Targeting Patients Undergoing Angioplasty for Thrombus Inhibition
A Cost-Effectiveness and Decision Support Model
Background—In recent clinical trials, glycoprotein IIb/IIIa blockers have demonstrated effectiveness in preventing adverse events after angioplasty in high-risk patients. However, uncertainty exists regarding the cost-effective selection of patients to receive antiplatelet therapy.
Methods and Results—All 4962 patients at Emory University Hospitals who underwent coronary intervention procedures (n=6062) from 1993 to 1995 were studied. Multivariate models to predict death and the composite of death, Q-wave and non-Q-wave myocardial infarction, and emergency additional revascularization were developed. Hospital costs and professional costs were determined. A cost-effectiveness analysis with therapy targeted to high-risk patients was performed. If patients with a >5% probability of events received antiplatelet therapy that reduced events by 24% and cost $1000, 40.1% of patients would receive therapy; complications would be reduced from 6.39% to 5.37%, and cost would increase $261 from $10 343 to $10 604, or $25 504 per event prevented. The marginal cost per event prevented by moving from a 7% to a 5% probability of an event cutoff would be $57 799.
Conclusions—For high-risk patients, there may be cost savings; for low-risk patients, therapy may not be cost effective; and for patients in the midrange (between 5% and 7% probability of an adverse event), events may be prevented at an acceptable level of cost.
Although platelet glycoprotein (GP) IIb/IIIa receptor blockers decrease complications after coronary intervention,1 2 3 4 5 6 the appropriate selection of patients to receive this form of therapy remains unclear, in part because of concerns about cost. The extreme options are to administer this therapy to all patients or to none. The alternative option is to try to target therapy to patients on the basis of their level of risk of developing a complication.
The goal of this study was to assess the cost-effectiveness of a therapeutic strategy involving the administration of antiplatelet therapy to a segment of the population who are targeted according to their level of risk.
Data from 4962 patients undergoing 6062 coronary angioplasty procedures to 7604 coronary sites at the Emory University hospitals between January 1, 1993, and December 31, 1995, were used to develop a model to predict in-hospital events. Patients in this study included those having procedures (1) for stable or unstable angina pectoris, (2) after several days of stabilization after an acute myocardial infarction (MI), or (3) in the setting of acute ischemia. The model was validated in 3663 patients undergoing 4232 angioplasty procedures in 1996 and 1997. In addition, data from 21 535 patients undergoing angioplasty between 1980 and 1996 were used to develop a model to predict long-term survival.
An emergent procedure was defined as one performed in the setting of acute ischemia or infarction. MI was defined as new Q waves or an elevation of creatine phosphokinase levels to 3 times the upper limit of normal in patients with baseline levels in the normal range. Defined by patient history indicates hypertension, diabetes, severity of angina, or prior MI. Angina was defined by the Canadian Cardiovascular Society Classification.7 Congestive failure was defined by the New York Heart Association criteria.8
Data Collection and Statistical Analysis
Data were recorded prospectively on standardized forms and entered into a computerized database. All fields were defined in a data dictionary. Angiographic success was assessed by quantitative coronary arteriography.
Discrete variables are expressed as proportions and continuous variables as mean±SD. Statistical models were developed using multivariate logistic regression to examine the ability of demographic, clinical, angiographic, and procedural characteristics to predict in-hospital mortality and the composite outcome of death, Q-wave and non–Q-wave MI, or emergent revascularization. Candidate predictors were age, sex, hypertension, diabetes, prior coronary artery bypass grafting, prior PTCA, ejection fraction, angina class, number of diseased vessels, lesion length, percent diameter stenosis, diffuseness, ulceration, calcification, branch, thrombus, and procedural priority (emergent versus elective). An S-plus procedure, making use of relationships among all variables, was used to impute missing covariate data. Potential nonlinear effects of each of the continuous predictor variables were checked using restricted cubic splines. The ability of the models to discriminate patients with respect to outcome was measured using the c-index. The model calibration was checked by dividing the population into deciles of expected risk and comparing expected to observed risk by linear regression.9 Long-term survival was determined by the Kaplan-Meier method.10 Quality-adjusted life-years were calculated as the sum of utility times discounted survival estimates. Utility was not measured; rather, it was estimated as a form of sensitivity analysis.
Cost and Cost-Effectiveness Analysis
The hospital charges were assessed from the UB-92 formulation of the hospital bill. Charges were reduced to costs using departmental cost-to-charge ratios. For the 3842 patients in 1994 and 1995, all professional charges and Current Procedural Terminology codes for the episode of care were gathered. These codes were converted to relative value units using the Resource-Based Relative Value Scale.11 The relative value units were summed and converted to dollars using the Medicare conversion factor.11A
The model presented in Figure 1⇓ was used to assess the cost-effectiveness of antiplatelet therapy on the basis of observed in-hospital costs and event rates and the published efficacy of antiplatelet therapy. The model was specified by a set of definitions and equations that are available on request. The model does not include the possibility of bleeding caused by the therapy.
The assumptions underlying the cost-effectiveness model are based on an analysis of data from Emory, except for the cost of therapy, which was assumed to be $750 or $1000, and the effectiveness of therapy. A meta-analysis (similar to that of Kong et al12 ) of 6 published trials1 2 3 4 5 6 revealed a 24% relative event reduction from 11.79% to 8.95% at 30 days, with events defined as death, MI, or urgent revascularization (or revascularization if urgency was uncertain). However, the generalization of the event rate reduction to 24% at 30 days, which is available for all published series, may not be entirely appropriate to the in-hospital event data. In the Randomized Efficacy Study of Tirofiban for Outcome and Restenosis (RESTORE),6 the absolute difference in event rates changed little from 2 to 30 days, whereas the relative event rate reduction fell because of continuing events in both arms. If there was an event incidence of 5% in both arms from discharge to 30 days in the datasets used for the meta-analysis, then the in-hospital event reduction would be from 6.79% to 3.95%, a 42% relative reduction. The in-hospital event rate reduction may reasonably be estimated between the extremes of 24% and 42%.
Cost-effectiveness was then assessed using the methods of Weinstein and Fineberg.13 Sensitivity analyses were performed to assess the influence of varying the cost of therapy, the efficacy of therapy, and the cut-off probability of complications for initiating therapy on the cost-effectiveness ratio. The effect of therapy on long-term outcome estimated in cost per quality-adjusted life-year gained was also estimated. Assumptions about the effect of events on future costs, appropriate discount rates, utility, and effect of events on utility, as well as the impact of events prevented on future survival were made.
In Table 1⇓, the characteristics of the patient population and sites are presented, including clinical, angiographic, procedural, and outcome data. Stent usage was still low in 1994 and 1995, but in 1997, it was much higher (30%). This population is similar to those in RESTORE4 and the Evaluation of 7E3 for the Prevention of Ischemic Complications (EPIC) study1 ; it more closely resembles the Evaluation in PTCA to Improve Long-term Outcome with Abciximab GP IIb/IIIa blockade (EPILOG)5 study in clinical and outcome descriptions, including death and Q-wave MI.
Multivariate correlates of the composite end point are shown in Table 2⇓. The ability to predict death was greater than the ability to predict the composite end point (c-index, 0.85 versus 0.71). However, the incidence of death was only 0.7% versus 6.4% for the composite end point, and the importance of preventing these other complications is easily recognized.
The stability of the model using data from 1993 to 1995 was explored by using it to predict events in patients undergoing angioplasty from 1996 to 1997, which yielded a c-index of 0.73. The strongest and most stable predictors of events were emergent or urgent angioplasty (odds ratio, 4.69), higher percent diameter stenosis (odds ratio, 1.35 for a 10% increase), and thrombus (odds ratio, 2.01).
Average costs at Emory University hospitals were used to develop the cost-effectiveness analysis. The mean hospital plus professional cost was $10 383±5223 for interventional procedures at Emory University hospitals. For patients without the composite end point, the cost was $9467±3795; for those with the composite end point, the cost was $23 168±11 648.
Two additional assumptions underlying this cost-effectiveness model are that (1) the cost of events in treated patients is equal to the cost of events in untreated patients and (2) the cost of complications in high-risk patients is equal to that in low-risk patients. The first assumption seems reasonable, but it cannot be tested with the available data. To examine the second assumption, a test for a correlation between the predicted composite end point and total hospital cost was performed. The Spearman rank correlation coefficient was −0.11 (P=0.098); this lack of dependence of cost on probability of an event supports the validity of the second assumption.
The probability of an event, based on the logistic model for the composite end point, was used as the criteria for determining whether a patient should receive therapy. The cost-effectiveness of the strategy of targeting therapy to patients at varying levels of underlying risk was evaluated. Table 3⇓ presents the cost-effectiveness analyses assuming a cost of $750 or $1000 and an effectiveness of 24% in preventing events with GP IIb/IIIa blocker therapy. Assuming a 0% cut point, 100% of the patients would be treated, and the event rate would drop from 6.39% to 4.86%; it would fall to 3.84% at 40% effectiveness. The mean in-hospital cost per patient, when assuming 24% effectiveness and $750 of cost per patient for the therapy, would be $10 883, as shown in Table 3⇓. Alternatively, the mean cost per patient at a 100% cut point, in which case no patients would receive therapy, would be $10 343. The average cost per event averted for the strategy of treating all patients versus no patients is $35 173.
There is a great degree of variability in cost-effectiveness because the probability of an event in the targeted population varies. As the cutoff probability of a composite event for determining who will receive therapy is increased, the incidence of events rises and the cost-effectiveness ratio falls. A dominant strategy is one in which administering treatment at any level of risk results in events being prevented at a cost savings. Table 3⇑ shows the average cost-effectiveness ratio for treating all patients above specific cutoff points at $750 and $1000 for cost of therapy. Table 3⇑ also shows the incremental (marginal) cost per event averted; it reveals the additional costs associated with preventing an event at the next lowest cutoff point. The incremental cost per event averted moving from a 7% to a 5% probability of an event cutoff was $39 923 at $750 and $57 799 at $1000 for cost of therapy. The marginal cost-effectiveness decreased with increasing effectiveness of therapy. At 40% effectiveness, the marginal cost per event averted when comparing a 7% cutoff to a 5% cutoff was $18 474 at a cost of $750 for therapy and $29 199 at a cost of $1000 for therapy.
A 2-way sensitivity analysis on cost of therapy and effectiveness of therapy is presented in Figure 2⇓. The marginal cost-effectiveness for treating patients with a 3% versus 5% probability of an event is on the left, that for a 5% versus 7% probability is on the right. Isobars of cost from $500 to $1500 in $250 steps are shown. The x axis reflects the effectiveness of therapy from 0.20 to 0.80, and the y axis reflects the cost-effectiveness ratio in dollars per event prevented. The cost-effectiveness ratio rises as the cost of therapy rises. The ratio falls as therapy becomes more effective. The marginal cost-effectiveness ratio is larger when moving from the 5% to the 3% cutoff than when moving from the 7% to the 5% cutoff. For reasonably inexpensive, even reasonably effective therapy, the therapeutic strategy dominates. However, at the other extreme, for therapy that is more expensive and less efficacious, the cost per event prevented ultimately becomes prohibitive.
A 2-way sensitivity analysis on cost of therapy and the probability cutoff for treatment is presented in Figure 3⇓. Therapy is assumed to be 24% effective in preventing events on the left and 40% effective on the right. On the x axis, the probability cutoff varies from 0 to 0.15. Points are computed as the marginal cost-effectiveness at each probability of an event compared with the cutpoint 1% higher. Again, as the cost of therapy rises, the cost-effectiveness ratio also rises, and as the probability cutoff for treatment rises, the cost-effectiveness ratio falls. Thus, a favorable ratio can be achieved by using a high cutoff probability. The problem is that as the probability cutoff is raised, the proportion of patients treated will fall (Table 3⇑). That only a small proportion of patients will have even a 10% probability of a complication limits the ability to define a high-risk group.
The incremental cost-effectiveness of targeted therapy measured in cost per quality-adjusted life-year gained and assuming 24% effectiveness is presented in Table 4⇓. The cases for patients with probability of event cutoffs from 0% to 3% and 10% to 15% were considered assuming a cost of therapy of either $750 or $1000. From the 21 535 patients who underwent angioplasty between 1980 and 1996, the mean survival was 12.8 years (95% confidence interval, 12.6 to 13.0 years) without an event and 11.5 years (95% confidence interval, 11.0 to 12.0 years) with the composite end point (P<0.0001). Mean discounted life-year over 15 years of follow-up as determined from the Kaplan-Meier estimates using an annual 3% discount rate was 9.44 years for patients without an event and 8.63 years for patients with an event. No additional survival benefit of an event-free initial hospitalization was assumed beyond 15 years. It was assumed that costs after discharge would be similar in the patients with and without events, such that any difference in cost would be reflected in the hospital cost.
Five models were developed. Utility was assumed to decrease for some of the models by 20% for the first year if there was an event, which would reduce quality-adjusted survival to 8.45 years; otherwise, the utility was assumed to remain at 1.0. Therapy that decreases in-hospital events may or may not decrease future mortality. In the first 2 models, an event averted was assumed to increase survival to that of a patient without an event. In the last 3 models, the increase in survival attributed to an event averted was first assumed to be 50%, then 25%, and finally 0% of that in models 1 and 2. In the final model, the effect of events averted is only to improve quality of life in the first year. The cost-effectiveness ratio varied widely depending on the assumptions made. For each model, the cost-effectiveness ratio fell as the probability cutoff was raised. At a cost of therapy of $750, the marginal cost-effectiveness comparing a 5% cutoff to a 7% cutoff probability of an event for the best case (model 2) was $40 205 at 24% effectiveness and $18 604 at 40% effectiveness. At a $1000 cost of therapy, these ratios rose to $58 206 at 24% and $29 405 at 40% effectiveness.
We developed a model to predict complications after PTCA, which we used to evaluate the cost-effectiveness of GP IIb/IIIa blockade therapy targeted to patients according to their level of risk. That the prediction models offered a limited ability to predict outcome was not entirely surprising, because it is likely that clinicians would avoid PTCA when possible in patients for whom there is a high probability of failure. These relatively imprecise models also influence the cost-effectiveness analyses; a better ability to predict outcome would permit a better ability to target therapy. The information presented in Tables 3⇑ and 4⇑ and in Figures 2⇑ and 3⇑ can be used to guide the decision-making process regarding which patients should receive therapy. If an acceptable upper limit of cost per event averted or quality-adjusted life-year gained is known, then all patients for whom the cost of preventing an event is less than this upper limit should receive therapy, given the limitations and assumptions of the models. It is appropriate to use marginal rather than average cost-effectiveness ratios for decision-making about the appropriate cut point, because it is the additional cost associated with treating the next patient that should drive the therapeutic decision. However, the mean ratio provides the cost-effectiveness for the overall population to be treated once the cut point is chosen.
According to our cost-utility analysis (Table 4⇑), if society was willing to pay up to $50 000 per quality-adjusted life-year gained and the cost of therapy is $750, then 5% would be a reasonable cut point for initiating therapy according to the best case models (models 1 and 2). If society was willing to pay up to $100 000 per quality-adjusted life-year gained, then 5% would be a reasonable cut point for initiating therapy, according to all but the worst-case models. This would result in treating ≈40% of patients at the highest level of risk. If no long-term benefit exists in terms of improved survival, quality of life, or cost savings after hospitalization, the decision regarding therapy should be made on the in-hospital cost alone. A nomogram (Table 5⇓⇓⇓⇓) has been developed that may be used with some caution to evaluate the probability of an event. Given the consistent predictive effects of urgent or emergent procedures and the presence of thrombus on the risk of events from 1993 to 1997, the presence of either may be reasonably used to help select patients for therapy.
The extension of the in-hospital cost-effectiveness analysis to a long-term cost-utility analysis involved multiple additional assumptions, some of which are not verifiable.14 The cost-effectiveness ratio may be quite sensitive to these assumptions, as Table 4⇑ suggests. However, with cost-effectiveness expressed in cost per quality-adjusted life-year gained, it is possible to benchmark therapy against other forms of therapy; this is not possible for cost per event prevented.
However, the cost-effectiveness ratios expressed in cost per event averted were highly data-driven, and event rates and costs (exclusive of drug costs) were based on actual data rather than assumptions. Drug efficacy was based on the literature.1 2 3 4 5 6 Although the cost of therapy is uncertain, $750 to $1000 is approximately what current GP IIb/IIIa agents cost, and the sensitivity analysis from $500 to $1500 covers the expected range well. Drug efficacy and costs were the 2 primary forms of sensitivity analysis.
Another potential limitation is the assumption that the efficacy of the drug is independent of the probability of an event. Perhaps in low-risk patients, the therapy prevents a lower or a higher proportion of events.
It is also possible that the trend toward the greater use of intracoronary stents may alter costs and/or event rates. However, the recently completed Evaluation of IIb/IIIa Platelet Inhibitor for Stenting (EPISTENT) study,15 in which the benefits of GP IIb/IIIa blockade were shown to extend to patients undergoing coronary stenting, suggests that the results of the present study will remain applicable to a large extent. In principle, the present study could be repeated as more stent data become available, although future studies will be limited by frequent antiplatelet therapy in the patient population.
Although this study takes a societal perspective overall, hospital costs had a hospital perspective and professional costs a payer perspective, which were used as proxies for societal costs. The cost-effectiveness ratio should be from a societal perspective, because it is the responsibility of society as a whole to prevent cardiovascular events, because society would have to pay the cost of these events. However, the hospital would have to pay the cost of the drug and, depending on reimbursement scheme, may derive little or no benefit from the events prevented.16
By examining patient level data and creating a risk profile, the ability to examine cost-effectiveness according to predicted level of risk has been demonstrated. Although a similar approach was used in a study of lipid lowering,17 we are unaware of any studies that examine the cost-effectiveness of targeting high-risk segments of the population for treatment using inputs derived from an empirical dataset. Although the study was based on a single-center experience, the population resembled that seen in several multicenter trials.1 2 3 4 5 6 This kind of modeling is limited in a number of ways; however, this study has shown quantitatively and in a real patient population what is often stated: as the level of risk rises, therapy to decrease events can become more cost effective. In terms of decision-making for angioplasty patients, the results must be interpreted with some caution. Clearly, highly effective, inexpensive therapy could be used in most patients. For more expensive therapy with modest effects, patients at a higher risk, including patients undergoing coronary intervention in the setting of uncontrolled unstable angina, acute MI, or intracoronary thrombus, should be selected. From a societal or policy standpoint, studies such as this may help in the development of informed guidelines for the use of expensive therapies.
The authors thank Lesley Wood for her careful editorial review. Funded by a grant from Merck US Human Health, West Point, Pa.
Presented in part at the 46th Annual Scientific Sessions of the American College of Cardiology, Anaheim, Calif, March 16 through 19, 1997.
- Received September 29, 1999.
- Revision received February 16, 2000.
- Accepted February 21, 2000.
- Copyright © 2000 by American Heart Association
Tcheng JE, Harrington RA, Kottke-Marchant K, et al, for the IMPACT investigators. Multicenter, randomized, double-blind, placebo-controlled trial of the platelet integrin glycoprotein IIb/IIIa blocker Integrelin in elective coronary intervention: IMPACT Investigators. Circulation. 1995;91:2151–2157.
The RESTORE Investigators. Effects of platelet glycoprotein IIb/IIIa blockade with tirofiban on adverse cardiac events in patients with unstable angina or acute myocardial infarction undergoing coronary angioplasty. Circulation. 1997;96:1445–1453.
Campeau L. Grading of angina pectoris. Circulation. 1975;54:522–523. Letter.
The Criteria Committee of the New York Heart Association. Nomenclature and Criteria for Diagnosis of Diseases of the Heart and Great Vessels. 8th ed. Boston: Little, Brown and Co; 1979.
Becker ER, Mauldin PD, Bernadino ME. Using physician work RVUs to profile surgical packages: methods and results for kidney transplant surgery. Best Pract Benchmarking Healthc. 1996;1:1–7.
Commerce Clearing House. Physician’s Medicare Fee Schedule, January 1, 1995. Chicago, Ill: CCH Incorporated; 1994:5.
Kong DF, Califf RM, Miller DP, et al. Clinical outcomes of therapeutic agents that block the platelet glycoprotein IIb/IIIa integrin in ischemic heart disease. Circulation. 1998;98:2829–2835.
Weinstein WC, Fineberg. Clinical Decision Analysis. Philadelphia, Pa: Saunders; 1980:256–261.
Gold MR, Siegel JE, Russell LB, et al. Cost-Effectiveness in Health and Medicine. New York: Oxford University Press; 1996.