Table 2 Univariable Cox proportional hazards analysis of the risk of recurrence.

From: Deep learning predicts postsurgical recurrence of hepatocellular carcinoma from digital histopathologic images

Patient characteristics

TCGA-HCC test cohort (n = 53)

Stanford-HCC (n = 198)

Internal test set

External test set

Hazard ratio (95% CI)

p value

Hazard ratio (95% CI)

p value

Risk score (binarized)

6.52 (1.83, 23.2)

0.0038

3.72 (2.17, 6.37)

 < 0.0001

Age (at surgery)

 > 60 years

0.83 (0.38, 1.8)

0.65

1.1 (0.64, 1.8)

0.77

Gender

Female

0.99 (0.44, 2.2)

0.98

0.99 (0.53, 1.9)

0.98

Hepatitis B virus infection

Positive

0.57 (0.22, 1.5)

0.25

1.1 (0.62, 1.9)

0.78

Hepatitis C virus infection

Positive

1.2 (0.4, 3.6)

0.74

0.66 (0.4, 1.1)

0.11

Alcohol intake

Positive

0.89 (0.37, 2.2)

0.80

0.19 (0.027, 1.4)

0.10

Non-alcoholic fatty liver disease

Positive

1.6 (0.46, 5.4)

0.48

1.8 (0.81, 3.9)

0.15

AJCC stage grouping

 > II

1.3 (0.49, 3.6)

0.58

4.4 (2.3, 8.3)

 < 0.0001

Largest tumor diameter (mm)

 > 50

1.1 (0.52, 2.5)

0.74

3.5 (2.1, 5.8)

 < 0.0001

Tumor multifocality

Positive

1.3 (0.53, 3.4)

0.53

1.1 (0.64, 1.9)

0.71

Histologic grade

 > Moderately-differentiated

1.1 (0.44, 2.6)

0.90

2.1 (1.2, 3.9)

0.013

Microvascular invasion

Positive

1.4 (0.6, 3.1)

0.46

3.9 (2.4, 6.5)

 < 0.0001

Macrovascular invasion

Positive

1.6 (0.46, 5.4)

0.48

5.3 (2.1, 1.3)

0.00043

Surgical margin

Positive

0.74 (0.22, 2.5)

0.63

6.8 (1.6, 28)

0.0090

Fibrosis stage

 > 2

2.7 (0.98, 7.7)

0.054

0.33 (0.2, 0.55)

 < 0.0001

  1. CI, confidence interval.