Journal Logo

Clinical and Translational Research

Elective Liver Transplant List Mortality: Development of a United Kingdom End-Stage Liver Disease Score

Barber, Kerri1,3; Madden, Susanna1; Allen, Joanne1; Collett, Dave1; Neuberger, James1; Gimson, Alexander2 on Behalf of the United Kingdom Liver Transplant Selection and Allocation Working Party

Author Information
doi: 10.1097/TP.0b013e318225db4d
  • Free

Abstract

In both Europe and the United States, there is an increasing disparity between the number of patients currently on a liver transplant list and organs available for transplantation. Many transplant services have introduced criteria for the selection of patients to a transplant list to define those patients in whom transplantation is deemed most appropriate. Programs have also developed systems to allocate donor organs to recipients. In many programs this has been organized on a sickest first basis, therefore requiring simple, reproducible, and accurate prognostic scoring for assessing transplant list mortality and priority for transplantation. In the United States, since February 2002, the Mayo End-Stage Liver Disease (MELD) score has been used (1).

The MELD score, based on three commonly used reproducible laboratory parameters, serum bilirubin, creatinine, and international normalized ratio (INR) for prothrombin time, was originally developed to predict mortality after transjugular intrahepatic portosystemic shunt (2). A number of studies have demonstrated the accuracy of an MELD system in predicting short-term mortality risk in patients with cirrhosis (3–5). Nevertheless, one weakness of the MELD score is that it was not developed in a cohort of patients specifically selected for transplantation. Such patients may have subtly different predictors of their transplant list mortality because they represent only a small subset of all patients with cirrhosis. Furthermore, some evidence suggests that although MELD is an accurate predictor of outcome at the high end of the range, it may be a poorer predictor at lower MELD scores, in patients with noncholestatic liver disease and in those with persistent ascites (6, 7). Previous single and multicenter studies have attempted to improve the predictive power of an MELD-based score by incorporating clinical parameters such as diuretic resistance or persistent ascites or another common easily reproducible biochemical parameter. Hyponatremia is associated with a number of complications of chronic liver disease (8) and might be a factor predictive of transplant list mortality (9). The addition of serum sodium to an MELD score (MELD-Na) has improved predictive power in some but not all studies (10–12).

Our aim was to assess factors predictive of mortality in a large cohort of patients on the UK Liver Transplant list derived from a prospective multicenter database of all adult patients awaiting their first elective liver transplant. We present a new model, United Kingdom End-Stage Liver Disease (UKELD), for prediction of transplant list mortality and compare its accuracy with an MELD and MELD-Na score.

RESULTS

Patient Characteristics

The study cohort (cohort 1) comprised 1103 patients with complete data. Patient demographic characteristics of this cohort together with cohort 2 are summarized in Table 1. The most common causes were alcoholic liver disease (ALD) and hepatitis C virus (HCV). All patient characteristics, cause, and laboratory parameters were comparable across the two cohorts.

TABLE 1
TABLE 1:
Patient characteristics

Outcome of Patients on the Transplant List

Similar proportions of patients died on the transplant list or were removed from it because of their condition deteriorating, 14% and 12% in cohorts 1 and 2, respectively. Fewer patients in cohort 2 had received a liver transplant (Table 1) in the study period because of shorter follow-up time, up to 969 and 372 days in cohorts 1 and 2, respectively.

Factors Predicting Transplant List Survival

The three laboratory parameters at registration that predicted transplant list survival, adjusted for all relevant patient-specific factors, were INR, serum bilirubin, and sodium. Serum creatinine did not predict transplant list survival but was included in the model because of its clinical relevance. The median values and data ranges for each of these four variables are shown in Table 1, for cohorts 1 and 2, separately. For all variables, the distributions of the values are similar across the two cohorts.

From cohort 1, the formula for calculating the UKELD score is as follows:

where valid serum creatinine values range between 1 and 400 μmol/L, the minimum value for bilirubin is 1 μmol/L, and for INR is also 1. The sodium values range between 112 and 150 mmol/L. Values outside of these ranges are capped.

Figure 1 shows the distribution of UKELD and MELD for patients in cohorts 1 and 2. For patients in cohort 1, UKELD ranges from 40 to 79, with a median score of 55 (Fig. 1A), with a similar range for patients in cohort 2 (42–78, median score of 55, Fig. 1B). The corresponding MELD scores range from 6 to 40, median of 15 (Fig. 1C) and 6 to 37, median 16 (Fig. 1D) in cohorts 1 and 2, respectively. UKELD has a fairly symmetrical distribution, but on the same cohort of patients, the MELD score is skewed to the right.

FIGURE 1.
FIGURE 1.:
Comparison of UKELD and MELD distributions in cohort 1 (modeling dataset (A, C)) and cohort 2 (validation dataset (B, D)). UKELD and MELD scores for patients in cohort 1 (modeling dataset), including linear trend (E).

Figure 1(E) shows the relationship between UKELD and MELD for patients in cohort 1, with a line of best fit superimposed.

Patients have been divided into five UKELD bands of equal spacing (UKELD 40–47 up to UKELD 70–79), and Kaplan-Meier survivor functions are shown in Figure 2. It is clear that the survival rate on the transplant list is significantly different for patients categorized into the different UKELD bands (log-rank test: P<0.0001 for both cohort 1 (A) and cohort 2 (B)). A similar level of discrimination is seen between the bands in cohort 2.

FIGURE 2.
FIGURE 2.:
Comparison of UKELD bands in cohort 1((A) modeling dataset) and cohort 2 ((B) validation dataset). Comparison of the observed and fitted survivor functions (C).

Validation of UKELD Scores

The observed and fitted survivor functions for the 452 patients in cohort 2 are shown in Figure 2C separately for patients at a low, medium, and high risk for death. For each of the three risk groups, the model-based survivor functions track the observed survivor functions reasonably well.

Comparison of UKELD, MELD, and MELD-Na

The UKELD score had a significant effect on the hazard of death on the transplant list having adjusted for all relevant patient-specific factors. Table 2 shows the hazard ratio associated with one unit increase in the UKELD score; a unit increase in patient UKELD score increases the chance of death on the transplant list by approximately five times. In addition, Table 2 shows the comparable MELD and MELD-Na hazard ratios for the same risk-adjusted models. The Cox model that contains UKELD had the lowest value of −2 log L (1677.6), indicating that this model is a much better predictor of mortality on the transplant list compared with the model with MELD (−2 log L=1754.8) and MELD-Na (−2 log L=1703.4). The value of the Gönan and Heller (13) concordance statistic is 0.76 for the Cox model that contains UKELD and 0.72 and 0.73 for the models that contain MELD and MELD-Na, respectively, confirming the superiority of UKELD as a predictor of transplant list mortality.

TABLE 2
TABLE 2:
Comparing the influence of patient UKELD, MELD, and MELD-Na score at registration on time to death on the transplant list

The Probability of Death by Patient Score

The probability of death on the transplant list within 3 months and 1 year, calculated from the survivor function estimate at 89 and 365 days, respectively, together with 95% confidence intervals for each estimated probability are presented in Figure 3 for a range of UKELD, MELD, and MELD-Na scores. The plots illustrate that the probability of death increases with increasing UKELD score. The probability of death increases more steeply within 1 year compared with 3 months. This is the same for all three patient scores. Compared with MELD, the UKELD plots show sharper increases in the probability of death, both within 3 months and 1 year, which demonstrates that UKELD is better at discriminating between the sickest patients on the transplant list.

FIGURE 3.
FIGURE 3.:
Probability of death on the transplant list within 3 months and 1 year by patient score.

UKELD Score and Transplant List Survival

Overall, there was a significant effect of UKELD score recorded at time of registration on transplant list survival within 1 year postregistration (P<0.0001). For each of the indications, ALD (n=318), cholestatic liver disease (n=273), and HCV (n=164), there was a significant effect of UKELD score recorded at the time of registration on transplant list survival within 1 year postregistration (P<0.0001 for all three indications, separately).

UKELD Score, Survival, and Resource Utilization After Liver Transplantation

There was no significant effect of UKELD score recorded at time of transplantation on posttransplant survival at 1 year in the 842 adult patients (768 with data available) registered for their first elective liver transplant between April 1, 2003, and March 31, 2006, and were subsequently transplanted (P=0.81). When considering different indications, there was no association between UKELD and posttransplant survival for neither patients with ALD (n=199, P=0.93) nor those with cholestatic liver disease (n=140, P=0.71); however, for patients with HCV (n=140), there was some evidence of an association between UKELD and posttransplant survival (P=0.06).

Although UKELD was not associated with outcome, there was an association between UKELD and both the duration of stay in intensive care unit (n=695, P=0.003) and overall initial hospital stay (n=766, P=0.02).

DISCUSSION

In this article, we have developed a transplant list mortality risk score incorporating INR, serum creatinine, bilirubin, and sodium. In a prospective validation cohort, this predicted outcome while waiting for an adult elective liver transplant demonstrated improved predictive power compared with both MELD and MELD-Na.

A number of other authors have attempted to describe transplant list mortality scores using MELD alone or in conjunction with other parameters (14–16). The main strengths of this analysis are the size of the modeling cohort, the fact that the model was developed only in patients already selected for the transplant list and has been internally validated on a prospectively collected cohort of similar patients selected under the same criteria. Importantly, the score included analysis of the individual weights of the four component parameters rather than using MELD in its original formula. It is of importance that in the modeling cohort the weighted coefficient for serum creatinine was significantly lower than that included in MELD, an observation previously noted (17). Ruf et al. (16) did not observe any difference in serum creatinine in those who died on a transplant list within 3 months compared with those who survived. This may be because MELD was developed a number of years ago, initially in a cohort of patients following variceal hemorrhage and transjugular intrahepatic stent shunt insertion, where the effect of some parameters may be different. It may also reflect differences in the management of ascites and renal impairment on a transplant list population where stringent attempts are made not to allow increases in serum creatinine because of its known effect on posttransplant mortality (18, 19).

Although numerous studies have demonstrated the ability of the MELD score to predict transplant list mortality, some studies have exposed the limitations in the predictive power of an MELD score (7). UKELD score has again emphasized the predictive power of serum sodium on transplant list mortality. Heuman et al. (6) demonstrated that in patients with an MELD score less than 21 and persistent ascites, low serum sodium was associated with increased transplant list mortality. Ruf et al. (16) showed that addition of hyponatremia (defined as serum sodium ≥130 mmol/L) to MELD identified a subgroup of patients with a poor outcome in a more efficient way than MELD alone and increased the efficacy of the score. In their logistic regression model, hyponatremia had the highest odds ratio for predicting 3-month mortality. Biggins et al. (10) have also described a model (MELD-Na) where the addition of serum sodium to MELD was associated with improved predictive power. Neither of these two latter studies estimated the coefficients of the individual parameters of the MELD score as in our study, where UKELD had improved predictive power compared with MELD or MELD-Na.

In this study, the transplant list mortality of patients with a UKELD score between 40 and 47 was low and less than the current 1-year unadjusted mortality from elective liver transplantation in adult patients in the United Kingdom (10%; 95% confidence interval, 9%–12%) (20). Although a proportion of these patients may have been transplanted because of severe reductions in quality of life (including persistent intractable pruritus, polycystic liver disease, and amyloidosis), it emphasizes the importance of minimum listing criteria before selection onto a transplant list for elective liver transplantation. A UKELD score of less than 49 in patients without a hepatocellular carcinoma represents an expected 1-year mortality of 9% and might be used as a suitable level. Similarly, an MELD score of less than 10 has been associated with a low transplant list mortality and has been recommended as a minimum criterion.

Although transplant list mortality scores such UKELD may be developed to inform decisions about which cases to select for a transplant list, it does not follow that they are necessarily the best measure to inform the allocation of donor organs to transplant list candidates. The effect of the parameters of both MELD and UKELD on posttransplant survival is complex. Although serum creatinine may have only a minor effect on transplant list mortality, it has a more significant effect posttransplant. The effect of hyponatremia on mortality up to 1 year after transplantation was significant in a large national cohort; effect of serum sodium pretransplant may not be reflected posttransplant (21), although not observed in other studies (22).

Certain limitations apply to the interpretation of the UKELD score. First, UKELD has been developed in a cohort where patients with hepatocellular cancer have been excluded. The principles behind their selection to a transplant list and prioritization for a transplant in the United Kingdom are different and include risk of both short-term mortality and tumor recurrence after transplantation. Second, concerns over standardization of serum creatinine (23) and INR (24) estimations and whether female transplant list candidates are prejudiced because of a lower population mean serum creatinine level remain with all such scoring systems. Third, this score has been developed with artificial high and low cutoff values for the parameters. There are arguments both for and against such a device. In developing the MELD-Na score, Biggins et al. (10) argued in favor to avoid excess weighting to low sodium or high serum creatinine, which may significantly affect posttransplant outcome. Because that may be most appropriate when developing a score for prioritization, we have capped values when developing a score predominantly for transplant list selection. Finally, patients on a transplant list may be removed for a number of reasons other than death. If those patients are removed for transplantation and one criterion for the removal is severity of liver disease and mortality risk, undue bias in estimating true transplant list mortality may occur (25). Allowance may then need to be made for the fact that censoring at time of transplant is informative about subsequent survival time. In this analysis, the four laboratory parameters were obtained at the time of registration. Using data subsequent to registration may demonstrate different weighted coefficients. However, whether the use of MELD scores at subsequent time points after registration or the change in MELD (delta-MELD) will significantly improve transplant list mortality prediction remains controversial (7, 26, 27).

MATERIALS AND METHODS

Study Population

Data for this study were obtained from the UK Transplant Registry maintained by National Health Service Blood and Transplant, on behalf of transplant services in the United Kingdom and Republic of Ireland. These registry data have previously been described and validated.

This study is based on data from 1555 adult patients registered for their first elective liver transplant between April 1, 2003, and March 31, 2007. Patients with any type of cancer recorded as their primary liver disease (n=86) were excluded. A modeling dataset (cohort 1) comprised 1103 adult patient registrations at the seven liver transplant centers in the United Kingdom between April 1, 2003, and March 31, 2006. A validation dataset, which will be referred to as cohort 2, comprised 452 adult patients registered for their first elective liver transplant from all seven liver transplant centers in the United Kingdom between April 1, 2006, and March 31, 2007.

During this time period, patients were selected for the elective liver transplant list if they had decompensated chronic liver disease where the survival without transplantation was considered to be less than 1 year and a greater than 50% predicted survival with transplantation at 5 years.

Once included on the transplant list, patients where managed by experienced hepatologists within the seven liver transplant centers according to recognized protocols. Donor organs were allocated to transplant centers proportionate to the size of the annual transplant volume. Within each center, clinicians were able to select the recipient they believed most appropriate to receive that specific donor organ, considering both the quality of the donor organ and severity of liver disease of the recipient. Selection and allocation procedures within centers during the time of this study did not considerably change.

End Points of the Study and Definitions

Patient demographics, primary liver disease and laboratory values available at the time of registration, together with year of registration, were considered in the analysis. The patient demographics were age, sex, ethnicity, blood group, body mass index, height, and weight, and the laboratory values were INR, serum creatinine (μmol/L), bilirubin (μmol/L), and sodium (mmol/L). Laboratory tests were those taken on the day of registration. Patients were followed up until death, liver transplantation, or last known follow-up before April 10, 2007. Patient survival time was defined as the active time spent on the transplant list with patient death while on the list treated as an event. Registrations that resulted in the removal from the transplant list for reasons defined collectively as condition deteriorated were included as deaths on the list, and the removal date taken as the death date. The survival times of patients who received a transplant or were removed from the list for reasons other than condition deteriorated were censored at their date of transplant or removal, respectively. Patients who were suspended from the list were censored at the date of suspension, and patients remaining on the list were censored at the time of analysis.

The MELD score was calculated according to Kamath et al. (3). The following formula was used:

where serum creatinine (mg/dL)=serum creatinine (μmol/L)/88.4 and bilirubin (mg/dL)=bilirubin (μmol/L)/17.1.

MELD scores were rounded to the nearest whole number and ranged from a minimum of 6 to a maximum of 40. INR had a minimum value of 1, serum creatinine and bilirubin had a minimum value of 1 mg/dL, and serum creatinine had a maximum value of 4 mg/dL. We also assessed the MELD-Na score (10) defined as:

where the minimum and maximum values for sodium were 125 and 140 mmol/L, respectively, and MELD was the standard MELD score calculated without upper and lower limits on the input variables or an upper limit of MELD score at 40.

Statistical Methods

Summary statistics for continuous variables and frequency counts and percentages for categorical variables were calculated where appropriate. Percentages were calculated out of nonmissing values.

Survival rates were obtained using the Kaplan-Meier estimate. A Cox proportional hazards model was used to assess the effect of prognostic and clinical factors on the risk of death on the transplant list. A complete case analysis was performed. Transplant center was included in the model as a fixed effect.

The clinical components considered for inclusion in the UKELD score, namely INR, serum creatinine (μmol/L), bilirubin (μmol/L), and sodium (mmol/L), were added to the model that contained relevant patient-specific factors. A model containing logarithms of the clinical components was also considered and from the value of the log likelihood statistic was found to provide a better fit.

The UKELD score was expressed as a linear combination of the logarithms of the clinical components. The estimated coefficients from the fitted model, adjusted for relevant patient-specific factors, were rounded to three decimal places, and the score rescaled and relocated to give positive scores over a suitably wide range of values.

The distributions of the MELD and UKELD scores were compared using histograms.

To compare the predictive ability of UKELD, MELD, and MELD-Na in models for the hazard of death while waiting for a transplant, Cox models were fitted that included all relevant patient-specific factors together with UKELD, MELD, or MELD-Na. The log likelihood statistic (−2 log L), a measure of agreement between the fitted model and the observed data, was then used to compare the three models. The predictive ability of these three different Cox models was also compared using a concordance statistic of Gönan and Heller (13), which can vary from 0.50 (no predictive ability) to 1.00 (perfect predictive ability).

The relationship between the unadjusted probability of death within 3 months and 1 year and UKELD was obtained from fitting a Cox proportional hazards model with UKELD alone.

The effect of UKELD recorded at time of registration on transplant list survival was assessed overall and for selected indications.

The effect of UKELD recorded at time of transplantation on posttransplant survival was assessed overall and for selected indications. The effect of UKELD on both the duration of stay in an intensive care unit and initial hospital stay was also investigated.

Validation

The UKELD score was validated using the previously defined cohort 2 (validation dataset).

The patients, cohort 2, were divided into three groups based on their risk score, corresponding to the 33rd and 66th percentiles, with patients at a high, medium, and low risk for death on the transplant list. The Kaplan-Meier estimate of the survivor function for each group was then compared with the model-based estimates of the survivor function, adjusted for patient specific-factors (28).

ACKNOWLEDGMENT

United Kingdom Liver Transplant Selection and Allocation Working Party members: Alexander Gimson (Addenbrooke's Hospital, Cambridge); David Mutimer (Queen Elizabeth Hospital, Birmingham); John O'Grady (King's College Hospital, London); Andrew Burroughs (Royal Free Hospital, London); Mervyn Davies (St James' Hospital, Leeds); Derek Manas (The Royal Victoria Hospital, Newcastle); Mark Hudson (The Royal Victoria Hospital, Newcastle); and Murat Akyol (Royal Infirmary, Edinburgh).

REFERENCES

1. Wiesner R, Edwards E, Freeman R, et al. United Network for Organ Sharing Liver Disease Severity Score Committee. Model for end-stage liver disease (MELD) and allocation of donor livers. Gastroenterology 2003; 124: 91.
2. Malinchoc M, Kamath PS, Gordon FD, et al. A model to predict poor survival in patients undergoing transjugular intrahepatic portosystemic shunts. Hepatology 2000; 31: 864.
3. Kamath PS, Wiesner R, Malinchoc M, et al. A model to predict survival in patients with end-stage liver disease. Hepatology 2001; 33: 464.
4. Said A, Williams J, Holden J, et al. Model for end-stage liver disease score predicts mortality across a broad spectrum of liver disease. J Hepatol 2004; 40: 897.
5. Forman LM, Lucey MR. Predicting the prognosis of chronic liver disease: An evolution from child to MELD. Mayo End-stage Liver Disease. Hepatology 2001; 33: 473.
6. Heuman DM, Abou-Assi SG, Habib A, et al. Persistent ascites and low serum sodium identify patients with cirrhosis and low MELD scores who are at high risk for early death. Hepatology 2004; 40: 802.
7. Cholongitas E, Marelli L, Shusang V, et al. A systematic review of the performance of the model for end-stage liver disease (MELD) in the setting of liver transplantation. Liver Transplantation 2006; 12: 1049.
8. Arroyo V, Rodés J, Gutierrez-Lizarriaga MA, et al. Prognostic value of spontaneous hyponatremia in cirrhosis with ascites. Am J Dig Dis 1976; 21: 249.
9. Biggins SW, Rodriguez HJ, Bacchetti P, et al. Serum sodium predicts mortality in patients listed for liver transplantation. Hepatology 2005; 41: 32.
10. Biggins SW, Kim WR, Terrault NA, et al. Evidence-based incorporation of serum sodium concentration into MELD. Gastroenterology 2006; 130: 1652.
11. Kim WR, Biggins SW, Kremers WK, et al. Hyponatremia and mortality among patients on the liver-transplant waiting list. N Engl J Med 2008; 359: 1018.
12. Londoño MC, Cárdenas A, Guevara M, et al. MELD score and serum sodium in the prediction of survival of patients with cirrhosis awaiting liver transplantation. Gut 2007; 56: 1283.
13. Gönan M, Heller G. Concordance probability and discriminations power in proportional hazards regression. Biometrika 2005; 92: 965.
14. Hsu CY, Lin HC, Huang YH, et al. Comparison of the model for end-stage liver disease (MELD), MELD-Na and MELD-Na for outcome prediction in patients with acute decompensated hepatitis. Dig Liver Dis 2010; 42: 137.
15. Mathur S, Gane EJ, McCall JL, et al. Serum sodium and hydration status predict transplant-free survival independent of MELD score in patients with cirrhosis. J Gastroenterol Hepatol 2008; 23: 239.
16. Ruf AE, Kremers WK, Chavez LL, et al. Addition of serum sodium into the MELD score predicts waiting list mortality better than MELD alone. Liver Transplantation 2005; 11: 336.
17. Sharma P, Schaubel DE, Sima CS, et al. Re-weighting the model for end-stage liver disease score components. Gastroenterology 2008; 135: 1575.
18. Nair S, Verma S, Thuluvath PJ. Pretransplant renal function predicts survival in patients undergoing orthotopic liver transplantation. Hepatology 2002; 35: 1179.
19. Bahirwani R, Campbell MS, Siropaides T, et al. Transplantation: Impact of pretransplant renal insufficiency. Liver Transplantation 2008; 14: 665.
20. One year unadjusted mortality estimates from elective liver transplantation in adult patients. Available at: http://www.organdonation.nhs.uk/ukt/statistics/centre-specific_reports/centre-specific_reports.jsp. Accessed October 13, 2010.
21. Dawwas MF, Lewsey JD, Neuberger JM, et al. The impact of serum sodium concentration on mortality after liver transplantation: A cohort multicenter study. Liver Transplantation 2007; 13: 1115.
22. Yun BC, Kim WR, Benson JT, et al. Impact of pretransplant hyponatremia on outcome following liver transplantation. Hepatology 2009; 49: 1610.
23. Goulding C, Cholongitas E, Nair D, et al. Assessment of reproducibility of creatinine measurement and MELD scoring in four liver transplant units in the UK. Nephrol Dial Transplant 2010; 25: 960.
24. Porte RJ, Lisman T, Tripodi A, et al.; Coagulation in Liver Disease Study Group. The International Normalized Ratio (INR) in the MELD score: Problems and solutions. Am J Transplant 2010; 10: 1349.
25. Kim WR, Therneau TM, Benson JT, et al. Deaths on the liver transplant waiting list: An analysis of competing risks. Hepatology 2006; 43; 345.
26. Bambha K, Kim WR, Kremers WK, et al. Predicting survival among patients listed for liver transplantation: An assessment of serial MELD measurements. Am J Transplant 2004; 4: 1798.
27. Merion RM, Wolfe RA, Dykstra DM, et al. Longitudinal assessment of mortality risk among candidates for liver transplantation. Liver Transplantation 2003; 9: 12.
28. Collett D. Modelling survival data in medical research [ed. 2]. London: Chapman and Hall/CRC; 2003.
Keywords:

Liver transplantation; Transplant list mortality; Allocation; UKELD score

© 2011 Lippincott Williams & Wilkins, Inc.