Establishment and internal evaluation of risk prediction model
for severe CRF
According the Table 3, we can draw the CPMs of severe CRF as follows:
Logit(P )=1.276-0.947 monthly income+0.989 long-term passive
smoking -0.952 physical exercise+1.512 diagnosis type+1.040 coping style
-0.726 PSS -2.350 SOC.
Visualize the model in the form of nomograms, as shown in Figure 2. The
C-Index of nomogram model calculated by Bootstrap method is 0.921 (95%
CI: 0.877~ 0.958), which indicates that the model has
good discrimination (Figure 2). It can be seen from the ROC curve that
the best cutoff value of the prediction probability of the nomogram
model is 0.412, which corresponds to the maximum Jordan index of 0.721
(Figure 3-1). At this time, the sensitivity of the model is 0.821, the
specificity is 0.900, and the accuracy is 0.857. AUC is 0.916 (95% CI:
0.876~0.957), which further indicates that the model has
high discrimination. The calibration curve shows that the predicted
probability of the model fits well with the actual probability. The
Hosmer Lemeshow verification shows thatχ2 =9.021, P =0.340, greater than 0.05,
further indicating that the model has good calibration (Figure 3-2). It
can be seen from the DCA curve that when the prediction probability is
greater than about 10%, the benefit from using this model is positive,
and there is a wide threshold range, which indicates that the use of
this nomogram can benefit better (Figure 3-3).