Fig 4. The utility of protein combinations to distinguish GA and associated distant metastasis.

For multi-protein models, linear discriminate analysis was performed using 3 protein and 4 protein combinations. The diagnostic performance of each model was evaluated using leave one out cross validation method. The utility of serum proteins as cancer biomarkers was evaluated using the area-under-curve (AUC) of the ROC curves for different models. Several protein combinations possess excellent diagnostic value (AUC >0.9) in distinguishing both non-metastatic and metastatic samples from healthy controls.