Machine learning in the problem of adverse cardiovascular events prognosis in patients after coronary artery bypass surgery

• Ovcharenko E.A.
• Klyshnikov K.Yu.
• Kutikhin A.G.
• Frolov A.V.

Abstract

Background. Despite existing prediction algorithms to stratify the risk of adverse cardiovascular complications in the postoperative period of cardiac surgery, there is still no unity in understanding, which is the most accurate.

Aim is to explore advanced machine learning algorithms as a predictive tool for assessing the risk of developing adverse cardiovascular events that occur in the long-term period after coronary artery bypass surgery.

Material and methods. Based on retrospective analysis of 152 case histories of patients operated on in 2004–2006 years for coronary artery disease, 59 factors of pre-, intra- and early postoperative periods were identified and we studied algorithms or machine learning models for predicting long-term adverse cardiovascular events (stroke, myocardial infarction in combination with its subsequent revascularization, death and combined endpoint). We analyzed algorithms were such as logistic regression, decision trees, random forest method, super-random decision trees, gradient boosting, multilayer neural networks, K-means neighbor’s method and combined algorithms.

Results. The area under the ROC-curve, as an indicator of the effectiveness of the algorithms of various types used, varied significantly. Multivariate logistic regression reached 0.40–0.56, while K-means neighbors, multilayer neural networks and single decision trees had 0.74–0.75. However, the largest area was demonstrated by advanced variants of models based on a combination of decision trees that use gradient boosting, and especially the combined algorithm, the area under the ROC curve for which reached 0.77–0.91. For single models with maximum area, 4 predictors (interventricular septal thickness, left ventricle ejection fraction, body mass index, and age) were also selected and quantified as contributing the most to their performance.

Conclusion. The most effective and accurate models for predicting distant adverse events are combined models that use several sub-algorithms, mainly based on a combination of decision trees. When analyzing single algorithms, the interventricular septal thickness factor was identified as the most significant predictor of prognosis.

Keywords:coronary artery bypass grafting; machine learning; prediction; cardiovascular events

References

1.     Bokeriya L.A., Milievskaya E.B., Kudzoeva Z.F., Pryanishnikov V.V., Skopin A.I., Yurlov I.A. Cardiovascular surgery. Diseases and congenital abnormalities of the circulatory system. Moscow: FGBU «NMITsSSKh im. A.N. Bakuleva» MZ RF. 2019: 270 p. (in Russian)

2.     Nashef S.A.M., Roques F., Sharples L.D., Nilsson J., Smith C., Goldstone A.R., et al. EuroSCORE II. Eur J Cardiothorac Surg. 2012; 41 (4): 734–45. DOI: https://doi.org/10.1093/ejcts/ezs043  

3.     O’Brien S.M., Feng L., He X., Xian Y., Jacobs J.P., Badhwar V., et al. The Society of Thoracic Surgeons 2018 adult cardiac surgery risk models: part 2 – statistical methods and results. Ann Thorac Surg. 2018; 105 (5): 1419–28. DOI: https://doi.org/10.1016/j.athoracsur.2018.03.003  

4.     Xie W., Li D., Shi Y., Yu N., Yan Y., Zhang Y., et al. Serum FGF21 levels predict the MACE in patients with myocardial infarction after coronary artery bypass graft surgery. Front Cardiovasc Med. 2022; 9: 850527. DOI: https://doi.org/10.3389/fcvm.2022.850517  

5.     Kalyoncuoglu M., Ozturk S., Sahin M. Does CHA2DS2-VASc score predict MACE in patients undergoing isolated coronary artery bypass grafting surgery? Braz J Cardiovasc Surg. 2019; 34 (5): 542–9. DOI: https://doi.org/10.21470/1678-9741-2018-0323  

6.     Hung D.Q., Minh N.T., Vo H.-L., Hien N.S., Tuan N.Q. Impact of pre-, intra-and post-operative parameters on in-hospital mortality in patients undergoing emergency coronary artery bypass grafting: a scarce single-center experience in resource-scare setting. Vasc Health Risk Manag. 2021; 17: 211–26. DOI: https://doi.org/10.2147/VHRM.S303726   

7.     Shawon M.S.R., Odutola M., Falster M.O., Jorm L.R. Patient and hospital factors associated with 30-day readmissions after coronary artery bypass graft (CABG) surgery: a systematic review and meta-analysis. J Cardiothorac Surg. 2021; 16 (1): 172. DOI: https://doi.org/10.1186/s13019-021-01556-1  

8.     Benuzillo J., Caine W., Evans R.S., Roberts C., Lappe D., Doty J. Predicting readmission risk shortly after admission for CABG surgery. J Card Surg. 2018; 33 (4): 163–70. DOI: https://doi.org/10.1111/jocs.13565  

9.     Gatti G., Rochon M., Raja S.G., Luzzati R., Dreas L., Pappalardo A. Predictive models of surgical site infections after coronary surgery: insights from a validation study on 7090 consecutive patients. J Hosp Infect. 2019; 102 (3): 277–86. DOI: https://doi.org/10.1016/j.jhin.2019.01.009  

10. Forte J.C., Wiering M.A., Bouma H.R., Geus F., Epema A.H. Predicting long-term mortality with first week post-operative data after Coronary Artery Bypass Grafting using Machine Learning models. In: F. Doshi-Velez, J. Fackler, D. Kale, R. Ranganath, B. Wallace, J. Wiens (eds). Proceedings of the 2^nd Machine Learning for Healthcare Conference. PMLR, 2017: 39–58.

11. Gel’tser B.I., Shakhgel’dyan K.I., Rublev V.Yu., Kotel’nikov V.N., Kriger A.B., Shirobokov V.G. Machine learning methods in predicting hospital deaths in patients with coronary heart disease after coronary bypass surgery. Kardiologiya [Cardiology]. 2020; 60 (10): 38–46. DOI: https://doi.org/0.18087/cardio.2020.10.n1170  (in Russian)

12. Castela Forte J., Mungroop H.E., de Geus F., van der Grinten M.L., Bouma H.R., Pettilä V., et al. Ensemble machine learning prediction and variable importance analysis of 5-year mortality after cardiac valve and CABG operations. Sci Rep. 2021; 11 (1): 3467. DOI: https://doi.org/10.1038/s41598-021-82403-0  

13. Hancock J.T., Khoshgoftaar T.M. CatBoost for big data: an interdisciplinary review. J Big Data. 2020; 7 (1): 94. DOI: https://doi.org/10.1186/s40537-020-00369-8  

All articles in our journal are distributed under the Creative Commons Attribution 4.0 International License (CC BY 4.0 license)