SPC365

AIMING-PACE: Artificial Intelligence and Machine learnING-based prediction model for PACEmaker implantation after transcatheter aortic valve replacement

Session: Comunicações Orais - Sessão 05 - Arritmologia: da cardioneuroablação até à inteligência artificial
Speaker: Francisco Barbas de Albuquerque

Congress: CPC 2024
Topic: C. Arrhythmias and Device Therapy
Theme: 04. Arrhythmias, General
Subtheme: 04.4 Arrhythmias, General – Treatment
Session Type: Comunicações Orais
FP Number: ---
Authors: Francisco Barbas De Albuquerque; Ricardo Carvalheiro; Bárbara Lacerda Teixeira; André Grazina; António Fiarresga; Inês Rodrigues; Tiago Mendonça; Rui Ferreira; Tomás Barbas de Albuquerque; Mário Oliveira; Rúben Ramos; Duarte Cacela

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><strong><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">Background</span></span></strong></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">Pacemaker (PM) implantation is a common complication after transcatheter aortic valve replacement (TAVR). Accurate identification of factors that contribute to PM implantation is crucial for clinical practice. Artificial intelligence (AI)- and machine-learning (ML) technologies may contribute to developing better prediction models in this clinical context.</span></span></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><strong><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">Aim</span></span></strong></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">To develop a ML-based Binary Classification Model for predicting PM implantation after TAVR and to compare it with conventional regression-based models.</span></span></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><strong><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">Methods</span></span></strong></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">Single-center retrospective study on patients with severe aortic stenosis who underwent TAVR between 2018 and 2022. PM implantation group was considered until the hospital discharge date after TAVR. Both pre- and intra-procedural variables were included in the dataset, after a careful review. We developed a Python script to build the Binary Classification model. For dataset balancing, a SMOTE-based upsampling technique was performed on the minority class. The XGBoost (eXtreme Gradient Boosting) open-source software library and algorithm were used to train the final model. To achieve a better performance, we applied a 5-fold cross-validation. Model performance metrics were computed using the confusion matrix of predictions on the validation set and are as follows: accuracy, precision, and recall. A logistic regression was executed for performance comparison between models. AUC of ROC were computed for each model.</span></span></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><strong><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">Results</span></span></strong></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">From a total of 673 TAVR procedures, 560 patients entered the analysis. Median age was 83 years and 44% were male. PM implantation occurred in 150 (26.8%) patients after TAVR. The mean time until PM implantation was 4.2 days after the procedure.</span></span></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">Using our XGBoost ML algorithm, a scoring model was generated, representing the variables from the highest to the lowest weighted (Figure 1). The highest weighted variables were the presence of right bundle branch block, history of smoking, male sex, history of myocardial infarction and history of peripheral artery disease. </span></span></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">ML-model performance metrics were as follows: accuracy 82%, precision 85.1% and recall 77.6%. Figure 2 represents the model metrics after cumulative column removal from the least to the highest weighted variable.&nbsp;</span></span> </span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">The logistic regression performance was: accuracy 63.8%, recall 59.7% and precision 64.9%. Figure 3 represents the relative weight of each variable computed in logistic regression.</span></span></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">AUC from ROC curves of ML-model (AUC=0.89) and logistic regression (AUC=0.67) model are depicted in figure 4.</span></span></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><strong><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">Conclusion</span></span></strong></span></span></p>

<p style="text-align:justify"><span style="font-size:11pt"><span style="font-family:Aptos,sans-serif"><span style="font-family:&quot;Times New Roman&quot;,serif"><span style="color:black">Our ML-based prediction model provides valuable and novel insights regarding the factors contributing PM implantation after TAVR. It outperformed the traditional regression-based models by a great margin. This underscores the move towards a more personalized medicine, where AI enhances clinical decision-making for better patient outcomes.</span></span></span></span></p>