Skip to main content

Table 4 Performance metrics for determining colorectal cancer recurrence using pre-defined variable algorithms and conditional inference tree algorithms

From: Evaluation of algorithms using administrative health and structured electronic medical record data to determine breast and colorectal cancer recurrence in a Canadian province

 

Pre-defined variable algorithm, % (95% CIc)

Conditional tree algorithm, % (95% CI)

Training cohort (N = 933)

Validation cohort (N = 1811)

Training cohort (N = 620)

Validation cohort (N = 693)

 

Unweighted

Weighted

 

Unweighted

Weighted

Sensitivity

88.1 (81.7–92.9)

79.4 (72.8–86.0)

77.7 (74.1–80.7)

71.4 (63.5–79.4)

64.7 (56.6–72.8)

62.6 (58.9–66.4)

Specificity

89.9 (84.2–95.1)

92.5 (87.4–96.4)

92.8 (90.6–94.7)

96.2 (92.1–99.2)

96.9 (93.4–99.3)

97.8 (96.4–98.8)

PPVa

68.9 (59.0–81.8)

72.0 (60.2–85.1)

70.7 (64.8–76.7)

82.6 (70.3–95.7)

83.8 (71.0–95.9)

86.4 (79.8–92.3)

NPVb

96.7 (95.1–98.1)

94.8 (93.2–96.4)

94.9 (94.1–95.6)

93.0 (91.2–94.8)

91.4 (90.1–93.6)

92.2 (91.4–92.9)

Correct classification

89.5 (84.8–93.7)

89.9 (85.6–93.5)

90.1 (88.1–91.8)

91.1 (87.9–94.0)

90.6 (87.6–93.4)

91.4 (90.1–92.5)

Scaled Brier

0.35 (0.06–0.61)

0.36 (0.09–0.59)

0.33 (0.20–0.45)

0.45 (0.25–0.63)

0.41 (0.21–0.58)

0.42 (0.34–0.50)

  1. aPositive predictive value
  2. bNegative predictive value
  3. cConfidence interval