Skip to main content
. 2018 Apr 18;20(3):485–498. doi: 10.1093/biostatistics/kxy013

Table 4.

Estimates (EST) and standard errors (ESD) of measures of predictive capacity summarizing predictions of hepatocellular carcinoma based on des-Inline graphic-carboxyprothrombin biomarker and a partly conditional logistic model with the logit link function (Inline graphic)

HALT-C NCC study prediction evaluation results (Inline graphic = 12 months)
  s = 6 months s = 1 year s = 2 years s = 3 years
  (Inline graphic/Inline graphic = 4/109) (Inline graphic/Inline graphic = 5/105) (Inline graphic/Inline graphic = 11/94) (Inline graphic/Inline graphic = 17/76)
  EST (ESD) EST (ESD) EST (ESD) EST (ESD)
PE(Inline graphic10) 0.042 (0.021) 0.055 (0.024) 0.107 (0.035) 0.214 (0.056)
AUC 0.709 (0.125) 0.782 (0.121) 0.858 (0.067) 0.748 (0.077)
PCF(0.2) 0.500 (0.223) 0.600 (0.207) 0.818 (0.112) 0.646 (0.120)
PNF(0.8) 0.637 (0.197) 0.747 (0.285) 0.147 (0.210) 0.523 (0.192)

PE(Inline graphic10) = prediction error Inline graphic 10; AUC = area under the ROC curve; PCF(0.2) = the proportion of events in Inline graphic timeframe from Inline graphic captured if 20% of the population at risk at time Inline graphic were followed; PNF(0.8) = the proportion of the population at risk at time Inline graphic that would need to be followed in order to capture 80% of the events in the timeframe Inline graphic from Inline graphic. The estimates were obtained Inline graphic months and Inline graphic months. The number of events between Inline graphic and Inline graphic, and the number of subjects with no events before Inline graphic, are denoted by Inline graphic and Inline graphic, respectively. The standard errors were estimated with 500 perturbations.