Skip to main content

Table 3 Comparison of the performance of revised 4-variable and 9-variable CKD-EPI equations and 9-variable ANN model: internal validation

From: Improving accuracy of estimating glomerular filtration rate using artificial neural network: model development and validation

 OverallMeasured GFR < 60 mL/min/1.73 m2Measured GFR ≥ 60 mL/min/1.73 m2
Bias—median difference (95% CI)
 4-variable CKD-EPI equation4.67 (3.55 to 5.90)11.04 (9.47 to 12.32)0.03 (− 1.73 to 1.18)
 9-variable CKD-EPI equation5.00 (3.82 to 6.54) P = 0.510.95 (9.08 to 12.60) P = 0.8− 0.10 (− 1.54 to 2.09) P = 0.3
 9-variable ANN2.77 (1.82 to 4.10) P = 0.00710.54 (8.40 to 11.78) P = 0.2− 2.91 (− 4.60 to − 1.32) P = 0.01
Precision – IQR of the difference (95% CI)
 4-variable CKD-EPI equation20.11 (18.46 to 21.80)15.90 (13.90 to 17.87)21.08 (19.34 to 23.80)
 9-variable CKD-EPI equation18.91 (17.43 to 20.48) P = 0.0516.73 (14.67 to 19.05) P = 0.321.01 (18.69 to 23.78) P = 0.9
 9-variable ANN19.33 (17.77 to 21.17) P = 0.316.03 (14.15 to 17.72) P = 0.920.80 (19.19 to 22.84) P = 0.7
Accuracy—P30, % (95% CI)
 4-variable CKD-EPI equation75.8 (72.9 to 78.6)53.2 (47.9 to 58.2)91.5 (88.8 to 93.6)
 9-variable CKD-EPI equation76.6 (73.7 to 79.5) P = 0.454.6 (49.6 to 59.7) P = 0.491.9 (89.6 to 94.0) P = 0.7
 9-variable ANN80.0 (77.4 to 82.7) P < 0.00159.9 (54.6 to 64.9) P < 0.00194.0 (91.5 to 95.9) P = 0.01
  1. GFR glomerular filtration rate, CKD-EPI chronic kidney disease epidemiology collaboration, ANN artificial neural network, IQR interquartile range, CI confidence interval