Long-term Outcomes and Costs of Ventricular Assist Devices Among Medicare Beneficiaries FREE

Mechanical circulatory support with a ventricular assist device is an evolving technology for patients with end-stage heart failure. Ventricular assist devices may be considered in several clinical scenarios but are used principally in patients with end-stage heart failure who are awaiting heart transplantation, as “destination,” or permanent, therapy for patients who are not candidates for transplantation, or as a rescue procedure for patients with refractory shock after open-heart surgery.^1- 3 Clinical scenarios for ventricular assist devices often involve patients with rapidly deteriorating conditions for which an assist device is considered emergent because of imminent death or urgent because of end-organ dysfunction. The Randomized Evaluation of Mechanical Assistance in the Treatment of Congestive Heart Failure (REMATCH) demonstrated that destination therapy with an assist device could reduce mortality by 48% among patients with end-stage heart failure and improve quality of life.⁴ However, overall survival in REMATCH was poor among patients who received an assist device, with 52% surviving after 1 year and 23% surviving after 2 years.

In 2003, the Centers for Medicare & Medicaid Services responded to the REMATCH findings by increasing Medicare coverage of ventricular assist devices to include destination therapy for patients meeting certain end-stage eligibility requirements.⁵ However, questions remain regarding long-term outcomes of device placement, especially given expanded indications for destination therapy and use in elderly patients. Therefore, we analyzed trends in the use of ventricular assist devices alone or after open-heart surgery for all Medicare fee-for-service beneficiaries from February 2000 through June 2006. We followed up patients through the end of 2006 to examine acute and long-term outcomes, including death, rehospitalization, and inpatient costs to Medicare.

We obtained research-identifiable inpatient claims and corresponding denominator files from the Centers for Medicare & Medicaid Services for all Medicare beneficiaries discharged between 2000 and 2006. The inpatient files contain institutional claims for services covered under Medicare Part A. The denominator files contain beneficiary demographic data and information about program eligibility and enrollment. We included all persons living in the United States who were continuously enrolled in fee-for-service Medicare for at least 30 days before placement of a ventricular assist device, including patients younger than 65 years who were enrolled in Medicare because of disability. We restricted the analysis to claims filed during periods of fee-for-service coverage.

The analysis included patients for whom implantation of a ventricular assist device (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] codes 37.62, 37.65, and 37.66) was reported on a single inpatient claim. If a patient had more than 1 such procedure during the study period, we retained the first as the index procedure. Because an inpatient stay may consist of multiple claims and involve transfers between facilities, we examined multiple claims to generate a single episode of care. We considered contiguous inpatient claims 1 day apart or less to be part of a single continuous episode. We included patients whose episode of care began on or after February 1, 2000, and who received a device on or before June 30, 2006.

We divided patients into 2 groups for the analysis based on the clinical scenarios typically encountered: (1) patients for whom the device was the first or primary therapy and (2) patients who received the device after cardiac surgery. Patients in the primary device group had received a ventricular assist device without having undergone cardiotomy in the 30 days before implantation. Patients in the postcardiotomy group had received an assist device for postcardiotomy circulatory support. Inclusion in the postcardiotomy group required a claim for coronary artery bypass graft surgery (ICD-9-CM code 36.1x), open-heart valvuloplasty without replacement (codes 35.11, 35.12, and 35.14), open-heart valve replacement (codes 35.21-35.28), or cardiotomy (code 37.11) during the episode of care or up to 30 days before device implantation.

We followed up all patients through December 31, 2006. We searched claims after device implantation for evidence of heart transplantation (ICD-9-CM codes 33.6 and 37.5), implantation or repair of an assist device (codes 37.62, 37.63, 37.65, and 37.66), removal of an assist device (code 37.64), and all-cause readmission. We conducted 2 sensitivity analyses. First, to understand whether outcomes for the primary device group differed based on the primary diagnosis, we restricted the analysis to patients with a primary diagnosis of acute myocardial infarction (ICD-9-CM code 410.xx). Second, because ICD-9-CM procedure codes were redefined in 2005 to distinguish between implantable (code 37.66) and nonimplantable (codes 37.62 and 37.65) assist devices, we compared outcomes for implantable and nonimplantable devices placed in 2005 and 2006.

For patients discharged alive, we calculated the average number of days spent in the hospital during the subsequent 2 years. For all device recipients, we determined 1-year direct inpatient costs to Medicare for the 2000 through 2005 cohorts by summing Medicare reimbursement amounts for inpatient care from the beginning of the episode of care until death or censoring. We calculated costs for each day patients were alive and out of the hospital by dividing the observed total payments for all patients receiving a device by the number of days on which those patients were alive and out of the hospital for the first year and the first 2 years after implantation.

For baseline patient and hospital characteristics, we present categorical variables as frequencies with percentages and continuous variables as either means with SDs or medians with interquartile ranges. To test for differences between the primary device group and the postcardiotomy group, we used χ² tests for categorical variables and Wilcoxon rank sum tests for continuous variables. To examine trends in characteristics or volume of device implantation, we used Cochran-Mantel-Haenszel correlation statistics.

We identified comorbid conditions using coding algorithms described by Birman-Deych et al⁶ and Quan et al.⁷ Specifically, we searched all inpatient claims for the 30 days preceding the index procedure through the end of the episode of care for evidence of myocardial infarction, congestive heart failure, stroke, valvular heart disease, diabetes mellitus, coronary heart disease, hypertension, cerebrovascular disease, chronic obstructive pulmonary disease, peripheral vascular disease, or renal disease. Race/ethnicity is reported by Medicare beneficiaries at the time of enrollment. We used the reported category “black” and combined all others and missing values as “nonblack.”⁸ Procedure volume per hospital reflects counts of all assist device implantations (ICD-9-CM codes 37.62, 37.65, and 37.66) at a given facility during the index year.

We determined the proportion of patients who underwent heart transplantation, had a device removed, died with a device, or were alive with a device at discharge. Using the cumulative incidence function,⁹ we estimated the proportions of these competing events at 1 year after implantation. If a device removal was performed in conjunction with heart transplantation, we counted the transplantation only. Each of these end points was a terminal state for the purposes of the analysis, so the proportions sum to 100% for each year and cohort. We also used the cumulative incidence function to estimate rates of device replacement or repair among all patients and rates of readmission among patients discharged alive with a device. We considered mortality, device removal, and heart transplantation to be competing risks for these other events. We counted readmissions for heart transplantation as transplantations, not readmissions. In all cases, censoring occurred when patients were no longer eligible for fee-for-service Medicare or when they reached the end of the study period. We used Kaplan-Meier methods to calculate overall survival, regardless of subsequent heart transplantation or device removal.

We used Cox proportional hazards models to examine unadjusted and adjusted relationships between study variables and survival. The multivariable models included indicators for age, race, sex, comorbid conditions and risks, hospital procedure volume, geographic region, year of the index procedure, teaching hospital status, and heart transplant hospital status. To account for the competing risks of heart transplantation and device removal, we considered patients to be at risk as long as they were uncensored and remained alive with a device. We used robust standard errors to account for clustering at the hospital level.¹⁰

We used SAS software, version 9.1.3 (SAS Institute Inc, Cary, North Carolina) for all analyses. The analyses were 2-tailed and we considered P < .05 to be statistically significant. The institutional review board of the Duke University Health System approved the study.

From February 2000 through June 2006, 2943 patients received a ventricular assist device at 570 hospitals. Of these, 1476 patients (50.2%) at 335 hospitals had no evidence of prior cardiotomy (ie, the primary device group) and 1467 patients (49.8%) at 448 hospitals had undergone cardiac surgery in the 30 days before implantation (ie, the postcardiotomy group). The median annual volume of device implantations per hospital was 1 (interquartile range, 1-2). The number of hospitals with 5 or more implantations per year ranged from 19 in 2000 to 28 in 2005. Median follow-up was 294 days in the primary device group (interquartile range, 41-1056) and 18 days in the postcardiotomy group (interquartile range, 5-582), reflecting the high rate of early mortality.

There were small differences in age, sex, and race between the cohorts (Table 1). Both groups had substantial comorbidity, including previous heart disease, heart failure, diabetes mellitus, chronic obstructive pulmonary disease, and renal disease. Among patients in the postcardiotomy group, 887 (60.5%) had undergone isolated coronary artery bypass graft surgery, 193 (13.2%) had undergone both bypass surgery and valve surgery, 148 (10.1%) had undergone valve replacement surgery only, 75 (5.1%) had undergone bypass surgery and some other cardiac procedure, and 164 (11.2%) had undergone some other combination of open-heart procedures. Most of the postcardiotomy patients (n = 1358 [93%]) had an assist device implanted within 1 day of the original cardiac surgery. Compared with patients in the primary device group, fewer postcardiotomy patients were treated at a teaching hospital or heart transplant hospital (P < .001 for both hospital types).

From 2000 through 2006, mean age declined in both groups (from 66 to 60 years in the primary device group [P < .001] and from 70 to 68 years in the postcardiotomy group [P < .001]). The percentage of black patients increased from 8.7% in 2000 to 18.2% in 2005 (P = .02) in the primary device group and from 3.1% to 12.5% (P = .05) in the postcardiotomy group. The number of patients in the primary device group increased from 184 in 2000 to 263 in 2005. Procedure volume for postcardiotomy patients did not significantly change from 2000 (n = 196) through 2005 (n = 229).

During the initial inpatient stay, 140 patients (9.5%) in the primary device group underwent heart transplantation, 66 (4.5%) had a device removed, and 815 (55.2%) were discharged alive with a device (Table 2). Median length of stay was 30 days (interquartile range, 10-62 days). In the postcardiotomy group, 21 patients (1.4%) underwent heart transplantation, 129 (8.8%) had a device removed, and 493 (33.6%) were discharged alive with a device. Median length of stay was 10 days (interquartile range, 5-25 days).

At 1 year, 299 patients (20.7%) in the primary device group had undergone heart transplantation, 72 (4.9%) had a device removed, and 417 (32.2%) were alive with a device. Seventy-five patients (5.2%) had a device replaced. In the postcardiotomy group, 51 (3.5%) had undergone heart transplantation, 132 (9.0%) had a device removed, and 333 (24.1%) were alive with a device. Overall 1-year survival, regardless of subsequent heart transplantation or device removal, was 51.6% (n = 669) in the primary device group and 30.8% (n = 424) in the postcardiotomy group (Figure, A).

Of the 815 patients in the primary device group who were discharged alive with a device, 450 (55.6%) were readmitted within 6 months. On average, these patients spent 29.8 (SD, 45.0) days in the hospital during the subsequent 2 years; survival was 64.7% (n = 354) at 2 years (Figure, B). Among patients in the postcardiotomy group who were discharged alive with a device (n = 493), 237 (48.3%) were readmitted within 6 months. These patients spent an average of 16.7 (SD, 31.6) days in the hospital during the subsequent 2 years; survival was 69.4% (n = 283) at 2 years (Figure, B).

The results of the sensitivity analyses were qualitatively similar. Among 342 patients in the primary device cohort with a primary diagnosis of acute myocardial infarction, 176 (51.5%) were discharged alive with a device. One-year survival was 45.8% (n = 148). Moreover, among patients in both cohorts who received implantable devices, 1-year survival was 44.4% (127/306) compared with 26.1% (47/186) among patients who received nonimplantable devices (P < .001). Six-month survival was 50.6% (76/163) among patients who received an implantable device during the first half of 2006 and 29.1% (23/79) among patients with nonimplantable devices. In comparison, 6-month survival was 46.5% (72/162) among patients with implantable devices during the first half of 2005 and 33.0% (32/97) among those with nonimplantable devices.

In the primary device group, age, history of congestive heart failure, peripheral vascular disease or renal disease, and geographic region were independently associated with a greater hazard of mortality (Table 3). History of hypertension was associated with a lower hazard of mortality. In the postcardiotomy group, device implantation volume at the hospital level was significantly associated with a lower hazard of mortality. Annual volume of 5 or more assist devices was associated with a 31% lower hazard. Other factors associated with a greater hazard of mortality in the postcardiotomy group were peripheral vascular disease, valvular heart disease, and geographic region. Hypertension, stroke/transient ischemic attack, and male sex were associated with a lower hazard of mortality.

For patients in the 2000 through 2005 cohorts, the mean payment to hospitals by Medicare in the first year after implantation of a ventricular assist device was $144 298 (SD, $125 173). The mean payment was $178 714 (SD, $142 549) in the primary device group and $111 769 (SD, $95 413) in the postcardiotomy group. One-year Medicare payments for inpatient care of primary device patients totaled $228 039 342. Collectively, primary device patients were alive and out of the hospital for 221 892 days during the 365-day period beginning with the date of the device admission, resulting in a Medicare payment of $1028 per day for the first year.

Similarly, 1-year Medicare payments for inpatient care of postcardiotomy patients were $150 887 516. These patients were alive and out of the hospital 149 100 days following the device admission date, resulting in a Medicare payment of $1012 per day for the first year. For patients in the 2000 through 2004 device cohorts, Medicare payment per day alive outside the hospital during the 2-year period beginning with the episode admission date was $482 for primary device patients and $444 for postcardiotomy patients.

In this large retrospective study examining the use of ventricular assist devices in the Medicare population, we found that patients receiving an assist device had prolonged index hospitalizations and high perioperative mortality, a high likelihood of readmission within 6 months, and a resulting high cost of inpatient care. The analysis included all Medicare fee-for-service beneficiaries, so it provides a complete picture of the use of ventricular assist devices beyond clinical trial settings alone. Our findings highlight the importance of the perioperative clinical course for long-term mortality and morbidity.

Mechanical ventricular assistance is a challenging technology to assess, in part because patients with heart failure are already at high risk of mortality and readmission. Among Medicare beneficiaries with heart failure, mortality is 27% and the readmission rate is 65% at 1 year.¹¹ Patients with heart failure, especially those with end-stage heart failure, incur high medical costs.^12- 13 Mortality among patients with severe heart failure who are inotrope-dependent is more than 75% at 1 year,¹⁴ and mortality approaches 100% among those who undergo cardiac surgery and cannot be weaned from cardiopulmonary bypass. Moreover, among all patients receiving medical therapy in trials of assist devices as destination therapy, survival is less than 25% at 1 year.^4,15

Although REMATCH and other studies found that ventricular assist devices improve survival compared with medical therapy,^4,15 uncertainty remains regarding their routine use. In 2004, a National Heart, Lung, and Blood Institute working group established a goal of 50% survival at 2 years and minimal time in the hospital after device implantation as a prerequisite for widespread adoption of assist devices as destination therapy.¹⁶ Although experience with the devices has improved with device evolution,^17- 18 overall outcomes in the Medicare population have not met this goal. Continued postmarketing surveillance will be essential as the use of assist devices expands to populations outside clinical trials, which tend to include patients who are older and sicker than patients in research settings.^18- 20

The Interagency Registry for Mechanical Assisted Circulatory Support (INTERMACS) is now reporting outcomes on patients who have US Food and Drug Administration–approved durable devices and who consent to participate in the protocol. The initial results reported from this registry are notably different from the results of our study, including 6-month survival of 74%.²¹ Reasons for these differences are severalfold. First, our study included only Medicare beneficiaries, who are likely older and have a higher burden of comorbid illness than the overall INTERMACS population. Second, our study may have a higher proportion of patients receiving ventricular assist devices as destination or emergent (rescue) therapy and who are potentially ineligible for heart transplantation and, thus, have a higher risk of mortality. Moreover, many destination therapy patients are included in clinical trials covered by Medicare through evidence development policies but are not included in INTERMACS. Third, enrollment in INTERMACS requires informed consent. Patients and their family members may be less willing to participate in a registry in emergent situations. To the extent that patients in emergent situations have a higher risk of mortality and are less willing to participate, survival among participants in INTERMACS may be higher.

Another goal in the use of ventricular assist devices is improved quality of life, which in part assumes reduced inpatient hospital days. Previous studies have shown improved quality of life for patients with ventricular assist devices, but the results were limited to small samples due to overall mortality rates.¹⁴ Thus, number of days in the hospital has become a surrogate measure for quality of life. For example, inotrope-dependent patients in REMATCH who were randomly assigned to receive a left ventricular assist device had a median of 255 days out of the hospital compared with 105 days for patients assigned to medical therapy. Although a substantial number of patients who were discharged alive in the present study were readmitted within the first year, the average number of inpatient days was less than 30 during the following 2 years. Initial length of stay increased over time from a median of 18 days to 30 days or more. This observation warrants further attention and may become a basis for the development of new processes of care to reduce length of stay and costs. Number of days in the hospital could also be considered a global ranking end point in future studies because it captures key elements of the clinical course involving this unique patient population.²²

We also examined inpatient costs associated with ventricular assist devices. Mean Medicare payments for inpatient costs at 1 year after implantation were $178 714 in the primary device group and $111 769 in the postcardiotomy group. The costs are comparable with or lower than costs associated with liver transplantation.²³ These estimates do not include costs for physician services, skilled nursing, outpatient visits, medications, home health care, physical and occupational therapy, and durable medical equipment, but they do include costs for subsequent hospitalizations, including hospitalizations for heart transplantation and device removal. Although the cost of ventricular assist devices in the Medicare population will be relatively high, Medicare administrators have suggested that payments for innovative technology may spur investment, particularly in areas with otherwise limited therapeutic alternatives or significant market barriers.²⁴

Although other studies have found improved outcomes over time in postcardiotomy patients with ventricular assist devices, we did not observe this trend among Medicare beneficiaries.² For unknown reasons, 1-year survival from 2000 through 2006 for these patients declined significantly from 41% to 22%. A potential explanation is that patients' risk profiles may have shifted, so that patients who previously would not have been considered for an assist device are now undergoing salvage therapy.²⁵ Also, the increasing availability of ventricular assist devices for salvage therapy in patients undergoing complicated cardiac surgery may allow opportunities for use of the devices that were previously unavailable. Previous work focused on operative mortality from 1995 through 2004; the study populations were younger and other patient characteristics differed from this study. Future studies should examine what factors are contributing to the trends in outcomes that we observed.

Improving outcomes will require a focus on the high perioperative mortality found in this and other studies.^4,15 Identifying patients who are likely to benefit from ventricular assist devices and excluding those whose likelihood of survival is low is warranted.^2,26 Because we found a volume-outcome relationship between device implantation and mortality in the postcardiotomy group, development of systems that foster the transfer of these patients to experienced centers may improve outcomes.^27- 28 In addition, there appears to be a suggestion of improved outcomes in high-volume centers in the primary device cohort. Other studies have suggested that complex procedures should be concentrated at centers that are likely to translate early experience into acceptable outcomes.^28- 29 Moreover, our observation of 65% survival at 2 years among patients discharged alive warrants further clinical investigation to identify factors associated with survival. These findings support a 2008 National Heart, Lung, and Blood Institute working group recommendation to conduct a randomized trial of ventricular assist devices in patients who have not developed serious complications of heart failure.³⁰

Our study has several limitations. Medicare claims do not include clinical details about the type of device and site of implantation, so we had limited ability to determine the indication or identify important clinical variables associated with mortality. However, in a sensitivity analysis using data from 2005 and 2006, outcomes of patients receiving implantable vs nonimplantable devices were similar to outcomes for patients in the primary device and postcardiotomy groups. Most devices implanted during the study period were pulsatile devices, and we were unable to distinguish patients with continuous-flow devices who were enrolled in a clinical trial. In addition, the analysis included only patients enrolled in fee-for-service Medicare, so other patients are not represented in the study. Although the primary aim of the study was to examine outcomes of Medicare beneficiaries, outcomes may differ because of patients' age or comorbidity compared with other patient populations. However, selection bias is likely to be minimal because the study included all Medicare fee-for-service beneficiaries and all hospitals that implanted a ventricular assist device in a Medicare beneficiary. Although quality of life is an important consideration for ventricular assist devices in patients with refractory heart failure, Medicare claims do not enable a direct assessment of this domain. Studies focused on quality of life will be needed.

Ventricular assist devices are an evolving technology with modest adoption in the Medicare population. Mortality, morbidity, and costs remain high, so periodic surveillance using Medicare claims may complement other postmarketing surveillance efforts. Improving patient selection and reducing perioperative mortality will be critical factors for improving overall outcomes.