Impact of computational modeling on transcatheter left atrial appendage closure efficiency and outcomes: the PREDICT-LAA trial

Selected in JACC: Cardiovascular Interventions by L. Biasco , F. Giacobbe

The results of PREDICT-LAA remain fascinating and point focus on a current topic such as the use of AI models in planning structural cardiac interventions.

Reviewers

Luigi Biasco

@BiascoDr

Interventional cardiologist / Cardiologist

OSPEDALE CIRIÈ - Ciriè, Italy

Federico Giacobbe

@fedegcb

Interventional cardiologist / Cardiologist

Azienda Ospedaliero-Universitaria Città Della Salute E Della Scienza Di Torino - Metropolitan City of Turin, Italy

Our Comment

Why this study – the rationale/objective? 

Artificial intelligence (AI) algorithms are particularly well suited to manage big data, and can be used to tailor interventional strategies for each patient with a level of accuracy never seen before. This is particularly relevant in the field of structural interventional cardiology, where pre-procedural planning is crucial to achieve optimal outcomes. Left atrial appendage occlusion (LAAO) is a relatively recent procedure which integrates different imaging techniques in the device selection process to reduce the risk of residual leaks and thus increase procedural success and efficacy.

CT-based computational modeling may help in personalizing the choice of a LAA closure device so as to optimize safety and efficacy of procedure. Nevertheless, the real advantage given by these new tools remains a matter of debate, since their efficacy and safety has not yet been proven, as well as the cost-benefit ratio remains to be demonstrated.

How was it executed? - the methodology

PREDICT-LAA was a prospective, multicenter, randomized, open-label clinical trial which enrolled 200 patients among 10 participating centers from 2020 to 2022.

Pre-procedural planning was performed with standard CT based LAAO planning vs CT+ three-dimensional simulation planning with FEops HEARTguide, an artificial intelligence supported three-dimensional reconstruction software dedicated to LAAO.

Computational CT based AI derived models are able to provide three dimensional reconstructions of heart and foresee the mechanical interactions between the peculiar, patient’s specific left atrial appendage anatomy and different LAAO device sizes. This in fact enables an accurate pre-procedural choice of the optimal device, along with a precise planning of implantation depth, potentially reducing procedural time and number of implant attempts and procedural success rates.

• Inclusion criteria: patients with indication to LAAO undergoing Amplatzer Amulet device implantation.
• Exclusion criteria: reduced renal function, iodine contrast allergy or any other condition that prohibited cardiac CT imaging.
• Treatment: LAAO pre-procedural planning with CT-based FEops HEARTguide simulations.
• Control: standard LAAO planning with CT. with or without Transesophageal echocardiogram (TEE).
• All patients underwent a pre-procedural CT and were planned for a postprocedural cardiac CT scan at 90-30 days after LAA closure to assess presence of peri device leaks and device-related thrombosis (DRT).
• 29 patients of the control arm and 31 of the experimental arm received TEE in addition to pre-procedural CT according to enrolling site protocols.
• Primary endpoint:
  • composite of incomplete LAA closure with residual III-IV grade contrast leakage into the LAA (distal of the Amulet lobe) and/or
  • Device related thrombosis assessed at post procedural CT scan.
• Secondary endpoints:
  • individual components of the primary endpoint,
  • achievement of complete LAA closure (CT assessed and defined as no leak, full coverage of all LAA trabeculations and no disk retraction into the LAA),
  • procedural efficacy (number of devices used, number of repositions, total procedure time, radiation time, dye volume),
  • procedural safety,
  • clinical endpoints at follow-up (thromboembolisms, major bleedings and/or mortality).
  • Trial was sponsored by Abbott and FEops NV.

What is the main result?

Among the 200 patients enrolled, 100 were randomized to CT+AI simulation planning, and 100 to standard planning. 3 randomized patients did not undergo LAAO while post-procedural CT scan was available in 181.

No statistically significant difference for the rate of the primary endpoint was evident, although a trend in favor of AI based simulation planning was evident (rate of primary endpoint: 41.8 % in the standard treatment group and 28.9 % in the CT + simulation group; RR: 0.69; 95 % CI: 0.46 to 1.04; p = 0.08).

Notably, significance was reached for the secondary endpoints assessing procedural safety and efficacy.

As a first, LAAO planning with CT computational models demonstrated a significant benefit in achieving complete LAA occlusion without residual leak or disc retraction (61.1 % in the AI based simulation group vs 44.0 % CT standard planning, p = 0.03).

Furthermore, the use of AI models led to a reduction of the number of devices used and repositioned as well as the number of disc retractions. Notably, single device-deployment was achieved in 58.0 % of experimental group as compared to 29.9 % of control group (p < 0.001).

Finally, a significant reduction of both radiation time and dye volume was noted in the experimental arm (< 0.001 and p = 0.02 respectively).

Owing to the low number of patients enrolled and consequent events observed at follow up, clinical endpoints were not evaluated. Nonetheless, procedural results are overwhelming, especially if compared to those reported in two the recent trials such as SWISS-APERO and AMULET IDE trials^1,2.

Critical reading and the relevance for clinical practice

The results of PREDICT-LAA remain fascinating and point focus on a current topic such as the use of AI models in planning structural cardiac interventions.

The hypothesis generated by this study goes beyond and above the lack of significance observed by the primary endpoint, defined as the reduction of III-IV grade contrast leakage and/or device thrombosis.
In this study, CT computational models demonstrated a relevant improvement in both procedural safety and efficacy. As a matter of fact, the definition of this primary endpoint is controversial as we know from the most recent data that even minor peridevice leaks may be associated with worse outcomes^3,4. Therefore, much more value should be placed on the higher percentage of complete closures obtained through FEops simulation models rather than only high-grade leaks.

The computational models were able to reduce procedural time, dye volume, and number of devices used, disc retractions, and numerically reduce the number of DRTs (p = 0.21)

Clearly, several considerations on PREDICT-LAA have to be done since it presents many limitations. At first, the study tested only the Amulet device: this design may have limited the potential of AI-enabled computational models which could be useful precisely in choosing the most appropriate type of device depending on anatomical model of LAA. Then, there was no standardization of pre-procedural planning in control arm as it was defined on the basis of the standard patient selection protocol of each enrolling center. Finally, the sample size was relatively small, thereby underpowered to assess any potential impact on clinical outcomes.

Despite these limitations, PREDICT-LAA is a trial of pivotal importance suggesting the new potentials of artificial intelligence in planning structural interventions. Whether increasingly accessible and affordable software will provide a wider availability of this tools and improve the cost-benefit ratio remains to be defined.

References

1. Galea R, De Marco F, Meneveau N, Aminian A, Anselme F, Gräni C, et al. Amulet or Watchman Device for Percutaneous Left Atrial Appendage Closure: Primary Results of the SWISS-APERO Randomized Clinical Trial. Circulation. 2022 Mar 8;145(10):724–38.
2. Lakkireddy D, Thaler D, Ellis CR, Swarup V, Sondergaard L, Carroll J, et al. Amplatzer Amulet Left Atrial Appendage Occluder Versus Watchman Device for Stroke Prophylaxis (Amulet IDE): A Randomized, Controlled Trial. Circulation. 2021 Nov 9;144(19):1543–52.
3. Alkhouli M, Du C, Killu A, Simard T, Noseworthy PA, Friedman PA, et al. Clinical Impact of Residual Leaks Following Left Atrial Appendage Occlusion: Insights From the NCDR LAAO Registry. JACC: Clinical Electrophysiology. 2022 Jun 1;8(6):766–78.
4. Dukkipati SR, Holmes DR, Doshi SK, Kar S, Singh SM, Gibson D, et al. Impact of Peridevice Leak on 5-Year Outcomes After Left Atrial Appendage Closure. Journal of the American College of Cardiology. 2022 Aug 2;80(5):469–83.