This retrospective analysis was approved by the Ethics Committee of The First Affiliated Hospital of Ningbo University and Xuzhou Central Hospital and the requirement for informed consent was waived. A training cohort consisting of 289 consecutive patients with GC from Xuzhou Central Hospital between January 2021 and December 2023 was established. Furthermore, a testing cohort comprising 90 consecutive GC patients from The First Affiliated Hospital of Ningbo University during the same period was established. Data from the training cohort were utilized to establish the PPM, and the same data from the training cohort were used to validate the PPM.

Identical inclusion and exclusion criteria were applied to the training and validation cohorts. To be considered for inclusion, patients had to meet the following criteria: (1) they must have received a diagnosis of GC based on a preoperative gastroscopy, (2) their GC must be of the adenocarcinoma GC pathological type, (3) they must have undergone serum testing, gastroscopy, and CT scanning within 7 days before surgical resection, (4) they must have undergone gastrectomy with D2 lymphadenectomy, and (5) their LNM status must have been confirmed through pathological examination. Patients were excluded if they (1) lacked complete clinical data needed for this study, (2) exhibited any other comorbid tumors, or (3) had undergone preoperative antitumor treatment.

The data utilized for these analyses encompassed age, gender, tumor size, tumor location, depth of invasion, CT-based LN status, pathological findings, and the results of tumor marker tests.

A 64 Dual-Source CT instrument was used for all CT scanning. Patients were directed to fast for >8 hours before scanning, and received an intravenous injection of anisodamine (20 mg) to prevent gastric motility. Patients also orally consumed 1 L of warm water before scanning for stomach dilation, and were directed to hold their breath during scanning. Both non-enhanced and enhanced CT scanning were performed as detailed in the Supplementary Materials . Two radiologists with 8 and 15 years of expertise interpreting abdominal CT pictures independently assessed the resulting images, keeping the patient’s pathological diagnosis confidential. A third radiologist with 20 years of experience interpreting abdominal diagnosis resolved any disagreements. CT-reported LN positivity was determined based on a short axial diameter for the target LN exceeding 5 mm.

During the gastroscopy, information was obtained on the tumor’s location, tumor size, and presence or absence of ulcers. The depth of tumor invasion [clinical T stage (cT stage)] was determined through endoscopic ultrasonography. Every patient in the research population had GC that was verified by biopsy pathology. Moreover assessed were the tumor’s pathological type and differentiation.

Through univariate and multivariate logistic regression analyses that evaluated clinical features, CT results, gastroscopy findings, and the amounts of investigated tumor markers. LNM-related risk variables were identified in the training cohort and then the PPM was established based on the risk variables. Data from the testing cohort was then used for PPM validation.

SPSS 27.0 and R 4.1.2 were used for all analyses. Categorical data were compared using χ² tests or Fisher’s exact test. Continuous data that were normally (non-normally) distributed were compared using Student’s t-tests (Mann-Whitney U tests). LNM-related risk factors were selected through logistic regression analyses. The PPM was established based on the LNM-related risk factors and a risk scoring formula was listed. The nomogram was constructed based on the risk scoring formula by using the “rms” R package based on the training cohort. Comparisons of area under the ROC curve (AUC) values were made with the DeLong test. The inter-observer agreement analyzes were conducted with Intraclass Correlation Coefficient test for continuous data or Kappa analysis for categorical data. The Intraclass Correlation Coefficient or Kappa values were used to assess the degree of agreement (< 0.21, poor; 0.21–0.4, fair; 0.41–0.6, moderate; 0.61–0.8, good; and 0.81–1, very good).

Of the GC patients in the training cohort for this study, 67 (23.2%) and 222 (76.8%) were respectively negative and positive for LNM ( Table 1 ). In LNM positive group, patients exhibited significantly higher rates of cT3 invasion, the presence of ulcers, moderate/poor differentiation, and CT-reported LN-positive status than those in the LNM negative group. Furthermore, LNM positive group also exhibited significantly larger tumor diameter and higher CEA value than LNM negative group. No other analyzed parameters differed between patients who were and were not LNM positive in this training cohort. The inter-observer agreements of maximum tumor diameter and CT-reported LN status were very good in both training and test cohorts ( Supplementary Table 1 ).

In univariate analyses, tumor location at the fundus (P = 0.084), maximum tumor diameter (P = 0.001), cT2 invasion (P = 0.061), cT3 invasion (P = 0.001), CT-reported LN (+) (P = 0.026), the presence of ulcers (P = 0.001), CEA value (P = 0.077), and CA199 value (P = 0.021) were significantly related to LNM. The tumor diameter, CEA value, and CA199 value were positively correlated with LNM. In multivariate analyses, cT3 invasion (P = 0.001), CT-reported LN positivity (P = 0.044), and CA199 value (P = 0.030) were found to be independently associated with LNM risk. The CA199 value was positively correlated with LNM ( Table 2 ).

The risk factors identified above were used to establish a PPM and nomogram ( Figure 1 ). The risk scoring formula for the nomogram was the following: score = -2.382 + 0.694×CT-reported LN status (+: 1; -: 0)+2.497×invasion depth (cT1: 0; cT2: 1; cT3: 2)+0.032×CA199 value. The cut-off score of 3.4217 demonstrated the highest sensitivity (52.3%) and specificity (86.6%), Patients were classified as LNM positive if their score at or above this threshold value, while they were otherwise classified as LNM negative.

The respective AUC values for the PPM and CT-reported LN status of 0.753 and 0.609 differed significantly (P < 0.001) ( Figure 2A ).

In the testing cohort, 29 (32.2%) and 61 (67.8%) patients were respectively diagnosed as negative and positive for LNM ( Table 1 ). The AUC values for the PPM and CT-reported LN status differed significantly when clinical data from the testing cohort were included in the PPM (0.756 vs. 0.568, P < 0.001) ( Figure 2B ).

Calibration curves generated for the training and testing cohorts revealed good consistency between actual and expected LNM status for these GC patients ( Figures 3A, B ). In all the training and testing cohorts, decision curve analyses showed a significant net clinical benefit associated with the PPM. The risk criteria for each of these two groups were 0.2-1.0 and 0.07-1.0, respectively ( Figures 4A, B ).