A Novel Prediction Model of Prognosis After Gastrectomy for Gastric Carcinoma: Development and Validation Using Asian Databases.

Objective: The prognoses of gastric cancer patients vary greatly among countries. Meanwhile, tumor-node-metastasis (TNM) staging system shows limited accuracy in predicting patient-specific survival for gastric cancer. The objective of this study was to create a simple, yet universally applicable survival prediction model for surgically treated gastric cancer patients.
Summary Background Data: A prediction model of 5-year overall survival for surgically treated gastric cancer patients regardless of curability was developed using a test data set of 11,851 consecutive patients.
Methods: The model's coefficients were selected based on univariate and multivariate analysis of patient, tumor, and surgical factors shown to significantly impact survival using a Cox proportional hazards model. For internal validation, discrimination was calculated with the concordance index (C-statistic) using the bootstrap method and calibration assessed. The model was externally validated using 4 data sets from 3 countries.
Results: Our model's C-statistic (0.824) showed better discrimination power than current tumor-node-metastasis staging (0.788) (P < 0.0001). Bootstrap internal validation demonstrated that coefficients remained largely unchanged between iterations, with an average C-statistic of 0.822. The model calibration was accurate in predicting 5-year survival. In the external validation, C-statistics showed good discrimination (range: 0.798-0.868) in patient data sets from 4 participating institutions in 3 different countries.
Conclusions: Utilizing clinically practical patient, tumor, and surgical information, we developed a universally applicable prediction model for accurately determining the 5-year overall survival of gastric cancer patients after gastrectomy. Our predictive model was also valid in patients who underwent noncurative resection or inadequate lymphadenectomy.