Table 2 (A) Model selection according to the posterior probability of model by Bayesian model averaging and (B) Bayesian logistic regression. Best model according to Bayesian model averaging.

From: Molecular profiling of advanced solid tumours. The impact of experimental molecular-matched therapies on cancer patient outcomes in early-phase trials: the MAST study

(A)

 

p!=0

EV

SD

Model 1

Model 2

Model 3

Model 4

Model 5

Intercept

100

−1.175

2.053

−2.674

−0.527

−1.377

−3.602

1.084

Age

48.7

−0.028

0.035

–

−0.057

−0.063

–

−0.053

PS

6.0

−0.031

0.237

–

–

–

–

–

Treatment

86.8

1.644

0.921

1.847

1.904

1.897

1.943

–

Leucocytes

6.3

<0.001

<0.001

–

–

–

–

–

Lymphocytes

19.5

<0.001

<0.001

–

–

<0.001

<0.001

–

Neutrophils

6.1

<0.001

<0.001

–

–

–

–

–

RMH score

8.1

0.068

0.342

–

–

–

–

–

nVar

   

1

2

2

2

1

BIC

   

−311

−311

−309

−308

−308

Post prob

   

0.232

0.222

0.078

0.071

0.050

(B)

Parameters

Mean

SD

2.5%

25%

50%

75%

97.5%

Rhat

N.eff

Intercept

−2.790

0.560

−4.009

−3.123

−2.749

−2.392

−1.871

1

900

Treatment (ref. no)

1.938

0.724

0.664

1.434

1.951

2.400

3.482

1

1000

Deviance

60.008

2.040

58.010

58.570

59.400

60.790

65.359

1

520

  1. Models are ordered according to posterior probability; estimates are in the log odds scale.
  2. Best mode includes only Treatment (No targeted therapy as reference) as an independent factor.
  3. p!=0 posterior probability that the variable is in the model, EV BMA posterior mean, SD posterior standard deviation, nVar number of variables in the model, BIC Bayesian Information Criterion, post prob the posterior probability of the model.
Back to article page