1Universitätsklinikum Schleswig-Holstein Innere Medizin III mit den Schwerpunkten Kardiologie, Angiologie und internistische Intensivmedizin Kiel, Deutschland; 2Universitätsklinikum Schleswig-Holstein, Campus Kiel, Klinik für Angeborene Herzfehler und Kinderkardiologie Kiel, Deutschland
Background
The integration of contact force (CF) technology in the ablation catheters and the development of Ablation Index (AI) allowed the prediction of the transmurality and durability of lesions. The LOCALIZE trial has shown that a local impedance (LI) drop-guided ablation of atrial fibrillation can be performed without calculation of any lesion indexes. Today, the next generation of ablation catheters combines CF technology and LI monitoring (Stablepoint, RHYTHMIA HDx™ with DIRECTSENSE™). As no proprietary lesion index is available for this platform, our study aims to compare AI-guided ablation and LI drop-guided ablation and to estimate whether the LI drop-guided ablation is sufficient to reach AI-target values.
Methods
1) As the AI Formula is unavailable, the direct calculation of AI for the ablations performed on non-CARTO® platforms is impossible. We built a machine learning model trained on the ablation data of eight patients undergoing PVI exported from the CARTO®3 mapping system into a Python programming environment. A total of 529,684 data points at different sampling times were used to train a Random Forest Regressor. During cross-validation on a subset of the original data (30% split), the model highly predicted the AI based on duration, power, and CF dynamics (R2 0.972, MAE 6.62).
2) We performed a retrospective analysis of atrial fibrillation ablation data of 27 patients treated using LI drop-guided ablation (RHYTHMIA™). Raw impedance data has been filtered by a moving mean filter with a window length of 1.5s, and the LI drop point/plateau has been defined as the impedance minimum (Fig. 1).
We chose a minimum AI-target of ≥400, which corresponds to transmural lesion in the atrial myocardium.
|
Table: Ablation parameters of the lesions | |||
|
|
AI-target ≥400 matched |
AI-target not matched |
p |
|
Number of lesions, n (%) |
880 (74) |
307 (26) |
- |
|
Time to LI drop plateau, s |
14.0 (10.4 - 19.8) |
7.21 (3.9 - 12.2) |
<.001 |
|
Time to AI ≥400, s |
6.58 (4.3 - 8.4) |
9.59 (7.5 - 15.4) |
<.001 |
|
AI at LI drop plateau |
512.6 (468.2 - 547.6) |
359.1 (309.5 - 383.6) |
<.001 |
|
Starting CF, g |
19.1 (12.9 - 27.2) |
10.4 (7.3 - 16.6) |
<.001 |
|
Mean CF, g |
21.0 (15.3 - 28.5) |
11.9 (7.7 - 18.1) |
<.001 |
|
LI drop, Ohm |
21.6 (16.3 - 28.6) |
15.35 (11.1 - 24.3) |
<.001 |
|
Starting LI, Ohm |
147.6 (136.2 - 159.0) |
141.07 (132.6 - 155.9) |
.001 |
|
Closest distance to neighbouring lesion, mm |
3.9 (2.9 - 4.8) |
3.6 (2.6- 4.7) |
.024
|
Results
The main results are presented in Table 1. AI-target was reached in 74% of points (Target-group). Ablation duration was shorter in the Target-group. The median ablation time to AI-target was 7.4s shorter than the time to the LI drop plateau for the Target-group. Starting LI and LI drop were of higher value in the matching group. Most importantly, the starting and mean CF during the ablation are shown to be of higher value in the Target-group.
Conclusions
LI drop-guided ablation achieved the AI-target of 400 in only 74% of the lesions. The primary factor influencing the attainment of the AI-target was the CF during ablation. However, further clinical studies are necessary to determine the comparative clinical effectiveness of LI drop-guided ablation versus AI-guided ablation.