Echocardiographic Identification of Cardiovascular Sources of Emboli To Guide Clinical Management of Stroke: A Cost-Effectiveness Analysis

Methods

Design

A Markov decision analysis [20] was done by using a commercially available computer program (Decision-Maker 7.0, Pratt Medical Group, Boston, Massachusetts) to evaluate nine diagnostic strategies in a hypothetical cohort of 65-year-old patients in normal sinus rhythm with new-onset stroke (Figure 1). (See the Glossary for definitions of terms.) The treat-none and treat-all strategies did not include imaging. In the all-transthoracic, all-transesophageal, and all-sequential strategies, imaging was done in all patients. In the selective-transthoracic, selective-esophageal, and selective-sequential-1 strategies, imaging was done only in patients who had a history of cardiac problems (left ventricular dysfunction or valvular disease). The selective-sequential-2 strategy used a sequential approach in patients with a history of cardiac problems and used transesophageal echocardiography in patients with no such history.

The cohort consisted of patients with four types of underlying pathologic condition: thrombi in the left atrium, other potential cardiac sources of emboli, aortic plaque only, and no identifiable cardiovascular source of emboli (Figure 2). Identification of the underlying condition depended on the imaging strategy used (Figure 1). For example, the all-transesophageal strategy identified all potential sources of emboli, the all-transthoracic strategy identified only some of the sources, and the treat-none strategy identified none of the sources. Anticoagulation was started on the basis of the results of the imaging studies. For the imaging strategies included in the base-case analysis, only patients with thrombus received anticoagulants; other patients received aspirin.

The cohort was then followed through monthly cycles for events, including recurrent cerebrovascular accident, intracranial hemorrhage, gastrointestinal bleeding, and death (Figure 2). Event and mortality rates were calculated on the basis of the underlying condition and treatment. Costs and outcomes for each health state were summed over the cycles. Patients in the cohort were followed over a lifetime horizon to provide the most relevant analysis from a societal perspective.

Major Clinical Assumptions

Several assumptions were necessary to implement the Markov model. First, patients had no obvious clinical cause of stroke (that is, no evidence of periprocedural stroke, recent myocardial infarction, prosthetic valve, known endocarditis, and so on). Second, patients were not receiving anticoagulants or antiplatelet agents at the time of stroke. Third, the subtype of ischemic stroke (such as lacunar or cortical) was independent of the underlying condition. Fourth, transesophageal echocardiography did not cause enough discomfort to decrease quality of life. Fifth, given the same underlying condition, the risk for recurrent stroke was independent of the history of cardiac disease (angina, valvular disease, and so on). Sixth, only the first major complication after stroke was considered in the analysis (for example, patients could not have both recurrent stroke and intracranial hemorrhage). Finally, new-onset and recurrent strokes were assumed to have similar relative effects on quality of life and on risk for death.

Results of Diagnostic Imaging and Occurrence of Subsequent Events

We performed a systematic review of the literature to gather the best available evidence about the prevalence of various cardiovascular sources of emboli in patients who have had stroke. A MEDLINE search from 1990 to 1995 was done by using cerebrovascular accident and transesophageal echocardiography as keywords (before 1990, transesophageal echocardiography was not a keyword). References from papers identified in the MEDLINE search were used to find relevant studies published before 1990. Studies were chosen if they reported the prevalence of potential cardiovascular sources of emboli identified by transesophageal echocardiography in a defined cohort of patients who had had stroke [8-1417, 21, 22]Table 1 and Table 4. Studies were distinguished with respect to four characteristics: patient selection (all patients who had had stroke rather than only patients referred for echocardiography), performance of intracardiac contrast studies, blinding of image readers to clinical history, and use of multiple readers. Only single-plane and biplane probes were used in the studies. Thrombi in the left atrium or left atrial appendage were more often identified in studies with selected patients, but other cardiac sources of emboli were identified at similar rates in selected and nonselected patients. Major findings on transesophageal echocardiography did not significantly differ among the studies with regard to use of echocardiographic contrast, blinded readers, or multiple readers. In the papers in which it was mentioned, positive cardiac history was generally defined as atrial fibrillation (patients with atrial fibrillation were excluded from our cohort), decreased left ventricular function, or valvular disease known at the time of transthoracic or transesophageal echocardiography. Rates of identification of aortic plaques varied widely, in part because of poor standardization of diagnostic criteria. Because no group of studies was clearly superior in design and conduct, base-case imaging results were obtained from the mean of the results weighted by the number of participants in each study.

Secondary event rates and mortality rates (including the rate of death from nonvascular causes) were determined by using the best available estimates in the literature (Table 2) [1-418, 23-30]. The estimated relative risk reduction for recurrent stroke (that is, efficacy) with anticoagulation ranged from 0% [23] to 84% [24]. Meta-analyses of studies of patients with atrial fibrillation estimated the efficacy of anticoagulation to be 60% to 67% and the efficacy of aspirin to be 33% compared with that of no therapy [18, 25]. Ezekowitz and colleagues [26] estimated a relative risk reduction of 40% for anticoagulation in patients with a history of atrial fibrillation and stroke, a patient population that may be similar to our patients who had stroke and documented thrombus in the left atrium or left atrial appendage. We chose a relative risk reduction of 33% for anticoagulation compared with aspirin and used a wide range (17% to 67%) for sensitivity analysis.

Prospective information on the incidence of intracranial hemorrhage with anticoagulation after a stroke is also scarce. In a previous decision analysis, Eckman and colleagues [27] estimated the annual incidence of major bleeding to be 4.5% with anticoagulation. In a case series, Landefeld and Goldman [28] found that one third of all major bleeding episodes with anticoagulation were intracranial hemorrhage. Thus, we used a 1.5% annual incidence of anticoagulant-related hemorrhagic stroke. This estimate closely reflects that found in prospective studies of patients with atrial fibrillation that showed an overall incidence of about 1% per year, with higher incidence in patients with a history of stroke [24]. Because of the uncertainty and importance of this factor, we chose a wide range for sensitivity analysis (0.5% to 4.5%).

Cost and Utilities

Direct medical costs and indirect costs due to lost productivity after stroke were assigned on the basis of the estimated cost to society (Table 2). Professional and technical costs for transesophageal and transthoracic echocardiography were calculated on the basis of Medicare reimbursement rates. Other relevant costs were obtained from the literature [31-37], converted to 1995 dollars [38], and discounted at 3% per year [39].

Utilities are numerical estimates of quality of life in various conditions of health. We obtained utilities from a previous study [40] of patient preferences. Because present health is usually valued more highly than future health, the utilities for future health states were discounted at 3% per year [39]. All utilities and associated assumptions were assessed in a sensitivity analysis. Details about costs and utilities are given in the Appendix.

Analyses

The base-case analysis was first performed by using the best available input data and the conservative assumption that only patients with a thrombus in the left atrium or left atrial appendage were treated with and benefited from anticoagulation. A cost of less than $20 000 per quality-adjusted life-year was considered cost-effective, and a cost of greater than $100 000 per quality-adjusted life-year was considered not cost-effective. Additional analyses were performed with less conservative assumptions and over shorter time horizons. Sensitivity analyses were performed to determine whether clinical outcome was sensitive to any of the input variables.

Results

Base-Case Analysis

The base case was analyzed for each of the nine diagnostic strategies by using the best estimates of the input variables (Figure 1). We used the conservative assumption that indication for anticoagulation was limited to thrombus, except in the treat-all strategy.

Efficacy of Diagnostic Strategies and Cost of Testing

The most efficacious strategies (those that identified the most patients with thrombus in the left atrium or left atrial appendage) were the all-transesophageal, all-sequential, and selective-sequential-2 strategies. The selective-transesophageal and selective-sequential-1 strategies identified 59% of these patients. The all-transthoracic and selective-transthoracic strategies identified only 20% and 12% of these patients, respectively. The two strategies that did not use imaging (treat-none and treat-all), did not identify patients with thrombus.

The all-sequential strategy had the most expensive tests and cost $686 per patient on average. The average test cost for the selective-sequential-2 strategy was $481 per patient. For the all-transesophageal and all-transthoracic strategies, the average costs per patient were $360 and $330, respectively. The selective-sequential-1 strategy cost $254 per patient on average, and the selective-transesophageal and selective-transthoracic strategies cost an average of $133 and $122 per patient, respectively. The treat-none and treat-all strategies did not incur any testing costs because imaging was not done.

Cost-Effectiveness of Strategies with Follow-up

On the basis of the diagnostic yield of the strategies, the patients in the hypothetical cohort were treated and followed throughout their lifetimes (Figure 2). Treat-none was the least aggressive and the least expensive strategy. The cost of recurrent events using this strategy was $4740 per patient who, on average, had 4.854 quality-adjusted life-years. Each of the other diagnostic strategies was then compared with the treat-none strategy (Figure 3). Relative costs per quality-adjusted life-year for the selective-transesophageal and all-transesophageal strategies were modest (about $9000 and $13 000, respectively). Although these strategies had relatively high initial test costs, they achieved improved outcomes. The sequential strategies that used transthoracic echocardiography followed by transesophageal echocardiography (selective-sequential-2, selective-sequential-1, and all-sequential) cost $24 000 to $32 000 per quality-adjusted life-year. These strategies had higher initial cost (because more tests were done) without improvement in outcomes compared with strategies that used transesophageal echocardiography only. The strategies that used transthoracic echocardiography only (selective-transthoracic and all-transthoracic) cost about $36 000 and $57 000 per quality-adjusted life-year, respectively. Although the strategies that used only transthoracic echocardiography had lower test costs, patients had worse outcomes, resulting in higher subsequent costs. Finally, the most aggressive strategy, treat-all, led to a high incidence of intracranial hemorrhage, resulting in higher cost and worse outcomes than any other diagnostic strategy. In general, compared with the treat-none strategy, strategies that used only transesophageal echocardiography had more favorable cost-effectiveness ratios than those that used a sequential approach and those that used transthoracic echocardiography only.

It was also useful to compare strategies with each other rather than with the treat-none strategy. By incremental cost-effectiveness, the all-transesophageal strategy cost $20 000 more per quality-adjusted life-year than the selective-transesophageal strategy (Table 3). In other words, it cost $9000 per quality-adjusted life-year to perform transesophageal echocardiography in patients with a history of cardiac problems and $20 000 per quality-adjusted life-year to perform transesophageal echocardiography in patients without such a history; both are within a reasonable range of cost-effectiveness. Direct comparison found that the all-transesophageal strategy had lower cost and was more effective than all other strategies except selective-transthoracic, which was clearly less cost-effective.

Effect of Time Horizon

For the base-case analysis, patients in the cohort were followed over their expected lifetimes. The cost-effectiveness of the strategies at earlier points in time was determined by doing a cumulative analysis after each additional month had elapsed after the initial stroke and implementation of the strategy. Both the selective-transesophageal and all-transesophageal strategies cost less than $100 000 per quality-adjusted life-year in less than 2 years and less than $20 000 per quality-adjusted life-year in less than 8 years (Figure 4).

Sensitivity Analysis

Our results were most sensitive to two factors: efficacy of anticoagulation (relative reduction in rate of recurrent stroke with anticoagulation) and rate of intracranial hemorrhage with anticoagulation (Figure 5). They were also mildly sensitive to prevalence of thrombus on transesophageal echocardiography, rate of recurrent stroke in patients with thrombus who did not receive anticoagulants, utility of the baseline stroke state, cost of transesophageal echocardiography, and specificity of transesophageal echocardiography. Of note, the base-case results were not sensitive to estimates of the sensitivity of transesophageal echocardiography, to the cost of recurrent events or complications, or to the discounting factor.

The relation between efficacy of anticoagulation and occurrence of intracranial hemorrhage with anticoagulation was analyzed by varying both inputs simultaneously in a two-way sensitivity analysis (Figure 6). The value at which the cost-effectiveness of doing transesophageal echocardiography in all patients approached $20 000 per quality-adjusted life-year or $100 000 per quality-adjusted life-year for one of the two variables was highly dependent on the value of the other. For example, if anticoagulation in patients with thrombus reduced stroke recurrence by 40%, the all-transesophageal strategy would remain cost-effective as long as the rate of intracranial hemorrhage remained less than approximately 4%.

Effect of Sensitivity of Transthoracic Echocardiography

Compared with the all-transesophageal and selective-transesophageal strategies, strategies that used transthoracic echocardiography alone or in sequence with transesophageal echocardiography were not cost-effective in sensitivity analysis. Even at half the examination cost, transthoracic echocardiography would have to identify at least half as many thrombi as transesophageal echocardiography to be cost-effective; this is unlikely, given the high proportion of thrombi in the left atrial appendage.

Effect of Clinical History

To evaluate the effect of clinical history, we varied the proportion of thrombi that would be found in patients with a history of cardiac problems. In the base-case model, 59% of thrombi were assumed to be present in patients with such a history [8-11, 13]. When we increased this proportion to 80%, the incremental cost-effectiveness ratio for the all-transesophageal strategy compared with the selective-transesophageal strategy increased to $39 000 per quality-adjusted life-year. Decreasing this proportion improved the ratio for the all-transesophageal strategy.

Discussion

Our analysis of cardiac imaging strategies for patients who have had stroke and are in normal sinus rhythm strongly suggests that the initial cost of transesophageal echocardiography is substantially offset by improved outcome and reduction in resource use for management of recurrent stroke. The cost-effectiveness of anticoagulation as determined on the basis of results of transesophageal echocardiography in all patients after initial stroke ($13 000 per quality-adjusted life-year) compares favorably with a commonly used threshold for cost-effectiveness, $50 000 per quality-adjusted life-year for hemodialysis treatment for end-stage renal disease [41, 42] and with other treatments in cardiovascular patients [43, 44]. For example, the cost-effectiveness of the pharmacologic treatment of hypertension has been estimated to be $12 000 to $64 000 per quality-adjusted life-year, that of electrocardiographic exercise testing has been estimated to be $30 000 to $124 000 per quality-adjusted life-year, and that of the pharmacologic treatment of asymptomatic hypercholesterolemia has been estimated to be $48 000 to $2 000 000 per year of life saved. Our findings are consistent with those of Seto and colleagues [45], who recently showed that transesophageal echocardiography was cost-effective before cardioversion for patients with atrial fibrillation.

Three factors may explain the favorable cost-effectiveness of transesophageal echocardiography. First, transesophageal echocardiography is very sensitive and relatively specific for identifying left atrial thrombus [7, 46]. Second, the cost of transesophageal echocardiography is relatively low ($360). Even when we increased the examination cost twofold, the cost-effectiveness ratio remained within a reasonable range. Third, and most important, cost savings and outcome improvement from a small reduction in the rate of recurrent stroke compensate for the initial cost of transesophageal echocardiography. In a sensitivity analysis, a wide range of costs for subsequent events did not substantially alter our results. Overall, the balance of complications (such as intracranial hemorrhage) compared with prevented recurrent strokes influenced the results more strongly than the cost of transesophageal echocardiography. These results are similar to the favorable cost-effectiveness of warfarin found in analyses of primary stroke prevention in patients with atrial fibrillation [25, 27, 47, 48].

Although transthoracic echocardiography is widely used, no strategy using this test (alone, sequentially with transesophageal echocardiography, or selectively in patients with a history of cardiac problems) was found to be cost-effective. This was because many thrombi are found in the left atrial appendage, an area that is rarely visualized on transthoracic echocardiography [8-1114, 49]. Neither increasing the sensitivity of transthoracic echocardiography nor decreasing its cost sufficiently improved its cost-effectiveness. Even extending the potential benefits of anticoagulation therapy to all cardiac sources of emboli did not make transthoracic echocardiography cost-effective compared with transesophageal echocardiography. In addition, we assumed that the specificity of transthoracic echocardiography is high and similar to that of transesophageal echocardiography. This conservative assumption may not be true, and the cost-effectiveness of transthoracic echocardiography may be even less favorable than our results suggest.

The ambiguity that surrounds what constitutes a relevant history of cardiac problems precludes definitive ascertainment of the role of selective imaging. The prevalence of thrombi in patients with various cardiac histories has not been adequately determined. For these reasons, we chose to compare nonselective transesophageal echocardiography in all patients with no imaging. Future research may identify subgroups of patients in whom selective imaging is more cost-effective.

The selective-transesophageal and all-transesophageal strategies had reasonable costs per quality-adjusted life-year within approximately 3 years after initial stroke (Figure 4). This result is important for two reasons. First, if any of the clinical or economic variables change substantially over time, shorter horizons may be more relevant. In addition, cost-conscious managed care organizations may emphasize earlier “return on investments” from initial outlays.

In the sensitivity analysis, the two most influential factors were the efficacy of anticoagulation (relative reduction in recurrent stroke with anticoagulation) and the rate of intracranial hemorrhage with anticoagulation. Our analysis confirms a common clinical impression. After embolic stroke, patients are at high risk for subsequent emboli [17], providing an appropriate subgroup for secondary prevention with anticoagulation. However, these patients also have increased risk for anticoagulation-induced hemorrhagic strokes. In general, the stroke risk reduction and the rate of intracranial hemorrhage are very sensitive to the extent of anticoagulation as measured by the international normalized ratio [50]. The values used in our analysis reflected an attempt to estimate rates in the targeted range for the international normalized ratio; substantial deviation in actual practice would probably greatly affect both of these rates (Figure 6).

The major limitation of our analysis was our reliance on the best available data in the literature; these data were often retrospective and heterogeneous. For example, the criteria for what constitutes a history of cardiac problems varied among the published studies [8-14]. Prevalence rates of lesions identified by transesophageal echocardiography have been fairly well studied, but more prospective studies of event rates, health state utilities, and costs, particularly after complications, are needed. Age is also an important factor in the current clinical recommendations of cardiac imaging after stroke [19]; however, because lesion prevalence, event rates, utilities, and costs were not adequately reported by age groups, subgroup analysis was in-appropriate. Finally, our analysis approaches the issue of secondary prevention from a societal perspective, whereas decision making in the clinical setting primarily considers the risks and benefits to the individual patient. We hope that our sensitivity analysis will allow physicians to make judgments on the basis of appraisal of event rates and circumstances of individual patients.

In conclusion, despite higher initial costs, transesophageal echocardiography in patients who had new-onset stroke and were in normal sinus rhythm was cost-effective compared with other commonly used diagnostic and therapeutic strategies. Cardiac history may play a role in selecting patients for transesophageal echocardiography. Conversely, transthoracic echocardiography was not cost-effective either alone or in combination with cardiac history or transesophageal echocardiography. Two factors were particularly influential and deserve more concentrated research: the efficacy of anticoagulation in patients who have had stroke and who have documented thrombus and the rate of intracranial hemorrhage in patients who have had stroke and are receiving anticoagulants.

Appendix

Costs

Costs of transthoracic echocardiography and transesophageal echocardiography were obtained from the Medicare reimbursement schedule for Johns Hopkins Hospital. These rates are substantially lower than those of most private indemnity rates; considering the advanced age of most patients who have had stroke and the increasing presence of managed care, our estimated costs may be more relevant than those of private agencies. The cost of anticoagulation was obtained from a decision analysis of anticoagulation in heart disease [27] and was calculated on the basis of six ambulatory visits per year and two prothrombin times per month.

Costs associated with recurrent stroke were obtained from the median values found in the relevant studies. Specifically, the cost associated with acute hospitalization for stroke was obtained from a study by Lee and colleagues [35], who examined a sample of Medicare patients admitted to an acute-care hospital with a diagnosis of stroke during the first 6 months of 1991. The cost of the first year after recurrent stroke was obtained from a study by Ashraf and coworkers [34], who used data from the Agency for Health Care Policy and Research and the 1995 Physicians Fee and Coding Guide. This cost was adjusted to account for nonprocedural and nonphysician costs, such as nursing home and home health care costs [37]. The cost of subsequent years after a recurrent stroke was obtained from a study by Eckman and associates [27], who used outpatient costs from the Tufts Associated Health Plans, Inc.

The costs associated with intracranial hemorrhage were estimated to be 60% higher than the cost of an ischemic stroke based on data from Lee and coworkers [35]. The cost of gastrointestinal bleeding was estimated from a study by Oster and colleagues [36], who used reimbursement rates from the Health Care Financing Administration Common Procedure Coding System. Because of substantial uncertainty about these costs, we halved and doubled each of the costs for our sensitivity analyses.

Utilities

The utilities for the health states associated with stroke were estimated from the literature. In most of the previously reported decision analyses, the initial health state was assumed to be that of a completely healthy person and was assigned a utility value of 1.0 [20]. Other utilities were then determined in relation to this healthy state. In our analysis, however, patients in the initial state had already had stroke. Thus, we chose to assign this initial state according to the criteria in the study by Solomon and associates [40], who determined utilities by obtaining preferences for various health states in stroke from outpatients visiting the cerebrovascular laboratory. They asked patients to rank various stroke scenarios. From these rankings, the patients were asked to assign a value of 0.50 to the scenario midway between perfect health and death. Values for other scenarios were then assigned by the patients, using the values for perfect health, death, and the midway scenario as reference points. Solomon and associates found somewhat lower utilities than have been published in other studies [48, 51-54], ranging from 0.03 for severe hemiplegia to 0.54 for mild dysarthria. Thus, we used the highest estimate in the study by Solomon and associates but evaluated a wide range during sensitivity analysis.

Utilities have been shown to change over time for such conditions as rheumatoid arthritis [55] and breast cancer [56] but to remain stable in such conditions as chronic renal failure requiring dialysis [57] and myocardial infarction [58]. To our knowledge, the time course of variation of the utility in patients who have had stroke has not been assessed. In the absence of complications, stroke utilities would be expected to increase or stay the same over time. To estimate an initial improvement in patients with stroke who are frequently seen clinically [1], the baseline utility of 0.54 was increased at a fixed rate over the first 6 months, reaching a utility of 0.77 (halfway between the healthy-person utility of 1.0 and the recent-stroke utility of 0.54).

We assumed that the occurrence of a complication of stroke would decrease utility to the same relative degree as the same complication would in a healthy person; this assumption is conservative by some estimates. Gage and coworkers [48] estimated a utility of 0.12 for recurrent strokes by using a survey. To account for potential clinical improvement, the utility in patients with recurrent stroke was assumed to be worse in the first year than in subsequent years. Death was assigned a utility of zero.

Glossary

Diagnostic Strategies

All-sequential: Transthoracic echocardiography was done in all patients who had had stroke, and transesophageal echocardiography was done in patients who had negative findings on transthoracic echocardiography.

All-transesophageal: Transesophageal echocardiography was done in all patients who had had stroke.

All-transthoracic: Transthoracic echocardiography was done in all patients who had had stroke.

Selective-sequential-1: Transthoracic echocardiography was done in patients who had had stroke and who had a history of cardiac problems, transesophageal echocardiography was done in patients with negative findings on transthoracic echocardiography, and no echocardiography was done in patients who did not have a cardiac history.

Selective-sequential-2: Transthoracic echocardiography was done in all patients who had had stroke and who had a history of cardiac problems, and transesophageal echocardiography was done in patients who had negative findings on transthoracic echocardiography and all patients who did not have a history of cardiac problems.

Selective-transesophageal: Transesophageal echocardiography was done only in patients who had had stroke and had a history of cardiac problems.

Selective-transthoracic: Transthoracic echocardiography was done in all patients who had had stroke and who had a history of cardiac problems.

Treat-all: No imaging was done, but all patients received anticoagulants.

Treat-none: No imaging was done, and no anticoagulants were given.

Terms

Discounting: The process of counting future costs and utilities less than present cost and utilities to account for time preferences for resource outlays or health status.

Markov model: A type of decision analysis model in which a patient is in one of a finite number of health states and in which transitions from one state to another occur over time.

Quality-adjusted life-year: A measure of effectiveness or health outcome that considers both quality and length of life.

Utility: A numerical estimate of quality of life in a particular health state.

From The Johns Hopkins Medical Institutions, The Johns Hopkins University School of Hygiene and Public Health, and The Johns Hopkins University School of Medicine, Baltimore, Maryland; and Tulane University Medical Center, New Orleans, Louisiana.

Acknowledgments: The authors thank colleagues at The Johns Hopkins Medical Institutions and The Johns Hopkins University School of Medicine for feedback on our analysis and Mark Danese, MHS, for helpful comments.

Dr. Lima: Johns Hopkins Hospital, Blalock 569, 600 North Wolfe Street, Baltimore, MD 21287.

Dr. Whelton: School of Public Health and Tropical Medicine, Tulane University Medical Center, 1501 Canal Street, New Orleans, LA 70112.

Dr. Powe: Welch Center for Prevention, Epidemiology, and Clinical Research, The Johns Hopkins Medical Institutions, Suite 2-645, 2024 East Monument Street, Baltimore, MD 21250-2223.