Author + information
- Received November 11, 1998
- Revision received May 20, 1999
- Accepted June 29, 1999
- Published online November 1, 1999.
- ↵*Reprint requests and correspondence: AVID Clinical Trial Center, 1107 NE 45th, Room 505, Seattle, Washington 98105
This study analyzed the causes of death in the Antiarrhythmics Versus Implantable Defibrillators (AVID) Trial.
Both implantable cardioverter-defibrillators (ICDs) and antiarrhythmic drugs (AADs) are used as mainstays of treatment for life-threatening ventricular arrhythmias in patients who have survived either ventricular fibrillation or sustained ventricular tachycardia with hemodynamic compromise and serious symptoms. The AVID Trial compared the effectiveness of these two therapies. Survival was better with the ICD. Assessment of the cause of death should help to determine the mechanism of improvement in survival with the ICD.
Of 1,016 patients enrolled in the AVID Trial, 202 patients died. The mode of death was determined by the unblinded Principal Investigator and independently by an Events Committee, which reviewed materials meticulously blinded with respect to treatment. Deaths were classified as cardiac or noncardiac. Cardiac deaths were further classified as arrhythmic or nonarrhythmic, and causes of noncardiac death were identified.
Deaths were more frequent in patients treated with an AAD (n = 122), compared with patients treated with the ICD (n = 80), unadjusted p < 0.001, p = 0.012 adjusted for sequential monitoring. In AVID, 157 deaths were cardiac, and 79 were arrhythmic. The major effect of the ICD was to prevent arrhythmic death (AAD = 55, ICD = 24, nominal unadjusted p < 0.001). Nonarrhythmic cardiac deaths were equal (AAD = 39, ICD = 39). Patients treated with an AAD had a slightly greater incidence of noncardiac deaths (28 vs. 17, p = 0.053), primarily due to pulmonary and renal causes.
The ICD is more effective than an AAD in reducing arrhythmic cardiac death, while nonarrhythmic cardiac death is unchanged. Of note, apparent arrhythmic death still seems to constitute 38% of all cardiac deaths despite treatment with an ICD. However, the ICD remains superior to an AAD in prolonging survival after life-threatening arrhythmias.
The implantable cardioverter-defibrillator (ICD) is designed to terminate life-threatening ventricular arrhythmias (1). These devices can also pace to control serious bradycardias. Thus, ICDs should prevent only deaths due to arrhythmias and have no effect on deaths due to other causes. If other causes of death, such as congestive heart failure, predominate in any patient population, these competing causes of mortality could diminish or eliminate any beneficial effect of the ICD compared with drug therapy.
The Antiarrhythmics Versus Implantable Defibrillators (AVID) Trial (2,3)has demonstrated that the ICD is superior to an antiarrhythmic drug (AAD) in prolonging survival for patients who have survived life-threatening ventricular arrhythmias. At three years, the survival rate for ICD-treated patients was 84%, compared with 76% for patients treated with an AAD (p = 0.02). In this paper, the causes of death of patients in the AVID Trial are reported.
The design of the AVID Trial has been reported (2). It was a randomized, multicenter comparison of treatment of patients resuscitated from ventricular fibrillation (VF) or sustained ventricular tachycardia (VT) that had caused hemodynamic compromise. Therapy was randomized to either: 1) an ICD or 2) an AAD, primarily amiodarone. All study sites had the protocol approved by the local Institutional Review Board, and all patients signed an informed consent.
The first patient was enrolled in June 1993. The study was terminated in April 1997, when an interim analysis showed that the ICD was the superior treatment. The original report (3)described the results based upon all-cause mortality. Subsequently, all deaths were also classified by cause. For the purposes of the AVID Trial, the term “death” was defined as the time when respiration and blood circulation (pulse) ceased without recovery, that is, the time listed on a death certificate. Any arrhythmia that reverted spontaneously or was converted to a perfusing rhythm by any technique was not counted as a death. As such, sustained VT specifically was not included. Likewise, a cardiac arrest in which the patient was successfully resuscitated was not counted as a death but simply as a recurrent arrhythmia. In addition, patients whose resuscitation yielded a comatose, unresponsive or vegetative condition were also not included as a death, until the patient’s pulse and respiration ultimately ceased. The time of brain death was not considered to be relevant in our analysis.
For each death event, the Principal Investigator completed a death form that categorized the terminal event in detail. Data on this form included the timing of events leading to death, the date and time of clinical death, symptoms before loss of consciousness, associated symptoms, current medications, documentation of rhythm associated with the event (if rhythm monitoring occurred), summary cause of death (cardiac vs. noncardiac), cause of cardiac death (arrhythmic vs. nonarrhythmic), evidence of congestive heart failure (CHF) or shock, evidence of myocardial ischemia or myocardial infarction (MI), recent surgery, recent invasive cardiac therapeutic or diagnostic procedure and the opinion of the Principal Investigator whether this patient would have survived for four months if a terminal arrhythmia had not occurred (to distinguish the patient whose terminal arrhythmia was simply the result of end-stage heart failure). Furthermore, the Principal Investigator explained the circumstances surrounding the death in a detailed letter to the Clinical Trial Center, including a description of the documented and/or presumed sequence of arrhythmias and events from information obtained from relatives, bystanders or other persons who might have witnessed the event. Detailed data were obtained that included prehospital notes, emergency room notes, all available electrocardiograms (ECGs) and rhythm strips, hospital notes (including the admission history and physical examination, plus a minimum of the last three days of physicians’ progress notes), nurses’ notes for the three days preceding death, physicians’ orders, drug administration notes, enzyme tests, other relevant laboratory tests, ICD interrogation electrograms and data output, the discharge summary and an autopsy report (if performed). In addition, the Clinical Site Coordinator completed a concurrent drugs form for medications being taken by the patient, and an ICD evaluation form for patients who had an ICD. If the patient was hospitalized in conjunction with the event, a hospitalization form also was completed. Thus, extensive clinical data regarding the terminal event were obtained.
After the Principal Investigator reviewed and classified the death, all data were sent to the Clinical Trial Center. Medical records were then blinded to remove all references to therapy either with an AAD or the ICD. Furthermore, “sham-blinding” was performed, in which chart items were excluded to try to mimic items that would have been deleted if the patient had been randomized to the alternative arm of the study. For example, a medication might have been deleted from an ICD patient’s medication list so that a blank would appear on that portion of the progress notes. Therefore, not all patients who had obvious deletions of drugs in admission notes and progress notes would have been patients randomized to the drug arm of AVID. Conversely, patients randomized to the drug arm also had “sham-blinding” performed by deleting, for example, sentences from X-ray reports to suggest that a description of an endocardial electrode had been deleted from the X-ray report. Members of the Events Committee were aware that “sham-blinding” could occur.
A single member of the Events Committee reviewed each blinded death event without prior knowledge of the death classification by the site’s Principal Investigator. If the Events Committee Reviewer agreed completely with the Principal Investigator, no further review occurred. If disagreement in classification existed, the event was reviewed by the entire Events Committee, consisting of three cardiologists experienced in Events Committee deliberations. Determination of cause of death was made by consensus. The Events Committee classification was final, except in three cases categorized as “unable to classify” as cardiac or noncardiac by the Events Committee, in which case the Principal Investigators’ determinations were used. In one other case, classified as cardiac both by the Events Committee and the Principal Investigator, the Events Committee was unable to determine whether the cardiac death was arrhythmic or nonarrhythmic, in which case the Principal Investigator’s judgment of nonarrhythmic was used.
Deaths were classified as cardiac or noncardiac. Cardiac deaths were subsequently classified as arrhythmic cardiac or nonarrhythmic cardiac. The assignment of a diagnosis of arrhythmic cardiac death required the absence of severe CHF or shock. Furthermore, an arrhythmic cardiac death was one in which long-term survival was judged to have been possible, had it not been for the arrhythmia itself. Whereas the time frame of the events leading to death was recorded, timing of symptoms was not necessarily a determinant of the mechanism of death. Nonarrhythmic cardiac deaths were secondary to CHF or shock. If the CHF or shock was not reversible and the reviewer believed that the patient would die soon of pump failure, the death was considered to be cardiac, but nonarrhythmic. In cases where CHF or shock caused other end-organ failure, a separate cardiac classification was allowed. The presence or absence of myocardial ischemia was noted, with evidence for ischemia being symptoms, ECG changes or enzyme abnormalities. Patients with MI and death were classified as arrhythmic if end-stage CHF or shock was not present. If the death in a clinically stable person was unwitnessed (the patient not being seen or heard for more than 5 min), it was presumed to be arrhythmic unless evidence to the contrary was discovered. Patients who died after cardiac surgery were considered to have had a cardiac death. If death occurred after a cardiac procedure or operation in which there was an unbroken string of nonarrhythmic events leading inexorably downhill, the event was classified as nonarrhythmic cardiac, even though the terminal item in the sequence might have been noncardiac, such as sepsis. In all cases, an assessment of anticipated long-term survivability was crucial. A patient who was resuscitated from out-of-hospital cardiac arrest but who died without regaining neurologic function during hospitalization was considered to have had an out-of-hospital arrhythmic death. A patient who died from a pulmonary cause was classified based on whether it could have been due to amiodarone, even though the data regarding drug administration were blinded. Cases of suspected amiodarone pulmonary toxicity were then unblinded regarding whether the patient had actually received amiodarone.
Patients in whom therapy was changed during the study (crossovers) were included in this analysis, but they were analyzed in the group to which they were randomized (intention-to-treat). Because sequential monitoring rules were based on total mortality, it is not possible to compute adjusted p values for comparisons of cause-specific outcomes. Thus, nominal p values (unadjusted for sequential monitoring) based on the log-rank statistic are provided for cause-specific deaths.
Mode of death
More deaths occurred in patients assigned to an AAD (n = 122), compared with patients treated with an ICD (n = 80), p < 0.012, adjusted for sequential monitoring. The majority (78%) of deaths were classified as cardiac (Table 1), subclassified equally frequently as arrhythmic and nonarrhythmic. The onset of terminal symptoms occurred nearly equally in-hospital versus out-of-hospital (74 patients [47%] vs. 83 patients [53%], Table 2). Thirty-four deaths were witnessed arrhythmic events: 15 were asymptomatic (5 ICD, 10 AAD); 12 were >5 min from onset of symptoms to death (5 ICD, 7 AAD); and 7 were missing data to calculate time from onset of symptoms to death (4 ICD, 3 AAD). Twenty-four unwitnessed arrhythmic deaths occurred unexpectedly during sleep (5 ICD, 19 AAD). Most arrhythmic deaths were out-of-hospital. Very few patients had evidence of myocardial ischemia or MI directly related to death. Even with assessment of multiple markers of ischemia (symptoms, ECGs, enzymes), only 8 of 78 nonarrhythmic cardiac deaths and 4 of 79 arrhythmic cardiac deaths were associated with suspected ischemia, 6 of 8 and 1 of 4 with MI documented, respectively. Noncardiac deaths were more common in the group treated with an AAD (28 vs. 17, p = 0.053; relative risk = 1.78, 95% confidence interval = 0.98 to 3.26; Table 3), with deaths due to cancer and pulmonary disease predominating. Three deaths were definitively attributed to pulmonary toxicity due to amiodarone.
Figure 1depicts survival free of cardiac death, with noncardiac death censored. Cardiac death was less frequent in patients treated with an ICD. Survival free of cardiac death at one and two years was 90.9% and 85.0% in the ICD group and 85.1% and 81.2% in the AAD group, respectively (p = 0.0042).
Figure 2depicts survival to arrhythmic death, with both noncardiac and nonarrhythmic cardiac deaths censored. Survival was 96.6% and 94.2% at one and two years for patients treated with ICDs and 91.9% and 89.1% for AADs, respectively (p = 0.0002). The benefit of the ICD is most evident in this analysis.
The 24 apparent arrhythmic deaths in patients with an ICD are of particular interest. Five were unwitnessed and occurred unexpectedly during sleep; five were witnessed and occurred within 5 min of onset of symptoms or asymptomatic; 10 were unwitnessed but did not occur during sleep and were classified as presumably arrhythmic. The remaining four deaths were witnessed and occurred >5 min from the onset of symptoms. Thus, there was little opportunity to gather much data about these terminal events. Only seven had an ICD interrogation after death. In three of seven, the ICD shocked appropriately but simply failed to convert the tachyarrhythmia. In four of seven, no shocks were delivered, and bradycardia may have been the terminal arrhythmia with the ICD pacing function not preventing death.
Overall, an ICD interrogation was performed after death in only 17 patients, rendering this method of assessment generally unhelpful in classifying deaths. Most ICDs were not interrogated because the physician attending the patient at the time of death was not the AVID physician and was unable to interrogate the device or unaware that such interrogation might be useful. However, in those 17 patients in whom an interrogation was performed, 10 had been classified as nonarrhythmic.
Autopsies, performed in only 14 patients, were too infrequent to be generally useful to aid in classifying the cause of death.
Figure 3illustrates that the ICD conferred no survival benefit for nonarrhythmic cardiac deaths, p = 0.804 (both noncardiac and arrhythmic deaths censored). Though protection against arrhythmic death might be expected to increase the number of patients with an ICD at risk of nonarrhythmic cardiac death, such a trend was not seen.
Arrhythmic death was more common, but of borderline significance (p = 0.07), in patients whose index arrhythmia was VF than in patients who initially qualified for AVID with VT. There was no difference between patients with VF and VT regarding the length of time until nonarrhythmic cardiac death (p = 0.54). Survival free of arrhythmic death for patients whose presenting arrhythmia was VT was improved by the ICD (Fig. 4, p = 0.025), but not as marked as it was for patients who qualified with VF, where the survival advantage of the ICD is obvious, with more than twice as many deaths in the AAD group (Fig. 5, p = 0.0019). There was no difference in the survival between ICD and antiarrhythmic drug for nonarrhythmic cardiac death in patients with either VT or VF (p = 0.72, 0.98, respectively).
Table 4lists the covariates analyzed in AVID by treatment group. Neither these nor any other factors analyzed explained the survival benefit seen with the ICD.
Consistency of classification of death
Initial review by the single Events Committee Member agreed with the Principal Investigator in 150 of 202 cases (74%). In the 52 disputed cases reviewed by the entire Events Committee, 50 were decided by unanimous vote (three to zero), while two were decided by a two-to-one vote. Table 5compares the final classification of the death event between the Principal Investigator and the final Events Committee determination. Agreement was generally good between these classifications. In those 147 deaths that both the Principal Investigator and the Events Committee agreed were cardiac, the assessment of arrhythmic versus nonarrhythmic was likewise good, with only 8% disagreement between the two classifications. After completion of the review of deaths, the Events Committee examined all disagreements regarding arrhythmic versus nonarrhythmic causes between the Events Committee and the Principal Investigator. Six of the 11 disagreements had resulted from the unawareness on the part of the Principal Investigator that unwitnessed, unexpected death was to be classified as arrhythmic.
The ICD was designed to prevent arrhythmic death. In the AVID Trial, the ICD decreased arrhythmic death, whereas nonarrhythmic cardiac death was unaltered. Noncardiac death was more frequent in the AAD-treated group (p = 0.053), with the difference primarily an excess of pulmonary (AAD = 9, ICD = 2) and renal (AAD = 4, ICD = 0) deaths in the AAD group.
Comparison with previous studies
The mechanism of death is frequently difficult to determine in patients with severe cardiac disease (both coronary disease and left ventricular dysfunction) who are at high risk both for arrhythmic and nonarrhythmic mortality (4–6). Previous nonrandomized studies have shown the benefit of the ICD in the reduction of arrhythmic events (7–13), which are usually thought to be due to either VF or VT degenerating to VF (though bradycardia deaths may be common in end-stage heart failure). However, because these studies were not blinded with respect to ICD therapy, there is a potential for bias to be introduced by investigator classification of events. Because most previous ICD studies (7–13)were small or lacked randomization, patient selection bias may have entered into the choice of therapy and, therefore, the outcome. Evidence of this bias can be seen in the improvement in nonarrhythmic or noncardiac death in patients treated with the ICD. Previous studies also have a significant potential for event classification bias, where fewer deaths might have been classified as arrhythmic because of known ICD therapy. Randomized studies, such as AVID, minimize this type of bias (14–19). In the AVID Trial, the Events Committee classified events without knowledge of treatment assignment. The Coordinating Center meticulously blinded all references to arrhythmia treatment from the charts before review, ensuring an unbiased classification.
Although precise classification of death is always difficult, in the AVID Trial, agreement in the classification of deaths between the unblinded Principal Investigator and the blinded Events Committee was good.
Mechanism of death
In the AVID Trial, the primary end point (all-cause mortality) was decreased by the ICD. Our study on the mechanism of death confirms that the majority of the benefit of the ICD results from the prevention of death due to arrhythmia, presumably from the treatment of VT or VF. It is also possible, though unlikely, that antiarrhythmic drugs actually increased arrhythmic deaths. The relative contribution of the beneficial effect of the ICD and any postulated deleterious effect of an antiarrhythmic drug cannot be estimated.
Clinical and electrophysiologic differences exist between patients who have been resuscitated from VF and those who have experienced sustained VT (20,21). The AVID Trial further emphasizes this distinction. Patients with a history of VF were more likely to have an arrhythmic death than patients with VT. This finding suggests that the substrate for an episode of VF is more likely to be associated with subsequent arrhythmic mortality.
In the AVID Trial, the majority of events were unrelated to measurable ischemia, whether detected by patient symptoms, electrocardiographic changes or enzyme abnormalities. In fact, only 4 of 79 arrhythmic deaths were associated with markers of ischemia. This finding is consistent with previous studies that have demonstrated that most arrhythmic events are not necessarily associated with detectable transient myocardial ischemia or an acute MI (22,23). However, clinically inapparent ischemia as an arrhythmic trigger in this population with a high prevalence of coronary artery disease cannot be excluded. In the AVID Trial, 45 of 79 deaths classified as arrhythmic were unwitnessed, and witnessed arrhythmic deaths usually occurred abruptly. Often, ECGs and enzymes were not obtained. The suddenness and lack of objective laboratory data could account for some of the apparent lack of manifest ischemia.
In the AVID Trial, randomly assigned therapy was expected to minimize treatment biases. It would be expected that nonarrhythmic and noncardiac deaths would be equal in patients treated with the ICD and patients treated with an AAD, unless nonarrhythmic cardiac deaths were increased in patients with an ICD because these patients have a longer exposure time during which a nonarrhythmic cardiac death could occur. However, no survival advantage or disadvantage from the ICD was seen for nonarrhythmic cardiac death. This finding is contrary to what had been shown in some other studies where enrollment bias might have been present. There was a small, though statistically insignificant, difference in noncardiac death, with a survival advantage in patients treated with the ICD, compared with patients treated with an AAD. The trend was not eliminated by adjusting for ejection fraction, age, diabetes or history of CHF. This difference was eliminated, however, when pulmonary deaths were censored. Thus, a possible explanation may be a deleterious effect of AADs on noncardiac mortality, but if the pulmonary deaths were indeed caused by AADs, then these deaths should have been classified as cardiac.
The use of beta-blocking drugs was imbalanced, with more patients in the ICD group receiving these agents. Statistical adjustment for this covariate imbalance did not affect the main result of the AVID Trial (the primary end point being death from any cause), though it is possible that beta-blockers could have affected the mechanism of death.
In the AVID Trial, most arrhythmic deaths were out-of-hospital. However, out-of-hospital events were also more likely to be unwitnessed, perhaps leading to an overestimate of the number of arrhythmic events. On the other hand, hospitalized patients would be more likely to have a condition that would lead to a nonarrhythmic event, such as CHF, cancer or pneumonia.
The ICD is an effective strategy to prevent arrhythmic cardiac deaths. Despite this significant benefit, up to 38% of cardiac deaths in patients treated with an ICD still appeared to be arrhythmic. Nevertheless, applied to a sufficiently high-risk population, the ICD is superior to an AAD. Appendix.
Events Committee:Chairman: David W. Richardson, MD, FACC, Richmond, VA; Leonard A. Cobb, MD, FACC, University of Washington, Seattle, WA; Craig M. Pratt, MD, FACC, The Methodist Hospital, Houston, TX.
Writing Committee (listed alphabetically):Jeffrey L. Anderson, MD, FACC, University of Utah/LDS Hospital, Salt Lake City, UT; Leonard A. Cobb, MD, FACC, University of Washington, Seattle, WA; Frederick Ehlert, MD, FACC, St. Luke’s/Roosevelt Hospital Ctr., New York, NY; Andrew E. Epstein, MD, FACC, University of Alabama at Birmingham, Birmingham, AL; Dennis Flynn, RN, MBA, University of Rochester, Rochester, NY; H. Leon Greene, MD, FACC, University of Washington, Seattle, WA; Alfred P. Hallstrom, PhD, University of Washington, Seattle, WA; Soo Kim, MD, FACC, Montefiore Medical Center, Bronx, NY; Linette R. Klevan, RN, Sentara Norfolk General Hospital, Norfolk, VA; Richard Moore, University of Washington, Seattle, WA; Paul Mounsey, MD, University of Virginia Medical Center, Charlottesville, VA; Craig M. Pratt, MD, FACC, The Methodist Hospital, Houston, TX; David W. Richardson, MD, FACC, Richmond, VA; Eleanor Schron, RN, MS, National Heart, Lung, and Blood Institutes, NIH, Bethesda, MD.
For a list of AVID Investigators, see the Appendix to reference 3.
☆ This work was supported by a contract (N01 HC-25117) with the National Heart, Lung, and Blood Institute, Bethesda, Maryland.
- antiarrhythmic drug
- angiotensin-converting enzyme
- Antiarrhythmics Versus Implantable Defibrillators
- congestive heart failure
- implantable cardioverter-defibrillator
- myocardial infarction
- ventricular fibrillation
- ventricular tachycardia
- Received November 11, 1998.
- Revision received May 20, 1999.
- Accepted June 29, 1999.
- American College of Cardiology
- Epstein A.E.,
- Carlson M.D.,
- Fogoros R.N.,
- et al.
- Pratt C.M.,
- Greenway P.S.,
- Schoenfeld M.H.,
- et al.
- Akhtar M.,
- Avitall B.,
- Jazayeri M.,
- et al.
- Fogoros R.N.,
- Elson J.J.,
- Bonnet C.A.,
- et al.
- Zipes D.P.
- Greene H.L.
- Cobb L.A.,
- Baum R.S.,
- Alvarez H. III.,
- Schaffer W.A.
- Goldstein S.,
- Landis J.R.,
- Leighton R.,
- et al.