Table 2 Performance of the new and Korean undiagnosed diabetes screening method in the development and validation datasets.

From: Comparisons of the prediction models for undiagnosed diabetes between machine learning versus traditional statistical methods

 

Model

Screening method

Feature

AUC

(95% CI)

Youden index

Sensitivity (%)

Specificity (%)

PPV

NPV

PLR

NLR

Train & Internal Validation Set

Park*

Risk score

Sex, Age, WC, RHR

0.745

(0.717 to 0.773)

37.00

70

66

0.08

0.98

2.09

0.45

Logistic Regression

Logistic Regression

0.780

(0.754 to 0.806)

41.90

80.94

60.92

0.09

0.98

2.07

0.31

Random Forest

Random Forest Classifier

0.781

(0.756 to 0.806)

41.20

84.60

56.60

0.08

0.99

2.1

0.16

LGBM

LightGBM Classifier

0.778

(0.752 to 0.804)

41.70

82.00

61.60

0.08

0.99

2.14

0.29

XGB

XGBoost Classifier

0.778

(0.752 to 0.804)

41.50

82.40

59.10

0.08

0.98

2.12

0.23

Ada

AdaBoost Classifier

0.780

(0.754 to 0.806)

41.80

82.60

59.20

0.08

0.99

2.03

0.29

External Validation set

Park*

Risk score

Sex, Age, WC, RHR

0.740

(0.721 to 0.759)

35.00

75

61

0.09

0.98

1.9

0.42

Logistic Regression

Logistic Regression

0.786

(0.77 to 0.802)

43.30

80.25

63.04

0.11

0.98

2.2

0.31

Random Forest

Random Forest Classifier

0.788

(0.772 to 0.804)

44.00

87.40

56.50

0.18

0.99

2.01

0.22

LGBM

LightGBM Classifier

0.788

(0.772 to 0.804)

43.70

82.90

60.80

0.1

0.99

2.12

0.28

XGB

XGBoost Classifier

0.788

(0.772 to 0.804)

44.00

85.80

58.20

0.1

0.99

2.05

0.24

Ada

AdaBoost Classifier

0.779

(0.762 to 0.796)

42.40

81.20

61.30

0.1

0.98

2.1

0.31

  1. *Park et al. 20226, When Park model’s performance was tested, data from 2019, 2020 were used to build prediction model and data from 2014, 2015, 2016, 2017, 2018 were used to validate. WC: Waist circumference, RHR: Resting heart rate, LGBM: Light Gradient Boosting Machine, XGB: Extreme Gradient Boosting), Ada: Ada Boost. AUC: The receiver operating characteristics curve under the curve. For this study, five different machine learning classification algorithms were used to predict undiagnosed diabetes. Based on their performance assessed by AUC, results from the best performed machine learning classification was used.