Machine learning paving the way for successful antegrade crossing of total chronic coronary occlusions

In this issue of the journal, Rempakos et al. (1) reported on a machine learning (ML) model which can be used for predicting the success of antegrade recanalization with coronary chronic total occlusive (CTO) lesions. ML, a subset of artificial intelligence (AI), is revolutionizing healthcare by leveraging the analysis of large clinical datasets. Through this analysis, patterns can be identified, and these learned patterns can subsequently support risk prediction in patients and inform clinical decision-making. This is particularly significant in the clinical context, where the volume of available clinical data now may significantly exceed the processing capacity of the human brain. The integration of ML into healthcare systems, therefore, holds immense potential to enhance the precision and efficiency of medical decisions, ultimately improving patient outcomes (2).

Despite advances with percutaneous interventions and pharmacologic therapy, deaths attributable to cardiac diseases are steadily increasing during the last decade (3). Coronary artery disease (CAD) represents the leading cause of cardiovascular death in the Western world in this context (3). The presence of CTO lesions due to atherosclerotic disease is quite common based on contemporary studies. Thus, between 15–20% of patients with CAD exhibit at least one CTO lesion during cardiac catheterization (4,5). A CTO lesion is defined as coronary artery occlusion with thrombolysis in myocardial infarction (TIMI) 0 antegrade flow and probable or definite duration of more than 3 months, based on the presence of mature collaterals and the absence of thrombus or contrast agent staining at the proximal cap (6). Importantly, the interventional treatment of CTO lesions is technically more challenging than in non-occlusive lesions, requiring interdisciplinary cooperation between experts from different disciplines such as interventional cardiology, cardiac surgery and cardiovascular imaging (7).

In this regard, the development of non-invasive imaging techniques like coronary computed tomography angiography (CCTA) during the last years may aid the pre-interventional planning of CTO lesions and can be used to guide such complex procedures (8). This together with advanced percutaneous coronary intervention (PCI) equipment, including dedicated guidewires, catheter extensions, and microcatheters contributed to an increase in the success rate of CTO PCI, which varies between 80% and 90% with experienced operators (9). In addition, the rate of complications during CTO PCI, which has been described between 1% and 3%, although being higher than with normal coronary interventions, has reached a clinically acceptable level. This is attributed to the expertise of the operators and the presence of dedicated materials for complication management (10,11).

Several crossing algorithms are currently available, suggesting four different CTO crossing strategies, including (I) antegrade wiring (AW); (II) antegrade dissection and re-entry; (III) retrograde wiring; or (IV) retrograde dissection and re-entry along with an array of available techniques and equipment to facilitate crossing of the CTO (9). With all these different strategies however, the importance of dual angiography and careful angiographic review to guide the selection of initial and subsequent crossing strategies is crucial (12). AW, including microcatheter back-up and wire escalation to penetrate the proximal cap of the CTO, is the most frequently used strategy to cross the CTO lesion, since this strategy exhibits relatively high success and low complication rates (13). Therefore, AW is widely accepted as the preferred strategy to begin with crossing of CTO lesions (9,10,12). With some specific lesions characteristics however, including proximal cap ambiguity, long occlusion length, and severe calcification a limited antegrade approach followed by a prompt change to a retrograde strategy or even a primary retrograde strategy is recommended (12). Such hybrid or bidirectional algorithms considering a prompt change of strategy are essential to achieve progress in case of antegrade recanalization failure in the interest of radiation exposure, contrast administration, and procedure time. From a pathophysiological perspective, CTO lesions may be easily crossed from retrograde since the distal part of the CTO usually has a softer, thinner, and less calcified cap (14-16).

In the present retrospective multicenter trial by Rempakos et al. (1), the authors developed several ML models incorporating 14 features, many of which had been partially but not fully explored in previous studies on CTO success. Using data from 12,136 primary cases, performed at 48 centers in the PROGRESS CTO registry, the results of the trial indicate that occlusion length, followed by the presence of a blunt or no stump and of interventional collaterals, vessel diameter, and proximal cap ambiguity, significantly impact the success of antegrade guidewire crossing followed by successful CTO recanalization. The models used, demonstrated an acceptable predictive capacity for successful primary AW crossing in CTO PCI. After optimizing the hyperparameters, the final XGBoost model exhibited an acceptable discriminative ability for outcome prediction, with an area under the curve (AUC) of 0.780 in the testing set. The study employed a cohort distribution ratio of 80:20 for training and testing, ensuring robust model validation.

Thus, the present study provides preliminary evidence that supports the pre-procedural decision to choose AW in CTO recanalization. Although randomized data are lacking, it is recommended to select an initial strategy and establish a rank-order hierarchy based on CTO characteristics to achieve higher success rates, fewer complications, reduced contrast media use, lower radiation exposure, and shorter procedure times (17,18). This finding underscores the importance of informed decision-making prior to intervention to maximize patient outcomes. Moreover, the study combines the latest interventional methods with cutting-edge technologies, including AI, to enhance the precision and effectiveness of CTO PCI.

Traditionally, decision-making in studies and clinical practice has relied on classic flowcharts, where decisions are made sequentially based on a limited set of parameters. However, the study by Rempakos et al. (1) marks a significant shift, by integrating ML into this process. While the parameters used in their model are drawn from previous studies, the ML approach offers a distinct advantage by creating neural networks that interconnect these parameters. This allows for dynamic weighting and adjustment of variables based on individual patient characteristics, far surpassing the linear approach of traditional methods (19). Furthermore, the inclusion of 14 variables highlights the potential of ML, given that the human brain is generally limited to effectively managing only 5 to 8 pieces of information (19,20).

In addition, ML models have a significant edge over traditional regression models due to their ability to handle nonlinear variables, better account for interactions between variables, and manage a larger number of predictors. This capacity ultimately enhances predictive accuracy, providing a more sophisticated and reliable tool for clinical decision-making (21). There are already several exciting ML models that demonstrate the potential for clinical prediction. For example, ML has been used to predict aortic valve stenosis from ECG data, achieving an adjusted AUC of up to 0.9 when accounting for risk factors (22). Similarly, ML has been used to predict left ventricular ejection fraction (LVEF) in patients with dyspnea from electrocardiogram (ECG) data more accurately than using N-terminal pro-B-type natriuretic peptide (NT-proBNP) levels (23). These examples underscore the high potential of ML in medical science and daily practice.

The study by Rempakos et al. (1) now highlights the potential of ML in interventional cardiology. However, while the model demonstrated an acceptable discriminative ability with its AUC, the 20% testing cohort relative to the training cohort is notably small compared to other ML studies. To improve outcomes, future research should focus on developing ML models with better AUCs, consider integrating a validation cohort between the training and testing phases, and ultimately validate these models in prospective studies.

Importantly, the greatest limiting factor for predicting AW success is not the ML model itself, but rather that the 14 variables used are derived from the subjective analysis of angiograms by investigators. A more desirable approach would involve AI-driven analysis of angiograms, like the highly accurate AI applications in computed tomography analysis of coronary arteries and plaque burden, which could then determine the relevant variables (24). In such models, the optimal CTO strategy could be automatically suggested, further enhancing clinical decision-making.

Another important limitation of the study is that it included only patients for whom AW was the initial strategy, and it only predicted the success of this approach. It did not assess whether patients were candidates for the retrograde technique. Therefore, if the algorithm predicts a low likelihood of success with AW, it does not necessarily indicate that starting with this approach is incorrect, as the alternative retrograde strategy may be even less suitable for the patient’s anatomy. Finally, the results of the study may aid in enhancing CTO success to a higher extent in non-expert than in experienced CTO operators.

Despite these limitations, the study by Rempakos et al. (1) clearly demonstrates the potential of combining classical medical techniques with cutting-edge digital models to significantly enhance future patient care.

Acknowledgments

Funding: None.

Footnote

Provenance and Peer Review: This article was commissioned by the editorial office, Cardiovascular Diagnosis and Therapy. The article has undergone external peer review.

Peer Review File: Available at https://cdt.amegroups.com/article/view/10.21037/cdt-24-423/prf

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://cdt.amegroups.com/article/view/10.21037/cdt-24-423/coif). P.B. reports consulting activities and honoraria for lectures: AstraZeneca, Bayer Vital, Thieme, Diaplan, BNK (Bundesverband Niedergelassener Kardiologen), Bristol-Myers Squibb, Daiichi Sankyo, Novo Nordisk, Pfizer and Sanofi Aventis; grants from German Heart Foundation; and travel fees from Bayer Vital and Daiichi Sankyo. D.W. reports institutional grants (trial support) from Abiomed, and honoraria for lectures from AstraZeneca, Bayer Vital, Edwards, Medtronic, Meril and Novartis. The other author has no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Rempakos A, Alexandrou M, Mutlu D, et al. Predicting Successful Chronic Total Occlusion Crossing With Primary Antegrade Wiring Using Machine Learning. JACC Cardiovasc Interv 2024;17:1707-16. [Crossref] [PubMed]
2. Radke P. Künstliche Intelligenz in der kardiovaskulären Medizin-Status und Perspektiven. Aktuelle Kardiologie 2023;12:433-8. [Crossref]
3. Martin SS, Aday AW, Almarzooq ZI, et al. 2024 Heart Disease and Stroke Statistics: A Report of US and Global Data From the American Heart Association. Circulation 2024;149:e347-913. [Crossref] [PubMed]
4. Azzalini L, Jolicoeur EM, Pighi M, et al. Epidemiology, Management Strategies, and Outcomes of Patients With Chronic Total Coronary Occlusion. Am J Cardiol 2016;118:1128-35. [Crossref] [PubMed]
5. Fefer P, Knudtson ML, Cheema AN, et al. Current perspectives on coronary chronic total occlusions: the Canadian Multicenter Chronic Total Occlusions Registry. J Am Coll Cardiol 2012;59:991-7. [Crossref] [PubMed]
6. Ybarra LF, Rinfret S, Brilakis ES, et al. Definitions and Clinical Trial Design Principles for Coronary Artery Chronic Total Occlusion Therapies: CTO-ARC Consensus Recommendations. Circulation 2021;143:479-500. [Crossref] [PubMed]
7. Galassi AR, Vadalà G, Werner GS, et al. Evaluation and management of patients with coronary chronic total occlusions considered for revascularisation. A clinical consensus statement of the European Association of Percutaneous Cardiovascular Interventions (EAPCI) of the ESC, the European Association of Cardiovascular Imaging (EACVI) of the ESC, and the ESC Working Group on Cardiovascular Surgery. EuroIntervention 2024;20:e174-84. [Crossref] [PubMed]
8. Opolski MP, Achenbach S, Schuhbäck A, et al. Coronary computed tomographic prediction rule for time-efficient guidewire crossing through chronic total occlusion: insights from the CT-RECTOR multicenter registry (Computed Tomography Registry of Chronic Total Occlusion Revascularization). JACC Cardiovasc Interv 2015;8:257-67. [Crossref] [PubMed]
9. Brilakis ES, Mashayekhi K, Tsuchikane E, et al. Guiding Principles for Chronic Total Occlusion Percutaneous Coronary Intervention. Circulation 2019;140:420-33. [Crossref] [PubMed]
10. Brilakis ES, Banerjee S, Karmpaliotis D, et al. Procedural outcomes of chronic total occlusion percutaneous coronary intervention: a report from the NCDR (National Cardiovascular Data Registry). JACC Cardiovasc Interv 2015;8:245-53. [Crossref] [PubMed]
11. Tanaka H, Morino Y, Abe M, et al. Impact of J-CTO score on procedural outcome and target lesion revascularisation after percutaneous coronary intervention for chronic total occlusion: a substudy of the J-CTO Registry (Multicentre CTO Registry in Japan). EuroIntervention 2016;11:981-8. [Crossref] [PubMed]
12. Wu EB, Brilakis ES, Mashayekhi K, et al. Global Chronic Total Occlusion Crossing Algorithm: JACC State-of-the-Art Review. J Am Coll Cardiol 2021;78:840-53. [Crossref] [PubMed]
13. Kostantinis S, Simsek B, Karacsonyi J, et al. In-hospital outcomes and temporal trends of percutaneous coronary interventions for chronic total occlusion. EuroIntervention 2022;18:e929-32. [Crossref] [PubMed]
14. Thind A, Strauss B, Teitelbaum A, et al. A novel method for the measurement of proximal fibrous cap puncture force in chronic total occlusions: the effect of increasing age. EuroIntervention 2011;6:997-1002. [Crossref] [PubMed]
15. Korosoglou G, Schmidt A, Lichtenberg M, et al. Best crossing of peripheral chronic total occlusions. Vasa 2023;52:147-59. [Crossref] [PubMed]
16. Korosoglou G, Schmidt A, Lichtenberg M, et al. Crossing Algorithm for Infrainguinal Chronic Total Occlusions: An Interdisciplinary Expert Opinion Statement. JACC Cardiovasc Interv 2023;16:317-31. [Crossref] [PubMed]
17. Brilakis ES, Grantham JA, Rinfret S, et al. A percutaneous treatment algorithm for crossing coronary chronic total occlusions. JACC Cardiovasc Interv 2012;5:367-79. [Crossref] [PubMed]
18. Karatasakis A, Danek BA, Karmpaliotis D, et al. Approach to CTO Intervention: Overview of Techniques. Curr Treat Options Cardiovasc Med 2017;19:1. [Crossref] [PubMed]
19. Meder B, Duncker D, Helms TM, et al. eCardiology: ein strukturierter Ansatz zur Förderung der digitalen Transformation in der Kardiologie. Die Kardiologie 2023;17:12-26. [Crossref]
20. Miller BM, Moore DE Jr, Stead WW, et al. Beyond Flexner: a new model for continuous learning in the health professions. Acad Med 2010;85:266-72. [Crossref] [PubMed]
21. Goldstein BA, Navar AM, Carter RE. Moving beyond regression techniques in cardiovascular risk prediction: applying machine learning to address analytic challenges. Eur Heart J 2017;38:1805-14. [PubMed]
22. Cohen-Shelly M, Attia ZI, Friedman PA, et al. Electrocardiogram screening for aortic valve stenosis using artificial intelligence. Eur Heart J 2021;42:2885-96. [Crossref] [PubMed]
23. Adedinsewo D, Carter RE, Attia Z, et al. Artificial Intelligence-Enabled ECG Algorithm to Identify Patients With Left Ventricular Systolic Dysfunction Presenting to the Emergency Department With Dyspnea. Circ Arrhythm Electrophysiol 2020;13:e008437. [Crossref] [PubMed]
24. Weichsel L, André F, Renker M, et al. Effects of high- versus low-intensity lipid-lowering treatment in patients undergoing serial coronary computed tomography angiography: results of the multi-center LOCATE study. Clin Res Cardiol 2024; Epub ahead of print. [Crossref] [PubMed]