Validation of an assessment tool: Estimation of Postoperative Overall Survival for Gastric Cancer

Abstract Background Benchmarking of long-term surgical outcomes has rarely been attempted. We previously devised a prediction model for assessing the outcome of late survival after surgery, termed the Estimation of Postoperative Overall Survival for Gastric Cancer (EPOS-GC). This study was undertaken to validate EPOS-GC in an external data set. Methods A retrospective cohort study was conducted in 11 cancer care hospitals in Japan, analyzing a consecutive series of patients who underwent elective gastric cancer resection between April 2007 and March 2009. EPOS-GC consists of three tumor-related variables and three physiological variables. The primary endpoint was postoperative overall survival. The observed-to-expected (O/E) ratio of 5-year survival rates was defined as a metric of quality of care. The sample size for O/E was determined as 42. Results We included 2045 patients for analysis. The median (95% confidence interval) follow-up time was 5.1 (1.2–6.8) years for censored patients. Although EPOS-GC demonstrated a good discriminative power (Harrell's C-index, 95% confidence interval: 0.80, 0.79–0.83), the calibration plot revealed that EPOS-GC underestimated 5-year survival rates in the high-risk group. Therefore, we recalibrated the model with Cox's regression analysis. The recalibrated EPOS-GC showed a good calibration, preserving the high discriminative power (C-index, 95% confidence interval: 0.80, 0.78–0.82). The O/E among hospitals according to the recalibrated EPOS-GC ranged between 0.87 and 1.27. The O/E correlated with hospital volumes (Spearman's correlation = 0.76, n  = 11, p  = .006). Conclusion EPOS-GC with recalibration can convey risk-adjusted quality assurance regarding late survival following gastric cancer resection.