Table 3 Performance statistics of the different machine learning models, and EKFCcrea-equation, in the external validation dataset using only serum creatinine as biomarker.

From: Comparison between the EKFC-equation and machine learning models to predict Glomerular Filtration Rate

External validation cohort (n = 8378)

Methods

Median bias (95%CI)

IQR (P25-P75)

P10 (95%CI)

P30 (95%CI)

MSE

EKFC

−0.90 [−1.30; −0.60]

17.15 (−9.20, 7.50)

0.43 [0.42; 0.44]

0.87 [0.86; 0.87]

296.49

RF

−0.39 [−0.73; −0.01]

17.13 (−9.51, 8.20)

0.41 [0.40; 0.42]

0.85 [0.85; 0.86]

294.24

LR

−3.06 [−3.51; −2.50]

22.21 (−15.49, 9.77)

0.31 [0.30; 0.32]

0.76 [0.75; 0.77]

509.88

XGBoost

−0.79 [−1.19; −0.41]

17.86 (−10.71, 8.43)

0.38 [0.37; 0.39]

0.84 [0.83; 0.84]

320.56

  1. EKFC = European Kidney Function Consortium equation; RF = Random Forest model; LR = Linear Regression model; XGBoost = eXtreme Gradient Boosting model; IQR = Interquartile range; P10/P30 = fraction of patients with eGFR within 10%/30% of mGFR; MSE = Mean Square Error.
Back to article page