Table 3 Equation performance in kidney donor dataset (n = 287).

From: Deep learning body-composition analysis of clinically acquired CT-scans estimates creatinine excretion with high accuracy in patients and healthy individuals

Formula

Bias

Precision

Accuracy

MPE [95%CI]

R2 [95% CI]

RMSE [95% CI]

MAPE [95% CI]

P15 [95% CI]

P30 [95% CI]

Cockcroft–Gault

− 1.66 [− 1.88; − 1.43]

0.64 [0.57; 0.70]

2.57 [2.37; 2.76]

16.5% [15.3–17.7%]

47% [41%; 53%]

90% [86%; 93%]

Ix

− 1.08 [− 1.29; − 0.87]

0.70 [0.64; 0.75]

2.10 [1.92; 2.27]

13.1% [12.1–14.2%]

63% [57%; 68%]

97% [95%; 99%]

CRAFT 1

− 0.71 [− 0.91; − 0.51]

0.73 [0.67; 0.78]

1.86 [1.70; 2.01]

12.2%[11.1–13.3%]

70% [65%; 75%]

97% [94%; 99%]

CRAFT 2

− 0.73 [− 0.94; 0.52]

0.69 [0.63; 0.74]

1.97 [1.80; 2.14]

12.7% [11.6–13.8%]

67% [62%; 72%]

95% [93%; 98%]

  1. Confidence intervals were calculated with the combined variance of multiple imputation (10×) and bootstrap (1000×).
  2. MPE mean prediction error (mmol/day), MAPE mean absolute percentage error, RMSE root mean squared error (mmol/day), R2 the R2-value calculated with linear regression, p15/p30 the percentage of points that fall within 15%/30% of the outcome.
Back to article page