We consecutively reviewed clinicopathological data of the 4,375 patients who underwent surgically resection with lymph node dissection for gastric cancer in our hospital between January 2014 to January 2022 and 655 patients diagnosed EGC by postoperative pathology were selected. Exclusion criteria were as follows: (I) Pathological types (neuroendocrine tumor, carcinoma with lymphoid stroma and hepatoid carcinoma) outside the scope of this study (6 cases were excluded). (II) Cases with multiple synchronous cancers (12 cases were excluded). (III) Preoperative endoscopic treatment (2 cases were excluded). (IV) Local recurrence (2 cases were excluded). (V) Patients who had undergone chemotherapy and/or radiation (0 cases were excluded). (VI) Cases with no preoperative endoscopic images or reports (7 cases were excluded). A total of 626 patients with 626 lesions were finally enrolled. Clinical and pathological data including sex, age, tumor location, tumor size, macroscopic type, ulceration, invasion depth, LVI, perineural invasion, histological type, scope of LND (lymph node dissection), number of LND LNM status were collected. Cut value of age was set to 60 (years) as mean age of the cohort was 63.6 (±8.8) years. For tumor size, according to the calculated mean value (2.4 ± 1.3 cm), we graded lesions into small group(≤2 cm) and large group (>2 cm). For tumor location, we classified the lesions into upper, middle or lower 1/3 of stomach (2). Classification of tumor gross morphology were based on Paris endoscopic classification (9). Invasion depth included three following grades: mucosal(M), SM1 (depth of invasion <500 μm) and SM2 (depth of invasion ≥500 μm). Judgement of LND scope involved in this study (D1, D1+, D2) was in accordance with Japanese gastric cancer treatment guidelines (2). Cut value of node dissection number was set to 22 as mean number of the cohort was 22.3 (±9.6).

All lesions were sliced at intervals of 3 to 5 mm. Lymph nodes (at least 15 nodes were harvested per case) were sliced and stained with hematoxylin and eosin to assessment the presence of LNM. Pathological parameters were evaluated according to guidelines of Japanese Gastric Cancer Association (JGCA) for EGC (2, 23). To assess proportion of undifferentiated components, in this study, the panel of three pathologists examined all slides of all the specimens. Different histological cancer areas of all slides were evaluated under microscope and finally summarized to determine the PUC level.

To further investigate differences of clinicopathological characteristics between different PUC levels, we grouped mixed histologic type lesions into five groups (M1:0% < PUC ≤ 20%, M2:20% < PUC ≤ 40%, M3:40% < PUC ≤ 60%, M4:60%<PUC ≤ 80%, M5:80% < PUC < 100%). Together with PD (PUC = 0%) and PU lesions (PUC = 100%), clinicopathological features according to PUC level were explored. According to guidelines of ESD for EGC, tumor differentiation is determined according to its quantitative dominant component (23). Thus in our study, mixed histological type lesions were regrouped into four groups: G1(0% < PUC ≤ 25%), G2 (25% < PUC < 50%), G3 (50% ≤ PUC ≤ 75%) and G4 (75% < PUC < 100%). Compared with PD, ESD indications for differentiated EGC were verified in G1 and G2. Compared with PU, ESD indications for undifferentiated EGC were verified in G3 and G4.

Statistical analyses were conducted by SPSS 26.0 software and R software version 4.2.0. Continuous variables (age and tumor size) were translated into categorical variables. Differences of clinical and pathological features between groups were analyzed by Pearson chi-square or Fisher`s exact test. Variables that were statistically associated with LNM in univariate analysis were then entered into a logistic regression model to investigate independent risk factors. Accuracy of the nomogram was validated by the area under the curve (AUC), concordance-index(C-index) was also applied to estimate nomogram performance. Under Bootstrap method, repetitive sample of the same size were constructed. The sample was used as the training set, and the corresponding unsampled queue was used as the verification set. This performance evaluation was repeated 1,000 times to obtain the calibration curve. Hosmer-lemeshow test was used to calculate the goodness of fit of the model. P < 0.05 means the difference is regarded as statistically significant.

A total of 626 patients with a total of 626 lesions were finally included in this study. Of the patients, 429(68.5%) were male and 197(31.5%) were female. Incidence of LNM in this study was 11.3% (71 of 626). Forty-eight (7.7%) lesions were in the upper third of the stomach, 156(24.9%) in the middle third, and 422(67.4%) in the lower third. Proportion of PD, mixed and PU histology type in this study was 49.8% (n = 312), 30.7% (n = 192) and 19.5% (n = 122), respectively. Together with the parameters mentioned above, clinicopathological characteristics of the whole cohort were shown in Table 1.

To further investigate differences of clinicopathological characteristics between different level of PUC, we grouped mixed histologic type lesions in to five groups: M1 (n = 26, 4.2%), M2 (n = 34, 5.4%), M3 (n = 30, 4.8%), M4 (n = 41, 6.5%), M5 (n = 61,9.7%). Clinicopathological features of the patients based on different level of PUC were shown in Table 2. Distribution of age, sex, tumor location, macroscopic type, presence of ulcer, scope of LND and number of LND in lesions with different PUC levels showed no difference (p > 0.05 after Bonferroni correction). Compared with pure undifferentiated lesions, pure differentiated lesions tend to have less presence of perineural invasion (p = 0.025 after Bonferroni correction). When compared with PD group, mixed type tumors in M4 were larger (p = 0.025), tumors in M5 were more frequently with submucosal invasion to SM2 (p < 0.05), and lesions in M4 (p < 0.001) or M5 (p = 0.035) were prone to LVI (All p value were corrected under Bonferroni method).

Comparison between each two groups showed that LNM rate was higher in group M4(M4 vs. PD: p < 0.001, M4 vs. M1: p = 0.010, M4 vs. M2: p = 0.004, M4 vs. M3: p = 0.038, M4 vs. M5: p = 0.040, M4 vs. PU: p < 0.001, all p value corrected by Bonferroni method). LNM rate of M5 was higher than PD (p = 0.009 after Bonferroni correction). While the LNM rate of M5(19.7%) was higher than PU (9.0%), there was no statistics difference between the two groups. Comparison between other each two groups also showed no significant difference (p > 0.05 after Bonferroni correction). Such results suggested that we should be more cautious about recommendations for mixed type lesions with PUC level over 60% after ESD. LNM rate according to PUC level was showed in Figure 1.

Based on the results above, to identify the clinicopathological predictive factors of LNM, univariate and multivariate logistic regression analyses were performed. The univariate analysis showed that larger tumor size (OR 4.015; 95% CI 2.245–7.179; p < 0.001), presence of ulcer (OR 2.038; 95% CI 1.212–3.424; p = 0.007), submucosa invasion to SM2 (OR 7.809; 95% CI 3.952–15.431; p < 0.001), presence of LVI (OR 18.082; 95% CI 10.201–32.052; p < 0.001), presence of perineural invasion (OR 2.973; 95% CI 1.261–6.843; p = 0.009), PUC level [(M4:OR 14.687; 95% CI 6.812–31.666), (M5:OR 3.777; 95% CI 1.725–8.267), p < 0.001] was significantly associated with LNM rate (Table 3). Based on enter method multivariate analysis, tumor size over 2 cm (OR 3.157; 95% CI 1.581–6.303; p = 0.001), submucosa invasion to SM2 (OR 2.869; 95% CI 1.262–6.523; p = 0.012), presence of LVI (OR 12.648; 95% CI 6.246–25.611; p < 0.001) and PUC level M4 (60% < PUC ≤ 80%) (OR 12.205; 95% CI 4.791–31.088; p < 0.001) significantly predicted LNM in EGC (Table 4).

Mixed type lesions were regrouped into four groups: G1 (n = 34, 5.4%), G2 (n = 39, 6.2%), G3 (n = 52, 8.3%) and G4 (n = 67, 10.7%). One case in PD (1/172, 0.6%) and one case in G2 (1/23,4.3%) who met with absolute ESD indications for differentiated EGC developed with LNM, while no statistical difference was found in cases which met the absolute indications or in cases who was out of the absolute ESD indications for differentiated EGC. No LNM occurred in patients who met the absolute ESD indications for undifferentiated EGC in group G3 (0 of 3), G4 (0 of 6) and PU (0 of 20). For cases beyond the absolute ESD indications for undifferentiated EGC, LNM rate was significant higher in G4 (G4 vs. PU: 44.9% vs. 10.8%, p < 0.001; G4 vs. G3: 44.9% vs. 18.0%, p = 0.002) (Shown in Tables 5, 6).

Based on the results of multivariate logistic analysis, a nomogram was developed to predict LNM in EGC. Showed in Figure 2, the nomogram score was 39 for submucosa infiltration depth ≥ 500, 42.5 for tumor size > 2 cm, 100 for M4 PUC level and 93 for LVI, respectively. AUC of the model was 0.899 (range 0.724–0.915) (Figure 3). Predictive performance internal validation by Bootstrap method showed a consistency index of 0.899, the internal calibration curve showed optimal agreement between actual observations and model predictions (Figure 4). And a good fitting effect was demonstrated by Hosmer - Lemeshow test (χ² = 7.187, p > 0.05).