Figure 3.

Receiver Operating Characteristic (ROC) curves showing the ability of three logistic regression models to classify death within three years of follow-up for head and neck cancer patients with a prior smoking history. The first model (black line) uses only HPV status. The second model (red line) uses only the pretreatment plasma metabolomic subtype biomarker. The third model (green line) adds the metabolomic subtype to a model of HPV status. Comparing the area under the curve (AUC) of the first and third models indicates a 20% relative increase in predictive ability when the metabolomic subtype biomarker is added to a model of HPV status (0.73/0.61 = 1.197).