determining priority for liver transplantation

Appl Health Econ Health Policy 2005; 4 (4): 249-255ORIGINAL RESEARCH ARTICLE 1175-5652/05/0004-0249/$34.95/0

© 2005 Adis Data Information BV. All rights reserved.

Determining Priority for Liver TransplantationA Comparison of Cost per QALY and Discrete Choice Experiment-GeneratedPublic Preferences

Julie Ratcliffe,1 Martin Buxton,2 Tracey Young2 and Louise Longworth2

1 Sheffield Health Economics Group, ScHARR, University of Sheffield, Sheffield, UK2 Health Economics Research Group, Brunel University, Uxbridge, UK

Objective: A comparison of the implications of the application of the principles of equity and efficiency as twoAbstractdesirable but competing attributes of the organ allocation system. Efficiency is defined in economic terms as thestandard cost per QALY model and equity considerations are included in a model based on public preferencesgenerated from a discrete choice experiment in determining priority for donor liver graft allocation.Methods: A survey of the general public (n = 303) using a discrete choice experiment was undertaken. Theresults enabled estimation of the relative weights attached to several key factors which might be used to prioritisepatients on the waiting list for liver transplantation. These weights were then used to develop a patient-specificindex (PSI) for all patients diagnosed with one of three main chronic liver diseases who had received a livertransplant during an 18-month period at all Department of Health designated liver transplant centres in Englandand Wales (n = 207). The cost per QALY model comprised net total costs from assessment to 27 monthsfollowing assessment as the numerator of the ratio. Net survival over the same time period, adjusted for HR-QOLusing population values for the EQ-5D descriptive system, formed the denominator.Results: Priority for liver transplantation differed markedly according to whether patients were rankedaccording to efficiency (net cost per QALY) or equity considerations (PSI) and the differences in ranks werefound to be statistically significant (Wilcoxon signed rank test p < 0.001).Conclusions: This study emphasises that the priorities of the general public may not accord with those arisingfrom a pure efficiency objective and quantifies the extent of the efficiency loss in terms of lost QALYs andincreased net programme costs associated with the incorporation of equity concerns as reflected in publicpreferences for the allocation of donor livers for transplantation.

Liver transplantation is now widely accepted as the treatment of The methodology of economic evaluation in healthcare tradi-tionally identifies the maximisation of health, subject to a costchoice for patients with end-stage liver disease.[1] In recent years,constraint, as the main criterion by which priorities for resourceas the technology has developed, survival rates from the procedureallocation should be identified. Typically, the benefits of health-have improved dramatically and this has resulted in a broadeningcare interventions are measured in terms of life-years gained orof the indications for which transplantation is considered appropri-QALYs gained. Applying the cost per QALY model in the context

ate. As a consequence an increasing number of patients are beingof liver transplantation would imply giving priority to patients on

referred and accepted for liver transplantation. Unfortunately,the waiting list with the greatest capacity to benefit in QALY

however, the number of donated livers has remained relatively terms relative to the estimated costs associated with their treat-static in recent years.[2,3] Given the imbalance between the demand ment. Indeed, such a policy seems to be in broad concordance withfor and supply of donor organs, it is necessary to employ decision the views of the majority of transplant physicians, who believe thatcriteria to determine which patients should be given priority in patients should be treated irrespective of the cause of their liverreceiving a donor organ. failure and based upon their capacity to survive and benefit.[4]

250 Ratcliffe et al.

However, it is not clear that the general public share this view. A tient stay, outpatient visits, high cost/high volume drugs, bloodUK survey[5] and similar US-based surveys[6,7] of the general products, nutrition, physiotherapy sessions, dietitian sessions,public’s allocation decisions in liver transplantation have suggest- tests, treatments and the length of the transplant operation provid-ed that the general public place a high priority on equity in terms of ed at the transplant centre. Unit costs for resources were soughtgiving everyone a chance to receive a scarce organ (including from each of the six centres and mean costs, weighted by thethose with an anticipated poor capacity to benefit) even if this number of transplants performed at each centre, were applied toresults in a substantial decrease in the chance that the available each item of resource use. Indirect costs to the transplant centresorgans will save lives. There is also evidence that the general were included in the estimates of unit costs. Drug costs were takenpublic would give a higher priority to younger patients (particular- from the British National Formulary.ly those with dependants) and to those whose liver damage could The pre-transplant costs of each patient during the waiting listnot be described as self-induced, e.g. through alcoholism or illegal period were estimated by observing the resource use incurred anddrug use.[5] applying the unit cost estimates. Costs in the absence of transplan-

It has been argued that it is important for the general public to tation were estimated by calculating an average cost per day forbe consulted on this issue, since ultimately the supply of trans- the waiting list period and applying it to the predicted number ofplantable livers is entirely dependent on the public’s willingness to days of their survival in the absence of transplantation. Analysis ofdonate. A US-based study of people who declined to donate livers the pre-transplant cost data indicated that on average the costs ofshowed that one of the main reasons for not donating was a patients increased in the month prior to death. Based on the resultsperceived lack of confidence in the fairness of donor liver alloca- of an ordinary least squares regression model an adjustment wastion.[8] Similarly, a more recent US survey of factors associated made to the costs of the final month of patients whose predictedwith (un)willingness to be an organ donor found that those who survival in the absence of transplantation was less than 27 months.were unwilling more often thought that the transplant waiting The regression model allowed for patients’ ages, disease types andsystem is based on race and income.[9] whether they were considered an emergency candidate (defined as

This paper compares the allocation of donor livers based on not expected to survive for more than 3 days without a transplant).public preferences with those implied by cost per QALY calcula- A self-completion questionnaire containing the EQ-5D wastions. The public preferences are derived from the results of a administered to all study patients (n = 207) at the point of listing,quantitative survey to members of the general public.[10] The and then to the patients who were still waiting to receive a liversurvey was based on a discrete choice experiment (DCE) which transplant at 3, 6 and 12 months after listing. The questionnaireallowed the relative weight attached to specific equity versus was re-administered to all study patients following transplantationefficiency considerations in determining the general public’s allo- at 3, 6, 12 and 24 months after transplantation. A ‘tariff’ of valuescation decisions to be determined. The results from the DCE were representative of the UK general population generated using theused to derive a patient-specific index (PSI) for a sample of time trade-off (TTO) method was applied to the EQ-5D classifica-patients with chronic liver disease who were listed for transplantation system to generate single-index health state values.[12]

tion at six regional liver transplant centres in the UK. The net For the purposes of the calculation of net survival (observedQALY gains were also estimated for the same study patients on survival with transplantation minus predicted survival in the ab-the basis of results of a cost utility analysis from assessment for sence of transplantation), survival in the absence of transplantationtransplantation until 27 months post-assessment.[11] The cost per was estimated using validated prognostic indicators for primaryQALY implied preference orderings for donor liver graft alloca- biliary cirrhosis (PBC), alcoholic liver disease (ALD) and primarytion were then compared with the PSI implied preference order- sclerosing cholangitis (PSC) patients to estimate the probability ofings. surviving over 27 months from date of listing.[13] These models

have been generated on the basis of the survival experience ofMethods cohorts of non-transplant patients with end-stage liver disease and

can be used to predict survival in the absence of transplantation onthe basis of the values of several clinical variables immediatelyCalculating Cost per QALYprior to transplantation.

In brief, detailed information on resource use incurred at the For the estimation of net QALYs (observed QALYs withtransplant centres was collected for each patient prospectively. transplantation minus predicted QALYs in the absence of trans-Resource use data were measured for 27 months from the point of plantation) several assumptions have been made. With transplan-assessment for transplantation and included all subsequent inpa- tation, QALYs have been estimated using observed EQ-5D and

© 2005 Adis Data Information BV. All rights reserved. Appl Health Econ Health Policy 2005; 4 (4)

Determining Priority for Liver Transplantation 251

Characteristics Group A Group B

Age 50 years 40 years

Alcoholic liver disease Yes No

Expected length of post-transplant survival

5 years 5 years

Time on waiting list 3 months 3 months

Re-transplanted No No

How would you allocate the available livers between the two groups of individuals?(The total for the two groups should add up to 100.)Please write the number of livers allocated to each group in the boxes below.

Group A Group B

TOTAL = 100

Fig. 1. Example of choice question from discrete choice experiment (DCE) questionnaire.

survival data over 27 months from time of listing. Health-related academic staff. A useable response rate of 38% was achievedquality of life in the absence of transplantation has been approxi- (n = 303).mated by calculating patients’ observed QALYs up to the time of Respondents were presented with eight choice situations intransplantation. It is assumed that patients would remain in the which they were asked to allocate 100 donor liver grafts betweensame EQ-5D health state as their last EQ-5D state reported prior to two groups of 100 individuals in urgent need of a transplant. Thetransplantation, for the length of time they were predicted to groups of individuals differed in terms of the length of time spentsurvive by the prognostic indicator. Full details on the calculation waiting, the life-years gained following transplantation, age, alco-of net costs and QALYs have already been published.[11] holic or non-alcoholic liver disease and whether they were primary

or re-transplant candidates. An example of a choice questionincluded in the questionnaire is presented in figure 1. The vari-Development of the Patient-Specific Index (PSI)ables included in the data analysis and their codings are presentedin table I.To investigate the nature of public preferences in the allocation

of donor liver grafts for transplantation, a DCE was designed for a Only two respondents chose to allocate all available organs toself-completion questionnaire survey (for further details of the the group with the highest expected benefit in terms of healthmethods and results of this survey, see Ratcliffe[10]). The question- outcome; the overwhelming majority chose not to abandon thenaire was administered to 800 randomly chosen employees from a group with the lower expected survival. In the extreme case, sevenBritish university; the sample included non-academic as well as respondents (2%) chose to allocate the donor organs equally

Table I. Discrete choice experiment (DCE) attributes and their codings for data analysis

Variable Type Coding

attribute description

DL Difference in the number of livers allocated as move from group A to group B Continuous –100 to +100

AGE Difference in age in years as move from group A to group B Continuous –20 to +20

ALCO Difference in alcoholism as move from group A to group B Discrete –1, 0, 1alcohol: yes = 1alcohol: no = 0

SURV Difference in expected survival in years as move from group A to group B Continuous –10 to +10

WAIT Difference in waiting time in months as move from group A to group B Continuous –9 to +9

RETRANS Difference in transplantation status as move from group A to group B Discrete –1, 0, 1re-transplanted: yes = 1re-transplanted: no = 0



between the two groups of individuals regardless of their charac- Solving this equation for DL gives a value of +54, which meansteristics. that on average the group would allocate 54 more livers to group B

than to group A.The data from the DCE were analysed in the statistical packageThe equation generated from the DCE was then used to developSTATA®[14] using a random effects linear regression model.1 The

a PSI which could be used to generate a score for chronic liverresults (table II) indicated that all of the attributes were significantdisease patients who had received a transplant. The PSI for eachin influencing respondents’ allocation decisions. The sign (+/–)study patient was generated using the following model:attached to each coefficient indicates the direction of preferences.PSI = –1.49(AGE) –38.18(ALCO) +4.05(SURV) +1.14(WAIT)It was found that, on average, members of the general public+7.90(RETRANS) –0.50would give greater priority to younger patients, those without

ALD, those with a greater expected length of post-transplantRanking by PSI and Cost per QALYsurvival, those who had been on the waiting list for the longest

period and patients who are being re-transplanted. Both total costs and net QALYs gained beyond the base yearUsing the estimated equation generated in table II, the follow- were discounted at the current UK treasury rate of 3.5%.[15] For

ing model was generated: each study patient, cost per QALY estimates were then obtainedDL = –1.49(AGE) –38.18 (ALCO) +4.05(SURV) +1.14(WAIT) by dividing the total net discounted costs by the total net dis-+7.90(RETRANS) –0.50 counted QALY gain. Patients were then ranked according to PSIwhere DL is the difference in the numbers of livers allocated and cost per QALY scores. It was then hypothesised that 100between groups A and B, AGE is the difference in age between the donor livers would become available for transplantation. Thetwo groups, ALCO is the difference in alcoholic versus non- donor livers were allocated until exhausted firstly on patientsalcoholic status, SURV is the difference in expected length of ranked according to PSI and secondly on patients ranked accord-survival, WAIT is the difference in time spent waiting and RE- ing to cost per QALY. This enabled a direct comparison of theTRANS is the difference in primary versus re-transplant status. characteristics of the two patient groups who would receive priori-

Using this model it is possible to predict how respondents ty according to each allocation mechanism.would make allocation decisions between two groups of individu-

Resultsals on the basis of their characteristics. In the choice replicated infigure 1, for example, the individuals in group B are younger and Table III summarises the characteristics of the 100 top candi-have naturally occurring liver disease, whereas those in group A dates according to PSI and cost per QALY rankings. (The fullare older and have ALD. In all other respects the groups are results for all patients in the sample are available from the au-assumed to be the same. thors.) A total of 64 patients appeared in both groups. The mean

Using the estimated equation for DL above and the codings for age and transplantation status were similar, although there werethe levels of the attributes presented in figure 1, this gives: relatively more males in the PSI group. There were fewer patientsDL = –1.49(–10) –38.18 (–1) +4.05(0) +1.14(0) +7.90(0) –0.50 with ALD in the PSI group and the difference was statistically

significant (chi-squared 64.98, p < 0.001). The mean cost perQALY was also higher for the PSI group (£38 559 compared with£18 781) and the difference was statistically significant (Wilcoxontest, p < 0.001). The programme of 100 transplants would cost£4.8 million based on PSI criteria and £3.0 million based on costper QALY criterion and would generate 153 QALYs using PSIand 162 QALYs using cost per QALY.

Figure 2 presents a comparison of the PSI and cost per QALYrankings in graphical format. If the rankings for each patient werethe same using either approach the points would represent astraight line at 45° from the origin. The Wilcoxon signed rank testwas used to test for statistically significant differences between the

Table II. Random effects regression model of results of discrete choiceexperiment (DCE)a

Attribute Coefficient p-Value 95% CI

AGE –1.49 <0.001 –1.73, –1.25

ALCO –38.18 <0.001 –41.28, –35.08

SURV +4.05 <0.001 3.62, 4.47

WAIT +1.14 0.006 0.33, 1.94

RETRANS +7.89 0.003 2.68, 13.10

CONSTANT –0.50 0.606 –2.42, 1.42

a Number of observations = 2413; n = 303; chi-squared = 1632.4(p < 0.001); R2 = 0.43.

1 A choice between the random effects linear regression model and a fixed effects model was made using the Hausman test, which indicated that therandom effects model was superior (chi-squared 7.3, p chi-squared 0.1994). The data were also analysed using a tobit regression model. The results ofthis analysis were very similar to those of the random effects model.



Table III. Summary of characteristics of patients receiving priority for transplantation according to patient-specific index (PSI) and cost per QALY

Variable PSI Cost per QALY

Patients appearing in both groups (%) 63 63

Mean age (range) [y] 49 (22–67) 51 (22–69)

Sex (%)

male 57 45

female 43 55

Re-transplanted (%) 9 8

Primary liver disease (%)

alcoholic liver diseasea 11 33

primary biliary cirrhosis 53 43

primary sclerosing cholangitis 36 24

Mean waiting time [mo] (range) 2.7 (1–17) 3 (1–17)

Mean incremental costa [£; 1999 values] (range) 48 035 (16 974–212 974) 30 055 (14 921–49 754)

Mean cost per QALYa [£; 1999 values] (range) 35 889 (9175–272 609) 18 781 (9175–28 641)

Mean PSIa (range) 24.23 (–9.65 to 70.18) 11.07 (–41.95 to 67.34)

Mean net QALY (range) 1.54 (0.38–2.17) 1.63 (1.06–2.17)

Total net programme costs (£) 4 847 508 3 045 800

Total net programme QALYs 153 162

a Significant difference in means at 5% level.

cost per QALY and PSI rankings and it was found that there were been noted in other surveys of the public’s views regarding donorhighly statistically significant differences (p < 0.001) in rank allocation, thereby highlighting the tension between judgementalorderings of patients using either approach. views and the ethical view that social desirability criteria are

unjust ways to allocate resources. Such findings have promptedTable IV documents the characteristics of the 63 patients incalls for public education to explain the transplant community’stotal who appear in both cost per QALY and PSI priority listings.rationale for distributing donor organs to patients regardless ofIt can be seen that the vast majority of these patients have atheir role in causing their illness.[16]primary diagnosis of either PSC or PBC and all exhibit reasonably

high net QALY gains and lower incremental costs at 2 years after Although the theoretical existence of equity-efficiency trade-transplant relative to the total sample. offs in the provision of healthcare is well documented, this paper

provides a rare policy-relevant, real-world example. The empiricalDiscussion implications are presented of the choice between two equally

common and potentially equally justifiable policy positions: toThe findings of this study indicate that priority for liver trans-elicit and follow public preferences, or to estimate cost-effective-plantation based on public preferences differs markedly from the

priority implied by the health economists’ construct of cost perQALY and the differences are statistically significant. The resultsfrom the DCE exercise indicated that the majority of respondentswere prepared to allocate a proportion of donor organs to the groupwith the lower expected survival, thereby sacrificing some gain inthe efficiency of the transplantation programme for an increase inequity or fairness in the allocation of donor livers. When compar-ing patients’ rankings according to cost per QALY and PSI, thissocial preference translated into an efficiency loss of 9 QALYsand an increased total net programme cost of £1 801 708.

The results from the PSI rankings also indicate that respondentswould give greater priority to patients whose liver disease couldnot be attributed to alcohol consumption. Similar results have also

0

20

40

60

80

100

120

140

160

180

200

0 50 100 150 200

PSI ranking

Cos

t per

QA

LY r

anki

ng (

£)

QALY and PSIQALY onlyPSI only

Fig. 2. Scatter plot comparison of patient-specific index (PSI) and cost perQALY rankings for top 100 candidates in each group (£; 1999 values).



naire, some individuals may have experienced difficulties in un-derstanding and interpreting the QALY concept. Although thecorrelation between post-transplant survival and the net QALYgain was high (Pearson’s correlation coefficient 0.710, p < 0.001)it is possible that the inclusion of quality of life, either as a separateattribute within the DCE or presented in combination with survivalin the form of QALYs, may have improved comparability betweenthe PSI and cost per QALY estimates for comparative purposes.

The DCE study was in many ways an opportunistic study.Individuals were surveyed at one timepoint only and were provid-ed with limited information about the transplantation process.There is evidence to suggest that the public’s views on prioritysetting in healthcare may change as a result of discussion anddeliberation.[18] If the public were given the opportunity to reflectupon the implications of the implementation of their preferences,e.g. in terms of the high opportunity costs associated with theallocation of donor livers according to the PSI, this may lead to a

Table IV. Summary of characteristics of patients appearing in top 100using either patient-specific index (PSI) or cost per QALY

Variable PSI and cost per QALY

Mean age (range) [y] 49 (41–57)

Sex (%)

male 52

female 48

Re-transplanted (%) 10

Primary liver disease (%)

alcoholic liver disease 13

primary biliary cirrhosis 52

primary sclerosing cholangitis 35

Mean waiting time [mo] (range) 2.7 (1–17)

Mean incremental cost [£] (range) 30 473 (16 974–47 725)

Mean cost per QALY [£] (range) 18 664 (9175–28 440)

Mean PSI (range) 23.86 (–21.65 to 67.33)

Mean net QALY (range) 1.67 (1.06–2.17)change in their thinking, which may lead ultimately to a change inpreferences. Further research in this area should attempt to address

ness and base choices on efficient use of scarce resources (donor these issues by assessing the role of education, reflection andorgans). The results illustrate the extent to which choices based on deliberation on people’s views concerning priority for liver trans-the economists’ construct of cost per QALY and preferences plant allocation.elicited from the public may diverge in practice and the nature of However, the findings of this study suggest that, in the case ofthe inconsistencies that arise. The analysis also provides an indica- liver transplantation, there may well be an important discordancetion of the magnitude of the potential efficiency loss in terms of between allocation based on public preferences and allocationlost QALYs and increased net costs for the liver transplantation according to cost per QALY principles.programme if public preferences are followed.

It is important to note that neither of the two studies on which Acknowledgementsthe results reported in this paper are based can be considered as

We would like to acknowledge the assistance of the cost effectiveness ofentirely representative of the underlying ‘ideals’. The DCE wasliver transplantation (CELT) study team in the collection of the cost peradministered to a convenience sample of university employees andQALY data. We thank colleagues in ScHARR, in particular Aki Tsuchiya formay therefore not be representative of the general population. Thehelpful comments upon earlier drafts of this paper. The research was financial-

cost per QALY estimates are based on a relatively short time ly supported by the Department of Health’s Policy Research Programme. Theperiod of post-transplant follow-up and might change over a views expressed and any errors or omissions are the responsibility of the

authors alone.longer period of follow-up.The authors have no conflicts of interest that are directly relevant to theIdeally a detailed estimate of the benefits of liver transplanta-

content of this study.tion would be based on longer-term follow-up. Given the evidencethat the survival rate in liver transplant patients beyond 2 years

Referencesremains fairly constant,[17] it is likely that the cost effectiveness of1. Neuberger J. Developments in liver transplantation. Gut 2004; 53: 759-68transplantation would improve over time. However, it is not clear2. 2001 annual report of the US Organ Procurement and Transplantation Network and

that longer-term follow-up would necessarily lead to significant the Scientific Registry for Transplant Recipients: Transplant Data 1991-2000.Rockville (MD): Department of Health and Human Services, Health Resourceschanges in patients’ ranking according to cost per QALY, sinceand Services Administration, Office of Special Programs, Division of Trans-

one would expect improvements in cost per QALY ratios to be plantation; Richmond (VA): United Network for Organ Sharing; Ann Arbor(MI): University Renal Research and Education Association, 2001fairly constant across the patients in the group.

3. National Health Service (NHS) UK Transplant. Transplant activity in the UKIn estimating the PSI, the expected length of post-transplant 2002-2003. Stoke Gifford: UK Transplant, 2003

4. Ghent CN. Overall evaluation: screening and assessment of risk factors. Liversurvival (rather than quality-adjusted post-transplant survival) wasTranspl Surg 1996; 2: 2-8

used as a proxy for health outcome. At the time of administration it 5. Neuberger J, Adams D, McMaster P, et al. Assessing priorities for allocation ofwas considered that, given the mode of delivery of the question- donor liver grafts: survey of public and clinicians. BMJ 1998; 317: 172-5



6. Ubel P, Loewenstein G. The efficacy and equity of retransplantation: an experi- 14. StataCorp. Stata statistical software: release 8.0. College Station (TX): Statamental survey of public attitudes. Health Policy 1995; 34: 145-51 Corporation, 2003

7. Ubel P, Loewenstein G. Distributing scarce livers: the moral reasoning of the 15. Great Britain HM Treasury. The green book appraisal and evaluation in centralgeneral public. Soc Sci Med 1996; 42: 1049-55 government. London: Her Majesty’s Stationery Office (HMSO), 2003

8. Peters TG. Organ donors and non-donors. Arch Intern Med 1996; 156: 2419-24 16. Neuberger J, Ubel PA. Finding a place for public preferences in liver allocation9. Haustein SV, Sellers MT. Factors associated with (un)willingness to be an organ decisions. Transplantation 2000; 70: 1411-3

donor: importance of public exposure and knowledge. Clin Transplant 2004;17. Neuberger J, Lucey M. Liver transplantation practice and management. London:

18: 193-200BMJ Publishing Group, 1994

10. Ratcliffe J. Public preferences for the allocation of donor liver grafts for transplan-18. Dolan P, Cookson R, Ferguson B. Effect of discussion and deliberation on thetation. Health Econ 2000; 9: 137-48

public’s views of priority setting in health care: focus group study. BMJ 1999;11. Longworth L, Young T, Buxton M, et al. Mid term cost effectiveness of the liver318: 916-9transplantation programme of England and Wales for three liver disease groups.

Liver Transpl 2003; 9: 1295-30712. Dolan P. Modelling valuations for Euroqol health states. Med Care 1997; 35 (11): Correspondence and offprints: Julie Ratcliffe, Sheffield Health Economics

1095-108Group, ScHARR, University of Sheffield, 30 Regent St, Regent Court,13. Young T, Longworth L, Ratcliffe J. Survival on the waiting list, post transplant andSheffield S1 4DA, UK.estimated survival in the absence of transplantation: final report to the Depart-

ment of Health. Uxbridge: Brunel University, 2000 E-mail: [email protected]


determining priority for liver transplantation

Documents