An Easy-to-use Clinical Nomogram for Predicting the Prognosis of Patients with Hepatocellular Carcinoma Concomitantly Suffer from Severe Fibrosis or Cirrhosis: A Population-based Analysis. CURRENT STATUS: UNDER REVIEW

Background Patients with hepatocellular carcinoma (HCC) concomitantly suffer from liver cirrhosis may have worse prognosis. Based on Surveillance, Epidemiology, and End Results (SEER) database, we evaluated the overall survival (OS) and cancer-specific survival (CSS) of these patients. Methods A total of 2,369 patients were selected from the SEER database. They were classified into F0 (n=691) and F1 (n=1,678) groups by different Ishak fibrosis score. Propensity score matching (PSM) and Kaplan-Meier method were performed to evaluate the OS and CSS. The F1 group were randomized into training sub-set (n = 1,176, 70%) and validation sub-set (n = 502, 30%) for further construction and validation of nomogram . Results After matched, there were statistically significant worse outcome for F1 group patients compared with F0 group (n=587, OS: P<0.001, CSS: P<0.001). Six independent predictors for both OS and CSS were identified to construct the nomograms by COX regression analyses. The nomogram performed well concerning its ability of discrimination and calibration and its net benefits compared with the conventional staging system. Conclusions Patients with HCC concomitantly suffer from severe fibrosis or cirrhosis has a significant worse survival compared with none or moderate fibrosis patients. The validated nomograms provided useful prediction of survival.


Introduction
Hepatocellular carcinoma(HCC), ranking as the third leading cause of cancer-specific deaths around the world, is a global health issue with its high mortality ratio. [1] Patients with cirrhosis are exposed to high risk of HCC, ranging from 1-8% per year, and HCC can emerge at any stage of cirrhosis [1][2][3]. For several common reasons, a great deal of patients with HCC concomitantly suffer from liver cirrhosis. [3,4] Accordingly, in addition to the clinicopathological characteristics of tumor itself, the outcome of these patients is strongly associated with the stage of liver cirrhosis [3,5,6] and its severe complications. [7] Liver biopsy is still the recognized standard for histologic evaluation of liver disease activity and degree of fibrosis, even if some limitations existed. [8] Over the past decade, the Ishak scoring system has extensively applied in clinical trials, particularly in America.
Histologically, higher Ishak score reflects more scarring, thus, investigators assume that patients with a higher fibrosis stage should have an growing risk of prognosis compared to its lower counterpart. However, since clinical researches always exclude cases with underlying cirrhosis, the correlation between the degree of fibrosis and the clinical prognosis has rarely been confirmed in a prospective randomized controlled trials. Most of these studies were retrospective and cases were mostly derived from small sample size trials. [9,10] Surveillance, Epidemiology and End Results (SEER) database, containing the long followup duration and a great number of cases of patients with Ishak fibrosis score, giving us registered adequate events to discuss the differences in patients with HCC concomitantly suffer from liver cirrhosis in all-cause mortality and cancer-specific mortality.
Nomograms, a novel statistical predictive model, can accurately calculate and estimate individual survival [11,12]. A wide variety of nomograms have been extensively established to give help to formulating personalized therapy and follow-up management strategies in various diseases. For all we know, the study to estimate prognosis of patients with HCC concomitantly suffer from cirrhosis using data derived from the SEER database has never been carried out.
In present study, we aimed to estimate the survival difference in HCC patients between no to moderate fibrosis and advanced/severe fibrosis. Then, based on clinicopathological risk factors derived from the SEER database, we aimed to construct and validate an easy-touse and effective nomogram model to compute and estimate survival of patients with HCC concomitantly suffer from severe fibrosis or cirrhosis.

Materials And Methods
Patient and data collection SEER database is one of the most representative cancer registration databases. In this study, we used the SEER 18 Regs Research Data (1975-2016 to search for cases with HCC, and a total of 108711 cases between 1975 and 2016 were fetched. These factors were eventually selected in present study: age at diagnosis, racial status, gender, marital status, international classification of diseases for oncology, third edition (ICD-O-3) [13], Edmondson-Steiner classification, American Joint Committee on Cancer (AJCC) 7th edition TNM Staging System, alpha-fetoprotein (AFP), fibrosis (or Ishak) score, treatment, months of survival, cancer-specific survival status and overall survival status,. The Ishak score [14] is a reliable and significant prognostic indicator of liver disease. The characteristics of fibrosis were classified by scores defined by the AJCC.
Ishak fibrosis scores were defined by AJCC: 1) scores from 0 to 4 means none fibrosis to moderate fibrosis, 2) scores 5 and 6 means severe fibrosis or cirrhosis. In this study, the severity of the liver fibrosis or cirrhosis were classified as F0 and F1. Patients with complete fibrosis score data in SEER database during 2010-2016 were selected. After deleting cases with an unknown data of racial status, Edmondson-Steiner classification, AFP, tumor size, AJCC 7th stage and survival time, we eventually selected 2,369 patients for this study. The flow chart (Fig. 1) shows the selection of study population in this study.
Overall survival (OS) and cancer-specific survival(CSS) were both the interest endpoints in this study. OS was regarded as the duration from the date of diagnosis to death from any cause, while CSS was defined as the duration from the date of diagnosis until death due to HCC in the absence of other causes. This study was approved by the Ethics Committee of Jinan University Ethics Committee, and written informed consent was waived.

Propensity Score Matching
Patients were divided into F0 and F1 groups on the base of different fibrosis score. To ensure well-balanced characteristics for the two comparisons, propensity score matching (PSM) with 1:1 proportion was performed. The chi-square test was used to compare the categorical variables of both groups described as the number and percentage, and student's t-test was used to compare the continuous variables expressed as the mean and standard deviation. The short-term and long-term outcomes of patients between two groups were estimated using Kaplan-Meier survival analysis (log-rank test).

Statistical Analysis
Those selected 1,676 patients with HCC concomitantly suffer from severe fibrosis or cirrhosis were randomized to a training set (n = 1176, 70%) and a validation set (n = 500, 30%). One was for the construction of the nomogram and another was for validating it. Cox proportional hazards model were performed to analysis the potential confounders and identify the independent predictors in training set. [15]. Hazard ratios(HR) and their 95% confidence intervals (CI) were estimated using the univariable and multivariable Cox regression analysis. We used the backward step-down process to obtain the final factors for the development of the nomogram based on the principle of the Akaike information criterion (AIC) [16,17].
The discrimination ability of the model was assessed by the concordance index(C-index) [18]. The accuracy of predictions was measured by plotting the calibration curves of the nomogram. The precision of the prediction for survival was estimated by using area under receiver operating characteristic (ROC) curve (AUC). In addition, the clinical efficacy and net benefit of the models was evaluated by decision curve analysis(DCA) [19,20]. The survival differences between the different level of risk groups were compared by using Kaplan-Meier method.
The data was originated from the SEER*Stat software version 8.3.6. X-tile software was applied to calculate the optimal cut off age, tumor size and the score of nomogram based on the OS and CSS of patients. The DCA curves was plotted by R software with file "stdca.R"(available from the site www.mskcc.org) and relevant packages. Statistical analysis was performed by using R(version 3.6.1). A two-sided P < 0.05 was regarded as statistical significance.

Result
Propensity score matching A total of 2369 patients in 2004-2016 were eventually selected from the SEER database, they were divided into two groups according to Ishak score: F0 (Ishak score was 1-4) and F1 (Ishak score was 5-6) groups. There were 1678 cases in group F0 and 691 patients in group F1. After 1:1 propensity score matching, 587 patients from each group were identified. The balance test showed that the distribution of covariates of selected cases in the two groups was adequately balanced (Fig. 2a, 2b, 2c and 2d). Table 1 shows the demographic and pathological characteristics before and after matching between the matched groups, indicating that the potential confounders in two groups were minimized.    DCA is a novel method for estimating clinical usefulness of a prediction model based on the threshold probability [19,20]. It was built to assess whether the novel model improves decision-making of the 1-, 3-and 5-year OS and CSS prediction in training set. The DCA curves of the nomogram model, the AJCC 7th edition system and the AJCC 8th edition system were shown in Fig. 5(a, b, c,

Discussion
In this study, we estimated the role of liver fibrosis or cirrhosis in hepatocellular carcinoma using the national SEER database for the first time. In order to eliminate selection bias, PSM was performed in this population-based research. Processed with the combination of clinical and pathological covariates, the propensity score matching made a comparable distribution of the clinicopathological characteristics between the F0 and F1 cohorts, thus bringing about a result that was similar to random allocation [21]. Based on population-based database, we constructed a nomogram to evaluate the definite 1-, 3-and 5-year OS and CSS probabilities of patients with HCC concomitantly suffer from severe fibrosis or cirrhosis, then, the ability of the nomograms were verified concerning its discrimination and calibration. As a result, the nomograms were performing well in both the validation set and the training set. In the perspective of net benefit and clinical efficacy, the novel models showed wider range of threshold probabilities than some conventional systems.
According to our results, the type of therapy was the strongest predictor of outcome. In fact, liver transplantation(LT), derived from both cadaveric and living donation, often represents the only curative treatment which is able to simultaneously cure the HCC and the liver cirrhosis. Mazzaferro et al described that selected patients can achieve more considerable survival benefit compared with those of whom transplanted for benign cases in a published landmark paper. [24,25] The success of LT is not subjected to the severity of liver dysfunction, it could be able to improve survival and the quality of life in selected patients with end-stage liver disease [2,3,24]. However, this option may be precluded to a significant number of patients because of age, comorbidities, tumor characteristics, shortage of donor organs and some other limitations [26]. Certainly, when these factors hinder the chance of LT, liver resection should be considered as a precious option. Due to severe complications (severe portal hypertension, thrombocytopenia etc.) in patients with advanced cirrhosis, Barcelona Clinic Liver Cancer (BCLC) algorithm recommend hepatectomy as the preferred therapy for patients with single tumor and well-preserved liver function (Child-Pugh A). But a recent study pointed out that hepatectomy for HCC in Child-Pugh B cirrhosis could be also feasible, after careful preoperative assessment based on patients' features, liver function and tumor pattern as well as decreasing surgical stress [27]. Unlike hepatectomy, terrible liver function does not mean a contraindication of real-time image-guided local radiofrequency ablation(RFA). Patients with tumors less than 3 cm and Child-Pugh A or B who have no indication for hepatectomy are candidates for RFA [3]. A retrospective study that contained 7,185 patients concluded that hepatectomy might contribute a lower rate of recurrence than RFA in small HCC. Another single-center study described that patients who underwent hepatectomy had a longer OS [28,29].
Nearly 40% of the cases did not receive operation, and the reason for the non-operation therapies was that it was contraindicated or was not recommended. It indicated that severe fibrosis or cirrhosis patients with HCC were more likely to suffer from an end stage of HCC.
Tumor size is the core element of AJCC 8th edition T stage and an integral part of the AJCC 8th edition TNM stage system. T stage was always considered as a crucial predictor for HCC [30,31]. The present study has indicated that advanced T stage meant higher risks of OS and CSS. Meanwhile, heavier weight from T stage in calculating overall survival than cancer-specific survival was observed, indicating that CSS were more largely depended on the intrinsic nature of HCC. Tumor grade, another factor showed inherent feature of tumor, is widely accepted as one of the most effective prognostic factors of outcome in patients with HCC both after hepatectomy and LT, as shown by many series study of these topics [32][33][34][35][36][37][38]. In accordance with the previous study, our study indicated that a poorly differentiation was associated with a worse prognosis. AFP has been highly applied not only in diagnostic biomarker but in evaluating the outcome of HCC for years. Previous studies found that high AFP level before treatment was an independent predictor concerned with tumor grade, progression and survival [39].The present study supported that the AFP level is an negative and independent predictor for both CSS and OS of severe fibrosis or cirrhosis patients with HCC.
The predictive value was not observed for N stage in this study. The proportion of patients with lymph node (LN) metastasis was fairly low. T diagnosis of LN metastasis was based on intraoperation exploration and pathological confirmation, rather than imaging examination in present study. It might result in the lower rate of LN metastasis.
Additionally, based on the risk stratification of prognostic curves, the novel nomogrampredicted models had better discrimination than that of AJCC systems both in OS and CSS, as shown in Fig. 5. For overall survival, the AJCC 7th edition system was underperformed in stratifying stages I and II patients. The AJCC 8th stage systems shown good prognostic stratification for patients, but its C-index was only 0.650(95%CI, 0.626-0.674, training set) and 0.662(95%CI,0.629-0.695, validation set) respectively. However, in both sets, four sub-groups(the low-A risk, low-B risk, moderate risk and high risk groups) in our nomograms had significant differences in OS and CSS, and showed better accuracy and discrimination in both short-term and long-term survival prediction. The most likely reason is that the AJCC system only take the tumor size, positive regional LN and metastasis into consideration. These result indicated that our nomograms were able to use as a conventional tool for predicting the short-term and long-term outcome of severe fibrosis or cirrhosis patients with HCC. However, it was ignored that racial differences, differentiation grades, and therapy methods were also independent predictors for prognosis.
As is known to us, this study established the first nomogram model for predicting the outcomes of severe fibrosis or cirrhosis patients with HCC. By calculating the score of the variables, clinicians can not only predict the prognosis immediately and accurately but also obtain valuable information to choose treatments and predict survival rates before therapy option. Meanwhile, doctors can easily distinguish different level risk of patients after treatment, careful and regular follow-up should be made in high-risk populations.
Then, SEER database provided multicenter clinical data, made our results more applicable to the general population than that at a single institution.
However, some limitations were still existed. Firstly, This was an retrospective observational study exposed to potential confounding bias. Therefore, a 1:1 PSM analysis was performed to simulate a realistic scenario of two homogeneous populations. Then, liver function is also a vital factor for the prediction of both cirrhosis and HCC, but the data was not available from the SEER database in our nomograms. Therefore, external validation should be performed by using an independent external dataset for the prognostic nomograms developed in this study in the future. and further randomized evidence is required to verify the conclusions of our study.

Conclusion
The present study shows the potential survival risk of severe fibrosis or cirrhosis in HCC by analyzing the population-based SEER database. Given the independent prognostic impact of the severe fibrosis or cirrhosis on both OS and CSS, these nomograms were

Consent for publication
Not applicable

Availability of data and materials
All data generated or analyzed during this study are included in the published articles

Competing interests
The author reports no conflicts of interest in this work.

Funding
This study was supported by The Natural Science Foundation of Guangdong Province(1714050005583)

Author contributions
WHY and RSY acquisition of data, analysis and interpretation of data; ZYP and JL contributed to data analysis; JHS conception and design of the study, acquisition of data, analysis and interpretation of data, drafting the article, final approval; HCF critical revision, final approval.