Table 1 Parameter estimates, model goodness of fit, model complexity, and predictive accuracy (case study I)
Whole data analysis200 CVs
ModelPredictorsLog likelihoodbEffective number of parameters (pD)Deviance information criteria (DIC)Average CV-AUCcProportion of times (of 200 CVs) model in column had AUC > model in row
Age at diagnosisRaceaLobular (Y/N)Tumor subtypePathological stageGene expressionM2M3M4M5M6M7 (COV)M8 (COV + WGGE)
M1X−146.12.1294.30.557d (0.007)0.14<0.01>0.99>0.99>0.99>0.99>0.99
M2X−147.52.0296.90.525d,e (0.023)0.59>0.99>0.99>0.99>0.99>0.99
M3X−144.32.0290.60.526e (0.020)>0.99>0.99>0.99>0.99>0.99
M4X−138.64.1281.30.618f (0.013)0.14>0.99>0.99>0.99
M5X−142.42.0286.90.596f (0.012)>0.99>0.99>0.99
M6X−132.415.5280.30.659g (0.011)>0.99>0.99
M7: COVXXXXX−146.33.2295.80.704h (0.007)>0.99
M8: COV + WGGEXXXXXX−131.317.6280.30.721i (0.010)
  • a African American, Y/N.

  • b Estimated posterior mean of the log likelihood.

  • c Average over 200 tenfold CVs.

  • d,e,f,g,h,i The same letter indicates that the models are no different (empirical P < 0.05).