Current noninvasive liver reserve models do not predict histological fibrosis severity in hepatocellular carcinoma

The Ishak scoring system has been used to stage liver fibrosis. Ten noninvasive liver reserve models were proposed to assess the severity of liver fibrosis, but their performance in hepatocellular carcinoma (HCC) is unknown. We aimed to evaluate the correlation between these models and severity of fibrosis in patients with HCC. A total 464 patients with HCC undergoing surgical resection were retrospectively analyzed. Multivariate logistic regression analysis was performed to determine independent factors associated with advanced fibrosis (Ishak score 4 or higher). There were no significant correlations between all noninvasive models and severity of fibrosis in HCC (p for trend all >0.1). In subgroup analysis, cirrhosis discriminant index (CDS) and Lok’s index in hepatitis B-, and fibrosis index based on 4 factors (FIB-4), CDS and Lok’s index in hepatitis C-associated HCC, best correlated with the severity of liver fibrosis. Low platelet count, prolonged prothrombin time, hepatitis C and multiple tumors were independently associated with advanced fibrosis. Among the 10 models, CDS was the best model to predict cirrhosis. Currently used noninvasive liver reserve models do not well correlate with severity of histological fibrosis in HCC. New noninvasive models are required to improve the predictive accuracy of liver fibrosis in HCC.


Material and Methods
Patients. Patients with newly diagnosed HCC in our hospital during the period from 2009 to 2016 were prospectively identified and retrospectively analyzed. A total 464 patients undergoing surgical resection were enrolled in this study. Their baseline information, including patient's demographics, etiology of liver disease, performance status, tumoral status, liver functions and serum biochemistry, were comprehensively recorded at the times of diagnosis. The inclusion criteria of surgery were (1) tumor involving no more than three Couinaud segments, (2) CTP class A or B, (3) no main portal vein trunk involvement or distant metastases, and (4) absence of other major diseases that contradict surgical resection 32 . Patients were followed every 3-4 months until death or dropout from follow-up. This study complies with the standard of the Declaration of Helsinki and current ethical guidelines and has been approved by the Institutional Review Broad of Taipei Veterans General Hospital. Waiver of consent was obtained, and patient records/information was anonymized and de-identified before analysis.
Diagnosis and definition. The pre-operative diagnosis of HCC was based on the findings of typical four-phase multidetector contrast-enhanced dynamic computed tomography (CT) scan or magnetic resonance imaging (MRI) 6,33 , and as all histologically confirmed post-operatively. Patients who were seropositive for hepatitis B antigen (HBsAg), seronegative for anti-HCV antibody, and without a history of alcoholism were classified as HBV-related HCC. HCV-related HCC was defined as seropositive for anti-HCV, seronegative for HBsAg and no history of alcoholism. Dual HBV-and HCV-related HCC was defined as seropositive for HBsAg and anti-HCV 34 . The performance status was assessed by using the Eastern Cooperative Oncology Group Performance scaling ranging from 0 (asymptomatic) to 4 (confined to bed).

Surgical resection.
After the diagnosis was confirmed, patients were reviewed at our multi-disciplinary HCC team for treatment discussion. Share-decisions regarding treatment modalities were made by patients and physicians after individual counseling. Written informed consent was obtained prior to definite treatment. The operative procedures have been previously described in detail 34,35 . The resected liver tissue was sent for gross and microscopic examinations, and the recorded tumor size was based on the largest dimension of the resected specimen.
Histological analysis. Histology slides of all eligible patients were retrieved and carefully reviewed for tumoral part and non-tumoral part by the pathologists who were blinded to clinical information. The degree of hepatic inflammation and stage of fibrosis in non-tumoral part of the specimen were graded according to the Ishak scoring system. The Ishak staging was defined as the following: 0, no fibrosis; 1, fibrous expansion of some portal area, with or without of short fibrous septa; 2, fibrous expansion of most portal area, with or without short fibrous septa; 3, fibrous expansion of most portal area, with occasional portal to portal bridging; 4, fibrous expansion of portal area with marked bridging as well as portal-central; 5, marked bridging with occasional nodules (incomplete cirrhosis); 6, cirrhosis, probable or definite 36 . Grading of 10 noninvasive liver reserve models. The calculation of the 10 noninvasive liver reserve models was based on clinical variables and serum biochemistries at the time of diagnosis. The grading of these models was determined according to published studies [15][16][17][18][20][21][22]24,25,30,31 . Generally, grade 1 indicates no or mild liver fibrosis, and grade 3 shows advanced liver fibrosis or cirrhosis (Table 1). Statistical analysis. The χ 2 test or Fisher's exact test was used to analyze categorical variables, and the Mann-Whitney ranked sum test was used for continuous variables. Factors that showed significant difference in univariate analysis were subjected to multivariate analysis by forward logistic regression to identify independent predictors and determination of odds ratio (OR) and 95% confidence interval (CI). The predictive accuracy of noninvasive models for cirrhosis was determined by calculating the area under receiver operating curve (AUROC) 37 . Spearman's correlation analysis was used to estimate the correlation between noninvasive liver reserve models and Ishak fibrosis stage in HBV-and HCV-related HCC patients. For all tests, a p < 0.05 was considered statistically significant. All statistical analyses were conducted using the SPSS for Windows version 21 release (SPSS Inc., Chicago, IL, USA). Table 2, the median age of the study patients was 63 years and 77% were male. The most frequent cause of chronic liver disease was hepatitis B (55%), followed by hepatitis C (16%). Approximately 65% of patients were classified as performance status 0, and 26% had diabetes mellitus. The majority (83%) of patients had single tumor, and 63% of patients had main tumor size larger than 3 cm. In these patients, 118 (26%) and 177 (38%) received lobectomy and bi-segmentectomy, respectively. Another 149 (32%) and 20 (4%) patients received segmentectomy and sub-segmentectomy, respectively. All patients had histologically confirmed HCC and the resected tumors were free of surgical margin. The numbers of patients of each grade in different noninvasive liver reserve models are described in Table 2. Correlation of noninvasive liver reserve models and Ishak fibrosis score in HCC. The distributions of Ishak fibrosis scores were as following: score 0, 18 (4%) patients: score 1, 85 (18%) patients; score 2, 60 (13%) patients; score 3, 55 (12%) patients; score 4, 65 (14%) patients; score 5, 71 (15%) patients and score 6, 110 (24%) of patients ( Table 2). The relationship of these models and Ishak fibrosis score were assessed. There were no significant correlations between the Ishak score and all 10 (ALBI, APRI, CTP, FIB-4, GUCI, King's score, MELD, PALBI, CDS and Lok index) models (Figs 1 and 2; p values for trend all >0.1).
Correlation of noninvasive models and Ishak fibrosis score according to viral factors. The correlation between liver reserve models and the stage of fibrosis according to viral factors was analyzed (Table 5). There was a relatively high correlation for CDS and Lok index (correlation coefficient, 0.340 and 0.277, respectively, p < 0.001) and the stage of fibrosis specicially in HBV-related HCC. In HCV-related HCC, the correlation was more significant for FIB-4, CDS, Lok index, and APRI with Ishak fibrosis (correlation coefficient: 0.591, 0.546, 0.546, and 0.464, respectively, p < 0.001).

Correlation of noninvasive models and tumor burden.
We analyzed the correlation between the surrogates of tumor burden (number and size of tumor, serum AFP level) and 10 noninvasive models (Table 6). Single tumor was associated with lower scores of APRI, GUCI, King's score and PABLI (all p < 0.05), and smaller size of main tumor was linked with lower ALBI, CTP and PABLI score (all p < 0.05). Serum AFP levels tended to be lower with lower scores of APRI, CDS, FIB-4, GUCI, King's score, Lok index and MELD (all p < 0.05).

Discussion
Although the CTP and MELD scoring system are widely used in patients with liver diseases, these models are not optimal for evaluating liver fibrosis. Alternatively, noninvasive liver reserve models were proposed to assess the severity of liver fibrosis mainly in patients with chronic hepatitis B or C 18,23,25,28,38-40 . However, the correlation between these models and liver fibrosis in HCC patients is unclear. We have recruited a large cohort of  patients undergoing surgical resection to evaluate the degree of liver fibrosis and its association with these models. Surprisingly, our data reveal that none of these models significantly correlate with the severity of histological fibrosis in HCC patients. This finding suggests that the performance of currently used liver reserve models in predicting the severity of liver fibrosis is far from satisfaction. The main possible reason is that the etiologies of fibrosis greatly vary in HCC patients, while these models were generated from patients with distinct clinical characteristics. We further examined the factors associated with advanced fibrosis, defined as an Ishak score of 4 or higher. Consistent with previous studies 18,20,27,[39][40][41] , low platelet count and prolonged INR, which are known surrogate markers of cirrhosis, are independent predictors of advanced fibrosis. With increasing fibrosis and worsening portal hypertension, there is increased sequestration and destruction of platelets in the enlarged spleen 42 . In addition, progression of liver fibrosis was linked with decreased production of thrombopoietin by hepatocytes, and hence reduced platelet production 43 .
An interesting finding in this study is that the factor of multiple tumors was also independently associated with advanced fibrosis. Previous studies showed that non-cirrhotic HCC patients may have larger tumor size than those with cirrhosis; one of the possible explanations is that non-cirrhotic patients were often diagnosed outside the surveillance program 44    burden and CTP score and MELD score 46 . However, in the current study, we found that the surrogates of tumor burden tended to associate with the severity of liver fibrosis. Smaller tumor burden may predict a lower score in most noninvasive models except for the comparison between CDS and tumor size. These findings partly explain why these noninvasive models correlate poorly with histological fibrosis, and suggest that the selection of tools in evaluating liver injury for different clinical entities is crucial. Altogether, not only surrogate markers of cirrhosis, the extent of tumoral involvement should also be taken into consideration in predicting non-tumoral part liver damage in HCC. Noninvasive models, including APRI, CDS, FIB-4, GUCI, King's score and Lok's index, were reported to correlate with the degree of fibrosis in HCV-infected patients 20,[22][23][24]27,29,38 . In the current study, we confirm that these 6 models may reflect the severity of liver injury as defined by the Ishak score in HCV-related HCC patients. Additionally, the ALBI score is a new prognostic marker for HCC, and our data indicate that it can also be used to assess liver fibrosis in HCV-related HCC.    Table 6. Correlation of tumor burden, serum AFP level and noninvasive liver reserve models.
Several models, such as APRI, CDS, FIB-4, Lok's index and King's score, were also reported to associate with histological fibrosis in HBV-infected patients 23,47,48 . In the current study, however, the correlation is apparent only for CDS and Lok's index. Our result implies that noninvasive models derived from HCV-infected patients may not be necessarily feasible in predicting histological fibrosis in HBV-related HCC. The pathogenesis of liver fibrosis in chronic hepatitis B is distinct and could be different from that of hepatitis C. Activity of hepatitis B can become quiescent after a period of severe activity, such as recurrent hepatitis flares, or resolution of hepatic necroinflammatory activity following HBV e antigen seroconversion after the development of cirrhosis 49 . In contrast, chronic hepatitis C is a slow but progressive disease with persistent inflammation that ultimately leads to cirrhosis 50 . HCC patients due to HBV or HCV are usually at a late stage of infection and may thus have heterogeneous patterns of liver fibrosis that make prediction with noninvasive models much more difficult.
Noninvasive liver reserve models were also adopted as surrogate markers for discriminating cirrhosis from chronic hepatitis. In accordance with previous studies 20,22,24,38,39,48 , we found that APRI, CDS, FIB-4, GUCI, Lok's index and King's score had moderate power to predict cirrhosis in HCC, with an AUROC between 0.703-0.729. Among these models, CDS was identified as the best model to predict cirrhosis. Given so, the predictive accuracy is considered not satisfactory, and new models are needed to refine the predictive ability for cirrhosis in HCC.
This study has a few limitations. Firstly, in this single-hospital study, the major etiology of HCC is HBV infection. This feature is apparently different from Western counties where HCV infection is the predominant etiology of HCC. Secondly, our hospital is a tertiary medical center. Therefore referral bias cannot be completed avoided. Lastly, since this study is retrospective in nature, external validation from independent patient cohorts is required.
In conclusion, the currently used noninvasive liver reserve models do not well correlate with the severity of histological fibrosis in HCC. Among these models, CDS is more accurate in predicting the presence of cirrhosis. Different models should be used for HCC patients with different viral etiology. In addition to traditional cirrhosis surrogates, the extent of tumoral involvement and viral factor are crucial determinants that contribute to liver injury. We advocate that new models should be explored to enhance the predictive ability for liver fibrosis in the setting of HCC.