Українська  English  Русский  

DOI: 10.31071/kit2015.12.08

Inventory reference ISSN 1812-7231 Klin.inform.telemed. Volume 11, Issue 12, 2015, Pages 57–62

Author(s) M. S. Abramovich1, Е. S. Atroschenko2

Institution(s) 1Research Institute for Applied Problems of Mathematics and Informatics of the Belarusian State University, Minsk, Belarus 2Republican Scientific and Practical center of Cardiology, Minsk, Belarus

Article title On prediction of chronic heart failure therapy effectiveness

Abstract (resume)

Introduction. A correctly formed training sample and a set of informative features are required to predict effectiveness of chronic heart failure (CHF) therapy using statistical methods of classification. An actual problem is to study prediction effectiveness using different methods of intellectual data analysis.

The study objective. The goal of the research is to develop forecasts (predictors) for effectiveness of CHF therapy using statistical methods of classification on the basis of informative features measured before the therapy.

Study results. An approach based on comparison of the expert and statistical classifications is considered to estimate the quality of the training sample formation. An informative set of instrumental and clinical features is constructed to characterize the state of patients with CHF: the 6 minute walking test, the middle arterial pressure in pulmonary artery, the ejection fraction of left ventricle, the number of heart beats per minute, integral parameter of the life quality. In case of non-normal distribution of informative features the use of robust discriminant analysis is proposed instead of classical discriminant analysis. Various classification methods to predict the effectiveness of CHF therapy are studied, such as discriminant analysis, robust discriminant analysis, support vector machine, decision trees and boosting decision trees.

Conclusion. The obtained prediction accuracy of the CHF therapy effectiveness is 80% for discriminant analysis, 82.1% for robust discriminant analysis, 81.1% for nonlinear SVM, 89.5% for decision trees and 95.4% for boosting decision trees.

Keywords chronic heart failure, training sample, k-means clustering, informative features, discriminant analysis, robust discriminant analysis, decision trees, support vector machine, boosting, therapy effectiveness


1. Atroshenko E.S. Khronicheskaya serdechnaya nedostatochnost’ [Chronic heart failure]. Мinsk, Belprint, 2011, 149 p. (In Russ.)

2. Atroshenko E.S., Abramovich M.S. Prognozirovaniye effektivnosti kompleksnoy terapii bol'nykh s khronicheskoy serdechnoy nedostatochnost'yu [Predicting the effectiveness of complex therapy of patients with chronic heart failure]. Kardiologiya v Belarusi [Cardiology in Belarus], 2013, iss. 6 (31), pp. 23–30. (In Russ.)

3. Afifi A.A., Azen S.P. Statistical analysis: A Computer Oriented Approach, 2nd ed. New York, Academic Press, 1979, 442 p.

4. Duda R.O. Pattern classification, 2nd ed. New York, Wiley-Interscience, 2000, 456 р.

5. Kharin Yu.S. Robustness in Statistical Pattern Recognition. Dordrecht, Kluwer Academic Publishers, 1996, 302 p.

6. Harrington P. Machine Learning in Action. New York, Manning, 2012, 382 p.

7. Hastie T., Tibshirani R., Friedman J. H. The Elements of Statistical Learning, 2nded. New York, Springer, 2009, 764 p.

Full-text version