EXTERNAL VALIDATION OF THE PERI-GASTRIC MODEL FOR PREDICTING PERITONEAL FAILURE AFTER CURATIVE-AIM GASTRECTOMY
Annamaria Agnes*, Yuki Hirata, Laura R. Prakash, Paul Mansfield, Brian Badgwell, Naruhiko Ikoma
The University of Texas MD Anderson Cancer Center Division of Surgery, Houston, TX
Background: The prediction of peritoneal failure, or metachronous peritoneal recurrence, after curative-aim gastrectomy, could identify patients who might benefit from prophylactic intraperitoneal treatments. In a recent Italian study, a 5-variables model, the PERIGastric 2 (based on pT, pN, linitis plastica, remnant gastric cancer status and signet ring cell features) was developed to predict peritoneal failure. The aim of the present study was to externally validate the model to assess its generalizability in an independent population.
Methods: Patients with Stage Ib-III gastric adenocarcinoma undergoing curative-aim gastrectomy between February 1995 and December 2019 at MD Anderson Cancer Center (MDACC) (Houston, TX, USA) were included. The PERIGastric 2 model was used to predict peritoneal recurrence, and discrimination was tested with the area under the curve (AUC) of receiver operating characteristic (ROC) curves. Calibration of PERIGastric 2 model was evaluated by plotting observed against predicted risk of peritoneal recurrence. A multivariable backward logistic regression (including age, gender, type of neaodjuvant chemotherapy, type of lymphadenectomy, linitis plastica, remnant gastric cancer, pT3-4, pN2-3, presence of SRC features and adjuvant therapy) was used to select variables associated with peritoneal recurrence in the MDACC cohort.
Results: Three-hundred fifty patients were included. The PERIGastric 2 model had a fairly good discrimination to predict peritoneal recurrence in the MDACC cohort (AUC 0.74, 95% CI 0.67-0.82), but had poor calibration (Figure 1) with possible underestimation of the risk of peritoneal failure. The underestimation of peritoneal recurrence may be due to a different reported rate of peritoneal recurrence (9.9% in the original PERIGastric cohort, 15.7% in the MDACC cohort) caused by the differences in the use of neoadjuvant/adjuvant treatment in the two populations. By logistic regression model, 4 variables (linitis plastica, pT3-4, pN2-3, signet ring cell features) of the 5 variables included in the Italian model were selected as independent variables associated with peritoneal failure in the MDACC cohort.
Conclusion: The PERI-Gastric 2 kept a good discrimination in the independent patient cohort but calibration was likely affected by the difference in the reported rate of peritoneal recurrence and in patient characteristics. Further accumulation of international data of gastric cancer recurrence pattern and recalibration of the model is needed in order to improve the accuracy and generalizability of prediction model of peritoneal recurrence of gastric cancer.
Back to 2022 Abstracts