Skip to main content

Table 5 Logistic regression models with six different approaches

From: Bayesian predictors of very poor health related quality of life and mortality in patients with COPD

Very poor HRQoL/death outcome

Logit 1

Logit 2

Logit3

Logit 4

Logit5

Logit6

Parameter

OR

SE

OR

SE

OR

SE

OR

SE

OR

SE

OR

SE

Age at onset of COPD

0.9772

0.0178

0.9860

0.0148

  

0.8432*

0.0585

  

0.9342*

0.0273

Age at onset asthma

        

0.8469*

0.0560

  

Year of birth

      

0.8579

0.0680

0.8663

0.0647

0.9163*

0.0313

Diagnosed cerebrovascular disease

2.2090

0.9669

2.4865*

0.8858

2.5121**

0.7810

      

Diabetes

2.0537*

0.6558

2.2233**

0.6105

2.2204**

0.5260

      

Alcohol abuse

2.1063*

0.6814

2.4580**

0.6461

1.9996**

0.4726

    

2.4388**

0.7556

Cancer

2.5036

1.3626

2.1930**

1.2808

2.5107*

0.8961

      

Any psychiatric disease

4.6815***

1.7979

3.3283***

1.0637

2.5081**

0.7054

3.3262

2.4100

4.0301

2.9826

3.8206***

1.4034

Body mass index

1.2048

0.2493

1.4440*

0.2496

        

FEV1% of predicted spirometry

0.9761**

0.0071

1.4440***

0.0580

      

0.9851*

0.0065

Atrial fibrillation

3.1544*

3.1544

        

2.6895*

1.1314

Corrected QT-time

1.0093*

0.0047

          

Tests

Value

p

Value

p

Value

p

Value

p

Value

p

Value

p

N included to the model

365

 

547

 

647

 

75

 

75

 

389

 

Probability of the model (chi2)

52.90

<0.0001

72.00

<0.0001

48.21

<0.0001

10.25

0.0166

10.28

0.0163

36.94

<0.0001

Pseudo R2

12.18%

 

11.69%

 

6.38%

 

12.07%

 

12.11%

 

7.88%

 

Sensitivity

33.01%

 

24.09%

 

15.43%

 

21.05%

 

21.05%

 

21.24%

 

Specificity

95.80%

 

95.61%

 

93.64%

 

94.64%

 

94.64%

 

94.20%

 

Positive predictive value

75.56%

 

64.71%

 

47.37%

 

57.14%

 

57.14%

 

60.00%

 

Negative predictive value

78.44%

 

79.03%

 

74.92%

 

77.94%

 

77.94%

 

74.50%

 

Log likelihood

−190.73

 

−271.87

 

−353.57

 

−37.32

 

−37.31

 

−215.93

 

Akaike information criteria

403.46

 

561.74

 

719.14

 

82.65

 

82.61

 

445.87

 

Bayesian information criteria

446.36

 

600.48

 

745.98

 

91.92

 

91.88

 

473.61

 

Proportion of observations used

56.41%

 

84.54%

 

100.00%

 

11.59%

 

11.59%

 

60.12%

 

Correct classification of used N

78.08%

 

77.70%

 

72.49%

 

76.00%

 

76.00%

 

73.01%

 

Worst possible correct classification among all (N 647)

44.05%

 

65.69%

 

72.49%

 

8.81%

 

8.81%

 

43.89%

 
  1. NBCMM = naïve Bayesian classification merger model. Logit1 = all variables included in the NBCMM (comparable to the NBCMM in terms of predictors included to the logit model). Logit2 = variables which had at least 80% of observations present. Logit3 = variables which had 100% of observations present (comparable to the NBCMM in terms of number of observations included). Logit4 = forward stepwise elimination with p-value 0.10 threshold. Logit5 = backward stepwise elimination with p-value 0.10 threshold (shows how well the greedy hill-descending predictor selection of naïve Bayes approach performed in comparison to the p-value selection). Logit6 = manual one-by-one dropping of the predictor variable with the poorest p-value until all logit predictors have a p-value below 0.10. *p<0.050. **p<0.010. ***p<0.001.
Back to article page

BMC Medical Informatics and Decision Making

ISSN: 1472-6947

Contact us