3. Establishment and verification of the prediction model for post-CPVA recurrence in patients with PAF
In the modeling group, by using AF recurrence after radiofrequency ablation as the dependent variable, and using the patient’s gender, age, body mass index (BMI), LAD, LVEF, left ventricular end-diastolic dimension (LVEDD), LAAV, degree of MR, B-type brain natriuretic peptide (BNP), CHA2DS2-Vasc as independent variables, we performed multivariate logistic regression analysis. The results showed that the age of PAF patients, the degree of MR, and LAAV were all independent risk factors for recurrence after CPVA, Table 4. The equation for predicting AF recurrence after radiofrequency was: Logit(P)= -3.253 +0.092×age+1.263×mild MR +2.325×moderate MR +5.111×severe MR -0.113×LAAV.
The ROC curve of the prediction model in the modeling group was plotted. The AUC was 0.889 (95%CI: 0.793-0.986), the sensitivity was 76.5%, the specificity was 94.6%, the positive likelihood ratio was 14.224, and the negative likelihood ratio was 0.249. The AUC of the prediction model was better than the single-factor parameters LAD, LVEF, LVEDD, LAAV, BNP, CHA2DS2-Vasc (allP <0.05) (Figure 1A).
The ROC curve of the prediction model in the verification group was plotted. The AUC was 0.866 (95%CI: 0.711-1.000), the sensitivity was 71.4%, the specificity was 97.0%, the positive likelihood ratio was 23.571, and the negative likelihood ratio was 0.295. There was no significant difference in AUC between the modeling group and verification group (0.889 vs 0.866, P >0.05) (Figure 1B).
The nomogram (Figure 2) and calibration curve (Figure 3) of the prediction model were plotted. The calibration curve showed that the prediction model had good consistency between the predicted value and observed value in both modeling group and verification group.