Skip to main content

Table 3 Performance of different cox-regression models to predict outcome combining radiomics features with clinical parameters (significant P values are written in bold)

From: Multiparametric detection and outcome prediction of pancreatic cancer involving dual-energy CT, diffusion-weighted MRI, and radiomics

 

Endpoint

(all-cause mortality)

Model

b

SE

Exp(b)

95% CI of Exp(b)

Chi-squared

P value

Unadjusted Model 1

1.256

0.593

3.513

1.099 to 11.224

5.088

0.0241

Adjusted Model 2

1.434

0.615

4.196

1.258 to 13.995

11.207

0.0037

Adjusted Model 3

2.173

0.702

8.788

2.220 to 34.797

23.444

 < 0.0001

Adjusted Model 4

2.415

0.739

11.192

2.630 to 47.627

29.014

 < 0.0001

Adjusted Model 5

1.638

0.638

5.145

1.474 to 17.952

14.632

0.0055

  1. Model 1: unadjusted radiomics model. Model 2: additionally adjusted by T stage. Model 3: additionally adjusted by alcohol abuse and smoking. Model 4: additionally adjusted by T stage, alcohol abuse, and smoking. Model 5: additionally adjusted by TNM stage
  2. • Variables that did not reach univariate significance:
  3. ◦ Death: age (P = .75), sex (P = .29), arterial hypertension (P = .35), diabetes mellitus (P = .60), family history (P = .44), hypercholesterolemia (P = .75), chronic pancreatitis (P = .26), obesity (P = .21), C-reactive protein (P = .77), lipase (P = .44), lactate dehydrogenase (P = .47), creatinine (P = .49), glomerular filtration rate (P = .61), leucocytes (P = .99), CA 19–9 (P = .58), CEA (P = .76), N stage (P = .54), M stage (P = .53)
  4. • Variables that reached univariate significance:
  5. ◦ Death: alcohol abuse (P = .04), smoking (P = .03), T stage (P = .03)
  6. Abbreviations: b Regression coefficient, SE Standard error, Exp(b) Ratio of hazard rates, CI Confidence interval