The Economic Implications of Broader Sharing of Liver Allografts

Introduction

As liver transplantation enters its fourth decade as a widespread clinical treatment for end-stage organ failure, the liver transplant community is facing unprecedented challenges both clinically and economically (1). Broader application of this life-saving technology has led to waiting lists that increasingly outpace organ supplies, resulting in higher average severity of illness at the time of transplantation. Adoption of the severity of illness-based allocation system in 2002 using the Model of End Stage Liver Disease score (MELD) has further increased the acuity of patients reaching transplantation nationally (2). Finally, dramatic differences in the severity of illness at transplant and corresponding waitlist mortality exist across allocation regions due to geographic disparities in supply–demand ratios (3,4).

Simultaneously, there is a broad recognition that the current rate of increase in the cost of health care is unsustainable (5,6). Liver transplantation offers a life-saving treatment to many patients certain to die without transplant, but also ranks among the highest cost Medicare-approved procedures (7,8). Restrictions on insurance premiums, lower Medicare spending and an increasing number of patients with Medicaid only insurance are likely to further increase the financial risk of liver transplantation. Given the geographic variation in access to liver transplant, areas with a relative donor shortage are forced to transplant patients with higher MELD scores using more marginal organs. Consequently, these transplant centers face disproportionate financial and clinical risk (9,10). Broader organ-sharing proposals have been considered by the United Network for Organ Sharing (UNOS) Liver and Intestine Transplant Committee. These proposals are designed to increase the clinical benefit of available liver allografts by raising the MELD thresholds for local use and thereby increasingly sharing within a geographic region. While simulation models have demonstrated that these systems are likely to save lives, the economic implications of alternative sharing schemes are not defined.

Prior assessment of the relationship between MELD and cost has been limited by a lack of nationally representative cost data (1,9,11–14). Using a novel database that links clinical data from the Organ Procurement and Transplantation Network (OPTN) with cost accounting data from a national sample of liver transplant centers, we developed a detailed analysis of the relationship between severity of illness and cost. We use resulting estimates to assess the economic implications of novel allocation proposals, and to determine the incremental costs incurred through the shift in organs from less expensive low-MELD to more costly high-MELD patients, the expected benefit in terms of life-years gained, and the corresponding cost per year of life saved if these proposals were adopted given current simulation modeling results.

Methods

Organ allocation reform analysis

Data prepared for the OPTN Liver and Intestine Transplant Oversight committee by the Scientific Registry of Transplant Research (SRTR) was used to evaluate the cost-effectiveness of alternative organ sharing proposals. On March 4, 2010, the SRTR presented a series of analysis of the clinical impact of revising the current liver allocation system to provide a two-tiered system of liver allocation (Figure 1) (17). As summarized in Table 3, five alternative allocation systems using OPTN 2006 data were considered in which there were varying MELD thresholds for local and regional utilization of available allografts. The results of these proposed allocation systems were analyzed using the Liver Simulated Allocation System model (LSAM), a simulation program that estimates the clinical impact of alternative allocation systems on key transplant outcomes (18).

Table 3. Modeled allocation systems and LSAM results LSAM results for transplants performed and lives saved

	Current	15/35	15/32	15/29	15/25	15/22
Transplants performed
Status 1	440	453	454	455	450	455
MELD 35+	692	769	744	733	719	703
MELD 30–34	595	600	652	667	654	626
MELD 25–29	1000	1047	1045	1062	1197	1081
MELD 20–24	2134	2215	2194	2179	2124	2383
MELD 15–19	1153	1205	1196	1182	1125	1011
MELD 10–14	255	25	25	28	27	25
MELD <10	55	12	12	12	12	12
Total	6324	6324	6320	6319	6309	6296
Deaths
Waitlist	1659	1592	1583	1574	1572	1585
Posttransplant	602	606	624	618	621	616
Postremoval	406	392	387	393	386	389
Total	2667	2589	2594	2585	2579	2589
Lives saved	0	78	73	82	88	78

We utilized the number of transplants done at each MELD score and Status category under the current allocation system and each of the proposed allocation systems modeled through LSAM. The expected cost of transplant for patients in each illness severity category was estimated using each MELD and status-specific cost estimate from the categorical regression analysis; the distribution of other clinical characteristics in the cost regression was drawn according to illness category distributions for patients in the OPTN in 2006 (Appendix B) and multiplied by the corresponding categorical cost regression parameters. Finally, the total cost of a sharing proposal was computed as product of expected costs for patients transplanted in each illness severity category with the count of transplants performed at each severity stratum.

To compute the benefit of an alternative-sharing proposal, we quantified the posttransplant survival expectations (quality-adjusted life-years, QALYS) according to the expected counts of net lives saved in each illness severity category at transplant produced by the LSAM simulations. Long-term survival was estimated using a two-part method in which a Kaplan–Meier curve was fit to observed survival data over the first year after transplant, and a parametric survival function with a Weibull form is fit to observed survival data in years 1–5 (19,20). The Weibull form is then applied to project survival to 20 years posttransplant and QALYs saved are computed as the area under the 20-year projection, adjusted for a transplant utility of 0.8 and discount rate of 5% (21). Posttransplant QALYs were computed for each illness severity category at transplant using historical OPTN data for patients transplanted in 2002–2008. The benefit of a sharing proposal was computed as the product of net lives saved per illness severity category with stratum-specific QALY expectations, and used to calculate the costs per QALY gained.

Results

Economic analysis

The mean cost of the transplant hospitalization in the study sample was $122 862 (standard deviation $96 669; interquartile range $65 744). Multivariate regression confirmed a strong association between total hospital cost and key clinical variables (Table 2). Recipient factors associated with significantly increased cost in this analysis included recipient age ≥60 years, female gender, other race, low body mass index and hepatitis B (p < 0.05 for all). While DRI > 1.17 was associated with an increased cost of transplant, there was not a completely linear increase in costs associated with donor risk in this population. There was a gradual increase in the overall cost of transplant care in more recent years despite adjustment for medical inflation. Thus, mean cost of transplant increased by more than $60 000 between 2002 and 2008 in this cohort (p < 0.0001).

Table 2. Regression model for prediction of transplant hospitalization costs based on illness severity at transplant and other clinical and demographic factors

Characteristic Regression intercept	Parameter estimate, (Standard error) $ 129 519 (6338)	p-Value <0.0001
Illness severity at transplant MELD score at transplant
<10	Reference
10–14	6876 (2853)	0.02
15–19	7876 (2826)	0.005
20–24	21 782 (2939)	<0.0001
25–29	42 171 (3214)	<0.0001
30–34	79 136 (3507)	<0.0001
≥35	87 625 (3402)	<0.0001
Status 1	60 012 (4758)	<0.0001
Recipient age
12–17	–6999 (6552)	0.28
18–29	–5642 (4346)	0.19
30–45	Reference
45–59	3560 (2308)	0.12
60+	8547 (2588)	0.001
Female gender	8613 (1597)	<0.0001
Race
White	Reference
Black	510 (2431)	0.83
Other	16 882 (3074)	<0.0001
Hispanic ethnicity	–5197 (3586)	0.15
Blood type
AB	1900 (3155)	0.55
A	Reference
B	5493 (2238)	0.01
O	3687 (1592)	0.02
Cause of liver failure
Hepatitis C	Reference
Hepatocellular carcinoma	1900 (244)	0.44
Autoimmune	3678 (2281)	0.11
Fulminate Hepatic Failure	5479 (3803)	0.15
Alcoholic cirrhosis	1221 (2377)	0.61
Primary sclerosing cholangitis	–4807 (3180)	0.13
Primary biliary cirrhosis	6960 (4013)	0.08
Hepatitis B	9311 (4142)	0.02
Nonalcoholic steatohepatosis	–2990 (11 863)	0.80
Cancer	13 339 (11 219)	0.23
Other	6341 (3927)	0.11
BMI category (kg/m2)
<18.5	Reference
18.5 to <25	–21 554 (4949)	<0.0001
25 to <30	–24 485 (4933)	<0.0001
30 to <35	–23 971 (5051)	<0.0001
35+	–16 302 (5215)	0.002
Missing	–34 282 (5259)	<0.0001
Donor risk index
<1.17	Reference
1.17–1.36	2883 (2351)	0.22
1.37–1.61	7735 (2356)	0.001
1.62–1.92	5241 (2392)	0.03
>1.93	11 013 (2516)	<0.0001
Missing	11 306 (2598)	<0.0001
Year of transplant
2002	–61 183 (2827)	<0.0001
2003	–51 603 (2723)	<0.0001
2004	–47 946 (2682)	<0.0001
2005	–34 244 (2618)	<0.0001
2006	–23 411 (2544)	<0.0001
2007	–13 312 (2570)	<0.0001
2008	Reference

Increasing calculated MELD score was strongly associated with overall cost of liver transplantation (p < 0.0001). However, the increase in cost associated with higher MELD score was nonlinear (Figure 2). Above MELD of 25, the cost of liver transplant increased exponentially in the continuous analysis. For example, incremental costs for patients with MELD >30 exceed those with MELD <10 by approximately $80 000 (p < 0.0001). The analysis of costs according to illness severity category (categorical MELD strata and status 1), which was employed to integrate with the LSAM models, shows similar patterns albeit with less granularity across MELD values.

Discussion

Liver transplantation has matured into a well-accepted treatment for patients with end-stage liver failure, but application of this therapy is constrained by access to acceptable donor allografts. This analysis of a novel database containing linked clinical and cost-accounting data from a large, national sample of liver transplant centers confirms prior single center-based observations that severity of illness, patient age, diagnosis and body mass index affect the cost of liver transplantation. Among these factors, severity of illness is the most dramatic cost driver, affecting costs according to a nonlinear functional form. Thus increases in MELD from 4 to 25 have a minimal impact on cost while MELD scores above this level are associated with rapidly escalating costs. The analysis also demonstrates marked differences in center-reported costs of transplant as a function of OPTN region, even after local wage-price adjustment to allow comparability across geography. Using this unique database, we have also assessed the economic impact of novel allocation systems on the incremental cost of transplant care expected with broader sharing of available donor allografts. Our findings suggest that broader sharing of organs may be highly cost effective, with a transplant hospitalization-related cost per year of life saved of less than $20 000.

Prior analyses of factors influencing the cost of liver transplantation using single center data have demonstrated that severity of illness is associated primarily with longer length of stay (LOS). Thus, high MELD patients use more pharmaceuticals, laboratory tests, radiographs and dialysis services (1). There was no significant association with operating room costs, blood bank or anesthesia. These findings are consistent with the MELD-cost curve identified in the current study. At low MELD scores, the ‘fixed’ costs of liver transplant are dominant and consistent. Thus, all liver transplants require operating room time, anesthesia and, in general, a period of ICU care. This corresponds with the generally flat lower part of the cost curve. Above a MELD of 25, the ‘variable’ costs dominate which reflect the cost of delivery care to the individual patient as a function of degree of illness. We have previously demonstrated that the shift to a MELD-based system of organ allocation was associated with increased costs and reduced profits for liver transplant centers as a result of an increase in high-MELD, costly patients (11). Washburn et al. recently assessed factors, which influence LOS and resource utilization in two large, geographically disparate centers (12). In this experience, the relationship between MELD and cost was consistent across the institutions. Interestingly, however, adjusted LOS was actually reduced at the hospital with the higher MELD score, suggesting, perhaps, some learning curve in managing high MELD recipients.

The desire to expand access to liver transplant for patients dying on the waiting list has given rise to multiple strategies to broaden the pool of potential organs. We have previously explored the impact of the DRI on LOS and estimated cost at a registry level (9). In that analysis, it was principally the highest DRI livers, which drove costs. In the current analysis, increasing DRI was associated with cost, though the effect was small in comparison to recipient factors. This may reflect our emphasis on inpatient cost or a learning curve associated with the use of marginal donors (22). MELD and DRI do interact to increase costs of care in both the inpatient stay and over the life of the transplant, with variation according to MELD score dominating the cost during the initial transplant hospitalization and DRI predominantly affecting long-term costs (10). Other analyses have demonstrated that using organs from Donors following Cardiac Death (DCD), dramatically increases the cost of transplant within the first posttransplant year while decrease graft and patient survival (23–25). Jay et al. demonstrated that compared with livers from donors after brain death, the cost of DCD liver transplant was over 25% higher driven principally by the high incidence of biliary complications (13). Finally, centers have turned to living donors to attempt to bridge the gap between supply and demand. Data from the NIH sponsored Adult to Adult Living Donor Liver Transplant (A2ALL) registry were used to construct a Markov model the estimated the cost-effectiveness of living donor liver programs. Northrup et al. estimated the incremental cost-effectiveness ratio of adding a living donor program to a deceased donor liver transplant (direct evaluation costs, organ recovery and cost of the transplant) to be >$100 000 per quality adjusted life year (26).

Unlike the use of marginal donor organs, this analysis suggests that a re-allocation of available liver allografts has the potential to increase the benefit of transplantation far more cost effectively. Through the implementation of a 15/25 tiered allocation system, we could increase the life-years benefit from transplant by 551 at a cost of less than $20 000 per life year saved. This compares favorably with the incremental costs effectiveness of procedures accepted by CMS for Medicare coverage, which range from <$50 000 to more than $100 000 per life year saved (27). In general, a ratio less than $50 000–$100 000 per QALY is considered cost effective in the US (28,29). This estimate is likely a significant under estimate of the true benefit of this re-allocation, because we did not include the differential in cost of caring for patients on the waiting list. We have shown through previous analyses that pretransplant costs and hospitalization are highly correlated with MELD score (30). Thus, transplanting a high MELD patient rather than maintaining that patient on the waiting list results in more savings than that achieved through the transplantation of a low MELD patient. Posttransplant costs, however, are minimally associated with MELD score. Thus, after their transplant hospitalization, we would not expect a marked difference in the cost of posttransplant care if more high MELD patients were transplanted. We also did not capture the costs incurred by patients who are forced to travel long distances in search of organs under the current system. For these patients, the direct medical costs are likely to be similar, but the costs of travel, lost work and the need to find housing are likely significant.

Although this is a robust database including OPTN clinical data and hospital reported cost data, several limitations must be acknowledged. First, there are likely to be some clinical factors, which are poorly captured in UNOS and affect the cost of transplant (e.g. poor social support). However, these factors are unlikely to diminish the association between MELD and transplant cost. They may, however, increase the cost of transplant disproportionately in areas with long waiting times, which would benefit from re-allocation of available livers. Second, these analyses include only the cost of the initial transplant hospitalization. Thus, physician fees and posttransplant costs are not captured in this database. This limitation is mitigated by the fact that the vast majority of the costs of liver transplant are the hospital costs incurred during the initial hospitalization (10). In addition, MELD score prior to transplant has a limited impact on posttransplant costs (30). Third, we have included the cost of liver–kidney transplantation within the overall cost of liver transplant at each MELD score. Thus, a portion of the increased cost of transplanting higher MELD patients reflects the greater need for liver–kidney transplants. Given current reimbursement under Medicare, which does not include additional payment for liver–kidney transplant, it is appropriate to include these cases in the cost of liver transplantation. Fourth, we did not directly factor in the increase in transportation costs incurred through broader sharing. To address this concern, we considered two additional factors. The SRTR model estimates that implementation of the 15/25 sharing policy would increase the average distance from donor hospital to transplant center by only 32 miles. We also examined the proportion of shared allografts. Using of the proposed national sharing system of 15/25 would increase the proportion of allografts used out of the DSA by approximately 10% or 630 transplants. We estimated that even if this sharing these organs resulted in a $10 000 per organ incremental transportation cost, this would increase the total cost of liver transplantation by < 1%.

It is also very important to acknowledge the fundamental limits of the simulation models (LSAM) to completely predict what will happen under alternative allocation systems. First, although the LSAM models consider the use of exception points in the allocation assignment, the models reported in this simulation use laboratory MELD score and status 1 only. We also use the laboratory MELD/status 1-stratified costs to calculate costs as this score is more closely tied to costs than exception MELD. Because the exception patients are likely to be lower cost than predicted by considering MELD in terms of allocation MELD score at transplant, it is possible our cost-effectiveness models over-estimate the cost consequences of the shifting of organ to higher MELD patients, as some of these patients may, in fact, be tumor patients who will cost less than predicted here. This would further reduce the cost-per-QALY incurred in reallocation of available allografts. Second, the simulation models are based upon current listing and acceptance practices. Changing allocation may lead to increased acceptance of marginal livers in regions where the average MELD at transplant begins to increase. These changes may increase the cost of transplantation in these regions. However, there will likely be an offsetting reduction in the use of marginal organs in what are currently the most under-supplied regions, and corresponding savings by reducing the need to use high DRI organs in the highest MELD recipients (9,10). Third, we could not directly account for changes in the proportion of livers that would be used for liver–kidney transplants. However, it is unlikely that broader sharing would impact the use of liver–kidney transplant as a proportion of all transplants performed within a given MELD strata. Thus, the average costs we used to estimate cost effectiveness are marginally higher than costs incurred for liver transplants alone, as our cost data includes patients who receive liver–kidney transplants at a given MELD score.

In conclusion, use of this detailed source of clinical and economic data has provided an improved method to estimate the economic implications of changes in transplant practice. Detailed understanding of transplant costs allows a better understanding of the impact of recipient and donor factors on cost. Among the factors identified, perhaps MELD at transplant has the greatest potential to be affected by local factors. This analysis is reassuring in that it provides support that broader organ sharing offers an opportunity to reduce the death from end-stage liver disease in a durable, highly cost-effective manner that can be implemented without risk to living donors. Further implementation of this policy should be strongly considered in light of the ongoing disparity in access to transplant organs and the need to maximize the benefit of each available organ.

Funding Sources

This work was supported by an American Recovery and Reinvestment Act grant from the National Institute of Diabetes Digestive and Kidney Diseases, RC1 1RC1DK086450–01. Dr. Axelrod was supported by a grant from the Hitchcock Foundation. Dr. Lentine also received career development support from an NIDDK grant, K08DK073036.

Disclosure

The authors of this manuscript have no conflicts of interest to disclose as described by the American Journal of Transplantation.