Salvage Liver Transplantation Leads to Poorer Outcome in Hepatocellular Carcinoma Compared with Primary Liver Transplantation

Hepatocellular carcinoma is the most common liver malignancy. Salvage liver transplantation (SLT) is viewed as a feasible cure for recurrence of HCC after resectomy, but the effect is under dispute. A retrospective study examined data at Renji Hospital for 239 transplants from January 2006 to December 2015, including 211 who received primary liver transplantation (PLT) and 28 who underwent SLT. A multivariable cox regression model was employed to pick out relative factors to overall survival (OS) and recurrence free survival (RFS). Propensity score matching (PSM) was used to balance the bias. Both OS and RFS were worse in SLT group than in PLT group, especially for those patients within Milan criteria. Our study demonstrates that SLT bears higher risk of recurrence and death than PLT, indicating that SLT should be given a more careful thought at performance.

clinical files. All methods employed in this study were performed in accordance with the relevant guidelines and regulations. Retrospectively, we reviewed the clinicopathological data on patients who underwent LT for HCC between 2005 and 2011 at the Renji Hospital Shanghai, China. SLT here is defined as LT performed in recurrent HCC patients who received previous radical resection. Those excluded according to flowchart in Fig. 1 were: patients without histological confirmation of HCC in the resected livers; patients with concurrent other malignancies, or with pathologically-confirmed other tumor types in removed samples, including interhepatric cholangiocellular carcinoma (ICC), combined hepatocellular carcinoma (cHCC-CC), Fibrolamellar hepatocellular carcinoma (FLC) and secondary metastatic tumor; patients received palliative downstaging therapy (resectomy, hepatic arterial chemoembolization, radiofrequency ablation etc.); patients died within 1 month after surgery; pediatric patients (< 18 years); patients who received liver transplantation to treat liver function deterioration after resections. Among the 239 patients finally included in this retrospective study, 211 received PLT and 28 were recurrent cases who received radical hepatectomy before and then treated by LT protocol.
Perioperative Assessment. Parameters compared between PLT and SLT are: age, sex, donor type, serum α fetal protein (AFP) level, Carbohydrate Antigen 19-9 (CA19-9) level, positive rate of hepatitis B surface antigen, hepatitis B (HBV) DNA and hepatitis C virus (HCV) RNA, CHILD-PUGH Score, tumor feature information such as tumor DIAMETE, number and metastasis. In this study, CHILD-PUGH Score was recognized as an index reflecting liver functional reserve.

Surgical Procedure and Postoperative Management.
In all the surgeries, 82.8% of the donors were deceased donors and 17.2% were living donors. Only prior hepatectomies with R0 margin status were considered radical and then, recruited in this study. All the transplantations were performed following standard techniques by experienced specialists in the Department of Liver Surgery, Ren Ji Hospital, Shanghai, China. Classic orthotopic LT was the only surgery type in the cadaveric donor group. All LDLT cases were operated on using right liver grafts without the middle hepatic vein. Biliary tract anastomosis was performed in a duct-to-duct form. The largest diameter of tumor as well as the number of tumor will be verified on the sample. Then, with the help of pretransplantation imaging information, all cases were classified as either within Milan criteria or beyond it. After LT, the primary immunosuppressive therapy involved tacrolimus (FK506) or cyclosporine (CsA) combined with methylprednisolone and mycophenolate mofetil (MMF). Steroids were withdrawn in 3 months. 41.8% of the patients received Rapamycin immunosuppression in later treatment. Cases with HBcAb and/or HBeAb positive, which indicating a potential or history HBV infection, all received prophylactic antivirus therapy. For those who still had HBVsAg and/or HBV DNA positive before transplantation, intravenous anti-HBV globulin was added after surgery. Outcome Definition. RFS was defined as the interval between surgery and recurrence. If recurrence didn't happen, patients were censored at death, retransplantation or the date of the latest follow-up. The OS was calculated from the date of operation to the date of death or retransplantation. Patients who were alive at latest follow-up were treated as censored. The definition of recurrence includes imaging evidence of both recurrent tumor mass in liver or elsewhere, ascites or pleural effusion proved malignant by biopsy and elevating AFP alone without tumor finding.
Statistical Analysis. Mean ± SD or median (interquartile range [IQR]) was used for description of continuous variables and No. (%) was for categorical factors. Difference between groups was assessed by Student t test for parametric variables and Mann-Whitney U test for nonparametric ones. A χ 2 test was used to generate the univariate models describing the association of variables with OS or RFS. Variables with p values < 0.2 by univariate analysis were chosen for multivariate analysis by the use of a cox regression model. To reduce potential bias in this retrospective study, propensity-score based matching analysis (PSM) was employed, which included all possible variables. We performed caliper matching within a range of 0.2 multiplied by the standard deviations on the PS logit 10 . A Kaplan-Meier plot was constructed and log-rank test was adopted to compare the survival. A P value of < 0.05 was considered to be statistically significant. All analyses were performed with the aid of SPSS version 18.0.

Results
Overall and propensity-score-matched Patients' characteristics were listed in detail in Table 1. Of the total 239 patients, 211 were de novo HCC cases and 28 were recurrent cases. There was imbalance in some variants between the two LT groups ( Table 1). The median pre-transplantation CA19-9 level was 36.7 U/mL (range, 18.60~56.90 U/mL) in PLT group and 20.4 U/mL (range, 8.50~37.15 U/mL) in SLT group respectively (p < 0.001). 92.9% of the entire patients had HBsAg positive, 56.9% still had HBV DNA proliferation active at the time of LT. The DNA positive rate was 63.03% in PLT group significantly higher than 32.14% in SLT group (p = 0.001). PLT group also had a worse liver functional reserve before LT (Child-Pugh A 36.02% compared with 60.71%, p = 0.026). No other characteristics showed significant differences between groups.
PSM of all possible variants generated 23 matched pairs of PLT and SLT. No significant differences were found between matched cases (Table 1).
Factors recruited in multivariate model are listed in Tables 2 and 3. According to cox regression model, factors associated with RFS and OS include serum AFP level, tumor mass and Milan stage (Tables 2 and 3). Rapamycin didn't manifest significant effect on OS or RFS.
The entire median follow-up was 35 months (range, 11~52 months). The entire median follow-up was 35 months (range, 11~52 months). The entire median OS was 35 months, with the 1 yr, 3 yr and 5 yr OS respectively 80.00%, 63.86% and 59.02% in PLT group while 65.22%, 52.99% and 42.39% in SLT group. The entire median RFS was 32 months. The 1 yr, 3 yr and 5 yr RFS were 66.96%, 57.86% and 54.95% significantly better in PLT group than 48.25%, 32.17% and 32.17% in SLT group (Table 4; Fig. 2B). No significance was found between two groups in OS (P > 0.05, Table 4), but a trend of poorer outcome can be seen from Kaplan-Meier survival curves ( Fig. 2A). On multivariate analysis, SLT is significantly associated with poorer RFS (HR 1.98, 95% CI 1.05-3.72, p = 0.035) ( Table 4). Its unfavorable effect on OS is also significant (HR 2.17, 95% CI 1.18-4.01, p = 0.013) (  Fig. 2C and D). When compared through multivariate Cox regression, SLT turned out to be a significantly unfavorable factor for RFS (OR = 3.53, 95% CI = 1.02-12.20, p = 0.037) and OS (OR = 2.84, 95% CI = 1.06-7.67, p = 0.038) ( Table 4). If we set apart those within Milan criteria (n = 126) and those who exceed the criteria (n = 113), we will find interestingly that SLT is a significant discordant factor on both OS and RFS in patients within Milan criteria ( Fig. 3A and B). When matched by propensity scores of all the possible factors as mentioned before, gap between SLT and PLT become even deeper. After PSM, the 10 matched PLT patients had a median follow up time of 53  months compared to 14.5 months of their SLT counterpart, and all remain alive without recurrence at the end of follow up. SLT patients had a recurrence rate at 30% with median RFS at 3.5 months (Fig. 4A and B). For those beyond Milan criteria, no significant difference was told between two groups. 50% of the cases obtained recurrent tumor at around 9 months after transplantation, 50%of them died before the 26 th month after surgery (Figs 3C,D and 4 C,D).

Discussion
Though, The 5-year overall survival of HCC has reached 75% after liver transplantation 11 , against the background of organ shortages, improvements in the prognostic tools for predicting outcomes after LT for HCC are necessary. In our study, we found that both OS and RFS were significantly worse in SLT groups than in PLT group, when compared under a PSM model. This result is contrast to most of the conclusion in other researches. Such divergence may credit to a variety of factors. Generally, different method in measurements and statistical analysis can come to different conclusion. Over the past decades, restrictive Milan criteria has been adopted to select HCC patients for LT. LT fell within Milan criteria came out in excellent prognosis: expected 4-year survival rate of 85% and an HCC recurrence-free survival rate of 92% after LT 3,12 . So, this study employed Milan criteria to grade our cases. We also utilized PSM to reduce selecting biases, particularly for those risk factors once proved to associate with poorer outcomes like AFP level, tumor mass and Milan criteria, etc. In the past researches, however, grading was not all based on Milan criteria. Moreover, since most preoperative scoring was only based on radiological images, it added to the potential bias of underestimation. It has been reported that underestimation of Milan criteria happened from time to time 13,14 . In this study, all those scorings were validated by measurement on dissected samples, which added accuracy to the grading.
Chronic HBV infection is the most common predisposed liver diseases in Chinese HCC patients, with 100-fold more likely to develop this malignancy 15 . The prestransplantation HBV-infectious rate is fairly high at our center at 92.89% entirely, respectively 89.29% in SLT population and 93.36% in PLT population, which was comparable with each other. We failed to find an association of either HBV infection history or positive tilter of virus DNA before transplantation with the increasing risk of tumor recurrence. This absence of correlation may either lie in lack of HBV negative patient group, or potent and regular anti-virus treatment after transplantation.
As for salvage transplantation itself, in the past 5 years, many articles, meta-analysis and reviews were published focusing on this topic. PLT was proved to be significantly associated with better OS or RFS only in 4 researches 8, [16][17][18] . However, most studies showed a literally improved OS or RFS for PLT rather than SLT 6,7,9,[19][20][21][22][23] . Dispute might lie in the heterogeneous nature in SLT definition or process. First, in some researches with optimistic results, later SLT was performed as a rescue of deteriorating liver function rather than recurrence 8 . A recent research has verified that de principe SLT has greater OS or RFS than LT performed following recurrence diagnosis 24 . We therefore narrow the recruiting standard of SLT in this study to HCC recurrence only. Second, the survival was either calculated from the point of LR or SLT 7,8,20,22,23,25 . Though, with consideration of all-dimensional analysis of patients' prognosis, the intention-to-treat model was employed by some researches, a calculation based on individual survival can't reflect the utilization efficacy of every donor. In order to better serve our research purpose, we chose to calculate OS or RFS from the final transplantation surgery.
It has been reported that the presence of microvascular invasion increased hazard ratios of recurrence in both resection and liver transplantation 12,26 . Due to the absence of more detailed pathological information, this factor was pitifully omitted from this research. Finally, since this research adopted a retrospective per-protocol model, it inevitably had the risk of selection bias. A future prospective study targeting larger samples is expected to give more reliable conclusion.

Conclusion
In conclusion, this study demonstrates that HCC recurrence and survival after SLT is significantly inferior to after PLT. These findings go against the common realized concept that SLT is as good as PLT. A wide feasibility of SLT should be given a second consideration.