With the development of ICDs that can be implanted without thoracotomy (and possibly during an outpatient procedure), use of an ICD is now a practical therapeutic alternative to antiarrhythmic drug therapy [4]. Although ICDs are remarkably effective in terminating ventricular arrhythmias [17-20], they are expensive ($40 000 to $60 000 for implantation), and the extent to which they extend life is unknown [21]. Ongoing or planned randomized, controlled trials [19, 20, 22-27] will clarify the role of ICDs and drug therapy for the prevention of sudden cardiac death, but their results will not be available for several years. Economic analyses have suggested that ICDs have favorable cost-effectiveness ratios [28-32], but these analyses were based on controversial assumptions about efficacy in improving survival [28, 30, 31]; compared the implantation of an ICD with an expensive alternative, such as electrophysiologically guided therapy [29, 31, 32]; or limited the use of ICDs to extremely high-risk patients [28-32].

In this study, we used data from ongoing randomized trials and data on the costs of third-generation ICDs to evaluate the cost-effectiveness of treatment with an ICD (implanted without thoracotomy) relative to empirical therapy with amiodarone. We determined the reduction in total mortality that ICD use would have to confer to reach specified cost-effectiveness ratios. Because the indications for ICD use may expand the use of this therapy into new patient populations, we evaluated how the cost-effectiveness of treatment with an ICD varies when the device is used in a population that has a lower risk for sudden cardiac death than do survivors of cardiac arrest.

Methods

Top Methods Results Discussion References

We used a decision model to estimate the quality-adjusted length of life and expenditures for a population of patients who received amiodarone or an ICD. We used the perspective of society and incorporated benefits and costs accordingly. We examined three treatment strategies (Figure 1). Patients who received the ICD-only regimen began treatment with an ICD and continued to receive this therapy regardless of subsequent arrhythmic events. Patients who received amiodarone only began treatment with amiodarone and continued to receive this drug as sole therapy regardless of subsequent arrhythmic events. They crossed over to receive an ICD only if they had intolerable side effects as a result of amiodarone use. Patients who received the amiodarone-to-ICD therapy began treatment with amiodarone and crossed over to ICD if they were subsequently resuscitated from ventricular fibrillation (all survivors) or from ventricular tachycardia (50% of survivors) or if severe drug toxicity occurred. We did not evaluate treatment strategies that used ICD and amiodarone simultaneously because evidence was not sufficient to assess the efficacy of this combined therapy. We discounted health benefits and costs using a 3% annual discount rate, as recommended by a panel on cost-effectiveness analysis in health care [33], and we did sensitivity analyses on all model variables.

View larger version (35K): [in this window] [in a new window]   Figure 1. Schematic representation of the decision model. The square node represents a decision to use one of three treatment regimens: implantable cardioverter defibrillator (ICD) only, amiodarone only, or amiodarone-to-ICD (see Methods). Circles represent chance nodes. After a regimen is chosen, the patient enters a Markov tree (denoted by rectangles containing circles and an arrow). A patient who receives the ICD-only regimen enters the Markov tree only if he or she survives ICD implantation. The Markov trees represent the clinical events that can occur during each 1-month period as a patient is followed until death. For example, a patient receiving the ICD-only regimen is at risk each month for ventricular tachycardia (VT), ventricular fibrillation (VF), nonarrhythmic cardiac death, and noncardiac death. If none of these events occur, the patient remains "well" for the 1-month period. A patient who is well at the end of a month reenters the Markov tree. The braces represent subtree notation and indicate that a patient who has VT or VF enters the arrhythmia subtree, which is shown in the top panel of Figure 2. Subtrees are denoted by rounded rectangles. Patients receiving the amiodarone-only or the amiodarone-to-ICD regimen are at risk for VT, VF, nonarrhythmic cardiac death, noncardiac death, and toxicity from amiodarone. If toxicity from amiodarone occurs, the patient enters the amiodarone toxicity subtree (Figure 2, bottom). For the amiodarone-to-ICD treatment regimen, the events that occur after a patient has an arrhythmia depend on the type of arrhythmia experienced. A patient who has VT enters the amiodarone-to-ICD VT subtree (Figure 2, upper middle); a patient who has VF enters the amiodarone-to-ICD VF subtree (Figure 2, lower middle).

View larger version (33K): [in this window] [in a new window]   Figure 2. Decision model subtrees. The subtrees show events that may occur to patients during a 1-month cycle. Top. The arrhythmia subtree. A patient who receives the implantable cardioverter defibrillator (ICD)-only regimen or the amiodarone-only regimen and has ventricular arrhythmia enters the arrhythmia subtree. If the patient survives the arrhythmia, he or she may have neurologic impairment or may become well (have no neurologic sequelae). A patient who is well continues to receive the assigned therapy (Rx) and returns to the appropriate Markov node (see Figure 1) to begin another 1-month period. Upper Middle. The amiodarone-to-ICD ventricular tachycardia (VT) subtree. Fifty percent of patients who survive ventricular tachycardia without neurologic impairment receive an ICD; the other 50% continue to receive amiodarone. Lower Middle. The amiodarone-to-ICD ventricular fibrillation (VF) subtree. All patients who have VF receive an ICD, as denoted by the chance node with only one branch (ICD implantation). Bottom. The amiodarone toxicity subtree. This represents the events that may occur in patients in whom amiodarone toxicity occurs. A patient may have an acute side effect, withdraw from treatment, or die as a result of toxicity. Acute toxicity represents time-limited symptoms that require evaluation and management (for example, thyroid dysfunction) but not discontinuation of amiodarone therapy.

We developed a Markov model [34, 35] (Figure 1, Appendix) using SMLTree software (version 2.9, J. Hollenberg, New York, New York); the model tracked a hypothetical cohort of patients over time. Each cohort began receiving one of the three therapeutic regimens: ICD only, amiodarone only, or amiodarone-to-ICD. Each month, a patient was at risk for ventricular fibrillation, ventricular tachycardia, nonarrhythmic cardiac death, noncardiac death, and illness or death from drug toxicity (the latter was applicable only to patients who received amiodarone). Patients who had an ICD were also at risk for perioperative death. If a patient had ventricular fibrillation or ventricular tachycardia, the patient either died, survived with neurologic impairment, or survived without neurologic impairment (Figure 2). The model included a decrement in quality of life for patients who survived an arrhythmic event with neurologic sequelae [36-39]. Patients who were treated with amiodarone were at risk for acute drug toxicity (Figure 2).

Quality of Life

The Markov model incorporated adjustments for quality of life associated with current health, ICD or amiodarone therapy, arrhythmic events, ICD discharges (shocks), and amiodarone toxicity. We used the time-tradeoff technique to calculate quality-adjusted life-years [40, 41]. In the base-case analysis, we assumed that the quality of life of current health was 0.75 [39], and we assumed that quality of life did not change as a result of ICD or amiodarone therapy. In sensitivity analyses, we evaluated the importance of changes in quality of life caused by ICD or amiodarone therapy.

Effectiveness of Implantable Cardioverter Defibrillators

We assumed that treatment with an ICD did not affect the frequency of arrhythmias but did increase the chance for surviving an arrhythmic event if one occurred. Evidence from randomized trials and patient registries indicates that ICDs successfully treat life-threatening ventricular arrhythmias. In a registry that contained more than 600 patients with third-generation ICDs [17], ventricular tachycardia was terminated successfully in 98.7% of cases and ventricular fibrillation was converted in 98.9% of cases. The Cardiac Arrest Survivors in Hamburg (CASH) study, a randomized trial that compared ICD with pharmacologic therapy in survivors of cardiac arrest, reported a cardiac death rate of 0% at 1 year among 59 patients treated with an ICD [16, 19, 20, 42]. Other studies [18, 43-46] have also reported low rates of sudden cardiac death in ICD recipients. In our base-case analysis, we assumed that ICDs successfully terminated arrhythmias at rates similar to those reported in the patient registry [17] (Table 1).

The effect of ICD use on total mortality is less clear [67-70]. At approximately 1 year of follow-up, total mortality rates in the CASH study were 14.3% in the group that received ICDs and 14.7% in the group that received amiodarone; the difference between the groups was not statistically significant [16]. However, this trial is incomplete and relatively small: Each treatment group contained fewer than 60 patients. In the Multicenter Automatic Defibrillator Implantation Trial (MADIT), 196 patients who had a history of previous Q-wave infarction, documented nonsustained ventricular tachycardia, an ejection fraction of 0.35 or less, and inducible sustained ventricular tachycardia (shown on electrophysiologic testing) that could not be suppressed by the infusion of procainamide were randomly assigned to receive either an ICD or conventional pharmacologic care [25]. The trial was stopped early because 15 deaths occurred in the ICD group and 39 occurred in the usual care group, producing a hazard ratio for total mortality of 0.46 (95% CI, 0.26 to 0.82). This corresponds to Kaplan-Meier survival rates of 87% in the ICD group and 65% in the usual care group at approximately 2 years (Moss AJ for the MADIT Investigators. Multicenter Automatic Implantable Defibrillator Trial [MADIT]. 17th Annual Scientific Session of the North American Society of Pacing and Electrophysiology. Seattle; 1996). The risk reduction found in MADIT may particularly favor ICD because selected patients were not suppressed by drug therapy and because the trial was discontinued early. Other large randomized trials of ICD treatment are still ongoing and have not reported any results.

Cost-effectiveness studies [28, 30, 31] (with one exception [32]) have assumed that patients who receive ICDs have substantial survival advantages. This assumption has been based on the results of nonrandomized trials, which are subject to selection bias because healthier patients may have had ICD implantation. Such bias precludes definitive inferences about the effect of ICDs on total mortality [21, 71]. For our base-case analysis, therefore, we assumed that ICD use would reduce total mortality at 1 year by 20% to 40% relative to amiodarone therapy in patients who survive ICD implantation. This reduction was approximately constant over time. In sensitivity analyses, we examined reductions in total mortality by ICDs compared with amiodarone that varied from 5% to 60%. A reduction in total mortality of 30% is a reasonable point estimate, given the current evidence.

We assumed that the implantation of an ICD was associated with a perioperative mortality rate of 1.8% [17] and was complicated by infection [31, 72] or lead failure [17] at rates of 2.0% and 3.0% per year, respectively (Table 1). The perioperative mortality rate of 1.8% was that seen in a registry of more than 600 patients in whom ICD implantation was attempted without thoracotomy, but it included patients who subsequently had thoracotomy because of unsuccessful implantation of endocardial leads; we explored lower estimates of perioperative mortality in sensitivity analyses. In our base-case analysis, we assumed that the ICD generator would have to be replaced every 4 years [47, 50, 51].

Effectiveness of Amiodarone

We assumed that treatment with amiodarone both reduced the incidence of arrhythmic events and improved the likelihood of surviving an episode of ventricular tachycardia if one occurred (Table 1). We assumed that the annual mortality rate from sudden cardiac death with amiodarone therapy was identical to the rate of sudden cardiac death seen at approximately 1 year among patients treated with amiodarone: 8.6% [16]. We assumed that the total mortality rate with ICD treatment was 20% to 40% less than the total mortality rate with amiodarone treatment. We deliberately provided a survival advantage to patients treated with ICD so that we could evaluate the cost-effectiveness of ICD use when it modestly improved longevity. Our approach to modeling the toxicity of amiodarone [10-1214, 56, 58-64] is described in the Appendix and shown in Figure 2.

Patient Population

Our base-case analysis evaluated 57-year-old patients who had survived previous cardiac arrest and thus were at high risk for sudden cardiac death (Table 1). We also evaluated the cost-effectiveness of the three therapeutic regimens in a population of patients at intermediate risk for sudden cardiac death (the rate of arrhythmia and nonarrhythmic cardiac death was approximately one half the base-case rate). This level of cardiac mortality and total mortality is similar to that noted in patients with congestive heart failure [73].

Costs

Our analysis included the direct costs of medical care associated with treatment (Table 1, Appendix) and excluded indirect costs, such as patient travel time and inconvenience. We included the costs of hospitalization for the initiation of treatment with amiodarone or with an ICD; costs of ongoing therapy, including physicians' visits, laboratory tests, and rehospitalization; and costs of ICD generator and lead replacement. Our cost estimates reflect the fact that ICD generators can be replaced during an out-patient procedure. The cost of generator replacement is an important component of the cost of ICD treatment (50% to 65% of the initial implantation cost) and depends on the frequency of replacement and the type of unit used. Further details about model inputs, assumptions, and structure are available from the authors or on the World Wide Web (http://www-smi.stanford.edu/projects/scd/).

Results

Top Methods Results Discussion References

Treatment with an ICD is the most expensive of the regimens that we studied, but it results in the greatest quality-adjusted life expectancy (Table 2). The amiodarone-to-ICD regimen results in intermediate costs and intermediate quality-adjusted length of life. The amiodarone-only regimen results in the lowest quality-adjusted life expectancy, but it is the least expensive strategy. Patients in the intermediate-risk group live substantially longer than do high-risk patients (because of lower mortality from sudden and nonsudden cardiac death); thus, their costs are higher for each of the treatment regimens (Table 3). Table 2 and Table 3 show the outcomes that occur if ICD reduces the total mortality rate by 20% and indicate expected length of life with treatment, un-adjusted for quality of life. If use of an ICD reduces the total mortality rate by 40% relative to amiodarone therapy, high-risk patients who receive the ICD-only regimen live 1.6 years (1.17 quality-adjusted years) longer than do high-risk patients who receive the amiodarone-only regimen, at an additional cost of $43 700. Intermediate-risk patients who receive the ICD-only regimen live 1.74 years (1.28 quality-adjusted years) longer than do intermediate-risk patients who receive the amiodarone-only regimen, at an additional cost of $46 300.

In high-risk patients, the marginal cost-effectiveness of treatment with an ICD relative to treatment with amiodarone alone ranges from $74 400 per quality-adjusted life-year gained (if ICD use reduces the total mortality rate by 20%) to $37 300 per quality-adjusted life-year gained (if ICD use reduces the total mortality rate by 40%) (Table 4). If ICD use reduces the total mortality rate by 30%, it costs $49 300 per quality-adjusted life-year gained relative to amiodarone therapy. If it reduces the mortality rate by 20%, ICD use is modestly less cost-effective in intermediate-risk patients (Table 4); if it reduces the mortality rate by 40%, the cost-effectiveness ratios are similar in high-risk and intermediate-risk patients. The marginal cost-effectiveness of ICD alone relative to the amiodarone-to-ICD regimen is slightly more favorable. However, the amiodarone-to-ICD regimen provides little benefit relative to the amiodarone-only regimen and is expensive relative to both the amiodarone-only and the ICD-only regimens. We thus emphasized the amiodarone-only and ICD-only regimens in the sensitivity analyses.

Sensitivity Analyses

Our sensitivity analyses indicated that estimates of the cost-effectiveness of ICD therapy were substantially influenced by the relative reduction in the total mortality rate due to use of an ICD, frequency of generator replacement, quality of life with therapy, and cost of initial implantation (Figure 3 and Figure 4). We recalculated the cost-effectiveness of treatment with an ICD, assuming relative risk reductions that ranged from 5% to 60% (Figure 3, top). At a relative risk reduction of 10%, the marginal cost-effectiveness of ICD treatment relative to amiodarone therapy is $158 700 per quality-adjusted life-year gained; a relative risk reduction of 15% results in a cost-effectiveness of $100 600 per quality-adjusted life-year gained. In contrast, if treatment with an ICD reduces the total mortality rate by more than 50% relative to amiodarone therapy, treatment with ICD costs less than $30 200 per quality-adjusted life-year gained. The bottom panels of Figure 3 and Figure 4 are based on the assumption that ICD use reduces the total mortality rate by 20%; results are qualitatively similar when different levels of risk reduction are assumed.

View larger version (17K): [in this window] [in a new window]   Figure 3. Sensitivity analyses. Base-case estimates for each sensitivity analysis are indicated by arrows. Top. Effect of the increased survival advantage produced by an implantable cardioverter defibrillator (ICD) on the cost-effectiveness of therapy with an ICD compared with amiodarone therapy. Bottom. The frequency of generator replacement substantially affects marginal cost-effectiveness because the devices are expensive.

View larger version (25K): [in this window] [in a new window]   Figure 4. Effect of cost of implantation of an implantable cardioverter defibrillator (ICD) on cost-effectiveness. The middle line shows the base-case cost estimate for ICD implantation; the top and bottom lines show how cost-effectiveness varies if implantation costs are increased or decreased by 20%.

If ICD generators are replaced frequently, the cost-effectiveness of ICD treatment becomes unfavorable (Figure 3, bottom). However, if the generators are replaced every 5 years (as manufacturers currently predict) rather than every 4 years, as we assumed in our base-case analysis, the cost-effectiveness of ICD implantation relative to amiodarone therapy improves from $74 400 to $63 800 per quality-adjusted life-year gained. Variation in the implantation cost of an ICD has a moderate effect on the marginal cost-effectiveness of ICD treatment relative to amiodarone therapy (Figure 4). In addition, sensitivity analyses indicate that if quality of life with amiodarone treatment is 0.65 and quality of life with use of an ICD remains at the base-case estimate of 0.75, the marginal cost-effectiveness ratio would improve from $74 400 to $43 300 per quality-adjusted life-year saved. Alternatively, if quality of life with use of an ICD is 0.65 and quality of life with amiodarone treatment is 0.75, the cost-effectiveness of ICD use would change from $74 400 to $447 700 per quality-adjusted life-year gained.

Other variables have less influence on our estimates of the cost-effectiveness of ICD use. The ease of implantation of third-generation ICDs has led several observers to predict that the indications for ICD treatment will expand substantially. As shown in the bottom panel of Figure 3, treatment with an ICD becomes only modestly less cost-effective as the device is used in a population that has an intermediate risk for sudden cardiac death. Using a discount rate of 5% rather than 3% reduces the cost-effectiveness of ICD treatment relative to amiodarone therapy from $74 400 to $85 900 per quality-adjusted life-year gained. Reduction of the perioperative mortality rate from 1.8% to 0.5% improves the cost-effectiveness of ICD use from $74 400 to $69 900 per quality-adjusted life-year gained.

Discussion

Top Methods Results Discussion References

We compared the cost-effectiveness of ICD implantation done without thoracotomy with that of empirical amiodarone therapy for the prevention of sudden cardiac death. Our base-case analysis indicated that in high-risk patients, treatment with an ICD costs $37 300 (if it reduces the total mortality rate by 40%) to $74 400 (if it reduces the total mortality rate by 20%) per additional quality-adjusted life-year saved relative to amiodarone therapy. Our study suggests that use of ICDs may be less cost-effective than was estimated in previous studies [28, 30-32]. In these studies, the cost-effectiveness of treatment with an ICD ranged from $30 500 [28] to $47 700 [30] per life-year saved, adjusted to 1995 dollars.

Our study differs from previous analyses in several important ways. First, we evaluated the cost-effectiveness of treatment with an ICD relative to empirical amiodarone therapy rather than to conventional therapy with type I agents or electrophysiologically guided therapy [74]. Amiodarone may be superior to type I agents in preventing sudden cardiac death, and it compares favorably with ß-blockers [10, 14, 62, 75, 76]. A recent randomized trial suggests that conventional, electrophysiologically guided pharmacologic and surgical therapy is unlikely to provide an effective, or cost-effective, alternative to the early use of an ICD [32, 49]. Second, we evaluated the use of third-generation ICDs (implanted without thoracotomy), examined strategies of crossover between amiodarone and ICD, explicitly examined the influence of quality of life on the cost-effectiveness of pharmacologic and device therapy, and evaluated the cost-effectiveness of ICD use in patients at intermediate risk for sudden cardiac death. Third, we estimated that ICD use reduced total mortality rates by 20% to 40% relative to amiodarone, a reduction that is less than that assumed in some previous studies.

The importance of the effect of ICD implantation on total mortality is apparent from our sensitivity analyses. Currently, the mortality benefit produced by an ICD relative to amiodarone therapy is uncertain. Two relatively small randomized trials have reported divergent results, and several trials, some of which are larger, are still ongoing [26, 27, 77]. In the CASH study [19], total mortality at 1 year was almost identical in patients who received amiodarone and those who received ICDs (14.7% in the former group and 14.3% in the latter group [16]), but this study is small and investigators have not reported any results of long-term follow-up. In contrast, MADIT [25] recently reported a 54% reduction in total mortality with ICD relative to conventional pharmacologic care. Our analysis assumed a risk reduction between these two estimates (20% to 40%). Our findings indicate that the cost-effectiveness of ICD use is strongly dependent on the reduction in mortality that the device provides. The evidence about the efficacy of ICD that will be provided by ongoing trials, such as the Canadian Implantable Defibrillator Study [27], the Coronary Artery Bypass Graft Patch trial [22], the Multicenter Unsustained Tachycardia trial [77], and the Arrhythmics versus Implantable Defibrillator trial [26] will greatly improve our understanding of the effect of ICDs on total mortality. If the device is ultimately shown to reduce the total mortality rate by 30% or more, the cost-effectiveness of ICD will be in the more favorable portion of the range that we have estimated.

A surprising finding was that the amiodarone-to-ICD regimen was expensive and resulted in only a small incremental benefit (0.02 to 0.03 quality-adjusted life-years) relative to amiodarone alone. The amiodarone-to-ICD regimen still has a relatively high mortality rate (similar to that seen with the amiodarone-only regimen) but is associated with the costs of amiodarone plus the costs of ICD in patients who have implantation. Our analysis suggests that if clinicians choose to treat patients with ICDs, early implantation is more cost-effective than delayed implantation. This finding is similar to those of previous analyses [29, 32].

We also found that the cost-effectiveness of ICD implantation changes only modestly when the device is used in patients who are at intermediate risk for sudden cardiac death. Although the likelihood of sudden cardiac death is lower in intermediate-risk patients, the longer life expectancy of these patients leads to gains in quality-adjusted life-years that are similar to those estimated for high-risk patients. In addition, if patients at intermediate risk for sudden cardiac death have a higher quality of life than do high-risk patients, ICD use would be more cost-effective than we have estimated. Scant information is currently available on the effectiveness of ICD in intermediate-risk patients, and thus our analysis is preliminary.

Our sensitivity analyses identified other important considerations in decisions about ICD use. The effect of alternative therapies on quality of life should be an important factor in therapeutic decisions. This effect has not been well studied, but it is being assessed in several ongoing clinical trials [26, 27] and in cohort studies being done as part of the Cardiac Arrhythmia and Risk of Death Patient Outcome Research Team. The frequency with which generators will have to be replaced strongly influences the cost-effectiveness of treatment with an ICD. Although device manufacturers predict improved generator longevity, little information is available on the lifespan of current devices. These uncertainties highlight areas for further study.

Is the use of an ICD cost-effective? Currently, there is no consensus about levels of expenditures that are cost-effective [78-82]. Our analysis suggests that using an ICD to prevent sudden cardiac death will cost more than $50 000 per quality-adjusted life-year gained unless the device reduces all-cause mortality by 30% or more. More precise estimates of cost-effectiveness will be possible when ongoing randomized trials have measured the effect of ICDs on all-cause mortality in large samples of patients.

Appendix

The decision model estimated quality-adjusted length of life and costs for cohorts of patients who received one of three regimens: ICD only, amiodarone only, and amiodarone-to-ICD. Each of the regimens is represented by a Markov model. A patient who received one of the regimens was at risk for the events shown in Figure 1 and Figure 2 during each 1-month cycle of the Markov model. During each cycle, the model also allowed for death from nonarrhythmic cardiac and noncardiac causes for patients who survived an arrhythmic event. The model assumed that arrhythmic events occurred at a constant rate. We assumed that patients in all strategies received appropriate non-antiarrhythmic therapy (for example, patients with congestive heart failure would receive angiotensin-converting enzyme inhibitors). For the base-case analysis, we adjusted the total mortality rates in the patients receiving ICD only to provide a reduction in the total mortality rate of 20% relative to the amiodarone-only regimen at the initiation of therapy. This relative risk reduction was approximately constant over time. For the ICD-only regimen, we assumed that generator replacement occurred deterministically at fixed intervals (4 years in the base-case analysis). We applied quality adjustments to each health state as noted in Table 1. We assumed that decreases in quality of life caused by therapies used to prevent sudden cardiac death resulted in an incremental decrement to the quality of life of current health. For example, if a patient assessed the quality of life with amiodarone therapy as 0.9 relative to their current health and rated the quality of life with their current health as 0.75 relative to perfect health, we assigned a utility of 0.9 x 0.75 = 0.675 to amiodarone therapy.

To estimate the costs of the treatment strategies, we used published data [30, 48], data from hospitals in northern California (Owens DK. Unpublished data), and analysis of Medicare claims (McClellan M. Unpublished data) (Table 1). Our upper and lower estimates of costs represented the most and least expensive of these different estimates. The base-case analysis used the average of costs at a private hospital, costs in a health maintenance organization, and published estimates [30]. We developed cost estimates for the private hospital and the health maintenance organization by using the TSI cost accounting system (Transition Systems, Inc., Boston, Massachusetts). We used a similar procedure to estimate the cost of maintenance therapy with an ICD. To estimate the annual cost of this therapy, we assumed that patients would require hospitalization to evaluate lead infection and lead failure at annual rates of approximately 2% and 3%, respectively [30], with corresponding costs per hospitalization of $25 800 [83] and $14 100 [17]. To estimate the annual costs of these events, we multiplied the annual probability of the event by the cost. We calculated an annualized cost for generator replacement of $5992 (including hospitalization costs and professional fees) on the basis of our base-case estimate of replacement every 4 years. Finally, we estimated the annual costs of outpatient visits as $683. The sum of each of these costs provides our total annualized cost ($7700) for treatment with an ICD.

Because the cost of maintenance therapy with amiodarone was not available from our hospital survey, we used published estimates of this cost [30], adjusted to 1995 dollars and modified to reflect current practice, such as the use of low-dose amiodarone therapy. The estimate assumed that of all patients treated with amiodarone, approximately 7% per year would require rehospitalization after surviving a ventricular arrhythmia (cost per hospitalization, $5615). We also included costs of laboratory tests ($970), drugs ($2136), and outpatient visits ($485) in our estimate of annual costs, for a total annual cost of approximately $4000 for amiodarone therapy. We separately estimated the costs associated with amiodarone toxicity. On the basis of our review of 11 randomized trials of amiodarone therapy [10-1214, 56, 59-64], we assumed that 10.0% of patients would discontinue amiodarone therapy during the first year because of intolerable side effects and that 5% would discontinue amiodarone therapy in each subsequent year. We assumed that 1.8% of patients would have acute mild toxicity ($350) and that patients who had to discontinue therapy because of toxicity would require hospitalization for evaluation ($7020).

From the Veterans Affairs Palo Alto Health Care System, Palo Alto, California; and Stanford University, Stanford, California.

Ms. Sanders and Dr. Dembitzer: Medical School Office Building X215, Stanford University, Stanford, CA 94305.

Mr. Harris and Ms. McDonald: Health Research and Policy, Redwood Building, Room 246, Stanford University, Stanford, CA 94305-5092.

Dr. Hlatky: Health Research and Policy, Redwood Building, Room 150, Stanford University, Stanford, CA 94305-5092.