|Home | About | Journals | Submit | Contact Us | Français|
The economic outcomes of clinical management strategies are important in assessing their value to patients.
BARI 2D randomized patients with Type 2 diabetes and angiographically documented, stable coronary disease to strategies of 1) prompt revascularization vs. medical therapy with delayed revascularization as needed to relieve symptoms, and 2) insulin sensitization vs. insulin provision. Prior to randomization, the physician declared whether CABG or PCI would be used if the patient were assigned to revascularization. We followed 2005 patients for medical utilization and costs, and assessed the cost-effectiveness of these management strategies.
Medical costs were higher for revascularization than medical therapy, with a significant interaction with the intended method of revascularization (p<0.0001). In the CABG stratum, four-year costs were $80,900 for revascularization vs. $60,600 for medical therapy (p<0.0001). In the PCI stratum, costs were $73,400 for revascularization vs. $67,800 for medical therapy (p<0.02). Costs also were higher for insulin sensitization ($71,300) vs. insulin provision ($70,200). Other factors that significantly (p<0.05) and independently increased cost included insulin use and dose at baseline, female sex, white race, body mass index ≥30, and albuminuria.
Cost-effectiveness based on four-year data favored the strategy of medical therapy over prompt revascularization and the strategy of insulin provision over insulin sensitization. Lifetime projections of cost-effectiveness showed that medical therapy was cost-effective compared with revascularization in the PCI stratum ($600 per life-year added) with high confidence. Lifetime projections suggest revascularization may be cost-effective in the CABG stratum ($47,000 per life-year added), but with lower confidence.
Prompt coronary revascularization significantly increases costs among patients with Type 2 diabetes and stable coronary disease. The strategy of medical therapy (with delayed revascularization as needed) appears to be cost-effective compared with the strategy of prompt coronary revascularization among patients identified a priori as suitable for PCI.
Patients with Type 2 diabetes are at increased risk of developing coronary artery disease, which in turn is the leading cause of death in this population. Patients with diabetes and coronary disease need effective treatments for both the metabolic derangements of diabetes and for the myocardial ischemia due to coronary atherosclerosis. The Bypass Angioplasty Revascularization Investigation 2 Diabetes (BARI 2D) trial used a two-by-two factorial design to compare alternative treatment strategies for diabetes and coronary disease in patients with both conditions (1, 2).
The economic consequences of clinical strategies for diabetes and coronary disease are also important outcomes. The net costs of a clinical strategy are determined not only by the inherent costs of the intended treatment itself, but also by the subsequent costs related to clinical consequences of that treatment. Since the economic outcomes are determined in part by clinical outcomes, they are best compared in the setting of a randomized clinical trial, which provides unbiased estimates of the outcomes of alternative treatments. The goal of this study was to conduct a prospective evaluation of the economic outcomes, including medical costs and cost-effectiveness, of the alternative treatment strategies for Type 2 diabetes and stable coronary disease that were tested in the BARI 2D clinical trial.
The design of the BARI 2D trial and of the economic evaluation have been described in detail (3, 4). In brief, patients with Type 2 diabetes and stable, angiographically documented coronary disease were eligible for randomization. Patients were excluded if they required immediate coronary revascularization or had significant left main disease, a creatinine >2.0 mg/dl, a HbA1c >13%, or either coronary artery bypass graft (CABG) surgery or percutaneous coronary intervention (PCI) within 12 months.
Patients were randomly assigned according to a two-by-two factorial design to a 1) Glycemic Management Strategy comparing insulin sensitization (with metformin or rosiglitazone, or both) versus insulin provision (with insulin or sulfonylurea, or both) to achieve the target of HbA1c 7.0% or lower, and 2) Revascularization Strategy comparing prompt coronary revascularization combined with intensive medical management versus intensive medical management alone, with coronary revascularization at a later date only if clinically indicated. A key feature of BARI 2D was that before randomization the responsible physician determined whether CABG or PCI would be used if the patient were assigned to prompt coronary revascularization and randomization was stratified by the intended revascularization strategy (CABG or PCI). The primary endpoint of the trial was total mortality, and the principal secondary endpoint was major cardiovascular events; economic outcomes and cost-effectiveness were secondary endpoints of the trial. As reported previously, total mortality did not differ significantly in either randomized comparison, but major cardiovascular events were significantly reduced by revascularization in the CABG stratum (1).
Patients were enrolled at one of 49 clinical sites in the United States, Canada, Brazil, Mexico, the Czech Republic and Austria; 46 of these clinical sites agreed to assess economic outcomes. Economic data were collected by the staff of the Economic Core Laboratory at Stanford University for 41 sites and by the local study staff at five sites. Patients were queried every three months on their use of specific medical resources, including hospital admissions, physician visits, outpatient tests and procedures, and prescription medications.
Medical care costs were measured for each patient by applying a standardized cost weight to each medical resource used. Hospital admissions were assigned to a Diagnosis Related Group and costs were calculated using the fiscal year 2007 weights and the national conversion factor. Physician fees from the 2007 Medicare schedule were used for office tests and physician visits and for inpatient procedures. Prescription drugs were assigned costs based on 2007 average wholesale prices. Indirect costs, such as those related to employment, were not measured in this study. In sensitivity analysis, we converted hospital charges to costs, and used medication prices from drugstore.com.
Cumulative costs and medical utilization over four years of follow-up were calculated using an actuarial approach in which surviving patients under observation provided data on costs/utilization incurred during each three month follow-up interval, which was then multiplied by the Kaplan–Meier survival rate to estimate mean incremental costs/utilization (5). Costs were expressed in 2007 United States dollars, and discounted at a 3% annual rate, as is conventional in health economic studies.
Cumulative costs/utilization were compared on an intention-to-treat basis between the two glycemic control strategies (insulin provision versus insulin sensitization), and between the two revascularization strategies (prompt revascularization versus optimal medical therapy with delayed revascularization if needed to relieve symptoms). These endpoints were also compared within the predefined strata according to the mode of revascularization chosen prior to randomization as most appropriate for the patient (CABG or PCI). Statistical significance was assessed using permutation tests, and confidence limits were calculated using percentiles of bootstrap resamples.
We assessed the effects of baseline clinical factors upon cumulative medical costs using multivariable regression, after log transformation of cost. We performed this analysis using costs at two years because almost all patients had complete data at that point. The effect of each clinical factor was first tested in a model that adjusted only for the two randomization assignments, the stratum of intended revascularization, the length of follow-up, and the country of enrollment. The effect of each baseline clinical factor on cost was subsequently assessed in a multivariable model that adjusted for the other baseline characteristics. We tested for interaction between baseline characteristics and randomization assignment on two-year costs in a model that included all main effects of baseline characteristics. In view of the multiplicity of factors to be tested, a priori we hypothesized that prior insulin use would have a significant interaction with assignment to insulin provision or insulin sensitization, and that the stratum of intended revascularization (CABG or PCI) and history of prior revascularization (none, prior PCI only, prior CABG) would have significant interactions with assignment to prompt revascularization or medical therapy.
We assessed the cost-effectiveness of the randomized strategies using survival and cost data from the trial. The incremental cost-effectiveness ratio at time “t” was calculated as:
where Cost(t i) indicates the cumulative cost to time t for patient group i, and LY(ti) indicates the cumulative life-years of survival to time t for patient group i. In a sensitivity analysis, we calculated quality adjusted life-years of survival with an adaptation of our previously reported method (6), estimating utility based on BARI 2D data on the Duke Activity Status Index, self-reported health status, Canadian Cardiovascular Society Class for angina, and health rating.
We assessed cost-effectiveness up to four years using observed follow-up data, and projected long-term cost-effectiveness using a two-step procedure (7). The observed mortality for the entire patient cohort, irrespective of randomization, after one year of follow-up was compared with the age, sex and race specific expected mortality from the 2007 US Life Tables. We assumed the excess mortality in the BARI 2D patient cohort would continue, and estimated remaining life expectancy for each patient alive at last follow-up (mean 5.3 years) (7). In the base case analyses, we assumed that the differences in medical costs between randomized groups observed between one and four years would decline linearly over the next four years of follow-up, and that subsequent follow-up costs would be equivalent. We assessed the variability of the cost-effectiveness ratios using 1,000 bootstrap resamplings of the patient population. We tested sensitivity of the cost-effectiveness ratios to different scenarios, including: quality adjustment of survival; reduced survival after non-fatal myocardial infarction (by two years) and non-fatal stroke (by three years); and assuming cost differences seen between years one and four would either persist indefinitely or cease immediately after four years.
Analyses were performed using SAS 9.1.3 (SAS Institute, Cary, NC), and R 2.4.0.
Of the 2,368 patients randomized in BARI 2D, 2,246 (95%) were enrolled at sites that provided economic follow-up data. One hundred eighty-two patients at participating sites refused economic follow-up, and a further 55 patients withdrew before providing any economic data. Over subsequent follow-up, a further 67 patients withdrew and 33 patients were lost. Follow-up for economic outcomes extended to one year for 96% of surviving participants, to two years for 88%, to three years for 61%, and to four years for 34%.
Four-year cumulative costs of patients randomized to the strategy of prompt coronary revascularization ($75,900) were significantly higher (p<0.001) than those of patients randomized to medical therapy ($65,600). The cost difference between these strategies emerged immediately, but narrowed over follow-up from $16,800 at one year, to $14,900 at two years, $13,000 at three years, and $10,200 at four years (Figure 1, bottom panel). Most of the difference in cost between the randomized strategies was due to the higher initial costs of the assigned coronary revascularization procedure (Table 1A), which were only partially offset by subsequently lower costs for medications and cardiovascular testing (Table 1A).
Prior to randomization, the responsible physician chose whether CABG or PCI would be used if the patient were assigned to the prompt revascularization strategy. The cumulative cost was significantly higher (p<0.001) in the patients assigned to prompt revascularization, regardless of whether CABG ($20,300 higher than medical therapy) or PCI ($5,700 higher than medical therapy) was selected a priori as the intended method of revascularization. The difference in costs between the randomized strategies narrowed progressively over follow-up in both strata (Figure 2), but the patterns of resource use and cost differed somewhat according to the stratum of intended revascularization (Tables 1B, ,1C).1C). There were greater differences in hospital days and hospital costs between the patients assigned to prompt revascularization and the patients assigned to medical therapy in the CABG stratum than in the PCI stratum, but also greater savings in medication costs (Tables 1B, ,1C1C).
Cumulative medical costs among patients randomized to the insulin sensitization strategy ($71,300) were higher than the cost of patients randomized to the insulin provision strategy ($70,200), but this was not significant at four years (p=0.81). The difference in cost between the insulin sensitization and insulin provision strategies widened between one year ($1,200) and two years ($2,000) of follow-up, then narrowed at three years ($1,000) and four years ($1,100) (Figure 1, top panel). The difference in cost between the glycemic control strategies was almost entirely due to the significantly higher cost of diabetes-related medications among patients randomized to the insulin sensitization strategy (Table 2). There were fewer diabetes-related hospitalizations among patients assigned to insulin sensitization, but no other significant differences in cost or utilization between the glycemic control strategies (Table 2).
We tested the effect of baseline factors on cumulative costs using a multiple regression model after applying a log transformation to two-year cost (Table 3). Randomization to insulin sensitization increased cost by 11% at two years (p<0.0001). There was a significant interaction (p<0.0001) between assignment to prompt revascularization and the intended mode of revascularization on two-year cost. In the CABG stratum, prompt revascularization increased two-year costs by 112% (p<0.0001) compared with medical therapy, whereas in the PCI stratum prompt revascularization increased costs by 54% (p<0.0001) compared with medical therapy.
Use of insulin at the time of study entry increased costs overall by 22% (p<0.0001), but the hypothesized interaction with randomization to insulin sensitization was not significant (p=0.94). A history of prior coronary revascularization did not affect two-year cost (p=0.33) and the hypothesized interaction with randomization to prompt revascularization was not significant (p=0.77).
Few of the remaining baseline factors affected medical costs significantly. Baseline HbA1c levels and duration of diabetes each had a graded effect on two-year cost (p<0.0001) when tested individually, but were not significant predictors of cost after adjustment for other baseline factors (Table 3). Women had 10% higher costs (p=0.002), as did patients with a body mass index of 30 or more (p=0.0008). Costs were increased 8% by microalbuminuria and 16% by macroalbuminuria (Table 3). None of these baseline factors had a significant interaction with treatment assignment.
In the fully adjusted model, randomization assignment, baseline insulin use and dose, race, female sex, body mass index, and albuminuria each had significant, independent effects on cumulative costs (Table 3).
Mean cumulative costs over four years were increased when the analysis was restricted to the 1,279 patients (64%) enrolled in United States clinical sites (Table 4), and when hospital costs were estimated using charges and the ratio of cost to charges rather than diagnosis-related group reimbursements (Table 4).
The incremental cost-effectiveness of treatment strategies was assessed using the within-trial data (Table 4) and also after projecting lifetime outcomes (Table 5). Since the prompt revascularization strategy had a significant interaction with the intended form of revascularization upon both clinical outcomes (1) and costs, we evaluated its cost-effectiveness separately in the stratum of patients with CABG intended and the stratum of patients with PCI intended.
Within the PCI stratum, medical therapy yielded more life-years of survival over four years at lower cost, and was preferred over revascularization in 99.9% of bootstrap replications at the $50,000 per life-year added benchmark (Table 4). In the lifetime projection, medical therapy had slightly higher costs ($238,100 vs. $237,900) but more life-years of survival (14.03 vs. 13.70). The cost-effectiveness ratio was $600 per life-year added in the base case analysis, and the medical strategy was preferred over the revascularization strategy in 95% of bootstrap replications (Table 5, Appendix Figure).
Within the CABG stratum, prompt revascularization provided slightly fewer life-years of survival over four years follow-up, at significantly higher cost (Table 4). However, in the lifetime projection CABG increased survival from 12.90 to 13.42 years, and increased costs from $210,900 to $235,500, yielding a favorable lifetime cost-effectiveness ratio of $47,000 per life-year added. Using a $50,000 per life-year added benchmark, CABG was preferred in 56% of bootstrap replications in the lifetime projection (Table 5, Appendix Figure).
Over four years of follow-up, the insulin provision strategy had lower costs and higher survival (Table 4). In a lifetime projection of cost and survival, however, the insulin sensitization strategy had slightly higher costs ($239,100 vs. $236,800) and survival (13.66 vs. 13.61 life-years), yielding an incremental cost-effectiveness of $52,000 per life-year added. This lifetime cost-effectiveness estimate was highly variable in bootstrap replications, with insulin sensitization favored in 51% of replications and insulin provision in 49% using the benchmark of $50,000 per life-year added (Table 5, Appendix Figure).
The results of the lifetime cost-effectiveness estimates were not changed appreciably by adjusting for quality of life, by adjusting for reduced life-expectancy after myocardial infarction or stroke, or by assuming there were no further differences in costs between randomized groups after four years (Table 5). When the cost differences seen between one and four years of follow-up were assumed to continue indefinitely, however, medical therapy became less attractive in the PCI stratum ($57,000 per life-year added), as did insulin sensitization ($82,000 per life-year added), whereas prompt revascularization in the CABG stratum became more attractive ($12,000 per life year added).
Much of the cost of medical care in the population is generated by the management of patients with chronic illnesses such as coronary disease and diabetes. Effective treatments for chronic diseases may increase overall medical costs, even after accounting for subsequent savings due to prevention of costly complications. These higher net costs may be acceptable, however, if clinical outcomes are sufficiently improved. The assessment of the value provided by treatments therefore requires measuring both their clinical and economic consequences, and weighing these outcomes to assess their cost-effectiveness compared with alternative treatments.
In this study, we found that prompt coronary revascularization was significantly more costly than medical therapy for patients who have coronary disease and diabetes. The high initial procedural costs of CABG and PCI were only partially offset by later cost savings over four years of follow-up, a finding consistent with previous studies. The cost-effectiveness of coronary revascularization in the BARI 2D patient population was more difficult to assess owing to the limited duration of follow-up (four years) compared with the projected life-expectancy of the population (over 15 years). The limited time horizon of the trial introduces a bias into the cost-effectiveness estimates of procedures, since the full costs are captured but only a portion of the benefits. Life-time projections of cost-effectiveness can remove this bias, but introduce uncertainties due to the model.
Medical therapy was quite cost-effective compared with prompt revascularization among patients in the PCI stratum, both in the within trial analysis (Table 4) and the lifetime projection ($600 per life-year added). These results were robust in bootstrap replications and sensitivity analyses (Table 5), strongly suggesting that a strategy of medical therapy, with delayed revascularization only if clinically indicated, is more economically attractive than a strategy of prompt revascularization with PCI.
This result is consistent with the economic outcomes reported by other randomized trials of PCI and medical therapy in stable coronary disease. In the COURAGE trial, the cumulative costs among patients assigned to PCI were $11,000 higher over three years of follow-up than those of the patients assigned to medical therapy, and the lifetime cost of PCI patients was projected to be $9,500 higher. The cost-effectiveness ratio for PCI compared with medical therapy was $262,000 per life-year added in COURAGE (8). The RITA-2 trial found that costs were significantly higher over three years follow-up (by 2685 pounds, or $4,385) among patients randomized to PCI compared with medical therapy (9). Costs over one year follow-up were higher among PCI assigned than medically assigned patients in the MASS II trial (10), and in the TIME trial patients assigned to invasive therapy (mostly PCI) had higher costs over one year of follow-up than patients assigned to medical therapy (11). In decision models the cost-effectiveness of PCI depends primarily on the severity of angina, since highly symptomatic patients benefit from relief of angina (12, 13). Patients with severe symptoms despite medical therapy were excluded from BARI 2D, however. In patients with stable coronary disease and mild symptoms, PCI was not cost-effective, since it did not improve survival and led to much higher costs.
Patients in the CABG stratum had much higher costs after random assignment to prompt revascularization than to medical therapy, driven by the higher initial procedure costs. Patients in the CABG stratum were, however, significantly more likely to be free of major cardiovascular events after revascularization (77.6% at five years) than after medical therapy (69.5%). Over four years of follow-up, this reduction in clinical events did not increase life-expectancy sufficiently to be considered cost-effective by conventional benchmarks. Four years of follow-up is sufficient to capture all the costs of CABG, but only part of the benefit, however, and a lifetime projection is necessary to provide a fair perspective for the economic evaluation. These projections suggest that CABG may well be cost-effective compared with medical therapy for patients with diabetes ($47,000 per life-year added). This estimate was, however, variable in the bootstrap analysis and somewhat sensitive to model assumptions, and thus must be interpreted cautiously (Table 5).
BARI 2D represents, to our knowledge, the first trial-based comparison of the economic outcomes of CABG and medical therapy. The major trials comparing CABG with medical therapy were conducted before economic data were collected alongside clinical data in randomized trials. Decision models suggest that CABG is cost-effective relative to medical therapy among patients with either extensive anatomic disease or severe angina symptoms (12, 14). Patients with left main disease, extensive coronary disease, or severe angina requiring CABG were excluded from BARI 2D, however. Patients in the CABG stratum typically had more severe coronary disease, with three-vessel disease in 53% and reduced left-ventricular ejection fraction in 18% (15). Our results are broadly consistent with the earlier decision models that demonstrated CABG to be cost-effective compared with medical therapy in higher risk patients (12–14).
The insulin sensitization strategy in BARI 2D was more expensive than the insulin provision strategy, largely because of the higher cost of thiazolidinediones (Table 2). There was little evidence that the higher cost of insulin sensitizing drugs was offset by reductions in other cost categories (Table 2) or by fewer clinical complications. The cost-effectiveness of the insulin sensitization strategy was therefore not favorable over the four-year time horizon of the trial (Table 4), but was more favorable in the lifetime projection (Table 5). This lifetime cost-effectiveness estimate was essentially a “toss-up”, however, since there were insignificant differences in long-term cost and survival between the insulin sensitization and insulin provision strategies (Appendix Figure).
The glycemic control strategies tested in BARI 2D differ from the treatments evaluated in other trials, which generally assessed specific drugs or intensive versus conventional management approaches. Several prior studies suggest that intensive treatment to lower a HbA1c target may be cost-effective compared with more conventional management of diabetes (16–21). In BARI 2D, however, the HbA1c target was the same in both the insulin provision and insulin sensitization strategies. There have been relatively few economic evaluations of the newer thiazolidinediones for patients with diabetes (22). An economic model based on the PROactive trial (23) suggested that use of pioglitazone among patients with type 2 diabetes and evidence of macrovascular disease may be cost-effective (24). Over three years of follow-up, the patients assigned to pioglitazone had higher total costs (by 102 pounds, or $167) and greater survival, yielding a cost-effectiveness ratio of 5396 pounds ($8,811) per QALY, with a projected lifetime value of 4060 pounds ($6,631) per QALY. The PROactive trial design differed from that of BARI 2D in several ways, most notably in being a trial of a specific drug rather than a management strategy, and in achieving significantly different levels of glycemic control between the study groups.
This study has a number of limitations, the most important of which is that economic follow-up extended only to four years, less than the average of 5.3 years of clinical follow-up. Therefore, the within-trial cost-effectiveness evaluation did not capture the full effect of the assigned treatments on patient survival. While we projected survival and costs to assess lifetime cost-effectiveness, these estimates are subject to various uncertainties, and consequently must be interpreted cautiously. Furthermore, other clinical trials in diabetes have shown that survival differences may become evident only after ten years or more (25, 26). Longer follow-up of the BARI 2D patients, which is under consideration, would clarify the clinical effectiveness and cost-effectiveness of these strategies.
In conclusion, the strategy of prompt revascularization in patients with diabetes and coronary disease is significantly more costly than the strategy of medical therapy with delayed revascularization as needed. The medical strategy was cost-effective compared with the revascularization strategy in the stratum of patients with less severe coronary disease most suitable for PCI. Within the stratum of patients with more severe coronary disease identified as most suitable for CABG, the revascularization strategy may ultimately provide sufficient clinical benefits to be considered cost-effective.
The BARI 2D clinical trial randomized patients with diabetes and coronary disease to prompt coronary revascularization versus medical therapy, as well as to strategies of glycemia control based on use of either insulin sensitizers or drugs that increase insulin provision. Mortality over five years was not significantly different, but major cardiovascular events were reduced by revascularization in the CABG stratum. The present study shows that revascularization increases four-year cost significantly, by roughly $5,700 (PCI) to $20,300 (CABG), whereas insulin sensitization increases costs by $1,100. Medical therapy was highly cost-effective compared with prompt revascularization in the PCI stratum ($600/life-year added), suggesting that revascularization can be delayed until clinically indicated in patients with less extensive coronary disease. Revascularization may be cost-effective in patients with more extensive disease amenable to CABG, but this result was less certain with the limited follow-up available.
As an NIH funded trial, we are required to abide by the NIH PubMed Central Policy that we retain the right to provide a copy of the final manuscript to the NIH upon acceptance for publication by your journal, for public archiving in PubMed Central as soon as possible, but no later than 12 months after publication.
We thank Robert L. Frye, MD, and Sheryl F. Kelsey, PhD, for their support of this study.
BARI 2D was funded by the National Heart, Lung and Blood Institute and the National Institute of Diabetes and Digestive and Kidney Diseases (HL061748, HL061744, HL061746, HL063804).
BARI 2D received significant supplemental funding provided by GlaxoSmithKline, Collegeville, PA, Lantheus Medical Imaging, Inc. (formerly Bristol-Myers Squibb Medical Imaging, Inc.), North Billerica, MA, Astellas Pharma US, Inc., Deerfield, IL, Merck & Co., Inc., Whitehouse Station, NJ, Abbott Laboratories, Inc., Abbott Park, IL, and Pfizer, Inc, New York, NY. Generous support was given by Abbott Laboratories Ltd., MediSense Products, Mississauga, Canada, Bayer Diagnostics, Tarrytown, NY, Becton, Dickinson and Company, Franklin Lakes, NJ, J. R. Carlson Labs, Arlington Hts., IL, Centocor, Inc., Malvern, PA, Eli Lilly and Company, Indianapolis, IN, LipoScience, Inc., Raleigh, NC, Merck Sante, Lyon, France, Novartis Pharmaceuticals Corporation, East Hanover, NJ, and Novo Nordisk, Inc. Princeton, NJ.
ClinicalTrials.gov Registration Number: NCT00006305
The authors declare no financial conflicts of interest with respect to this work.