A meta-analysis of adjuvant therapy after potentially curative treatment for hepatocellular carcinoma

*Co-First Authors. 1Department of General Surgery, The First Hospital of Lanzhou University; 2Department of General Surgery, Lanzhou University Second Hospital; 3Department of Burn Surgery, People’s Hospital of Gansu Province, Gansu, China Correspondence: Dr You Cheng Zhang, Department of General Surgery, Lanzhou University Second Hospital, Lanzhou 730030, Gansu, China. Telephone 86-931-894-2287 , fax 86-931-845-8109, e-mail zhangychphd@yahoo.cn Received for publication April 10, 2012. Accepted October 25, 2012 Worldwide, hepatocellular carcinoma (HCC), the most common primary cancer of the liver, ranks sixth among malignant tumours in incidence and is the third leading cause of cancer-related death (1). In China, owing to the high prevalence of hepatitis B viral infection and associated liver cirrhosis, HCC accounts for >54% of the world annual incidence, with an estimated 372,079 mortalities in 2008 (1-3). For many years, surgery (hepatic resection and transplantation) has been considered the only curative option for HCC. Locoregional ablation therapy, particularly percutaneous radiofrequency ablation (RFA), was recently demonstrated to have similar efficacy to surgical resection for small HCC nodules (<3 cm in diameter) (4). Although these treatments offer a possibility of cure for HCC, the long-term outcomes after surgery or ablation therapy are disappointing because of the high frequency of recurrence. HCC often recurs as a result of intrahepatic dissemination of primary cancer cells or the development of a de novo tumour in the cirrhotic liver, which, in total, complicates 70% of cases at five years (5-7). However, because it is difficult to distinguish the two types of HCC recurrence in routine clinical practice, the presence of tumour originating from either primary or metachronous multicentric carcinogenesis is regarded as a recurrence (8). Therefore, adjuvant therapy that aims to reduce or delay the incidence of recurrence by eradicating the growth of pre-existing minute tumour foci not detected before initiation of adjuvant therapy, or by inhibiting the occurrence review

Several adjuvant modalities have been developed in the past decades; nevertheless, the clinical use of these therapies is either controversial or requires further evaluation (9).To date, there have been five reviews or meta-analytic studies (10)(11)(12)(13)(14) published between 2002 and 2009 assessing the role of neoadjuvant and/or adjuvant therapy for HCC.While the early analyses (10)(11)(12) had achieved modest survival benefits from some adjuvant therapies, two recent studies with updated evidence (13,14) only applied qualitative descriptive approaches for therapeutic evaluation.Furthermore, these studies (10)(11)(12)(13)(14) neither considered modalities other than surgical resection as potentially curative treatment for HCC, nor quantitatively measured the effects of adjuvant therapies on survival using time-to-event outcomes.On the other hand, although previous individual meta-analyses have also shown encouraging results with adjuvant chemotherapy (15,16), adjuvant immunotherapy (17)(18)(19)(20)(21)(22)(23) and vitamin analogue chemoprevention (24), some of these may be inconclusive or of doubtful accuracy because of bias due to limited sample size and statistical methodological flaws (15,16,24).Nevertheless, emerging evidence from randomized controlled trials (RCTs) of novel adjuvant modalities and from trials with updated information is currently available.Given that no therapies after potentially curative treatments have been accepted as standard of care in HCC to date, the effects of adjuvant therapies on recurrence and survival are even less clear.Using currently available RCT evidence and knowledge of survival analysis (25)(26)(27), the aim of the present study was to evaluate the efficacy of adjuvant therapies of any modality in terms of the hazard ratio (HR) of recurrence-free survival (RFS) and overall survival (OS), and the safety of these adjuvant therapies after potentially curative treatment with surgical resection, liver transplantation or ablation therapy for HCC.

Literature search
The Cochrane Library, MEDLINE (via PubMed) and Embase were searched using the keywords "hepatocellular carcinoma" and "recurrence".Both medical subject and text terms were used and combined, and the search strategy was not restricted to languages or publication date.However, only RCTs were considered in the present metaanalysis, for which searches in MEDLINE and Embase were limited by study type.The searches were performed mainly in June 2011, and the result was updated in November 2011.In addition, published metaanalyses and reviews of relevance were scrutinized for other potential studies, and reference lists of included trials were manually searched.

inclusion and exclusion criteria
RCTs published as full text assessing adjuvant therapy in patients with HCC who had undergone potentially curative treatment with surgical resection, hepatic transplantation or locoregional ablation therapy were eligible if recurrence-related outcomes were analyzed using survival analysis.However, to guarantee that adjuvant therapy was truly adjuvant, only trials in which randomization was performed after patients had been treated with initially curative treatment were considered.Trials concurrently comparing sequential combinations of curative and adjuvant therapies versus curative treatment alone were excluded because adjuvant effects in such trials could not be separately evaluated due to the presence of initially curative treatment.Similarly, given the nature of neoadjuvant trials for which patient allocation is achieved before curative treatment, neoadjuvant therapy was not included in the present study.Furthermore, because no well-accepted adjuvant therapeutic modality has been currently established, the meta-analysis only compared adjuvant therapy with no active adjuvant therapy (no treatment or concurrent placebo).Other exclusions were trials with nonrandomized design, studies involving noncurative or palliative treatment for HCC, and trials comparing different adjuvant therapies or different schedules of one adjuvant therapy.

study selection and outcomes measurement
All references records retrieved from the searches were stored in an EndNote (Thomson Reuters, USA) file and duplicates were removed.Two authors independently assessed eligibility against the inclusion criteria by scanning the title and abstract of each record, with disagreement resolved by discussion.Where studies had multiple publications, the most recent report was included and secondary articles were also considered.Trials containing three or more study groups were retained if at least two groups addressed an eligible comparison.
The primary outcome was RFS, which is also referred to as diseasefree survival (DFS).However, for studies in which neither of the two outcomes was reported, time to recurrence (TTR) was used as a surrogate outcome.The secondary outcomes were OS and side effects.

Data extraction and analysis
Information regarding adjuvant therapy protocols of each trial was extracted and tabulated, along with tumour histological factors (size, number and vascular invasion) and staging of HCC, and underlying liver disease, which are the variables that are the most important predictors of recurrence and survival (28)(29)(30).Details of methodological quality assessment of the included RCTs were also abstracted.The number of patients developing events (recurrence or death) during follow-up was recorded.Locations and types of recurrence, and reasons for death were also presented.Furthermore, to determine the risk of patients developing events, recurrence and survival rates at various time points (one, two, three, four and five years) for patients in the control group (did not receive an active adjuvant therapy) were reported.
The HR of time-to-event outcomes (ie, RFS, TTR and OS) was directly extracted from trial publication, if available, or was estimated indirectly using the reported number of events and the corresponding P value for the log-rank statistics, or by reading survival curves, as described by Parmar et al (25).An Excel spreadsheet (Microsoft, USA) developed by the Meta-analysis Group of the MRC Clinical Trials was used for the calculations (26).
A pooled analysis was performed for RCTs testing a similar modality of adjuvant therapy, for which all trials included were analyzed in several subgroups to assess their effects on RFS and OS.Sensitivity analyses were performed to examine effects of excluding extraordinary studies with either quality concerns or confounding data.In the safety analyses, however, because the criteria for assessment varied across trials, only side effects requiring a discontinuation of adjuvant therapy were reported.All outcomes data were extracted on worksheets and were cross-checked for accuracy by two authors before combining for analysis.

Quality assessment
The methodological quality of RCTs was assessed independently by two authors, using the criteria outlined in Cochrane Handbook for Systematic Reviews of Interventions (Version 5) (31).Any disagreement was resolved by discussion or by consulting a third author.Six items assessing components of internal validity of RCTs were applied: generation of random sequence, allocation concealment, completeness of outcomes data reporting, whether free of selective reporting, and other bias including whether balanced in baseline characteristics and whether there was a priori sample size calculation.Blinding was, however, removed in the assessment of risk of bias, given the impracticability to mask adjuvant therapy in most trials using no treatment as a comparator.Each item was scored as 'yes' for low risk of bias, 'unclear' for either lack of information or uncertainty over the potential for bias, and 'no' for high risk of bias.

statistical analysis
Time-to-event outcomes were combined using the inverse variance method in Review Manager Software (Version 5.1.1,The Nordic Cochrane Centre, The Cochrane Collaboration) with a fixed-effect model.A pooled HR represents the overall risk of an event on adjuvant therapy over control, in which HR <1.0 favoured adjuvant therapy and HR ≥1.0 favoured control.Respective 95% CIs were calculated for each estimate.Statistical heterogeneity of the results of the trials was assessed using the χ 2 statistic and the proportion of variation due to heterogeneity was expressed as I 2 , where I 2 <25% is considered to be low-level heterogeneity, 25% to 50% as moderate-level, and I 2 >50% as high-level heterogeneity (32).Recurrence and survival rates were expressed as median values with the minimum to maximum range.

resuLts search results and characteristics of the included rCts
The databases searches indentified 527 records, and another two were obtained from references lists.Of these, 474 were initially judged as irrelevant, yielding 55 potential study reports that were assessed for eligibility at full-text level, resulting in retrieval of another prospective trial.Finally, 32 publications corresponding to 28 RCTs met the inclusion criteria, from which 31 publications (33-63) of 27 trials, involving a total of 2614 randomized patients, were eligible for analysis (one study [64] was excluded due to insufficient reporting of survival outcomes data).Figure 1 summarizes the study identification flow according to the recommendations of the PRISMA statement (65).
The searches identified three publications of one trial by Kubo et al and others (42)(43)(44).Although the latest report (44) provided final outcomes data, the previous two publications (42,43) were also considered for analysis because they reported other important information of the trial accrual.The polyprenoic acid study by Muto et al (50,51) and internal radiation study by Lau et al (60,61) also presented two *Two reports from one trial were presented in 1999 and in 2008, respectively; † Includes ablation therapy with or without transhepatic arterial therapy; ‡ In terms of patients analyzed; § One study reported time to second primary hepatocellular carcinoma; ¶ According to presented recurrence curves.RCT Randomized controlled trial unduplicated articles; hence, data from both reports of each of the studies were used, separately (50,51) or in combination (60,61), in the outcomes analysis.Six RCTs (33,46,49,55,58,59) involved three or more study groups.In two of these (46,55), because no difference in recurrence rates and in RFS was found between two adjuvant groups, respectively, data from both adjuvant groups were combined for analysis.However, for the other four trials (33,49,59,63), due to the lack of combined outcomes data, only selected adjuvant groups were compared with control.The HR of RFS, TTR or OS was estimated using indirect methods (25,26) for most of the RCTs, because only a few studies (34,35,55,59) directly provided data for these time-to-event outcomes.Overall, 13 trials (33,(35)(36)(37)46,49,50,52,54,(56)(57)(58)(59) found adjuvant therapy could confer a significant advantage in RFS or TTR, and six (38,44,48,51,59,62) indentified a significant OS gain with adjuvant therapy.In four other trials (34,39,40,55), however, adjuvant therapy demonstrated worse outcomes compared with no adjuvant therapy, although differences in RFS or OS were insignificant.

efficacy and safety evaluation of adjuvant therapy
For assessment and analysis of adjuvant therapies for which there was potential for clinical benefit, these 27 RCTs were classified into five categories: chemotherapy, IFN therapy, vitamin analogue therapy, adoptive immunotherapy and other therapies (each therapy was tested in a single trial) including cancer vaccine therapy, internal radiation therapy, radioimmunotherapy and heparanase inhibitor PI-88 therapy.A detailed description of treatment protocols, study characteristics and methodological quality assessment of the RCTs are presented in Tables 3 to 12.  (35,56,57,59) of which also included TTR as a primary outcome.However, for the other nine trials (41,45,46,50,(52)(53)(54)62,63) that reported recurrence or recurrence-free rates, TTR was used as surrogate outcome.Eight subgroup analyses were performed for RFS (Figure 2).Chemotherapy: The meta-analysis for eight trials (33)(34)(35)(36)(37)(38)(39)(40) demonstrated that adjuvant chemotherapy failed to improve RFS compared with no treatment (HR 0.94 [95% CI 0.81 to 1.10]).Nevertheless, a significant statistical heterogeneity among the trials was found (I 2 =55%).After excluding any of these eight trials by sensitivity analysis, heterogeneity was consistently high (data not shown).Therefore, the eight trials were further subdivided according to the different administration routes of chemotherapy (Figure 3), and the results showed that adjuvant oral chemotherapy (HR 0.  (41) appeared to be the source of the heterogeneity.This was the only trial identified that used IFN-β, and the sample sizes were small (10 patients each in IFN and control groups); it was also restricted by methodological limitations.The sensitivity analysis excluding the study (41) showed that pooled HR was 0.75 (95% CI 0.61 to 0.92; Figure 2), significantly favouring adjuvant IFN therapy with acceptable heterogeneity (I 2 =5%).vitamin analogue therapy: The pooled analysis for five trials testing polyprenoic acid (50,51) or menatetrenone (52-55) indentified a significant RFS gain with these vitamin analogues (HR 0.76 [95% CI 0.60 to 0.96]; data not shown).However, an elevated heterogeneity was observed across the studies (I 2 =78%), which was likely due to the inclusion of the VK2 study by Yoshida et al (55).Despite having the largest sample size (n=548), this trial (55) enrolled 21% patients with first intrahepatic recurrence of HCC and was terminated after a maximum follow-up of only 36 months, all of which may confound the combined result.Nevertheless, restricting the analysis to the other four RCTs (50-54) did not change the effects of adjuvant vitamin analogue therapy on RFS (HR 0.43 [95% CI 0.30 to 0.62]; Figure 2), while heterogeneity was eliminated (I 2 =0%).
Adoptive immunotherapy: The pooled HRs for two trials (56,58) testing adoptive immunotherapy demonstrated there was no significant difference between adjuvant and no adjuvant groups in terms of OS (HR 0.79 [95% CI 0.55 to 1.14]).An acceptable heterogeneity was observed between the trials (I 2 =25%).other therapeutic modalities: The HR of OS for the trial of cancer vaccine therapy (59) was 0.07 (95% CI 0.02 to 0.27), and that for the radioimmunotherapy study (62) was 0.26 (95% CI 0.08 to 0.83), each of which significantly favoured adjuvant therapy.However, OS in the internal radiation trial (60,61) did not differ significantly between adjuvant and the control groups (HR 0.68 [95% CI 0.30 to 1.55]).

DisCussioN
HCC is a major health problem in hepatitis-prevalent countries such as China (1-3).Despite great improvement in diagnostic and  therapeutic techniques, the long-term outcomes of HCC remain unsatisfactory, even after treatment with cure intent.From our analysis for patients receiving no active adjuvant therapy, the median five-year recurrence rate after surgical resection, liver transplantation or ablation therapy was up to 82% (range 55% to 94%), and that for survival rate at five years was 48% (range 29% to 75%).The high recurrence rate was explained by intrahepatic metastasis of primary HCC, which accounted for >79% of recurrence cases; the main cause of death after these potentially curative treatments was recurrent HCC (67% to 100%) that occurred mostly in residual liver.
The current meta-analysis included a broad spectrum of adjuvant therapies with either antitumour or chemopreventive effects to be evaluated.Adjuvant chemotherapy failed to confer any benefit to RFS and OS, regardless of what agents or administration modes were used.On the contrary, the combination of systemic and transarterial chemotherapy may have a deleterious effect on patient prognosis.Although the efficacy results of the adjuvant IFN studies were encouraging, one caveat is that the survival benefits from IFN therapy should be weighed against the risks of adverse effects.The findings of chemoprevention with vitamin analogues require further examination because the meta-analysis of RFS and of OS was performed only in a limited number of small RCTs with methodological weaknesses.Adoptive immunotherapy with either LAK or CIK cells may be as promising a strategy as adjuvant therapy for HCC because it increased the RFS by 45% and resulted in few adverse effects, although the approaches are too cumbersome and costly for use in large clinical trials.Nevertheless, follow-up periods in these trials (56)(57)(58) were too short to confer a statistically significant benefit in OS.Postoperative adjuvant therapy using either transhepatic arterial 131 I-lipiodol infusion or subcutaneous injection of heparanase inhibitor PI-88 did not appear able to decrease recurrence and to improve survival, while cancer vaccine therapy and Licartin radioimmunotherapy showed some promise after radical surgery for HCC.However, each of these modalities was examined in single trial with small-scale and preliminary settings.
The results of the current meta-analysis suggest that the effectiveness of an adjuvant modality depends largely on front-line therapy for HCC.In fact, because the cancer is insensitive to chemotherapy and radiotherapy (66), adjuvant use of any chemotherapeutic agent or radioactive material may not be effective.Furthermore, the beneficial effects of adjuvant IFN therapy on recurrence and survival may contribute to its efficacy in preventing hepatitis from developing into HCC (67,68).On the other hand, the lack of effective therapeutic agents remains the main challenge in the provision of adjuvant therapy for HCC (9) because the presence of underlying cirrhosis limits the capability of remnant liver insulted by initially curative treatment to tolerate any adjuvant cytotoxic therapy.Nevertheless, because intrahepatic recurrence can either represent metastasis from primary HCC or de novo tumour formation in a cirrhotic liver, an agent or regimen with both tumouricidal and chemopreventive effects, but less toxicity, may be effective to prevent recurrence after curative treatment for HCC (7,61,69).In the current analysis, however, the effect of adjuvant therapy on metachronous de novo carcinogenesis may have been obscured because of the presence of recurrence resulting from intrahepatic metastases of HCC, and because most of the trials lacked a precise description of the difference between primary metastasis and de novo HCC.For future trials, therefore, measures for Figure 3) Subgroup analysis of recurrence-free survival in randomized controlled trials evaluating adjuvant chemotherapy Figure 2) Meta-analysis (sensitivity analysis) of recurrence-free survival in randomized controlled trials evaluating adjuvant therapy metastatic recurrence in the residual liver may be necessary, and molecular diagnostic techniques, such as comparative genomic hybridization, integration pattern of HBV, or DNA fingerprinting using loss of heterozygosity assays or microarray analysis (70-72) are recommended, if feasible, to differentiate the two types of recurrence.
Theoretically, reduced recurrence is expected to improve survival.Nevertheless, several trials (38,44,48) found an improvement in OS but not in RFS.These were studies that had relatively small number of participants and insufficient follow-up, and there was little consideration of statistical power in the design.These limitations, however, may prevent or delay recognition of potentially beneficial therapies.Furthermore, effects of adjuvant therapy may differ over time.This was apparent in studies involving polyprenoic acid (50,51) and internal radiation therapy (60,61), in which the benefits in decreasing recurrence became more evident or were lost after expending follow-up visits.Therefore, larger sample sizes and longer periods of observation are emphasized for future trials.
For trials testing adjuvant therapy, any adjuvant effect should be evaluated independently from other potential effects.The database searches initially identified 19 reports on adjuvant therapies in which randomization was performed before patients underwent mainly curative resection for HCC (Figure 1).Although postoperative recurrence was addressed, these studies were not considered for inclusion because adjuvant effects in treatment group may have been confounded by the introduction of initially curative treatment.Also, it is crucial to distinguish recurrent disease after a curative treatment from residual tumour after a palliative therapy when testing an adjuvant therapy that aims primarily to decrease the recurrence of HCC.Therefore, proper documentation of curability of initial treatment for HCC is necessary.The curability in most of the trials included was defined as the complete elimination or necrosis of all macroscopically detectable tumours, with no demonstrable evidence of residual or recurrent tumours on image studies before initiation of adjuvant therapy.However, it was either not described at all, or insufficiently described by other trials (34,38,41,55,62), which may have resulted in inadvertent enrollment of patients with residual HCC before randomization.Nevertheless, given that there have been no established assessment criteria of curability after surgical resection or ablation therapy, a more stringent definition of curability is required.
One of the major weakness of the present meta-analysis is that the generalizability of the results was limited by predominantly including RCTs conducted among Japanese and Chinese patients.Conceivably, there could be differences in the natural history of HCC among geographical regions, although these potential differences have not been well understood.Therefore, it is possible that the findings of the current meta-analysis may not be extrapolated to the non-Asian population.
Another limitation is that the study failed to detect the impact of underlying liver disease, particularly cirrhosis, on survival outcomes.A meta-regression analysis could be helpful to capture the effect of liver cirrhosis on survival.Nevertheless, it is hardly believable that a metaregression analysis using cirrhosis as a covariate could affect the overall results of the current study because the majority of deaths were due to HCC recurrence rather than to liver cirrhosis (Figure 2).On the other hand, the use of RFS as primary outcome could potentially confound effects of adjuvant therapies on survival by involving death from liver failure.In the current meta-analysis, we initially planned to include TTR as the primary outcome.However, because only a few studies reported TTR (35,56,57,59) or presented data for estimating the HR of the outcome (41,45,46,(52)(53)(54)62), RFS was used as a surrogate.Furthermore, we found the difference in the estimates of the HR between TTR and RFS in the trials (35,56,57,59) that reported either of the outcomes was minor (data not shown).Again, RFS or DFS may be preferable to TTR as a correlate of OS, because it is able to capture fatal toxicity in trials where the majority of deaths are expected to be related to cancer (73).Nevertheless, given the confounding composite nature of RFS that concerns not only recurrence but also death from any cause, death resulting from the natural history of cirrhosis may confound detection of potential benefits from adjuvant therapy (8,74).Therefore, the overall results of RFS in the meta-analysis should be interpreted with care, and TTR is recommended for further adjuvant trials in HCC.
Third, the present study has the typical weakness of an aggregated data meta-analysis of time-to-event outcomes.Of the 31 study reports included, only four (34,35,55,59) directly presented HR and associated 95% CIs for RFS or OS, while results of these outcomes for most of the trials were obtained by performing calculations using other statistics or data extracted from published survival curves (25,26).Bias may have been produced in estimating the HR with the use of these indirect methods, which may, however, partially explain the statistical heterogeneity among individual comparisons.Nevertheless, in the absence of necessary statistics, a meta-analysis based on data from published curves can be the only practical alternative (25).Future publications reporting time-to-event outcomes should, therefore, provide more detailed statistical information, preferably in the form of the results of log HRs and their variances, or their estimators (75).
The current meta-analysis assessed methodological quality of the included RCTs using rigorous criteria proposed by the Cochrane systematic review (31).As a result, only nine RCTs included were assessed as having adequate quality (34,44,(47)(48)(49)(50)55,61,62), while nearly one-half of the trials were subject to methodological weakness.Hence, the overall results of the present meta-analysis were prone to bias that was incurred from the bias of these original studies.Therefore, more prospective studies of good methodological quality are needed for adjuvant HCC clinical trials.

DisCLosures:
The authors have no financial disclosures or conflicts of interest ot declare.

Figure 4 )Figure 5 )
Figure 4) Meta-analysis of overall survival in randomized controlled trials evaluating adjuvant therapy

TAble 7 Characteristics of the randomized controlled trials evaluating adjuvant vitamin analogue therapy Study (reference) Adjuvant protocol and number of patients Patients' baseline characteristics
NR C-P Child Child-Pugh class; HCV Hepatitis C virus; LCSGJ Liver Cancer Study Group of Japan; NR Not reported os Twenty-two trials

TAble 8 Methodological quality assessment of the randomized controlled trials evaluating adjuvant vitamin analogue therapy Study (reference) Random sequence generation Allocation concealment Incomplete outcome data Selective reporting Other bias balanced in baseline characteristics Sample size calculation
YesAST Aspartate aminotransferase; DCP Des-gamma-carboxy prothrombin; ITT Intention-to-treat; TBIL Total bilirubin

Child Child-Pugh class; CIK
Cytokine-induced killer; HBV Hepatitis B virus; HCV Hepatitis C virus; iv Intravenous; LAK Lymphokine activated killer;