Introduction
The most frequent kind of cancer and the third leading cause of death from cancer worldwide is hepatocellular carcinoma (HCC).1 Approximately 14.0–36.7% of patients have distant metastases at the time of first diagnosis, and the five-year overall survival (OS) for HCC patients remains below 20%.2,3 Furthermore, the lungs, brain and bones are the most often extrahepatic metastasized locations in HCC. Although the clinical treatment and diagnosis of HCC patients have considerably improved in the last few years, the prognosis for patients with HCC, who also have distant metastases (DM), remains dismal.4 As a result, predicting models for estimating DM in HCC patients and the prognosis of HCC patients with DM are required. Numerous risk factors and prognostic variables have been identified in prior studies, such as higher T stage, surgery, tumor size, etc.5,6 However, limited studies have focused on the predictive and prognostic nomograms of DM in HCC patients. A nomogram is a convenient and effective tool for clinical management and diagnostics, because this allows for the prediction and quantification of the probability of clinical patient outcomes.7 The present study aims to construct diagnostic and prognostic nomograms for DM in HCC patients based on data obtained from the SEER database.
Materials and methods
Patients and study design
The data of HCC patients were extracted from 18 registry studies in the SEER database from 2010 to 2015. Inclusion criteria: (1) patients with a pathological diagnosis of HCC, with the liver as the primary site; (2) the demographic data of patients, such as race, age and gender, is accessible; (3) the tumor features, such as T stage, M stage, N stage and tumor size, are available. Exclusion criteria: missing information on the demographic data, T stage, N stage and M stage of the tumor, and missing or less than one month OS. Finally, a total of 14,508 HCC patients, which included 1,638 patients diagnosed with DM, was included for the present study. In order to build the diagnostic nomogram, all patients were examined to identify the risk factors for DM. In addition, all HCC patients with DM were evaluated to determine the prognostic factors, and construct the prognostic nomogram.
All HCC patients and HCC patients with DM were randomly divided into two groups, at a ratio of 3:2: training group and testing group. For the predictive and prognostic risk factors, the training group was used to construct the nomogram, and the testing group was used to verify the nomogram.
Data collection
In the present study, the characteristics used to detect the risk factors for developing DM in patients with HCC included the following: age, gender, race, T stage, N stage, histological type, and tumor size. In addition, three treatment-related variables (radiotherapy, surgery and chemotherapy) were included in the survival analysis. The major outcome measure in this section was OS, which was described as the gap of time between the diagnosis date and death date from any cause.
Statistical analysis
The SPSS 19.0 and R software (version 4.1.2) were applied to conduct all statistical analyses in the present study. Chi-square test was utilized to compare the distribution of characteristics between the testing and training groups. The statistical significance was defined as p < 0.05 (both sides).
Univariate logistic regression analysis was performed to determine the factors linked with DM in the diagnostic group, and variables with p < 0.05 were included in the multivariate logistic regression analysis to identify the risk factors correlated to DM in patients with HCC. The multivariate Cox regression analysis included the significant variables from the univariate Cox regression analysis, and determined the correlated prognostic factors for HCC patients with DM.
The prognostic and predictive nomograms were established using the “rms” package in the R software. In addition, the receiver operating characteristic (ROC) curve for the nomogram was drawn, the independent prognostic risk factors were identified, and the corresponding area under the curve (AUC) was used to determine the discrimination of the nomogram and risk factors.8 In order to further assess the nomograms, decision curve analysis (DCA) was used, and the calibration curve was plotted.9 Finally, all HCC patients with DM were divided into two groups based on their median risk score: high-risk and low-risk groups. Then, the Kaplan-Meier (K-M) survival curve was plotted, and log-rank test was performed to determine the difference in OS between the two groups.10
Results
Clinical and pathological characteristics of the study population
A total of 14,508 HCC patients were included for the present study. These patients were divided into two groups: training group (8,705 patients) and testing group (5,803 patients). The baseline clinical and pathological characteristics of HCC patients are presented in Table 1. The Chi-square test demonstrated that the difference in the two groups was entirely random (p > 0.05).
Table 1Demographic and pathological information of HCC patients
| Characteristics | Training group (n = 8,705) | Testing group (n = 5,803) | X2 | p |
|---|
| Age, years | | | 1.206 | 0.272 |
| ≤65 | 4,987 (57.3%) | 3,271 (56.37%) | | |
| >65 | 3,718 (42.7%) | 2,532 (43.63%) | | |
| Gender | | | 0.089 | 0.766 |
| Female | 2,035 (23.4%) | 1,369 (23.59%) | | |
| Male | 6,670 (76.6%) | 4,434 (76.41%) | | |
| Race | | | 1.834 | 0.400 |
| Black | 1,219 (14%) | 769 (13.25%) | | |
| Other | 1,535 (17.6%) | 1,017 (17.53%) | | |
| White | 5,951 (68.4%) | 4,017 (69.22%) | | |
| T | | | 0.196 | 0.987 |
| T1 | 4,292 (49.3%) | 2,867 (49.41%) | | |
| T2 | 2,040 (23.4%) | 1,372 (23.64%) | | |
| T3 | 2,092 (24%) | 1,378 (23.75%) | | |
| T4 | 281 (3.2%) | 186 (3.21%) | | |
| N | | | 1.111 | 0.292 |
| N0 | 8,116 (93.2%) | 5,384 (92.78%) | | |
| N1 | 589 (6.8%) | 419 (7.22%) | | |
| M | | | 0.000 | 0.992 |
| M0 | 7,722 (88.7%) | 5,148 (88.71%) | | |
| M1 | 983 (11.3%) | 655 (11.29%) | | |
| Tumor size, mm | | | 0.392 | 0.822 |
| ≤20 | 1,056 (12.1%) | 712 (12.27%) | | |
| >50 | 3,883 (44.6%) | 2,558 (44.08%) | | |
| 20–50 | 3,766 (43.3%) | 2,533 (43.65%) | | |
| Histological type | | | 0.020 | 0.887 |
| 8170 | 8,522 (97.9%) | 5,683 (97.93%) | | |
| Other | 183 (2.1%) | 120 (2.07%) | | |
Risk factors of distant metastasis in HCC patients
In the present study, 1,638 patients were diagnosed with DM (11.29%) and 12,870 patients (88.71%) were diagnosed without DM. The six predictors screened by the univariate logistic analysis (p < 0.05) were further analyzed using multivariate logistic analysis, and four DM-related characteristics were identified: gender, T stage, N stage and tumor size (Table 2 and Fig. 1a).
Table 2Univariate and multivariate logistic analysis of distant metastasis in HCC patients
| Characteristics | Univariate analysis
| Multivariate analysis
|
|---|
| OR | 95% CI | p | OR | 95% CI | p |
|---|
| Age, years | | | | | | |
| ≤65 | Reference | | | | | |
| >65 | 0.941 | 0.848–1.044 | 0.252 | | | |
| Gender | | | | | | |
| Female | Reference | | | Reference | | |
| Male | 1.344 | 1.180–1.530 | 0.000 | 1.232 | 1.072–1.417 | 0.003 |
| Race | | | | | | |
| Black | Reference | | | Reference | | |
| Other | 0.758 | 0.630–0.911 | 0.003 | 0.836 | 0.686–1.019 | 0.076 |
| White | 0.866 | 0.749–1.002 | 0.053 | 0.935 | 0.799–1.094 | 0.400 |
| T | | | | | | |
| T1 | Reference | | | Reference | | |
| T2 | 1.250 | 1.066–1.464 | 0.006 | 1.404 | 1.186–1.662 | 0.000 |
| T3 | 4.260 | 3.759–4.827 | 0.000 | 2.234 | 1.933–2.581 | 0.000 |
| T4 | 8.477 | 6.858–10.479 | 0.000 | 4.515 | 3.576–5.701 | 0.000 |
| N | | | | | | |
| N0 | Reference | | | Reference | | |
| N1 | 9.315 | 8.118–10.688 | 0.000 | 6.137 | 5.305–7.098 | 0.000 |
| Tumor size, mm | | | | | | |
| ≤20 | Reference | | | Reference | | |
| 20–50 | 1.426 | 1.110–1.832 | 0.006 | 1.282 | 0.994–1.655 | 0.056 |
| >50 | 4.910 | 3.874–6.223 | 0.000 | 2.652 | 2.052–3.427 | 0.000 |
| Histological type | | | | | | |
| 8170 | Reference | | | Reference | | |
| Other | 1.648 | 1.217–2.232 | 0.001 | 1.161 | 0.823–1.639 | 0.395 |
Construction and validation of the predictive nomogram
Gender, T stage, N stage and tumor size were identified as independent risk factors correlated to DM in the multivariate logistic regression analysis. Based on these independent characteristics, a nomogram for the DM risk assessment of HCC patients was constructed (Fig. 2a). Meanwhile, the ROC curve was plotted for the training and testing groups, and the AUC for the nomogram in the training and testing groups was 0.766 and 0.776, respectively (Fig. 2b, e). Moreover, we also plotted the). Furthermore, ROC curves were plotted for each independent risk factor (Fig. 2). The calibration curves revealed the excellent calibration of the nomogram in both groups (Fig. 2c, f). Furthermore, the DCA indicated that the nomogram is a robust clinical decision-making tool for DM in HCC patients (Fig. 2d, g). Moreover, the AUCs for all risk factors were lower than that for the nomogram, in both the training (Fig. 3a) and testing (Fig. 3b) groups.
Prognostic factors for HCC patients with DM
The prognostic factors for HCC with DM were investigated in the present study by evaluating 1,638 patients. Among these patients, there were more male patients (80.95%) than female patients (19.05%). Furthermore, 68.80% of these patients were white, 15.57% of these patients were black, and 15.63% of these patients were of other race. The details of all HCC patients with DM are presented in Table 3. The multivariate and univariate Cox regression results are presented in Table 4 and Figure 1b. T stage, N stage, surgery, radiation and chemotherapy were identified as independent prognostic variables for DM in patients with HCC.
Table 3Demographic and pathological information of HCC patients with DM
| Characteristics | Training group (n = 984) | Testing group (n = 654) | X2 | p |
|---|
| Age, years | | | 2.712 | 0.100 |
| ≤65 | 557 (56.6%) | 397 (60.7%) | | |
| >65 | 427 (43.4%) | 257 (39.3%) | | |
| Gender | | | 1.173 | 0.279 |
| Female | 179 (18.2%) | 133 (20.34%) | | |
| Male | 805 (81.8%) | 521 (79.66%) | | |
| Race | | | 1.090 | 0.580 |
| Black | 146 (14.8%) | 109 (16.67%) | | |
| Other | 153 (15.5%) | 103 (15.75%) | | |
| White | 685 (69.6%) | 442 (67.58%) | | |
| T | | | 0.756 | 0.860 |
| T1 | 271 (27.5%) | 174 (26.61%) | | |
| T2 | 161 (16.4%) | 100 (15.29%) | | |
| T3 | 454 (46.1%) | 310 (47.4%) | | |
| T4 | 98 (10%) | 70 (10.7%) | | |
| N | | | 0.707 | 0.401 |
| N0 | 708 (72%) | 458 (70.03%) | | |
| N1 | 276 (28%) | 196 (29.97%) | | |
| Tumor size, mm | | | 0.147 | 0.929 |
| ≤20 | 45 (4.6%) | 32 (4.89%) | | |
| 20–50 | 229 (23.3%) | 155 (23.7%) | | |
| >50 | 710 (72.2%) | 467 (71.41%) | | |
| Histological type | | | 0.415 | 0.520 |
| 8170 | 955 (97.1%) | 631 (96.48%) | | |
| Other | 29 (2.9%) | 23 (3.52%) | | |
| Surgery | | | 0.110 | 0.740 |
| No | 916 (93.1%) | 606 (92.66%) | | |
| Yes | 68 (6.9%) | 48 (7.34%) | | |
| Radiation | | | 1.517 | 0.218 |
| No | 732 (74.4%) | 504 (77.06%) | | |
| Yes | 252 (25.6%) | 150 (22.94%) | | |
| Chemotherapy | | | 0.004 | 0.947 |
| No | 450 (45.7%) | 298 (45.57%) | | |
| Yes | 534 (54.3%) | 356 (54.43%) | | |
Table 4Univariate and multivariate Cox analysis for HCC patients with DM
| Characteristics | Univariate analysis
| Multivariate analysis
|
|---|
| HR | 95% CI | p | HR | 95% CI | p |
|---|
| Age, years | | | | | | |
| ≤65 | Reference | | | | | |
| >65 | 1.037 | 0.939–1.147 | 0.473 | | | |
| Gender | | | | | | |
| Female | Reference | | | | | |
| Male | 1.034 | 0.911–1.173 | 0.606 | | | |
| Race | | | | | | |
| Black | Reference | | | | | |
| Other | 1.022 | 0.855–1.22 | 0.813 | | | |
| White | 0.989 | 0.861–1.135 | 0.871 | | | |
| T | | | | | | |
| T1 | Reference | | | Reference | | |
| T2 | 1.080 | 0.924–1.262 | 0.335 | 1.185 | 0.996–1.411 | 0.056 |
| T3 | 1.361 | 1.207–1.536 | 0.000 | 1.237 | 1.089–1.405 | 0.001 |
| T4 | 1.233 | 1.028–1.48 | 0.024 | 1.095 | 0.908–1.32 | 0.344 |
| N | | | | | | |
| N0 | Reference | | | Reference | | |
| N1 | 1.267 | 1.137–1.413 | 0.000 | 1.160 | 1.037–1.297 | 0.009 |
| Tumor size, mm | | | | | | |
| ≤20 | Reference | | | Reference | | |
| 20–50 | 1.040 | 0.806–1.342 | 0.760 | 1.035 | 0.801–1.336 | 0.794 |
| >50 | 1.324 | 1.041–1.686 | 0.022 | 1.255 | 0.967–1.63 | 0.087 |
| Histological type | | | | | | |
| 8170 | Reference | | | | | |
| Other | 0.797 | 0.601–1.056 | 0.114 | | | |
| Surgery | | | | | | |
| No | Reference | | | Reference | | |
| Yes | 0.372 | 0.301–0.459 | 0.000 | 0.327 | 0.264–0.405 | 0.000 |
| Radiation | | | | | | |
| No | Reference | | | Reference | | |
| Yes | 0.762 | 0.679–0.855 | 0.000 | 0.768 | 0.684–0.864 | 0.000 |
| Chemotherapy | | | | | | |
| No | Reference | | | Reference | | |
| Yes | 0.650 | 0.588–0.718 | 0.000 | 0.583 | 0.526–0.645 | 0.000 |
Construction of the prognostic nomogram for HCC patients with DM
According to the independent prognostic factors, the nomogram for the prognosis of HCC patients with DM was constructed (Fig. 4). Then, the ability of the nomogram, and the independent factors to discriminate the prognosis of patients in the training (Fig. 5a–c) and testing group (Fig. 5d–f), and thegroups were assessed. The AUC for the nomogram was significantly greater, when compared to that for all independent prognostic factors, at 6, 9 and 12 months. Furthermore, the 6-, 9- and 12-month OS probability calibration curves demonstrated a strong agreement between the predicted OS of the nomogram and the actual result for HCC patients with DM, both in the training and testing groups (Figs. 6a–c and 7a–c). Moreover, the DCA curves revealed that the nomogram has good predictive efficiency for OS in HCC patients with DM in the training and testing groups (Fig. 6d–f and Fig. 7d–f). The AUC for the nomogram was 0.732, 0.727 and 0.719 (Fig. 8a) in the training group and 0.697, 0.722 and 0.731 (Fig. 8c) in the testing group at 6, 9 and 12 months, respectively. According to the K-M survival curves, patients categorized as high-risk have a poorer prognosis, when compared to those classified as low-risk (Fig. 8b, d).
Discussion
HCC is a common malignant tumor that is highly aggressive and susceptible to DM. The incidence of DM from HCC is 14–37%, and the existence of DM is commonly coupled with a poor clinical prognosis.11 The intra-abdominal lymph nodes, peritoneum, lungs and bones are all common metastasis sites in HCC. Although surgical resection, locoregional, and systemic treatments can improve the OS of HCC patients, the prognosis of these patients remains poor.12,13 Therefore, the prognostic and risk factors must be identified to establish the early diagnosis, and estimate the prognosis of DM in HCC patients. Nomograms were developed and assessed to determine the occurrence possibility of DM in HCC patients, and a prognostic nomogram was constructed for HCC patients with DM. The total score may be derived from the nomogram by acquiring a number of easily accessible criteria from each HCC patient. Scores associated with the prognosis can be estimated using a prognostic nomogram. These nomograms can be facilitated for personalized clinical decisions and clinical management.
Even though the OS was low in HCC patients with DM, earlier detection and prevention can significantly improve the prognosis. Therefore, it is critical to determine the clinical and molecular risk factors for HCC with DM. The expression of FOXC1,14 LncRNA CDKN2BAS,15 and exosomal circRNA-10033816 were determined to be correlated with DM in HCC, and serum long noncoding RNAs were constructed for the early detection of metastasis in HCC.17 However, these molecules are difficult to apply in clinical decision-making and treatment. In a clinical and pathological features research, Yan et al. reported that hepatitis B virus (HBV) DNA loading, portal hypertension, Barcelona liver clinic (BCLC) stage, Child-Pugh classification, and computerized tomography (CT) imaging features are independent risk factors for metastasis in patients with HCC.18 Furthermore, for HCC patients, Hu et al. reported that gender, histological grade, T stage and N stage are independent risk factors of bone metastasis.19 In the present study, the most recent thorough clinical data were collected from a large sample obtained from the SEER database, and it was determined that the prevalence of DM was 11.29%. Furthermore, the findings of the present study revealed that gender, T stage, N stage and tumor size are independent risk factors for DM in HCC. Earlier studies have established a link between DM in HCC patients and these factors. The feature of the nomogram is that it has a larger discriminating power, when compared to any single predictor, demonstrating the necessity of an integrated prediction model.
In addition, the present study revealed that HCC patients with DM, who had a higher T stage, local lymph node metastasis, or no surgery, radiation, or chemotherapy, had the worse OS. Based on the independent prognostic factors, a prognostic nomogram was constructed, and these exhibited a higher discriminatory power, when compared to any of the independent prognostic variables. The prognostic nomogram may be effective in identifying the survival probability of HCC patients with poorer OS. Similarly, earlier studies have established a link between tumor size, stage and prognosis in patients with HCC. Liang et al. reported that a tumor diameter of more than 5 cm is an independent risk factor for tumor recurrence and long-term survival in patients with HCC.20 Furthermore, Zhang et al. reported that tumor size, but not small vessel infiltration, can be applied to predict the survival of HCC patients.21 In the present study, tumor size was a prognostic factor, rather than an independent characteristic.
The present study revealed that radiotherapy is an independent prognostic factor for HCC patients with DM. Stereotactic body radiation therapy (SBRT) is one of the treatment modalities for small HCC. Related studies have revealed that the OS rate for SBRT in small HCC is approximately 60–70%.22,23 SBRT has a similar OS rate for small HCC, when compared to radiofrequency ablation and surgery, and has a lower local recurrence rate, when compared to radiofrequency ablation.24,25 For bone or soft tissue metastases,26 lung metastases,27 and brain metastases28 in HCC, external radiotherapy can benefit clinical patients by reducing the size of the metastases and relieving symptoms.
Previous prediction nomograms for HCC patients have focused on elderly or young HCC patients.29,30 However, the sample size of these studies wass small. Furthermore, tumor progression-associated genes may play an important role in the prognosis of HCC patients.31–33 Although similar studies have been reported, the two nomograms in the present study, which were constructed on the basis of several characteristics that can be easily obtained from clinical records, can be used as convenient tools for assessing the prognosis of HCC patients in clinical practice. However, the present study had some limitations. First, the limited number of DM patients (n = 1 638) may have led to possible errors. Second, the SEER database does not provided access to complete laboratory information, which would be explored in future studies. Third, the present study was retrospective in nature, with unavoidable selection bias.