Skip to main content

Table 3 Bias, precision and accuracy of each model in the external validation data set

From: Improving precision of glomerular filtration rate estimating model by ensemble learning

Variable Measured GFR (ml/min/1.73 m2)
Overall < 30 ≥ 30 and < 60 ≥ 60
Bias = median difference (95% CI)
 Regression model 2.3 (1.0–3.4) 4.4 (2.9–5.9) 3.1 (1.5–6.5) −1.9 (−4.5 to 0.9)
 ANN model 3.2 (2.2–5.4) 5.4 (3.1–7.4) 5.4 (2.4–7.6) 0.8 (−3.9 to 2.7)
 SVM model 3.6 (2.6–4.9) 6.8 (4.9–9.0) 4.0 (2.2–6.4) −0.2 (−2.3 to 2.6)
 Ensemble model 3.4 (2.3–4.4) 5.6 (3.7–8.2) 4.0 (2.1–6.7) −0.5 (−3.9 to 2.3)
Precision = IQR of the difference (95% CI)
 Regression model 14.0 (12.4–15.9) 9.2 (7.3–11.8) 13.5 (11.2–18.0) 19.6 (16.8 to 23.5)
 ANN model 15.1 (13.6–17.0) 11.1 (9.1–14.8) 14.9 (13.1–17.7) 20.5 (17.9 to 25.1)
 SVM model 14.2 (12.4–16.0) 9.5 (7.5–12.1) 12.9 (10.3–16.2) 18.5 (14.9 to 21.5)
 Ensemble model 13.5 (11.8–14.9) 8.9 (7.0–11.0) 12.7 (10.4–16.0) 17.9 (15.44 to 21.9)
Accuracy = 30% accuracy (95% CI)
 Regression model 75.1 (70.7–79.4) 52.4 (42.7–61.2) 75.2 (67.8–81.9) 89.1 (83.6 to 93.3)
 ANN model 73.4 (69.0–77.2) 54.4 (44.7–64.1) 70.5 (63.11–77.2) 87.9 (82.4 to 92.1)
 SVM model 73.1 (68.8–77.2) 47.6 (37.9–57.3) 71.1 (63.11–77.9) 90.9 (86.1 to 94.5)
 Ensemble model 75.5 (71.5–79.6) 52.4 (42.7–62.1) 73.8 (65.11–79.9) 91.5 (86.7 to 95.2)
  1. GFR glomerular filtration rate, ANN artificial neural network, SVM support vector machine, IQR interquartile range, CI confidence interval
  2. P < 0.05 compared with regression model-GFR