66; p = 0.98; I²= 89%). Results are shown in Figure 2A. In total there were 1,455 patients, 468 patients (32.2%) were treated with nCTX + surgery and 987 patients (67.8%) underwent surgery only. Three primary studies indicated an OS benefit for surgery alone over nCTX + surgery whereas one study revealed a significant OS benefit for nCTX + surgery over surgery alone (12). The chemotherapy regimens used differed between and within the four studies (see Table 1).

Risk of bias assessment resulted in low to moderate risk of bias in one study, moderate risk of bias in two studies and some concerns in one study. The certainty of evidence was very low.

The second analysis compared three studies analyzing OS for patients treated by nCTX followed by surgery comparing SRCC to nSRCC patients (28–30). OS of SRCC patients was significantly worse compared to nSRCC patients (HR, 1.45, 95% CI = 1.21–1.74; p < 0.01; I²= 0%). For this analysis 1,276 patients were included, of which 362 patients (28.4%) had SRCC and 914 patients (71.6%) nSRCC. Two of the analyzed primary studies showed a significant OS benefit for nSRCC patients. The other study showed an OS benefit as well for nSRCC patients but was not significant. All results are shown in Figure 2B.

All but one study used chemotherapy regimens consisting of a platinum compound and a fluoropyrimidine, sometimes combined with a taxane.

Risk of bias assessment resulted in low risk of bias in one study and moderate in two studies. The certainty of evidence was very low.

In the third analysis response rates were compared between patients treated by nCTX followed by surgery comparing SRCC patients to nSRCC patients (30–33). For this meta-analysis 1,186 patients were available, of which 458 patients (38.6%) were SRCC and 728 patients (61.4%) were nSRCC. Overall, histopathological response rates to nCTX were significantly higher in nSRCC patients (OR, 2.47; 95% CI, 1.78–3.44; p < 0.01; I²= 7%). Three of the included studies showed significantly higher histopathological response rates for nSRCC patients in comparison to SRCC patients. The fourth study yielded the same result, however, the difference in histopathological response between SRCC and nSRCC did not reach statistical significance. All results are shown in Figure 2C.

Two studies used platinum compounds and a fluoropyrimidine, two studies used a combination of platinum compounds plus a taxane and/or epirubicin for CTX treatment.

The four studies analyzing histopathological response rates used different tumor regression grading (TRG) systems. Response rate was 13.6% for SRCC vs. 38.1% for nSRCC in Xu et al., 16.6% vs. 28.9% in Heger et al., 54% vs. 74.2% in Jiang et al. and 0% vs. 35% in Jary et al.

Heger et al. and Jary et al. used the tumor response grading (TRG) system of Becker (34, 35) defining grade 1a + 1b as histopathological responders. Xu et al. used the Mandard-TRG (36) defining grade 1 + 2 with a tumor residuum of <10% as pathological responders. In contrast, Jiang et al. used the CAP-TRG (37) by the College of American Pathologist (38) and defined grade 0 + 1 + 2 as pathological responders.

Whereas histopathological response definition of Heger, Jary, and Xu et al. can be considered correspondent, Jiang applied a vaster definition of histopathological response including also residual cancer outgrown by fibrosis, which equals Mandard-TRG grade 3 + 4 (Supplementary Table S1). We therefore performed a sensitivity analysis without the study of Jiang et al. but yielded basically unchanged results (OR, 2.72; 95% CI = 1.56–4.76; p < 0.01; I²= 38%) (Supplementary Figure S1).

Risk of bias assessment resulted in low risk of bias in one study, low to moderate risk in one study and moderate risk in one study. The certainty of evidence was very low.

An overview of the rating of certainty of evidence was made with the grading of recommendations, assessment, development, and evaluation (GRADE) approach is shown in Table 3. The certainty of the evidence of the main analysis for all three outcomes was very low largely due to the study designs of the included studies (retrospective) and study biases.