Figure 1.
Comparison of the area under the time dependent receiver operating characteristic curve (AUC) at cut-off time year = 1, 2, and 3 among two risk predictors: ML and Lung-RADS in the NLST training (N=25,097) samples, and three risk predictors, including volume doubling time (VDT), in the PanCan validation (N=2,294) samples. The VDT cannot be reliably estimated in the NLST sample. In the NLST, the p-values of AUC difference between ML and Lung-RADS were <0.001 in all three years. In PanCan, the p-values of AUC difference between ML and Lung-RADS were 0·202, 0·048, and 0·028 for years 1, 2, and 3 respectively; the p-values of AUC difference between ML and VDT were <0.001 for all three years.