Methods

Top Methods Results Discussion References

Decision Model

We used standard cost-effectiveness analysis methods [7, 8] to evaluate a screening program that would identify high-grade carotid stenosis in a cohort of 65-year-old men who did not have such neurologic symptoms of carotid disease as transient ischemic attack, reversible ischemic neurologic deficit, amaurosis fugax, or previous stroke. Our decision tree compared screening for carotid disease done using duplex Doppler ultrasonography with no screening. If significant carotid stenosis ( 60%) was found on ultrasonography, disease would be confirmed by angiography. Patients who have angiography bear a risk for a major complication, such as stroke; we assumed that patients who had stroke or another major complication during the evaluation were no longer candidates for surgery. If angiography had a positive outcome and was uneventful, carotid endarterectomy would be done.

A patient having carotid endarterectomy could have several outcomes. First, the operation could be uncomplicated and technically successful. Second, death could result from surgery. Third, a nonfatal stroke could occur. Fourth, the operation could be technically successful but could result in a nonfatal myocardial infarction. Transient complications of surgery were not explicitly modeled; Figure 1 shows a schematic of the decision model.

View larger version (9K): [in this window] [in a new window]   Figure 1. Decision tree schematic showing options and outcomes of screening compared with no screening for carotid stenosis. The square node indicates a decision; round nodes indicate chance outcomes; M nodes represent Markov models of health states.

Data software, version 2.5 for Macintosh (Tree-Age Software, Inc., Boston, Massachusetts), was used to build the decision tree and to do model calculations for initial 5-year analyses. Programming for extended time horizon modeling was done with Excel, version 4.0 (Microsoft Corp., Redmond, Washington).

Assumptions

Our evaluation included direct medical costs; indirect societal costs, such as loss of work, were omitted. Our baseline case considered screening in a cohort of 65-year-old men. Men were chosen because they have higher rates of stroke and carotid disease than women do [9] and because about two thirds of the persons in ACAS were male. We modeled for a 65-year-old person because 65 years was the approximate mean age of patients in ACAS and because risk for stroke increases rapidly between 60 and 70 years of age [9]. Because of this increase, screening was expected to have the greatest benefit when applied to 65-year-old men: If screening was not cost-effective for this group it could not be cost-effective for any group. As did ACAS, we defined carotid stenosis as an occlusion of 60% or more of at least one carotid vessel. Outcomes of stroke (of varying levels of severity), deaths, and lasting complications of angiography and endarterectomy were the health state outcomes of interest. Myocardial infarction due to endarterectomy was also included as an outcome. Transient ischemic attacks were not included as outcome measures because they are by definition transient and by themselves affect neither mortality nor quality of life.

No empirical data exist on the efficacy of endarterectomy more than 5 years after surgery. Because surgery imposes short-term risks and costs while offering the promise of future benefit, we evaluated the cost-effectiveness of screening under the assumption that surgery has prolonged benefits. To do this, we used extended time-horizon modeling, in which the time horizon was the lifetime of the cohort. Beyond the 5 years studied by ACAS, age-specific mortality is modeled correctly and reasonable assumptions about disease-specific (in this case, stroke-specific) mortality and morbidity are made [10]. We assumed that the reduction in the rate of stroke conferred by surgery would diminish gradually over time, and we used an exponential hazard function to model this decline. For the baseline case, we assumed that 10 years after surgery, the survivors of endarterectomy who had not had stroke would have the same rate of stroke as a person without stroke who did not have surgery. Using this assumption, our model determined that the survival curves would equalize at year 30 (and therefore the survival advantage conferred by surgery would disappear). We then did a sensitivity analysis that modeled for rates of stroke equalizing at years 20 and 30 (the lifetime of the entire cohort).

In our model, duplex ultrasonography was chosen as the screening method because it is considered to be one of the best methods for detecting carotid occlusive disease, because it is well tolerated and safe, and because its use is accepted practice for the initial evaluation of carotid stenosis. The sensitivity and specificity ranges for duplex ultrasonography are approximately 81% to 90% and 82% to 95%, respectively [11, 12]. Conventional carotid angiography is considered to be the reference diagnostic test for the evaluation of carotid artery disease. In current clinical practice, it is recommended that angiography be done before carotid endarterectomy [11]. Although ACAS reported that angiography had a complication rate of 1.2%, large series [13, 14] have shown rates of 0.45% to 1.3% for transient neurologic complications, 0.1% for permanent neurologic deficits, and 0% to 0.1% for death. Patients with carotid ischemic disease, however, can have an increased risk for complications [14]. Magnetic resonance angiography is rapidly gaining acceptance as a method for the diagnosis of extracranial vascular disease, but most clinicians do not consider it sufficient to substitute for contrast angiography in evaluation before endarterectomy. Recent series show that, compared with conventional angiography, magnetic resonance angiography has a sensitivity of 84% to 92% and a specificity of 76% to 81% [15-17].

The Glasgow Neurologic Outcome Scale was used in ACAS to define stroke severity [18]. Our model classified strokes as either major (Glasgow score of 2 to 4) or minor (Glasgow score of 1), consistent with the categories of stroke used in ACAS. We assumed that the distribution of strokes in the major and minor categories was the same as that in ACAS. To capture differences in cost of stroke care, major strokes were further designated as "severe" or "moderate." We defined severe stroke as the presence of permanent neurologic deficits leading to institutionalized care. This encompassed Glasgow scores 3 and 4, which are respectively described as "severe disability with lack of independence" and "persistent vegetative state." Although follow-up care is also required after a moderate stroke, the costs of this care are much less than those of severe stroke. Moderate stroke (Glasgow score of 2) corresponds to "moderate disability but with functional independence" [18]. Gresham and associates [19] found that approximately 15% of clinically relevant strokes led to institutionalization. Therefore, 15% of strokes were considered to be severe. Data from ACAS showed that 53% of strokes were mild. The remaining 32% were considered to be moderate. This distribution was used to determine costs of stroke care and the future likelihood of death.

Markov modeling [20, 21] was used to estimate annual transitions to different health states. On the basis of the presence or absence of stenosis and whether the patient had had surgery, persons were assigned probabilities of remaining in good health or moving into the stroke or death states; once in a stroke state, a person could remain in the same state or proceed to a more severe stroke state or to death [6, 9, 22, 23].

In the following discussion of specific Markov states for the rightmost branches of our decision tree, the states are ordered from mildest to most severe. Patients are distributed among three possible Markov states for major complications of angiography: minor stroke, major stroke, and death. The rule of thumb is that each year, patients in each group have a transition probability of staying in the same state or moving to a more severe health state but cannot move to a less severe state. For example, patients with minor strokes can remain within the minor stroke state or move into the major stroke state or the death state. Patients with major strokes can stay in the major stroke state or move into the death state. Once in the death state, patients remain there. If sufficient time should pass, everyone will be in the death state. Patients having endarterectomy will be distributed over the well, minor stroke, major stroke, and death states. In addition, these patients can start out in a myocardial infarction state but cannot "transition in" to the myocardial infarction state from other states. Patients starting in the myocardial infarction state can remain in that state or can move into any other state except the well state. Once in another state, they follow the usual rule for direction of movement. Patients who had no procedures and no complications from angiography are distributed over the well, minor stroke, major stroke, and death states. These distributions differ according to whether patients have a carotid lesion 60% or greater.

The Gompertz exponential survival function derived from the life Table ofthe general population was used to determine mortality rates of the initial 5 years for members of the cohort with carotid stenosis less than 60% [24]. See Table 1 for the equation. The variable "age" in the Equation is65 years; the variable "year" is the number of years after the age of 65 years. For lifetime modeling, we used the life Table formen [33]. For deaths from major and minor strokes, we used rates reported in the literature [22].

Costs

Because accurate data on costs were not available, charges were used as a proxy for costs. Hospital and professional charges for screening, angiographic confirmation of disease, surgery and hospitalization, stroke work-up and acute treatment, follow-up care for stroke (which depends on the clinical course of the patient but can include use of a rehabilitation unit, outpatient visits, outpatient professional services, physical and speech therapy, home health care, appliances, and aspirin) were included. These incorporated medical consequences other than stroke but omitted opportunity costs, such as loss of work and other nonmedical economic consequences. All monetary outcomes were adjusted to 1994 U.S. dollars using the medical component of the Consumer Price Index rates; a 3% annual discount rate was used for the adjustment of future costs [34]. For the purposes of calculation, the model assumes that events occur in the middle of each year. Charges related to carotid endarterectomy were calculated from the California Office of Statewide Health Planning and Development (OSHPD) 1991 database.

Utilities

Because both stroke and the side effects of screening and surgery can decrease quality of life, we assigned utility values to each health state; utilities ranged from zero (equivalent to death) to unity (equivalent to life without disability). Stroke utility values were obtained from a survey using the time-tradeoff method to measure the preferences of patients at risk for stroke [35]. The utility of acute myocardial infarction was obtained from a time-tradeoff study in survivors of this event [36]. Future utilities were discounted at a rate of 3% [34]. When costs are discounted, it is recommended that health benefits (such as utilities) be discounted at the same rate [37, 38]. When benefits are discounted, years of life saved in the future are given less value than are years of life saved earlier. The rationale for discounting benefits is not to demean the value of future life. Rather, because lives and other utilities are being valued relative to dollars and because dollars are discounted to their present value, benefits must be discounted at the same rate [39]. In addition, Keeler and Cretin [38] show how the failure to discount utilities when costs are discounted leads to logical inconsistencies. When costs are discounted and utilities are not, it is always better to delay initiation of an intervention so that no program with a finite starting date can be selected. It is thus impossible to make a decision to begin a program on the basis of its cost-effectiveness.

Data Summary

The baseline model incorporated the following information from the ACAS report: rate and types of complications from carotid endarterectomy, annual rate of stroke, and mortality rates for patients having carotid endarterectomy and patients with untreated carotid stenosis. To determine the rate of stroke in the group of patients with untreated carotid stenosis, we took the rate of stroke from the nonsurgical arm of ACAS. That cohort was treated with 325 mg of enteric-coated aspirin. Cote and colleagues [40] found that aspirin did not have a substantial long-term protective effect in asymptomatic patients who had carotid stenosis of greater than 50%; thus, empirical treatment with aspirin should not have affected the rate of stroke in the medical arm of ACAS. For other values, we relied whenever possible on published medical literature that reflected current technology and was specific to elderly men [9, 22, 25, 31, 32, 35, 36, 41]. When a range of values was reasonable, we selected a value that would favor screening. Data on costs were taken from a review of admissions for elective carotid endarterectomy in California [42], literature identified through a search of the MEDLINE database [11, 43-47], published average charges for Medicare diagnosis-related groups [48], and the 1994 Physicians Fee and Coding Guide [49]. Values used are shown in Table 1 and Table 2.

Charges related to carotid endarterectomy were calculated from the OSHPD 1991 database. Data for each hospitalization include demographic variables for patients (age, sex, race, ZIP code of residence, and county of residence); admission information; diagnosis-related groups; principal payment sources; total charges; and as many as 24 International Classification of Diseases, Ninth Revision (ICD-9), diagnosis fields and 20 ICD-9 procedure fields. To derive an appropriate cohort, we selected admissions that had a primary principal procedure code for carotid endarterectomy [ICD-9 code 38.12] and met the following conditions: 1) The patient was between 60 and 70 years of age, 2) the length of hospital stay was between 3 and 7 days, 3) there were 2 days or less between admission and procedure, 4) the admission was coded as "routine" [that is, the patient was not admitted through the emergency department], 5) the admission was coded as "elective," and 6) the patient did not have coronary artery bypass graft surgery or angiography during the admission. Furthermore, any record with a total charge of $0 was excluded.

We examined this pool of in-patient admissions for diagnosis codes that possibly indicated the occurrence of a stroke from surgery. We found 818 "uncomplicated" carotid endarterectomy admissions and 247 carotid endarterectomy admissions that had an ICD-9 code for stroke. Because the database lacks temporal information about procedures and diagnoses, we could not determine whether these 247 cases represented patients who presented with small strokes and received endarterectomy to prevent other cerebral events or patients who were initially asymptomatic but had a procedure-related event. Because the mean total charges of all uncomplicated carotid endarterectomy admissions were similar for men ($13 500) and women ($13 400), we used the combined mean charge of $13 400 in our analysis. The mean total charge for all admissions for carotid endarterectomy with stroke was $25 800. The difference between these two mean charges (approximately $12 400) was considered to represent the cost of acute care for a stroke induced by carotid endarterectomy.

Results

Top Methods Results Discussion References

Baseline Analysis

Over a lifetime (30 years), screening resulted in an additional cost of $1553 per person and generated 0.013 quality-adjusted life-years, or 4.75 days, more than the no-screening strategy, producing a cost of $120 000 per quality-adjusted life-year. This ratio is the additional cost for the gain in quality-adjusted life-years between the screening strategy and the no-screening strategy and is known as the marginal cost-effectiveness of screening relative to no screening.

Sensitivity Analysis

We explored the effect on marginal cost-effectiveness produced by varying prevalences of disease, costs of screening and surgery, procedure complication rates, and durations of reduction in risk for stroke from endarterectomy. Key findings are shown in Table 3. The marginal cost-effectiveness was sensitive to disease prevalence in the screening population. If a population with a 40% prevalence of carotid stenosis could be identified, the cost-effectiveness ratio of screening decreased to $50 000 per quality-adjusted life-year. We simulated future trends that could decrease marginal cost-effectiveness. Published reports assert that the costs of carotid endarterectomy can be reduced by 50%, primarily by reducing length of hospital stay by doing carotid endarterectomy under local instead of general anesthesia and by selectively using postoperative intensive care [50, 51]. We found that endarterectomy and hospitalization at half the cost would reduce the marginal cost-effectiveness to $91 000 per quality-adjusted life-year. If a new screening technique were free, marginal cost-effectiveness would decrease to $77 000 per quality-adjusted life-year. To anticipate new trends in screening technology, a complication-free, perfectly accurate screening method that cost no more than Doppler ultrasonography (eliminating the need for angiography) was modeled; when this was done, the marginal cost-effectiveness of screening became $71 000 per quality-adjusted life-year.

In examining the effects of procedure complication rates, we found that as the complication rate of angiography approached 2%, the no-screening strategy became dominant over the screening strategy; this means that not screening patients (that is, providing usual care) generated more quality-adjusted life-years and was less expensive than screening. If the complication rate of endarterectomy were reduced to zero, marginal cost-effectiveness would be $90 000 per quality-adjusted life-year. If this complication rate were 2%, however, the marginal cost-effectiveness would be $142 000. With the more typical complication rates of 3% and 4%, the marginal cost-effectiveness ratios are $196 000 and $318 000, respectively.

If reduction in rates of stroke from endarterectomy were prolonged to 20 years, the marginal cost-effectiveness would be $96 000. If this reduction were prolonged for the lifetime of the cohort (30 years), marginal cost-effectiveness would be $74 000 per quality-adjusted life-year.

Discussion

Top Methods Results Discussion References

The ACAS investigators showed that carotid endarterectomy done under carefully controlled conditions can reduce the risk for stroke in patients with asymptomatic carotid stenosis. Asymptomatic carotid stenosis can be systematically identified only through screening programs. Therefore, the results from ACAS suggest that screening may be beneficial. The exact limits of acceptable cost-effectiveness are controversial, but interventions that cost $50 000 or less per quality-adjusted life-year are generally acceptable. Interventions that cost $50 000 to $100 000 are borderline, and those costing more than $100 000 are not cost-effective [52]. Another way to examine this issue is to look at the marginal cost-effectiveness of accepted interventions. For example, screening for hypertension costs $12 200 to $42 600 per year of life saved (in 1993 U.S. dollars) [53], coronary artery bypass graft surgery for left main artery disease costs $6300 to $7000 per quality-adjusted life-year (in 1991 U.S. dollars) [54], and renal dialysis costs $30 000 to $35 000 per year of life saved (in 1990 U.S. dollars) [55].

Our analyses indicate that according to generally accepted criteria, screening for asymptomatic carotid stenosis to identify candidates for carotid endarterectomy would be less cost-effective than most accepted health interventions. Our baseline case generated a marginal cost-effectiveness ratio of $120 000. In a sensitivity analysis, we examined factors that might make screening more cost-effective: the identification of high-risk and high-prevalence populations; improved technology; and lower costs for diagnosis, surgery, and angiography. Although the cost-effectiveness ratio was sensitive to these factors in the range of values considered, implausible conditions were needed for the marginal cost-effectiveness of screening to approach what is commonly considered to be acceptable. We found that if the costs of carotid endarterectomy (including hospitalization) were reduced by half, the marginal cost-effectiveness of screening decreased to $91 000 per quality-adjusted life-year. We also found that screening became cost-effective if it was applied to groups with a 40% prevalence of carotid atherosclerosis. However, it is unlikely that an asymptomatic group with such a prevalence could be found. The prevalence of carotid stenosis greater than 50% was 12.9% in a Veterans Administration hospital population with peripheral vascular disease and 15.2% in persons with cardiovascular disease in the same population [30]. In addition, we would expect that patients in the latter two groups would have higher rates of procedural complications, thus increasing cost per quality-adjusted life-year.

Screening had to be free of both complications and costs to reduce the marginal cost-effectiveness of screening to $71 000 per quality-adjusted life-year. Complication-free endarterectomy with current costs produced a marginal cost-effectiveness of $91 000. In our baseline case, we assumed that endarterectomy produced a complication rate of 1.33%, as found in ACAS (we excluded deaths occurring in the perioperative interval in surgical patients who did not actually have endarterectomy). It is highly improbable that all hospitals can reproduce such a low complication rate. Sensitivity analysis showed that complication rates in the more achievable range of 3% to 4% give a marginal cost-effectiveness in the range of $197 000 to $318 000 per quality-adjusted life-year.

The model was sensitive to the duration of reduction in stroke incidence resulting from surgery. In the baseline case, in which we assumed that the absolute 5-year difference in stroke rates would disappear after 10 years, the marginal cost-effectiveness would be $120 000 per quality-adjusted life-years. The lowest marginal cost-effectiveness of $74 000 per quality-adjusted life-year assumes that the reduction in stroke rate lasts for 30 years. In all of our modeling, the survival benefit from surgery is maintained for 30 years. Unfortunately, the literature contains little information about the long-term benefits of prophylactic carotid endarterectomy compared with no treatment. An observational study [56] of the benefits of a series of prophylactic carotid endarterectomy procedures over a 10-year period showed lower survival rates among the surgically treated patients at year 5 than were seen in the nonsurgically treated arm of ACAS and included a comparison group for only 48 months. Other studies with information on 10-year survival did not include comparison groups [57, 58]. Similarly, the data in the literature on the rate of restenosis after carotid endarterectomy is limited to 30 to 44 months of follow-up data [59, 60]. Because data on the durability of prophylactic carotid endarterectomy are lacking, it may be helpful to examine data on the benefits of another procedure. Sixteen years of observational data comparing the survival of patients having coronary artery bypass graft surgery with that of patients receiving medical treatment for left main coronary artery disease are available [61]. The survival curves for these two treatments begin to approach each other at year 7, converge faster thereafter, and appear to meet before 20 years. In some subgroups, the survival curves converge at 15 years or less. The survival benefit conferred by coronary artery bypass graft surgery seems to extend to less than 20 years. In all of our lifetime modeling, the stroke survival curves meet at 30 years. Given that coronary artery bypass graft surgery and carotid endarterectomy are different procedures, the benefits of carotid endarterectomy would still need to last several decades before the marginal cost-effectiveness of screening approaches acceptability.

Our study has several potential limitations. First, 5 years of data were available from ACAS. Evaluating the benefits of screening over this period and assuming that no further benefits would occur after 5 years may penalize screening. To overcome this potential bias, we assessed the effects of alternative assumptions about long-term reduction in the rate of stroke. Because the results were sensitive to these assumptions, data on long-term reduction of rate of stroke may influence the cost-effectiveness of screening. Second, our analysis focused on the screening of a healthy cohort, whereas patients in ACAS were recruited from vascular laboratories, ultrasonography suites, and similar places. The fact that patients in ACAS were identified through case finding rather than through the screening of a truly asymptomatic group may have caused overestimation of the benefits of screening. In addition, our model favored screening whenever possible; thus, our results may underestimate the actual magnitude of marginal cost-effectiveness ratios. Finally, our cost perspective was restricted to direct medical costs.

Kent and coworkers [62], using a decision analysis model based on the experience of the North American Symptomatic Carotid Endarterectomy Trial (NASCET), found that the combination of Doppler ultrasonography and magnetic resonance angiography used for the preoperative detection of carotid stenosis of 70% to 99% can have an acceptable marginal cost-effectiveness ($22 400 per year of life saved, in 1993 U.S. dollars). Their conclusion is not incompatible with our findings. Kent and coworkers compared the cost-effectiveness of different diagnostic strategies for the preoperative evaluation of symptomatic patients who are potential candidates for carotid endarterectomy, not a screening strategy for asymptomatic persons as we did.

Physicians who would screen in order to recommend endarterectomy to asymptomatic patients should note the small gain in quality-adjusted life-years (4.75 days over 30 years) achieved by screening. The small difference in quality-adjusted life-years between the two strategies may seem counterintuitive. If carotid endarterectomy prevents future strokes, why doesn't screening significantly improve overall outcome? The primary explanation is that although ACAS showed a statistically significant reduction in the incidence of ipsilateral strokes over the 5-year study period in surgically treated patients compared with medically treated patients (P = 0.004), we were concerned with all strokes, regardless of side or type. To assess whether a screening program to identify persons with asymptomatic carotid stenosis would be a cost-effective stroke prevention strategy, the appropriate risk reduction to compare is that of the "any stroke or death" category in the ACAS report, not just ipsilateral stroke. In the case of "any stroke or death," the 5-year relative risk reduction from surgery in ACAS was 20% and was not statistically different for the two treatment groups (P = 0.08). This relative risk reduction of 20% (the difference between the 31.9% event risk in the medically treated group and the 25.6% event risk in the surgically treated group divided by the medical event risk) translates to only a 5% reduction in absolute risk (the difference between the event risks in the medically treated group and the surgically treated group). In addition, ACAS did not find a statistically significant difference in event risk between the treatment groups with respect to the "any major stroke or death" category (P = 0.16). This means that endarterectomy did not afford protection against more disabling strokes. Furthermore, strokes caused by endarterectomy occur early in the observation period, whereas those caused by carotid disease occur later. The model incorporates the added time that patients with strokes from carotid disease spend in good health. Finally, because stroke occurs infrequently in asymptomatic populations, the number of strokes that can be prevented will not be large.

Despite modeling the almost ideal conditions found in ACAS and assuming that the survival benefit produced by endarterectomy lasts 30 years, our analysis indicates that screening is not as cost-effective as most accepted interventions (>$100 000 per quality-adjusted life-year). Screening approaches acceptability ($50 000) only under implausible conditions-for example, if a free screening instrument with perfect test characteristics is used or an asymptomatic population with a 40% prevalence of carotid stenosis is found. Screening is not cost-effective because the yield is low, and when carotid stenosis is found in an asymptomatic population, surgery offers a real but modest absolute reduction in the stroke rate at a substantial cost. Our evaluation indicates that a program of screening asymptomatic populations for carotid stenosis to find candidates for endarterectomy costs more per quality-adjusted life-year than is usually considered acceptable. As a strategy for the prevention of stroke, it would therefore be an unwise use of public resources. However, because our results were sensitive to assumptions about the duration of reduction in the risk for stroke from endarterectomy, studies examining this issue would be helpful.

Presented in part at the Society of General Internal Medicine, 10 March 1995, San Diego, California.

Dr. Solomon: Kaiser Permanente, 1 Kaiser Plaza, 23rd Floor, Oakland, CA 94612.

Mr. Oehlert: Division of Immunology and Rheumatology, Stanford University School of Medicine, 701 Welch Road #3301, Stanford, CA 94305-0107.

Dr. Garber: Stanford School of Medicine, 204 Junipero Serra Boulevard, Stanford, CA 94305.