Criteria and prognostic models for patients with hepatocellular carcinoma undergoing liver transplantation

Criteria based on morphological characteristics

Traditional morphological criteria based on the number and size of tumors were easily accessed through preoperative imaging and regarded as surrogate markers of HCC recurrence after LT (Table 1). Because the Milan criteria restrict the number of patients eligible for LT, several groups achieved comparable post-LT outcomes by expanding the original morphological criteria. Among the most commonly used expanded criteria were those proposed by Yao et al. in 2001, known as the University of California, San Francisco (UCSF) criteria (single tumor ≤6.5 cm in size, up to three tumors ≤4.5 cm in size, total tumor diameter ≤8 cm [16]. Using an expanded tumor size and total tumor diameter, the 5-year RFS rate reached 81% while expanding the number of patients eligible for LT by 5–20% versus the Milan criteria [9]. The UCSF criteria were further validated by the University of California Los Angeles group in a larger cohort of patients, demonstrating a 5-year survival rate of 64% for patients beyond the Milan criteria but within the UCSF criteria [17]. In 2009, Mazzaferro et al. [18] proposed the “up-to-seven criteria” based on the findings in 1,556 patients within 36 LT centers. The criteria illustrated that HCC with a total sum of tumor size and number not exceeding 7, but without microvascular invasion (MVI), could achieve equivalent survival outcomes as the Milan criteria. The 5-year survival rate of 283 patients was 71.2% within the up-to-seven criteria, similar to the rate of 73.3% for those within the Milan criteria. The main limitation of these criteria was the difficulty in obtaining histological features of MVI in the pretransplant setting, which limited its widespread adoption globally.

Table 1.

Results and criteria based on morphological characteristics for liver transplantation in patients with HCC

Apart from tumor diameter and number, the total tumor volume (TTV), first introduced by Toso et al. [19] in 2008, is another morphometric selection criterion adopted by several Canadian groups. The results indicated that patients with TTV within 115 cm³ had a similar survival rate as those within the Milan criteria (5-year OS rate 74% vs. 79%, P=0.3; 5-year RFS rate 78% vs. 80%, P=0.3). Additionally, the use of TTV significantly increased the number of included recipients compared with both the Milan (28–53% increase) and UCSF criteria (16–26% increase). These results were further validated in patients at centers in Toronto and Colorado [20].

Although the expansion of morphological features has permitted the identification of more patients with acceptable post-LT outcomes, there are differences in accuracy in detecting liver lesions because of the great heterogeneity in liver imaging techniques. In addition, increasing numbers of studies have suggested that the traditional tumor number and diameter do not precisely reflect tumor biology, which necessitates the use of other tumor biological markers [21].

Criteria based on markers of tumor biology

Alpha-fetoprotein (AFP) is among the most commonly adopted serum biomarkers for HCC development and differentiation, and it has been adopted by United Network for Organ Sharing (UNOS) and multiple centers to screen patients on waiting lists [22,23]. Hameed et al. [24] identified an AFP cutoff of 1,000 ng/mL as the independent predictor of vascular invasion and tumor recurrence. Patients with AFP levels lower than 1,000 ng/mL had a 5-year RFS rate of 80.3%, significantly exceeding the rate of 52.7% for patients with elevated AFP levels. Consequently, several criteria have incorporated AFP levels and morphological characteristics to better screen patients. The Hangzhou criteria select patients with HCC for LT based on a tumor diameter smaller than 8 cm or a tumor size exceeding 8 cm but with concurrent AFP serum levels <400 ng/mL and a histological grade of I or II [25]. The 5-year OS rate for patients meeting these criteria was 70.7%, comparable with that among patients meeting the Milan criteria. In 2009, Toso et al. [20] proposed similar criteria combining AFP levels and TTV. Recipients with AFP levels <400 ng/mL and TTV <115 cm³ experienced significantly better survival after LT. Another selection model was developed by the Liver Transplantation French Study Group based on a large cohort from 1988 to 2001 [26]. This model categorized patients according to AFP levels of <100, 100–1,000, and >1,000 ng/ mL and assigned different scores to each variable. Patients were considered at high risk of HCC recurrence if the final score exceeded 3 points. This model has replaced the Milan criteria for liver allocation in France and has been validated in various countries [27-29], confirming its utility in predicting excellent outcomes beyond the Milan criteria. Several other models, including RETREAT, MORAL and Metroticket 2.0, also classified survival risk by incorporating AFP levels and tumor number, size, or grade, and these models outperformed the original Milan criteria (Table 2) [30-33].

Table 2.

Results and criteria based on tumor biology for liver transplantation in patients with HCC

The limitations of the aforementioned criteria are attributed to the difficulty in consistently measuring AFP levels. The fluctuation of AFP levels during the waiting interval could lead to inconsistent outcomes. Thus, variations in AFP levels, rather than a cutoff, are considered more accurate for predicting post-LT outcomes. In 2009, Vibert et al. [34] reported that an increase in AFP levels of more than 15 μg/L per month was a negative prognostic factor in waiting recipients. The results were further validated by Lai et al. [35], who recorded superior 5-year outcomes in patients with an AFP change of less than 15 ng/mL per month (5-year OS rate 66.0% vs. 36.7%; 5-year RFS rate 92.3% vs. 53.8%) [29]. More recently, Halazun et al. [36] proposed a novel system using dynamic AFP levels to predict RFS. In a study consisting of 1,450 patients, the change in AFP levels between the maximum and final value was identified as an independent prognostic factor. It was demonstrated that patients with persistent AFP levels of <200 ng/mL had the best outcomes. Moreover, patients with the last recorded AFP level of <1,000 ng/mL and a >50% decrease had a comparable prognosis as those with a maximum AFP level of 200–1,000 ng/mL and a decrease to <200 ng/mL.

Apart from AFP levels, other serum markers reflective of tumor biology have been considered important in patient selection. Des-gamma-carboxyprothrombin (DCP), also known as protein induced by vitamin K absence or antagonist II, is an abnormal prothrombin caused by the absence of vitamin K, and it is expressed by some HCC cells [37,38]. DCP has been widely used in Asian countries to predict outcomes after LT, especially following living donor liver transplantation (LDLT). In a Japanese study, Todo et al. [39] found that patients who were beyond Milan criteria with DCP <100 mAU/mL and AFP <200 ng/mL had a 5-year RFS rate of 83.5% after LDLT [33]. Another large-scale Korean study confirmed that patients with a combined AFP+DCP level of <300 experienced 5-year RFS rates exceeding 50%, even among those with multiple tumors >10 cm in size [40]. Most recently, Norman et al. [41] proposed selection criteria based on AFP-L3 (fraction of AFP bound to Lens culinaris agglutinin) and DCP. A dual-biomarker combination of AFP-L3 ≥15% and DCP ≥7.5 mAU/mL predicted 61.1% of HCC recurrences, which outperformed AFP with C-statistics of 0.81 and 0.86, respectively, compared with 0.74 for AFP alone. The 3-year RFS rate was 43.7% for patients with dual-positive biomarkers, compared with 97.0% for all others. However, DCP levels can be elevated in certain situations other than HCC, such as biliary obstruction and vitamin K deficiency caused by malnutrition. Additionally, DCP levels can be strongly influenced by drugs including rifampicin and warfarin. In addition, the cutoff of DCP remains under debate because of variability in measurement techniques. Considering these limitations, AFP remains the most useful biomarker for predicting the clinical outcomes of patients with HCC post-transplant.

Using the aforementioned biomarkers, the selection of patients with HCC for LT has greatly improved beyond simple morphometrics. Recent developments in downstaging therapy, including multikinase inhibitors and locoregional therapy (LRT), have displayed promising effects on advanced HCC during the waiting interval. Therefore, selection criteria based on the response to downstaging treatments with greater accuracy are gaining increasing attention (Fig. 1).

Figure 1.

Evolving criteria for the selection of patients with hepatocellular carcinoma for liver transplantation. After the introduction of the Milan criteria in 1996, the subsequently expanded criteria mainly focused on the morphological characteristics of the tumor. Starting in 2008, the addition of biological markers facilitated further expansion of the original Milan criteria. More recently, new concepts for patient selection focused on successful downstaging and the response after locoregional or systemic treatment. UCSF, University of California, San Francisco; TTV, total tumor volume; AFP, alpha-fetoprotein; UNOS, United Network for Organ Sharing; TACE, transarterial chemoembolization; TARE, transarterial radioembolization; RFA, radiofrequency ablation; PIVKA-II, vitamin K absence II.

Criteria based on the response to bridging and downstaging treatments

Tumor “bridging” describes treatments for accepted HCC transplant candidates aimed at reducing the risk of waiting list dropout, whereas “downstaging” defines treatments that reduce the tumor burden to meet acceptable criteria and thus achieve expected survival [42,43]. The advantage of both approaches is that they permit dynamic assessments of tumor biology over time. Additionally, a positive response to pre-LT anti-HCC treatments often implies favorable tumor biology, aiding in the selection of suitable candidates and improving post-LT outcomes [44].

Regarding downstaging treatments, the priority is determining the extent to which we aim to reduce the tumor burden. Many studies have used the Milan criteria as the endpoint of downstaging. In 2015, Yao et al. [45] revealed that patients beyond the Milan criteria who underwent downstaging to within the Milan criteria had similar 5-year OS (77.8% vs. 81%) and RFS rates (90.8% vs. 88%) as patients within the Milan criteria without downstaging. Similar outcomes were demonstrated in other studies [46,47]. However, less than 10% of successfully downstaged patients underwent LT. This might be attributable to discrepancies in the original tumor burden between studies before downstaging. Therefore, the current UNOS policy uses the upper limits of the tumor burden defined by the UCSF group in 2008, including ≤8 cm for one tumor, ≤5 cm each for two or three tumors and sum of the maximal tumor diameters ≤8 cm, and ≤3 cm each for four or five tumors and sum of the maximal tumor diameters ≤8 cm (UCSF downstaging criteria) [48]. Using the upper limits of the UCSF group, Sinha et al. [49] recorded a lower dropout rate (25% vs. 54%) and superior survival outcomes (56% vs. 21%) compared with those in no limit group. The Italian Bologna group proposed the criteria of ≤6 cm for one tumor, ≤5 cm each for two tumors, and ≤4 cm each for three to five tumors with a sum of maximal diameters ≤12 cm [50].

In addition to the original tumor burden and downstaging criteria, the response of HCC to downstaging treatments is the most crucial marker for survival. The modified Response Evaluation Criteria in Solid Tumors represent a method for measuring treatment response [51], which is divided into several categories: complete response (disappearance of arterial enhancement in the tumor), partial response (a minimum of 30% reduction in the sum of the diameters of viable tumors versus baseline), and stable disease (neither partial response nor progressive disease). DiNorcia et al. [52] reviewed data from the United States Multicenter HCC Transplant Consortium to evaluate whether complete pathological response following pre-LT LRT affects post-LT outcomes [45]. Their results illustrated that patients with complete response had significantly lower 1-, 3-, and 5-year recurrence rates than those without complete response. Another large-scale European study confirmed that a poor radiological response after bridging treatment represented a strong independent risk factor for post-LT recurrence [53].

As previously mentioned, the AFP response is another surrogate marker of successful downstaging. Patients with persistent AFP levels >1,000 ng/mL despite anti-HCC treatment before LT achieved 5-year OS and RFS rates of only 49% and 35%, respectively [54]. A significant decrease in AFP from >1,000 ng/mL to <500 ng/mL was associated with a 3-fold reduction in HCC recurrence. Thus, the AFP response has been implemented in the US allocation system. Along with the response to downstaging treatments, the waiting time can also be used to identify tumor aggressiveness and biology. The “ablate and wait” strategy suggests at least 3 months of observation to ensure the absence of tumor progression and success of downstaging. Halazun et al. [55] conducted a large-scale study including more than 6,000 patients from the UNOS database. Patients in long wait-listing regions (median, 7.6 months) were more likely to drop out, resulting in tumor progression. However, the OS rate in long wait-listing regions was significantly better than that in short wait-listing regions (median, 1.6 months; 75% vs. 67%). The results were further validated by Mehta et al. [56], who recorded an increased 3-year survival rate in patients waiting for >9 months than for those waiting for <3 and 3–9 months (92% vs. 79% vs. 73%). To date, the optimal waiting period from downstaging to LT has not been clarified. A minimum observation period of 6 months is mandated by the UNOS policy for recipients with HCC (Table 3) [57].

Table 3.

Results and criteria based on the response to downstaging treatments for liver transplantation in patients with HCC

The treatment options for downstaging represent another important issue. Transarterial chemoembolization (TACE), transarterial radioembolization, and radiofrequency ablation are the most commonly used LRTs [58]. In addition to LRT, tyrosine kinase inhibitors (TKIs) including sorafenib and lenvatinib have displayed excellent effects as bridging treatments to LT [59]. In 2018, Golse et al. [60] reported a case series of five patients with HCC who received sorafenib as a downstaging therapy before LT. No recurrence was observed after 27 months of follow-up [60]. Another study from France in 2022 found that 62 of 327 patients with HCC were treated with TKIs. Of these patients, 26 underwent LT, and their 5-year RFS and OS rates were 48% and 77%, respectively [61]. Combination therapy with LRT and TKIs has also been investigated. In 2022, a retrospective study of 128 patients with HCC discovered that those who received TACE plus TKIs before LT achieved significantly better 5-year RFS rates than those who underwent TACE alone [62]. Apart from TKIs, immune checkpoint inhibitors (ICIs) have also exerted dramatic antitumor effects and led to prolonged survival in HCC in recent years [63]. Tabrizian et al. reported a case series of nine patients with HCC who received nivolumab and successfully bridged to LT [64]. Surprisingly, there was no tumor recurrence or death at a median of 16 months after LT. Similar results were also reported by several case studies [65,66]. To improve bridging strategy, the integration of ICIs and TKIs was further investigated. Abdelrahim et al. reported a patient who received atezolizumab plus bevacizumab prior to LT, and no recurrence occurred after 12 months of follow-up [67]. Another cohort of seven patients with HCC who received lenvatinib in combination with ICI therapy prior to LT also experienced satisfactory survival outcomes [68]. However, it must be noted that acute rejection after LT is a major concern in the context of ICI treatment. A safe washout period before LT and cautious post-LT immunosuppression strategies are required [69].

Risk scoring systems based on tumor clinicopathological features

Several scoring models combining tumor clinicopathological features have been proposed to predict the risk of HCC recurrence after LT (Fig. 2). In 2000, Iwatsuki et al. [70] proposed a Cox proportional hazards regression-based prognostic scoring system that included the bilobar tumor distribution, maximum tumor size, and vascular invasion. This scoring system classified patients into five grades and found that higher grades were associated with a lower tumor-free survival rate. In 2008, another study proposed and validated a prognostic score based on three different preoperative variables (maximum tumor size, tumor differentiation, and number of nodules), and this score had better accuracy than the Milan criteria in predicting HCC recurrence after LT [71]. Interestingly, this study revealed that tumor differentiation alone has no significant additive value for predicting HCC recurrence.

Figure 2.

Features used to develop risk scoring systems for predicting the prognosis of patients with hepatocellular carcinoma after liver transplantation. The parameters included recipient features, tumor clinicopathological characteristics, and serological biomarkers.

Based on its ability to reflect tumor biology and screen patients, the AFP level is also an important parameter predicting tumor recurrence after LT. A Cox score threshold of 0.7 from the AFP model, consisting of the largest tumor size, number of nodules, and log₁₀AFP, was deemed useful in stratifying patients with HCC at higher risk of recurrence after LT (50.6% vs. 8.8%) [26]. Another Cox proportional hazards regression-based model consisting of the Child–Pugh score, positive HBV detection time, number of tumors, tumor size, AFP levels, and tumor differentiation grade was proposed, and its sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) were 72.5%, 90.7%, and 0.887, respectively [72]. A Japanese study evaluated the prognostic impact of AFP and DCP, finding that satisfying two or more of the identified criteria (tumor size ≤5 cm for ≤5 tumors, AFP <250 ng/mL, DCP <450 ng/mL) was associated with higher 5-year RFS (96.8% vs. 20.0%) and OS rates (84.0% vs. 20.0%) [73]. In addition to the pre-LT AFP level, Ma et al. [74] found that AFP levels 7 days after LT are predictive of HCC recurrence. A model based on tumor size, tumor thrombus, MVI, and 7-day postoperative alanine aminotransferase and AFP levels was validated to predict recurrence with an AUC of 0.732.

Some other parameters have also been incorporated into risk score models. The systemic immune–inflammation index (SII), calculated as absolute platelet count × absolute neutrophil count/absolute lymphocyte count, was reported to be more effective than the platelet-to-lymphocyte ratio, neutrophil-to-lymphocyte ratio, and monocyte-to-lymphocyte ratio in patients with HCC after LT within the Hangzhou criteria [75]. Although a high SII did not significantly distinguish patients at higher risk of recurrence (5-year RFS rate 64.1% vs. 78.4%), it was significantly predictive of worse OS (5-year OS rate 56.1% vs. 82.4%). Wang et al. [76] developed a model including d-dimer and plasma fibrinogen levels (0.91×fibrinogen concentration+0.967×d-dimer concentration+0.585×AFP concentration+1.623×Milan criteria+0.68×MVI−3.159), and the model had satisfactory performance in predicting recurrence (AUC=0.828) and survival (AUC=0.764). Kornberg et al. [77] evaluated the prognostic nutritional index (10×albumin (g/dL)+0.005×lymphocyte count) for predicting HCC recurrence and identified a cutoff of 42 as a threshold for predicting tumor recurrence (AUC=0.896). Huang et al. [78] found that the combination of the preoperative albumin–globulin score and skeletal muscle index achieved predictive accuracy for OS (AUC=0.710) and RFS (AUC=0.700). The summarized results of the multivariable risk scoring systems are presented in Table 4.

Table 4.

Summary on the multivariable-based risk scoring systems based on clinicopathological features

Imaging radiomics features

The relationship between glucose metabolism, as evaluated by ¹⁸F-fluorodeoxyglucose positron emission tomography (¹⁸F-FDG PET), and HCC progression has been acknowledged by several studies [79,80]. Lee at al. [81] demonstrated that the ratio of the tumor maximum standardized uptake value to the normal liver tumor maximum standardized uptake value as determined by ¹⁸F-FDG PET significantly distinguished patients with a higher risk of recurrence after LT (1-year RFS rate 97% vs. 57%, P<0.001). A study from 16 Japanese centers further confirmed a PET-positive status (increased FDG uptake in the tumor as compared with non-tumor liver tissue) as an independent risk factor for HCC recurrence after LT [82].

Computed tomography (CT)-based radiomics features, including both non-textural and textural features, were reported to be effective in predicting HCC recurrence after LT [83]. The AUCs of an arterial phase radiomics model for predicting HCC recurrence among patients within and exceeding the Milan criteria were 0.748 and 0.661, respectively. Another integrated study of contrast-enhanced CT-based features, clinical characteristics, and laboratory values demonstrated that a model including peritumoral enhancement, tumor number, tumor size, AFP levels, and the presence of a tumor capsule had good utility (5-year AUC=0.85) for predicting HCC recurrence after LT [84].

In addition, magnetic resonance imaging (MRI) has also been adopted in prognostic prediction in patients with HCC after LT. Kim et al. [85] found identified the presence of satellite nodules and peritumoral hypointensity in the hepatobiliary phase as independent predictors of HCC recurrence after LT. Specifically, the researchers reported that patients with least one of two MRI features had a 3-year RFS of 27.5%, versus 84.6% for the remaining patients. Another study of 140 cases of pretransplant contrast-enhanced MRI in patients with treatment-naïve HCC found that patients with probable or definitive HCC based on the Liver Imaging Reporting and Data system had a 5-year RFS rate of 36.9%, compared with 95.8% for those with probable or definite malignancy but not specific for HCC [86]. In addition, Chiang et al. [87] revealed that the transverse relaxation rate attributable to the characteristics of tissue and field inhomogeneity on blood oxygen level-dependent MRI is significantly associated with liver rejection after liver transplantation (AUC=0.878). The summarized results of the imaging radiomics feature-involved models are presented in Table 5.

Table 5.

Summary on the prognostic effects of imaging radiomics features-involved models

Biological marker-based models

In addition to AFP, several other serum biomarkers have been discovered to be closely associated with HCC recurrence after LT (Fig. 3). Fiorentino et al. [88] reported that the MIB-1 proliferation index, E-cadherin level, and nuclear beta-catenin level are effective in identifying patients at a higher risk of HCC recurrence after LT. The presence of any biomarker alone and that of all three biomarkers were associated with recurrence rates of 88% and 99%, respectively. Another study utilized Feulgen staining and semi-automatic image analyses, which automatically calculate the DNA index by comparing the DNA content of the peak tumor cells to that of diploid reference cells. The results found that DNA index ≤1.5 is associated with higher 5- and 10-year survival rates (86% and 80%, respectively) than DNA index >1.5 (27% and 6%, respectively) [89,90].

Figure 3.

Features of biological marker-based models for predicting the prognosis of patients with hepatocellular carcinoma after liver transplantation. The markers can be divided into tumor proliferation and pathology markers, angiogenesis and inflammatory markers, circulating tumor cells, microRNAs, and metabolic profiling.

In addition, a CpG island methylator phenotype based on the P16, CDH1, SOCS1, GSTP1, STK, XAF1, and DAPK1 genes identified patients with HCC and a lower 3-year RFS rate after LT (25% vs. 64%) [91]. Campillo et al. [92] identified high expression of angiogenesis and proliferation markers (COX2, VEGF, and VEGF-2) in the cirrhotic liver, but not in the tumor, as predictive of recurrence in patients with HCC and liver cirrhosis after LT. The study evaluated preoperative plasma VEFG levels and indicated that 5-year RFS levels were significantly worse in patients with plasma VEFG levels >44 pg/mL (47.7% vs. 8.7%). In addition, Atanasov et al. [93] revealed that the hepatic infiltration of TIE2-expressing monocytes and CD68⁺ tumor-associated macrophages was predictive of decreased survival after LT in patients with HCC.

A microRNA (miRNA) microarray was adopted to predict the prognosis of patients with HCC after LT. Liese et al. [94] found that addition of miR-214 and miR-3187 expression to the Milan criteria more effectively stratified patients at higher risk of recurrence after LT (AUC=0.869) than the Milan criteria alone (AUC=0.640). Meanwhile, Ng et al. [95] identified circulating miRNAs as predictors of recurrence (miR148a and miR-1246) and survival (miR-1246) in patients with HCC after LT. Plasma metabolomics profiling identified phosphatidylcholine 16:0/P-18:1 and 18:2/OH-16:0 as independent predictors of HCC recurrence after LT [96]. Wang et al. [97] reported that peripheral blood circulating tumor cell count ≥1/5 mL is predictive of HCC recurrence after LT.

Artificial intelligence-based prognostic models

In 1997, a pilot study first developed fully connected artificial neural network models to predict HCC recurrence after LT based on the lymph node status, margins, vascular invasion, sex, tumor size and number, lobar distribution, age, and the presence of hepatitis B or C virus infection [98]. The mean AUC was 0.971±0.034 in the test set. However, the single-center-based nature and small number of patients (n=178) were cited as limitations. With the rapid development of artificial intelligence, novel models based on machine learning algorithms have been constructed to predict HCC recurrence after LT. Liu et al. [99] developed a deep learning model from pathology images that was trained using a pre-trained U-net. The images were fed into the MobileNetV2 model (a convolutional neural network model), and aggregation was performed using a generalized mean with a sign. The predictive accuracy was significantly higher (0.75) at 12 months after LT than that of American Joint Committee on Cancer staging. Using clinical variables, the US Multicenter HCC Transplant Consortium constructed a random survival forest machine learning model, achieving an AUC of 0.82 for predicting 5-year recurrence [100]. Most recently, Qu et al. [101] established a deep pathomics score for predicting tumor recurrence after LT. The results identified immune cells as the most significant tissue category for predicting post-LT recurrence (hazard ratio 1.907, 95% confidence interval 1.490–2.440).