Figure 1. Physician level variation in the use of drug-eluting stents (DES). The physicians include the individual physicians (n = 2715) in the National Cardiovascular Data Registry (NCDR) CathPCI Registry, version 4 data, who performed more than 75 percutaneous coronary interventions (PCIs) annually. These physicians are ranked in ascending order of rate of DES use, such that those using the least DES are to the left and those using the highest DES are to the right.
Figure 2. Incremental cost-effectiveness (A) and cost-effectiveness acceptability curve (B) for drug-eluting stent (DES) use. Illustrated is the comparison between DES lower use strategy (base case) in patients at low risk of target-vessel revascularization (TVR) vs DES existing use strategy in the 1000 NCDR samples generated by sampling-based probabilistic sensitivity analysis. A, For incremental effectiveness (TVR events avoided), points to the left of 0 indicate decreased efficacy (more TVR events with lower DES use strategy vs existing DES use strategy). For incremental costs (measured in thousands of dollars) points below 0 indicate costs saved by a lower DES use strategy. The ellipse represents the 95% confidence ellipse and contains 95% of the simulations. B, The solid vertical line represents the $10 000/TVR avoided threshold. Cost-effective probability indicates the proportion of simulations that are cost-effective. The curve is the cumulative probability density function of the 1000 trial replicates; 98.3% of the 1000 trial replicates showed an incremental cost-effectiveness ratio for DES existing use vs lower use strategy that exceeded $10 000/TVR avoided.
Figure 3. Sensitivity of projected cost savings to assumed reductions in drug-eluting stent (DES) use in patients at low risk for target-vessel revascularization (TVR). The x-axis shows modeled rates of DES use in the low-TVR-risk group. The first bar (far left) represents 100% bare-metal stent (BMS) use in this predicted low-TVR-risk group, while the last bar (far right) represents existing use rate or 74% DES use in the low-TVR-risk group. The y-axis on the left shows the US annual costs saved per year in patients undergoing percutaneous coronary intervention (bars). The y-axis on the right shows the absolute increase in TVR event rate (black line).
Amin AP, Spertus JA, Cohen DJ, Chhatriwalla A, Kennedy KF, Vilain K, Salisbury AC, Venkitachalam L, Lai SM, Mauri L, Normand S-LT, Rumsfeld JS, Messenger JC, Yeh RW. Use of drug-eluting stents as a function of predicted benefit: clinical and economic implications of current practice [published online July 9, 2012]. Arch Intern Med. doi:10.1001/archinternmed.2012.3093.
eAppendix. Online Supplementary Materials Appendix
eTable. MassDAC TVR risk prediction logistic model
eFigure 1. Inclusion and exclusion criteria
eFigure 2. Distribution of the predicted TVR risk
eFigure 3. Rates of DES use by predicted TVR risk categories across years
This supplementary material has been provided by the authors to give readers additional information about their work.
Amin AP, Spertus JA, Cohen DJ, Chhatriwalla A, Kennedy KF, Vilain K, Salisbury AC, Venkitachalam L, Lai SM, Mauri L, Normand ST, Rumsfeld JS, Messenger JC, Yeh RW. Use of Drug-Eluting Stents as a Function of Predicted BenefitClinical and Economic Implications of Current Practice. Arch Intern Med. 2012;172(15):1145-1152. doi:10.1001/archinternmed.2012.3093
Author Affiliations: Cardiovascular Division, Department of Medicine, Barnes Jewish Hospital, Washington University School of Medicine, St Louis, Missouri (Dr Amin); Saint Luke's Mid America Heart and Vascular Institute, Kansas City, Missouri (Drs Spertus, Cohen, Chhatriwalla, and Salisbury, Mr Kennedy, and Ms Vilain); The University of Missouri, Kansas City (Drs Spertus, Cohen, Salisbury, and Venkitachalam); Department of Preventive Medicine, The University of Kansas Medical Center, Kansas City (Dr Lai); Cardiology Divisions, Departments of Medicine, Brigham and Women's Hospital (Dr Mauri) and Massachusetts General Hospital (Dr Yeh), Harvard Medical School, Boston; Department of Biostatistics, Harvard School of Public Health, Boston (Dr Normand); Department of Health Care Policy, Harvard Medical School, Boston (Dr Normand); Division of Cardiology, University of Colorado School of Medicine, Aurora (Drs Rumsfeld and Messenger); and The Denver Veterans Affairs Medical Center, Denver, Colorado (Dr Rumsfeld).
Background Benefits of drug-eluting stents (DES) in percutaneous coronary intervention (PCI) are greatest in those at the highest risk of target-vessel revascularization (TVR). Drug-eluting stents cost more than bare-metal stents (BMS) and necessitate prolonged dual antiplatelet therapy (DAPT), which increases costs, bleeding risk, and risk of complications if DAPT is prematurely discontinued. Our objective was to assess whether DES are preferentially used in patients with higher predicted TVR risk and to estimate if lower use of DES in low-TVR-risk patients would be more cost-effective than the existing DES use pattern.
Methods We analyzed more than 1.5 million PCI procedures in the National Cardiovascular Data Registry (NCDR) CathPCI registry from 2004 through 2010 and estimated 1-year TVR risk with BMS using a validated model. We examined the association between TVR risk and DES use and the cost-effectiveness of lower DES use in low-TVR-risk patients (50% less DES use among patients with <10% TVR risk) compared with existing DES use.
Results There was marked variation in physicians' use of DES (range 2%-100%). Use of DES was high across all predicted TVR risk categories (73.9% in TVR risk <10%; 78.0% in TVR risk 10%-20%; and 83.2% in TVR risk >20%), with a modest relationship between TVR risk and DES use (relative risk, 1.005 per 1% increase in TVR risk [95% CI, 1.005-1.006]). Reducing DES use by 50% in low-TVR-risk patients was projected to lower US health care costs by $205 million per year while increasing the overall TVR event rate by 0.5% (95% CI, 0.49%-0.51%) in absolute terms.
Conclusions Use of DES in the United States varies widely among physicians, with only a modest correlation to patients' risk of restenosis. Less DES use among patients with low risk of restenosis has the potential for significant cost savings for the US health care system while minimally increasing restenosis events.
Drug-eluting stents (DES) are effective in reducing restenosis, with an estimated 50% to 70% relative risk (RR) reduction in target vessel revascularization (TVR) rates.1,2 These benefits have led to the rapid adoption of DES after 2003, such that by 2005, their use in the United States was nearly 90%3- 6 and has remained higher than 75%.3- 5 Several studies have demonstrated that the benefits of DES in reducing the need for TVR are largely confined to subsets of patients at high risk of restenosis with bare-metal stents (BMS).7,8 Accordingly, some have suggested that DES should be targeted selectively to the higher-TVR-risk lesions.9- 11 Whether DES are preferentially used among patients at higher risk of restenosis in current clinical practice is unknown.
Examining the use of DES as a function of patients' TVR risk can have important implications for the costs of percutaneous coronary intervention (PCI), which remain a concern.12- 16 While trial-based economic analyses have shown DES to be generally cost-effective for patients at moderate to high risk of restenosis, a recent population-based analysis has found that the annual costs associated with DES use were $1.57 billion between 2002 and 2006.17 Moreover, DES use currently requires prolonged dual-antiplatelet therapy (DAPT),18- 20 which not only increases long-term medication costs but also increases patient risk of bleeding events and potentially subjects them to serious complications if DAPT is prematurely discontinued.18- 21
To determine current patterns of DES use as a function of TVR risk and the potential clinical and economic implications of more tailored DES use, we analyzed data from the National Cardiovascular Data Registry (NCDR) CathPCI Registry.22 Specifically, we assessed (1) variation in DES use among US physicians participating in the NCDR; (2) whether predicted TVR risk with BMS is associated with DES use; and (3) the estimated clinical and economic consequences of lower DES use among patients with low TVR risk.
The NCDR CathPCI Registry, cosponsored by the American College of Cardiology (ACC) and the Society for Cardiovascular Angiography and Interventions (SCAI), is the largest US clinical registry of patients undergoing PCI. Details of the CathPCI Registry have been previously described.22 In brief, participating hospitals collect detailed baseline clinical characteristics, in-hospital care processes, and outcomes retrospectively via chart review using a standardized set of data elements and definitions, which are available at http://www.ncdr.com/WebNCDR/elements.aspx.
Data from 2 120 659 PCI admissions from 1119 hospitals participating in the registry from January 2004 through September 2010 were initially included. To ensure a sample of patients who were “eligible” for both stent types, we then excluded patients receiving stents smaller than 2.25 mm and larger than 4.00 mm in diameter for which DES were not available throughout the period of observation. We next developed a propensity-score model to predict DES (vs BMS) use via logistic regression conditioned on 46 demographic and clinical variables. After plotting the distribution of propensity scores by stent type, we excluded patients falling into regions of nonoverlapping propensity scores. These were patients in whom either DES or BMS were used almost exclusively, and the choice of using an alternative stent was not likely feasible. The remaining 1 506 758 PCI admissions were included. For admissions during which multiple PCIs were performed, we analyzed only the first PCI.
For each patient, we estimated the risk of TVR assuming treatment with BMS using a validated prediction model developed from the Massachusetts Data Analysis Center (MassDAC) database (eTable).23 This model incorporates sociodemographic, clinical, and angiographic variables to predict TVR and provides better discrimination than the 3 commonly used variables of diabetes, vessel diameter, and lesion length, which are all components of the MassDAC model.23
Estimates of 1-year TVR risk were categorized into the 3 clinically relevant groups of low (<10%), moderate (10% to <20%), and high (≥20%). Baseline clinical and demographic patient characteristics by TVR risk group were compared using the χ2 test for categorical variables and analysis of variance for continuous variables. We then compared the rates of DES use in low-, medium-, and high-TVR-risk groups and estimated the unadjusted association of TVR risk with DES use by means of modified Poisson regression.24- 26 Because this association might have changed after concerns regarding stent thrombosis led to declines in DES use after 2006,27,28 we included an interaction term between time (before and after October 2006) and TVR risk on the outcome of DES use (eFigure 3).
Finally, we estimated the economic and clinical impact of a hypothetical reduction in the rate of DES use among low-TVR-risk patients within the US PCI population (approximately 600 000 PCIs per year).29 For this analysis, we assumed that the distribution of TVR risk as well as the use of DES among groups of TVR risk within the NCDR population were representative of that seen in the US PCI population. We used previously described assumptions12 to estimate clinical outcomes and costs from the perspective of the US health care system, as detailed in the eAppendix. The model considered the cost of stents, the cost of repeated TVR procedures for the treatment of restenosis (and their associated hospitalizations), and the cost of DAPT after either DES or BMS. For patients whose PCI was performed electively, we assumed the duration of DAPT would be 1 month after BMS and 12 months after DES.30,31 However, for PCI in the setting of an acute coronary syndrome, we assumed that DAPT would be used for 1 year regardless of stent type.30,31 We modeled the uncertainty observed in real-world clinical practice around these assumptions used in estimating costs and TVR events by performing sampling-based probabilistic sensitivity analyses in which we executed the cost-effectiveness model repeatedly (1000 samples) for combinations of values sampled randomly from the probability density functions of the input factors known to vary in real clinical practice.
We also performed deterministic sensitivity analyses assuming alternate proportions of DES use with a “lower use” strategy only among patients at low TVR risk (ie, from the 74% existing rate of DES use, as reported in the “Results” section, to 0% DES use in 1% increments). Finally, we assumed that clopidogrel was available in generic form at a cost of $1.00/d. All analyses were conducted in SAS software, version 9.2 (SAS Institute) and TreeAge Pro 2011 software (TreeAge Inc).
A total of 1 506 758 PCI admissions met the inclusion criteria for the analysis (eFigure 1). Of these, 648 292 patients were predicted to be in the low-TVR-risk group (43.0%), 659 838 in the moderate-TVR-risk group (43.8%), and 198 628 in the high-TVR-risk group (13.2%). As expected, patients with a high predicted TVR risk were more likely to be older and male and have diabetes, chronic kidney disease, and prior PCI (Table). They were also more likely to present with stable angina rather than an unstable coronary syndrome. Finally, they were more likely to have severe 3-vessel coronary artery disease, with smaller-diameter vessels and longer lesions. Drug-eluting stents were used in 76.9% of the PCIs included in the study sample. We found extensive variation in physician patterns of DES use (Figure 1). Among the 2715 physicians performing 415 115 PCI procedures (at least >75 procedures per year) between July 2009 through September 2010, DES use ranged from 2% to 100%.
The median predicted TVR risk was 11% (interquartile range [IQR], 8%-16%) (eFigure 2). Drug-eluting stent use was 73.9% among those at a low risk for TVR, 78.0% among those at moderate risk for TVR, and 83.2% among those at the highest TVR risk (Table). We found a 0.53% (95% CI, 0.50%-0.56%) relative increase in the rate of DES use for each 1% increase in the predicted risk of TVR with BMS. In addition, despite an overall decline in DES use over time (30% decline in DES use after October 2006), the relationship between TVR risk and DES use was modest in both time periods (RR, 1.0020 [95% CI, 1.0016-1.0025] before October 2006 vs RR, 1.0086 [95% CI, 1.0082-1.0089] after October 2006) (P <.01 for the interaction).
A 50% reduction in the use of DES only among those patients at low TVR risk was estimated to result in potential net savings of $204 654 000 per year in the United States (95% CI, $189 899 520-$227 258 760), or $34 109 per 100 PCIs performed compared with current practice. These estimated savings occurred even after accounting for a modest estimated increase in repeated procedures due to TVR (absolute increase in TVR rate, 0.50% [95% CI, 0.49%-0.51%]), which were estimated to cost $64 728 000. The incremental cost-effectiveness ratio (ICER) of the “existing” use strategy vs the lower DES use strategy was at $68 230 per TVR event avoided. In probabilistic sensitivity analyses, the ICER remained higher than $10 000 per TVR in 98.3% of simulations (Figure 2).
Deterministic sensitivity analysis showed the projected impact of alternative rates of DES use among patients at low risk of TVR with BMS (Figure 3). As the rate of DES use among low-TVR-risk patients decreases, the potential cost savings are projected to increase substantially with more modest increases in TVR events. For example, use of only BMS among all patients at low risk of TVR would be projected to reduce current health care expenditures by $409 317 379 per year with a 0.99% absolute increase in the risk of TVR at a population level. Finally, with the assumption that the cost for DAPT would decrease to $1.00/d (with the expected approval of generic clopidogrel in 2012), the estimated net cost savings with a 50% reduction in DES use in the low-TVR-risk group was projected at $127 950 000 per year.
The present study demonstrates that in current US practice, DES use is prevalent, even among patients at low risk of developing restenosis. There was also significant variation in the rate of DES use by individual physicians. A reduction in DES use among patients at low risk for restenosis was projected to be associated with substantial costs savings with only a small increase in TVR events.
The use of DES remains an important driver of increasing health care costs in the United States and worldwide. Groeneveld et al17 analyzed Medicare data on 2 million beneficiaries from 2002 to 2006 and found that the additional costs associated with DES use were $1.57 billion annually. Given the costs to patients and society of DES technology, and the potential risks of the requirement for long-term DAPT after DES (increased bleeding with DAPT, increased stent thrombosis with premature discontinuation), there appears to be an important opportunity to tailor DES use to those with the greatest potential to benefit and reduce its use in those with favorable outcomes after BMS alone. From an economic perspective, this study projected that adopting a strategy that reduced the current use of DES in those with the lowest predicted risk of TVR by 50% could be associated with cost savings of about $200 million every year in the United States alone, even after accounting for a small increase (<0.5%) in the need for subsequent PCI for restenosis—which is a relatively benign condition in most patients.
Several previous studies have compared the clinical benefits of DES vs BMS among patients across different levels of restenosis risk. Tu and colleagues8 found that in Ontario's Cardiac Care Network, the benefits of DES were substantially greater in those patients with diabetes, small target vessels [<3 mm in diameter], and long lesions [stent length ≥20 mm]. The number needed to treat (NNT) to prevent 1 TVR event with DES ranged from 10 to 27 in those individuals with 2 or more of these TVR risk factors, while in those with fewer risk factors, the effectiveness of DES did not differ significantly from that of BMS, and the NNT ranged from 53 to 167. More recently, a post hoc analysis from the HORIZONS-AMI trial7 of patients with acute myocardial infarction (MI) demonstrated that in patients at highest risk for restenosis, use of DES resulted in a marked reduction in target-lesion revascularization (TLR) at 12 months, but that no differences in TLR at 12 months were present between DES and BMS in patients at low risk for restenosis. The current study extends these findings by using national clinical practice data to examine how DES are being used in relation to patient risk for restenosis and to estimate how changes in practice could affect health care costs on a population level.
The findings of this study also extend recent insights from the EVENT investigators,32 who found that the approximate 25% reduction in DES use after the 2006 was accompanied by a small increase in clinical restenosis but no differences in the rates of death or MI. However, this decrease in DES use after 2006 led to substantial reductions in the cost of cardiovascular care of about $400 per PCI patient. In this study, while reduced DES use after 2006 was associated with risk factors for restenosis, these associations were modest, implying that reductions in DES were not necessarily in low-TVR-risk patients. Our findings now build on this concept and suggest that further reductions in DES use only among patients at low risk of TVR may translate into additional cost savings with an even smaller impact on overall clinical TVR outcomes than that observed in the EVENT study.
We found that the projected cost savings associated with lower use of DES were extremely sensitive to the magnitude of reduction in DES use among low-TVR-risk patients, while estimated increases in TVR events were largely insensitive to these reductions, implying that even a small reduction in DES use practice patterns may result in substantial cost savings. However, successfully implementing strategies that incorporate the predicted benefit of interventions into practice remains a sizable challenge. The MassDAC TVR risk prediction model could potentially offer an evidence-based solution to this problem. The model is freely available as an online tool (http://massdac.org/riskcalc/revasc) and could be used to prospectively inform clinicians and patients of patients' TVR risk prior to stent implantation.33 Use of the model could not only promote evidence-based care but also shared decision making with patients so that patients' preferences for small reductions in TVR could be integrated with their desires to adhere to DAPT and its potential costs and bleeding risks. Because the immediate financial cost of stents is not borne by patients while the burden of increased downstream TVR events is, it might be argued that physicians should implant DES in all eligible patients regardless of the expected magnitude of benefit without consideration for societal costs. However, the clear delineation of the small magnitude of benefit of DES to low-TVR-risk patients balanced against the costs and consequences of prolonged DAPT, as well as the uncertainty of future events possibly requiring premature DAPT discontinuation and exposure to an increased risk of stent thrombosis, may lead many low-TVR-risk patients to favor PCI with BMS, even without a consideration for societal costs.
We purposely modeled a strategy to reduce DES use by 50% in the low-TVR-risk group rather than to disallow DES altogether in certain subgroups. This strategy preserves clinicians' and patients' abilities to exercise their judgment and preferences on a case-by-case basis and at the same time sets a target goal of substantial cost savings. Our intention was not to advocate a sweeping policy change that would limit physician and patient autonomy, but rather to illustrate the potential for costs savings, without a significant increase in patient morbidity, that could be achieved with an evidence-based approach to stent selection, and to encourage shared decision making with patients.
This study has several potential limitations. First, the discrimination of the model used to estimate predicted TVR risk was modest (C statistic = 0.66). However, this model had better discrimination than the more commonly applied risk factors of diabetes, vessel diameter, and lesion length (0.66 vs 0.60) (P < .001).23 In addition, model calibration, a metric for assessing a model's ability to identify a low-TVR-risk group of patients, was excellent (Hosmer-Lemeshow P = .90).23 Thus, the projections of the increase in TVR events with the lower DES use strategy among low-risk patients and associated TVR costs are likely to be accurate. Next, we did not account for either the potential costs of major or nuisance bleeding events or the potential ischemic benefits of prolonged DAPT because these represent areas of uncertainty in the literature that are currently under investigation.34 Third, we did not have any assessment of patients' preferences regarding stent type or willingness to accept the costs and risks of prolonged DAPT in exchange for a reduced risk of repeated procedures. Such patients could still be among the 50% of low-risk patients who receive DES in our proposed strategy. Fourth, the MassDAC model has not been validated in the entire NCDR CathPCI registry, but only in Massachusetts. Fifth, it is impossible to account for unmeasured confounders in the CathPCI registry. Sixth, while our model of projected clinical and economic outcomes of PCI procedures was based on the best available clinical and economic data, the resulting projections cannot be directly verified using empirical data at present—a prospective trial would be necessary to confirm these results. Finally, we recognize that newer technology and its costs are constantly evolving; but the principles guiding our analysis in this “cross-sectional snapshot” of PCI practices should be maintained.
In conclusion, although the benefits of DES are greatest among patients at the highest risk for restenosis, DES use is common even among those predicted to be at the lowest restenosis risk. Furthermore, DES use is extremely variable among physicians. Given the marked variation in physicians' DES use, a strategy of lower DES use among patients at low risk of TVR could present an important opportunity to reduce health care expenditures while preserving the vast majority of their clinical benefit.
Correspondence: Robert W. Yeh, MD, MSc, Cardiology Division, GRB8-843, Massachusetts General Hospital, Boston, MA 02144 (firstname.lastname@example.org).
Accepted for Publication: May 21, 2012.
Published Online: July 9, 2012. doi:10.1001/archinternmed.2012.3093
Author Contributions: Drs Amin, Spertus, and Yeh, Mr Kennedy, and Ms Vilain had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Amin, Spertus, Cohen, Chhatriwalla, Vilain, Lai, Normand, and Yeh. Acquisition of data: Amin, Spertus, and Messenger. Analysis and interpretation of data: Amin, Spertus, Chhatriwalla, Kennedy, Salisbury, Venkitachalam, Lai, Mauri, Rumsfeld, Messenger, and Yeh. Drafting of the manuscript: Amin, Chhatriwalla, and Yeh. Critical revision of the manuscript for important intellectual content: Amin, Spertus, Cohen, Chhatriwalla, Kennedy, Vilain, Salisbury, Venkitachalam, Lai, Mauri, Normand, Rumsfeld, Messenger, and Yeh. Statistical analysis: Amin, Kennedy, and Lai. Obtained funding: Spertus. Administrative, technical, and material support: Spertus, Vilain, Rumsfeld, and Messenger. Study supervision: Spertus, Chhatriwalla, Lai, and Mauri. Interpretation of statistics: Normand.
Financial Disclosure: Dr Spertus is the primary investigator of a contract from the American College of Cardiology Foundation to serve as an Analytic Center for the NCDR. He also has an equity position in Health Outcomes Sciences. Dr David J. Cohen has received research grant support from Boston Scientific, Medtronic, Abbott Vascular, Eli Lilly, Daichi Sankyo, Accumetrix, BMS/Sanofi, Schering-Plough, and Edwards Lifesciences. He has also received consulting fees from Cordis and Medtronic and speaking honoraria from Eli Lilly and the Medicines Company. Dr Mauri receives institutional research support from Abbott, Boston Scientific, Cordis, Medtronic, Eli Lilly, Daiichi Sankyo, Bristol Myers Squibb, and sanofi-aventis and has consulted for Cordis and Medtronic. Dr Rumsfeld is Chief Science Officer for the NCDR. Dr Yeh is an investigator for the Harvard Clinical Research Institute and a consultant for the Kaiser Permanente Division of Research.
Funding/Support: Funding for the ACC NCDR CathPCI registry was provided by grants from the American College of Cardiology Foundation. This analysis was funded by the ACC. This research was supported by the American College of Cardiology Foundation National Cardiovascular Data Registry (NCDR). Dr Amin is funded, in part, by an award from the American Heart Association Pharmaceutical Round Table and David and Stevie Spina.
Role of the Sponsor: No sponsor participated in the design and conduct of the study; collection, analysis, or interpretation of the data; nor in the preparation, review, nor approval of the manuscript.