Endoscopic resection or radical gastrectomy? Risk factors for lymph node metastasis in early gastric cancer: a systematic review and meta-analysis Authors: Yike Zhang, Haozhe Wang, Tianhang Luo Published: 2026 ABSTRACT AbstractBackgroundThe management of EGC hinges on accurate LNM risk assessment, guiding the critical choice between endoscopic resection (ER) and radical gastrectomy. This study provides a comprehensive evidence synthesis to optimize treatment selection.MethodsThis systematic review and meta-analysis (PROSPERO CRD420251152013) was conducted in accordance with PRISMA guidelines. A comprehensive literature search of PubMed, Web of Science, Scopus, and Embase was performed for studies published up to September 5, 2025. Study quality and risk of bias were assessed using the ROBINS-I tool, and the certainty of evidence for each association was evaluated with the GRADE framework. Statistical heterogeneity was investigated, and the sources were explored through meta-regression where appropriate.ResultsOne hundred one studies were included. Meta-analysis identified predictors: LVI (OR = 9.04, with LVI negative as reference, 95% CI: 7.18–11.37), mucosal invasion (T1a) (OR = 0.20, with T1b as reference, 95% CI: 0.19–0.22), tumor size ≤ 2 cm (OR = 0.39, with > 2 cm as reference, 95% CI: 0.36–0.43), differentiated histology (OR = 0.45, with undifferentiated type as reference, 95% CI: 0.40–0.50), PNI (OR = 3.41, with PNI negative as reference, 95% CI: 2.46–4.71), ulceration (OR = 1.68, with ulceration negative as reference, 95% CI: 1.37–2.05), age ≤ 60 years (OR = 1.13, with > 60 years old as reference, 95% CI: 1.06–1.21), male (OR = 0.69, with females as reference, 95% CI: 0.65–0.72), middle location (OR = 0.96, with other parts as reference, 95% CI: 0.90–1.02), CEA > 0.5 ng/mL (OR = 1.35, with CEA ≤ 0.5ng/mL as reference, 95% CI: 1.04–1.75), CA199 > 37 U/mL (OR = 4.02, with CA199 ≤ 37 U/mL as reference, 95% CI: 1.46–11.11), CA125 (OR = 1.17, with CA125 ≤ 35 U/mL as reference, 95% CI: 0.48–2.81), CA724 (OR = 1.33, with CA724 ≤ 7 U/mL as reference, 95% CI: 0.84–2.11). All reported pooled ORs represent unadjusted estimates derived from univariate data.ConclusionThis meta-analysis provides a comprehensive evidence base for risk stratification in EGC. By distinguishing patients suitable for ER from those requiring radical operation, our findings serve as a critical guide for optimizing clinical practice and resolving the therapeutic dilemma between endoscopy and radical surgery.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12957-026-04455-8.Keywords: Early gastric cancer, Lymph node metastasis, Risk factor, Meta-analysis, Endoscopic resection Introduction Gastric cancer ranks as the fifth most frequently diagnosed malignancy worldwide, with more than 900,000 estimated new cases each year [1]. The definition of EGC is based on invasion depth and is independent of lymph node metastasis (LNM) status. However, the presence of LNM is a critical determinant for treatment, shifting the approach from endoscopic resection to radical surgery [2, 3]. The accurate assessment of LNM status is crucial for optimizing treatment selection in EGC, as it guides the choice between endoscopic resection and radical surgery, thereby mitigating the risks of both under-treatment and over-treatment. Existing risk stratification systems, such as e-Cura scoring system and guidelines from the European Society of Gastrointestinal Endoscopy (ESGE), incorporate variables including depth of invasion, ulceration, histologic type, tumor size, and lymphovascular invasion (LVI) to evaluate the risk of LNM [4, 5]. Furthermore, emerging evidence suggests that additional factors such as perineural invasion (PNI) and tumor location may also correlate with LNM [6, 7]. Studies on the risk factors of LNM in EGC often yield inconsistent results due to small sample sizes and limited statistical power. Systematic reviews and meta-analyses address this issue by integrating data from multiple studies, thereby providing more precise and reliable estimates. While several meta-analyses have been conducted [8, 9], certain limitations persist about the scope and timeliness of included studies, the comprehensiveness of predictive factors considered. Therefore, we conducted this comprehensive systematic review and Meta-analysis to analyze the strength of association between various clinicopathological factors and serum tumor markers with LNM risk in EGC patients. Method Search strategy The present study strictly complied with the relevant requirements of the PRISMA guideline and completed the PRISMA 2020 checklist [10]. The study protocol was prospectively registered in the PROSPERO database (registration number: CRD420251152013, https://www.crd.york.ac.uk/PROSPERO/view/CRD420251152013). A systematic literature research was conducted in Pubmed, Scopus, Web of Science, and Embase databases for studies published before September 5 2025 that reported risk factors of LNM in EGC, using the following searching terms: early gastric cancer, EGC, lymph node metastasis, lymph node involvement, LNM, risk factor, predictor, clinicopathologic, correlate. The detailed search strategy is available in Supplementary Material (Search strategy). Eligible studies were restricted to the English language but had no publication date limit. Furthermore, reference lists of included articles were manually searched to avoid omission of any relevant literature. Inclusion and exclusion criteria Included studies must meet the following criteria: (1) Enrolled a patient cohort with a definitive diagnosis of EGC; (2) A study that compared the relationship between LNM and potential variables in patients with EGC; (3) Studies must report univariate logistic regression results for associations between candidate risk factors and LNM. Studies were excluded if they included any of the following criteria: (1) Letters, reviews, conference reports, comments, case reports, and animal experimental studies; (2) Studies not published in English; (3) Patient received neoadjuvant therapy (chemotherapy, radiotherapy, or immunotherapy) before surgery; (4) Enrolled patients were diagnosed with advanced gastric cancer; (5) Enrollment of only a single pathological subtype of EGC; (6) Studies that exclusively enrolled patients from a specific age group; (7) Studies only reported OR values without available raw data. Literature screening was independently conducted by two reviewers (Y.Z. and H.W.) using software Rayyan (https://www.rayyan.ai/). An initial screening was performed based on titles and abstracts, after which the full texts of potentially eligible records were retrieved and reviewed. Any disagreements arising during the screening process were resolved through discussion and consultation between the two reviewers. If a consensus could not be reached, the matter was referred to a third senior researcher (T.L.) for arbitration. Data extraction and quality assessment Data extraction and quality assessment were conducted independently by two reviewers, with disagreements resolved by consensus or by a third reviewer. The following data were extracted from each study: author, publication year, country, study size, age (≤ 60 or > 60 years old), sex (male or female), LNM status (positive or negative), ulceration status, tumor size (≤ 2 or > 2 cm), invasion depth (mucosal invasion T1a or submucosal invasion T1b), histology type (differentiated or undifferentiated), tumor location (middle location or other parts), PNI status (positive or negative), CEA (≤ 0.5ng/mL or > 0.5 ng/mL), CA199 (≤ 37 U/mL or > 37 U/mL), CA125 (≤ 35 U/mL or > 35 U/mL), CA724 (≤ 7 U/mL or > 7 U/mL). To enable a consistent synthesis, all extracted variables were harmonized into predefined binary categories. The harmonization approach and the choice of cut-points were based on common clinical practice, and the most frequently reported formats in the included literature to maximize data inclusion and relevance. Age: ≤ 60 years (exposed) vs. > 60 years (reference). The age of 60 is a commonly used demographic threshold in oncology studies. Tumor Size: ≤ 2 cm (exposed) vs. > 2 cm (reference). This aligns with the classic criteria for endoscopic resection in many guidelines. Depth of Invasion: mucosal (T1a) invasion (exposed) vs. submucosal (T1b) invasion (reference). This dichotomy is the most consistently reported and carries fundamental prognostic significance. Finer stratification (e.g., SM1, SM2) was not consistently reported across studies. Therefore, our meta-analysis synthesized data at the level of T1a versus T1b comparison. Studies were included in this analysis only if they reported histology using the differentiated vs. undifferentiated classification. This criterion was chosen because the differentiated/undifferentiated dichotomy is directly relevant to assessing eligibility for endoscopic resection in current clinical practice. Studies that reported histology solely using the Lauren classification (intestinal, diffuse, mixed) without providing corresponding differentiation status were not included in the synthesis for this variable, as a reliable and consistent cross-mapping could not be assured based on the reported data alone. Ulceration, LVI and PNI: Presence (exposed) vs. absence (reference). Serum tumor markers (CEA, CA19-9, etc.): Elevated level (exposed) vs. normal level (reference), using the standard laboratory cut-offs (e.g., CEA > 0.5 ng/mL, CA19-9 > 37 U/mL) as reported in the majority of studies. Minor variations in cut-offs between studies were addressed by adopting the most prevalent standard value. All included studies were non-randomized observational studies. The risk of bias for each study was assessed using the ROBINS-I (Risk of Bias in Non-randomized Studies - of Interventions) tool. This tool evaluates studies across seven core domains and assigns an overall risk-of-bias judgment for each study as “Low risk,” “Moderate risk,” “Serious risk,” or “Critical risk.“. The certainty of evidence for each association was assessed using the Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) framework. The evidence from observational studies was initially rated as “low” certainty and was subsequently rated down or up across five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. The final certainty for each predictor was categorized as high, moderate, low, or very low. These assessment were performed independently by two reviewers (Y.Z. and H.W.), and any discrepancies were resolved through discussion or by consulting a third senior researcher (T.L.) Statistical analysis The pooled proportion of LNM rate in EGC and its 95% confidence interval (95% CI) was calculated by the Freeman–Tukey double arcsine method and the inverse variance method, as this method provides more robust estimates when dealing with potential extreme proportions and the substantial between-study heterogeneity observed in our data. The pooled results of the meta-analysis were expressed with odds ratios (ORs) and 95% CI, which were derived from the unadjusted estimates of univariate analyses in the included studies. The I2 statistics was utilized to evaluate the heterogeneity. I2 < 25%, 25% ≤ I2 < 50%, and I2 ≥ 50% were regarded as low, moderate and high heterogeneity. A random-effect model was used if the test of heterogeneity was high (I2 > 50% or P < 0.05). Otherwise, a fixed-effect model was adopted. Visual inspection of funnel plot symmetry was performed for the overall pooled estimate. For each clinicopathological feature included in the meta-analysis, potential publication bias was assessed using Egger’s linear regression test [11]. This test evaluates funnel plot asymmetry by performing a linear regression of the effect size against its precision, with the standard error as the predictor and the inverse variance as the weight. A statistically significant publication bias was suggested if the p-value for the intercept of the regression was < 0.05. To explore potential sources of the observed between-study heterogeneity (I2), multivariable random-effects meta-regression analyses were performed for the primary predictors. The covariates included in the models were: study region (East Asian vs. non-East Asian), year of publication (continuous), risk of bias (categorical variable based on the ROBINS-I tool), surgical type (salvage surgery after endoscopic resection vs. primary radical surgery), and sample size (continuous). The proportion of heterogeneity jointly explained by these covariates was quantified by calculating the coefficient of determination (R2). This analysis aimed to evaluate whether the above study-level characteristics could explain the variation in effect estimates across studies. To test the robustness of the meta-analysis, sensitivity analysis was conducted by leave-one-out method. P < 0.05 and 95% CI that did not encompass 1 was considered to be statistically significant. All of the statistical analysis was performed by R (version 4.4.2). Results Selected studies After searching the Pubmed, Scopus, Web of Science and Embase databases, 8491 potential articles were initially included in the analysis. After removing 4565 duplicates, 3462 articles were removed by reviewing titles and abstracts because they were conference summaries, case reports, reviews, animal experimental studies, comments, letters, and meta-analyses. 31 articles were removed because they were not published in English and 305 articles were removed due to advanced gastric cancer or patients receiving neoadjuvant therapy by abstract scanning. 27 articles were removed after reviewing the full text with detailed exclusion reason illustrated in Supplementary Table 2.4. Finally, a total of 101 retrospective studies were included in the meta-analysis according to the inclusion and exclusion criteria, as shown in Fig. 1. Flow chart of literature search strategies Study characteristics and quality 101 studies, involving a total of 59,220 patients, of which 8736 patients were LNM positive, were included. The basic characteristics of the included studies are shown in Table 1. 61 of the included studies were from China, 18 from Korea, 10 from Japan, 1 from Turkey, 1 from Canada, 3 from Brazil, 2 from the US, 2 from Germany, 1 from Italy, 1 from Peru, and 1 from Lithuania. Among the 101 included studies, 9 studies conducted endoscopic resection before radical operation, while others only performed radical operation. The ROBINS-I tool was used to assess the risk of bias for all 101 included studies. The results showed that 73 studies were judged to be at moderate risk of bias, 26 at serious risk, and 2 at critical risk (Fig. 2). No study was judged to be at low risk, primarily due to the potential confounding bias inherent in the retrospective design of all studies. The inter-reviewer agreement in the initial assessment was good, and all discrepancies were resolved through discussion. Detailed risk-of-bias assessments are provided in the Supplementary Material Table 2.1. Table 1Main characteristics of studies included in the meta-analysisAuthorYearCountryTime rangeCasesLNM+ casesRobins-IStudy typeSurgery typeSeo JW et al. [12]2025Korea2007–20192426550ModerateRetrospective studyRadical operationBozkurt HA et al. [13]2025Turkey2011–20224110ModerateRetrospective studyRadical operationHe Y et al. [14]2025China2009–202422821ModerateRetrospective studyRadical operationLong X et al. [15]2025China2019–202429743SeriousRetrospective studyRadical operationSong K et al. [16]2025China2014–20231423233ModerateRetrospective studyRadical operationWang X et al. [17]2024China2016–201821518ModerateRetrospective studyRadical operationLiu S et al. [18]2024China2018–202228641CriticalRetrospective studyRadical operationChiarello MM et al. [19]2024ItalyNA12976SeriousRetrospective studyRadical operationYu Z et al. [20]2024China2010–20231438189ModerateRetrospective studyRadical operationSon MY et al. [21]2024Korea2014–202213010ModerateRetrospective studyRadical operationMei Y et al. [22]2024China2019–20222722388ModerateRetrospective studyRadical operationHe JY et al. [23]2024China2010–2019750148ModerateRetrospective studyRadical operationLiu DY et al. [24]2024China2015–20231011239ModerateRetrospective studyRadical operationYu Z et al. [25]2023China2010–2020824140ModerateRetrospective studyRadical operationSun F et al. [26]2023China2013–202213314ModerateRetrospective studyER-Radical operationZhang P et al. [27]2023China2014–202134669ModerateRetrospective studyRadical operationLi C et al. [28]2023China2014–202262671ModerateRetrospective studyRadical operationParedes O et al. [6]2023Peru1990–201950196ModerateRetrospective studyRadical operationWatanabe A et al. [29]2022Canada2000–20178626ModerateRetrospective studyRadical operationSun F et al. [30]2022China2010–2019730144ModerateRetrospective studyRadical operationFan X et al. [31]2023China2010–201940242SeriousRetrospective studyRadical operationYou H et al. [32]2023China2014–202218340ModerateRetrospective studyRadical operationYang JJ et al. [33]2023China2012–20221922278ModerateRetrospective studyRadical operationMakimoto S et al. [34]2022Japan2012–202010014SeriousRetrospective studyER-Radical operationDuan K et al. [35]2022China2016–2019305ModerateRetrospective studyRadical operationFeng M et al. [36]2022China2010–202056185ModerateRetrospective studyRadical operationZheng C et al. [37]2022China2005–201633562ModerateRetrospective studyRadical operationLiu Z et al. [38]2022China2017–202129233ModerateRetrospective studyRadical operationLee JH et al. [39]2022Korea2000–20174264450SeriousRetrospective studyRadical operationCai F et al. [40]2022China2016–201835137ModerateRetrospective studyRadical operationYang Q et al. [41]2021China2002–201726549ModerateRetrospective studyRadical operationRen MH et al. [42]2021China2007–2018691114ModerateRetrospective studyRadical operationWu Y et al. [43]2021China2010–201835465ModerateRetrospective studyRadical operationZhou ZY et al. [44]2021China2008–201830240ModerateRetrospective studyRadical operationOh YJ et al. [45]2021Korea2012–2019164891SeriousRetrospective studyRadical operationJin X et al. [46]2021China2016–2019605109ModerateRetrospective studyRadical operationWang J et al. [47]2021China2010–2016611122ModerateRetrospective studyRadical operationShin SY et al. [48]2020Korea2001–20125114ModerateRetrospective studyRadical operationWang Z et al. [49]2020China2012–201936361SeriousRetrospective studyRadical operationZhang M et al. [50]2021China2010–201528559ModerateRetrospective studyRadical operationMilhomem LM et al. [51]2020Brazil1988–201517824ModerateRetrospective studyRadical operationZhu ZL et al. [52]2020China2012–2016517114ModerateRetrospective studyRadical operationLiang XQ et al. [53]2020China2012–201720340ModerateRetrospective studyRadical operationZhao B et al. [54]2020China2002–201430259SeriousRetrospective studyRadical operationJiao J et al. [55]2020China2012–201525338ModerateRetrospective studyRadical operationLi L et al. [56]2020China2014–201738152ModerateRetrospective studyRadical operationChen J et al. [57]2020China2012–20171033173ModerateRetrospective studyRadical operationYin XY et al. [58]2020China2015–201959698ModerateRetrospective studyRadical operationDu M et al. [59]2019China2013–2014621172ModerateRetrospective studyRadical operationMu J et al. [60]2019China2010–2018746144ModerateRetrospective studyRadical operationChu YN et al. [61]2019China2007–20181262182ModerateRetrospective studyRadical operationTian YT et al. [62]2019China2013–2019455ModerateRetrospective studyER-Radical operationZhang Y et al. [63]2018China2010–201527237SeriousRetrospective studyRadical operationLi X et al. [64]2018China2009–20141004123ModerateRetrospective studyRadical operationDu MZ et al. [65]2018China2010–2016734104ModerateRetrospective studyRadical operationPereira MA et al. [66]2017Brazil2009–201510514ModerateRetrospective studyRadical operationGu L et al. [67]2018China2000–20141029228ModerateRetrospective studyRadical operationTran TB et al. [68]2018US2000–201217638ModerateRetrospective studyRadical operationLi H et al. [69]2018China1990–201513821SeriousRetrospective studyRadical operationJung DH et al. [70]2017Korea2007–201532123ModerateRetrospective studyER-Radical operationKikuchi S et al. [71]2017China2004–2013738SeriousRetrospective studyER-Radical operationBausys R et al. [72]2017Lithuania2005–201521843SeriousRetrospective studyRadical operationSuzuki H et al. [73]2017Japan1999–201033818SeriousRetrospective studyER-Radical operationSun B et al. [74]2017China2010–201642655SeriousRetrospective studyRadical operationChen L et al. [75]2017ChinaNA1620251ModerateRetrospective studyRadical operationGuo CG et al. [76]2017China2002–2014952175ModerateRetrospective studyRadical operationLee HW et al. [77]2017Korea2011–201431030SeriousRetrospective studyRadical operationKim JY et al. [78]2017Korea2006–201110259ModerateRetrospective studyRadical operationPark JW et al. [79]2017Korea2000–20111237SeriousRetrospective studyER-Radical operationKawata N et al. [80]2017Japan2002–201232330ModerateRetrospective studyER-Radical operationSunnagawa H et al. [81]2017Japan2005–201520015SeriousRetrospective studyER-Radical operationZheng Z et al. [82]2016China1996–201259758ModerateRetrospective studyRadical operationFang C et al. [83]2016China2005–201237949SeriousRetrospective studyRadical operationSekiguchi M et al. [84]2016Japan1997–20133131386ModerateRetrospective studyRadical operationAhamd R et al. [85]2015US1995–20116715SeriousRetrospective studyRadical operationHaist T et al. [86]2016Germany2000–201312417SeriousRetrospective studyRadical operationGuo CG et al. [87]2016China2002–201525633ModerateRetrospective studyRadical operationRonellenfitsch U et al. [88]2016Germany1972–200527536ModerateRetrospective studyRadical operationZhao LY et al. [89]2016China2000–2015598151ModerateRetrospective studyRadical operationPyo JH et al. [90]2016Korea2002–2015116961ModerateRetrospective studyRadical operationFeng H et al. [91]2015China2009–2014976120ModerateRetrospective studyRadical operationLi H et al. [92]2016China1987–2007817SeriousRetrospective studyRadical operationLee JH et al. [93]2014Korea2001–201384714ModerateRetrospective studyRadical operationWang Y [94]2015China2008–201519828ModerateRetrospective studyRadical operationFang WL et al. [95]2015China1987–200639140ModerateRetrospective studyRadical operationJin EH et al. [96]2015Korea2003–20111104104ModerateRetrospective studyRadical operationRen G et al. [97]2013China2003–200920229ModerateRetrospective studyRadical operationMilhomem LM et al. [98]2012Brazil1998–201012621SeriousRetrospective studyRadical operationLee JH et al. [99]2012Korea2002–20091005162ModerateRetrospective studyRadical operationKim H et al. [100]2011Korea2000–20107910ModerateRetrospective studyRadical operationLee K et al. [101]2010Korea2003–200614153ModerateRetrospective studyRadical operationSung CM et al. [102]2010China1994–200755676ModerateRetrospective studyRadical operationShen L et al. [103]2009China1998–200741050SeriousRetrospective studyRadical operationKunisaki C et al. [104]2009Japan1985–200057374SeriousRetrospective studyRadical operationPark YD et al. [105]2008Korea1999–200523429ModerateRetrospective studyRadical operationAn JY et al. [106]2007Korea2002–20051043202SeriousRetrospective studyRadical operationXu YY et al. [107]2007China1972–200632247SeriousRetrospective studyRadical operationNasu J et al. [108]2006Japan1989–200533245ModerateRetrospective studyRadical operationNakahara K et al. [109]2004Japan1994–200142230ModerateRetrospective studyRadical operationKim DY et al. [110]2004Korea1985–199974875CriticalRetrospective studyRadical operationAbe N et al. [111]2002Japan1987–200027638ModerateRetrospective studyRadical operation Main characteristics of studies included in the meta-analysis Assessment result of included studies by ROBINS-I tool Results of the meta-analysis Pooled data from all 101 studies that involved 59,220 patients with EGC identified 8736 patients with LNM. The pooled prevalence of LNM in EGC was 15% (95% CI 14%–16%, I2 = 93.7%), as shown in Supplementary Fig. 3.1.1. Gender 94 studies included were involved in the meta-analysis of gender (36686 males and 18826 females). The overall results showed that the male patients had a lower risk of LNM than the female patients (OR = 0.69, with females as reference, 95% CI: 0.65–0.72, I2 = 29.0%), as shown in Fig. 3 & Supplementary Fig. 3.1.2. Summary of meta-analysis results of LNM and potential risk factors Age Data regarding age (≤ 60 or > 60 years old) were reported in 41 studies. The meta-analysis of age included 26,755 ECG patients, of whom 13,295 were > 60 years old. The results reflected that younger patients (≤ 60 years old) were more likely to have LNM (OR = 1.13, with > 60 years old as reference, 95% CI: 1.06–1.21, I2 = 45.3%), as shown in Fig. 3 & Supplementary Fig. 3.1.3. Tumor size Seventy-one studies with 47,064 patients (22117 patients ≤ 2 cm, 24947 patients > 2 cm) met the criteria of the meta-analysis of tumor size. The results showed that the small tumor size (≤ 2 cm) represents a lower risk of LNM (OR = 0.39, with > 2 cm as reference, 95% CI: 0.36–0.43, I2 = 55.6%), as shown in Fig. 3 & Supplementary Fig. 3.1.4. LVI Sixty-six studies including 45,787 patients reported patients’ LVI status (7172 patients LVI positive, 38615 patients LVI negative). The analysis demonstrated that LVI is a strong risk factor for LNM (OR = 9.04, with LVI negative as reference, 95% CI: 7.18–11.37, I2 = 91.8%), as shown in Fig. 3 & Supplementary Fig. 3.1.5. Ulceration Sixty-three studies including 39,189 patients reported ulceration status (15288 patients had ulcer, 23901 patients didn’t have ulcer). The pooled results indicated that ulcers were significantly related to LNM (OR = 1.68, with ulceration negative as reference, 95% CI: 1.37–2.05, I2 = 80.4%), as shown in Fig. 3 &Supplementary Fig. 3.1.6. Depth of invasion Eighty-three studies with 49,134 patients (24217 patients with mucosal invasion T1a, 24917 patients with submucosal invasion T1b) were included in the meta-analysis of the depth of invasion. The pooled results revealed that patients with mucosal invasion had a lower risk of LNM than patients with submucosal invasion (OR = 0.20, with T1b as reference, 95% CI: 0.19–0.22, I2 = 38.5%), as shown in Fig. 3 & Supplementary Fig. 3.1.7. Histologic type Seventy studies with 39,377 patients (20790 patients with differentiated lesions, 18587 patients with undifferentiated lesions) were included in the meta-analysis of the histologic type. Differentiated lesions were identified as a protective factor against LNM (OR = 0.45, with undifferentiated type as reference, 95% CI: 0.40–0.50, I2 = 66.2%), as shown in Fig. 3 & Supplementary Fig. 3.1.8. Tumor location Tumor location data were reported in 69 studies, which included a total of 49,606 patients (15008 middle part, 34598 other parts). The meta-analysis revealed no statistically significant association between tumor location and LNM (OR = 0.96, with other parts as reference, 95% CI: 0.90–1.02; I² = 28.8%), as shown in Fig. 3 & Supplementary Fig. 3.1.9. PNI Thirty-seven studies with a total of 24,685 patients (1857 PNI positive, 22828 PNI negative) were included for the meta-analysis of PNI. The pooled results demonstrated that PNI was a significant risk factor for LNM (OR = 3.41, with PNI negative as reference, 95% CI: 2.46–4.71, I2 = 74.0%), as shown in Fig. 3 & Supplementary Fig. 3.1.10. Serum tumor markers Nine studies with a total of 3979 patients (1239 > 0.5 ng/mL, 2740 ≤ 0.5ng/mL) were included for the meta-analysis of CEA. The pooled results demonstrated that higher CEA has a significant correlation with LNM (OR = 1.35, with CEA ≤ 0.5ng/mL as reference, 95% CI: 1.04–1.75; I2 = 16.3%). CA199 data were reported in 4 studies with 1821 patients (75 patients > 37 U/mL, 1746 patients ≤ 37 U/mL). The meta-analysis revealed a significant correlation between higher CA199 level and LNM (OR = 4.02, with CA199 ≤ 37 U/mL as reference, 95% CI: 1.46–11.11; I2 = 76.8%). The result of CA125 based on 3 studies with 1225 patients (30 patients > 35 U/mL, 1195 patients ≤ 35 U/mL) revealed no statistically significant association between CA125 and LNM (OR = 1.17, with CA125 ≤ 35 U/mL as reference, 95% CI: 0.48–2.81; I2 = 0.0%). The result of CA724 based upon 3 studies with 1218 patients (109 patients > 7 U/mL, 1109 patients ≤ 7 U/mL) showed no statistically significant association between CA724 and LNM (OR = 1.33, with CA724 ≤ 7 U/mL as reference, 95% CI: 0.84–2.11; I2 = 0.0%). Detailed data were shown in Fig. 3 & Supplementary Fig. 3.1.11. Sensitivity and meta-regression analyses The leave-one-out sensitivity analysis confirmed the robustness of our primary findings. For the vast majority of factors examined except CEA, the point estimates and their statistical significance were not materially altered by the removal of any single study, underscoring the stability of our conclusions. Forest plots were shown in the Supplementary Fig. 3.2. To investigate potential sources of inter-study heterogeneity, we conducted a multivariate random-effects meta-regression analysis on risk factors exhibiting higher heterogeneity (I2 > 50%). Included covariates comprised region, publication year, risk of bias, surgical procedure type, and sample size. Analysis of the overall LNM rate revealed that region (East Asia vs. non–East Asia), year of publication, and type of surgery (surgery after endoscopic resection versus radical surgery alone) were independent significant predictors of the effect size. These covariates collectively accounted for approximately 7.93% of the heterogeneity. Meta-regression analysis for tumor size found that the risk of bias assessed by ROBINS-I tool in the studies was the only significant predictor: compared with studies at low risk of bias, those at moderate and serious risk of bias reported significantly higher effect sizes, explaining approximately 12.6% of the heterogeneity. Region, year of publication, type of surgery, and sample size did not show significant moderating effects. Analysis of histological differentiation type revealed that the year of publication was a significant moderator, suggesting a trend toward a weaker effect size for differentiated histology as a protective factor in more recently published studies. For LVI, ulceration, and PNI, the examined covariates did not show significant moderating effects and did not explain the high degree of heterogeneity between studies. Detailed meta-regression results are summarized in Supplementary Material Table 2.2. Results of GRADE assessment The certainty of evidence for all included predictors of lymph node metastasis in EGC was assessed using the GRADE framework. The evaluation revealed that the majority of predictors were rated as having “low” certainty evidence. Among them, submucosal invasion was the only predictor that achieved a “moderate” certainty rating. The evidence for several factors, including tumor size ≤ 2 cm, differentiation, LVI, ulceration, PNI, and CA199, was downgraded due to serious inconsistency. Furthermore, the certainty for the tumor markers CA125 and CA724 was assessed as “very low”, primarily due to very serious imprecision, as indicated by wide confidence intervals that crossed the line of no effect. The complete GRADE profile for each predictor, detailing the ratings across all domains and the final certainty, is provided in the supplementary material. Risk of bias The funnel plots appeared symmetrical suggesting no significant publication bias. Funnel plots are shown in the Supplementary Fig. 3.3. To assess potential publication bias, Egger’s linear regression test was performed for all included predictors. The results of Egger’s linear regression test did not reveal statistically significant publication bias for any of the included predictors. However, borderline signals were noted for depth of invasion (P = 0.0662) and histologic type (P = 0.0513), suggesting potential trends of asymmetry that warrant cautious interpretation. Detailed test statistics are provided in Supplementary Material Table 2.3. Discussion The incidence of LNM in EGC is about 15%, and the prognosis of EGC patients with LNM is worse than those without LNM [112]. This systematic review and meta-analysis of 101 studies involving 59,220 patients comprehensively evaluated clinicopathological predictors for LNM in EGC to guide treatment decisions between ER and radical surgery. LVI emerged as the most powerful predictor of LNM (OR = 9.04, with LVI negative as reference), a finding consistent with previous meta-analyses [8, 9]. It is important to note that LVI is primarily a histopathological criterion diagnosed post-resection, rather than a reliably assessed preoperative indicator. Consequently, when LVI is identified in the post-endoscopic resection specimen, it represents the most significant high-risk factor for lymph node metastasis. Therefore, in line with current guidelines [113], the presence of LVI typically necessitates a strong recommendation for additional radical gastrectomy with lymph node dissection to achieve curative intent. The biological plausibility of this association is strong, as LVI represents a direct pathway for metastatic spread. Similar to LVI, PNI is also a postoperative pathological assessment indicator. When identified, it suggests that the tumor exhibits higher aggressiveness (OR = 3.41) and warrants serious consideration for additional radical surgery. The depth of invasion was a critical determinant. The risk of LNM increased from mucosal invasion (T1a) to submucosal invasion (T1b) (OR = 0.20, with T1b as reference), underscoring the importance of accurate endoscopic staging. This aligns with existing guidelines that consider submucosal invasion an indication for surgical resection [114]. Smaller tumor size was also a factor extensively reported in previous studies as a significant predictor [4, 115, 116]. Ulceration (OR = 1.68, with ulceration negative as reference) was also a significant predictor, though with lower effect sizes than LVI. Its inclusion in pre-operative risk assessment models could enhance predictive accuracy [4, 117]. Interestingly, this analysis revealed that female sex and younger age (≤ 60 years) were associated with a slightly elevated risk of LNM. Although these factors have been reported by previous studies [118–121], clinical decision-making does not currently take these factors into account. While the mechanisms remain unclear, this may relate to hormonal factors or more aggressive tumor biology in younger patients, warranting further investigation [122]. As the first meta-analysis to evaluate serum tumor markers, this study provides preliminary, exploratory evidence suggesting a potential association of elevated CEA and CA199 with LNM in EGC. However, given the limited number of studies, these findings must be interpreted with caution and require validation in larger, prospective cohorts before clinical application can be considered. While our findings align with the e-Cura and ESGE frameworks, this study expands the prognostic landscape by identifying additional risk factors (such as gender, age, serum markers) that are currently underrepresented in standard guidelines. When discussing the clinical implications of the findings from this study, it is essential to clearly distinguish the timing of assessment for different risk factors and their corresponding decision-making contexts [113]. The predictors of lymph node metastasis confirmed in this study can be divided into two main categories: preoperatively assessable factors and postoperative pathological factors. Factors such as age, sex, tumor location, and serum tumor marker levels are primarily obtained from preoperative clinical, endoscopic, and laboratory evaluations, and are used for initial patient risk stratification and initial treatment decisions (e.g., assessing the suitability of endoscopic resection). However, factors such as precise depth of tumor invasion (T1a vs. T1b), LVI, PNI, histological differentiation type, and pathologically defined ulceration are definitively diagnosed through microscopic evaluation of the completely resected specimen. These are core postoperative pathological indicators and provide the basis for considering further treatments, such as additional radical surgery. Due to the retrospective nature of all studies, the findings may be influenced by selection bias and unmeasured confounding. This study has certain limitations in the pooled analysis of histological variables. To ensure consistency in classification criteria, we only included studies that directly reported “differentiated/undifferentiated” types. This approach may introduce selection bias by excluding studies that used only the Lauren classification. However, this conservative strategy was necessary because the two systems are not entirely equivalent, and forced mapping could lead to misclassification. Furthermore, the pooled estimates were affected by significant heterogeneity across studies, and the potential for publication bias remains. Additionally, Egger’s test has limited power with few studies and is sensitive to high heterogeneity. Although our meta-analysis included 101 studies, providing sufficient power for Egger’s test, the borderline P-values for depth of invasion and histologic type may still reflect residual asymmetry, particularly given the high heterogeneity across studies. These findings should be interpreted with caution. Conclusion This meta-analysis of 101 studies confirms established high-risk factors for LNM in EGC, most notably LVI and submucosal invasion (T1b). It further identifies younger age, female sex, and elevated CEA/CA199 as associated factors. The clinical utility of these predictors is stage-dependent: preoperative factors aid initial ER candidacy assessment, while postoperative pathological findings (e.g., LVI, PNI) are decisive for recommending complementary radical surgery. This synthesis provides a refined, evidence-based framework to guide individualized treatment selection between endoscopic and surgical resection, aiming to optimize outcomes. Supplementary Information Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Acknowledgments Not applicable. Authors’ contributions Yike Zhang and Haozhe Wang collected the data, conducted the analyses and prepared the manuscript. Tianhang Luo designed the study and review the manuscript. Funding This work was supported by grants from the Multi-center clinical research project of Changhai Hospital (2024LYB02). Data availability All data generated or analysed during this study are included in this published article and its supplementary information files. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Yike Zhang and Haozhe Wang contributed equally to this work and share first authorship. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Data Availability Statement All data generated or analysed during this study are included in this published article and its supplementary information files.