Background
Primary hepatic carcinoma (PHC), including hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC), is one of the commonest cancers leading to death worldwide. The clinic strategy and prognosis of patients with PHC strongly depends on the tumor stage at the time of diagnosis [
1]. Patients with advanced-stage PHC, defined according to the presence of extrahepatic metastasis, are generally considered candidates for palliative therapy instead of curative treatment, with poor prognosis (median survival of < 1 year) [
2] in contrast to the > 70% 5-year survival of early-stage PHC without extrahepatic metastasis [
3]. Extrahepatic metastasis is the independent predictor of poor survival of PHC [
4], and the most frequent metastatic sites occur in chest and abdomen (exceeding 90% of extrahepatic metastasis), [
5,
6] including lungs, lymph nodes, bones (ribs and thoracolumbar vertebraes), adrenal glands, gastrointestinal tract, and pleuroperitoneum. Therefore, warning for extrahepatic spread in the chest and abdomen is crucial to determine the optimal clinical strategy for improving the prognosis of pretreatment patients with PHC.
The current diagnosis of extrahepatic metastasis of PHC mainly relies on biomarkers, biopsy, and imaging scanning. Several biomarkers had been proposed as predictors, such as alpha-fetoprotein (AFP) mRNA, glypican-3, CK19, CD44, and vascular endothelial growth factor [
7‐
9]. However, their pragmatic value remains controversial so that they still not serve for the clinical applications. The biopsy may result in additional injury to the patient and be not suitable for repeated. At present, the most valuable diagnostic strategy are comprehensive scanning of medical imaging and regular monitoring. However, these workups are costly, complicated, time-consuming, and probably unnecessary for majority which may not benefit more than tumor treatments after essential examination. There is still no available method to predict thoracoabdominal extrahepatic metastasis in PHC. Therefore, it is necessary to conduct risk prediction for thoracoabdominal extrahepatic metastasis in PHC, in order to facilitate the individualized clinical risk management for different pretreatment population.
In recent years, some new techniques provide a powerful tool for multivariate combination analyses and strongly facilitate the progress of diagnostics. A typical example is the combination of logistic regression analysis with the least absolute shrinkage and selection operator (LASSO) [
10,
11] for data dimension reduction and variable selection, nomogram [
12,
13] for visionally scoring probability, and decision curve analysis (DCA) [
14] for the net benefit of clinical decision-making, which has provided a systematic strategy for multivariate combination analyses and timely the individualized decision-making. Huang et al. [
15] utilized the strategy to develop a nomogram capable of predicting lymph node metastasis in patients with colorectal cancer, and assessing the net benefit at different threshold probabilities.
In the present study, we aimed to develop and validate a novel nomogram for risk management of thoracoabdominal extrahepatic metastasis in PHC for the first time. We also attempted to facilitate timely individualized clinical decision-making based on pretreatment risk management for different risk population.
Discussion
In this study, we developed and validated a nomogram model based on four clinical indices for individualized risk management of thoracoabdominal extrahepatic metastasis in pretreatment PHC. To our knowledge, this study is the first to develop and validate a predictive nomogram for thoracoabdominal extrahepatic metastasis in PHC based on large-scale datasets of multicenter 437 patients. The nomogram incorporated size, PVTT, infection and CA125, and demonstrated valuable prediction performance with AUROC of 0.830 (0.803 and 0.773 in internal and independent validations, respectively). Majority can be accurately predicted with accuracy of 77.8% (75.2% and 72.9% in internal and independent validations, respectively). Good calibration was observed in each set, suggesting that no departure can perfectly fit. Using the nomogram, the risk probability can be scored easily for thoracoabdominal extrahepatic metastasis in a patient with PHC. According to weighing the net benefit of individualized clinical decision-making, the differentiated risk management can be derived.
To develop a simple but efficient predictive model, we utilized the LASSO method to data dimension reduction and screen the optimized predictors. Four significantly independent predictors (size, PVTT, infection, and CA125) were selected from 55 collectable high-dimension clinical data for modeling. The LASSO is performed for both variable selection and regularization to enhance the accuracy and interpretability of the predictive model [
10,
11]. This method surpasses the methods using the strength of univariate differences with outcome, and enables the most optimized predictors into modeling.
In the four predictors incorporated into the model, size and PVTT were proven to be independently related to extrahepatic metastasis of HCC [
23‐
25]. One possible explanation is that tumor behavior, such as tumor size expansion, portal infiltration, or metastasis, is simply various manifestation of the same tumor stem cell with aggressive HCC biology [
25]. The intrinsic association may involve angiogenesis. A retrospective study found that tumor survival, growth, and dissemination are dependent on angiogenesis along with microvessel density, which are significantly higher in patients with than those without distant metastases [
26]. The principal route of recruitment of new blood vessels is as follows, tumor cells sustain growth, exit the primary sites, and enter the circulation [
27,
28].
Recently, the relationship between infection and tumor metastases is a hotspot issue. Mantovani et al. [
29] analyzed that “smoldering” inflammation in the tumor microenvironment has many tumor-promoting functions, such as supporting the proliferation of tumor cells, inducing angiogenesis, metastasis, and overturning adaptive immunoreactions. This may indicate that both infection and metastasis are of different aspects of immune imbalance in tumor. Matsumoto et al. [
30] observed the decrease in the number and activity of natural killer (NK) cells in a murine liver metastasis model with induced abdominal infection; Kawarabayashi et al. [
31] also found that decreased NK cells increased the susceptibility of bile duct-ligated mice to infection and tumor metastasis. Recently, some studies showed that complement promotes cancer metastasis through its contribution to epithelial-to-mesenchymal transition (EMT) [
32,
33]. Mechanistically, tumor cells reduce their attachment to neighboring surroundings, increase motility, and acquire the invasive ability through EMT induced by the activation of complement receptors [
34,
35].
Interestingly, as the “classic” biomarker for ovarian cancer, CA125 was a predictor of the nomogram. CA125 had been found as a tumor marker closely related to tumor metastasis, especially in gastrointestinal malignancies. Liu et al. [
36] analyzed the serum levels of eight tumor markers, CA19-9, CEA, CA242, CA72-4, CA50, CA125, CA153, and AFP, in 1047 patients with pancreatic cancer and found that CA125 was the most strongly associated with the metastasis of pancreatic cancer and the expression of a metastasis-associated gene signature. The association of serum CA125 levels with metastasis had been observed in the liver metastasis of colorectal, [
37] breast, [
38] and lung cancer [
39]. However, there have been no reports about the relationship between CA125 concentrations and PHC metastasis, although a few studies had observed significantly elevated serum CA125 concentrations in HCC [
40] and ICC, [
41] suggesting that other mechanisms may exist and more investigations should be conducted regarding the significance and mechanism of CA125 in extrahepatic metastasis of PHC.
The association of serum AFP concentration with PHC metastasis remains controversial. Some investigators reported that high AFP concentration was the adverse factors in extrahepatic metastasis of PHC [
23‐
25]. However, Ogawa et al. [
42] found no significant difference in AFP concentrations among the three groups patients with PHC (31 extrahepatic metastasis, 46 intrahepatic metastasis, and 14 no metastasis). Additionally, no significant correlation between circulating tumor cell number and serum AFP concentration [
43]. Actually, none of these proved to be predictive, indicating that the mechanism of elevated serum AFP concentrations may differ somewhat from the distant metastasis of PHC. Serum AFP concentration, except for HCC, may be elevated during liver regeneration following hepatic resection and recovery from massive hepatic necrosis [
44,
45]. Additionally, etiologies may also affect the serum AFP concentration. Adrian et al. [
46] reported that the baseline AFP concentration was ≥ 20 ng/mL in 191 of 1145 patients (16.6%) with advanced chronic hepatitis C without HCC; simultaneously, the mean AFP values were also significantly higher in cirrhosis than in bridging fibrosis (22.5 vs. 11.4 ng/mL). Moreover, due to nearly 40% of patients with PHC are of AFP-negative (< 20 ng/mL), [
47] the overall performance of AFP is easily disturbed and may be far from satisfactory as a metastatic marker.
We presented the nomogram model, which can visually score individual risk probability of PHC metastasis in the chest/abdomen according to four clinical predictors. For majority of patients with PHC, it is reasonable immediate tumor treatments after essential evaluation, thereby avoiding unnecessary surgical exploration and longer hospital stays in the high-risk population and excessive preoperative workups in the low-risk population. But for intermediate-risk population, further evaluation may drive more benefits, including comprehensive scanning of medical imaging and regular monitoring. The final value of the nomogram is to meet personalized clinical demands for different risk population. Therefore, DCA was employed in this study. This method provides insight into the clinical decision-making by weighing the net benefit at different risk threshold probability. Based on high negative predictive ratios (85.5%) in low-risk patients (risk threshold probability < 19.9%), further evaluation may not superior to immediate tumor-curative treatment as a result of these exhaustive workups cannot drive more benefit, while further evaluation should only be considered in highly selected cases. Similarly, for high-risk patients (risk threshold probability > 71.8%) with high positive predictive ratios (82.2%), gross thoracoabdominal metastasis portend a poor prognosis, and tumor-palliative treatment should be recommended for majorities while further evaluation and tumor-curative treatment may only be considered in highly selected cases. Patients who really benefit more from further evaluation should be intermediate-risk population between low-risk and high-risk patients (risk threshold probability from 19.9% to 71.8%). Due to the low positive/negative predictive ratios (42.6/57.4%), the net benefit during the risk threshold probability range superior to the baseline models of treatment-all and treatment-none. Both clinicians and patients could perform individualized risk management with this easy-to-use scoring system, which fits the current trend toward personalized medicine.
There are some limitations in our study. Firstly, the sample size in this study is not large enough, and because of incomplete data, some clinical parameters are not included such as hepatitis C virus (HCV). Secondly, the prediction performance of the nomogram is not enough excellent (especially in independent validation with AUROC < 0.8). As the first such study, there is no similar model for reference, and the proposed nomogram may be further optimized after incorporating more valuable variables and larger samples. Finally, as a retrospective study, we cannot avoid potential biases. Therefore, the reliability and stability of the nomogram remains to be further validated by prospective cases. In the next step, we will focus on conducting a prospective multi-center research for enrolling the large sample cases. In the prospective research, we will further evaluate the predictive value of variables that were valuable in reported studies but not into our model (including number of nodules, jaundice, hepatitis B surface antigen, HCV, Child–Pugh stage and so on) [
23‐
25,
48], and strive to improve our model by optimizing predictors.
Authors’ contributions
JH and KHZ conducted the study design, participated in data interpretation, and prepared the manuscript. YPJ and SX participated in coordination and acquisition of data. TW, SHC, YTH, HLY, YQW participated in data analysis and acquisition of data. All authors read and approved the final manuscript.