Session M9.2

Dysrhythmia Hazard After Hospitalization For Myocardial Infarction: Prognosis By Signal-Averaged ECG Compared With Ejection Fraction

J.J. Bailey, A.S. Berson, H. Handelsman

NIH
Bethesda, MD, USA

Prediction of long term risk or hazard of events such as sudden cardiac death or serious ventricular dysrhythmia (VT/ VF) in post myocardial infarct patients may dictate which patient gets a prophylactic implantable cardioverter/ defibrillator. We retrieved reports of signal-averaged ECGs (SAECG) and left ventricular ejection fraction (LVEF) used to predict risk of an event. From each report the number of cases with and without events was extracted to establish accurate values for true positive rate (tpr=sensitivity) and false positive rate (fpr=1-specificity). For all the SAECG reports (9,622 cases in 21 reports) these values were collected and tpr was plotted vs fpr. The (fpr,tpr) data were mapped to a (U,V) space by logistic transforms. The (U,V) data were mapped to (D,S) space by D= V-U and S=V+U. A straight line was fitted to the data using weights determined by the total number of patients in each report. A composite weighted mean value and 95% confidence interval (CI) were also derived. The straight line, mean, and 95% CI data were inverse transformed back to the original (fpr,tpr) space to form a meta-ROC curve, which summarized all the reports as well as revealing their diversity. A summary meta-ROC curve for the LVEF reports (5,173 cases in 17 reports) was obtained the same way.
Composite performance and hazard estimates (including 95% CI) for SAECG were, sensitivity 62.1% (55.8-68.0), specificity 77.0% (71.9-79.7), predicted hazard of a positive test (Haz+) 13.4% (12.7-14.2), predicted hazard of a negative test (Haz-) 2.9% (2.6-3.2). Composite estimates for LVEF were, sensitivity 62.1% (57.5-66.6), specificity 75.5% (73.5-76.9), Haz+ 17.6% (17.4-17.6), Haz- 4.1% (3.7-4.5). About 8 % of a post MI population could be expected to have both SAECG and LVEF tests positive with a Haz+ of 30.3%; whereas about 55% of the population could be expected to have both tests negative with a Haz- of 1.6%. The significance of results in individual reports is compromised by a low incidence of events. Meta-ROC analysis of multiple reports better summarizes the performances of different prognostic methods, and allows the effect of combining tests for a large population to be simulated.