A systematic review of the performance of the model for end-stage liver disease (MELD) in the setting of liver transplantation
The Model for End-Stage Liver Disease (MELD) score is now used for allocation in liver transplantation (LT) waiting lists, replacing the Child-Turcotte-Pugh (CTP) score. However, there is debate as whether it is superior to CTP score to predict mortality in patients with cirrhosis on the LT waiting list and after LT. We reviewed studies comparing the accuracy of MELD vs. CTP score in transplantation settings. We found that in studies of the LT waiting list (12,532 patients with cirrhosis), only 4 of 11 showed MELD to be superior to CTP in predicting short-term (3-month) mortality. In addition, 2 of 3 studies (n = 1,679) evaluating the changes in MELD score (ΔMELD) showed that ΔMELD had better prediction for mortality than the baseline MELD score. The impact of MELD on post-LT mortality was assessed in 15 studies (20,456 patients); only 6 (9,522 patients) evaluated the discriminative ability of MELD score using the concordance (c) statistic (the MELD score had always a c-statistic < 0.70). In 11 studies (19,311 patients), high MELD score indicated poor post-LT mortality for cutoff values of 24-40 points. In re-LT patients, 2 of 4 studies evaluated the discriminative ability of MELD score on post-LT mortality. Finally, several studies have shown that the predictive ability of MELD score increases by adding clinical variables (hepatic encephalopathy, ascites) or laboratory (sodium) parameters. On the basis of the current literature, MELD score does not perform better than the CTP score for patients with cirrhosis on the waiting list and cannot predict post-LT mortality. Liver Transpl 12:1049–1061, 2006. © 2006 AASLD.
Decompensated cirrhosis has a poor prognosis, and liver transplantation (LT) is the only curative therapy, giving excellent long-term survival.1 The Child and Turcotte classification (1964) and the Pugh's modification (1973) (Child-Turcotte-Pugh [CTP] score)2 were originally devised for the assessment of the severity of liver disease in predicting the outcome of patients with cirrhosis in whom surgical therapy for portal hypertension was planned. It was then extended for endoscopic or transjugular intrahepatic portosystemic shunt therapy, for prognosis in general, and more recently to stratify patients on the waiting list for LT.1
However, the use of CTP, particularly for prioritizing potential liver transplant recipients, has revealed several drawbacks.1, 3 First, 2 of the variables, ascites and encephalopathy, are subjective because they are assessed by physical examination alone; and second, when other methods are used (ultrasonography, psychometric testing, EEG), a different degree of severity is diagnosed. Both ascites and portosystemic encephalopathy are influenced by therapy such as diuretics, albumin, and lactulose. In addition, the “ceiling” and “floor” effect in terms of the limits set to the laboratory parameters of bilirubin, albumin, and prothrombin time in the grades A, B, and C and changes of serum bilirubin concentrations with therapy (e.g., with ursodeoxycholic acid) do not allow assessment using a continuous scale of severity. Moreover, the absence of an assessment of renal function, which is a well-established prognostic marker in cirrhosis,3 is another limitation of the CTP score.
The Model for End-Stage Liver Disease (MELD) is a new scoring system for severity of liver disease.4 In contrast to CTP score, it has been validated statistically. This model, published in 2000, was developed as “a model to predict poor survival in 231 patients who had undergone elective transjugular intra-hepatic porto-systemic shunt (TIPS).”5 In this study, serum bilirubin, serum creatinine, the international normalized ratio (INR) of prothrombin time and the underlying liver disease were the independent factors for short-term (3-month) survival, which were combined in a new prognostic score.
In 2001, the same group slightly modified this score to predict mortality of patients with cirrhosis with different causes and severity of liver disease.6 The MELD score was derived by multiplying the original score by 10, then rounding up to the nearest integer. The cause of liver disease was dropped (but the coefficient of this variable has remained in the formula, i.e., 6.4). The MELD score formula is: R = 9.6 × loge (creatinine mg/dL) + 3.8 × loge (bilirubin mg/dL) + 11.20 × loge (INR) + 6.4. The predictive ability of the MELD score was evaluated using the concordance c-statistic (i.e., the area under the receiver-operator characteristic curve, which ranges from 0 to 1, with 0.5 corresponding to what is expected by chance alone and 1.0 to perfect discrimination). In general, a c-statistic >0.70 indicates a useful test, whereas a value >0.80 implies excellent diagnostic accuracy, but still means that there will not be an accurate prediction in approximately 20% of occasions. In this study,6 MELD score, almost always gave a c-statistic for 3-month survival >0.80 in all groups of patients with cirrhosis, without any significant improvement by adding complications such as ascites, encephalopathy, variceal bleeding, and spontaneous bacterial peritonitis. The authors concluded that MELD score was superior to CTP score and proposed MELD score for prioritization of recipients for LT, a proposal that was adopted in the United States in February 2002.
However, the MELD score is not without limitations. Recent studies have revealed marked variations in serum creatinine measurement when different laboratory methodologies are used. It has been recognized for many years that bilirubin is one of the major chromogens, which gives negative interference when measuring serum creatinine. Various methods for creatinine measurement have been introduced by laboratories to overcome this interference, but there is little concordance between different assays and no accepted consensus on the best method.7 In addition, it is known that serum creatinine is also influenced by gender, age, ethnicity, and muscle mass, which may lead to discrimination against women, white, or malnourished patients with cirrhosis if creatinine is considered as a surrogate of renal function. Moreover, INR was designed to standardize the anticoagulation effect of warfarin and may not reflect the severity of liver disease.8 Trotter et al.9, 10 have shown in 2 studies that different assays to measure INR lead to significantly different MELD scores between transplant centres. These differences were more pronounced at higher MELD score, i.e., in patients with higher priority for LT.
In addition, the MELD prognostic score has never been formally assessed in terms of calibration—that is, whether it can predict equally well across the range of scores. Generally, it has been recognized that new statistical techniques are needed for the assessment of the accuracy prognostic scores for individual patients. The c-statistic is not the only parameter.11
Several studies have now been published that evaluate results after the implementation of MELD and compare these with the previous allocation system, which is based on CTP score and waiting time. In this review, we critically evaluate MELD score in the liver transplant settings in the available literature in order to assess the prognostic ability of MELD score compared with CTP score in patients with cirrhosis with or without hepatocellular carcinoma (HCC) on the transplant waiting list for a primary or re-LT; and to assess the use of the MELD score before LT for predicting short- and long-term post-LT survival.
MELD FOR PREDICTING MORTALITY WHILE ON LT WAITING LISTS
The continuing shortage of liver donors unfortunately results in some patients dying while on the waiting list. Thus, criteria for organ allocation are needed to establish a justice system and to improve utility—that is, to maximize better outcomes. The United Network for Organ Sharing (UNOS) has revised allocation and distribution policies on the basis of the ethical principles of justice for the individual patients vs. optimal use of the limited number of available organs. The MELD score was adopted for determining disease severity in order to allocate donor organs to the sickest patients first, rather than to those who had been on the waiting list longest (Table 1). The system has been in operation from February 27, 2002, in the United States with staged evolution after close audit and validation.
Table 1. Predictive Value of MELD and CTP Scores for 3-Month Survival in Patients With Cirrhosis on the Waiting List for Liver Transplants
|Wiesner et al.13||3,437*||0.83||0.76||<0.001|
|Kim et al.14||706||0.85||0.72||<0.01|
|Huo et al.15†||58||0.718||0.528||<0.001|
|Heuman and Mihas16||6,958‡||0.759||0.766||NS|
|Foxton et al.17||376||0.79||0.81||NS|
|Burak et al.18||320||0.79||0.77||NS|
|Abuassi et al.19||140||0.66||0.66||NS|
|Llado et al.21||70||0.75||0.78||NS|
Wiesner et al.12 applied the MELD model to estimate 3-month mortality in a group of 311 pre-LT patients with chronic liver disease on the UNOS waiting list. The mean MELD score for 26 patients, listed at 2A according to UNOS, was 27, and the score was 14 for the 285 patients listed as 2B. Within 3 months after listing, patients with a MELD score >18 experienced a 29% mortality rate compared with 6% in patients with a MELD score of <18 (P < 0.01). Evaluating 3-month mortality, the c-concordance statistic in the 311 patient cohort was 0.82. By comparison, the c-statistic for CTP score for 3-month mortality was 0.73.
A prospective study13 of 3-month mortality included 3,437 adult pre-LT patients with chronic liver disease on the UNOS waiting list (2A or 2B status, listed between November 1999 and December 2001), who had a mean MELD score of 28 in 536 patients listed as status 2A, and 18.3 for 2,901 patients listed as 2B. Within 3 months after listing, 412 (12%) had died, and 3-month mortality was greater in patients with higher MELD score (e.g., patients with MELD score <9 experienced a 1.9% mortality rate compared with 71.3% in patients with a MELD score >40). The c-statistic for 3-month mortality was 0.83 (95% confidence interval [95% CI] 0.81-0.94) for the MELD score, compared with 0.76 (95% CI 0.74-0.79) for the CTP score (P < 0.001).
A separate cohort of 706 patients waiting for LT was evaluated for factors associated with risk of death.14 MELD had consistent predictability across time (odds ratio [OR] 1.2 for 1990-1994; OR 1.3 for 1995-1999; both P < 0.01). When blood group (0 vs. other) and the time interval (1990-1994 vs. 1995-1999) were incorporated in the model, the effect of MELD remained unchanged (OR 1.2, P < 0.01). When CTP substituted MELD, there was also a significant association with 3-month mortality (OR 1.3, P < 0.01), but MELD had a significantly higher discriminating power than CTP score (c-statistic 0.85 vs. 0.72, P < 0.01). Recently, Huo et al.15 prospectively evaluated 58 patients and found that the baseline MELD score, compared with CTP score, was better for predicting 6-month mortality (c-statistic 0.718 vs. 0.528) and 12-month mortality (0.744 vs. 0.528) (P < 0.001).
However, in contrast to the 4 studies above, 7 others have failed to confirm the superiority of MELD score compared with CTP score for patients on the waiting list. Heuman and Mihas.16 analyzed the largest cohort of 6,958 patients with cirrhosis with complete concurrent data, listed between May 2001 and November 2001 as status 2A, 2B, and—in contrast to the study by Wiesner et al.13—also status 3 on the UNOS waiting list. A total of 306 died and 706 were withdrawn or underwent LT within 3 months. Although the authors extracted their data from the same UNOS database and for overlapping periods as did Wiesner et al., they found that the c-statistic for CTP score was slightly better, but not significantly different than, MELD score as a predictor of 3-month survival (0.766 ± 0.032 vs. 0.759 ± 0.034). This could be because of the inclusion of status 3 patients, but this would imply that the MELD score does not have the same prognostic ability when cirrhosis is not severe.
Their findings were confirmed by 2 recent studies. Foxton et al.17 found that in 376 candidates for LT, MELD score had a good predictive ability for 3-month mortality on the waiting list, but again was not superior to CTP score (c-statistic 0.79 and 0.81, respectively). Similarly, Burak et al.18 validated the MELD score in 320 candidates for LT in Canada, including patients with acute liver failure and HCC. The MELD score as well as CTP score had better discriminative ability for both 3-month mortality (c-statistic 0.79 and 0.78, respectively) and 12-month mortality (0.69 and 0.69, respectively) compared with the Canadian allocation system (c-statistic at 3 months, 0.58; at 12 months, 0.55). Interestingly, in both studies, MELD score had poor predictive ability for post-LT mortality.
In addition, Abuassi et al.,19 in a cohort of 140 U.S. veterans, examined the accuracy of MELD over 120, 180, 365, or 730 days. For the first 120 days, MELD was slightly more, but not significantly, accurate than CTP in predicting mortality. After 120 days, MELD and CTP scores were equally good predictors of survival. Baseline MELD values >7 and >13 were equivalent to CTP scores of ≥9 and ≥12. Subsequently, Adler et al.,20 in a preliminary report, did not find any significant difference between MELD and CTP scores to predict mortality on the liver waiting list in 124 patients. Moreover, Llado et al.21 retrospectively compared a small group of patients who died while on the waiting list (group D) with a group of patients who successfully underwent LT (group T) during the same time period. They examined 4 scores in 70 patients: CTP, MELD, Freeman scale, and Guardiola index. All the scores, except Freeman, were significantly higher in group D at the time of placement on the waiting list; their c-statistics were similar and in all cases <0.8, so none predicted particularly well while the patients waited for LT. Finally, Harrer et al.22 examined the accuracy of CTP or MELD score to predict survival in 32 patients who either died while waiting (n = 12) or who were not transplanted (n = 20). At listing, the MELD score (cutoff MELD score >18) did not predict reduced survival.
Accuracy of Changes in MELD Score for Predicting Mortality of Patients on LT Waiting Lists
Three studies have evaluated the change in MELD (ΔMELD) score in large cohorts of candidates on the LT waiting list. In a retrospective evaluation of 760 patients,23 the ΔMELD score had better prediction for mortality than the baseline MELD score. For example, a worsening of MELD score by >5 points during a 30-day period predicted a significantly increased risk of death (P < 0.0001). On this basis, Merion et al.23 suggested the ΔMELD score as a tiebreaker for patients on a waiting list with identical MELD scores. However, Bambha et al.24 retrospectively evaluated 861 patients and found ΔMELD score to be less predictive compared with the most up-to-date MELD score after adjustment with the number of serial MELD scores obtained for each patient and at the time of death. They claimed that Merion et al.23 had overestimated the prognostic ability of ΔMELD for 2 reasons: first, an increasing MELD score (positive ΔMELD) actually represented the dying process per se; and second, sicker patients had more frequent MELD assessments. The authors concluded that the current UNOS policy to use the most recent MELD for allocation is the proper one.
However, Huo et al.15 prospectively evaluated 58 patients and analyzed the risk factors for ΔMELD score at 3 months. In addition, they retrospectively compared ΔMELD, baseline MELD, and CTP in 351 patients. The c-statistic for ΔMELD per month at 6 and 12 months was 0.779 (95% CI 0.73-0.82) and 0.822 (95% CI 0.77-0.86), respectively, which was significantly better compared with baseline MELD (0.718 and 0.744, respectively) and CTP (0.528 for both periods). Interestingly, Huo et al. found in the prospective arm, that ΔMELD was independently associated only with the development of ascites (P = 0.005) and encephalopathy (P = 0.003), both prognostic variables in the CTP score. Despite this, baseline MELD was significantly better than CTP (P < 0.001). In another report, they found that the presence of complications of cirrhosis had similar predictive ability compared with baseline MELD score, and the baseline MELD score was similar between the groups with or without complications (11.6 vs. 12.2, P = 0.184), but precise data were not provided.25
Thus, considering 11 studies published to date, only 4 demonstrate a statistical superiority of MELD compared with CTP score (4,512 patients), whereas 7 show no statistical difference (8,020 patients). However, no study showed MELD to be statistically inferior to CTP score.
MELD SCORE BEFORE TRANSPLANTATION AS PROGNOSTIC FACTOR FOR POST-LT SURVIVAL
An ideal liver allocation model should not only allocate livers to patients with the highest probability of dying before LT, but should also be able to predict which candidates have the lowest post-LT predicted mortality in order to improve utility (Table 2). Pre-LT MELD score is a good predictor for survival on the waiting list, but its ability to predict post-LT survival is not well established. This is an important issue, because if MELD score could predict (at the same time) the patients with the highest post-LT mortality, this would strengthen the use of MELD as an allocation system for recipients. On the other hand, if higher MELD scores are associated with lower post-LT survival, this would worsen utility.
Table 2. Predictive Value of MELD Score for Post-LT Survival in Patients With Cirrhosis Who Received Liver Transplants
|Kim et al.26*†||1,185||0.62||Yes||<10 = 5%, >40 = 26%||MELD related with post-LT ICU and hospital stay|
|Wiesner et al.27*†||1,585||<0.70||Yes||…||CTP predicted post-LT mortality|
|Lally et al.28†||376||…||No||…||CTP did not predict post-LT mortality|
|Brown et al.29†||42||…||No||…||CTP did not predict post-LT mortality|
|Onaca et al.30||669||…||Yes||…||Cholestatic cirrhotics may receive less points|
|Saab et al.32||404||…||Yes||24 points||UNOS status just missed statistical significance|
|Yoo and Thuluvath33‡||3,227||…||Yes||30 points||18.3% had MELD >30 points|
|Kanwal et al.35§||4,245||…||Yes||…||No—measurable factors may affect post-LT mortality|
|Desai et al.36‖||2,565||0.54||Yes||24.45 points||…|
|Jacob et al.37||3,838||0.58||Yes||36 points||…|
|Bazarah et al.38||228||0.67||No||…||Multivariate regression (3 month): can WAIT and age (not CTP, MELD)|
|Habib et al.40||1,472||…||Yes||26 points||Recipient and donor age, gender, MELD, and re-LT were associated with poorer post-LT survival|
|Nagler et al.41∥||121||<0.61||Yes||30 points||Only 2 patients had MELD> 30 points|
|Island et al.42||…||…||Yes||30 points||Significant difference in hospital charges between group with high and low MELD score|
|Narayanan Menon et al.43‡||499||…||No||…||CTP did not predict post-LT mortality|
MELD and Short-Term Post-LT Mortality
Preliminary reports have given contrary results. In a cohort of 1,185 patients,26 post-LT outcome and resource utilization was progressively worse as the pre-LT MELD score increased. Thus, for MELD <10 and MELD >40, respectively, the death rate was 5% vs. 26%, graft failure was 7% vs. 33% at 3 months, the red blood cell transfusion requirement was 1.5 vs. 3.7, length of stay in the intensive care unit was 2 vs. 6 days and hospital stay was 14 vs. 19 days. The same group27 also compared the accuracy of pre-LT MELD, the CTP, and the Mayo LT Outcome Model in the prediction of post-LT outcome in 1,585 patients. The Mayo LT Outcome Model incorporates age, renal failure, CTP class, and patients status (home, hospitalized, or intensive care unit). Although in both studies MELD score had statistically significant predictive ability for post-LT outcome, its c-statistic for post-LT survival was low (c-statistic 0.62 and < 0.70, respectively), and thus was not clinically useful.
On the other hand, Lally et al.28 used MELD, CTP, and a region 1 Continuous Medical Severity Score (CMSS) (based on the Child score and associated medical condition) to predict post-LT survival in 376 patients. Post-LT survival at 1, 3, 6, 12, 24, and 36 months was not associated with any of the scores. Brown et al.29 examined 42 consecutive patients of 2A UNOS status undergoing LT for nonfulminant liver disease. Neither MELD nor CTP score was predictive of survival after LT (MELD OR 1.00, 95% CI 0.88-1.12, P = 0.99; CTP OR 1.49, 95% CI 0.65-3.41, P = 0.34), surgical time, or intraoperative transfusion requirement, although MELD score, but not CTP score, was predictive of pre-LT mechanical ventilation.
To date, the impact of pre-MELD score to predict post-LT survival has only been evaluated in a few large cohorts. In 669 consecutive patients,30 stratification according to pre-LT MELD score was made into 3 groups: <15, 15-24, and ≥25. Post-LT survival was lower with MELD ≥25 compared with MELD <15 at 3, 6, 12, 18, and 24 months after LT (P < 0.001 to 0.031). These post-LT survival differences, according to the pre-MELD score, remained significant (almost at all intervals) in the subgroup of 268 patients with hepatitis C virus (HCV)-associated cirrhosis and 263 patients with noncholestatic and nonviral cirrhosis, but not among the 115 patients with cholestatic cirrhosis. The latter patients had higher post-LT survival compared with those with HCV with the same MELD score (especially for patients with MELD score ≥15). This may be due to the higher bilirubin values in cholestatic cirrhosis (without parallel worsening of other measures of liver function). Thus, cholestatic patients with cirrhosis could be considered for receiving fewer points on the waiting list because they have better survival outcomes after LT.
The same authors31 further evaluated 287 patients with HCV with cirrhosis. The immediate pre-LT MELD score was <15 in 165, 15-24 in 86, and ≥25 in 36. Higher pre-LT MELD scores correlated with lower post-LT survival. In addition, the differences in survival for patients without bridging necrosis with pre-LT MELD score <15, 15-24, and ≥25 at 3 months and beyond were significant (e.g., for 3 months, 93.3%, 82.5%, and 80.5%, respectively, P = 0.012). There was no association with rejection rates or recurrence of HCV infection, and no cutoff of MELD score beyond which there was excessive increase in post-LT mortality. However, Saab et al.32 studied 404 patients of status 2A (23.8%), 2B (35.4%), and 3 (40.8%) and found that pre-LT MELD score >24 was associated with a significantly lower 1-year post-LT survival compared with a MELD score <24 (65% vs. 88%, P < 0.0001). Moreover, the overall 1-year survival was lower in patients with higher MELD scores immediately before LT (P = 0.0006). The difference in 1-year post-LT survival, using UNOS status, just missed statistical significance (2A, 78%; 2B, 87%; 3, 88%, P = 0.051
The last 3 studies29, 31, 32 used the UNOS database for overlapping periods of time and thus are not truly independent of each other. First, Yoo and Thuluvath33 compared the post-LT mortality between the patients who underwent LT before (n = 14,593, pre-MELD group) and after (n = 3,227, MELD group) the implementation of MELD score for liver allocation. No difference was found in 10-month patient and graft survival between MELD and pre-MELD groups. Patients with MELD score at LT ≥30 had significantly lower survival, compared with patients with MELD score < 30. Moreover, MELD score ≥30 was an independent predictor of post-LT mortality (hazard ratio [HR] 2.9, P < 0.001). Creatinine was the only component of MELD score significantly associated with post-LT mortality (HR 1.16, 95% CI 1.01-1.32), something that has been recognized for many years now.34 Interestingly, although MELD score was adopted to allocate the candidates according to the severity of their liver disease, only 18.3% of the patients in the MELD group had MELD scores >30 points at LT.
Kanwal et al.35 analyzed the UNOS database and again compared a pre-MELD LT group (n = 3,857) and post-MELD group (n = 4,245). Although the MELD score at LT was an independent factor for post-LT survival (HR 1.033, 95% CI 1.02-1.04, P < 0.0001) and the mean MELD score was significantly higher in post-MELD group compared with pre-MELD group (20.5 vs. 17, P < 0.0001), the authors found only a 3% drop in 1-year post-LT mortality in the post-MELD era. This post-LT survival difference could be influenced by other nonmeasurable factors and factors that did not change with the new allocation policy. On the other hand, the difference in severity of liver disease may be statistically significant but not clinically relevant. Finally, similar to the previous study,33 only a small proportion of the patients who received LT had MELD scores >30 points, so another interpretation is that the effect of pre-LT severity of liver disease may not be evident with MELD <30.
Desai et al.36 evaluated 2,565 patients for whom there was available INR. There were no significant differences in pre-LT variables between this group compared with the group without available INR (8,983 patients). However, more patients were UNOS status 2A in the first group, which could account for a lower 1-year survival compared with the second group (P = 0.002). Post-LT hospital stay and 1-year post-LT survival, according to the pre-LT MELD score, were significantly worse only for patients with the highest MELD score quintile (>24.45), but the predictive value of pre-LT MELD score for 3-month post-LT survival was very low (c-statistic 0.54). These data suggest that only high MELD scores affect post-LT survival and that MELD is not a good predictor of outcome after LT. However, because the sickest patients receive the highest priority, it is important to examine independent cohorts of patients to quantify the relationship because issues of utility (i.e., best outcome) are in particularly relevant in this group. In addition, the lack of predictive capacity of MELD also reflects the fact that donor and operative factors are not considered together with recipient characteristics. An external validation of MELD score in 3,838 elective transplant patients in the United Kingdom and Ireland showed that pre-MELD score did not accurately predict 3-month post-LT mortality (c-statistic 0.58).37 In this study, only patients with pre-MELD ≥36 had significantly post-LT mortality compared with those with pre-MELD <36. In addition, a Canadian study38 retrospectively analyzed 228 LT patients and concluded that the Canadian wait listing algorithm in LT (CanWAIT) status, MELD, and CTP scores had equal predictive ability for 3-month post-LT survival (area under the receiver-operator characteristic curve, 0.71, 0.67, and 0.65, respectively).
MELD and Short-Term Post-LT Mortality in Fulminant Hepatic Failure
Although all the previous studies evaluated short term post-LT mortality in UNOS status non-1 candidates, Kremers et al.39 retrospectively studied exclusively nonalcoholic patients (n = 720) classified as UNOS status 1, in order to assess the accuracy of MELD score to predict pre-LT and short-term post-LT survival and to identify subgroups of these patients with the greatest benefit associated with LT. Patients were classified into 4 categories: fulminant hepatic failure (FHF) due to acetaminophen (FHF-A) (12%), FHF without acetaminophen toxicity (FHF-NA) (43.3%), primary graft failure (PNF) (37.2%) and hepatic artery thrombosis within 7 days after transplantation (HAT) (8%). The FHF-NA group had the worst 30-day waiting list survival rate, and MELD score was the only significant predictor of mortality (HR 1.048, P = 0.0001). Kaplan-Meier analysis revealed that FHF-A had the worst, and FHF-NA and HAT groups the highest observed survival rates at 30 days after LT. Thus, among UNOS status 1 candidates, FHF-NA patients, and in particular those with higher MELD scores, benefited more from transplantation. The authors suggested that FHF-NA patients may warrant higher priority for transplantation because they died more frequently before transplantation than patients in other groups and could be stratified by MELD score.
MELD and Long-Term Post-LT Mortality
All the previous studies focused on short-term survival after LT. Habib et al.40 first evaluated the long-term post-LT survival in 1,472 patients. A pre-LT MELD ≥26 was associated with poorer short- and long-term (10 years) post-LT survival. Nagler et al.41 also evaluated the long-term post-LT mortality in 121 European patients who underwent LT for non-FHF. The mean follow-up was 5.4 years. The Kaplan-Meier curve showed no difference in post-LT survival for different cutoff points of the MELD score at any time points (1-12 years), except for a pre-LT MELD score cutoff of 30 points (P = 0.04), but this was present in only 2 patients; and pre-LT MELD score had a weak discriminative ability for post-LT survival (c-statistic < 0.61). Island et al.42 confirmed that patients with MELD score ≥30 have increased resource utilization and higher 12-month post-LT mortality (25% vs. 6%, P = 0.05) compared with those with MELD <30.
However, a retrospective review of 499 consecutive transplanted patients with cirrhosis (41% cholestatic liver disease) analyzing various pre-LT variables, including MELD and CTP score, with respect to early and long-term survival.43 In univariate analysis, MELD score, but not CTP score, was significantly associated with post-LT mortality. However, in the multivariate analysis, the MELD score was no longer significant. This was also true after exclusion of patients on hemodialysis. This result could be due to the high proportion of cholestatic patients with cirrhosis.
MELD SCORE AND RETRANSPLANTATION
The increasing number of patients who survive in the long term after primary LT will inevitably increase the proportion who need retransplantation as a result of chronic rejection or recurrence of primary liver disease. Generally, retransplanted patients have a worse prognosis compared with those being transplanted for first time (primary LT) (Table 3). A retrospective evaluation44 was performed of 22,120 patients from the UNOS database (9,564 HCV positive) with primary LT and 2,129 re-LT (899 HCV positive) from January 1996 to June 2002. Patients with malignancy or re-LT within 30 days after the primary LT were excluded. Survival after re-LT was significantly lower than after primary LT for the same preoperative MELD score (P = 0.001), particularly for pre-LT MELD scores >25 (5-year survival of less than 60%). Similar to Onaca et al.,30 HCV-positive patients (n = 362) with MELD scores of 21-25 had significantly lower survival rates after LT compared with HCV-negative patients with the same pre-LT MELD score.
Table 3. Predictive Value of MELD Score for Post-LT Survival in Patients With Cirrhosis Who Underwent Retransplantation
|Watt et al.44||2,129†||…||HCV re-LT had lower survival than non-HCV re-LT|
|Yao et al.45||40||MELD = 0.68, CTP = 0.82*||CTP >10 and MELD >25 associated with 1- and 5-year survival|
|Rosen et al.46||1,356‡||MELD = 0.635, CTP not provided||Independent factors of survival: Δt, recipient age, bilirubin, creatinine|
Yao et al.45 also evaluated the prognostic effect of MELD and CTP in the outcome of 40 consecutive patients who received re-LT more than 90 days after primary LT. By Kaplan-Maier analysis, survival at 1 and 5 years after re-LT were 69% and 62%, respectively. The preoperative CTP score was more strongly predictive compared with the preoperative MELD score (c-statistic 0.82 vs. 0.68, respectively), but the difference was not significant (P = 0.11). CTP score >10 and MELD score >25 correlated significantly with 1- and 5-year survival (P = 0.0006 and 0.038, respectively).
Rosen et al.46 developed a good prognostic model from the UNOS cohort to predict survival after re-LT on the basis of 5 variables [original UNOS model = 0.024 × (recipient age) + 0.112 × (√bilirubin) + 0.230(√creatine) − 0.974 × (cause of graft failure) + UNOS coefficient]. They analyzed 281 European and Australian patients who underwent re-LT for hepatic allograft failure with median length of follow-up 1.5 years. Patients with HCV underwent re-LT more frequently compared with other causes (P < 0.0001). Kaplan-Maier analysis of the 3-month survival after re-LT stratified patients into 3 risk groups according to the MELD score: <22, 22-31, and >32. Moreover, patients with >90 days' survival after re-LT had lower median MELD scores compared with those with <90 days' survival (P = 0.004). They also analyzed data from 773 UNOS patients and 206 non-U.S. patients who had undergone re-LT 15 days or more after primary LT with those retransplanted before 1994 having a worse prognosis than those retransplanted after 1994. Moreover, multivariable Cox regression analysis showed that time interval after primary-LT, recipient age, bilirubin, and creatinine were independent predictive factors of survival. On the basis of these independent variables, a new predictive model was developed. The c-statistic of this model (with or without recipient age) and pre-LT MELD score were comparable at 1-, 3-, and 12-month and 3-year time points (e.g., 0.65 and 0.635 for 12 months, respectively).
Finally, Pedersen et al.47 evaluated 23,034 LT recipients from the UNOS cohort between January 2001 and November 2004 (and thus overlapped with the study by Watt et al.44) comparing these to 955 candidates for re-LT (excluding urgent re-LT, i.e., interval between the first LT and reregistration <3 months). Although re-LT candidates had significantly higher MELD scores at listing, they had slightly better waiting list survival, compared with primary LT candidates with the same MELD at listing. In addition, there was no significant difference in waiting list mortality among re-LT candidates between pre-MELD and post-MELD eras. This study therefore contradicts the results of another study that uses patients from the same database.44 The only difference is that the first excluded patients received a new transplant within 30 days44 and the second within 90 days.47 More evaluation of the prognostic weight of the time interval to re-LT needs to be done in relation to MELD score. In the setting of re-LT, clinical utility may have more impact than in primary LT.
MELD AND HCC
LT is the treatment of choice for small HCC because it improves survival (80-87% at 1 year and 60-72% at 5 years), has a lower recurrence rate than after resection, and prevents new tumor by removing the cirrhotic field.48 Patients with HCC stage T1 (single tumor ≤1.9 cm) or T2 (single tumor 2-5 cm, or 2-3 nodules all <3 cm) (Milan criteria) are eligible for standard LT evaluation (Table 4). Allocation of cadaveric liver donors to patients on the basis of MELD score might not benefit patients with small HCC, particularly if they had preserved liver function. The risk of death and increased likelihood of recurrence after LT due to HCC progression need to be considered to ensure that these patients have a fair opportunity of receiving a cadaveric organ, without undue risk of recurrence. Cheng et al.49 used a Markov model to analyze 2 hypothetical cohorts of patients with small HCC who were assumed to have either (1) Gompertzian tumor growth (initial exponential growth decreases as tumor size increases) or (2) rapid exponential growth. On the basis of these results, in the initial HCC-adjusted MELD-based allocation policy, HCC patients with stage T1 were given a MELD score of 24 and with stage T2 a MELD score of 29, reflecting a 3-month pretransplantation death probability of 15% and 30%, respectively.
Table 4. Difference in Waiting List Survival Before and After Implementation of MELD in Patients With Cirrhosis With Hepatocellular Carcinoma
|Hayashi et al.50||20||Survival||NS|
|Yao et al.51||58 pre-MELD, 44 post-MELD||Cumulative probability of LT at 3, 6, and >8.5 months||0.0006|
|Sharma et al.52||535 pre-MELD, 644 post-MELD||5-months survival||<0.001|
Three recent studies have compared results in liver transplant candidates with HCC in the pre- and post-MELD era. Hayashi et al.50 retrospectively analyzed 20 HCC patients (stage T2 or less) who received transcatheter arterial chemotherapy and embolization. There were no statistical differences in waiting list survival between the pre-MELD and post-MELD group (50% vs. 64%, P = 0.35), despite the mean waiting time to transplantation being longer for the pre-MELD group than post-MELD group (327.8 vs. 175.5 days, respectively, P = 0.37). Although Hayashi et al. concluded that pre-LT transcatheter arterial chemotherapy and embolization is beneficial for listed patients with HCC stage T2 or less, the sample size was too small for conclusive results. Furthermore, Yao et al.51 studied the outcome of 58 patients before and 44 patients after MELD implementation. All patients had known HCC at the time of listing or developed HCC while on the waiting list. The cumulative probabilities of LT at 3, 6, and >8.5 months in the post-MELD group were 22.5%, 64%, and 88%, respectively, whereas in the pre-MELD group they were 17.2%, 24.7%, 35.8%, and 47.2% at 3, 6, 9, and 12 months, respectively. These differences were statistically significant (P = 0.0006).
Sharma et al.52 evaluated 535 pre-MELD and 644 post-MELD patients with HCC. The 5-month dropout rate before LT was 16.5% before MELD compared with 8.5% after MELD (P < 0.001), and the 5-month survival on the waiting list was 90.3% before MELD and 95.7% after MELD (P < 0.001). However, the 5-month survival after LT was 88.5% before MELD compared with 89.3% after MELD (P > 0.05). These findings raised concern that in the post-MELD period, patients with HCC had excessive priority compared with candidates without HCC. Similarly, Huo et al.53 found that patients with stage T1 and T2 of HCC who underwent locoregional therapy had a risk of disease progression no higher than patients without HCC and MELD score of 8-12 points (for stage T1) and 10-14 points (for stage T2). For these reasons, MELD score for HCC candidate was reduced from 24 to 20 points for stage T1 and from 29 to 24 points for stage T2 (reflecting a 3-month pretransplantation death probability of 8% and 15%, respectively), and in the latest proposal by UNOS, patients with T1 HCC are no longer eligible for MELD priority listing.54
The advantage of the MELD score is that it is based on multivariable analysis of objective and widely available laboratory tests for serum bilirubin, serum creatinine, and INR. In contrast to the CTP score, the MELD score incorporates an evaluation of renal function, which is an established marker of prognosis in patients with cirrhosis. In addition, there is no “floor” or “ceiling” effect with respect to quantitative variables such as bilirubin. Although there is a cap (i.e., a ceiling effect), this is minimized because the upper cap has been set to 40 points. Thus, severity of liver disease remains the major factor used to stratify patients with cirrhosis on the waiting list for LT in the new allocation system.
On the other hand, the use of MELD to prioritize LT recipients is not without limitations. First, the new allocation system based on MELD score is primarily a “justice” system. Thus, it is important to establish that there are no systematic biases that could influence prioritization of some patients over the others. However, serum bilirubin, creatinine, and INR may be influenced by therapeutic manipulations (e.g., Coumadin), as well as by disease progression, and for this reason, serial MELD calculations are needed in order to exclude transient changes. On the other hand, although renal function is a key component of the MELD score, and renal impairment, such as in hepatorenal syndrome, may improve after therapy with terlipressin and albumin or molecular adsorbent recycling system, the improvement of prognosis without LT is limited, so that the need to use terlipressin and albumin might be a better criterion of severity of liver disease rather than the creatinine value itself. In addition, the predictive ability of MELD score increases by adding clinical variables (hepatic encephalopathy, ascites) and laboratory (sodium) parameters.55–59 These parameters—hepatic encephalopathy, ascites, and hyponatremia (as an indicator of ascites)—are components of the CTP score, but are not included in the MELD score. Perhaps adding what is missing to the CTP score (i.e., serum creatinine), thus creating a Child D class, may be simpler for everyday clinical practice.60 Finally, in contrast to CTP score, it is necessary to use a calculator or an Internet connection to access the UNOS Web site (http://www.unos.org).
Although there was a concern that MELD score could give priority to patients more likely to die after LT, Freeman et al.61 analyzed the results of the first year (February 2002-February 2003) of the new MELD-based allocation system and found that post-LT survival remained unchanged. Furthermore, there were fewer new registrations to the waiting list, higher LT rates, significantly more HCC patients who underwent LT,4 and a small, but unfortunately not statistically significant, reduction in mortality on the waiting list, which may be as a result of the higher number of patients with more severe disease waiting for LT. However, new concerns have arisen because there has been an increase in combined kidney and LT as well as need for more renal support after the introduction of MELD score, suggesting that sicker patients are being transplanted and again raising the issue of utility.62 Future studies are needed to define the best approach in candidates for LT and renal dysfunction and which factors (cause, severity, or duration of renal dysfunction) are the most important for post-LT mortality.62
At the other end of the spectrum of severity of liver disease, Merion et al.63 analyzed 12,996 patients listed for LT after MELD implementation between September 2001 and June 2003. In patients with MELD score <15, there was a higher 1-year mortality after LT compared with those with the same score who remained on the waiting list—that is, they did worse with transplantation. In comparison, in recipients with MELD >18, there was a significant 1-year survival benefit, and this benefit increased as MELD score increased, even when MELD was >40. In this study, only 1-year survival was evaluated; it is possible that there could be benefit if longer-term survival or quality-of-life issues were to be assessed—that is, LT may be beneficial for patients with lower MELD scores, particularly when quality-of-life issues are prominent. Interestingly, 40% of LTs were performed in patients with MELD scores <18, the result of the current policy to first offer the donor organ to the organ procurement organization service area. Thus, at present, even though the liver allocation system in the United States is based on a “sickest first” policy that minimizes the role of waiting time, a large proportion of recipients do not have severe liver disease, as evaluated by MELD score.
In a previous study, we showed that the MELD score has a similar, but not better, discriminative ability compared with CTP score in predicting short- and medium-term survival in patients with cirrhosis who are not on a transplant waiting list.64 In this review, we focused on transplant settings. Although the studies reviewed compared mainly the MELD and CTP scores, and not the old UNOS status (1, 2A, 2B, 3) (Table 5), they do not establish MELD score to be superior to CTP score in predicting short-term mortality in patients with cirrhosis on the waiting list for LT (Table 1). In addition, MELD score has less discriminatory ability at lower values.65 However, in the CTP score, the assessment of severity of ascites and portosystemic encephalopathy is subjective, and this may explain why CTP score predicts similarly or slightly better than MELD score in single centers, compared with multicenter studies. In the United States, the new MELD-based allocation system has led to several advantages compared with the previous system because it has eliminated waiting time as a criterion, has slightly decreased mortality of patients on the waiting list,61 and, according to recent data, has provided utilitarian benefit, particularly in candidates with severe liver disease.63 In the United States, it is a step in the right direction, even if further fine-tuning and adjustments of the MELD variables, or additions to them,66 may be necessary—for example, by adding sodium assessment as a fourth component. However, one needs to consider that the benefit of MELD in the transplant setting may not be due to an intrinsic value of the scoring system itself, but because waiting time was removed as a criterion for prioritization.
Table 5. Classification of Candidates for Liver Transplants According to Old UNOS System
|1*||They have fulminant (sudden) liver failure or their new LT did not function (life expectancy <7 days without a LT).|
|2A||They have chronic liver disease and are in the hospital's critical care unit with a life expectancy <7 days without a LT. They have a CTP score ≥10 and meet other medical criteria.|
|2B||They have chronic liver disease and are becoming more urgently in need of a LT but do not meet the criteria for status 2A. They have a CTP score ≥10, or a CTP score ≥7 and meet at least one of the medical criteria.|
|3||They have chronic liver disease and are under continuous medical care, but are not in the hospital, except for possible short stays. These patients do not meet the criteria for status 2B.|
Ideally, an allocation system should ensure the best use of scarce organs to reduce waiting list mortality and improve post-LT survival. The optimal approach should be evaluation of preoperative recipient and donor characteristics, and validation of these factors. It may reflect “justice” to be top of an allocation list, but the patient will be hoping that because he or she is the sickest at this time, the best donor might be found to achieve the best chance of survival. On the other hand, no allocation system per se is able to overcome differences in social and education factors, possibly reflected in the study by Behari et al.,67 who showed that in the United States, white and Asian candidates are listed and transplanted at significantly lower MELD scores, whereas nonwhite candidates who are sicker do not undergo earlier LT, and in some cases wait for longer periods.
According to the available data, the MELD score, similar to the CTP score, which has been evaluated in fewer studies (Table 2), cannot accurately predict short-term post-LT survival (c-statistic always <0.70) (Table 2). For long-term post-LT survival, re-LT, as well as for validation of the current policy for patients with cirrhosis with HCC (no extra priority for T1 stage and 24 points for stage T2), further studies are needed. However, the ideal system should provide some system of matching donor to recipient, which currently occurs by clinical experience in center-based allocation systems and has been shown to correlate with MELD score.37 Current differences in short- and long-term complications and survival rates using extended donor criteria, non-heart-beating donors,68 and split donors69 necessitate that donor factors should be taken into account.36, 70 Haydon et al.,71 using artificial neural networks, recently developed a pioneer model for donor-recipient matching and prediction of survival of a particular donor-recipient pair. A recent evaluation of the European Liver Transplant Registry data demonstrates that donor age, total ischemic time, and other operative and recipient factors not included in MELD, significantly and independently affect outcomes after transplantation.72
In the future, determination of an optimal minimum listing score and performing more rapid LT in candidates with high MELD scores, especially for patients with worsening clinical condition not necessarily reflected by changes in MELD score, are needed. Standardization or use of normograms to equate different laboratory methods for creatinine and INR measurements will be necessary. In addition, other factors may need to be taken into account, such as age, cause of disease, changes in patient's clinical condition, comorbidities, and perhaps social factors (e.g., the likelihood of compliance and attendance at posttransplantation clinics). Last, more statistical work needs to be done to assess predictive accuracy in individual patients.11 All the studies cited, including the original MELD study (Tables 1–4), have evaluated discrimination (c-statistic), but they have not always providing confidence intervals, and none has evaluated calibration of the model (i.e., the observed vs. predicted outcome), which is a superior performance index.
Another consideration is that most prognostic models only evaluate short- and long-term mortality as an outcome, but other end points may be also important. For example, Saab et al.73 found that quality of life is not correlated with severity of liver disease as measured by MELD score, because hepatic encephalopathy and ascites, which have a strong relationship with quality of life, are not included in it. Byrne et al.74 found that MELD score at the time of LT was not related with quality of life during the first 4 months after LT. Thus, there is a need to evaluate this issue in well-designed prospective studies.75–77
MELD has been useful in stimulating a resurgence of interest in assessing prognosis in cirrhosis,60 but this review demonstrates it is imperfect and not necessarily better than the CTP score. Its merit is the inclusion of a criterion relating to renal function. It deserves to be explored with less of the fervor conferred to new dogma, as well as explored by combining donor characteristics. It seems implausible that any system, which needs to combine justice and utility in organ allocation, would not include donor variables that have been shown to be important.72 MELDD—with an added D for donor—is prime priority for evaluation.