Logit and Fuzzy Models in Data Analysis: Estimation of Risk in Cardiac Patients

Logit and Fuzzy Models in Data Analysis: Estimation of Risk in Cardiac Patients

Článek WoS

The aim of this study was comparison of risk stratification for death in patients after myocardial infarction and of risk stratification for malignant arrhythmias in patients with implantable cardioverter-defibrillator (ICD) using individual risk factors and more complex approaches, which take into account that a borderline between a risky and non-risky value of each predictor is not crisp (fuzzification of a critical value) and that individual risk factors have different weight (area under receiver operating curve - AUC or Sommers D - Dxy). The risk factors were baroreflex sensitivity, ejection fraction and the number of ventricular premature complexes/hour on Holter monitoring. Those factors were evaluated separately and they were involved into logit model and fuzzy models (Fuzzy, Fuzzy-AUC, and Fuzzy-Dxy ). Two groups of patients were examined: 1. 308 patients 7-21 days after having a myocardial infarction (23 patients died within period of 24 month); 2. 53 patients with left ventricular dysfunction examined before implantation of ICD (7 patients having malignant arrhythmia with electric discharge within 11 month after implantation). In myocardial infarction patients our results demonstrated that the application of logit and fuzzy models was superior over the risk stratification based on algorithm where the decision making is dependent on one parameter. In patients with implanted defibrillator only logit method yielded statistical significant result, but the reliability of such result was doubtful because all other tests were statistically insignificant. We recommend evaluating the data not only by tests based on logit model, but also by tests based on fuzzy models.

The aim of this study was comparison of risk stratification for death in patients after myocardial infarction and of risk stratification for malignant arrhythmias in patients with implantable cardioverter-defibrillator (ICD) using individual risk factors and more complex approaches, which take into account that a borderline between a risky and non-risky value of each predictor is not crisp (fuzzification of a critical value) and that individual risk factors have different weight (area under receiver operating curve - AUC or Sommers D - Dxy). The risk factors were baroreflex sensitivity, ejection fraction and the number of ventricular premature complexes/hour on Holter monitoring. Those factors were evaluated separately and they were involved into logit model and fuzzy models (Fuzzy, Fuzzy-AUC, and Fuzzy-Dxy ). Two groups of patients were examined: 1. 308 patients 7-21 days after having a myocardial infarction (23 patients died within period of 24 month); 2. 53 patients with left ventricular dysfunction examined before implantation of ICD (7 patients having malignant arrhythmia with electric discharge within 11 month after implantation). In myocardial infarction patients our results demonstrated that the application of logit and fuzzy models was superior over the risk stratification based on algorithm where the decision making is dependent on one parameter. In patients with implanted defibrillator only logit method yielded statistical significant result, but the reliability of such result was doubtful because all other tests were statistically insignificant. We recommend evaluating the data not only by tests based on logit model, but also by tests based on fuzzy models.

Risk prediction, myocardial infarction, implantable cardioverter-defibrillator, fuzzy logic, area under receiver operating characteristic, logistic regression

Risk prediction, myocardial infarction, implantable cardioverter-defibrillator, fuzzy logic, area under receiver operating characteristic, logistic regression

HONZÍK, P.; KŘIVAN, L.; LOKAJ, P.; LÁBROVÁ, R.; NOVÁKOVÁ, Z.; FIŠER, B.; HONZÍKOVÁ, N.

2011

01.06.2010

Institute of Physiology

Academy of Science of the Czech Republic, Prague

0862-8408

PHYSIOLOGICAL RESEARCH

59

Suppl. 1

Česká republika

89

96

8

http://www.biomed.cas.cz/physiolres/pdf/59%20Suppl%201/59_S89.pdf