Table 3 Parameter estimates, model goodness of fit, model complexity, and prediction accuracy (case study III)
Whole data analysis200 CVs
SetModelFactors IncludedVariance (90% posterior confidence region)Log likelihoodeEffective number of parameters (pD)Deviance information criteria (DIC)Average CV-AUCf (SD)Proportion of times model in column had AUC > model in the row
CovariatesaCNVbMETHcmiRNAdM13, M16, M19: omic onlyM14, M17, M20: COV + omic
Set 1 (n = 270)M12: COVX−125.58.1259.00.699g (0.009)<0.01>0.99
M13: CNVX0.637 (0.155; 1.124)−112.126.8250.90.653h (0.012)>0.99
M14: COV + CNVXX0.398 (0.070; 0.736)−110.524.5245.60.714i (0.009)
Set 2 (n = 199)M15: COVX−88.78.4185.70.667g (0.013)0.60>0.99
M16: METHX0.652 (0.086; 1.261)−76.618.7171.80.672g,h (0.017)0.76
M17: COV + METHXX0.402 (0.032; 0.739)−78.918.5176.30.684h (0.013)
Set 3 (n = 167)M18: COVX−71.28.2150.60.747g (0.011)<0.010.29
M19: miRNAX0.338 (0.072; 0.615)−75.213.5163.80.623h (0.018)>0.99
M20: COV + miRNAXX0.179 (0.029; 0.324)−67.313.8148.50.744g (0.011)
  • a Age: African American, Y/N; lobular (Y/N); cancer subtype and stage.

  • b Copy-number variants.

  • c Methylation.

  • d Whole-genome RNA-seq.

  • e Estimated posterior mean of the log likelihood.

  • f Average over 200 tenfold CVs.

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