Waiting List Removal Rates Among Patients with Chronic and Malignant Liver Diseases


* Corresponding author: Richard B. Freeman, rfreeman@tufts-nemc.org


Equitable liver allocation should ensure that nonelective removal rates are fairly distributed among waiting candidates. We compared removal rates for adults entered with nonmalignant (NM) (N = 9379) and hepatocellular cancer (HCC) (N = 2052) diagnoses on the Organ Procurement and Transplantation Network (OPTN) list between April 30, 2003, and December 31, 2004. Unadjusted removal rates for NM vs. HCC diagnoses were 9.4% vs. 8.7%, 13.5% vs. 16.9% and 19.1% vs. 31.8% at 90, 180 and 365 days, respectively after listing. For NM candidates, model for end-stage liver disease (MELD) score (RR = 1.16), age (RR = 1.03) and metabolic disease diagnoses (RR = 1.66) had higher risks of removal; and PSC (RR = 0.62) and alcoholic cirrhosis (RR = 0.82) had lower risks of removal. For HCC candidates, MELD score at listing (RR = 1.09), AFP (RR = 1.02), maximum tumor size (RR = 1.16) and age at listing (RR = 1.02) had increased risks of removal. The equation 1 − 0.920 exp[0.09369 (MELD at listing − 12.48) + 0.00193 (AFP − 97.4) + 0.1505 (maximum tumor size − 2.59) defined the probability of dropout for HCC candidates within 90 days of listing. We conclude that factors associated with the risk of removal for HCC are different from NM candidates, although MELD score at listing remains the most predictive for both groups. Liver transplant candidates with HCC may be prioritized using a risk score analogous to the MELD score.


The US liver allocation system for adult patients with chronic, nonmalignant (NM) liver disease employs a risk of mortality endpoint, defined by the model for end-stage liver disease (MELD) (1), for ranking patients (2). This endpoint, however, has been viewed as an inadequate definition of need for candidates with hepatocellular carcinoma (HCC) because these patients are thought to face a risk of progression of their malignancy that is greater than their risk of death while waiting. Several studies have identified the extent of HCC that is associated with low recurrence and excellent survival rates after liver transplantation (3–5). Rates of removal due to HCC progression beyond these favorable stages, the so-called ‘dropout rates’, have been studied in systems where liver allocation is center-based (6), or in single centers where deceased donor organs are allocated under the MELD system (7).

The original MELD-based US liver allocation system included rough estimates of the risk of HCC progression beyond criteria first described by Mazzaferro (so-called Milan criteria) where candidates with either one HCC lesion less than 5 cm in size, or up to three lesions with the largest not more than 3 cm in size, achieved excellent results with liver transplantation (3). Initially, policymakers estimated that HCC candidates with single tumors less than 2 cm in size (stage 1), and those with one tumor <5 cm or two or three tumors the largest of which was less than 3 cm in size (stage 2) would have 3-month risks of progression beyond these criteria of 15% and 30%, respectively. These risks of progression equated to an MELD score of 24 for stage 1 and 29 for stage 2 disease. Subsequently, single center (8) and Markov Model analyses (9) have indicated that these initial judgments overestimated the risks of tumor progression. Based on these and other data, policy developers reestimated the 3-month risk of HCC progression assigned to stage 1 and stage 2 disease to 8% (MELD = 20) and 15% (MELD = 24) in April 2003. The system was further refined to remove all increased priority for stage 1 HCC candidates in January 2004.

At this time, there is no published comprehensive analysis of risk factors that HCC candidates face for removal from the national waiting list. In addition, since one goal of equitable liver allocation is to be sure all registered candidates have similar risks for nonelective removal from the list, a comparison of the removal rates for NM and HCC liver candidates under the MELD allocation system is necessary. Finally, since risk of removal for HCC candidates may be equated with progressive HCC, construction of a model that predicts these risks, analogous to the MELD score for risk of removal due to death, is possible and may allow for more accurate prioritization of HCC candidates. These three objectives are the aims of this report.


We obtained listing and removal data for adult (>17 years of age) candidates registered on the United Network for Organ Sharing (UNOS)/Organ Procurement and Transplantation Network (OPTN) liver transplant waiting list between April 30, 2003, and December 31, 2004. Demographic, diagnostic, MELD score, blood type and HCC-specific data including stage, number and size of tumors, diagnostic imaging studies, use of ablative techniques, serum alpha-feto protein (AFP) and date of measurement, and reason for removal from the UNOS/OPTN system were assessed. Death rates were supplemented by cross-referencing UNOS/OPTN data with the Social Security Death Master Files. For NM cases, Status 1 patients, patients with a diagnosis of cancer and patients listed with MELD non-HCC exceptions were excluded. We excluded the non-HCC MELD exceptions because they represent a very small heterogenous group who has widely varying reasons for granting of exception and extremely few removal events from the list. NM dropout removal events included removals for death/too sick or removals for ‘other’ in which the stated reason for removal was interpreted as too sick to transplant. For HCC candidates, all first applications for increased HCC priority according to UNOS/OPTN policy (10) that were approved during April 30, 2003, to December 31, 2004, study period were included. Patients with granted HCC exceptions whose calculated MELD score subsequently exceeded their HCC assigned score (N = 66) were included in the HCC cohort to maintain an ‘intention to treat’ principle. HCC candidate removals for reasons of death/too sick, or ‘other’, tumor-related causes were counted as dropout events. Only cases meeting above specified criteria that were actually removed from the list were counted as dropout events. Patients who were made inactive but not removed from the list were counted as ‘at-risk’ and were included in all analyses. Unadjusted dropout rates for NM and HCC candidates were calculated using Kaplan-Meier methods. Comparisons for unadjusted rates were done using the log rank test. To assess risk factors for dropout removal we constructed Cox regression models first using all clinical variables. The endpoint for the Cox models was time to dropout with no assumptions made regarding number of events within a given time interval. Records for HCC patients were left truncated to correspond to the time of their first HCC application. All patients were censored at transplant or removal for other elective reasons or at the end of the study. We then calculated an equation incorporating only significant clinical variables to predict waiting list removal for HCC candidates. We purposefully did not include potentially discriminatory variables in accordance with the development of MELD allocation policy for NM patients (2,11). This equation was assessed for predictive accuracy on a more recent cohort of 624 HCC candidates accumulated from January 1, 2005, to April 30, 2005, using the area under the receiver operating curve or concordance (12), and a generalized method that accounts for censored data (13).


Between April 30, 2003, and December 31, 2004, 11 431 new adult candidates were added to the US liver transplant waiting list with 2052 candidates receiving increased priority for HCC based on initial applications. Demographic, diagnostic and HCC-related variables are displayed in Table 1.

Table 1.  Demographic characteristics of the 11,431 persons in this study
 Nonmalignant N = 9379HCC N = 2052
  1. 1This also includes cirrhosis due to drug exposure; cirrhosis type A; cirrhosis cryptogenic; other cirrhosis (specify) are not included in the HBV, HCV and HBV and HCV rows.

Age (mean ± SD)52.1 ± 9.455.1 ± 7.8
Gender (M/F) (%)63.3/36.777.6/22.4
Diagnosis (%)
Noncholestatic cirrhosis142.749.7
 HBV and HCV0.50.8
Blood type (%)
Race (%)
MELD at listing/first HCC application (mean ± SD)16.5 ± 8.112.0 ± 4.2
Transplanted within 90 days of listing20.735.2
 AFP > 500 (%) 10.8
 Maximum tumor size 2.7 ± 1.1
 Tumor number 
  1 71.1
  2 20.2
  3 8.5
  >3 0.2
 Ablation (%) 25.1
 CT (%) 61.6
 MRI (%) 42.3
 Ultrasound (%) 13.7

Overall unadjusted dropout rates are compared in Figure 1. Rates of dropout are lower for HCC candidates over time up to 90 days of listing. Ninety days after listing, rates of dropout are approximately equal for HCC and NM candidates. Thereafter, HCC dropout rates exceed NM dropout rates up to 1 year after listing.

Figure 1.

Unadjusted dropout fraction over time.*At one day after listing.

Results of a Cox regression analysis assessing the relative risk of dropout for candidates approved for increased HCC priority are displayed in Table 2. In this multivariate model, MELD score, maximum tumor size, AFP (per 10 unit increase continuously) and age at listing were all independently associated with an increased relative risk of removal. Ablation, diagnostic imaging modality, number of tumors, race, gender and blood type did not have significant relative risks for removal. Table 3 displays results of a second Cox model that included significant risk factors from the first model and compared HCC patient dropout rates for each OPTN region with Region 1. Compared to Region 1, one region (10) had a higher relative risk of dropout for HCC candidates and, overall, region as a variable was significantly associated with dropout rates (p = 0.006). We observed considerable variation when comparing unadjusted 3-month dropout rates across regions for HCC and NM candidates (Figure 2A) and the variation in HCC dropout rates was still present when a 6-month dropout time point was used (Figure 2B). In contrast to HCC dropout rates, dropout rates for NM cases were relatively similar among the regions at 3- and 6-month time points.

Table 2.  Cox regression model assessing relative risk of removal for candidates receiving increased HCC priority using calculated MELD score at the time of application without assigned HCC priority MELD points
FactorRR95% Lower95% Upperp-Value
  1. 1Per unit increase.

  2. 2Per 10 unit increase.

MELD at first HCC application1.0911.061.13<0.0001
AFP (ng/mL)1.0221.011.03<0.0001
Maximum tumor size (cm)1.1611.031.310.012
Age at listing1.0211.011.040.007
Ablation (yes vs. no)1.000.731.380.996
MRI vs. CT1.110.861.440.433
Ultrasound vs. CT1.050.701.590.801
2 tumors vs. 11.210.851.720.282
3 or 4 tumors vs. 11.100.681.800.690
Hispanic vs. white0.940.661.330.728
Black/other vs. white0.760.491.180.229
Asian vs. white0.640.401.030.065
ABO = A vs. O0.950.711.260.708
ABO = B vs. O0.920.601.420.711
ABO = AB vs. O0.760.242.420.646
Table 3.  Cox regression model using statistically significant risk factors from Table 2 and factors for region
FactorRR95% Low95% Upperp-Value
  1. 1Per unit increase.

  2. 2Per 10 unit increase.

  3. 3Chi square 10 degrees of freedom.

MELD at first HCC application11.101.071.13<0.0001
AFP2 (ng/mL)<0.0001
Max tumor size1 (cm)
Age at listing11.
Region3 0.006
 Region 2 vs. Region 11.440.792.630.229
 Region 3 vs. Region 11.770.883.550.111
 Region 4 vs. Region 11.160.562.370.692
 Region 5 vs. Region 10.860.511.470.584
 Region 6 vs. Region 10.850.282.560.767
 Region 7 vs. Region 10.860.451.650.641
 Region 8 vs. Region 10.810.331.960.635
 Region 9 vs. Region 10.810.451.450.471
 Region 10 vs. Region 12.561.125.850.026
 Region 11 vs. Region 11.780.923.460.089
Figure 2.

Figure 2.

(A) Unadjusted 3-month dropout rates by United Network for Organ Sharing (UNOS)/Organ Procurement and Transplantation Network (OPTN) region, (B) Unadjusted 6-month dropout rates by UNOS/OPTN region.

Figure 2.

Figure 2.

(A) Unadjusted 3-month dropout rates by United Network for Organ Sharing (UNOS)/Organ Procurement and Transplantation Network (OPTN) region, (B) Unadjusted 6-month dropout rates by UNOS/OPTN region.

For NM candidates (Table 4), significantly increased risks for dropout removal were MELD at listing, age at listing and metabolic liver disease. Candidates with alcoholic and primary sclerosing cholangitis liver diseases had independently significantly reduced risks of dropout removal.

Table 4.  Cox regression model for NM candidates
FactorRR95% Low95% Upperp-Value
  1. 1Per unit increase.

MELD at listing1.1611.151.17<0.0001
Age at listing1.0311.021.03<0.0001
Black vs. white0.750.491.160.196
Hispanic vs. white0.960.831.120.623
Asian vs. white0.790.591.050.107
Other vs. white1.150.781.710.480
ABO = A vs. O0.980.881.090.692
ABO = B vs. O0.900.751.070.241
ABO = AB vs. O1.060.801.420.679
AHN vs. cirrhosis (nonalcoholic)1.080.861.360.506
Alcoholic cirrhosis. vs. cirrhosis0.820.720.940.004
Metabolic vs. cirrhosis1.661.202.300.002
PBC vs. cirrhosis0.790.591.050.107
PSC vs. cirrhosis0.620.450.860.003
Other vs. cirrhosis0.920.801.060.259

The predicted probability of dropout for candidates assigned increased priority due to meeting the HCC inclusion criteria was based on an equation using the patient-specific nondiscriminatory variables (gender, race and age were not included) estimated from our Cox model. Thus, the probability of dropout for HCC patients meeting UNOS/OPTN HCC criteria is defined by 1 − 0.920 exp[0.09369 (MELD at listing − 12.48) + 0.00193 (AFP − 97.4) + 0.1505 (maximum tumor size − 2.59)]. We term this the HCC-MELD equation. Based on this equation, a patient meeting HCC UNOS/OPTN listing criteria with a calculated MELD score of 15 at the time of HCC application, a maximum tumor size of 3 cm and an AFP of 500 has a probability of dropout of 21.67%, equivalent to a standard MELD score of 24. Alternatively, for a patient with a calculated MELD score of 20, maximum tumor size of 2 cm and an AFP of 150, the probability of dropout would be 15.7%, equivalent to an MELD score of 22.

We validated our HCC-MELD model against a separate cohort of more recently listed HCC candidates and assessed the accuracy of this model to predict dropout using the area under the receiver operating curve or concordance (Table 5). The ROC AUC for our HCC-MELD model was 0.781 with an approximate 95% confidence interval [0.688, 0.853]. MELD alone was equally predictive of dropout but tumor characteristics of AFP and tumor size alone did not distinguish dropout candidates.

Table 5.  Concordance results from three models predicting dropout for HCC candidates from the waiting list
ModelConcordance95% Lower95% Upper
MELD + AFP + maximum tumor size0.7810.6880.853
AFP + maximum tumor size0.5560.4390.691


In the era of extreme organ shortage, inability to deliver organs in a timely fashion for waiting patients remains the most vexing clinical problem. Allocation systems should be designed so that risks for not receiving an organ can be evaluated and managed. This requires that allocation priorities be defined in quantifiable terms using measurable endpoints. The MELD-based liver allocation system provides such a system for candidates whose need for transplant can be defined by their risk of dying on the waiting list within 3 months. However, policymakers recognized that this definition of need would not be applicable to patients with HCC because these candidates face a risk of cancer progression greater than their risk of death in most cases. A mechanism to allow these HCC candidates to obtain higher priority was devised using arbitrary estimations of risk of cancer progression with several revisions of these estimates occurring since inception of the MELD system. Until now, there has been no examination of the relative and absolute effects of this policy on waiting list removal rates for death/too sick or cancer progression. Our data demonstrate that, during our study period, the dropout removal rate for adult candidates with HCC was similar to the dropout removal rate for candidates with NM liver diseases within 3 months of listing. This suggests that the relative priority established for HCC candidates allowed HCC candidates approximately the same access to transplantation within 90 days compared with NM candidates during our study period. These analyses were not limited to counting events within a time period, but rather, used time to events as a basis for comparison. The higher dropout rate for NM candidates less than 3 months after listing and lower dropout rate after 3 months compared with the HCC candidates suggests that the more ill NM patients either receive a transplant or die early after listing. Those that survive beyond 3 months are likely the less ill NM patients who have relatively low risk of dying or dropping out. Since the MELD system is targeted to the 90-day time frame, these data suggest that the prioritization scheme prevailing during our study period was relatively equitable for HCC and NM cases on a national level. Assigning priority for HCC candidates using a more continuous HCC policy may allow for better calibration of HCC and NM dropout rates in the future. In addition, such an approach would more fairly assign priority for patients with documented HCC and calculated MELD scores that exceed the HCC policy-defined MELD.

Our Cox models suggest, as in other reports (6,14), that AFP is independently associated with increased risk of dropout from the waiting list. Recent studies (6,14) have identified serum AFP as a marker for rapid tumor progression and dropout from the waiting list. Most controversial, however, has been the choice of the so-called ‘cutoff’ value of AFP used for prediction of HCC outcome (15,16). Our data suggest that the risk of dropout increases linearly with an increasing AFP level, with no discernible threshold. Previous studies have indicated that sensitivity increases and specificity decreases for lower AFP cutoff values (14).

Our finding of maximum tumor size as an independent risk for HCC dropout agrees with other recent reports citing tumor size (1,8,14) as a predictor of dropout from waiting lists. Additional investigators have also reported tumor number (8,17,18) as a discriminator of waiting HCC patients who will drop out from those who will not. Clearly, in our data set, the vast majority of cases meet Milan criteria, so presumably, only cases with three or fewer lesions and a single tumor maximum size of 5 cm are included for calculation of our model. It is possible that tumor number, in addition to maximum tumor size, would predict more rapid progression if a less restrictive inclusion of study subjects were used. Moreover, ablative treatments may influence dropout rates as recent studies have suggested (19–21), however, ablation in our analysis was not a significant predictor of waiting list dropout for HCC candidates. This may be due to variations in modalities, number of treatment repetitions and/or timing of application of ablative protocols for UNOS/OPTN-listed HCC candidates. Alternatively, since the majority of HCC candidates are transplanted within 6 months, there may be insufficient postablation follow-up before transplant to see an effect of ablation on dropout rates.

Dropout rates are different around the country for both HCC and NM patients within 3 months of listing. When we compared individual regions against Region 1, one region had an increased dropout risk relative to Region 1. In terms of the overall effect, region was a significant contributor to the model (chi square = 24.58 on 10 df, p = 0.006). We did not examine differences in dropout rates at the donor service area level due to an insufficient number of events (248). During our study period, HCC candidates enjoyed greater priority than they do presently, making it very possible that the disparity in dropout rates among the regions may have increased (or decreased) with the more recent HCC priority revisions. Interestingly, we observed relatively consistent dropout rates for NM candidates among the regions and much wider variation for the HCC candidates. Higher HCC dropout rates at 3 and 6 months after listing occurred in regions with relatively higher transplant rates and generally lower MELD scores at transplant. These results may indicate significant practice variation where some centers are more prone to remove HCC candidates for tumor progression while others more aggressively seek RRB exceptions. These geographic variations require more investigation.

We used time to event Cox models to calculate relative risks for various clinical parameters that were associated with dropout rates. By doing so, we avoided limiting our calculations to predefined time periods thereby allowing comparisons of widely varying waiting times around the country. Moreover, since HCC candidates who receive increased priority are removed based on policy-defined criteria, changes in policy will likely affect dropout rates. Thus comparisons between HCC and NM candidates must be done using patients listed under concurrent policy conditions. Furthermore, future assessment of candidates' dropout rates must take into account the changes in policy and priority assigned to HCC candidates.

This latter issue makes validation of our HCC-MELD equation in cohorts of HCC patients different from the derivation cohort difficult. Nonetheless, our initial validation findings suggest that our HCC-MELD model has reasonable predictive value for estimating dropout within 3 months of listing for HCC candidates meeting Milan criteria. We note, however, that our HCC-MELD equation did not have a better predictive value than the MELD score alone for predicting dropout for HCC patients. This may indicate that, in the time frame of our study, progression of underlying liver disease may be as important as tumor characteristics in influencing dropout. This is consistent with a recent comparison of prognostic scores for HCC where the one score that included liver function as a component was found to have superior prognostic accuracy (22). This effect of underlying liver disease may be even more evident as overall HCC priority has been reduced, making it more common for candidates to achieve higher calculated MELD scores than their HCC priority scores would dictate. We hope to make further refinements in our model to see if better concordance can be achieved and to revalidate the HCC-MELD on a larger cohort of patients under the lower priority system prevailing currently. With this lower priority, it may be possible that there are earlier dropouts for the HCC patients since they do not have as high a priority as during our study period. As with all risk models, reassessment of the calculations for the coefficients and validity of the variables is critical for maintaining accurate predictive power as the population of at-risk individuals changes over time. This study indicates that dropout rates converge at 3 months for HCC and NM candidates, but they diverge at other time points. We believe that these data suggest that a more continuous prioritization for HCC candidates might help bring these dropout rates closer together over larger periods of time.