Table 3.

Model Updating Results.^a,b

	No Updating	Recalibration		Model Revision		Model Extension			Shrinkage Model 6
	Method 1: Original Model	Method 2: Update Intercept	Method 3: Recalibration of Intercept and Slope	Method 4: Recalibration and Selective Reestimation	Method 5: Reestimation	Method 6: Recalibration (and Selective Reestimation) and Selective Reextension	Method 7: Reestimation and Selective Extension	Method 8: Re-estimation and Extension
Intercept	−2.776	−2.154	−1.049	−1.049	−0.849	−1.361	−1.119	−1.027	−1.115
Age	0.021c	0.021	0.011	0.011	0.006	0.011	0.004	0.004	0.011d
Highest lactatee	0.732c	0.732	0.384	0.384	0.508	0.384	0.537	0.515	0.384d
Aminoglycoside	0.951c	0.951	0.498	0.498	0.275	0.498	0.172	0.175	0.498d
Lowest phosphate						0.347	0.403	0.384	0.070f
Erythrocyte transfusion								0.083
Highest calcium								−0.06
Hosmer-Lemeshow test	<0.001	0.038	0.550	0.550	0.739	0.265	0.837	0.549	0.208
AUC-ROC	0.60 (0.54-0.66)	0.60 (0.54-0.66)	0.60 (0.54-0.66)	0.60 (0.54-0.66)	0.60 (0.54-0.66)	0.60 (0.54-0.66)	0.61 (0.55-0.67)	0.61 (0.55-0.67)	0.60 (0.54-0.66)

Abbreviation: AUC-ROC, area under the receiver operating characteristic curve.

^aMethod 1 is the original model. The model was recalibrated by adjusting only the intercept (method 2) or both the intercept and slope (method 3). With method 4, we investigated whether predictors were having a clearly different effect in the validation cohort, by selective reestimation of one or more of the included predictors. None of the models with reestimations improved the model; therefore, no selective reestimations were done. In method 5, the model was fitted in the validation data by reestimation of the intercept and regression coefficients for all predictors. In method 6, the 3 new predictors were one-by-one added to the recalibrated model. Only adding lowest phosphate improved the model. In method 7, model 5 was extended with new predictors. In method 8, a model with all old and new predictors was assessed.

^bShrinkage was applied to the improved model (method 6), and the intercept was recalculated.

^cUniform shrinkage factor applied.

^dShrinkage toward recalibrated values.

^eTransformed using the natural logarithm.

^fShrinkage toward zero.