Advertisement

B-011-06 USING MACHINE LEARNING TO IDENTIFY PATIENTS LIKELY TO REQUIRE REPEAT CATHETER ABLATION FOR ATRIAL FIBRILLATION: INSIGHTS FROM THE UC SAN DIEGO AF ABLATION REGISTRY

      Background

      Catheter ablation is superior to pharmacologic therapy in maintaining normal sinus rhythm in select patients with atrial fibrillation (AF). However, this procedure is not without risks, and many often require a repeat ablation.

      Objective

      To develop and evaluate whether a machine learning model can identify patients that will ultimately need a repeat left atrial ablation among those who undergo AF catheter ablation.

      Methods

      We performed a retrospective study of all patients who underwent radiofrequency catheter ablation enrolled in the University of California, San Diego AF Ablation Registry. We developed a machine learning model using a Naïve Bayes algorithm to identify patients that are likely to require repeat ablation after initial AF catheter ablation.

      Results

      550 consecutive patients who underwent AF catheter ablation were included in the study. During the follow up period, 172 (31.3%) patients underwent repeat ablation. Data was split into training (440) and test (110) sets for validation. When validated using the test set, the Naïve Bayes machine learning algorithm had a total test accuracy of 80.0% with a sensitivity of 93.8% and specificity of 41.3%. The area under the curve (AUC) was 0.844 (95% confidence interval (CI) 0.77-0.91) (Figure 1B). For reference, using the MB-LATER risk score for predicting very late AF recurrence after ablation, the best such clinical risk score for predicting AF recurrence has an AUC of 0.782.

      Conclusion

      Using a Naïve Bayes machine learning algorithm, we were able to reliably predict whether a patient would ultimately require repeat ablation or not after initial AF catheter ablation.
      Figure thumbnail fx1