LIV_1282.pdf

Clinical StudiesComparison of staging systems to predictsurvival in hepatocellular carcinomaSonia Pascual1,Pedro Zapater2,Jose Such1,Antonio Garc a-Herola1,Laura Sempere1,Javier Irurzun3,Jose Mar a Palazo n1,FernandoCarnicer1and Miguel Pe rez-Mateo11Liver Unit,2Clinical Pharmacology,3Interventional Radiology Unit,Hospital GeneralUniversitario de Alicante,Alicante,SpainPascual S,Zapater P,Such J,Garca-Herola A,Sempere L,Irurzun J,Palazo n JM,Carnicer F,Pe rez-Mateo M.Comparison of staging systems topredict survival in hepatocellular carcinoma.Liver International 2006:26:673679.r2006 The Author.Journal compilation r 2006 Blackwell MunksgaardAbstract:Purpose:Some new staging systems in hepatocellular carcinoma(HCC)have been described in the last years.The aim of this study was tocompare the survival-predicting capacity of some variables and theprognostic classifications.Methods:Demographic,clinical,analyticalvariables and tumour characteristics were collected in a study including 115patients with HCC.Predictors of survival were identified using the KaplanMeier test and the Cox model.Comparison between different staging systemswas carried out.Results:The 1-,2-and 3-year estimated survival was 65%,45%and 30%,respectively.ChildPugh score and a-fetoprotein level greaterthan 400UI/l were independent predictors of survival in the Cox model.Although all systems correctly differentiated between patients regardingsurvival(KaplanMeier,log rank o0.05 for all),the Barcelona Clinic LiverCancer(BCLC)showed a better discriminatory ability than the otherevaluated scores.In addition,the independent homogenizing ability andstratification value of BCLC was better than that of other systems.On thecontrary,model for end-stage liver disease(MELD)showed the worst results.Conclusions:ChildPugh score and a-fetoprotein levels were the onlyindependent predictors of survival in patients with HCC.ChildPugh scoreshowed a better prediction value for survival when compared with MELD.BCLC is more accurate than the other prognostic models evaluated in thisinvestigation.Key words:hepatocellular carcinoma prog-nosis staging systemDr.Sonia Pascual,Liver Unit,Hospital GeneralUniversitario de Alicante,C/.Pintor Baeza s/n.,03010 Alicante,Spain.Tel:134 965938350Fax:134 965938345e-mail:pascual_songva.esReceived 22 November 2005,accepted 25 March 2006?Hepatocellular carcinoma(HCC)arises on cir-rhotic liver in the vast majority of cases,oftenwith severe hepatic insufficiency that may affectsurvival per se(13).Considerable efforts havebeen made to identify the best prognostic classi-fication in patients with HCC and to unify criteriato accurately stratify patients,as shown in the2000 European Association for the Study of theLiver Consensus Conference held in Barcelona(4)and in AASLDs AHPBA/AJCC 2003 Meet-ing(5).Several models have been developed byseveral groups(613),but to date the results arenot uniform(1418)and different authors showdiscrepant data.Model for end-stage liver disease(MELD)is a new score initially designed to assessshort-term prognosis(3 months)in cirrhoticpatients being considered for transjugular intra-hepatic portosystemic shunt(TIPS)(19).Other-wise,MELD has also been useful in predictingshort-and medium-term prognosis(1 year)inpatients with cirrhosis(20,21),and in patientsrecovering from an episode of variceal bleeding(22).Supporting these facts,the United Networkfor Organ Sharing(UNOS)introduced MELDfor selection of candidates and allocation oforgans in patients awaiting liver transplantationin the United States.Prospective studies gavefavourable results(2325),although discrepan-cies exist(2628).Preliminary data suggest thatMELD allows a better managing of patients withHCC awaiting liver transplantation,but the beststrategy to follow remains to be defined(29,30).Little information is available regarding the use-fulness of the MELD score in predicting survivalin patients with HCC that are not candidates forliver transplant.The aim of this investigation was to comparethe survival-predicting capacity of available prog-Liver International 2006:26:673679r 2006 The AuthorJournal compilation r 2006 Blackwell MunksgaardDOI:10.1111/j.1478-3231.2006.01282.x673nostic classifications scales for HCC:CancerLiver Italian program(CLIP)(14),France classi-fication(8),BarcelonaClinicLiverCancer(BCLC)(9),Okuda(6)with MELD(31)andChildPugh score(32).MethodsSince January 1996,all patients diagnosed ofHCC in our area are currently enrolled in adatabase that includes demographic,analyticand clinical data,tumour characteristics,treat-ment modality,response to treatment(whenavailable),clinical course and end of follow-up(death or lost).Between November 2000 andFebruary 2004,patients were included in thepresent study and data were collected until Au-gust 2004 to complete a minimum follow-up termof 6 months.According to the conclusions of the consensusconference of the European Association for theStudy of Liver Disease,diagnosis of HCC isbased upon histological confirmation,two con-cordant imaging studies(computed tomography(CT)and magnetic resonance(MR)or oneimaging study in addition to an a-fetoproteinlevel greater than 400ng/ml(4).To assess thesize,number and location of tumours,and theexistence of portal vein thrombosis,abdominalCT and/or MR were performed.According totumour characteristics,and following the strate-gies recommended in the EASL consensus con-ference,therapeutic options considered in ourprotocol of care of patients with HCC were:(1)liver transplantation for patients younger than 65years old,with a solitary tumour?5cm or threenodules?3cm in diameter,without vascularinvasion or extrahepatic dissemination(33),(2)percutaneous ethanol injection(PEI)or radio-frequency thermal ablation(RF)in patients non-suitable for liver transplantation with small tu-mours(between less than 3.5cm),(3)transarterialchemoembolization(TACE)for patients withlarge/multinodulartumourswithoutportalthrombosis and preserved liver function and(4)symptomatic treatment for end-stage patients.In order to facilitate statistical analysis of thedifferent scoring classifications,the followingmodifications were carried out:first,patientswere stratified in groups regarding MELD scoresimilarly to what was previously reported byothers(34):o10 or?10 points;second,patientsincluded in group A of the BCLC score were notsubdivided further into stages A1A4;third,patients in categories 46 of the CLIP scorewere included in the same sub-group.Statistical analysisOverall survival was the only end point used inthe analysis.Survival was defined as the differ-ence(in months)between the diagnosis of HCCand death,or the finalization of the follow-upperiod(August 2004).Lost patients during thefollow-up period were censored at the last out-patient visit or the last known episode of hospi-talisation.Univariatesurvivalcurveswerecalculated using the KaplanMeier test andwere compared by means of the log rank test.Multiple comparisons were controlled using a Pvalue adjusted to the Bonferronis method.Wealso studied the relation between clinical featuresat diagnosis of HCC and survival in the univari-ate analysis.Pretreatment variables that weresignificant predictors of survival in the univariateanalysis were included in a multivariate analysis(Cox proportional hazard model)in order toestablish the relation with prognosis and survival.Following Ueno et al.(15)and Cillo et al.(16),the performance of the prognostic system wasrelated to(1)the homogeneity within classifica-tion groups(differences in survival must be smallbetween patients in the same sub-group),(2)thediscriminatory ability measured by the differencesamong groups(there must not be much greaterdifferences in the survival time among patientsclassified in the different sub-group)and(3)themonotonicity of gradients shown in the associa-tion between stages and survival rates(the meansurvival time for a group classified as favourablemust always be longer than the survival noted inthe less favourable groups).The likelihood ratio(LR)related to a Coxs proportional hazardregression model was used to evaluate the homo-geneity within categories of each classificationsystem.Using the prognostic scores as ordinaryvariables,this analysis was used to estimate themonotonicity of gradients.In order to neutralisethe potential bias in comparing prognostic scoreswith a different number of stages,the results ofCoxs regression are also expressed using theAkaike information criterion(AIC)(the smallerthe value,the better the model).The discrimina-tory ability of the staging systems and the mono-tonicity of gradients in mortality rates wasquantified using linear trend w2test.We evaluatedthe independent contribution of the six classifica-tions to explain survival comparing the LR testand the AIC values related to the full model withthe same values calculated in a reduced modelwhere alternatively the covariants related to theindividual classification were removed.The study protocol conformed to the ethicalguidelines of the 1975 Declaration of Helsinki674Pascual et al.and was conducted with the approval of theInstitutional Ethics Committee.Informed con-sent was obtained from each patient.ResultsPatient characteristicsClinical and main analytical features of all 115patients included in this study together with themodality of treatment used in every case arereflected in Table 1.a-fetoprotein level was higherthan 100 or 400U/l in only 27%and 15%ofpatients,respectively.HCC was diagnosed in thecourse of a surveillance program(serum a-feto-protein and ultrasonography every 6 months)in54%of the patients.Seventy percent of patientsrequired histological confirmation of HCC.Overall survivalMedian follow-up and survival was 14 months(range 0.540)and 20 months(95%CI 18.1424.28),respectively,and the estimated survival at12,24 and 36 months was 65%,45%and 30%,respectively.No patient was lost during the fol-low-up term and 64 died during the study period.Thirty-two patients(50%)died of liver insuffi-ciency,which was the commonest cause of death.Eight patients died of complications occurringafter the application of therapy:one case of sepsis,one haemoperitoneum after alcohol injection,twoinfectedpleuraleffusionpost-radiofrequency,three cases of variceal haemorrhage after TACEand one case of duodenal ischaemia after TACE.The reasons for death in the rest of the patientswere variceal haemorrhage,spontaneous haemo-peritoneum,pulmonary metastases and a miscel-laneous group that included septic complications,hepatorenal syndrome,encephalopathy and re-currence of cirrhosis in a patient with a previousliver transplantation.Survival analysisThe correlation between survival and clinicalfeatures at diagnosis is detailed in Table 2.Thevariables that independently predicted survival inthe multivariate analysis were ChildPugh scorewith a hazard ratio of 2.400(95%CI 1.414.07,P50.001),and the level of a-fetoprotein greaterthan 400UI/l with a hazard ratio of 4.57(95%CI2.528.29,P50.000).As expected,all modelsshowed a significant relationship to survival ac-cording to the KaplanMeier method(Table 3).Score systems designed to predict prognosis ofpatients with cirrhosis(ChildPugh and MELDscores)and those specifically developed to evalu-ate patients with HCC(BCLC,CLIP,France andOkuda)were considered separately.Figure 1shows how both ChildPugh and MELD scoresaccurately discriminate patients included in earlyand intermediate stages of cirrhosis.In our series,the France classification discriminated betweenintermediate and advanced stages(Po0.05)butnotbetweenearlyandintermediatestages(P50.6),while the Okuda classification did(Okuda 1 vs.2,P50.009,2 vs.3,P50.000).We could not show differences in the survival ofour patients in the different stages of CLIP score.On the contrary,BCLC was a good scale todistinguish between early and intermediate stagesand also between advanced and terminal stages,although differences were not found between theintermediate and advanced stages in our cohort(A vs.B,P50.00001;B vs.C,0.63,C vs.D,P50.0002).As shown in Table 4,BCLC stagingTable1.Clinical and analytical baseline characteristics of the overallgroup of patientsMean age(range)67.97(3984)Male sex72%Etiology of cirrhosis,n(%)HCV55(48)Alcohol30(26)Alcohol1virus15(13)HBV6(5)Others9(8)Albumin(g/dl)3.2?0.6Bilirubin(mg/dl)1.79?1.48AST(UI/l)94?70ALT(UI/l)74?68GGT(UI/l)199?308AP(UI/l)236?173Prothrombine activity(%)76?14.69International normalized ratio1.2?0.22a-fetoprotein(ng/ml)1921?9918Portal vein thrombosis,n(%)15(6)Tumor typen(%)Nodule o5cm45(39)Nodule?5cm24(21)Two nodules18(16)Three nodules10(9)Multiple nodules18(16)Treatment,n(%)Trasplantation9(8)Percutaneous ethanol injection13(11)Radiofrequency ablation16(14)Chemoembolisation39(34)Supportive care38(33)ChildPugh classification,n(%)A55(48)B45(39)C15(13)Ascites,n(%)57(50)Encephalopathyn(%)9(8)Data are expressed as mean?standard deviation(SD).Numbers inparentheses indicate percentages.Ascites and encephalopathy indicatesits presence at admission.AFP,a-fetoprotein;HCV,hepatitis C virus;HBV,hepatitis B virus.675Staging systems in hepatocellular carcinomahas the highest value in the linear trend w2test(35.4),confirming a better discriminatory abilitythan all other evaluated scores.In addition,theindependent homogenizing ability and stratifica-tion value of BCLC,as investigated by means ofthe LR test and Akaike within Coxs propor-tional hazard regression model,was better thanthat of other systems(w259.18,AIC 523.18).Onthe contrary,MELD showed the worst results.BCLC also provided the highest independentcontribution to the Coxs regression model in-cluding all classifications(Table 4):removal ofthe BCLC did significantly reduce the goodnessof fit of the model(AIC 534.38,LR w260.93).Table2.Univariateanalysisofvariablespredictiveofsurvival(KaplanMeier method and log rank test)VariablenMediansurvival(months)PAge(o65/?65)34/8119.63/21.90.51Sex(M/F)83/3220.9/21.70.91Etiology0.42Alcohol3020.6HVC5523.6HVB617Alcohol1virus1518.7Miscellaneous914.8ChildPugho0.0001A5528.7B4516.5C158.2MELD0.0002o106629?104911Noduleso0.0001o5cm4531.6?5cm2414Two nodules1816.9Three nodules1821.8Multicentre106Ascites(no/yes)58/5725.37/16.890.009Encephalopathy(no/yes)106/921.8/13.780.17PVT(no/yes)100/1522.89/15.790.21AFP(o400/?400UI/l)88/2726/8.8o0.0001Albumin(o3.5/?3.5mg/dl)40/7524.7/18.90.058Bilirrubin(o1.5/?1.5mg/dl)65/5026.6/14.3o0.0001Prothrombin activity ratio(?65%/o65%)86/2922.9/16.90.10Creatinine(o1.1/?1.1mg/dl)85/3020.8/22.60.82M,male;F,female;HCV,hepatitis C virus;HBV,hepatitis B virus;PVT,portal vein thrombosis;AFP,a-fetoprotein.Values that predict survivalare given in bold.Table3.Patient survival according to the BCLC,Okuda,CLIP andFrench classification(KaplanMeier method and log rank test)StagingsystemnMediansurvival(months)Deaths(n)Deaths(%)PBCLCo0.0001A4033.21025B3418.62161C2613.91818D153.215100Okudao0.0001I3730.181335II6320.363657III153.2315100CLIPo0.000103234.572213324.12154521914.8136831911178946123.8712100Franceo0.