Skip to main content

Table 2 Measurement of the plsRglm model based on the four-autoantibody signature in the diagnosis of ESCC and preclinical ESCC

From: Machine learning technique-based four-autoantibody test for early detection of esophageal squamous cell carcinoma: a multicenter, retrospective study with a nested case–control study

 

AUC (95% CI)

Sensitivity

Specificity

PLR

NLR

Training set

 ESCC (n = 388) vs. NC (n = 125)

0.860 (0.828–0.891)

68.8%/62.9%

90.4%/95.2%

7.17/13.10

0.34/0.39

 Early-stage ESCC (n = 53) vs. NC (n = 125)

0.817 (0.738–0.897)

62.3%/56.6%

90.4%/95.2%

6.49/11.79

0.42/0.46

Internal validation set

 ESCC (n = 312) vs. NC (n = 101)

0.826 (0.786–0.866)

65.3%/59.5%

89.1%/91.1%

5.99/6.68

0.39/0.44

 Early-stage ESCC (n = 72) vs. NC (n = 101)

0.842 (0.784–0.900)

62.5%/56.9%

89.1%/91.1%

5.74/6.39

0.42/0.47

External validation set 1

 ESCC (n = 237) vs. NC (n = 134)

0.851 (0.814–0.889)

69.2%/61.2%

87.3%/94.8%

5.45/11.71

0.35/0.41

 Early-stage ESCC (n = 76) vs. NC (n = 134)

0.854 (0.799–0.910)

63.2%/53.9%

87.3%/94.8%

4.98/10.33

0.42/0.49

External validation set 2

 Early-stage ESCC (n = 101) vs. NC (n = 101)

0.850 (0.795–0.904)

67.3%/61.4%

90.1%/95.0%

6.80/12.40

0.36/0.41

External validation set 3

 Prediagnostic ESCC (n = 24) vs. NC (n = 112)

0.723 (0.611–0.834)

54.2%/41.7%

86.6%/86.6%

4.04/3.11

0.53/0.67

  1. Data were generated according to the following criterion: the optimum cut-off values were determined by achieving the maximum sensitivity when the specificity was > 90% (for data after “/,” > 95%) as well as by minimizing the distance of the cut-off value to the top-left corner of the ROC curve, and were generated using whole ESCC versus all controls in the training set
  2. AUC Area under the curve, 95% CI 95% confidence interval, ESCC Esophageal squamous cell carcinoma, NC Normal controls, PLR Positive likelihood ratio, NLR Negative likelihood ratio