A statistical approach to evaluate the performance of cardiac biomarkers in predicting death due to acute myocardial infarction: time-dependent ROC curve

• Published in: Turkish journal of medical sciences Vol. 48; no. 2; pp. 237 - 245
• Main Authors: Karaismailoğlu, Eda, Dikmen, Zeliha Günnur, Akbıyık, Filiz, Karaağaoğlu, Ahmet Ergun
• Format: Journal Article
• Language: English
• Published: Turkey 30-04-2018
• Subjects: creatine kinase isoenzyme MB; troponin T; cardiac biomarkers; B-type natriuretic peptide; Time-dependent ROC curve; myoglobin
• ISSN: 1300-0144; 1303-6165
• DOI: 10.3906/sag-1708-108

Background/aim:
Myoglobin, cardiac troponin T, B-type natriuretic peptide (BNP), and creatine kinase isoenzyme MB (CK-MB)
are frequently used biomarkers for evaluating risk of patients admitted to an emergency department with chest pain. Recently, time-
dependent receiver operating characteristic (ROC) analysis has been used to evaluate the predictive power of biomarkers where disease
status can change over time. We aimed to determine the best set of biomarkers that estimate cardiac death during follow-up time. We
also obtained optimal cut-off values of these biomarkers, which differentiates between patients with and without risk of death. A web
tool was developed to estimate time intervals in risk.
Materials and methods:
A total of 410 patients admitted to the emergency department with chest pain and shortness of breath were
included. Cox regression analysis was used to determine an optimal set of biomarkers that can be used for estimating cardiac death
and to combine the significant biomarkers. Time-dependent ROC analysis was performed for evaluating performances of significant
biomarkers and a combined biomarker during 240 h. The bootstrap method was used to compare statistical significance and the Youden
index was used to determine optimal cut-off values.
Results
: Myoglobin and BNP were significant by multivariate Cox regression analysis. Areas under the time-dependent ROC curves of
myoglobin and BNP were about 0.80 during 240 h, and that of the combined biomarker (myoglobin + BNP) increased to 0.90 during
the first 180 h.
Conclusion:
Although myoglobin is not clinically specific to a cardiac event, in our study both myoglobin and BNP were found to be
statistically significant for estimating cardiac death. Using this combined biomarker may increase the power of prediction. Our web tool
can be useful for evaluating the risk status of new patients and helping clinicians in making decisions.