Pre-operative Predicting Model of Early Tumor Recurrence Outside the Milan Criteria for Hepatocellular Carcinoma

Background. Liver resection is a primary curative treatment for early stage hepatocellular carcinoma (HCC), but the outcomes are impeded by early tumor recurrence. This study aimed to establish a prediction model which could predict tumor recurrence outside the Milan criteria prior to liver resection. Methods. A cohort of 891 HCC patients who had curative liver resections were reviewed. HCC recurrence was divided into 4 groups: early recurrence inside the Milan criteria (ER-MI), early recurrence outside the Milan criteria (ER-MO), late recurrence inside the Milan criteria (LR-MI), and late recurrence outside the Milan criteria (LR-MO). The risk factors of ER-MO were analyzed to establish a prediction model. Results: During a median follow-up of 83.4 months, 589 (66.1%) patients developed recurrent HCC. Among 589 patients, 276 (31.0%), 86 (9.6%), 46 (5.2%) and 181(23.3%) patients were in ER-MO, ER-MI, LR-MO and LR-MI groups, respectively. In multivariate analysis of pre-operative factors, large tumors, non-single tumors, high α-fetoprotein level, high neutrophil-to-lymphocyte ratio and high indocyanine green retention were the independent predictors of ER-MO. By adjusting the weight of independent risk factors into scores and stratifying patients into class I, II, III and IV according to score 0, 2-6, 7-10 and ≥ 11, ER-MO events at 24 months were 12.5%, 22.5%, 45.4% and 66.3% for class I, II, III and IV, respectively (P< 0.05). Conclusions. A risk scoring model was built and could predict ER-MO based on pre-operative factors. This scoring model would be helpful for sharing the decision marking of treatments with patients. liver resection, local ablation, transcatheter hepatic artery chemoembolization (TACE), molecular targetting therapy, immune checkpoint inhibitor, radiotherapy, etc. According to Barcelona Clinic Liver Cancer (BCLC) treatment strategies, liver resection remains a primary curative treatment for early stage HCC.[1] However, high tumor recurrent rate after liver resection is the major obstacle in treating early stage HCC.[2, 3] Regarding the high tumor recurrent rate of liver resection for HCC, primary or salvage liver transplantation should be an option of treatments to achieve optimal outcomes. 14 RFA and 20 liver transplantation. For the 80 patients received re-hepatectomy, 15 patients were in ER-MO group, 24 patients in ER-MI group, and 41 patients in LR-MO/LR-MI groups. For the 20 patients (9 in ER-MO group and 11 in other groups) undergoing liver transplantation, 15 (75.0%) of them had TACE for disease-control or down-staging prior to salvage liver transplantation. The median waiting time between recurrence and liver transplantation were 15.2 months for all transplant recipients, but 30.0 months for the 9 patients in ER-MO group, who were able to have liver transplantation after successful down-staging of tumors. Six of 9 (66.7%) patients in ER-MO group achieved long-term post-transplant disease-free survival and 3 died of HCC recurrence after liver transplantation. Therefore, for the patients in ER-MO group, the curative chance was only 8.7% (24 in 276 patients) including 15 patients having re-hepatectomies and 9 patients having liver transplantation after successful down-staging of recurrent HCC.


Introduction
Hepatocellular carcinoma (HCC) is an aggressive malignancy resulting in the third leading cause of cancer-related death worldwide. The treatment modalities for HCC include liver transplantation, liver resection, local ablation, transcatheter hepatic artery chemoembolization (TACE), molecular targetting therapy, immune checkpoint inhibitor, radiotherapy, etc. According to Barcelona Clinic Liver Cancer (BCLC) treatment strategies, liver resection remains a primary curative treatment for early stage HCC. [1] However, high tumor recurrent rate after liver resection is the major obstacle in treating early stage HCC. [2,3] Regarding the high tumor recurrent rate of liver resection for HCC, primary or salvage liver transplantation should be an option of treatments to achieve optimal outcomes. HCC may recur early or late after liver resection. Most of early recurrence is intrahepatic metastasis from primary tumors within 2 years, while late recurrence may develop from multicentric lesions. [4] Early recurrent HCC is interrelated with aggressive tumor biology and microenvironment, and leads to di culty in treatment. Clinically, early recurrent tumors are frequently beyond the Milan criteria [5] once upon tumor recurrence is found. When the recurrent tumors are beyond the Milan criteria, the options of treatments are declined and the outcomes are less satisfactory than the tumors within the Milan criteria. Therefore, if the pattern of tumor recurrence can be predicted prior to liver resection, the treatment option may be altered to primary liver transplantation rather than liver resection if the primary tumors meet the Milan criteria.
When recurrent tumors are beyond the Milan criteria, only non-curative treatments can be offered to the patients. Salvage liver transplantation is curative, but is not indicated until the recurrent tumors can be down-staged successfully. Obviously, it is important to plan the therapeutic strategies precisely at the beginning of HCC diagnosed. The aim of this study was to investigate the prognostic factors which could predict the pattern of tumor recurrence prior to liver resection, and try to build up a pre-operative risk scoring model in predicting recurrent tumors within or beyond the Milan criteria.

Materials And Methods
Patients. 1005 HCC patients who had curative liver resection at Lin-Kou Chang Gung Memorial Hospital from 2003 to 2009 were reviewed. All patients' tumor status was at BCLC stage A and B. Curative liver resection was de ned as removal of all macroscopic tumors in pre-operative images and without any residual tumor at surgical margin microscopically. The patients who received other HCC treatments prior to liver resection (n=67) and who had other synchronous or metachronous malignancies (n= 41) were excluded. The patients with surgical mortalities (n=6) were also excluded since the present study was purposed to assess outcomes of tumor recurrence. The remained 891 patients were included in this study. The study was approved by local Ethic Committee of Chang-Gung Memorial Hospital (IBR No. 104-3900B).
Diagnosis of HCC recurrence. All patients were followed up regularly after liver resection. Liver function and α-fetoprotein (AFP) were measured every 3 months. Liver ultrasonography were performed every 3 months, too. When liver ultrasonography revealed a suspicious liver nodule or an elevated AFP level was noted, tri-phasic computed tomography (CT) or magnetic resonance imaging (MRI) would be performed to look for any evidence of tumor recurrence. HCC recurrence was de ned as typical imaging features of HCC (contrast uptake in arterial phase followed by washout in venous phase) in tri-phasic images of CT or MRI.
Strati cation of HCC recurrence. HCC recurrence was divided into early and late recurrence. Early recurrence was de ned as the tumor recurrence within two years after liver resection, and late recurrence was de ned as the tumor recurrence beyond two years after liver resection. [6] The recurrent tumors were further strati ed into inside or outside the Milan criteria: solitary tumor 5 centimeters in maximum diameter, 2 or 3 tumors without any one > 3 centimeters or more than 3 nodules in total tumor number, no major portal/hepatic vein branches invasion by tumor, and no evidence of extrahepatic metastasis. [5] Consequently, the patients with tumor recurrence were strati ed into 4 groups: early recurrence inside the Milan criteria (ER-MI), early recurrence outside the Milan criteria (ER-MO), late recurrence inside the Milan criteria (LR-MI), and late recurrence outside the Milan criteria (LR-MO) Data extraction and clinical outcome measurement. Demographic variables, routine hematology/biochemistry, clinicopathological parameters and tumor associated pathologic/histologic factors were documented. Disease-free survival was measured from the date of liver resection to tumor recurrence. Overall survival was measured from the date of liver resection to last follow-up or death which was registered in Bureau of Health, Taiwan.
Prediction model for early recurrence outside the Milan criteria. To establish the prediction model for tumor recurrence, the weight of each independent risk factor was determined and referenced to ''The Framingham Study risk score'' as Sullivan et al. described. [7] Statistical analysis. Pearson's chi-square test was used to compare clinical parameters and categorical variables between recurrent patterns. Kaplan-Meier methods were used to assess survival and the differences between subgroups were analyzed by the log-rank test. All potential risk factors of ER-MO recognized in the univariate analysis (p < 0.1) were recruited into multivariate analysis. The discrimination-ability of ER-MO was examined using the area under curve (AUC) under the receiver operating characteristic (ROC). P 0.05 was considered to be statistically signi cant. Analyses were done using IBM SPSS version 24.0 (SPSS Incorporation).

Results
Characteristics of patients and recurrent patterns. A total of 891 patients who underwent curative liver resection for HCC were enrolled in this study. Male predominance (n=706, 79.2%) and high prevalence of viral hepatitis (n=713, 80.0%) were observed. The mean age was 57.0 ± 13.1 with a range from 11.1 to 89.4 years. The majority (n=656, 73.6%) of the patients had single tumor. Near half of the patients (n=411, 46.1%) had cirrhosis, but most of them (n=880, 98.8%) were classi ed into Child-Pugh class A (supplement table 1 Survival in HCC patients according to recurrent patterns. After a median follow-up of 83.4 months, 413 patients (46.4%) died. The cumulative overall survival rates for all enrolled patients were 67.7% and 51.8% at 5 and 10 years, respectively. When the patients with tumor recurrence were strati ed into 4 groups according to recurrent patterns, cumulative overall survival rates at 5 and 10 years were 89.5% and 83.6% for the patients without tumor recurrence, 87.7% and 56.8% for LR-MI group, 74.0 and 40.6% for LR-MO group, 68.4% and 51.6% for ER-MI group, and 29.4 and 15.5% for ER-MO group, respectively. ER-MO group had an inferior outcome to other recurrent patterns ( Fig. 2, P < 0.001). The long-term survival between ER-MI and LR-MI was comparable. Therefore, ER-MO had the worst outcome and seemed different in natures from other recurrent patterns.
Clinical characteristics according to different recurrent patterns. According to 4 recurrent patterns, the clinicopathological features of the patients with tumor recurrence were presented in table 1. For patients in ER-MO and ER-MI groups, there were no signi cant difference between age, gender, hepatitis status, hematology/biochemistry data, indocyanine green 15-minute retention test (ICG R 15 ) and serum bilirubin. But, the patients in ER-MO group were highly associated with large tumors, multiple tumors, and micro-vascular invasion (P<0.05) and also had the tendency toward high serum AFP level and absence of tumor capsule. When the patients in ER-MO group were compared to LR-MI group patients who had a better outcomes, the patients in ER-MO group had higher population with high level of AFP and worse tumor characteristics including large-sized tumors, multiple tumors, poor histological differentiation, and microvascular invasion ( Table 1).
Development of ER-MO prediction model. The ER-MO group had the worst survival, compared to other groups. In this study, we tried to identify the risk factors of ER-MO and establish a predicting model of ER-MO prior to liver resection. The results of univariate and multivariate logistic regression analyses of prognostic factors for EM-MO were shown in table 2. In univariate analysis, seven pre-operative and three peri-/postoperative risk factors were identi ed including large tumor size, non-single tumor, hypoalbuminemia, high AFP level, high neutrophil-to-lymphocyte ratio (NLR), high ICG R 15 , history of tumor rupture, high intraoperative blood loss, presence of micro-vascular invasion and high histology grade. In the further multivariate analysis focusing only on pre-operative factors, large tumor size (for 3-5cm, HR = 1.90; for > 5cm, HR = 3.39), non-single tumor (HR = 1.76), high AFP (HR = 5.56), high NLR (HR = 1.60) and high ICG R 15 (HR = 1.58) were the independent predictors of ER-MO. The relative points were transformed from multiplying the coe cients (β) by ve and then run into integer to construct a scoring system. Different score allocations were given 2 points for AFP > 800ng/dL, 2 points for ICG R 15 >10%, 2 points for preoperative NLR > 2.5, 4 points for multiple tumors, 3 points for tumor size between 3 to 5 centimeters and 5 points for tumor size > 5 centimeters. Full score is 15 points. As expected, there was no patient situated in a sum score of 1, 12, and 14 by possible permutation, and we also did not have any patient with a full score of 15. To simplify the calculation, we grouped every two score and got 8 groups in sequence (0, 2, 3-4, 5-6, 7-8, 9-10, 11, 13). By comparing the Kaplan-Meier survival curves for the 8 groups, the risk of ER-MO was re-strati ed into 4 classes with no survival difference between intra-class groups. Class I, II, III and IV were corresponded to score 0, 2-6, 7-10 and ≥11, respectively (Table 3). Subsequently, this scoring system was used for analyses. The ER-MO event at 24 months after surgery was 12.5% for class I, 22.5% for class II, 45.4% for class III and 66.3% for class IV, respectively (Fig. 3a, P< 0.05 between any two groups). Our scoring model also deliberate long-term overall survival difference between each class. The median survival was 104.4, 88.8, 35.2 and 16.9 months for class I to IV, respectively. The 5-and 10-year overall survival were 86.7% and 58.6% for class I, 69.1% and 44.7% for class II, 43.8% and 31.6% for class III, and 27.2% and 19.6% for class IV, respectively (Fig. 3b, P< 0.05 between any two groups). The ROC curve of our scoring system to predict ER-MO had an AUC of 0.712 (95% CI: 0.675-0.749) with a best cut-off at 6. (supplement g. 1) Treatments for recurrent HCC. The principle of treatments for recurrent HCC is the same as primary diseases by considering disease extension, liver function and ECOG-Performance Status (PS). In the present study, TACE accounted for the most common treatment (369/589 patients,

Discussion
The prognosis of liver resection for HCC is impeded by high tumor recurrent rate. When HCC recurs, curative treatments such as re-hepatectomy or salvage liver transplantation still can be applied to rescue patients. However, the recurrent tumors may be beyond the Milan criteria once upon recurrent HCC is found and curative treatments are declined. According to multivariate analysis, large tumors, multiple tumors, NLR > 2.5, ICG R 15 > 10%, and AFP > 800 ng/mL were the signi cant pre-operative factors of ER-MO for the patients receiving curative hepatectomy for HCC. When transforming these risk factors to scores, four prognostic classes could be strati ed as score 0 for class I, score 1-6 for class II, score 7-10 for class III, and score ≥11 for class IV. This scoring system was well-correlated with long-term survival; and Class III and IV counted for 45.4% and 66.3% to develop worst prognostic ER-MO, respectively. Therefore, this pre-operative risk-scoring model for ER-MO prediction ts well for clinical needs. It provides predicting effects of surgery by ve easily available pre-operative factors and can serves as a reference for making clinical decisions before surgery.
ER-MO presents the worst outcomes for HCC patients with tumor recurrence after liver resection. HCC recurrence can be de ned as early and late while early recurrence was reappearance of tumor within 2 years after liver resection [4,8]. Early recurrence was commonly regarded as a consequence from intrahepatic metastases, which originated from the primary tumors [9,10] and highly associated with vascular invasion for tumor spreading [10]. Early recurrence was also highly linked with ER-MO as 76.2% of patients with early tumor recurrence had tumors beyond Milan criteria once upon the recurrent HCC was found. Thus, early recurrence indicates an inferior survival. [11] Fortunately, a minor portion of patients with early recurrence were within Milan criteria (ER-MI) and had compatible survival to the patient with LR-MI. Obviously, HCC with ER-MO were different from other recurrent patterns in nature. It could be the different intrinsic nature of ER-MO, which contributed to poor outcomes. [4,12] Several unfavorable factors were mentioned by Kim et al for recurrent HCC beyond Milan criteria after hepatectomy for solitary HCC [13]. However, there is still few studies speci cally discussing ER-MO after hepatecotmy. As ER-MO presented with aggressive tumor biology and was rarely curable in current practice, it is essential to identify high risk patients with ER-MO and to make an effort to detect disease recurrence early in order to arrange proper treatments. In spite of absence of de nite adjuvant therapy after hepatectomy, therapeutic approach to achieve a sustained viral suppression and even eradication of the virus via antiviral agents are important to prevent cirrhosis progression, HCC recurrence, and liver failure. [14] Moreover, primary liver transplantation may be considered if the patients are high risk to develop ER-MO after hepatectomy.
AFP level and tumor sizes re ected tumor biology. High serum AFP level not only re ects biological tumor burden but also causes tumor progression because it takes part in promoting metastasis of HCC by over-expressing certain oncogenes such as K19, EpCAM, MMP2/9, and CXCR4 [15]. Large-sized tumors are related to presence of micro-vascular invasion and micro-satellite nodules. [16] Both micro-vascular invasion and micro-satellite nodules often contribute to an aggressive tumor biology and poor clinical outcomes. Microscopic hidden lesions may be responsible for ER-MO after liver resections. A retrospective study based on prospective database emphasized the impact of pathologic pro le of the rst resected tumors, and demonstrated that the presence of either satellite nodules or microvascular invasion on the primary resected tumors declined the survival bene t of curative-intent therapy for recurrent diseases [17].
Pre-operative serum NLR has been considered as a reliable and easily-available information in evaluating cancer-related in ammatory microenvironment [18]. Evidences were shown that a higher level also indicated a negative result after liver resection [19,20]. Pro-tumor in ammatory microenvironment initiated by tumor-associated macrophages (TAMs) in HCC accelerates cancer progression [19,21]. A correlation between TAMs and serum NLR through upregulated in ammation-related cytokines was demonstrated in previous studies [22]. The degree of compactness of TAMs near micro-vascular regions may also explain the development of micro-metastases and subsequent eventful ER-MO [23].
Repeated hepatectomy and salvage liver transplantation are the curative treatments for recurrent HCC. Re-hepatectomy rate for recurrent HCC was between 15.8% to 25.1% [24,25], and long-term survival bene ts were compatible to primary liver resection with 31.8 to 50.9% of 5-year overall survival. [26,27] Salvage liver transplantation not only achieves tumor eradication but also cures liver diseases. The outcomes of salvage liver transplantation is comparable to primary liver transplantation. [28,29] When HCC recurs after liver resection, liver transplantation might be the rst choice of treatment with an optimal outcome. Previous retrospective studies [30] indicated salvage liver transplantation for HCC recurrence after primary hepatetomy is feasible, and it provided similar overall survival rate as the patients receiving primary liver transplantation . However, the possibility of salvage liver transplantation was only about 40% if tumor recurred after liver resection, and outcomes of salvage liver transplantation was probably impeded by extrahepatic spreading and aggressive tumor behavior of recurrent HCC. In the present study, we've conducted salvage liver transplantation for ER-MO HCC after a successful down-staging and 2-3 month waiting time to ensure tumor aggressiveness and obtained an acceptable outcomes.
A pre-operative risk scoring model could be established in this study to predict the possibility of ER-MO if tumor recurred after liver resection. This scoring model is supposedly applied to choose treatment options and will favor primary liver transplantation rather than liver resection if the preoperative risk score is more than 6. However, this is a retrospective study and no patients undergo primary liver transplantation according to this scoring model in this study. It is still unknown whether primary liver transplantation can avoid tumor recurrence and improve outcomes for the HCC with high possibility of ER-MO after liver resection.
In conclusion, early recurrence (ER-MO) is a clinical challenge in treating HCC after primary curative resection. In this study, we have built up a risk scoring model to predict ER-MO based on pre-operative factors. This scoring model can be helpful for choosing of treatment options. However, further prospective study is needed to verify this risk scoring model.

Declarations
Funding: The authors declare no funding was received for this study.
Con icts of interest/Competing interests: There is no con ict of interest among authors.
Ethics approval: The study was approved by local Ethic Committee of Chang-Gung Memorial Hospital (IBR No. 104-3900B).
Consent to participate: Not applicable. This is a retrospective study.
Consent for publication: Copyright will be transferred to APASL in the event the manuscript is accepted for publication.
Availability of data and material: All data generated or analyzed during this study are included in this published article (and its supplementary information les).
an intention-to-treat analysis.