Skip to main content

Table 3 Predictive performance in cross validation and external validation

From: Using real-world data to dynamically predict flares during tapering of biological DMARDs in rheumatoid arthritis: development, validation, and potential impact of prediction-aided decisions

 

Cross validation (cutoff 14.3%)

External validation (cutoff 14.3%)

External validation (cutoff 31.5%)

AUC

0.76 (0.69–0.83)

0.68 (0.62–0.73)

0.68 (0.62–0.73)

Sensitivity (%)

86.1 (81.9–90.1)

73.2 (64.4–82.0)

58.8 (49.0–68.6)

Specificity (%)

66.5 (60.1–72.5)

52.0 (0.48–56.0)

68.7 (64.9–72.4)

Positive predictive value (%)

33.0 (29.3–38.5)

20.1 (15.9–24.3)

23.7 (18.3–29.0)

Negative predictive value (%)

96.2 (95.4–98.4)

92.1 (89.2–95.0)

91.0 (88.3–93.6)

Accuracy (%)

70.6 (65.6–75.6)

55.0 (51.2–58.7)

67.3 (63.6–70.8)

  1. Results from the 5-fold cross-validation in development data are presented for an optimal cutoff point of 14.3% as determined with Youden’s index. The results from external validation in the DRESS trial [9] are presented for 2 different cutoff points: the optimal cutoff point from the development data (14.3%), and the optimal cutoff point in the DRESS data as determined by Youden’s index (31.5%). 95% confidence intervals are presented between brackets
  2. AUC area under the curve
Back to article page

Arthritis Research & Therapy

ISSN: 1478-6362

Contact us