Background: The sensitivity of computed tomography (CT) for the detection of peritoneal metastasis (PM) of advanced gastric cancer (AGC) is relatively low.
Purpose: To develop a predictive model to improve the sensitivity of PM detection and to externally validate this model.
Material and methods: We analyzed data from 78 patients with PM, who had undergone preoperative CT and subsequent surgery between January 2012 and December 2014, and 101 controls to form a derivation set, retrospectively. The following CT findings were evaluated: tumor size; Bormann type 4; enlarged lymph node; indirect findings of PM (peritoneal thickening, fat stranding, plaques or nodules on the peritoneum, and ascites); and definitive findings of PM (omental cake and rectal shelf). A predictive model was created using multivariate logistic regression. Receiver operating characteristic (ROC) analyses were performed to assess the diagnostic performance of the model. The accuracy was externally validated at other hospitals on 31 patients with PM and 48 patients without PM.
Results: Tumor size >5.2 cm, Bormann type 4, enlarged lymph node, peritoneal plaques or nodules, and ascites were independently associated with PM. It was able to predict PM with a higher area under the ROC curve (AUC) and sensitivity than definitive findings of PM (AUC 0.903 vs. 0.647, sensitivity 92.3% vs. 38.3%). External validation confirmed the predictive power with good inter-observer agreement.
Conclusion: The CT-driven model shows higher AUC and sensitivity for prediction of PM and may help decision-making with the aim of improving care for patients with AGC.
Keywords: Advanced gastric cancer; computed tomography; external validation; peritoneal metastasis; predictive model.