Agency for Healthcare Research and Quality

Address for Correspondence James J. Bailey, MD, MSc

National Institutes of Health

Building 12A, Room 2007, MSC 5620

Bethesda, Maryland 20892-5620

\ E-mail jjbailey@helix.nih.gov

Phone 301-402-5829

FAX 301-402-4544

In 1981 the annual incidence of sudden cardiac deaths (SCD), i.e. death from cardiac cause within minutes to hours after symptoms, was estimated at 300,000 to 400,000 1>; that statistic is still quoted despite a decline in coronary disease since then 2. Frequently SCDs (perhaps as many as 80%) are due to a ventricular dysrhythmia, i.e. ventricular fibrillation (VF) and/or ventricular tachycardia (VT), often related to coronary disease; in some cases, SCD may be the first manifestation of coronary disease. 1 After myocardial infarct (MI) a proportion of patients will have a high risk for VT, VF, or SCD within 2 years. However, the post MI events may represent only a small proportion of the total dysrhythmic events annually. 2

Implantable cardioverter/defibrillators (ICDs) are designed to abort ventricular dysrhythmias with an appropriate electrical stimulus or shock delivered to the myocardium. Five trials have shown superiority of ICDs in reducing mortality of patients at risk for SCD 3-6. In four of these studies, resuscitation from VT/VF was a criterion for acceptance. Based on five other studies Andresen has asserted that the recurrence rate of patients who have been resuscitated from ventricular fibrillation is about 30% and that such patients are now commonly treated with ICDs 7.

However, the proportion of post MI patients requiring resuscitation from VF/VT in the hospital may be small, perhaps 1% or less. There may be another 3- to 5% of post MI patients who did not require resuscitation, but who still are at risk for SCD. The present study compared two noninvasive methods, viz. signal-averaged electrocardiogram (SAECG) and heart rate variablity (HRV), used for assessment of dysrhythmic hazard as reported in the literature; comparison was accomplished by meta-analysis of the reports using ROC curves.

Report Retrieval: Relevant reports were found by searching MEDLINE for items such as: episodes of defibrillation/cardioversion, cardiac resuscitation, or sustained ventricular tachyarrhythmia and sudden cardiac death. A list of the reports used will be available upon request, but many reports can be found listed in the ACC/AHA Guidelines for Ambulatory Electrocardiography 8.

Data Extraction: In each report, the number of cases with and without events, and the number of true positives, true negatives, false positives, and false negatives were sought (as integer values only). Wherever not explicitly given (most reports), these integer values could be estimated using stated values for sensitivity, specificity, and/or predictive accuracy using well known formulae. These integer values were adjusted slightly to recalculate to three decimal places those values for sensitivity, specificity, and predictive accuracy which would most closely approximate (usually within 0.5%) the values stated in the report.

Data Analysis: For each diagnostic method, a global total over all the reports was collected, viz., the number of patients, number of events, number of true positives, ... etc. The first step was a meta-analysis of the reports by the method of Moses and Shapiro 9 to construct a summary receiver-operating-characteristics (ROC) curve for all the reports of a each diagnostic modality, viz. SAECG and HRV. Subsequently weighted composite mean values for sensitivity and specificity and 95% confidence intervals were located on the summary ROC curve (see Figures 1 and 2). The composite values of sensitivity and specificity were combined with the global event rate to derive estimates for predictive accuracy. Dysrhythmia hazard after a positive test is equal to the positive predictive accuracy (%) or after a negative test equal to 100 minus the negative predictive accuracy (%).

We counted up the number of events in 47 reports and that total (1,780) divided by the total number of patients (28,489) yielded an over event incidence of 6.25%. Using this value we estimated what would happen if a combination of SAECG and HRV were applied to a post MI population similar to a merged combination of those reports.

Table I shows the composite values for sensitivity and specificity as well as the event incidence for each group. One measure of diagnostic performance is the odds ratio which is independent of event incidence and is computed by:

OR = [ Se / (1 — Sp) ] / [ (1 — Se) / Sp ]

The composite OR for HRV (6.39) was higher than for SAECG (5.20).

The event incidence in Table I can be considered a prior probability and the hazards can be considered posterior probabilities given the results of the test. Another measure of diagnostic performance is relative risk (i.e. the ratio of the hazards) which was slightly higher for HRV group (5.10) than SAECG group (4.62). The relationship of the hazard or posterior probability and the relative risk to prior probability is not straightforward; in general, given values for sensitivity and specificity, an increase in event incidence will increase both hazards.

The pairwise combination of SAECG and HRV suggests that two thirds of a post MI population could be effectively stratified with dysrhythmic risk.

The significance of results in individual reports is compromised by a low number of events in each report surveyed. By accumulating global case and event counts and by constructing summary ROC curves, the different performances of diagnostic methods could be better appreciated and plausible values for expected outcomes from combining methods could be simulated. Population variation from one diagnostic method to the next was evident in the different global values for event incidence. There was also some diversity in the reported methods, especially in HRV. However each report of a given method reflects the same aspect of the basic pathophysiology. We assumed in the meta-analysis that the investigators in each report attempted to optimize criteria, i.e. to choose thresholds on parameters which gave the best separation of cases with dysrhythmic events from those without events. Of course, the choice of threshold affects the tradeoff between sensitivity and specificity. If that tradeoff were the only source of variation, the data for the individual reports would more closely fit the summary ROC curve. That they do not, probably also reflects the population variation between different reports. For purposes of the survey collection and comparison in Table I, report results were taken at face value, with no attempt to adjust them to make criteria uniform from report to report.

Choosing a threshold and arbitrarily segmenting the data is a fundamental problem with all reports surveyed. For example, a case with an LVEF of 0.39 and another with an LVEF of 0.41 should clearly belong to the same risk category, but they are arbitrarily separated into different risk categories by choosing a threshold of 0.40.

Many would accept that the 1% of post MI patients who meet the stringent criteria of the five ICD trials should so be treated and there are the two studies purporting to demonstrate cost effectiveness of doing so. 4,5 However, there may remain another 4 - 5% who do not meet such stringent criteria but who might meet a criterion of a 30 - 45% hazard. If, as Andresen et al suggest, these patients should be treated prophylactically with ICDs, 7 an immense expenditure in a large population is implied, in which perhaps 50% to 70% of the patients may not stand to benefit; the cost-effectiveness of ICDs in such patients may not be demonstrable. The further resolution of these issues may require another study with a much larger population, with the goal of constructing a robust hazard model with continuous parameters that would allow better individualized hazard predictions for each patient.

1. Arteriosclerosis 1981. Report of the Working Group on Arteriosclerosis of the National Heart, Lung, and Blood Institute. Bethesda, MD. U.S. Department of Health and Human Services, Public Health Service, National Institutes of Health 1981; DHEW, NIH Publ. No. 82-2035:114-22.

2. Zipes DP. Wellens HJJ: Sudden cardiac death. Circulation 1998; 98:2334-2351

3. The Antiarrhythmics Versus Implantable Defibrillators (AVID) Investigators. A comparison of antiarrhythmic-drug therapy with implantable defibrillators in patients resuscitated from near-fatal ventricular arrhythmias. New Eng J Med 1997; 337:1576-1583.

4. Mushkin AI, Hall WJ, Zwanziger J, et al. The cost-effectiveness of automatic implantable cardiac defibrillators: Results from MADIT. Circulation 1998; 97:2129-2135.

5. Wever EFD, Hauer RNW, Schrijvers G, et al. Cost-effectiveness of implantable defibrillator as first-choice therapy versus electrophysiologically guided, tiered strategy in postinfarct sudden death survivors. Circulation 1996; 93:489-496.

6. Schläpfer J, Kappenberger L, Fromer M. What risk should justify implantable cardioverter defibrillator therapy? Am J. Cardiol 1999; 101D-103D.

7. Andresen D, Bruggemann T, Behrens S, Ehlers C. Risk of ventricular arrhythmias in survivors of myocardial infarction. PACE 1997 Part 2; 20:2699-2705

8. ACC/AHA Guidelines for Ambulatory Electrocardiography. J Am Coll Cardiol 1999;912-48.

9. Moses LE, Shapiro D: Combining independent studies of a diagnostic test into a summary ROC curve: Data-analytic approaches and some additional considerations. Stat Med 1993; 12:1293-1316

Figure 1: Meta ROC Analysis of HRV reports

Figure 2: Meta ROC Analysis of SAECG reports