Development and Validation of a Prognostic Nomogram to Predict the Long-Time Prognosis in Non-B, Non-C Hepatocellular Carcinoma

Introduction

Hepatocellular carcinoma (HCC) is the sixth most common malignant tumor and the fourth cause of cancer-associated deaths globally.¹ Hepatitis B virus (HBV) and hepatitis C virus (HCV) infection are the main causes of HCC.² However, with the change of lifestyle, popularity of hepatitis B vaccines, and clinical use of effective antiviral drugs, the incidence of virus-related HCC is decreasing gradually^3,4 and accompanied with an increase of non-viral HCC, also known as non-B, non-C HCC (NBNC-HCC).^5,6

As with viral-associated HCC, surgical resection remains the main treatment that can provide long-term prognosis in NBNC-HCC.⁷ Yet, the long-term outcome in NBNC-HCC postoperatively is still not satisfactory. In previous studies, the 5-year overall survival (OS) rate in NBNC-HCC has been reported as only 42.6%-48.8%.^8,9 Because of the heterogeneous nature of HCC, accurate prediction of prognosis after treatment is of great significance. A good individualized prognostic prediction model may benefit some highly selected patients through well-selected therapeutic assignment.^10,11

Currently, there are several widely used HCC staging systems such as the Barcelona Clinic Liver Cancer staging system (BCLC)¹² and American Joint Committee on Cancer staging system, eight edition (AJCC^8th),¹³ which have important guiding significance for prognosis. However, they fail to achieve individualized prognostic prediction, so even with the same treatment, patients within the same stage of HCC tend to have different outcomes. In addition, these staging systems were not specifically constructed for patients with NBNC-HCC. As an individualized predictive tool to predict the prognosis of patients, a nomogram has been used and validated in various tumors in recent years.^14–16 The nomogram developed by Zhang et al¹⁷ was developed specifically for prognostic prediction in patients with NBNC-HCC who underwent hepatectomy, but the model was not validated in their study.

Therefore, this study aimed to construct a nomogram for individualized prediction in patients with NBNC-HCC after resection and validate its predictive performance through several complementary methods in order to determine whether the model can predict prognosis more accurately than other models.

Materials and Methods

Ethical Statements

The study was conducted in accordance with the 1975 Declaration of Helsinki. Written informed consent was obtained from all study patients. The institutional research ethics committee of Mengchao Hepatobiliary Hospital of Fujian Medical University approved the study (approval number: 2019_049_01).

Patient Selection

Data of patients who underwent hepatectomy as their primary anti-cancer therapy for HCC between August 21, 2008 and December 31, 2014 were identified and collected through primary liver cancer big data. Patients who met all of the following criteria were included in the research for further statistical analysis: (1) patients with NBNC-HCC, defined as seronegative for the hepatitis B surface antigen, HBV-DNA, hepatitis C antibody, and HCV-RNA test;⁶ (2) those with liver function of Child-Pugh class A/B7; (3) those who underwent R0 resection (removal of all detectable tumor nodes and negative surgical margin in postoperative pathological examination); (4) those without extrahepatic metastasis; (5) those without a medical history of other malignancy; and (6) those with complete clinical and follow-up data. All data were verified by three independent researchers.

Diagnosis and Hepatectomy

Routine preoperative assessments of patients included liver, renal, and cardiopulmonary function tests, alpha fetoprotein (AFP) analysis, hepatitis B/C immunology, and imaging examination. Imaging examination included radiography or computed tomography (CT) scan of the chest, abdominal ultrasonography, contrast-enhanced CT, or magnetic resonance (MRI). The diagnosis of HCC was determined by the appearance of typical radiological features on contrast-enhanced MRI, CT, or abdominal ultrasonography and confirmed by postoperative pathological examination. The use of anatomical or partial hepatectomy depended on the tumor variables, such as the diameter, number, and location and patient’s liver function status. Intraoperative ultrasonography was routinely performed to ensure that all detectable tumor nodes were removed. The removal of three or more Couinaud segments intraoperatively was regarded as major liver resection.¹⁸

Follow-Up

All patients were followed up regularly in the outpatient department after discharge. The interval of follow-up was about every 2 months during the first 2 years and every 6 months thereafter. The routine follow-up assessments included abdominal ultrasonography and laboratory examination of variables, such as liver function and the serum AFP level. When recurrence was highly suspected in patients on the basis of findings, such as an abnormal ultrasonography result or continuously elevated AFP level, contrast-enhanced CT or MRI was routinely performed. When tumor recurrence was diagnosed, appropriate treatments, which were based on the basic condition, reserved liver function, and tumor recurrence pattern of each patient, were prescribed according to the advice of the multidisciplinary team.

Outcome Measure

The end-point in this research study was OS, which was defined as the interval between the date of resection and the date of either death or the last follow-up.

Statistical Analysis

Continuous variables are expressed as mean (standard deviation), and they were compared using the Student’s t-test or Mann–Whitney U-test. Categorical variables are expressed as number (percentage), and they were compared using the chi-square test or Fisher exact test. Univariate and multivariate Cox proportional hazard regression analyses of the training cohort were performed to acquire the independent risk factor for OS. The statistically significant (p<0.05) variables in the univariate Cox regression analysis were chosen for further multivariate Cox regression analysis via the stepwise backward selection method. The nomogram was constructed on the basis of the independent risk factors of OS using the R package “rms” (Institute for Statistics and Mathematics, Vienna, Austria).

Predictive performance of the nomogram model was verified through discrimination and calibration.¹⁹ The discriminative ability of the nomogram was analyzed by the Harrell c-index and time-dependent areas under the receiver operating characteristic curve (tdAUC).¹⁰ Calibration of the nomogram was analyzed by using the calibration plot. Clinical utility of the nomogram was tested via the decision cure analysis (DCA).²⁰ Kaplan–Meier curves and the Log rank test for risk groups were further applied to measure the performance of the nomogram, and risk groups were generated by the previously reported cut-off values (50th and 85th percentiles) of the total points assessed using the nomogram.²¹

All the statistical tests were two-sided, and a p<0.05 was regarded as statistically significant. SPSS, version 20 (IBM Corp., Armonk, NY, USA) and R, version 3.5.2 (R packages “rms”, “CsChange”, “timeROC”, and “stdca”) were used to perform all statistical analyses in the study.

Results

Baseline Characteristics and Prognosis of Patients

According to the study’s inclusion criteria, 594 patients were enrolled and randomly categorized into the training cohort (n=396) or validation cohort (n=198) in a 2:1 ratio. The detailed flowchart for patient collection is shown in Figure S1. The baseline characteristics of the whole cohort and a comparison between the training and validation cohorts are shown in Table 1. Overall, non-alcoholic fatty liver disease (NAFLD) and alcoholic liver disease (ALD) were diagnosed in 6.9% and 9.4% of patients, respectively, and most patients (83.0%) were diagnosed with cryptogenic HCC. Most patients were male (87.2%); 26.8% of patients were diagnosed with hypertension, and 15.3% of patients had diabetes. Most patients had a solitary tumor node (83.8%), and the mean diameter of the tumor node was 7.39 ± 4.22 cm. According to the HCC staging system, 72.6% of patients remained BCLC A stage, and 56.4% of patients were classified as AJCC^8th IB stage.

Table 1 Baseline Clinical Characteristics of Patients with NBNC-HCC

The median follow-up times of patients were 59.2 months and 59.4 months in the training and validation cohorts, respectively. The 1-, 2-, 3-, 4-, and 5-year OS rates were 86.3%, 77.3%, 68.5%, 58.5%, and 50.8% in the training cohort and 84.0%, 76.3%, 69.3%, 61.2%, and 52.3% in the validation cohort, respectively.

Risk Factors for OS and Construction of the Nomogram

Results of the univariate and multivariate Cox proportional hazard regression analyses are shown in Tables 2 and 3. Multivariate analysis showed that five variables were independent risk factors of OS: large tumor diameter (p=0.015, hazard ratio [95% confidence interval]=1.042 [1.008–1.078]), multiple tumor nodes (p=0.003, 1.656 [1.184–2.315]), high preoperative serum GGT level (p=0.009, 1.517 [1.109–2.074]), presence of microvascular invasion (p=0.007, 1.593 [1.135–2.234]), and macrovascular invasion (p<0.001, 2.954 [1.937–4.505]) (Table 3). On the basis of these independent risk factors, a nomogram for predicting OS was constructed and verified in the training and validation cohorts (Figure 1).

Table 2 Univariate Cox Regression Analysis of OS in the Training Cohort

Table 3 Multivariate Cox Regression Analysis of OS in the Training Cohort

Figure 1 Nomogram for prognostic prediction in patients with non-B, non-C hepatocellular carcinoma who underwent hepatectomy.

Predictive Performance of the Nomogram

C-indexes of the nomogram model reached 0.702 (95% confidence interval [CI], 0.662–0.741) in the training cohort and 0.700 (95% CI, 0.643–0.758) in the validation cohort, which were higher than those of the BCLC staging system (0.636, 95% CI 0.599–0.672, p<0.001; 0.626, 95% CI 0.572–0.680, p<0.001), AJCC^8th staging system (0.658, 95% CI 0.620–0.695, p=0.002; 0.641, 95% CI 0.588–0.694, p=0.006), and Zhang et al’s model (0.658, 95% CI 0.618–0.698, p=0.003; 0.642, 95% CI 0.577–0.706, p=0.029).

The time-dependent AUC analysis also showed that the nomogram model had better discrimination than the other prognostic models (Figure 2A and B). Median tdAUC values of the nomogram were 0.743 (range, 0.736–0.775) in the training cohort and 0.751 (range, 0.686–0.793) in the validation cohort at various time points. Details of tdAUC values of the models are shown in Table S1.

Figure 2 Comparisons of the time-dependent area under the curve between the nomogram and other models at various time points in the training (A) and validation (B) cohorts.

For calibration of the nomogram, calibration plots depicted a good consistency between the predicted outcome of the nomogram and the observed outcome in terms of 3-, 4-, and 5-year OS in the training and validation cohorts (Figure 3A and B).

Figure 3 Calibration curves for predicting the 3-, 4-, and 5-year overall survival in the training (A) and validation cohorts (B).

In addition, with the nomogram, each patient received a corresponding total point. Median total points were 66.0 (range, 3.8–311.4) in the training cohort and 61.2 (range, 6.1–258.7) in the validation cohort. Values of 66.0 and 146.4 points, which represent the 50th and 85th percentiles of the total points in the training cohort, were used to categorize the patients into three different risk subgroups (low, intermediate, and high risk). The Kaplan–Meier analysis showed that the three survival curves were widely separated in the training and validation cohorts (both p<0.001), which further indicated the good predictive performance of the nomogram model (Figure 4A and B).

Figure 4 Kaplan–Meier plots for overall survival rates of risk groups defined by the nomogram model points. (A) Training cohort, (B) validation cohort.

Clinical Utility of the Nomogram

DCA was performed to measure the clinical utility of the model. DCA consisted of the continuous risk of the probability threshold (x-axis) and the net benefit (y-axis), which graphically demonstrated the model’s clinical utility. As shown in Figure 5A and B, DCA revealed that the nomogram had better net benefits than the BCLC staging system, AJCC^8th staging system, and Zhang et al’s model in the training and validation cohorts.

Figure 5 Decision curve analysis comparing the prognostic nomogram to other models in predicting 5-year overall survival. (A) Training cohort, (B) validation cohort.

Discussion

Because of different pathogenic mechanisms of hepatocarcinogenesis between virus-associated HCC and NBNC-HCC, the clinicopathologic features, especially tumor characteristics, of patients are very different between the two cancers. For instance, compared to patients with virus-associated HCC, those with NBNC-HCC tend to have a larger tumor diameter, solitary tumor node, and fewer vascular invasion events.^6,8 Although the conventionally used BCLC stage and AJCC^8th staging systems integrate the tumor-associated variables, such as the number of tumor nodes, tumor diameter, and vascular invasion, they do not consider that these factors are very different in HCC with a different etiology; thus, for patients with NBNC-HCC, they cannot achieve good predictive performance. In addition, none of these models includes the clinical serum biomarker, and this may be one reason why their prediction efficiency is poor. Furthermore, these staging systems cannot achieve individualized prognostic prediction.

In the present study, we analyzed the independent risk factors for the prognosis of NBNC-HCC and constructed a prognostic nomogram model. The results showed that the predictive performance of the model was satisfactory, which was verified by various methodologies. Further, the tdAUC analysis confirmed the aforementioned results. In addition, the DCA curve analysis displayed the superior net benefit of the nomogram compared to the other models. Compared with the Zhang et al’s study, our study included patients from 2008 to 2014, which ensured that patients received a long enough follow-up for the observation of outcome at 4 and 5 years after resection. This factor may be the reason for the better accuracy of our model in the prediction of long-time prognosis in patients with NBNC-HCC.

According to each individualized total point given by the nomogram model, we were able to divide patients into three different risk subgroups (high, intermediate, and low risk). For the whole cohort, the high-risk subgroup consisted of 15.7% of patients, with only a 5-year OS rate of 14.3%, whereas the low and intermediate risk subgroups accounted for 51.9% and 32.5 of patients, with 5-year OS rates of 66.1% and 44.3%, respectively. This easy-to-apply graphical model can be an additional tool for clinicians to make individual predictions of patients’ prognosis and identify high-risk patients among those with NBNC-HCC, which may be valuable in guiding corresponding follow-up strategies and postoperative adjuvant therapy.

Tumor diameter, multiple tumor nodes, vascular invasion and elevated serum GGT level were identified as the independent unfavorable factors of NBNC-HCC in this study. The effects of tumor diameter, multiple tumor nodes, and vascular invasion, which are relevant to aggressive behavior of a tumor, on the prognosis of HCC have been reported in many previous literatures.^6,9,22–25 In addition, several studies demonstrated the unfavorable effect of these factors in NBNC-HCC,^9,22 which is consistent with our results. GGT, a cost-effective and easily accessible serum biomarker, has been reported to have diagnostic and prognostic effects in various malignancies.^26,27 The possible mechanisms of GGT in HCC are as follows: (1) a high level GGT may induce DNA instability and result in tumor formation; and (2) the GGT level may be related to the aggressive behavior of HCC, such as poor differentiation, vascular invasion, or metastasis of tumor cells.^28–30 Li et al⁸ also demonstrated the unfavorable association between the elevated GGT level and prognosis of NBNC-HCC. These studies may support our clinical findings. However, the exact mechanism of the association between GGT and survival of NBNC-HCC remains unclear and requires elucidation.

It is worth noting that 83% of the patients in the present cohort had cryptogenic HCC. The risk factors of NBNC-HCC include NAFLD/non-alcoholic steatohepatitis (NASH), ALD, aflatoxin infection, and aristolochic acid intake.² Therefore, cryptogenic HCC in some of these patients may be caused by aflatoxin infection or aristolochic acid intake. Besides, NAFLD/NASH and ALD are mainly related to diet and lifestyle. The diet and lifestyle of Chinese individuals are different from those of western developed countries, which may be the reason why the proportion of NAFLD/NASH- and ALD-related HCC in our cohort differs from that in the western world.³¹ Currently, there are no data recording the contributions of ALD and NAFLD/NASH to HCC in China, although these entities will likely become leading causes of HCC in the future, and they deserve further exploration.³²

There are several limitations to our study. Firstly, the nomogram model was constructed using retrospective data; therefore, the results need to be further validated in prospective studies. Secondly, the etiology of most patients included in the present study was cryptogenic; considering regional disparity of etiology of NBNC-HCC, multicenter cohorts are necessary to validate the predictive performance of the nomogram. Lastly, the present study included only the patients who underwent resection, so the nomogram may not be suitable for patients who underwent other treatments.

Conclusions

In summary, we developed and validated a prognostic nomogram for individualized prediction in patients with NBNC-HCC who underwent resection. This novel nomogram model may provide clinicians with useful guidance for postoperative follow-up and treatments.