The prognostic classification of patients with

HEPATOLOGY, VOL. 67, NO. 6, 2018 AMERICAN ASSOCIATION FOR THE STUDY OFLIVERD I S E ASES HEPATOBILIARY MALIGNANCIES External Validation of the ITA.LI.CA Prognostic System for Patients With Hepatocellular Carcinoma: A Multicenter Cohort Study Mauro Borzio, 1 Elena Dionigi, 1 Angelo Rossini, 2 Massimo Marignani, 3 Rodolfo Sacco, 4 Ilario De Sio, 5 Emanuela Bertolini, 6 Giampiero Francica, 7 Anna Giacomin, 8 Giancarlo Parisi, 9 Susanna Vicari, 10 Anna Toldi, 11 Andrea Salmi, 12 Sergio Boccia, 13 Mario Mitra, 14 and Fabio Fornari 15 Several staging systems for hepatocellular carcinoma (HCC) have been developed. The Barcelona Clinic Liver Cancer staging system is considered the best in predicting survival, although limitations have emerged. Recently, the Italian Liver Cancer (ITA.LI.CA) prognostic system, integrating ITA.LI.CA tumor staging (stages 0, A, B1-3, C) with the Child-Turcotte-Pugh score, Eastern Cooperative Oncology Group performance status, and alpha-fetoprotein with a strong ability to predict survival, was proposed. The aim of our study was to provide an external validation of the ITA.LI.CA system in an independent real-life occidental cohort of HCCs. From September 2008 to April 2016, 1,508 patients with cirrhosis and incident HCC were consecutively enrolled in 27 Italian institutions. Clinical, tumor, and treatment-related variables were collected, and patients were stratified according to scores of the Barcelona Clinic Liver Cancer system, ITA.LI.CA prognostic system, Hong Kong Liver Cancer system, Cancer of the Liver Italian Program, Japanese Integrated System, and model to estimate survival in ambulatory patients with hepatocellular carcinoma. Harrell s C-index, Akaike information criterion, and likelihood-ratio test were used to compare the predictive ability of the different systems. A subgroup analysis for treatment category (curative versus palliative) was performed. Median follow-up was 44 months (interquartile range, 23-63 months), and median overall survival was 34 months (interquartile range, 13-82 months). Median age was 71 years, and patients were mainly male individuals and hepatitis C virus carriers. According to ITA.LI.CA tumor staging, 246 patients were in stage 0, 472 were in stage A, 657 were in stages B1/3, and 133 were in stage C. The ITA.LI.CA prognostic system showed the best discriminatory ability (C-index 5 0.77) and monotonicity of gradients compared to other systems, and its superiority was also confirmed after stratification for treatment strategy. Conclusion: This is the first study that independently validated the ITA.LI.CA prognostic system in a large cohort of Western patients with incident HCCs. The ITA.LI.CA system performed better than other multidimensional prognostic systems, even after stratification by curative or palliative treatment. This new system appears to be particularly useful for predicting individual HCC prognosis in clinical practice. (HEPATOLOGY 2018;67:2215-2225) SEE EDITORIAL ON PAGE 2076 The prognostic classification of patients with hepatocellular carcinoma (HCC) is extremely difficult because overall survival is mutually influenced by the tumor burden, dysfunction of the underlying cirrhosis, and patient general conditions (i.e., performance status). (1-3) Several multidimensional prognostic systems have been proposed over the years. The Barcelona Clinic Liver Cancer (BCLC) classification is currently considered by American and European guidelines the best Abbreviations: AFP, alpha-fetoprotein; AIC, Akaike information criterion; BCLC, Barcelona Clinic Liver Cancer; CLIP, Cancer of the Liver Italian Program; CTPS, Child-Turcotte-Pugh score; COG-PST, Eastern Cooperative Oncology Group; EpaHCC, epatologia-hepatocellular carcinoma; HCC, hepatocellular carcinoma; HKLC, Hong Kong Liver Cancer; IQR, interquartile range; ITA.LI.CA, Italian Liver Cancer; JIS, Japan Integrated System; MESIAH, model to estimate survival in ambulatory patients with hepatocellular carcinoma. Received May 15, 2017; accepted November 14, 2017. Supported by the Associazione Italiana Gastroenterologi Ospedalieri [Italian Association of Gastroenterology Hospitals]. EpaHCC is an AIGO scientific project approved in 2008, and EpaHCC consortium includes AIGO members. AIGO is currently the owner of the EpaHCC electronic database. No formal AIGO grant has ever been funded to any member of this project. 2215

BORZIO ET AL. HEPATOLOGY, June 2018 system to predict survival in patients with HCC. (1-3) However, several studies have pointed out that this staging system has several prognostic limitations, mainly among patients belonging to BCLC stage B and C due to their heterogeneity (4) ; the controversial prognostic role of Eastern Cooperative Oncology Group (ECOG) performance status (PST), which can be influenced by liver function, cancer symptoms, or comorbidities other than cirrhosis (5) ; and the lack of statistical prognostic weighting attributed to different relevant factors, such as tumor status, liver function variables, and ECOG PST. (6,7) Recently, a multicenter Italian study group proposed a new prognostic system for patients with HCC, the Italian Liver Cancer (ITA.LI.CA) system. (8) Its main feature is to re-establish the prognostic centrality of tumor staging. In fact, the core component of the ITA.LI.CA prognostic system was defined as ITA.LI.CA tumor staging and was based on four main stages: 0, A, B, and C (Table 1). These stages are similar to the BCLC stages but were designed to have some relevant differences from the latter in order to avoid several confusing aspects: a) ITA.LI.CA stages were based only on tumor characteristics (i.e., ECOG PST and Child-Turcotte-Pugh (CTP) classes did not contribute to stage definition); b) single tumor >5cm was considered in the intermediate stage, regardless of the treatment received, and was based on published data and clinical knowledge. (4,6) Patients in B stage were further stratified into three substages, B1, B2, and B3 (Table 1); patients with HCC with extrahepatic vascular invasion or metastases were included only in stage C, while patients with intrahepatic vascular invasion were placed in stage B3. (4,9) The centrality of ITA.LI.CA tumor staging is important because it has the potential to be useful for therapeutic recommendations and designing clinical trials. However, as stated in the original paper, this tumor staging risks a suboptimal prognostic accuracy at an individual level because it does not consider liver function or patient general condition-related variables. (8) For these reasons, the authors integrated ITA.LI.CA tumor staging with the Child-Turcotte-Pugh score (CTPS), ECOG PST, and serum alpha-fetoprotein (AFP) values in a multivariable survival model. All these variables were statistically weighted to create a final 0-13 point score (ITA.LI.CA integrated prognostic system) for individual prognostic prediction (8) (Table 1). The ITA.LI.CA prognostic system was created using a strong statistical methodology in a large Italian cohort of patients (i.e., using training and validation sets) and then validated in an external large cohort from Taiwan. In the original paper, it performed better than other available systems. (8) However, for the introduction in the clinical practice of this system, its external validation in an independent study is strongly recommended. The aim of this study was to externally and independently Copyright VC 2017 by the American Association for the Study of Liver Diseases. View this article online at wileyonlinelibrary.com. DOI 10.1002/hep.29662 Potential conflict of interest: Nothing to report. ARTICLE INFORMATION: From the 1 UOC Gastroenterologia ed Endoscopia Digestiva, ASST Melegnano e della Martesana, Milano, Italy; 2 Dipartimento di Medicina, SSVD di Epatologia, ASST Spedali Civili di Brescia, Brescia, Italy; 3 UOS Malattie delle vie Biliari e del Fegato, UOC malattie dell Apparato Digerente e del Fegato, AO S. Andrea, Universita Sapienza Roma, Roma, Italy; 4 UO Gastroenterologia e Malattie del Ricambio, Azienda Ospedaliero Universitaria Pisana, Ospedale Cisanello, Pisa, Italy; 5 Unita di Gastroenterologia, Ospedale Policlinico, Napoli, Italy; 6 UO Medicina VI Epatologia e Gastroenterologia, Ospedale San Paolo, Universita degli Studi di Milano, Milan, Italy; 7 Presidio Ospedaliero Pineta Grande, Unita di Ecointerventistica, Castel Volturno, Italy; 8 Dipartimento di Scienze Chirurgiche e Gastroenterologiche Ospedale Policlinico Padova, Padova, Italy; 9 Dipartimento di Medicina, Ospedale Santa Maria del Prato, Feltre, Italy; 10 UOS Gastroenterologia Ospedale di Bentivoglio, Bologna, Italy; 11 UO Gastroenterologia Ospedale Valduce, Como, Italy; 12 Dipartimento Medicina, Universita di Verona, Verona, Italy; 13 UOC Gastroenterologia, Ospedale S. Anna, Ferrara, Italy; 14 UO Medicina Interna I, Ospedale Civico e Benfratelli, Palermo, Italy; 15 Unita di Gastroenterologia ed Epatologia, Ospedale G da Saliceto, Piacenza, Italy. ADDRESS CORRESPONDENCE AND REPRINT REQUESTS TO: Mauro Borzio, M.D. Gastroenterology Unit ASST Melegnano e della Martesana, via Pandina 1, 20070 Vizzolo Predabissi Milan, Italy E-mail: mauro.borzio@gmail.com Tel: 139-2-92360317 2216

HEPATOLOGY, Vol. 67, No. 6, 2018 BORZIO ET AL. TABLE 1. The ITA.LI.CA Prognostic Score Score 0 1 2 3 4 5 ITA.LI.CA tumor staging O A B1 B2 B3 C Diameter (cm) 2 3 5 3-5 >5 3-5 >5 >5 any any Number of nodules 1 2-3 1 2-3 1 >3 2-3 >3 any Any Vascular invasion or metastases no no no no no no no no intra extra Child-Turcotte-Pugh score 5 6-7 8-9 10-15 ECOG performance status 0 1-2 - 3-4 Serum AFP (ng/ml) 1,000 - >1,000 The score ranged from 0 to 13 points. The lowest score (ITA.LI.CA score 5 0) of the model corresponds to the best prognosis, and the highest score (ITA.LI.CA score 5 13) is associated with the worst prognosis. Tumor staging covers the highest prognostic span (from 0 to 5 points), CTPS and ECOG PST from 0 to 3, AFP from 0 to 2. validate the ITA.LI.CA prognostic system value in a large multi-institutional cohort of Western patients, mirroring HCC management in real life. Patients and Methods STUDY GROUPS We analyzed prospectively collected data of 1,508 consecutive patients with HCC who had been recruited between 2008 and April 2016 at 27 Italian institutions participating in the Hepatology-HCC (EpaHCC) project. EpaHCC is an ongoing multi-institutional, infield, large cohort of newly diagnosed HCCs created to specifically validate different prognostic systems of HCC. (10,11) In April 2016, the project enrolled 1,780 cases whose data had been entered in an ad hocdeveloped electronic database (Airon Telematica, Milan, Italy). At the time of this analysis, complete baseline and follow-up data were available in 1,580 cases. To avoid the risk of introducing a significant referral bias, centers for liver transplantation or specialized centers of hepatobiliary surgery were excluded from the study. Exclusion criteria were: a) HCC already present at the time this protocol was implemented (September 2008); b) human immunodeficiency virus co-infection; c) absence of cirrhosis or chronic liver disease. Only patients with a diagnosis of HCC confirmed either by histology or cytology or by typical radiologic appearance were included in the study group. (2,12) Clinical and treatment-related variables, such as age, sex, etiology of underlying liver disease, presence of ascites and hepatic encephalopathy, main serologic parameters (total bilirubin, creatinine, prothrombin time and/or international normalized ratio, AFP, albumin, platelet count), CTPS, (13) tumor radiologic characteristics (number and size of lesions), ECOG PST, (5) and main treatment strategy (hepatic resection, percutaneous ablation, intra-arterial therapies, or other options) were recorded. ECOG PST was prospectively assessed by clinicians who participated in the EpaHCC project. Tumor number and size, major vascular invasion, and patterns of metastatic diffusion were assessed by imaging (either computed tomography or magnetic resonance imaging). Vascular invasion was classified as intrahepatic and/or extrahepatic following the criteria of the Hong Kong Liver Cancer (HKLC) staging system. (6) In detail, intrahepatic vascular invasion included intrahepatic portal vein branch, left or right portal vein invasion, and main hepatic vein invasion; extrahepatic vascular invasion included main portal trunk and inferior vena cava invasion. No predetermined therapeutic protocols were operative, and clinical decisions regarding patient management were totally dependent on the current clinical practice of each participating center and by locally available resources. HCC treatment was classified in six main therapeutic options: liver transplantation, hepatic resection, ablation (percutaneous alcoholization, radiofrequency, or microwave), intra-arterial therapies, sorafenib, and others (other palliative treatment or best supportive care). Moreover, for each patient the following composite variables were calculated and recorded: BCLC classification, ITA.LI.CA prognostic score, HKLC staging, Cancer of the Liver Italian Program (CLIP) score, Japanese Integrated System (JIS), and model to estimate survival in ambulatory patients with hepatocellular carcinoma (MESIAH) score. (2,6-8,14,15) In particular, the ITA.LI.CA integrated prognostic score was calculated by a single formula, TSFA, where TS is the tumor stage (points from 0 to 5 for stages from 0 to C), F is the point value of the ITA.LI.CA functional score based on the CTPS and PST (0 points for CTPS of 5 and PST of 0, 2 points for CTPS of 6-7 and PST of 1-2 or CTPS of 8-9 and so on), and A is the AFP point value (0 points for AFP 1,000 ng/ml, 2 points for AFP >1,000 ng/ml) ( Table 1). Specifically, as exemplified in the original paper, the formula 2217

BORZIO ET AL. HEPATOLOGY, June 2018 A22 indicates a patient with stage A HCC with an ITA.LI.CA functional score of 2 and with 2 points for AFP, resulting in a final prognostic score of 5. (8) DESCRIPTIVE STATISTICS Baseline characteristics were examined based on frequency distribution; continuous data are presented as median (interquartile range) unless otherwise indicated. Univariate comparisons were assessed using the Student t test, Wilcoxon rank sum test, or chi-squared test as appropriate. Missing data relative to study covariates always involved less than 10% of patients; thus, they were imputed using the maximum likelihood estimation method. (16) Overall survival was defined as the time from the date of HCC diagnosis to the date of death, last follow-up evaluation, or data censoring (April 30, 2016). Kaplan- Meier survival curves were used to estimate the median, 1-, 3-, 5-, and 8-year overall survival. Survival curves were also stratified according to stage, and the log rank test was used to compare differences in survival. VALIDATION OF THE ITA.LI.CA PROGNOSTIC SYSTEM By using survival time as an outcome measure, important criteria for assessing the performance of any prognostic system were the following: 1) agreement between observed and predicted outcomes (calibration); 2) difference in survival time is small among patients classified into the same group by that system (homogeneity); 3) compared with this difference, there are much greater differences in the survival times among patients classified into different groups (discriminatory ability); and 4) mean survival time for a group classified as favorable by that system is always longer than the survival times noted in less favorable groups (monotonicity of gradients). (17) Calibration, homogeneity, discriminatory ability, and monotonicity of gradients were evaluated to measure the prognostic ability of the ITA.LI.CA prognostic system in comparison with that of other systems. (8) In testing calibration, patients were divided into quintiles at the 20 th,40 th,60 th and 80 th percentiles of the risk score. Calibration was defined as the graphic agreement between observed outcomes and predictions. (17) ITA.LI.CA quintiles were also used to graphically show the discriminatory ability of the ITA.LI.CA prognostic system over that of BCLC classification. The homogeneity, discriminatory ability, and the monotonicity of gradients in mortality rates of different prognostic systems were then described using the Akaike information criterion (AIC), the test for trend chi-square, and Harrell s C-index. (17) Briefly, the test for trend (which is more suitable than the log-rank test for measuring both the discriminatory power and monotonicity of gradients across categories) and the AIC test were used for measuring homogeneity. In the AIC test, we used the ordinary prognostic score rather than transforming it into dummy variables; thus, the AIC test can also estimate the monotonicity of gradients. Harrell s C-index, which requires no assumption of the model, represents the proportion of correct predictions or concordance in all possible pairs of patients. Suppose that in a given pair, patient A has a worse prognostic scorethanpatientb.ifthesurvivaltimeofaisshorter (longer) than that of B, the prediction is concordant (discordant) with the actual outcome. If, on the other hand, one patient is censored earlier than the time of death or census for the other patient, that pair is not counted. (17,18) A lower AIC value results in higher homogeneity and monotonicity of gradients of the staging system. A higher C-index and the test for trend chi-square result in higher discriminatory ability and monotonicity of gradients of the staging system. To measure whether the performance of the ITA.LI.CA score in terms of homogeneity, discriminatory ability, and monotonicity of gradients was significantly better than that of other systems, (2,6-8,14,15) we used the likelihood ratio test. Due to the heterogeneity of received treatments, a subgroup analysis for treatment category (i.e., curative versus palliative) was also performed to evaluate and overcome potential time-related biases. All statistical tests were based on the proportional hazards model. All analyses were carried out with STATA version 13.0 (Stata Corp, College Station, TX) and the JMP package (1989-2010; SAS Institute Inc.). All tests were two-sided, and P < 0.05 was considered statistically significant. Results CHARACTERISTICS OF THE STUDY GROUP The characteristics of the study group are described in Table 2. The median age of enrolled patients was 71 years with a preponderance of male individuals (72.3%). 2218

HEPATOLOGY, Vol. 67, No. 6, 2018 BORZIO ET AL. TABLE 2. Patient Characteristics in the Study Group Variables Study Group (n 5 1,508) Sex M 1,091 (72.3%) F 417 (27.7%) Age (years) 71 (64-76) HBV 1 209 (13.9%) HCV 1 834 (55.2%) Alcohol abuse 319 (21.2%) Albumin (g/l) 35 (31-39) Bilirubin (mg/dl) 1.1 (0.8-1.7) INR 1.2 (1.1-1.3) Platelets 114 (81-167) Creatinin (mg/dl) 0.9 (0.7-1.1) Presence of ascites 341 (22.6%) Presence of encephalopathy 82 (5.4%) Child-Turcotte-Pugh score 6 (5-7) MELD 7 (6-11) ECOG PST 0 935 (62.0%) 1 378 (25.1%) 2 131 (8.7%) 3 1 4 64 (4.2%) AFP (ng/ml) 14 (5-88) Diameter of largest lesion (mm) 30 (21-50) Multinodular (>3 nodules) 260 (17.2%) Macrovascular invasion 222 (14.7%) Intrahepatic 125 (8.3%) Extrahepatic 97 (6.4%) Presence of metastases 56 (3.7%) BCLC stage 0 124 (8.2%) A 687 (45.5%) B 310 (20.6%) C 276 (18.3%) D 111 (7.4%) ITA.LI.CA Tumor stage 0 246 (16.3) A 472 (31.3) B1 371 (24.6) B2 116 (7.7) B3 170 (11.3) C 133 (8.8) Main treatment Resection 157 (10.4%) Transplantation 17 (1.1%) Ablation 490 (32.5%) IAT 416 (27.6%) Sorafenib 68 (4.5%) Other/BSC 360 (23.9%) (16.3%), 472 in the early stage (31.3%), 657 in the intermediate stages (B1 24.6%, B2 7.7%, B3 11.3%), and 133 in the advanced stage (8.8%) (Fig. 1). Surgery was carried out in 174 patients (liver transplantation 1.1%, hepatic resection 10.4%), ablation in 490 (32.5%), intraarterial therapy in 416 (27.6%), sorafenib in 68 (4.5%), and other palliative therapy/best supportive care in 360 (23.9%). The overall adherence to the BCLC therapeutic algorithm was 56.6%; in particular, adherence to the BCLC algorithm was 65% for liver transplantation, 65% for resection, 66% for ablation, 56% for transarterial chemoembolization, and 73% for sorafenib. The lowest adherence rate was registered for other palliative treatments or best supportive care (36%). CALIBRATION OF THE ITA.LI.CA PROGNOSTIC SYSTEM Median duration of follow-up was 44 months (interquartile range, 23-63 months). Median overall survival was 34 months (interquartile range, 13-82 months), and the 1-, 3-, and 5-year overall survival rates were 78%, 48%, and 34%, respectively. The lowest ITA.LI.CA score (score 5 0) corresponded to the best prognosis, whereas the highest score (score 5 13) was associated with the worst prognosis (Table 3; P < 0.0001, log-rank test). To test the prognostic calibration of the ITA.LI.CA score, patients were divided into quintiles at the 20 th,40 th, 60 th, and 80th percentiles of the risk score. Quintile 1 Abbreviations: BSC, best supportive care; IAT, intra-arterial therapy; INR, international normalized ratio; MELD, model for end stage liver disease. The majority of patients were hepatitis C carriers (55.2%), whereas only a minority of patients were hepatitis B carriers (13.9%). A non-negligible proportion of the study cohort (31.1%) had decompensated cirrhosis (CTPS, B-C) and 38% had an ECOG PST >0. In terms of tumor characteristics (i.e., ITA.LI.CA tumor staging), 246 patients were in the very early stage FIG. 1. Expected versus observed survival in the study cohort (calibration). Patients were divided into quintiles at the 20 th, 40 th,60 th, and 80 th percentiles of the risk score. Q1 coincided with ITA.LI.CA score 1, Q2 with score 2, Q3 with score 3-4, Q4 with score 5-6, and Q5 with values >6. 2219

BORZIO ET AL. HEPATOLOGY, June 2018 Score TABLE 3. Discrimination Ability of Different HCC Prognostic Systems Number of Patients (%) Observed Median Survival (Months) Lower 95% Higher 95% All 1,508 34 32 38 ITA.LI.CA score (8) 0 117 (7.8) 88 69 112 1 270 (17.9) 82 70 97 2 299 (19.8) 50 44 58 3 239 (15.8) 35 30 40 4 166 (11.0) 26 22 31 5 121 (8.0) 18 15 21 6 75 (5.0) 15 12 19 7-8 131 (8.7) 9 8 11 9-10 67 (4.4) 5 4 6 11-12-13 23 (1.5) 2 2 4 CLIP (13) 0 619 (41.1) 66 56 91 1 422 (28.0) 33 30 39 2 268 (17.8) 19 16 23 3 133 (8.8) 8 6 10 4-6 66 (4.4) 4 3 5 HKLC (6) I 608 (40.3) 76 62 92 IIa 304 (20.2) 36 31 43 IIb 179 (11.9) 30 25 36 IIIa 38 (2.5) 20 13 26 IIIb 26 (1.7) 13 8 17 IVa 75 (5.0) 13 9 17 IVb 56 (3.7) 7 5 9 Va 110 (7.3) 16 13 26 Vb 112 (7.4) 5 4 6 MESIAH (7) Q1 375 (24.9) 79 67 96 Q2 376 (24.9) 47 40 57 Q3 380 (25.2) 29 26 33 Q4 377 (25.0) 9 9 12 JIS (14) 0 151 (10.0) 76 63 115 1 583 (38.7) 62 54 79 2 326 (21.6) 30 26 33 3 260 (17.2) 17 14 20 4 138 (9.2) 8 6 10 5 50 (3.3) 4 2 5 BCLC (2) 0 124 (8.2) 83 74 116 A 687 (45.5) 60 50 64 B 310 (20.6) 28 25 32 C 276 (18.3) 12 10 14 D 111 (7.4) 5 4 8 Distribution of patients in different points/stages of the prognostic systems and corresponding observed median survivals. Log-rank test, P < 0.0001. (Q1) coincided with ITA.LI.CA score 1, quintile 2 (Q2) with score 2, quintile 3 (Q3) with score 3-4, quintile 4 (Q4) with score 5-6, quintile 5 (Q5) with values >6 (Fig. 1). The calibration of the ITA.LI.CA prognostic system was clearly shown to be optimal in the study cohort because predicted and observed median survivals and survival curves largely coincided for each quintile of the enrolled population (Fig. 1). HOMOGENEITY, DISCRIMINATION, AND MONOTONICITY OF GRADIENTS OF THE ITA.LI.CA PROGNOSTIC SYSTEM To evaluate the discriminatory power of the ITA.LI.CA system, we first described its prognostic 2220

HEPATOLOGY, Vol. 67, No. 6, 2018 BORZIO ET AL. FIG. 2. Kaplan-Meier survival curve of the study group stratified according to BCLC stages. Log-rank test, P < 0.001; AIC, 7,421; C-index 5 0.72. ability by Kaplan-Meier survival curves according to BCLC and ITA.LICA quintiles (Figs. 2 and 3). The discriminatory ability of the ITA.LI.CA score was better than that of BCLC (AIC, 7,421 for BCLC; AIC, 7,232 for ITA.LI.CA). The discriminatory ability of the ITA.LI.CA system compared with that of other available systems is shown in Table 3.When the ITA.LI.CA prognostic score was considered (from 0 to 13 points), it showed the best homogeneity, discriminatory ability, and monotonicity of gradients among the most common HCC staging systems (Table 4). In particular, the C statistic of the ITA.LI.CA score in the whole study group was 0.77, a value superior to that of the BCLC (0.74), CLIP (0.73), JIS (0.74), MESIAH (0.76), and HKLC (0.75). By using the likelihood ratio test to compare different survival models, the prognostic performance of the ITA.LI.CA system resulted once again as better than that of other systems (P < 0.0001). The superiority of the ITA.LI.CA score was also confirmed after stratification of the patient population for treatment strategy (curative versus palliative). Only the MESIAH score showed a prognostic performance not significantly inferior to the ITA.LI.CA score (P 5 0.0601) in patients undergoing curative therapies (i.e., surgery and ablation). Discussion Prognostic staging of HCC is still a challenge. (1-4) Several staging systems have been proposed, but none have been universally adopted. The difficulties in constructing an optimal staging system rely on the fact that the outcome of HCC is a composite result mutually influenced by tumor burden, residual function related to the underlying liver cirrhosis, and the general health status (i.e., performance status). (1-4) In addition, differences in epidemiology and risk factors for HCC in different geographic areas further account for these difficulties and may explain why several validation studies aimed at comparing the prognostic performance of different staging systems yielded conflicting results. (19-21) Recently the ITA.LI.CA group proposed a new prognostic system, the ITA.LI.CA system, integrating a tumor staging (six main stages: 0, A, B1, B2, B3, and C), a functional prognostic score (based on CTPS and ECOG-PST), and AFP. In the original paper, the ITA.LI.CA system was superior to other multidimensional systems in predicting survival in an internal validation cohort (ITA.LI.CA cohort) and in an external one from Taiwan. (8) In the present study, we proposed the first external and independent validation of the ITA.LI.CA prognostic system in a large real-life Italian cohort (EpaHCC multi-institutional cohort) of newly diagnosed HCCs. In our cohort, the ITA.LI.CA prognostic system offered the best predictive ability in terms of calibration, homogeneity, discriminatory ability, and monotonicity of gradients. (17) The rigorous statistical methodology by which this system was originally FIG. 3. Kaplan-Meier survival curve of the study group stratified according to ITA.LI.CA quintiles. Log-rank test, P < 0.001; AIC, 7,232; C-index 5 0.76. 2221

BORZIO ET AL. HEPATOLOGY, June 2018 TABLE 4. Discrimination Ability and Monotonicity of Gradients of Different HCC Prognostic Systems HCC Prognostic System AIC C-Index v 2 Test LR Test, P Value Study group (n 5 1,508) ITA.LI.CA score (8) 7,087 0.77 763 - CLIP (13) 7,233 0.73 575 161, <0.0001 HKLC (6) 7,194 0.75 659 44, <0.0001 MESIAH (7) 7,159 0.76 642 87, <0.0001 JIS (14) 7,206 0.74 563 123, <0.0001 ITA.LI.CA tumor stage (8) 7,307 0.72 482 221, <0.0001 BCLC (2) 7,234 0.74 514 158, <0.0001 Curative treatment (n 5 658) ITA.LI.CA score (8) 2,177 0.65 39 - CLIP (13) 2,186 0.61 34 19, 0.0075 HKLC (6) 2,189 0.62 33 18, 0.0112 MESIAH (7) 2,180 0.65 34 14, 0.0601 JIS (14) 2,193 0.61 21 21, 0.0003 ITA.LI.CA tumor stage (8) 2,202 0.59 13 28, <0.0001 BCLC (2) 2,193 0.60 23 25, 0.0001 Noncurative treatment (n 5 850) ITA.LI.CA score (8) 4,855 0.73 346 - CLIP (13) 4,957 0.70 220 109, <0.0001 HKLC (6) 4,907 0.72 280 18, 0.0015 MESIAH (7) 4,899 0.72 316 51, <0.0001 JIS (14) 4,945 0.70 233 91, <0.0001 ITA.LI.CA tumor stage (8) 5,014 0.68 192 140, <0.0001 BCLC (2) 4,952 0.70 215 92, <0.0001 Each column shows the AIC, C-index, and test for trend chi-square. The lower the AIC value, the higher the homogeneity and the monotonicity of gradients of the prognostic system; the higher the C-index and the test for trend chi-square, the higher the discriminatory ability and monotonicity of gradients of the prognostic system. In addition, in each column the ITA.LI.CA score was compared with other systems by using the likelihood ratio test. Abbreviation: LR, likelihood ratio. elaborated (8) may account for its superiority over the other five multidimensional prognostic systems considered in this study (Tables 3 and 4). In particular, the ITA.LI.CA system performed better than the BCLC system. This finding is not surprising and parallels the observations reported in two recent comparative studies carried out on Eastern and Western cohorts of HCC. These studies were aimed at comparing the accuracy of different prognostic systems, and the BCLC system performed less well than the CLIP score, which ranked first. (19,20) The relatively unsatisfying predictive ability of the BCLC classification is probably due to this system being mainly based on data extrapolated from systematic reviews of the literature and/or expert opinions rather than developed by a solid statistical methodology. The ITA.LI.CA system was also superior to the HKLC system in defining the prognosis of our Western population of HCC. This is possibly due to the epidemiologic characteristics of the large Asiatic population on which this latter system was developed (mostly composed of patients infected by the hepatitis B virus [HBV] (6) ). Indeed, the HKLC system is still waiting for a robust validation in Western countries, and its accuracy has been challenged. (22) However, to further increase robustness and exportability of ITA.LI.CA, new independent validations (from different authors) should be performed in the Eastern world where HBV-related HCC is prevalent. These validation studies are highly warranted and should be carried out in the near future. In the present study, the ITA.LI.CA integrated prognostic score confirmed a high calibration ability. In particular, the expected and observed survival in different score quintiles showed strict correspondence, a finding well in keeping with the findings of the original paper (Fig. 1). The performance of any HCC staging system should be interpreted within the specific context in which it was created. Therefore, a full characterization of the examined study population should anticipate the validation process. The ITA.LI.CA and EpaHCC cohorts are comparable as far as median age, etiology of cirrhosis, BCLC stratification, treatment offered, adequate follow-up, and median overall survival are concerned. (8) Moreover, both cohort characteristics are in line with the inherent characteristics of most of the Caucasian cohorts for which other staging systems were tested and validated and strictly reflect the life 2222

HEPATOLOGY, Vol. 67, No. 6, 2018 BORZIO ET AL. scenario of HCC distribution and management in most European countries. (2) We thus believe that our findings further reinforce the strength and exportability of our validation data. An additional advantage provided by the use of the EpaHCC cohort to validate the ITA.LI.CA system is its representativeness of the current presentation and management of HCC. Indeed, accrual of the EpaHCC cohort started in 2008, while the ITA.LI.CA cohort accrual period covered a larger time span (1987-2012) This aspect is not trivial because EpaHCC strongly adds homogeneity in terms of current diagnostic and treatment options, thus limiting the potential for time-related selection bias. Although the BCLC and ITA.LI.CA systems share a comparable tumor-staging framework, some relevant differences between the two systems deserve mention. First, unlike BCLC, in this new system, tumor stages are based only on oncologic characteristics while ECOG PST and CTPS are not considered. This may, at least in part, overcome the well-known limits of including these two variables when not weighted in a multivariable survival model. Second, ITA.LI.CA tumor stage 0 (Table 1) is clearly defined as a single nodule 2 cm. Conversely, the BCLC stage 0 definition (23) only recently included a single nodule 2cm, while previous definitions in guidelines (2,3) included single nodules <2 cm. Third, in an attempt to overcome the limits posed by the large heterogeneity of patients belonging to the BCLC intermediate stage and in keeping with a recent proposal, the ITA.LI.CA system further divided the ITA.LI.CA stage into three substages: B1, B2, and B3. Fourth, the ITA.LI.CA system simplified the stratification of patients harboring large HCC (>5 cm as a threshold), emending the confounding concept of resectability. Fifth, the presence of vascular invasion was better defined dividing thrombosis into intrahepatic and extrahepatic forms, resulting in a more precise calibration of the whole system. Finally, AFP was added to the prognostic score; this marker was used as a categorical variable rather than a continuous one with a cutoff >1,000 ng/ml because this value yielded the best prognostic power in the original survival model from Farinati et al. (8) Treatment per se is also a strong predictor of HCC outcome. Consequently, international guidelines recommend the use of multidimensional systems linking staging with treatment decisions. (2,3) However, treatment for HCC in cirrhosis is complex, varying widely among different geographic areas and with several therapeutic options being available for each stage. In addition, the high heterogeneity of patients included in any given stage, especially in BCLC intermediate/advanced ones, implies a wide range of possible managing treatments (including no treatment) with a consequent potential for a wide variation of their impact on overall survival. Rigidity of the BCLC therapeutic algorithm as well as its difficulty to be applied in clinical practice have been repeatedly underlined. (10,24) For these reasons, Farinati et al. (8) decided not to include a prefixed treatment algorithm in their integrated prognostic system. However, in contrast to the original study, we tested the prognostic performance of the ITA.LI.CA prognostic score in patients stratified by curative or palliative treatment. The ITA.LI.CA prognostic system performed better than other systems in patients undergoing curative and those undergoing palliative therapy. In patients treated by curative therapies, the MESIAH score also displayed a good predictive ability with only a marginally significant statistical difference (P 5 0.0601) between the ITA.LI.CA and MESIAH score. However, the main limit of the MESIAH score is that it does not offer tumor staging to guide and assist treatment decisions. (7) Conversely, we did not perform this analysis in single-therapy subgroups for two main reasons: 1) Some treatment subgroups (i.e., liver transplantation, resection, sorafenib) were too small to test the prognostic power of staging systems. 2) All HCC prognostic systems have been developed and designed for general HCC populations, including all different therapeutic options. Consequently, we considered that it was not appropriate to test them as single-therapy subgroups. A limitation of the present study is its retrospective nature. However, one of the most stringent prerequisites required from each participating center to the EpaHCC project was the consecutive enrollment of patients, and the EpaHCC database was specifically created to validate different prognostic systems in clinical practice. Consequently, each patient at entry was stratified according to staging systems available at that time and followed thereafter. Individual records entered into the database made it possible to stratify patients according to the ITA.LI.CA system as well. These methodologic precautions contribute to minimizing the inherent biases of retrospective studies. Another limitation is that the present study could not fully assess the role that different treatments for different stages might have had on overall survival. Therefore, further studies are warranted to validate the prognostic value of this scoring system in other welldesigned prospective cohorts. 2223

BORZIO ET AL. HEPATOLOGY, June 2018 A final limitation of this study is that the variable antiviral therapy for both HBV and hepatitis C virus (HCV) cases was not provided in the EpaHCC database. In particular, in light of the recent introduction of new anti-hcv therapies and their suspected negative role on HCC outcome, (25) the EpaHCC database will be improved by adding antiviral therapy information to analyze its prognostic impact. In conclusion, our study provides a robust validation of the integrated ITA.LI.CA prognostic system in a large, real-life, occidental cohort of HCC (EpaHCC multi-institutional cohort). In our prospective cohort in which HCV-related HCC is prevalent, the ITA.LI.CA prognostic score was accurate in predicting HCC outcome and performed better than the other staging systems considered. In addition, the predictive accuracy of the ITA.LI.CA system was maintained in both treated and untreated patients. This newly developed staging system has thus several advantages over the other commonly used multidimensional prognostic systems, making it more suitable for HCC prognostication at an individual level. Acknowledgment: The present study was sponsored by the Associazione Italiana Gastroenterologi Ospedalieri. Other members of the EpaHCC group are the following: UOC Gastroenterologia, ATSS Melegnano-Martesana, Italy: Dr.ssa R. Ferrini, Dr.ssa M. Quagliuolo; UO Gastroenterologia, Ospedale Valduce, Italy: Dr. G. Spinzi; Ospedale S. Maria del Prato, Italy: Dr. G. Parisi, Dr. M. Tollardo, Dr. C. Cardaioli; UO Medicina, Ospedale di Gavardo, Italy: Dr.ssa O. Bonzanini, Dr.ssa S. Polo; Ospedale Alessandro Manzoni Struttura Complessa di Medicina Generale, Italy: Dr. M. Andreoletti; Ospedale Fatebenefratelli, UO di Ecografia, Centro Zerbi, Radiologia, Milano, Italy: Dr. F. Borzio; UO Gastroenterologia ed Endoscopia Digestiva Ospedale San Carlo Borromeo Milano, Italy: Dr. A. Capretti; Cattedra di Gastroenterologia Seconda Universita di Napoli, Ospedale Policlinico, Italy: Dr. Niosi; Dipartimento Scienze Chirurgiche e Gastroenterologiche, Policlinico Universitario Padova, Italy: Dr.ssa A. Giacomin; UO Medicina Interna I, Ospedale Civico e Benfratelli, Palermo, Italy: Dr. A. Maranghini; Servizio di Gastroenterologia ed endoscopia ASL2 PG, Italy: Dr..ssa F. Sannella, Dr. A. Solinas; UO Gastroenterologia, Ospedale SS Cosma e Damiano, Pescia, Italy: Dr. P. Montalto; UOC Gastroenterologia ed Epatologia Ospedale G. da Saliceto, Piacenza, Italy: Dr. G.M. Prati, Dr.ssa P. Perazzo; UO Gastroenterologia Azienda Ospedaliera G.Salvini Ospedale di Rho, Italy: Dr.ssa A. Bortoli, Dr. A. Prada; UO Gastroenterologia, Ospedale Infermi, Rimini, Italy: Dr.ssa A. Miracolo, Dr. D. Onorato, Dr. L. Solmi; Azienda Ospedaliera S. Andrea, Roma, Italy: Dr.ssa P. Begini, Dr.ssa G. Anania; UO Gastroenterologia, San Giovanni Rotondo, Italy: Dr.ssa F. Terracciano, Dr. A. Andriulli; SC Gastroenterologia ed Epatologia, Presidio Sanitario Gradenigo, Italy: Dr. F. Rosina, Dr.ssa M. Berruti; UO Gastroenterologia, Ospedale San Bortolo, Vicenza, Italy: Dr. M. Salvagnini, Dr. A. D Alessandro; Ospedale di Vizzolo-Predabissi, ATSS Melegnano- Martesana, Medicina Epatologia, Italy: Dr. R. Centenaro; Dipartimento di Medicina, Chirurgia ed Odontoiatria, Ospedale San Paolo, Universita degli Studi di Milano, Italy: Dr. M. Zuin. REFERENCES 1) Bruix J, Sherman M, Llovet JM, Beaugrand M, Lencioni R, Burroughs AK, et al.; EASL Panel of Experts on HCC. Clinical management of hepatocellular carcinoma. Conclusions of the Barcelona-2000 EASL conference. European Association for the Study of the Liver. J Hepatol 2001;35:421-430. 2) European Association for the Study of the Liver; European Organisation for Research and Treatment of Cancer. EASL- EORTC clinical practice guidelines: management of hepatocellular carcinoma. J Hepatol 2012;56:908-943. 3) Bruix J, Sherman M; Practice Guidelines Committee, American Association for the Study of Liver Diseases. Management of hepatocellular carcinoma. HEPATOLOGY 2005;42:1208-1236. 4) Bolondi L, Burroughs A, Dufour JF, Galle PR, Mazzaferro V, Piscaglia F, et al. Heterogeneity of patients with intermediate (BCLC B) hepatocellular carcinoma: proposal for a subclassification to facilitate treatment decisions. Semin Liver Dis 2012;32: 348-359. 5) Hsu CY, Lee YH, Hsia CY, Huang YH, Su CW, Lin HC, et al. Performance status in patients with hepatocellular carcinoma: determinants, prognostic impact, and ability to improve the Barcelona Clinic Liver Cancer system. HEPATOLOGY 2013; 57:112-119. 6) Yau T, Tang VY, Yao TJ, Fan ST, Lo CM, Poon RT. Development of Hong Kong liver cancer staging system with treatment stratification for patients with hepatocellular carcinoma. Gastroenterology 2014;146:1691-1700. 7) Yang JD, Kim WR, Park KW, Chaiteerakij R, Kim B, Sanderson SO, et al. Model to estimate survival in ambulatory patients with hepatocellular carcinoma. HEPATOLOGY 2012;56: 614-621. 8) Farinati F, Vitale A, Spolverato G, Pawlik TM, Huo TL, Lee YH, et al.; ITA.LI.CA study group. Development and validation of a new prognostic system for patients with hepatocellular carcinoma. PLoS Med 2016;13:e1002006. 2224

HEPATOLOGY, Vol. 67, No. 6, 2018 BORZIO ET AL. 9) United Network for Organ Sharing. Important policy notice. 2011 Dec 15. https://optn.transplant.hrsa.gov/media/1200/optn_ policies.pdf#nameddest5policy_09. Accessed March, 2015. 10) Borzio M, Fornari F, De Sio I, Andriulli A, Terracciano F, Parisi G, et al.; EpaHCC Group. Adherence to American Association for the Study of Liver Diseases guidelines for the management of hepatocellular carcinoma: results of an Italian field practice multicenter study. Future Oncol 2013;9:283-294. 11) Borzio M, Dionigi E, Rossini A, Toldi A, Francica G, Fornari F, et al. Trend of improving prognosis of hepatocellular carcinoma in clinical practice: an Italian in-field experience. Dig Dis Sci 2015;60:1465-1473. 12) Italian Association for the Study of the Liver (AISF); AISF Expert Panel; AISF Coordinating Committee, Bolondi L, Cillo U, Colombo M, Craxı A, Farinati F, Giannini EG, et al. Position paper of the Italian Association for the Study of the Liver (AISF): the multidisciplinary clinical approach to hepatocellular carcinoma. Dig Liv Dis 2013;45:712-723. 13) Pugh RN, Murray-Lyon IM, Dawson JL, Pietroni MC, Williams R. Transection of the oesophagus for bleeding oesophageal varices. Br J Surg 1973;60:646-649. 14) [No authors listed]. A new prognostic system for hepatocellular carcinoma: a retrospective study of 435 patients: the Cancer of the Liver Italian Program (CLIP) investigators. HEPATOLOGY 1998;28:751-755. 15) Kudo M, Chung H, Osaki Y. Prognostic staging system for hepatocellular carcinoma (CLIP score): its value and limitations, and a proposal for a new staging system, the Japan Integrated Staging Score (JIS score). J Gastroenterol 2003;38:207-215. 16) Baraldi AN, Enders CK. An introduction to modern missing data analyses. J Sch Psychol 2010;48:5-37. 17) Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology 2010;21:128-138. 18) Stone M. Akaike s criteria. In: Armitage P, Colton T, eds. Encyclopedia of Biostatistics. Chichester, United Kingdom: Wiley;1998:123-124. 19) Liu PH, Hsu CY, Hsia CY, Lee YH, Su CW, Huang YH, et al Prognosis of hepatocellular carcinoma: assessment of eleven staging systems. J Hepatol 2016:64;601-608. 20) op dev Winkel M, Nagel D, Sappl J, op den Winkel P, Lamerz R, Zech CJ, et al. Prognosis of patients with hepatocellular carcinoma. Validation and ranking of established staging-systems in a large western HCC-cohort. Plos One 2012;7:e45066. 21) Maida M, Orlando E, Camma C, Cabibbo G. Staging systems of hepatocellular carcinoma: a review of literature. World J Gastroenterol 2014;20:4141-4150. 22) Adhoute X, Penaranda G, Bronowicki JP, Raoul JL. Usefulness of the HKLC versus. the BCLC staging system in a European HCC cohort. J Hepatol 2015;62:492-493. 23) Bruix J, Reig M, Sherman M. Evidence-based diagnosis, staging, and treatment of patients with hepatocellular carcinoma. Gastroenterology 2016;150:835-853. 24) Leoni S, Piscaglia F, Serio I, Terzi E, Pettinari I Croci L, et al. Adherence to AASLD guidelines for the treatment of hepatocellular carcinoma in clinical practice: experience of the Bologna Liver Oncology Group. Dig Liv Dis 2014;46:549-555. 25) Alberti A, Piovesan S. Increased incidence of liver cancer after successful DAA treatment of chronic hepatitis C: fact or fiction? Liver Int 2017;37:802-808. 2225