Skip to main content

Table 2 Percentage of biopsies that would be spared or delayed using machine-learning and logistic regression models at given sensitivity for detection of CSPCa in the validation cohorts

From: Development and head-to-head comparison of machine-learning models to identify patients requiring prostate biopsy

Models

Sensitivity for detection of CSPCa

Cut-off for predicted risk (%)

Biopsies sSpared † (n = 208), n (%)

Unnecessary biopsy avoided

Biopsy delayed

GS < 3 + 3 (n = 131), n (%)

GS = 3 + 3 (n = 12), n (%)

GS = 3 + 4 (n = 11), n (%)

GS = 4 + 3 (n = 18), n (%)

GS ≥ 4 + 4 (n = 36), n (%)

Biopsies spared or delayed using PCa models at given sensitivity for detection of CSPCa

ANN

64/65 (98%)

9

34 (16)

32 (24)

1 (8)

0 (0)

1 (6)

0 (0)

SVM

64/65 (98%)

11

57 (24)

55 (42)

1 (8)

0 (0)

0 (0)

1 (3)

CART

64/65 (98%)

NA

NA

NA

NA

NA

NA

NA

RF

64/65 (98%)

7

61 (29)

57 (44)

3 (25)

1 (9)

0 (0)

0 (0)

LR

64/65 (98%)

7

30 (14)

27 (21)

2 (17)

1 (9)

0 (0)

0 (0)

ANN

62/65 (95%)

11

56 (25)

51 (39)

2 (17)

1 (9)

1 (6)

1 (3)

SVM

62/65 (95%)

14

77 (37)

71 (54)

3 (25)

1 (9)

1 (6)

1 (3)

CART

62/65 (95%)

NA

NA

NA

NA

NA

NA

NA

RF

62/65 (95%)

11

79 (38)

73 (56)

3 (25)

1 (9)

2 (11)

0 (0)

LR

62/65 (95%)

10

53 (25)

47 (36)

3 (25)

1 (9)

1 (6)

1 (3)

ANN

59/65 (91%)

27

109 (52)

99 (76)

4 (33)

3 (27)

1 (6)

2 (6)

SVM

59/65 (91%)

23

110 (53)

100 (76)

4 (33)

4 (36)

1 (6)

1 (3)

CART

57/65 (88%)

10

104 (50)

90 (69)

6 (50)

4 (36)

1 (6)

3 (8)

RF

59/65 (91%)

20

101 (49)

91 (69)

4 (33)

4 (36)

2 (11)

0 (0)

LR

59/65 (91%)

24

107 (51)

97 (74)

4 (33)

4 (36)

1 (6)

1 (3)

Biopsies spared or delayed using CSPCa models at given sensitivity for detection of CSPCa

ANN

64/65 (98%)

7

60 (29)

58 (44)

1 (8)

0 (0)

0 (0)

1 (3)

SVM

64/65 (98%)

9

69 (33)

65 (50)

3 (25)

0 (0)

0 (0)

1 (3)

CART

64/65 (98%)

NA

NA

NA

NA

NA

NA

NA

RF

64/65 (98%)

7

79 (38)

75 (57)

3 (25)

1 (9)

0 (0)

0 (0)

LR

64/65 (98%)

6

61 (29)

57 (44)

3 (25)

0 (0)

1 (6)

0 (0)

ANN

62/65 (95%)

8

79 (38)

75 (57)

1 (8)

1 (9)

1 (6)

1 (3)

SVM

62/65 (95%)

10

82 (39)

76 (58)

3 (25)

1 (9)

1 (6)

1 (3)

CART

62/65 (95%)

NA

NA

NA

NA

NA

NA

NA

RF

62/65 (95%)

8

84 (40)

78 (60)

3 (25)

1 (9)

2 (11)

0 (0)

LR

62/65 (95%)

7

70 (34)

64 (49)

3 (25)

1 (9)

1 (6)

1 (3)

ANN

59/65 (91%)

9

96 (46)

86 (66)

4 (33)

4 (36)

1 (6)

1 (3)

SVM

59/65 (91%)

18

124 (60)

112 (85)

6 (50)

4 (36)

1 (6)

1 (3)

CART

58/65 (89%)

10

109 (52)

97 (74)

5 (42)

4 (36)

1 (6)

2 (6)

RF

59/65 (91%)

13

102 (49)

92 (70)

4 (33)

3 (27)

3 (17)

0 (0)

LR

59/65 (91%)

14

105 (50)

32 (24)

4 (33)

3 (27)

1 (6)

2 (6)

  1. PCa prostate cancer, CSPCa clinically significant prostate cancer, GS Gleason score, ANN artificial neural network, SVM support vector machine, CART classification and regression tree, RF random forest, LR logistic regression, NA not applicable
  2. †Number of biopsies spared = number of unnecessary biopsy avoided + number of biopsy delayed
Back to article page

BMC Urology

ISSN: 1471-2490

Contact us