Model validation
The above model can be used to calculate the probability of severe CRF
of each CC patient in the model validation group to form a prediction
probability set. Then, based on the actual occurrence of severe CRF of
each patient in the validation group, ROC curve (Figure 3-4),
Calibration calibration curve (Figure 3-5) and DCA curve (Figure 3-6)
can be generated in R software to evaluate the differentiation, accuracy
and clinical practicability of the model respectively. Finally, the AUC
of the area under the ROC curve is 0.928 (0.876~0.980),
and the best cut-off value of the prediction probability of the nomogram
model is 0.444, corresponding to the maximum Youdenindex of 0.748. At
this time, the sensitivity of the model is 0.889, the specificity is
0.860, and the accuracy is 0.875, indicating that the model has a high
degree of differentiation; Calibration calibration curve has good
consistency (χ2 =8.89,P =0.340>0.05); It can be seen from the DCA curve
that when the prediction probability is greater than about 12%, the
benefit of using this model is positive, and it has a wide threshold
range, which indicates that this model has good clinical practicability.