OBJECTIVETo explore the risk factors and establish an effective model to predict lymph node metastasis (LNM) for remnant gastric cancer (RGC).

METHODSClinicopathological characteristics of 91 RGC patients undergoing radical gastrectomy at Sun Yat-sen University Cancer Center from January 2000 to December 2017 were retrospectively analyzed. RGC was defined as cancer detected in the remnant stomach >5 years for primary benign diseases or >10 years for malignant diseases following distal gastrectomy. Risk factors of LNM in RGC were identified using logistic regression (P<0.1). Covariates were then scored according to the β regression coefficient in the multivariate analysis, and a score model was established, in which higher score indicated higher risk of LNM. Finally, receiver operating characteristic(ROC) curve was used to evaluate the diagnostic efficacy of risk factors and the score model.

RESULTSAmong the 91 RGC patients, 84 were male and 7 were female with the age ranging from 47 to 82 years (63.7±8.5) years. Mean harvested lymph node (LN) was 16.0±11.8, including ≥15 LNs in 42(46.2%) patients and <15 LNs in 49(53.8%) patients. Forty-six (50.5%) patients were identified as LNM. Univariate analysis showed that tumor size ≥4 cm (χ=8.106, P=0.004), Borrmann III(-IIII( gross type (χ=6.129, P=0.013), increased CEA level (χ=4.041, P=0.044) and T3-4 stage (χ=17.321, P=0.000) were associated with LNM in RGC. In Logistic multivariate analysis, tumor size ≥4 cm (OR: 2.362, 95%CI: 0.829-6.730, P=0.100, β regression coefficient: 0.859) and T3-4 stage (OR: 7.914, 95%CI: 1.956-32.017, P=0.004, β regression coefficient: 2.069) remained as the independent risk factors for LNM, and were scored as 1 and 2 point in the final prediction model. In the final score model, LNM of patients with 0, 1, 2, 3 point were 11.1%(2/18), 1/5, 51.6%(16/31) and 73.0%(27/37), respectively. The AUC of the prediction model was 0.756 (P=0.000).

CONCLUSIONSIncreased CEA level, tumor size ≥4 cm, Borrmann III(-IIII( gross type, and deeper T stage are associated with LNM in RGC. Moreover, the score model combining with tumor size and T stage can effectively predict the LNM in RGC.