Pre- and Postoperative Models for Prediction of Recurrence in Non-B, Non-C Hepatocellular Carcinoma

Background and Aims The incidence of non-B, non-C hepatocellular carcinoma (NBNC-HCC) is increasing. Like in hepatitis B virus (HBC)/HCV-associated HCC, treatment of NBNC-HCC after resection is challenging due to its high recurrence rate. However, few studies on the recurrence of NBNC-HCC have been published in the past decades. Hence, we aimed to investigate the risk factors for recurrence of NBNC-HCC and construct pre- and postoperative prognostic models for predicting recurrence in these patients who underwent curative resection. Methods We retrospectively analyzed 608 patients who underwent liver resection for NBNC-HCC. A multivariate Cox proportional hazard regression analysis was conducted to identify the independent risk factors of recurrence, based on which the prediction nomogram models were constructed and validated. The predictive performance of the models was assessed using the concordance index, time-dependent receiver operating characteristic curve, prediction error cure, and calibration curve. To facilitate clinical use, we stratified the patients into three distinct risk groups based on the score of the models. The cutoff scores of the models were determined by a survival tree analysis. Results Multivariable analysis identified neutrophil-to-lymphocyte ratio, alpha fetoprotein, tumor number, and tumor diameter as independent preoperative risk factors for recurrence. In addition to these variables, microvascular invasion was an independent postoperative risk factor for recurrence. The pre- and postoperative nomograms were constructed based on these variables. The C-index of the pre- and postoperative nomograms was 0.689 and 0.702 in the training cohort, 0.682 and 0.688 in the validation cohort, respectively, which were both higher than those of the conventional Barcelona Clinic Liver Cancer (BCLC) and 8th edition of the American Joint Committee on Cancer (AJCC8th) staging systems. In addition, the pre- and postoperative nomograms could also re-stratify patients with BCLC stage 0/A or AJCC8th stage IA/IB/II into distinct risk groups. Conclusions We constructed pre- and postoperative prognostic models for predicting recurrence in patients with NBNC-HCC who underwent curative resection. They can play a supplementary role to the traditional staging system.


INTRODUCTION
Hepatocellular carcinoma (HCC) is the fifth most common cancer and third leading cause of cancer-related deaths worldwide (1). Chronic infections with hepatitis B (HBV) and C viruses (HCV) are the prominent etiological factors for HCC. However, with the adoption of HBV immunization programs and control of HCV transmission, the incidence of HBV\HCVassociated HCC has decreased in recent years (2,3). The number of patients with HCC who are seronegative for both HBV and HCV, so called "NBNC-HCC," is gradually increasing (4,5).
Due to the shortage of donor organs, surgical resection remains the main treatment strategy for patients with HCC who have a good liver function and resectable tumors (6). However, long-term survival after surgery remains unsatisfactory due to the high incidence of tumor recurrence (7). Owing to the heterogeneity of HCC, some highly selected patients may benefit from a prognostic prediction model, wellselected therapeutic assignment, and strict postoperative monitoring (8,9). There are several studies on constructed prognostic models for HCC; however, very few studies have specifically focused on the recurrence of patients with NBNC-HCC. Considering the different clinical manifestations and prognostic outcomes of NBNC-HCC and viral-associated HCC, the study of its risk factors and establishment of prognostic models may provide important insight into novel strategies for the treatment and postoperative monitoring of NBNC-HCC.
In our study, we aimed to: (1) investigate the relationship between inflammatory indexes and recurrence of NBNC-HCC, (2) investigate the pre-and postoperative risk factors for recurrence of NBNC-HCC, and (3) construct a pre-and postoperative nomogram model for the prediction of recurrence of NBNC-HCC.

Patients
Data from patients diagnosed with NBNC-HCC who underwent hepatectomy as a primary anti-cancer therapy between April 28, 2008 andDecember 30, 2015 were extracted from the primary liver cancer big data (PLCBD) (21,22). In this study, patients with HCC who were seronegative for hepatitis B surface antigen, HBV-deoxyribonucleic acid (DNA), hepatitis C antibody, and HCV-RNA test were considered patients with NBNC-HCC (23). All data in this study were verified by three independent researchers, and the study was approved by the institutional ethics committee of Mengchao Hepatobiliary Hospital of Fujian Medial University.
Patients with Child Pugh A or B7 liver function, no extrahepatic metastasis, no macroscopic vascular invasion, and who underwent R0 resection (complete removal of all detectable tumor nodes with tumor-free margins confirmed by histological examination) were included. Patients who underwent palliative tumor resection or any preoperative anti-HCC therapy, with a history of other cancers, or incomplete clinical data were excluded. load, alpha-fetoprotein (AFP) levels, liver and kidney function examinations, chest radiography, abdominal ultrasonography, contrast-enhanced computed tomography, and magnetic resonance imaging. HCC was diagnosed according to the practice guidelines recommended by the American Association for the Study of Liver Diseases (24). The choice of performing anatomical or partial hepatectomy is commonly based on the patient's liver function status, tumor number, and location. In simple terms, anatomical hepatectomy was preferentially performed for patients with a good liver function and tumors located within a segment, sector, and hemiliver. Partial hepatectomy was performed for tumors that were peripherally located or patients with unsatisfactory liver function. Intraoperative liver ultrasonography was routinely used to ensure that all detectable tumors were completely removed. The follow-up of patients after discharge was performed in the outpatient clinic. The follow-up program and diagnostic criteria for tumor recurrence are reported in a previous study (9).

Clinicopathologic Variables
For hematological investigations, we used the results of the most recent test that was performed within 15 days prior to surgery. The formulae of the indices (NLR, PLR, GPR, PNLR, PNI, ANRI, ALRI, DNLR, and NgLR) are reported in a previous study (14,18,(25)(26)(27). The preoperative imaging data, including tumor number and diameter was obtained from contrast-enhanced Computed Tomography or contrast-enhanced magnetic resonance imaging. Tumor diameter was the diameter of the largest tumor. Histologic grading of the tumor was done according to the Edmondson-Steiner classification. The definition of microvascular invasion (MVI) from a previous study was used (28). The pathological review of all resected specimens was carried out independently by two pathologists.

Statistical Analysis
Categorical variables were expressed as number (percentage) and compared using the c2 test or Fisher's exact test. Continuous variables were expressed as mean (standard deviation SD) and compared using the Student t test or the Mann-Whitney U test. Univariate and multivariate Cox proportional hazard regression analyses were performed to determine the independent prognostic factors for recurrence. Clinical variables considered to be potentially relevant (p<0.05 in univariate analyses) were entered in the multivariate Cox proportional hazard regression analyses, and the final independent risk factors for recurrence were identified by the multivariable analyses with stepwise backward selection method.
The nomograms were constructed based on the results of multivariable analyses of recurrence in the training cohort. The R code used for construction of nomogram was shown in the supporting information. The preoperative nomogram was developed on the available preoperative clinicopathologic data. The postoperative nomogram was based on all available clinicopathologic variables. The predictive performance of the nomograms was measured using Harrell's concordance index (C-index), time-dependent areas under the receiver operating characteristic curve (tdAUC), prediction error curve, and calibration plot (29,30). The cutoff of models was determined by a survival tree analysis (29). The cumulative recurrence between each risk group was assessed and tested using Kaplan-Meier curves and the log-rank test, respectively. All statistical tests were two-tailed, and P < 0.05 was considered statistically significant. Statistical analyses were performed using SPSS version 20.0 and R version 3.0 (http://www.r-project.org/); the R packages of "table1," "rms," "CsChange," "survminer," "survival," "pec," "riskRegression," "timeROC," and "party" were used.

Patients' Characteristics
A total of 608 eligible patients were included in the study. They were randomly divided into a training cohort (n=456) and validation cohort (n=152) in a 3:1 ratio. A comparison of the clinicopathologic characteristics between the two cohorts is shown in Table 1. There were no significant differences in the clinicopathologic features between the two cohorts. The mean age was 59.2 ± 11.2 years and 58.3 ± 11.7 years in the training and validation cohorts, respectively. Most of the patients were male (85.5-87.5%), and few patients had cirrhotic livers (32.9-35.5%). The mean tumor diameter was 7.06 ± 3.99 cm and 6.80 ± 4.12 cm in the training and validation cohorts, respectively. Most of the patients harbored solitary tumors (86.2-83.6%).

Construction of Pre-and Postoperative Nomograms for Prediction of Recurrence
All the variables shown in Table 1 were included in the univariate and multivariate Cox proportional hazard regression analyses to identify independent risk factors for recurrence. Results of the univariate analysis are shown in Table 2. On univariate analysis, there was a significant relationship between six inflammatory indexes (NLR, PLR, ALRI, PNLR, NgLR, and DNLR) and recurrence. Other significant variables were neutrophil count, lymphocyte count, aspartate aminotransferase, AFP, tumor diameter, tumor number, MVI, and satellite nodules.
To construct the preoperative model, presence of MVI and satellite nodules, which were recorded only postoperatively, were excluded, and the remaining 12 factors were entered into the multivariate Cox regression analysis using a stepwise method. All 14 risk factors that could be acquired preoperatively were entered in the multivariate analysis.
Multivariate  Table 4). These independent risk factors were then used to build the pre-and postoperative nomogram models ( Figures 1A, B).
For the postoperative nomogram model, the corresponding tdAUC was 0.751 (range, 0.722-0.753) in the training cohort and 0.755 (range, 0.687-0.768) in the validation cohort. Both models had higher tdAUCs than those of the BCLC and AJCC staging systems (Figures 2A, B). In addition, the prediction error curve analysis was used to assess the overall performance of the models. The nomogram models had a lower prediction error rate than the conventional staging systems (Figures 2C, D). The calibration plots also displayed a good agreement between predictions of the pre-and postoperative nomogram models and the probability of 1-, 2-, and 3-year recurrence in the training and validation cohorts (Figure 3).

Risk Stratification Based on the Nomograms' Score
Each patient received an individualized risk score according to the individual scores calculated with the nomograms. We performed a survival tree analysis to determine the cutoff points of the risk score ( Figure 4). Based on the cutoff points, the patients were stratified into three different risk subgroups (low-risk, intermediate-risk, and high-risk). The Kaplan-Meier analysis showed that the recurrence curves were widely separated among the three different risk groups in the training and validation cohorts, which further indicated that the nomograms had good discrimination ability for recurrence ( Figure 5).    In addition, we combined the models with the BCLC/AJCC 8th staging system and found that the models had a good restratification effect on the traditional staging systems. Because there were only 12 patients with BCLC stage 0 and AJCC stage IA tumors, these patients were incorporated into the group of patients with BCLC stage A or AJCC 8 th stage IB. As shown in Figure 6, both models could re-stratify the patients with different recurrence risks in BCLC stages 0/A (p<0.001, preoperative nomogram model; p<0.001, postoperative nomogram model) well. The re-stratification ability of the models persisted in patients with BCLC B stage (p=0.044, preoperative nomogram model; p=0.037, postoperative nomogram model), but given that only a small number of patients were in the high-risk group, the re-stratification ability of models in patients with BCLC B stage needed be validated in a further large sample cohort. For the AJCC 8th staging system, we found that the models distinguished patients with stages IA/IB (p<0.001, preoperative nomogram model; p<0.001, postoperative nomogram model), and II (p<0.001, preoperative nomogram model; p=0.002, postoperative nomogram model) well (Figure 7).

DISCUSSION
In this study, we reported significant independent risk factors for recurrence of NBNC-HCC, which included large tumor diameter, multiple tumor number, elevated AFP levels, elevated NLR, and presence of MVI. We also constructed pre-and postoperative nomograms for individualized prediction of recurrence in patients with NBNC-HCC who underwent curative resection. Both nomograms had a better predictive performance than the currently used BCLC and AJCC 8th staging systems. Additionally, according to the individualized scores assessed using the nomograms, patients in cohorts can be stratified into three risk groups. These two easy-to-apply graphical models will be valuable in preoperative treatment planning, adjuvant therapy implementation, postoperative monitoring, and designing of clinical trials based on prognostic stratification.
Tumor diameter and number are important prognostic factors in the BCLC staging system; however, these tumor factors are insufficient to reflect the malignant characteristics  of HCC (32). Compared to the BCLC staging system, the AJCC 8th staging system included the presence of MVI as a stratification criterion; however, given that MVI is a pathological feature that can only be diagnosed after surgery, it limits the applicability of this system in preoperative clinical decision making. Moreover, both the BCLC and AJCC 8th staging systems were not developed specifically for the prediction of HCC recurrence. The nomograms in this study integrated the independent risk factors for recurrence of NBNC-HCC, including the above three tumor-associated factors; serum tumor biomarker, AFP; and inflammatory index, NLR; thus, making them more accurate recurrence predictors. Our recurrence nomograms were able to re-stratify patients in the same traditional staging system stages, and thus can play a supplementary role to the traditional staging system. The preoperative nomogram in this study, which had a significantly better predictive performance than the BCLC staging system, may be an additional tool for surgeons to identify high-risk patients before operation; thus, it may be valuable in preoperative treatment planning [e.g., preoperative transarterial chemoembolization (TACE), widened surgical margin]. Owing to the lack of consensus on follow-up procedures for the postoperative diagnosis of recurrence of HCC (33), using the postoperative nomogram can help surgeons to design stricter follow-up procedures (e.g., reduced interval of follow-up and more high-end imaging tests) and postoperative adjuvant therapy for the highrecurrence risk patients. For the current results, it is difficult to specify which model is better for the recurrence prediction. In terms of the predictive performance of the models, the C-index and tdAUC of the postoperative model were higher than those of the preoperative model, and the prediction error curve also shows that the postoperative model is better than the preoperative model, although the difference was not obvious, the postoperative model may be superior to the preoperative model considering that the postoperative model includes the indicator of MVI, MVI is currently widely recognized as an independent risk factor for recurrence in HCC. However, when the preoperative model shows poor prognosis of patients and the postoperative model  shows good prognosis, it may more proper to choose the preoperative model. After all, the cost of shortening the follow-up interval is much less than the cost of finding recurrence of HCC too late. Further prospective studies may be needed to distinguish which model has better predictive power. In addition, we included an inflammatory index in this study. To the best of our knowledge, this is the first study to investigate the association between a system inflammatory index and recurrence in NBNC-HCC. In this study, we analyzed nine inflammatory indices and found that the NLR was an independent risk factor for recurrence in NBNC-HCC. High NLR has been reported to be a poor prognostic factor for recurrence and OS in several malignancies, including HCC (13,34). Given that NLR consists of serum neutrophil count and lymphocyte count, elevation of NLR can be reviewed as elevated neutrophil and lymphocyte counts. Neutrophilia can promote cancer cell growth and progression by releasing angiogenic factors and inflammatory mediators (28,35,36). On the other hand, lymphocytes play an anti-cancer role in host immunity by inducing cytotoxicity and inhibiting proliferation, invasion, and migration of cancer cells; lymphopenia may weaken this anti-cancer effect (37)(38)(39). All these factors included in the NLR were adverse factors for HCC, which could possibly explain why NLR could be used to evaluate the recurrence probability of patients with HCC. However, the mechanism underlying the association between NLR and recurrence of NBNC-HCC remains unclear and needs further elucidation.
This study had some limitations. First, it was limited by its retrospective nature and limited sample size; thus, selection bias was unavoidable. A large sample, outer validation cohort, and prospective study are needed to validate the nomograms in future. Second, postoperative adjuvant therapy information, such as postoperative adjuvant TACE (PA-TACE), has not been included in our study. Given that PA-TACE may improve the prognostic outcome of high-risk patients (40,41), whether high-risk patients assessed by these nomograms can benefit from PA-TACE is to be determined. Third, the prognostic value of postoperative inflammatory indexes or the changed indexes after post-operation compared to pre-operation in HCC have been recognized in some studies, but this study did not include because the data of some postoperative inflammatory  indexes were missing. Last, the nomograms were generated using data of patients who underwent radical resection, which may not be applicable for patients receiving other therapies.

CONCLUSION
In summary, we developed pre-and postoperative nomograms for predicting the recurrence of patients with NBNC-HCC after resection. These two easy-to-apply graphical models will be valuable in guiding preoperative treatment planning, adjuvant treatment implementation, and postoperative monitoring for patients with NBNC-HCC.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

AUTHOR CONTRIBUTIONS
KL, QH, LW, WZ, and JL: study conception, design, quality control of this study, drafting of the manuscript, and statistical analysis. JZ, ZD, HL, JF, PG, and ZC: data acquisition and data interpretation.
YZ, WZ, and JL: resource and study supervision. All authors contributed to the article and approved the submitted version.

ACKNOWLEDGMENTS
We thank our colleagues and all the patients who participated in this study.