Can machine learning predict successful chronic total occlusion crossing with primary antegrade wiring?

Chronic total occlusions (CTOs) are found in about 25% of patients undergoing diagnostic angiography and are defined as 100% obstruction of a coronary artery of ≥3 months duration (1). Histological examination of CTO lesions has shown that they commonly consist of calcium, lipids, smooth muscle cells, an extracellular matrix, and neovascularization. CTO revascularization has been shown to significantly improve patients’ quality of life and reduce symptoms of angina. Although the success rate of percutaneous coronary intervention (PCI) for revascularizing CTOs is reported as fairly low (51–74%) up to 2009, recent technological advances and interventional strategies have improved the success rate of PCI of CTO. Antegrade wiring (AW) is the most commonly used strategy for wiring CTO lesions, however determining the suitability of specific lesions for CTO-PCI as well as determining the appropriate approach remains a challenge. While specific algorithms have been developed based upon angiographic features, limited data underlies their use. Machine learning (ML) encompasses several different algorithmic models and statistical methods to solve problems without specialized programming. Although a lot of importance has been given to artificial intelligence and ML in cardiology (2,3), their use in CTO-PCI has not been elucidated so far. The use of ML models may be helpful in determining the suitability of specific lesions for AW approaches.

In the article entitled “Predicting successful chronic total occlusion crossing with primary antegrade wiring using machine learning”, published on JACC: Cardiovascular Interventions, the authors analyzed the data of CTO-PCI using primary AW. The objective of this study was to develop and validate the ML model with high predictive capacity for successful primary AW in CTO-PCI (4). The authors utilized the data from 12,136 primary AW cases performed between 2012 and 2023 at 48 centers in the PROGRESS CTO registry (Prospective Global Registry for the Study of Chronic Total Occlusion Intervention; NCT02061436) to create five ML models. Fourteen lesion characteristics were used for the prediction of successful CTO crossing using primary AW. They were as follows; blunt/no stump, occlusion length, vessel diameter, aorto-ostial lesion, proximal cap ambiguity, in-stent restenosis, side branch at proximal cap, bifurcation at distal cap, poor distal landing zone, interventional collateral channels, calcification, tortuosity, target vessel, and prior attempt to open CTO. Hyperparameter tuning was conducted for the model with the best performance, and the SHAP (SHapley Additive exPlanations) explainer was performed to estimate feature importance. Primary AW was successful in 6,965 (57.4%) among 12,136 CTO-PCIs in their registry. Extreme gradient boosting was the best performing ML model with an average area under the receiver operating characteristic (ROC) curve of 0.775 (±0.010). After hyperparameter tuning, the average area under the ROC curve of the extreme gradient boosting model was 0.782 in the training set and 0.780 in the testing set. Among the examined parameters, occlusion length had the most significant impact on the prediction of the successful primary AW crossing followed by blunt/no stump, presence of interventional collaterals, vessel diameter, and proximal cap ambiguity, whereas the location of aorto-ostial lesion had the least impact on the outcome.

Although CTO-PCI techniques have advanced substantially, with retrograde approach and antegrade dissection re-entry (ADR) being performed by select operators, AW crossing remains the most common primary approach in CTO-PCI. Recent studies and consensus documents have proposed the parameters of successful AW in CTO lesions (5-11). From the PROGRESS CTO registry, Rempakos et al. showed that proximal cap ambiguity [odds ratio (OR): 0.52; 95% confidence interval (CI): 0.46–0.59], side branch at the proximal cap (OR: 0.85; 95% CI: 0.77–0.95), blunt/no stump (OR: 0.52; 95% CI: 0.47–0.59), longer lesion length [OR (per 10 mm increase): 0.79; 95% CI, 0.76–0.81], moderate to severe calcification (OR: 0.73; 95% CI: 0.66–0.81), moderate to severe proximal tortuosity (OR: 0.67; 95% CI: 0.59–0.75), bifurcation at the distal cap (OR: 0.66; 95% CI: 0.59–0.73), left anterior descending artery CTO [OR (vs. right coronary artery): 1.44; 95% CI: 1.28–1.62] and left circumflex CTO [OR (vs. right coronary artery): 1.22; 95% CI: 1.07–1.40], non-in-stent restenosis lesion (OR: 0.56; 95% CI: 0.49–0.65), and good distal landing zone (OR: 1.18; 95% CI: 1.06–1.32) were independently associated with primary AW crossing success, using multivariable logistic regression analysis (5). In the Japanese CTO-PCI expert registry, Tanaka et al. reported that reattempt CTO-PCI, long CTO length (≥20 mm), and no stump are significant predictors for antegrade guidewire failure in primary antegrade approach (6). From the data of RECHARGE (REgistry of CrossBoss and Hybrid procedures in FrAnce, the NetheRlands, BelGium and UnitEd Kingdom) registry, a primary AW strategy was not recommended when the CTO lesions have long lesion lengths, a blunt proximal stump, proximal cap ambiguity, and a diseased distal landing zone (7). Alessandrino et al. reported, using multivariable analysis, that previous coronary artery bypass graft surgery (OR: 2.49; 95% CI: 1.56 to 3.96), previous myocardial infarction (OR: 1.6; 95% CI: 1.17 to 2.2), severe lesion calcification (OR: 2.72; 95% CI: 1.78 to 4.16), longer CTOs (≥20 mm OR: 2.04; 95% CI: 1.54 to 2.7), non-left anterior descending coronary artery location (OR: 1.56; 95% CI: 1.14 to 2.15), and blunt stump cap (OR: 1.39; 95% CI: 1.05 to 1.81) were independently associated with CTO-PCI failure in the study where the majority of CTO-PCIs were performed via the antegrade approach (10). These previous reports support the validity of the 14 parameters proposed by Rempakos et al. in the ML model for predicting the successful AW in CTO-PCI (5-7,10).

To investigate the microstructure in CTO lesions, all clinical studies have a crucial limitation that the resolution of current imaging modalities such as intravascular ultrasound, optical frequency domain imaging, and coronary computed tomography angiography are not as precise as histology. Previous pathological studies analyzing CTO lesions lend support for the use of this ML model (12-14) as predictive for CTO-PCI success. From the pathological point of view, CTO lesions are classified into the short-duration CTO (SD-CTO) lesions or the long-duration CTO (LD-CTO) lesions based on the differences of structural constituents. The short- or long-duration refers to chronological duration, not lesion length. An SD-CTO is defined as a lesion including principally proteoglycans, which are shown by the light green or blue staining in Movat Pentachrome (MP), with fibrin, whereas an LD-CTO is characterized as a lesion containing type 1 collagen, shown by the green or gray staining in MP, without fibrin (12,13). Representative figures of SD- and LD-CTO lesions stained by MP are shown in Figure 1. Figure 1 shows that the LD-CTO has more calcification than the SD-CTO. Previous studies have shown that LD-CTOs had longer lesions, more blunt/no stump lesions, more calcification, smaller vessel diameter, and more proximal cap ambiguity, compared to SD-CTOs (12,13). Sakakura et al. showed that in LD-CTO lesions, the prevalence of the blunt type at the CTO proximal cap was significantly higher than that at the CTO distal cap (P=0.02), on the other hand, no significant difference was observed between the CTO proximal and distal caps in SD-CTO lesions (P=0.41) (12), which means that SD-CTOs have more tapered proximal caps, which can be associated with easier AW in CTO lesions. They also showed that SD-CTOs had greater necrotic core, compared to LD-CTO [18.6% (6.4–48.0%) vs. 7.8% (0–15.0%), P<0.05], which means soft plaque, leading to successful AW. In the analysis of 324 CTO sections, the LD-CTOs presented with smaller vessel diameter (negative remodeling) [2.81 (2.21–3.44) vs. 3.68 (3.04–4.08) mm, P<0.001], more calcification (77% vs. 62%, P=0.011), and decreased microchannels [0 (0–1) vs. 1 (0–1), P<0.001], compared to SD-CTOs (13). These data from the pathological studies support the validity of the ML system and its criteria for the prediction of successful AW in the paper.

Figure 1 Representative pathological images of CTO sections. (A) A SD-CTO is defined as a lesion including principally proteoglycan, which is shown by the green or blue staining in MP, and fibrin. (B) A LD-CTO is defined as a lesion consisting of type 1 collagen, shown by the green or gray stain in MP. LD-CTO (B) has more calcification, compared to SD-CTO (A). CTO, chronic total occlusion; Ca, calcification; SD-CTO, short-duration CTO; MP, Movat Pentachrome; LD-CTO, long-duration CTO.

Generally, ML models have a potential risk of increased computational time, overfitting, and reduced interpretability (15). To address these problems, the authors utilize the SHAP method to detect the effect that the predictors had on successful primary AW crossing in CTO-PCI. However, the generalizability of this ML system in CTO-PCI practice may be limited because the ML model in this paper is based upon observational registries and the operators in the PROGRESS-CTO registry are CTO PCI experts, and CTO lesions vary greatly in complexity. Nonetheless, the purpose of ML in this paper is to identify lesions which are easier to cross by AW within the data set from the more typical CTO-PCI operators. The features of CTO lesion contributing to their algorithm including occlusion length, the presence of stump, and vessel diameter are all identifiable by non-experts and not unique to CTO-PCI operators. This can make the algorithm of ML system more practical and utilizable for non-CTO experts. The prediction probability of AW success using the ML model can still be useful to evaluate and determine the primary CTO crossing strategy before attempting the CTO-PCI procedure. For example, the ML system may make it possible to do real-time decision-making before or during CTO-PCI. When the ML system indicates the highest chance of antegrade crossing success, operators can firstly choose AW strategy rather than other strategies such as retrograde wiring and ADR, which may decrease the procedural time. Moreover, in cases where the AW strategy is not working well during the procedure, other alternative strategies might be attempted earlier such as retrograde wiring based on the ML model prediction of occlusion length and presence of interventional collaterals channels, etc., saving radiation/contrast exposure and perhaps lowering procedural complications. The next steps would involve putting the ML algorithm to work in the real world to determine how its use might improve the safety and success of CTO-PCI procedures.

Acknowledgments

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-24-509/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://cdt.amegroups.com/article/view/10.21037/cdt-24-509/coif). R.V. and A.V.F. have received institutional research support from NIH (No. HL141425), Leducq Foundation Grant, 480 Biomedical, 4C Medical, 4Tech, Abbott, Accumedical, Amgen, Biosensors, Boston Scientific, Cagent Vascular, Cardiac Implants, Celonova, Claret Medical, Concept Medical, Cook, CSI, DuNing Inc., Edwards LifeSciences, Emboline, Endotronix, Envision Scientific, Lutonix/Bard, Gateway, Lifetech, Limflo, MedAlliance, Medtronic, Mercator, Merill, Microport Medical, Microvention, Mitraalign, Mitra Assist, NAMSA, Nanova, Neovasc, NIPRO, Novogate, Occulotech, OrbusNeich Medical, Phenox, Profusa, Protembis, Qool, Recor, Senseonics, Shockwave, Sinomed, Spectranetics, Surmodics, Symic, Vesper, W.L. Gore, and Xeltis. A.V.F. has received honoraria from Abbott Vascular, Biosensors, Boston Scientific, Celonova, Cook Medical, CSI, Lutonix Bard, Sinomed, and Terumo Corporation; and is a consultant to Amgen, Abbott Vascular, Boston Scientific, Celonova, Cook Medical, Lutonix Bard, and Sinomed. R.V. has received honoraria from Abbott Vascular, Biosensors, Boston Scientific, Celonova, Cook Medical, Cordis, CSI, Lutonix Bard, Medtronic, OrbusNeich Medical, Celonova, SINO Medical Technology, ReCor, Terumo Corporation, W. L. Gore, and Spectranetics; and is a consultant for Abbott Vascular, Boston Scientific, Celonova, Cook Medical, Cordis, CSI, Edwards Lifesciences, Lutonix Bard, Medtronic, OrbusNeich Medical, ReCor, Sinomedical Technology, Spectranetics, Surmodics, Terumo Corporation, W. L. Gore, and Xeltis. The other authors have 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. 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]
2. Lüscher TF, Wenzl FA, D'Ascenzo F, et al. Artificial intelligence in cardiovascular medicine: clinical applications. Eur Heart J 2024;45:4291-304. [Crossref] [PubMed]
3. Madaudo C, Parlati ALM, Di Lisi D, et al. Artificial intelligence in cardiology: a peek at the future and the role of ChatGPT in cardiology practice. J Cardiovasc Med (Hagerstown) 2024;25:766-71. [Crossref] [PubMed]
4. 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]
5. Rempakos A, Alexandrou M, Mutlu D, et al. Predictors of successful primary antegrade wiring in chronic total occlusion percutaneous coronary intervention. J Invasive Cardiol 2024; [Crossref] [PubMed]
6. Tanaka H, Tsuchikane E, Muramatsu T, et al. A Novel Algorithm for Treating Chronic Total Coronary Artery Occlusion. J Am Coll Cardiol 2019;74:2392-404. [Crossref] [PubMed]
7. Maeremans J, Walsh S, Knaapen P, et al. The Hybrid Algorithm for Treating Chronic Total Occlusions in Europe: The RECHARGE Registry. J Am Coll Cardiol 2016;68:1958-70. [Crossref] [PubMed]
8. 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]
9. Galassi AR, Werner GS, Boukhris M, et al. Percutaneous recanalisation of chronic total occlusions: 2019 consensus document from the EuroCTO Club. EuroIntervention 2019;15:198-208. [Crossref] [PubMed]
10. Alessandrino G, Chevalier B, Lefèvre T, et al. A Clinical and Angiographic Scoring System to Predict the Probability of Successful First-Attempt Percutaneous Coronary Intervention in Patients With Total Chronic Coronary Occlusion. JACC Cardiovasc Interv 2015;8:1540-8. [Crossref] [PubMed]
11. 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]
12. Sakakura K, Nakano M, Otsuka F, et al. Comparison of pathology of chronic total occlusion with and without coronary artery bypass graft. Eur Heart J 2014;35:1683-93. [Crossref] [PubMed]
13. Konishi T, Kawakami R, Vozenilek AE, et al. Mechanisms of Medial Wall Thinning in Chronic Total Occlusion. JACC Cardiovasc Interv 2024;17:1719-28. [Crossref] [PubMed]
14. Mori H, Lutter C, Yahagi K, et al. Pathology of Chronic Total Occlusion in Bare-Metal Versus Drug-Eluting Stents: Implications for Revascularization. JACC Cardiovasc Interv 2017;10:367-78. [Crossref] [PubMed]
15. 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]