Skip to main content
. 2016 Jul 7;11(7):e0158887. doi: 10.1371/journal.pone.0158887

Table 4. Survival prediction: summary of class performance and hazard ratios for associations between imaging features and overall survivala.

Variable ROC analysis Cox regression model
Cutoff AUC Sensitivity Specificity PPV NPV Hazard ratio (95% CI) p-value
Univariate Analysis
fCER (%) 9.860 0.893 (0.723–1.063) 1.000 0.714 1.000 0.800 1.193 (0.941–1.513) 0.145
D*CER (x10-3mm2/s) 21.712 0.857 (0.648–1.067) 1.000 0.714 0.800 1.000 1.000 (1.000–1.000) 0.068
rCBFCER (ml/100g/min) 59.010 0.821 (0.593–1.050) 0.625 1.000 1.000 0.700 1.025 (0.9934–1.057) 0.123
rCBVCER (ml/100g) 4.780 0.821 (0.599–1.044) 0.750 0.857 0.857 0.750 1.158 (0.7698–1.742) 0.481
max rCBFCER (ml/100g/min) 155.25 0.786 (0.533–1.000) 0.750 0.857 0.857 0.750 1.032 (0.995–1.069) 0.089
max rCBVCER (ml/100g) 10.765 0.821 (0.598–1.000) 0.875 0.714 0.750 0.714 1.044 (0.864–1.261) 0.658
Treatment 1.500 0.723 (0.490–0.956) 0.857 0.571 0.700 0.800 5.484 (1.162–25.88) 0.031
Multivariate Analysis
fCER and treatment 1.000 1.000 1.000 1.000 1.000
D*CER and treatment 0.929 1.000 0.857 0.889 1.000
rCBFCER and treatment 0.929 0.875 0.857 0.875 0.857
rCBVCER and treatment 0.893 1.000 0.750 0.778 1.000

aData are hazard ratio estimates, with 95% confidence intervals in parentheses, for variables included in the Cox regression model (imaging features plus clinical variables) for the analysis of the association between the imaging features and overall survival after adjusting for standard clinical variables. Likelihood ratio test of this model versus the null model: P = 0.047 (test statistic = 15.66 with eight degrees of freedom).