Liver transplantation (LT) is currently the treatment of choice for early unresectable hepatocellular carcinoma (HCC),1, 2 and with candidate selection according to the Milan criteria (MC; 1 nodule ≤ 5 cm or 2-3 nodules, all ≤ 3 cm), post-LT survival similar to that for benign disease can be achieved.3-5 It has also been clearly shown that patients suffering from HCC with macroscopic vascular invasion (VI) have a very high risk of post-LT HCC recurrence.6-8 There are, however, many patients with intermediate-stage HCC for whom the role of LT has yet to be clearly defined.9 According to some reports on relatively small series,9, 11 these patients, with HCC beyond MC but without macroscopic VI, have an intermediate likelihood of survival.
Investigators at the University of Pittsburgh12-14 and at the University of California San Francisco (UCSF)15-17 have advocated the adoption of specific models for the selection of transplant candidates with HCC that include some patients with intermediate-stage tumors.14-16 More recently, a web-based survey18 allowed the largest retrospective collection of pathologic data from transplants for HCC exceeding the MC (n = 1112). With these data, a mathematical model was generated to predict survival after LT on the basis of 3 covariates: size of the largest nodule, number of nodules, and microscopic VI. In particular, this study supported the adoption of the up-to-seven (Up-to-7) criteria [with 7 being the sum of the size (cm) and number of tumors for any given HCC] in order to identify potential candidates for LT among patients exceeding the MC. Unfortunately, this study did not investigate the potential prognostic role in this direction of other relevant pathologic variables, such as the sum of tumor diameters,15 grade of differentiation,19, 20 and macroscopic VI.21 Moreover, this study confirmed that the size and number effect on survival is mediated by the occurrence of tumor relapse. The effect of pathologic tumor characteristics on survival is indirect because it is mediated by their ability to influence the risk of post-LT recurrence, so in prognostic modeling for LT of HCC patients, the first step should be to identify the most important predictors of recurrence among pathologic characteristics,3, 15 whereas the second step should be to prospectively assess the prognostic impact on survival of the same tumor characteristics transferred into the preoperative setting.4, 17 We now therefore present this large, retrospective clinicopathologic study conducted by 2 major LT units sharing a liberal policy for LT in patients with HCC. Its aim was to analyze in detail the impact of available pathologic staging systems (Table 1) and pathologic tumor characteristics on post-LT HCC recurrence.
Table 1. Available Selection Criteria for Liver Transplantation of Hepatocellular Carcinoma Patients
Abbreviations: TNM, tumor node metastasis; UCSF, University of California San Francisco; UNOS, United Network for Organ Sharing; Up-to-7, up-to-seven.
UNOS-TNM stages I and II coincide with the Milan criteria.
Unlike Mazzaferro et al.,18 we do not consider microvascular invasion in the definition of Up-to-7 criteria.
Diameter of the largest nodule (cm)+ number of nodules ≤ 7
Diameter of the largest nodule (cm)+ number of nodules > 7 or T4b, N1, M1
AFP, alpha-fetoprotein; CI, confidence interval; HCC, hepatocellular carcinoma; LT, liver transplantation; MC, Milan criteria; MVI, macroscopic vascular invasion; PSEP, proportion separation index; TNM, tumor node metastasis; UCSF, University of California San Francisco; UNOS, United Network for Organ Sharing; Up-to-7, up-to-seven; VI, vascular invasion.
PATIENTS AND METHODS
Five hundred twenty-three patients transplanted with HCC were identified from prospectively collected databases at Mount Sinai Medical Center (New York, NY) and the University of Padua (Padua, Italy). Pathologic variables recorded for each patient included the number of lesions, maximal tumor diameter, VI (macroscopic and microscopic), and tumor grade. The sum of tumor diameters was calculated. HCC was graded according to a modified version of the Edmondson-Steiner classification.19 Tumors were thus classified as follows: G1 to G2, well differentiated; G3, moderately differentiated; and G4, poorly differentiated. Tumors exhibiting regions of varying differentiation were assigned the highest grade observed. All histology was reviewed by the senior liver pathologist from the relevant center; an ongoing collaboration between the 2 pathologists ensured uniformity of tumor grading at the 2 centers.
Forty-four patients for whom complete pathologic data were unavailable were excluded; the present analysis focuses on 307 patients transplanted between September 1988 and August 2003 at Mount Sinai and 172 patients transplanted between July 1991 and December 2007 in Padua.
Pathologic staging systems were calculated according to the definitions in Table 1. Unlike Mazzaferro et al.,18 we did not consider microscopic VI in the definition of Up-to-7 criteria. The following variables were also recorded for each patient: age, sex, underlying liver disease, donor type (deceased or living), alpha-fetoprotein level, and timing of HCC diagnosis (before or after LT).
The patients' baseline characteristics are expressed as medians (ranges) for continuous data and as frequencies (percentages) for categorical data. Comparisons were drawn between groups by logistic regression, the chi-squared test, or Fisher's exact test, as appropriate.
The length of follow-up was expressed as the median (range). Recurrence was calculated as the time to tumor recurrence and was censored at the time of last follow-up or death if at that time there was no evidence of tumor recurrence.
The prognostic performance in predicting recurrence of the available pathologic staging systems (Milan, UCSF, and Up-to-7 criteria) was calculated according to the Kaplan-Meier method and compared by the log rank test. Moreover, to obtain a better description of the recurrence discriminatory ability of the pathologic staging systems, a simple proportion separation index (PSEP)23 was also calculated. PSEP is the difference between the proportion of negative events (recurrence) in the good and bad prognostic groups; the higher the value is of this statistic, the better the criteria are. Finally, a Cox multivariate model was created that included only variables with P < 0.10 in the univariate analysis.
In consideration of the aim of the study, the pathologic system showing the best prognostic performance in the multivariate analysis was used to stratify the study population into 2 prognostic subgroups (within and beyond the criteria).
A separate analysis of the risk of recurrence was performed for simple pathologic tumor characteristics (tumor number, size of the largest nodule, sum of the tumor diameters, microscopic VI, macroscopic VI, and poorly differentiated grade): both continuous and categorical pathologic tumor characteristics were tested by univariate and multivariate Cox analyses in the overall study group and in the prognostic subgroups within and beyond the calculated criteria.
Calculations were performed with the JMP package (1989-2003, SAS Institute, Inc.).
Baseline Characteristics, Histologic Data, and Tumor Staging
The baseline characteristics of the 479 patients are summarized in Table 2. There were 374 men and 105 women with a mean age of 57 years. Hepatitis C was the most common etiology of cirrhosis, accounting for 311 of 479 (65%) cases. LDT was performed in 40 cases (8%). The median diameter of the largest nodule was 3 cm, the median sum of diameters was 4 cm, and the median number of nodules was 2. VI was present in 174 patients (37%): macroscopic in 36 and microscopic in 138. Tumor grade was poorly differentiated in 48 patients (10%). Lymph node metastases were not detected in any case. Tumors were beyond MC in 207 patients (43%), beyond UCSF criteria in 152 patients (32%), and beyond Up-to-7 criteria in 124 (26%).
Table 2. Patient and Tumor Characteristics
Within Up-to-7 Criteria (n = 355)
Beyond Up-to-7 Criteria (n = 124)
All Patients (n = 479)
Abbreviations: AFP, alpha-fetoprotein; HCC, hepatocellular carcinoma; UCSF, University of California San Francisco; Up-to-7, up-to-seven.
P < 0.05 in the comparison of the within and beyond Up-to-7 groups.
Tumors beyond MC but within Up-to-7 pathologic criteria had a significantly higher number of nodules and greater diameter of the largest nodule (Fig. 1).
Overall Survival and Recurrence Analyses
After a median follow-up of 32.2 months (range, 1-160) among survivors, post-LT overall survival rates were 82%, 67%, and 61% at 1, 3, and 5 years, respectively.
Recurrence developed in 70 of the 479 analyzed patients (15%) at a median of 12 months (range, 1.5-73.0 months) after LT; the recurrence rates were 9%, 18%, and 24% at 1, 3, and 5 years, respectively. Recurrence was limited to the liver in 13 patients (19%), was extrahepatic in 34 patients (49%), and was both intrahepatic and extrahepatic in 23 patients (32%).
Comparison of the Pathologic Staging Systems
The recently introduced Up-to-7 criteria18 were compared to the MC and UCSF criteria in terms of prediction of recurrence (Table 3). Although all the analyzed systems performed as significant predictors of post-LT recurrence (Fig. 2), the Up-to-7 pathologic staging showed the best discrimination power with PSEP analysis (0.27). The best prognostic power of Up-to-7 criteria was also confirmed by a multivariate proportional hazards model (hazard ratio = 2.32; 95% confidence interval = 1.51-3.71).
Table 3. Probability of Recurrence According to the Pathologic Staging Systems
Probability of Recurrence (%)
Proportion of Recurrences: PSEP (95% CI)
Hazard Ratio (95% CI)
Abbreviations: CI, confidence interval; PSEP, proportion separation index; UCSF, University of California San Francisco; Up-to-7, up-to-seven.
All cases of macroscopic vascular invasion were included in the group with the worst prognosis.
In Table 4, we evaluate the prevalence of aggressive biological features (microscopic and macroscopic VI and poorly differentiated grade) and the risk of recurrence in 3 different size and number groups of patients: within MC, beyond MC/within Up-to-7, and beyond MC/beyond Up-to-7. Interestingly, beyond MC/within Up-to-7 patients had the same risk of recurrence as patients within MC, although a significantly higher proportion of cases had microscopic VI.
Table 4. Prevalence of Aggressive Biological Features and Recurrence in Different Size and Number Group Categories
Within Milan Criteria (n = 272)
Beyond Milan Criteria but Within Up-to-7 Criteria (n = 83)
Beyond Milan and Up-to-7 Criteria (n = 124)
NOTE: Unlike Mazzaferro et al.,18 we do not consider microvascular invasion in the definition of Up-to-7 criteria.
On the contrary, beyond MC/beyond Up-to-7 patients had a significantly worse prognosis than the beyond MC/within Up-to-7 group: these 2 groups were similar in terms of microscopic VI but significantly different in terms of tumor grade and macroscopic VI prevalence.
Predictors of Recurrence Among Pathologic Tumor Characteristics
When a univariate analysis of recurrence was performed in the overall study group, all of the examined pathologic variables correlated significantly with HCC recurrence (Table 5); in multivariate analysis, on the contrary, only 2 biological variables, tumor grade and VI (both microscopic and macroscopic), independently predicted recurrence.
Table 5. Pathologic Predictors of Hepatocellular Carcinoma Recurrence After Liver Transplantation
Variable (479 Patients)
Hazard Ratio (95% CI)
Hazard Ratio (95% CI)
Abbreviation: CI, confidence interval.
Size of the largest tumor
Sum of the tumor diameters
Microscopic vascular invasion
Macroscopic vascular invasion
Poorly differentiated grade
Univariate and multivariate analyses stratified according to within Up-to-7 criteria versus beyond Up-to-7 criteria are shown in Table 6. Within the criteria, the size of the largest tumor, sum of tumor diameters, and microscopic VI were significant predictors of recurrence, but only the last remained significant in multivariate analysis. For HCC beyond the criteria, on the contrary, only biological variables (tumor grade and microscopic and macroscopic VI) were significant.
Table 6. Pathologic Predictors of Hepatocellular Carcinoma Recurrence After Liver Transplantation According to the Pathologic Up-to-7 Criteria
Because only macroscopic VI and tumor grade have the potential to be diagnosed before surgery and they both had a significant discriminative ability beyond the Up-to-7 criteria, we used these variables to create a 3-stage pathologic staging system. Figure 3 illustrates recurrence rates according to this new pathologic staging. Recurrence was observed in 8% of stage 1 patients (within Up-to-7 criteria). Stage 2 (beyond Up-to-7 criteria but without poorly differentiated grade and/or macroscopic VI) showed an intermediate risk of recurrence (24%). Stage 3 (beyond Up-to-7 criteria with macroscopic VI and/or poorly differentiated grade) had the worst prognostic profile, the overall recurrence rate being 45%.
Pathologic staging of HCC after LT correlates with the risk of tumor recurrence24-26 and is important for a number of reasons:
1It helps predict the risk of death or recurrence for individual patients after LT.12
2By stratifying patients into subgroups with different outcomes, it may guide the intensity of post-LT follow-up and the application of adjuvant strategies.13
3It may suggest clinical criteria for the selection of HCC patients to be considered for the LT waiting list.4, 17, 19
4It has the potential to address exclusion criteria in patients with tumor progression while awaiting LT.27
There are many published studies relating pathologic findings to post-LT outcome in patients with HCC, but most are limited by the small number of patients with HCC beyond MC.3, 5, 15
More recently, a web-based survey18 allowed the largest retrospective collection of pathologic data from transplants for HCC exceeding the MC (n = 1112). Among this large population, 283 patients without microscopic VI who fell within the Up-to-7 criteria achieved a 5-year overall survival of 71.2%, which was similar to that of patients within MC. Microscopic VI, however, may not be predicted preoperatively, and it is strongly related to the size and number of nodules. Also in that study, in fact, a relevant proportion (29.3%) of patients beyond MC and within Up-to-7 criteria had a pathologic detection of microscopic VI. Thus, the clinical applicability of the Up-to-7 criteria in a preoperative setting as formulated in Mazzaferro et al.'s article18 remains controversial. For this reason, in this article we use a definition of Up-to-7 criteria that does not account for microscopic VI.
In the particular setting of LT for HCC patients, the most reliable and unbiased endpoints are probably intention-to-treat survival and/or the LT survival benefit27, 28 because only these endpoints account for both pre-LT and post-LT deaths rates. On the contrary, when only pathologic variables are studied, as in the present analysis and in Mazzaferro et al.'s article,18 the most reliable and unbiased endpoint is probably the risk of post-LT recurrence because the effect of these variables on survival is “mediated by occurrence of tumor relapse.”
The present analysis includes a large population (207 patients) transplanted for HCC beyond MC (Table 1). In this view, this study has the potential to allow an effective prognostic evaluation of morphologic Up-to-7 criteria (not considering the presence of microscopic VI but excluding cases of macroscopic VI) with respect to Milan and UCSF criteria. This is a fundamental methodological step before hypothesizing the use of such criteria as potential selection tools before LT.
As shown in Table 3, Up-to-7 pathologic criteria showed the best prognostic performance by PSEP and multivariate analyses. More importantly, the subgroup of patients beyond MC but within Up-to-7 criteria (Table 4) had the same risk of recurrence as patients within MC.
The first result of this study is, therefore, that the expansion of MC to morphologic Up-to-7 criteria does not determine a significant increase in the risk of post-LT recurrence. In this view, our data represent a prognostic validation of morphologic Up-to-7 criteria because they showed a relevant ability in predicting post-LT recurrence independently of microscopic VI.
Similarly to Mazzaferro et al.'s study,18 we found a significant difference in the prevalence of microscopic VI in the 2 subgroups of within MC patients and beyond MC/within Up-to-7 patients. Unlike that study, however, the subgroup of patients with the worst prognosis (beyond MC/beyond Up-to-7) differed significantly from the beyond MC/within Up-to-7 group only for the prevalence of poorly differentiated tumors and macroscopic VI and not for microscopic VI. This result suggests the potential for a further prognostic refinement of this HCC pathologic staging system by identification of significant predictors of recurrence, especially in the area of patients beyond Up-to-7 criteria.
For this reason, we performed a univariate-multivariate Cox analysis testing the ability of the main pathologic tumor characteristics (both morphologic and biological) to predict recurrence in the overall study group and in the subgroups of patients within and beyond Up-to-7 criteria.
In this study, tumor size and number, the characteristics on which LT candidate selection is currently based at most centers, failed to predict post-LT recurrence of HCC (Table 5) in a Cox model that also included histologic grade and VI as covariates. Possible explanations for this finding, which differs from the findings of previous studies,3, 15, 16 are that a higher proportion of patients in this study had HCC beyond MC and that (1) the incidence of poorly differentiated histology and VI is greater in patients beyond MC than in those within MC and (2) the impact of poorly differentiated histology and VI may be less in patients with a smaller tumor burden. According to this view, the value of the tumor size/number in predicting post-LT recurrence is a surrogate for parameters such as grade and VI that are in fact more closely related to biological behavior.24-26
Histologic grade has been shown to correlate with HCC recurrence in a number of previous studies24-26, 29, 30; the present analysis demonstrates grade to be a strong independent predictor of post-LT HCC recurrence in the overall study group (Tables 2 and 3).
The ability of Up-to-7 criteria to identify patients with a low risk of post-LT HCC recurrence was confirmed in this series (Fig. 2); nevertheless, among patients within these criteria, microscopic VI was a significant predictor of recurrence, whereas tumor size and number were not significant in multivariate analysis (Table 6). On the contrary, tumor grade was not a significant recurrence predictor within these criteria. This result suggests that, unlike some experiences,20, 21 the diagnosis of poorly differentiated grade in a patient within Up-to-7 criteria should not be considered an absolute exclusion criterion for LT.
Interestingly, in the 124 HCC patients with a tumor beyond Up-to-7 criteria, morphologic variables (size of the largest nodule, sum of diameters, and tumor number) failed to significantly predict recurrence. In Mazzaferro et al.'s study,18 the prognostic effect of tumor number tended to plateau above 3 nodules (thus beyond MC), whereas it was linear for size. According to our results, it is probable that beyond the Up-to-7 criteria, the prognostic effect of tumor size on recurrence also gradually decreases to a plateau. Differently from morphologic variables, conversely, in our study, biological tumor characteristics were significant recurrence predictors also in patients beyond the Up-to-7 criteria (Table 5). In particular, by using macroscopic VI and poorly differentiated grade as discriminative factors, we identified a large group of patients (Fig. 3) at intermediate risk for HCC recurrence (24%); this group represents a target population for future efforts to refine LT selection criteria. HCC beyond Up-to-7 criteria with macroscopic VI and/or poorly differentiated histology was associated with a 45% likelihood of recurrence, a rate that nearly all centers would consider unacceptable (Fig. 3, group 3). Another significant finding was the separation according to biological features not only of the risk of recurrence but also of the time to its occurrence (Fig. 3), which has been shown to be a surrogate marker of recurrence aggressiveness.29
The retrospective nature of this analysis and the fact that it is based on explant pathology are bases for caution in applying the findings to prospective LT candidate selection. The issue of waiting list dropout becomes a greater consideration with the inclusion of patients having more advanced HCC31; decreased survival on an intention-to-treat basis is inevitable when patients with more advanced HCC are included27 and must be considered in weighing treatment alternatives. Center-specific variables including the median waiting time and neoadjuvant strategies employed32-36 must also be taken into account.
The majority of enrolled patients in the 2 LT centers participating in this study underwent liberally applied pre-LT treatment protocols.21 Because this variable has mainly an impact on pre-LT outcome (dropout rate), whereas the potential of pretherapies for post-LT outcome has not been demonstrated, we did not consider this variable in the present analysis.
Regional/national policies such as those established in the United States by the United Network for Organ Sharing are important practical issues, in that access to donor organs depends on priority based on meeting and remaining within MC until LT. As suggested by Bruix and Sherman,37 clearly defined dropout criteria for patients with HCC progression before LT are an essential element of any proposal for broadening conventional LT criteria.
Preoperative percutaneous tumor biopsy is a safe procedure.38 Its reliability in identifying patients with poorly differentiated HCC, however, has been questioned by Pawlik et el.,39 who in a retrospective study of 98 presurgical needle core biopsies for diagnosis found 35% sensitivity in identifying poorly differentiated histology among those cases ultimately found to contain poorly differentiated HCC upon pathologic examination of the entire specimen. This series was predominantly (81%) characterized by resection as opposed to LT. Only 51% of patients were cirrhotic, and tumors were remarkably large, ranging up to 23 cm and having a median diameter of 7 cm; as the purpose of biopsy was for diagnosis, a single core demonstrating HCC was sufficient for inclusion. Even in the presence of these relevant limits, they still found a very high (93%) specificity of needle biopsy in identifying poorly differentiated HCC.
Our protocol, in which biopsy is performed for prognosis as opposed to diagnosis, is quite different. Using primarily an aggressive approach (percutaneous or laparoscopic), we perform at least 2 core (16-18 G) biopsies as well as fine needle aspiration (21 G) with multiple passes. Biopsies are repeated over time in patients whose HCC demonstrates progression while they are awaiting LT. In a series of 84 patients biopsied for HCC larger than 3 cm and subsequently surgically treated (44 were retrospectively enrolled with a single biopsy technique, and 30 were prospectively enrolled according to the aforementioned protocol), we found the sensitivity to increase from 36% (retrospective group) to 64% (prospective group) when our protocol was applied (Cillo et al., unpublished data, 2009). The specificity of poorly differentiated grade HCC identified on biopsy was extremely high (100% in both groups), and it provided a reliable basis for decision making when present; compared to the 2 published criteria based on tumor size and number for selection of LT candidates with HCC beyond Up-to-7 criteria, the presence of poorly differentiated grade histology was a significantly better predictor of HCC recurrence (Table 6). This result is in accord with the findings of Marsh et al.,40 who demonstrated that the inclusion of data reflecting biological potential adds to the predictive power of models based on tumor size and number.40
Independently of future improvements in our ability to preoperatively predict macroscopic VI and poorly differentiated tumor grade, microscopic VI remains the strongest predictor of post-LT recurrence also in the present analysis (Tables 5 and 6). Probably only specific molecular/genetic signatures will give us reliable preoperative surrogate markers of pathologic microscopic VI,41 however.
In conclusion, although Milan, UCSF, and Up-to-7 criteria provide a simple and convenient way to select a substantial subset of patients with HCC who have a low likelihood of post-LT HCC recurrence, tumor size and number are an inadequate basis on which to surely exclude HCC patients beyond such criteria because a relevant proportion of these patients could have an acceptable risk of recurrence after LT if aggressive biological features were excluded.
Acceptance of expanded HCC criteria will likely await prospective studies that incorporate cytological/molecular data into the algorithm for LT candidate selection.41