Background and Aims
High-power-short-duration (HPSD) ablation is an effective treatment for atrial fibrillation but poses risks of thermal injuries to the oesophagus and vagus nerve. This study investigates incidence and predictors of thermal injuries, employing machine learning.
Methods
A prospective observational study was conducted at Leipzig Heart Centre, Germany, excluding patients with multiple prior ablations. All patients received Ablation Index guided HPSD ablation and subsequent oesophagogastroduodenoscopy. A machine learning algorithm categorized ablation points by atrial location and analysed ablation data, including Ablation Index, focusing on the posterior wall. The study is registered in clinicaltrials.gov (NCT05709756).
Results
Between February 2021, and August 2023, 238 patients were enrolled, of whom 18 (7.6%; 9 oesophagus, 8 vagus nerve, 1 both) developed thermal injuries, including 8 oesophageal erythemata, two ulcers and no fistula. Higher mean force (15.8±3.9g vs. 13.6±3.9g, p=0.022), ablation point quantity (61.50±20.45 vs. 48.16±19.60, p=0.007), total and maximum Ablation Index (24114±8765 vs. 18894±7863, p=0.008; 499±95 vs. 473±44, p=0.04, respectively) at the posterior wall, but not oesophagus location, correlated significantly with thermal injury occurrence. Patients with thermal injuries had significantly lower distances between left atrium and oesophagus (3.0±1.5mm vs 4.4±2.1mm, p=0.012) and smaller atrial surface areas (24.9±6.5 cm2 vs. 29.5±7.5cm2, p=0.032).
Conclusion
The low thermal lesion’s rate (7.6%) during Ablation Index guided HPSD ablation for atrial fibrillation is noteworthy. Machine learning based ablation data analysis identified several potential predictors of thermal injuries. The correlation between machine learning output and injury development suggests the potential for a clinical tool to enhance procedural safety.