Figure 2

The ROC curve of the final model. The naïve model was fitted with Cox proportional hazard regression model using combined score (continuous) adjusted for BMI (< 25 and ≥25 kg/m²), TNM stage (I, II, and III), ER status (positive and negative) and radiation therapy (yes and no). The prediction error fitted ROC was estimated using 10-fold cross validation. Among the 303 patients, 298 subjects were used in ROC analysis since the subjects with missing values of any covariates were excluded.