Skip to main content

Table 3 Multivariate Cox models

From: DNA methylation gene-based models indicating independent poor outcome in prostate cancer

  Model A:Gene-Only Model B:Genes + PSA Model C:Gleason + PSA Model D:Final model
Variable HR (95% CI) χ2 P -value HR (95% CI) χ2 P -value HR (95% CI) χ2 P -value HR (95% CI) χ2 P -value
Gleason -b - - - - - 2.20 (1.82, 2.67) 66.3 3.3*10-16 2.72 (2.09, 3.53) 56.3 6.4*10-14
PSA - - - 1.27 (1.18, 1.38) 36.5 1.5*10-9 1.27 (1.17, 1.37) 34.9 3.5*10-9 1.23 (1.13, 1.34) 24.7 6.7*10-7
DPYS 1.12 (1.02, 1.24) 5.8 0.016 1.12 (1.02, 1.24) 5.3 0.021 - - - 1.13 (1.03, 1.25) 6.4 0.012
HSPB1     0.88 (0.78, 0.99) 4.6 0.032 - - - 2.39 (1.15, 4.97) 5.5 0.019
Gleason x HSPB1 a - - - - - - - - - 0.89 (0.81, 0.97) 6.2 0.012
CCND2        - - - 0.86 (0.75, 0.98) 5.1 0.024
MAL 1.19 (1.05, 1.34) 7.6 0.006 1.17 (1.03, 1.34) 5.7 0.017 - - -    
GSTP1 1.15 (1.03, 1.27) 6.6 0.010     - - -    
TIG1     1.15 (1.03, 1.27) 6.5 0.011 - - -    
LR χ 2 (df) 49.4 (3) 76.6 (5) 111.4 (2) 125.6 (6)
Harrell’s c-index (se) 0.716 (0.034) 0.771 (0.034) 0.831 (0.034) 0.835 (0.034)
Gönen & Heller’s c-index c (se) 0.696 (0.022) 0.702 (0.019) 0.738 (0.016) 0.757 (0.017)
  1. a)Cross-product of Gleason score multiplied by HSPB1 methylation. For construction of a full model, all clinical variables and genes were included as well as interaction terms between each of the genes and the variables. The only significant interaction was found for Gleason score and HSPB1.
  2. b)Variable not included in model.
  3. c)The Gönen & Heller’s c-index is independent of the degree of censoring and is somewhat comparable to an area under the curve corresponding to a plot of the sensitivity versus positive predictive value of the predictor.
  4. (df) = degrees of freedom.
  5. (se) = standard error.