Cost-Effectiveness of Percutaneous Coronary Intervention with Drug Eluting Stents versus Bypass Surgery for Patients with Diabetes and Multivessel Coronary Artery Disease: Results from the FREEDOM Trial

Abstract

Background

Studies from the balloon angioplasty and bare metal stent eras have demonstrated that CABG is cost-effective compared with PCI for patients undergoing multivessel coronary revascularization—particularly among patients with complex CAD or diabetes. Whether these results apply in the drug-eluting stent (DES) era is unknown.

Methods and Results

Between 2005 and 2010, 1900 patients with diabetes and multivessel CAD were randomized to PCI with DES (DES-PCI; n=953) or CABG (n=947). Costs were assessed from the perspective of the U.S. health care system. Health state utilities were assessed using the EuroQOL. A patient-level microsimulation model based on U.S. life-tables and in-trial results was used to estimate lifetime cost-effectiveness. Although initial procedural costs were lower for CABG, total costs for the index hospitalization were $8,622/patient higher. Over the next 5 years, follow-up costs were higher with PCI, owing to more frequent repeat revascularization and higher outpatient medication costs. Nonetheless, cumulative 5-year costs remained $3,641/patient higher with CABG. Although there were only modest gains in survival with CABG during the trial period, when the in-trial results were extended to a lifetime horizon, CABG was projected to be economically attractive relative to DES-PCI, with substantial gains in both life expectancy and quality-adjusted life expectancy and incremental cost-effectiveness ratios <$10,000 per life-year or quality-adjusted life-year gained across a broad range of assumptions regarding the effect of CABG on post-trial survival and costs.

Conclusions

Despite higher initial costs, CABG is a highly cost-effective revascularization strategy compared with DES-PCI for patients with diabetes and multivessel CAD.

The optimal approach to revascularization in patients with diabetes and multivessel coronary artery disease (CAD) has been widely debated since the Balloon Angioplasty Revascularization Investigation (BARI) trial first reported improved survival for diabetic patients treated with bypass surgery (CABG) as compared with balloon angioplasty.¹ In addition to their relative clinical benefits, the long-term economic impact of alternative revascularization strategies is an important consideration for both clinical guidelines and health policy. A cost-effectiveness analysis based on data from the BARI trial demonstrated that for the average diabetic patient, CABG was a cost-effective strategy compared with conventional balloon angioplasty.^2,3 Since BARI, however, improvements in both surgical and percutaneous revascularization techniques, as well as medical therapy, have had an impact on the comparative clinical outcomes of the two revascularization strategies in both the short- and long-term. Nonetheless, most recent comparisons of revascularization outcomes in patients with diabetes in the bare metal and drug-eluting stent (DES) era have shown a trend toward more frequent major adverse cardiac and cerebrovascular events (MACCE) with PCI relative to CABG.^4–8 While the rate of repeat revascularization procedures has been reduced with stenting and DES, rates have remained higher than for CABG.^5,9 Moreover, economic analyses based on the available data (generally based on subset analyses of trials or observational studies) have suggested that CABG remains a cost-effective strategy.^3,10–14

The Future REvascularization Evaluation in patients with Diabetes mellitus: Optimal management of Multivessel disease (FREEDOM) trial is the largest prospective randomized trial to compare the outcomes of multivessel coronary artery revascularization with drug-eluting stents vs. CABG among diabetic patients.¹⁵ Recently, 5-year results from FREEDOM showed that CABG was associated with significantly lower rates of the primary endpoint of death, MI or stroke, compared with DES-PCI, with the benefit driven by reductions in both death and MI.¹⁶ To provide additional insight into the relative value of these alternative revascularization strategies, we performed a prospective health economic evaluation alongside the FREEDOM trial, the results of which are the focus of this report.

METHODS

The design and methods of the FREEDOM trial have been described previously.¹⁵ Between April 2005 and April 2010, 1900 patients with type 1 or 2 diabetes and angiographically confirmed multivessel CAD with a clinical indication for revascularization and who were deemed suitable for both DES-PCI and CABG were randomly assigned on a 1:1 basis to either technique. PCI procedures were performed using standard techniques. Given the timeframe of patient recruitment, both sirolimus-eluting stents (SES) and paclitaxel-eluting stents (PES) were the predominant stents used in the trial. The trial protocol recommended that patients randomized to PCI receive only one type of DES. Newer generation DES were allowed during the later phases of the trial, provided there was regulatory approval for use within the respective region. The use of abciximab was recommended for patients undergoing PCI and was provided free of charge; at least 12 months of dual antiplatelet therapy with aspirin and clopidogrel was recommended after any DES procedure. Patients randomized to CABG underwent treatment according to standard techniques. Following revascularization, optimal medical therapy including high dose statin therapy, aggressive blood pressure control, and normalization of hemoglobin-A1c was recommended for both groups.

In-person assessments were performed at 1 month post-procedure, and 6 and 12 months post- randomization, and annually thereafter. In addition, telephone follow-up assessments were performed at 18 months and all subsequent semiannual time points. All sites obtained Institutional Review Board approval of the protocol, and all patients provided informed consent. The trial is registered at the National Institutes of Health website (www.clinical-trials.gov) as identifier NCT00086450.

Estimation of Medical Care Costs

Medical care costs for the index hospitalization and in-trial follow-up period were assessed using a combination of resource-based and event-based methods as described below. All costs were assessed from the perspective of the U.S. healthcare system and are reported in 2010 U.S. dollars.

PCI and CABG Procedure Costs

Detailed resource use was recorded for each revascularization procedure, and the cost for each item was estimated on the basis of the mean hospital acquisition cost for the item at 3 surveyed U.S. hospitals. Each DES was assigned a cost of $1500. Costs of antithrombotic therapy were based on the most current wholesale acquisition cost obtained from Micromedex Red Book.¹⁷ Costs of additional disposable equipment, overhead and depreciation for the cardiac catheterization laboratory and the operating room, and non-physician personnel were estimated using data from the micro-cost accounting systems of Saint Luke’s Mid America Heart Institute and adjusted for actual procedure duration. Resource utilization and cost data for the initial PCI procedure and any planned staged PCI procedures were combined in the reporting of results for the index procedure.

Post-Procedure Hospitalization Costs

Post-procedure costs for index hospitalizations were estimated using regression models based on FREEDOM-eligible Medicare patients who underwent either PCI (n=113,921) or CABG (n=43,866) and whose hospitalization data were included in the 2010 Medicare Provider and Review (MEDPAR) database. Total costs for these hospitalizations were estimated by multiplying hospital charges by the hospital and cost-center-specific cost-to-charge ratio.^18,19 Linear regression models were then developed, using total Medicare hospitalization costs as the outcome, and socioeconomic factors, comorbidities and inhospital complications (identified on the basis of ICD-9 codes) as predictors (see Supplemental Table 1). Because of substantial variability in length of stay for revascularization procedures across the enrolling countries in FREEDOM, length of stay was not included as a predictor in these models. The final models for PCI and CABG were then used to predict non-procedural costs for each index hospitalization as well as any subsequent hospitalizations that involved a coronary revascularization procedure. To avoid double-counting procedural costs, the intercept for each model was adjusted to remove the costs directly related to the revascularization procedures, themselves, based on national averages for procedure duration and resource use.¹¹

For follow-up hospitalizations that did not involve a revascularization procedure, MS-DRGs were assigned based on the primary indication for hospitalization and procedures performed during the hospitalization. Costs were then assigned based upon mean 2010 Medicare reimbursement rates for the DRG obtained from the Medicare Part A data files.²⁰

Other Costs

Physician fees for PCI procedures and CABG procedures (including those for the primary surgeon, surgical assistant, and anesthesiologist) were based on the 2010 national Medicare fee schedule. Non-procedure related physician fees for revascularization-related hospitalizations were estimated for U.S. patients on the basis of post-procedure ICU and non-ICU length of stay and Medicare payment rates; for non-U.S. patients, post-procedure length of stay following CABG and PCI was estimated from regression models developed using 2010 MedPAR data and the same covariates as used in the cost models (see Supplemental Tables 1 and 2), and PCI and CABG-specific ratios of ICU vs. total post-procedure length of stay estimated from the trial data for U.S. patients. Physician costs for all other hospitalizations were estimated as a percentage of hospital costs according to DRG.^21,22 Costs for outpatient visits, tests and procedures, and inpatient rehabilitation and skilled nursing facility days were estimated using national average 2010 Medicare reimbursement rates. Because the length of stay in cardiac rehabilitation varied considerably at the country level, the average number of rehab days according to treatment group and follow-up year for U.S. patients, was used to estimate the cost of rehabilitation stays for patients enrolled at non-U.S. sites. Outpatient medication use was assessed at each follow-up visit, and costs were assigned using the most current average wholesale prices from Micromedex Red Book.¹⁷ To account for expected reductions in the cost of generic clopidogrel in the very near future, it was assigned a cost of $30/month.

Quality of Life

The EuroQOL (EQ-5D) health status instrument was used to assess quality of life for each study patient at baseline, 1 month post-procedure, 6 and 12 months after randomization, and annually thereafter. Health state utility weights (range 0–1, higher=better health) were obtained from the EuroQOL data using an algorithm developed from the U.S. population.²³

Statistical Analysis

Of the 1900 patients randomized in the FREEDOM trial, 36 patients assigned to CABG, and 9 patients assigned to PCI did not undergo any index procedure. All of these patients either died (n=3) or withdrew from the study (n=42) within 7 days of randomization. To avoid bias due to higher rates of withdrawal prior to CABG, the primary analytic population for the economic study consisted of all randomized patients who underwent at least one initial revascularization procedure, with patients categorized according to their assigned treatment (modified intention-to-treat [mITT] population, n=1855). A secondary analytic population included only those patients who underwent the revascularization procedure assigned (per protocol [PP] population, n=1832) and was used solely to examine initial treatment costs among those patients who actually underwent the specified procedures.

Categorical data are reported as frequencies, and continuous data are reported as mean ± standard deviation. Discrete variables were compared using Fisher’s exact test. Normally distributed continuous variables were compared using Student’s t-test, and non-normally distributed data were compared using the Wilcoxon rank-sum test. Treatment effects from Poisson regression models were used for the comparison of hospitalization rates. Kaplan Meier survival curves and logrank tests were used for the comparison of 5-year clinical events. Cost data are reported as both mean and median values, and confidence intervals for the differences in costs between treatment groups were obtained via bootstrapping.²⁴

Quality-adjusted life expectancy during the trial period was estimated for each patient as the time-weighted average of his or her utility value, using the midpoint between assessments as the transition between health states, starting at the 30-day visit. The baseline utility was applied to the time from randomization to the index procedure, and the 30-day utility value was applied to the period from the procedure through the midpoint between the 30 day and 6-month follow-up. Missing utility values were estimated using multiple imputation, with baseline patient characteristics, previous utility values, and previous in-trial clinical events informing the imputation.

In-Trial Analysis of Costs, Life-Years and QALYs

The prolonged recruitment period for the FREEDOM trial together with the fixed stopping point yielded a wide range of follow-up durations for enrolled patients. To accommodate this wide range of administrative censoring times, methods for the analysis of censored data were used to obtain estimates of cumulative costs and QALYs over time. An inverse probability-weighted estimator was applied, whereby the time axis was divided into 3 month intervals, and costs for each interval were estimated as the observed costs during the interval for patients with complete data divided by the probability of not being censored within the interval.²⁵ Similar methods were applied to estimation of quality-adjusted life expectancy. Life years gained at annual time points were estimated as the difference in the area between the Kaplan Meier survival curves for the two treatment groups. Confidence limits for the mean cumulative cost, life year and QALY estimates for each treatment group, as well as the difference between groups, were calculated using the bootstrap method.²⁴

Cost-Effectiveness

The cost-effectiveness of CABG vs. PCI was assessed over a lifetime horizon using both life-years and QALYs as measures of health benefit; the study protocol specified that the analysis based on QALYs would be considered the primary analysis, consistent with current US guidelines.²⁶ This analysis was based on a combination of (1) observed data up to the time of last follow-up for each patient, from which in-trial costs, life-years, and QALYs were estimated, and (2) projections of post-trial costs, life expectancy, and quality-adjusted life expectancy obtained from a Markov disease-simulation model. In this model, each surviving patient was assumed to face a monthly risk of death, with estimates of this risk based upon age-, sex- and race-matched risks of death obtained from U.S. life tables, calibrated to the observed 5-year mortality for the trial population.

For the PCI group, the comparison of the observed 5-year mortality for the trial population with that of an age, sex, and race-matched U.S. population yielded a mortality multiplier of 1.88. For the CABG group an additional multiplicative factor was applied in order to capture the prognostic benefit of CABG vs. DES-PCI. This multiplier was based on the hazard ratio derived from a landmark analysis of all-cause mortality from the FREEDOM patients, starting at 1 year after randomization. Three sets of analyses were performed based upon different assumptions regarding the duration of the survival benefit of CABG. The base case analysis assumed that the mortality hazard ratio for CABG vs. DES-PCI increased in a linear fashion from year 5 to year 10, and that there was no survival benefit of CABG beyond year 10 (i.e. hazard ratio=1). In sensitivity analyses, the survival benefit of CABG was assumed either (1) to remain constant through 10 years, with no further benefit beyond 10 years; or (2) to be in effect through 5 years only (i.e., no further benefit beyond the observed trial period).

Patient-level costs and utility weights applied to each projected year of life beyond the trial observation period were derived from regression models developed from the in-trial data (see Supplemental Tables 3 and 4). All projected life years, QALYs, and costs were discounted 3% annually based on time from randomization.

Uncertainty in the joint distribution of lifetime costs, life-years, and QALYs for each treatment group was estimated by the bootstrap method. To maintain consistency of the within-trial and post-trial CABG effect within each bootstrap sample, the effect of CABG on mortality, from the 1-year post-baseline landmark analysis, was re-estimated for each bootstrap replicate.

All analyses of data from the trial period, and the analyses of cost-effectiveness based on the combined trial data and long-term projections, were performed using SAS 9.3 (SAS Institute, Cary, NC). The Markov model used for projection of life expectancy beyond the trial was developed using TreeAge Pro 2012 (TreeAge Software, Inc., Williamstown, MA).

RESULTS

Patient Population

A total of 1900 patients with diabetes and multivessel CAD were randomized to either CABG (n=947) or DES-PCI (n=953). Of these, 45 patients did not undergo any form of revascularization (36 CABG, 9 PCI) and were excluded from the primary population for the economic analysis (Figure 1). Baseline characteristics for patients in the economic study population (mITT) and the 45 excluded patients (non-mITT) are summarized in Table 1. Patients who withdrew from the trial and had no index procedure tended to be older and have higher SYNTAX scores than the mITT population; otherwise there were no significant differences. Among the mITT population, there were no significant differences in any observed characteristics between the CABG and PCI groups. Nineteen percent of the mITT patients were enrolled in the U.S., more than 84% had 3-vessel CAD, and the median follow-up duration was 47 months.

Figure 1.

CONSORT Diagram. Black boxes represent the modified intention to treat (mITT) population that was the primary analytic population for the economic study. The grey boxes represent the per protocol (PP) population.

Table 1.

Baseline Characteristics

	Revascularized Patients (mITT)			Non-Revascularized Patients (non-mITT)			mITT vs non-mITT P value*
	CABG (n=911)	PCI (n=944)	P value	CABG (n=36)	PCI (n=9)	P value
Sociodemographic Characteristics
Age, years	62.9 ± 9.2	63.1 ± 8.9	0.67	66.2 ± 9.7	69.2 ± 8.5	0.399	0.006
Male, %	69.9	73.2	0.12	58.3	77.8	0.447	0.183
Body mass index (kg/m2)	29.8 ± 5.3	29.6 ± 5.4	0.52	29.8 ± 5.1	31.2 ± 4.3	0.491	0.686
Enrolling country, %
United States	19.1	18.7	0.86	16.7	33.3	0.354	0.848
Rest of world	80.9	81.2		83.3	66.7
Clinical Characteristics
Current smoker, %	16.7	14.8	0.28	13.9	11.1	1.000	0.836
Previous myocardial infarction, %	25.1	26.3	0.58	22.2	22.2	1.000	0.730
Peripheral Vascular Disease, %	10.5	10.1	0.76	8.3	22.2	0.258	0.804
COPD, %	5.4	3.2	0.02	5.6	22.2	0.173	0.130
Prior stroke, %	2.7	3.8	0.16	8.3	0.0	1.000	0.197
History of CHF, %	28.3	25.5	0.18	19.4	33.3	0.393	0.610
Angiographic Characteristics
LAD involvement, %	93.0	90.7	0.07	97.1	100.0	1.000	0.252
Three-vessel disease, %	84.3	82.1	0.21	88.2	100.0	0.564	0.297
SYNTAX score	26.0 ± 8.8	26.2 ± 8.4	0.71	28.2 ± 8.6	31.0 ± 9.8	0.411	0.040
SYNTAX score tertile
≤22	36.6	34.7	0.375	26.5	33.3	0.181	0.058
23–32	43.3	46.5		44.1	11.1
≥33	20.1	18.8		29.4	55.6
Initial treatment received, %
PCI	2.0	99.5	<0.001	0	0	--	--
CABG	98.0	0.5		0	0
None	--	--		100	100

COPD = chronic obstructive pulmonary disease; TIA = transient ischemic attack; CHF = congestive heart failure

^*comparison of combined treatment groups for mITT vs. non-mITT

Initial Treatment Costs

Among patients in the mITT cohort assigned to PCI, 99.5% underwent initial PCI and 0.5% underwent CABG. Among patients assigned to initial CABG, 98% underwent CABG and 2% underwent PCI. Resource utilization and costs for the initial revascularization procedures (including any staged procedures) are summarized in Table 2. Thirty-three percent of index PCI procedures were staged, with 31% involving 2 procedures, and 2% involving 3 or 4. On average, the initial PCI procedures required 2.3 guiding catheters, 3.1 guidewires, 3.4 angioplasty balloons, and 4.1 drug-eluting stents (range 0–13). While PCI procedure duration was considerably shorter than that for CABG, index procedure costs were significantly lower for CABG owing to the higher costs associated with stents and other consumable devices in the PCI group ($9,739 vs. $13,014 for the per protocol population, p<0.001). For the mITT population, the difference was slightly smaller ($9,776 vs. $12,998, p<0.001), since a small proportion of patients crossed over to the alternate treatment prior to revascularization.

Table 2.

Index Procedural Resource Utilization and Cost (Per Protocol Population)

	CABG (N=893)	PCI (N=939)	P value
Number of PCI Procedures, %
1	---	66.6 (625/939)
2	---	30.9 (290/939)
3	---	2.3 (22/939)
4	---	0.2 (2/939)
Procedure duration, minutes	248 ± 78	107 ± 67	<0.001
Drug-eluting stents	---	4.1 ± 1.9
Paclitaxel-eluting	---	1.9 ± 2.4
Sirolimus-eluting	---	2.1 ± 2.4
Other drug-eluting stents	---	0.1 ± 0.6
Bare metal stents	---	0.0 ± 0.3
Guiding catheters	---	2.3 ± 1.3
Guidewires	---	3.1 ± 2.3
Angioplasty balloons	---	3.4 ± 2.6
IABP	---	0.0 ± 0.2
Intravascular ultrasound catheters	---	0.1 ± 0.4
Atherectomy devices	---	0.1 ± 0.4
Contrast volume, ml	---	378 ± 186
Antithrombotic agents used, %
Unfractionated Heparin	---	87.2% (819/939)
Bivalirudin	---	16.0% (150/939)
Low molecular weight heparin	---	1.1% (10/939)
Abciximab	---	53.9% (506/939)
Other GP IIb/IIIa	---	28.6% (269/939)
Index Procedure Costs, $	9,739 ± 2,453 [9,477]	13,014 ± 5,173 [11,967]	<0. 001

Values in brackets represent medians

Clinical events and resource use during the index hospitalization for the mITT population are summarized in Table 3. Costs associated with the post-procedure hospital stay were significantly greater in the CABG group ($19,521 vs. $9,880, p<0.001) as were physician costs ($5,170 vs. $2,967, p<0.001). As a result, total index hospitalization costs were significantly higher for CABG than for DES-PCI ($34,467 vs. $25,845, p<0.001).

Table 3.

Index Hospitalization Events, Resource Utilization, and Costs (mITT population)

	CABG (N=911)	PCI (N=944)	Difference* (95% CI)	P value
Death, %	1.6% (15)	0.7% (7)	0.9% (−0.1%, 1.9%)	0.072
MI, %	1.3% (12)	1.6% (15)	−0.3% (−1.4%, 0.8%)	0.625
Stroke, %	1.6% (15)	0.2% (2)	1.4% (0.6%, 2.3%)	0.001
Unplanned CABG, %	0.2% (2)	1.3% (12)	−1.1% (−1.8%, −0.3%)	0.009
Complications, %
Major bleeding	3.4% (31)	2.8% (26)	0.6% (−0.9%, 2.2%)	0.124
Transfusion	32.8% (299)	3.8% (36)	29.0% (25.7%, 32.3%)	0.009
Atrial arrhythmia	18.0% (164)	1.7% (16)	16.3% (13.7%, 18.9%)	0.009
Ventricular arrhythmia	2.2% (20)	0.5% (5)	1.7% (0.6%, 2.7%)	0.42
Post-op infection	5.5% (50)	0.2% (2)	5.3% (3.8%, 6.8%)	< 0.001
Renal Failure	4.8% (44)	0.7% (7)	4.1% (2.6%, 5.6%)	< 0.001
Respiratory Failure	2.4% (22)	0.0% (0)	2.4% (1.4%, 3.4%)	0.002
Cardiogenic shock	1.1% (10)	0.4% (4)	0.7% (−0.1%, 1.5%)	< 0.001
Major vascular	4.3% (39)	3.5% (33)	0.8% (−1.0%, 2.5%)	0.25
Reoperation for bleeding	3.0% (27)	0.0% (0)	3.0% (1.9%, 4.1%)	0.42
Other procedures, %
Permanent pacemaker	1.1% (10)	0.7% (7)	0.4% (−0.5%, 1.2%)	0.38
ICD implantation	0.4% (4)	0.1% (1)	0.3% (−0.1%, 0.8%)	0.55
Initial hospitalization costs, $
Total procedural costs	9,776 ± 2,558 [9,447]	12,998 ± 5,165 [11,957]	−3,222 (−3,584, −2,859)	<0.001
Hospital stay + ancillary services	19,521 ± 7,912 [17,058]	9,880 ± 5,718 [6,227]	9,641 (9,019, 10,268)	<0.001
Physician fees	5,170 ± 768 [5,060]	2,967 ± 1,060 [2,571]	2,203 (2,114, 2,292)	<0.001
Total	34,467 ± 8,934 [31,910]	25,845 ± 9,861 [22,479]	8,622 (7,710, 9,483)	<0.001

^*Difference between CABG and PCI groups

Values in brackets represent medians

Follow-up Resource Utilization and Costs

Follow-up clinical outcomes, resource utilization and costs are summarized in Table 4. During the first year of follow-up, rates of repeat revascularization by either PCI or CABG were higher among patients assigned to initial PCI. Although costs for cardiovascular hospitalizations were thus greater for the PCI group, these higher costs were offset by higher costs for non-cardiovascular hospitalizations, outpatient visits, and rehabilitation services for the CABG group such that total 1-year follow-up costs were similar for the 2 groups. During each subsequent year of follow-up, patients assigned to PCI continued to experience higher rates of cardiovascular-related hospitalizations, driven mainly by differences in rates of MI and revascularization procedures. Costs for outpatient medications after PCI remained higher than those following CABG for the first 3 years (related mainly to higher rates of use of dual antiplatelet therapy) and then equalized. Annual medical care costs were thus substantially lower for the CABG group for each subsequent year with annual differences ranging from $793 in year 2 to $1,594 in year 5. As a result, the difference in cumulative medical care costs between the CABG and PCI groups narrowed progressively from $8,622/patient after the completion of the index revascularization procedures to $3,641/patient by the end of the 5-year follow-up period (Table 5 and Figure 2).

Table 4.

Follow-up Events, Resource Utilization, and Costs (mITT population)

	Year 1		Year 2		Year 3		Year 4		Year 5		5-Year Cumulative*
	CABG N=911	PCI N=944	CABG N=897	PCI N=845	CABG N=824	PCI N=784	CABG N=628	PCI N=659	CABG N=428	PCI N=419	CABG N=911	PCI N=944	p-value*
Clinical Outcomes
Death, %	4.1	3.4	2.1	3.0	1.3	2.4	1.8	2.3	1.0	4.0	10.7	16.0	0.06
MI, %	3.7	5.8	0.8	1.0	0.4	1.9	0.3	2.1	0.5	3.0	6.0	13.9	<0.001
Stroke, %	1.9	0.8	0.7	0.7	0.8	0.5	0.8	0.3	0.7	0.2	5.1	2.4	0.04
Resource Utilization (events/100 pts)
Repeat Revascularization
PCI procedures	4.6	8.4	3.8	6.0	2.3	5.3	0.8	4.6	2.4	5.1	3.3	6.8	<0. 001
CABG procedures	0.0	2.4	0.0	1.4	0.0	1.1	0.0	1.2	0.0	1.4	0.0	1.7	0.99
Total procedures	4.6	10.8	3.8	7.5	2.3	6.4	0.8	5.8	2.4	6.5	3.3	8.5	<0.001
Diagnostic catheterization	1.6	2.9	2.4	2.8	1.4	3.0	1.3	2.3	1.0	1.9
Re-hospitalization
Cardiovascular	18.8	22.2	10.3	14.6	6.0	12.5	3.3	14.1	5.3	10.7	10.8	17.2	<0.001
Non-cardiovascular	19.1	13.9	12.0	11.6	11.1	11.5	11.6	9.6	7.9	9.3	14.6	12.8	0.052
Costs per patient, $
Cardiovascular hospitalizations	2377	2991	1208	1749	581	1587	211	1782	461	1598	---	---	---
Non-cardiovascular hospitalizations	2380	1567	1356	1224	1050	1413	1437	1022	697	1104	---	---	---
Outpatient services	1200	1041	1061	831	954	848	856	885	839	754	---	---	---
Rehab/skilled nursing stays	1606	1229	339	486	714	240	529	454	264	787	---	---	---
Medications	1928	2275	1988	2323	1846	2172	2018	2053	1660	1719	---	---	---
Total	9491	9102	5952	6613	5145	6259	5051	6196	3921	5963	---	---	---

^*Kaplan-Meier rates for clinical outcomes with p-values from logrank test; rates per person-year for resource utilization, with p-values obtained from Poisson regression.

Table 5.

Cumulative Costs, QALYs and Life Years for Years 1–5, after adjusting for censoring

Time Since Randomization (Years)	Cumulative Costs ($)*			Cumulative QALYs*			Cumulative Life Years
	CABG	PCI	Δ	CABG	PCI	Δ	CABG	PCI	Δ
1	41,855	33,976	7878	0.8163	0.8497	−0.0333	0.9713	0.9796	−0.0083
2	47,111	40,025	7086	1.6197	1.6542	−0.0345	1.9209	1.9307	−0.0098
3	51,848	45,596	6251	2.3729	2.4015	−0.0286	2.8512	2.8518	−0.0006
4	56,551	51,316	5235	3.0773	3.0815	−0.0043	3.7676	3.7528	0.0148
5	60,501	56,860	3641	3.7191	3.6878	0.0312	4.6659	4.6130	0.0528

^*cumulative costs, QALYs, and life-years adjusted for censoring (see Methods for details)

Δ = difference between CABG and PCI groups

Figure 2.

Mean cumulative medical costs (lines) and mean annual follow-up costs (bars) in 2010 dollars, for the PCI and CABG groups.

Utility Weights and QALYs

Utility weights as assessed by the EQ-5D are summarized in Table 6. Overall, utility weights improved substantially for both treatment groups over the course of the trial. Not surprisingly, utility weights at 1 month follow-up were substantially lower after CABG than PCI (0.81 vs. 0.89, p<0.001), reflecting the more invasive nature and prolonged recovery of the former procedure. These differences were no longer apparent at 6 months, however, and utility weights remained similar for the 2 groups through the remainder of the 5-year follow-up period. As a result of these early differences, quality-adjusted life expectancy was lower with CABG than with PCI after the first year and remained lower through the first 4 years of follow-up (Table 5). However, owing to progressive differences in all-cause mortality, by the end of the 5-year follow-up period, life expectancy (4.665 vs. 4.613 years) and quality-adjusted life expectancy (3.719 vs. 3.688 QALYs) were actually greater with CABG than with PCI, although these differences were not statistically significant.

Table 6.

EQ-5D Utility Scores by Treatment (mITT population)

Time Point	CABG	PCI
Baseline	0.779 ± 0.185 [0.810]	0.782 ± 0.188 [0.810]
1 month	0.810 ± 0.180 [0.827]	0.894 ± 0.143 [1.000]
6 months	0.810 ± 0.180 [0.827]	0.895 ± 0.143 [1.000]
1 years	0.810 ± 0.180 [0.827]	0.881 ± 0.160 [1.000]
2 years	0.889 ± 0.164 [1.000]	0.895 ± 0.143 [1.000]
3 years	0.895 ± 0.162 [1.000]	0.891 ± 0.161 [1.000]
4 years	0.894 ± 0.148 [1.000]	0.868 ± 0.177 [0.854]
5 years	0.876 ± 0.175 [1.000]	0.884 ± 0.132 [0.854]

Values in brackets represent medians

Cost-Effectiveness

Results from the lifetime cost-effectiveness analyses are summarized in Table 7. Based on the landmark analysis of outcomes from year 1 to 5, the estimated mortality hazard ratio for CABG vs. PCI was 0.60 (95% CI, 0.42 to 0.86); CABG was associated with a reduction in follow-up costs of $1,672/year (95% CI, $942 to $2,403); and there was no significant difference in follow-up utility weights (see Supplementary Appendix, Tables 3 and 4, for details). When these results were used to project clinical and economic outcomes beyond the trial period (Figure 3), we estimated that CABG would be associated with lifetime incremental costs of $5,392 (95% CI, $399 to $10,320) together with an increase in overall quality-adjusted life expectancy of 0.663 QALYs (95% CI, 0.177 to 1.132). The resulting ICER for CABG vs. PCI was $8,132/QALY gained, with 99.2% of bootstrap replicates below a societal willingness to pay threshold of $50,000/QALY (Figures 4 and 5, and Table 7, row 1). When outcomes were assessed in life years, CABG was associated with a gain in life expectancy of 0.794 years and an associated ICER of $6,791/life year gained (Figure 5 and Table 7, row 2).

Table 7.

Lifetime Cost-Effectiveness Results for Base Case and Sensitivity Analyses

		Cost with CABG, $	Cost with PCI, $	Δ Cost ($) (95% CI)	QALYs with CABG	QALYs with PCI	ΔQALYs with CABG (95%CI)	ICER ($/QALY)	% Dominant	% Dominated	%< $50K
Tapered CABG Effect Between 5 and 10 Years	Base Case	114,571	109,179	5,392 (399, 10,320)	10.667	10.004	0.663 (0.177, 1.132)	8,132	1.5	0.4	99.2
	Lifetime analysis, Cost per Life Year Gained*	114,571	109,179	5,392 (399, 10,320)	12.177*	11.383*	0.794*	6,791*	1.5	0.2	99.7
	Lifetime analysis, Per Protocol population	114,855	108,878	5,976 (1,207, 10,925)	10.688	9.980	0.708 (0.202, 1.221)	8,440	0.7	0.2	99.3
	10 year time frame	81,710	80,295	1,416 (−2,061, 5,017)	6.684	6.455	0.230 (−0.008, 0.457)	6,156	22.8	1.7	95.8
	Lifetime analysis, no CABG effect on long term costs	121,244	109,179	12,045 (6,933, 17,103)	10.667	10.004	0.663 (0.177, 1.132)	18,167	0.0	0.5	97.4
Fixed CABG Effect Between 5 and 10 Years	Lifetime analysis	116,147	109,179	6,968 (1,273, 12,327)	10.867	10.004	0.864 (0.278, 1.426)	8,064	0.7	0.2	99.7
	10 year analysis	81,846	80,295	1,551 (−1,822, 5,292)	6.704	6.455	0.249 (0.016, 0.519)	6,229	18.8	1.4	96.7
No Effect of CABG after 5 Years	Lifetime analysis	118,664	109,179	9,485 (4,905, 13,995)	10.355	10.004	0.351 (−0.033, 0.713)	27,022	0.0	3.3	82.4
	10 year analysis	87,155	80,295	6,861 (3,408, 10,230)	6.541	6.455	0.086 (−0.088, 0.254)	79,779	0.0	17.8	27.0

^*Results presented in row 2 represent life years (columns 6–8) and cost per life year gained (column 9)

^**All costs, life-years, and QALYs discounted at 3% per year

Figure 3.

Observed survival through 5 years and predicted survival beyond 5 years for the CABG (solid) and PCI (dashed) groups.

Figure 4.

Joint distribution of projected lifetime incremental costs and quality-adjusted life expectancy for CABG vs. PCI based on bootstrap replication of the FREEDOM trial population-- plotted on the cost-effectiveness plane. The white circle represents the estimated mean values (incremental cost = $5392, incremental QALYs=0.663)

Figure 5.

Cost effectiveness acceptability curve of CABG vs. PCI, in $/QALY gained (black, solid line) and $/Life year gained (blue, dashed line). The probability that CABG is cost-effective is calculated as the proportion of bootstrap-derived estimates falling below a given cost-effectiveness threshold, is plotted across a range of possible cost-effectiveness thresholds.

These results were robust across a wide range of alternative assumptions regarding the duration and magnitude of the benefit of CABG over PCI on both survival and costs beyond the timeframe observed in the trial. When we assumed that the observed benefits of CABG would be sustained through 10 years after initial treatment, the estimated QALY gains with CABG increased to 0.864 QALYs with an associated ICER of $8,064/QALY gained-- similar to the base case results. Even under the highly conservative assumption of no further benefit of CABG on either survival or costs beyond 5 years, the ICER remained relatively favorable at $27,022/QALY gained and was <$50,000/QALY in 82.4% of bootstrap replicates. Similar results were seen when the analysis was based on life years rather than QALYs (see Supplemental Tables 5 and 6)

Subgroup Analyses

Results from prespecified subgroup analyses are presented in Table 8. Despite the greater uncertainty associated with reduced sample sizes, these results were generally consistent with our primary analysis. For all subgroups examined, with the exception of patients without LAD involvement, our analysis demonstrated that CABG was associated with greater quality adjusted life expectancy than PCI with mean differences ranging from 0.28 to 1.16 QALYs. For two subgroups, age 60–69 and HgbA_1c <7%, CABG was an economically dominant strategy, with lower lifetime costs and greater quality-adjusted life expectancy than PCI. For patients without LAD involvement (n=150), PCI was associated with greater overall costs and quality-adjusted life expectancy with an ICER of $20,661/QALY gained relative to CABG. However, these results were relatively unstable owing to the small sample size of the subgroup. Estimated ICERs for all other subgroups were <$25,000/QALY gained with CABG, and most were <$10,000/QALY gained.

Table 8.

Lifetime Cost-Effectiveness Results for Subgroups

Subgroup	Cost with CABG, $	Cost with PCI, $	ΔC-P Cost ($) (95% CI)	QALYs with CABG	QALYs with PCI	QALYs Gained with CABG (95% CI)	ICER: Cost ($) per QALY gained	% Dominant	% Dominated	% < $50K
Male (n=1,328)	107,527	104,468	3,059 (−2,304, 8,406)	10.856	10.078	0.778 (0.205, 1.358)	3,932	10.3	0.0	99.8
Female (n=527)	131,295	122,046	9,249 (−729, 18,900)	10.311	9.801	0.510 (−0.519, 1.427)	18,135	0.6	11.0	77.3
Age <60 (n=624)	136,342	125,152	11,190 (3,656, 18,572)	13.897	12.737	1.160 (0.526, 1.838)	9,647	0.3	0.0	99.8
Age 60–69 (n=621)	106,301	108,066	−1,765 (−9,533, 5,504)	9.947	9.671	0.276 (−0.486, 0.938)	CABG Dominant	45.7	1.1	80.5
Age ≥ 70 (n=610)	93,926	87,034	6,892 (−3,018, 15,804)	6.853	6.459	0.349 (−0.516, 1.159)	19,748	2.5	15.1	71.9
SYNTAX Score <23 (n=657)	113,201	104,417	8,784 (753, 16,272)	10.883	10.477	0.407 (−0.479, 1.180)	21,582	0.3	14.7	73.5
SYNTAX Score 23–32 (n=828)	115,602	111,441	4,160 (−3,670, 11,619)	10.728	9.731	0.997 (0.225, 1.699)	4,172	13.7	0.1	99.2
SYNTAX Score >32 (n=359)	114,220	113,247	973 (−10,177, 11,337)	10.177	9.863	0.315 (−0.843, 1.373)	3,088	22.9	7.3	72.4
LAD (n=1,695)	115,124	109,463	5,661 (505, 10,981)	10.713	9.991	0.722 (0.203, 1.239)	7,841	1.7	0.3	99.4
No LAD (n=150)	105,875	108,189	−2,314 (−52,357, 10,225)	10.126	10.239	−0.112 (−6.933, 1.313)	Not estimable	18.0	8.6	43.7
2 vessel disease (n=310)	103.264	92,313	10,950 (−2,486, 21,181)	11.029	10.312	0.718 (−1.343, 1.739)	15,251	0.6	11.2	78.8
3 vessel disease (n=1,534)	116,891	112,830	4,061 (−1,367, 9,483)	10.634	9.937	0.697 (0.141, 1.248)	5,826	6.6	0.3	99.2
HgbA1c < 7 (n=617)	103,579	103,852	−273 (−6,879, 7,142)	10.262	9.934	0.328 (−0.482, 1.132)	CABG Dominant	34.6	2.0	80.8
HgbA1c ≥ 7 (n=1,096)	120,399	111,893	8,507 (1,618, 15,266)	11.011	10.064	0.946 (0.297, 1.547)	8,993	0.5	0.4	99.5
US (n=351)	126,113	121,412	4,701 (−7,997, 17,253)	10.796	9.676	1.120 (0.084, 2.033)	4,197	21.8	0.3	98.1
Non-US (n=1,504)	111,978	106,356	5,622 (506, 10,653)	10.655	10.079	0.576 (0.006, 1.144)	9,760	1.0	2.0	96.5

^*All subgroup analyses based on mITT population using base case assumptions for the relative effect of CABG on survival and costs

DISCUSSION

This study is the first direct comparison of economic outcomes of DES-PCI vs. CABG among patients with diabetes and multivessel CAD. Our results reveal that although CABG was associated with an increase in initial costs of ~$9,000/patient, these up-front costs were partially offset by lower costs in subsequent years due principally to a lower rate of repeat revascularization procedures (and, to a lesser extent, less use of cardiac medications). Over the first 5 years of follow-up, CABG improved life expectancy by ~0.05 years and quality-adjusted life expectancy by ~0.03 QALYs while increasing total costs by ~$3,600. When the observed in-trial results were extrapolated over a lifetime horizon, CABG was associated with much larger gains in quality-adjusted life expectancy relative to PCI (0.66 QALYs in the base case) while projected lifetime costs remained ~$5,400/patient higher with CABG. Thus, under our base case assumptions regarding the duration and magnitude of benefit of CABG over DES-PCI, we found that CABG was associated with a lifetime ICER of $8,132/QALY gained. Although there are no universally accepted standards for cost-effectiveness in the U.S. health care system, ICERs <$50,000/QALY gained are commonly considered to be reasonably cost-effective, and ICERs <$20,000/QALY are considered highly cost-effective.²⁷ In our base case analysis, the probability that the ICER for CABG vs. DES-PCI would be <$50,000/QALY was >99%. These findings thus suggest that compared with DES-PCI, CABG is a highly attractive use of scarce societal healthcare resources.

These results were robust to a broad range of sensitivity analyses. Importantly, the ICER for CABG remained <$50,000/QALY gained (and in most cases <$10,000/QALY) in all analyses except those that were restricted to the first 10 years of follow-up. Although an “in-trial” cost-effectiveness analysis would have demonstrated an ICER for CABG of $68,958 per life-year gained and $116,699 per QALY gained, such an analysis would be misleading by failing to account for the significant 5-year survival advantage with CABG over DES-PCI (89.3 vs. 83.9%) and its impact on longer-term life expectancy. Indeed, our analysis demonstrates that the projected benefits of CABG on life expectancy and quality-adjusted life expectancy would be substantial even if there were no further prognostic benefit of CABG beyond the trial period. These findings underscore the importance of a long-term evaluation in understanding the clinical and economic benefits of procedures like CABG, where a higher initial cost must be balanced against clinical benefits that extend well into the future.

These results were also consistent across a wide range of subgroups defined by age, gender, angiographic extent of CAD (including SYNTAX score), and health care system (U.S. vs. other countries). The only possible exception to this finding was the subgroup of patients without significant LAD disease, for whom PCI was estimated to improve long-term survival at an ICER of ~$20,700/QALY gained relative to CABG. These findings were relatively uncertain, however, given the small sample size in this subgroup. Nonetheless, these results are broadly consistent with those of previous studies that have suggested that the prognostic benefit of CABG over PCI is restricted to patients who receive a LIMA graft to the LAD.²⁸

Comparison with Previous Studies

Overall, the cost and cost-effectiveness results from FREEDOM are similar to those reported from the BARI trial, in which CABG was estimated to have a lifetime incremental cost-effectiveness ratio of $14,300/life year gained compared with conventional balloon angioplasty.³ There are important differences in the results of these 2 studies, however. In BARI, most of the cost offsets for CABG accrued during the first 1–2 years of follow-up, due to relatively high rates of repeat revascularization procedures necessary to treat restenosis after conventional balloon angioplasty. In contrast, cost offsets were minimal during the first year in FREEDOM and accrued at a relatively constant rate throughout the 5-year follow-up period thereafter. Although detailed angiographic data were not assessed at the time of repeat revascularization in FREEDOM, it is likely that this event pattern reflects the fact that drug-eluting stents have led to dramatic reductions in restenosis compared with either balloon angioplasty alone or bare metal stents.^29,30 Nonetheless, diabetic patients remain at substantial risk of events related to progression of CAD^31,32—events that are unrelated to restenosis and are unlikely to be prevented by inherently focal therapies such as stenting.

Another difference between the FREEDOM and BARI results that impacts the cost-effectiveness of therapy is the timing of the mortality benefit. In BARI, improved survival with CABG was evident in the first year of follow-up, with increasing benefits over time. As a result, CABG was already an economically attractive revascularization strategy within the first several years after initial treatment. In contrast, in FREEDOM, mortality was similar after DES-PCI or CABG through the first 3 years of follow-up with divergence of the survival curves only in years 4 and 5. These findings may relate to (1) more complete and durable revascularization with DES-PCI in FREEDOM compared with “plain old” balloon angioplasty in BARI; (2) increased durability of left internal mammary artery grafts vs. saphenous vein grafts,³³ the former of which were used more often in FREEDOM; or (3) greater rate of use of preventive measures (including high dose statins, ACE-inhibitors or ARBs, and aggressive dual antiplatelet therapy) in FREEDOM which may have protected PCI patients from the progression of atherosclerosis and the development of atherothrombotic events in the earlier phases of follow-up.³⁴

Limitations

Our study should be considered in light of several important limitations. First, our economic analysis was carried out from a U.S. healthcare system perspective although FREEDOM enrolled patients from 18 countries. To address these issues, costs associated with the index procedures were estimated from detailed resource use data which would not be expected to vary significantly by geographic region. Since hospital length of stay varies considerably between countries, we used costing methodology that was independent of length of stay and depends only on the assumption that clinical outcomes and procedural complications are similar across healthcare systems. In addition, our study was limited by the need to extrapolate results from 5 years of follow-up data to a lifetime horizon in order to capture the full benefits of the alternative revascularization strategies. Our approach to extrapolation required several assumptions regarding the impact of CABG vs. PCI on long-term survival, health care costs, and quality of life that are not verifiable at the present time. Accordingly, we varied our assumptions of the magnitude and durability of the impact of CABG in a range of sensitivity analyses, the results of which demonstrated the robustness of the overall cost-effectiveness results to plausible variation in these key parameters.

Finally, most PCI patients in the FREEDOM trial received first generation DES, which were the only DES available for use at the time. Recently, second generation DES (primarily everolimus-eluting stents) have been shown to reduce rates of MI and stent thrombosis relative to first generation DES.³⁵ As a result, the cost results from our study may not generalize to a setting characterized by exclusive use of second generation DES for PCI. No studies, however, have demonstrated a mortality reduction with second vs. first generation DES,³⁶ and it is the mortality reduction obtained from CABG that primarily drives the favorable cost-effectiveness results in our study.

Conclusions

In FREEDOM, the largest randomized trial to compare CABG vs. DES-PCI for the treatment of diabetic patients with multivessel CAD, we found that CABG provides not only better long-term clinical outcomes than DES-PCI but that these benefits are achieved at an overall cost that represents an attractive use of societal health care resources. These findings suggest that existing guidelines that recommend CABG for diabetic patients with multivessel CAD^37,38 remain appropriate in current practice and may provide additional support for strengthening those recommendations.