Epidemiology and Prognostic Nomogram for Esophageal Cancer With Liver Metastasis Based on the Surveillance, Epidemiology, and End Results Database


 Background: The prognosis of esophageal cancer with liver metastasis (ECLM) is poor. This study evaluated the incidence and prognostic factors of ECLM at initial diagnosis and developed a visual nomogram to predict outcomes. Methods: 1892 eligible ECLM patients diagnosed between 2010 and 2016 were extracted from the Surveillance, Epidemiology, and End Results database (SEER). The incidence, patient overall survival (OS), and prognosis of liver metastasis were evaluated according to patient information, tumor characteristics, and therapy. Cox regression analyses were applied to evaluate prognostic factors. Based on the significant demographic and clinicopathologic factors, a novel visual nomogram was established to predict 6-month and 1-year survival for ECLM patients. Results: The incidence of ECLM increased with age, but survival rates showed the opposite trend. From 2010 to 2016, as age increased, the incidence also increased annually, peaking at age 70 to 79 years and then decreasing. The 1-, 3-, 6-, 9- and 12-month relative cancer survival rates for ECLM patients were 84.6%, 61.3%, 44.0%, 31.4%, and 21.8%, respectively. Multivariate cox regression analysis revealed that age, gender, grade, histological type, bone metastasis, lung metastasis, brain metastasis, chemotherapy and local treatment were independent prognostic factors related to OS. Moreover, our nomogram predicted the 6-month and 1-year survival times of the population, with a C-index of 0.735 (95% CI: 0.723-0.748). The nomogram demonstrated good clinical applicability. Conclusions: The incidence of ECLM increased with age, but survival rates showed the opposite trend. Age, gender, tumor characteristics, and therapy should be considered when predicting the prognosis of ECLM patients. The nomogram may improve clinicians’ abilities to predict individual survival and recommend treatments for ECLM patients.


Introduction
Esophageal cancer (EC) is a malignant tumor that represents the sixth leading cause of cancer-related deaths worldwide. There were 473,000 (95% uncertainty interval [95% UI] 459,000-485,000) new cases of EC and 436,000 (425,000-448,000) deaths due to EC in 2017 [1] . The incidence and mortality of EC vary geographically, with a high burden in East Asia and Eastern/Southern Africa, where esophageal squamous cell carcinoma predominates over adenocarcinoma [2] . However, esophageal adenocarcinoma (EAC) has become the most common esophageal malignancy in Western countries [2,3] . The most common sites of distant metastasis (DM) at initial diagnosis are the non-regional lymph nodes, liver, lung, bone, and adrenal gland [4] . The prognosis of metastatic EC is poor, and the ve-year survival rate is less than 5% [5] . The liver is one of the most common sites of metastases, occurring in 35-40% of patients at the time of EC diagnosis [6] . Chemotherapy is the standard treatment for these patients, however, survival is not satisfactory [7,8] . Small trials have reported that patients with a low hepatic tumor burden have a favorable prognosis after liver resection or another local therapy [9,10] .The optimal management of patients with liver metastases is still a question of debate.

Methods
This retrospective study was based on data obtained from the SEER database 18 registry, which is maintained by the National Cancer Institute of the United States. We extracted raw data through online access using SEER*Stat software version 8.3.5 [12] . Analysis of the raw data from the SEER program was exempt from medical ethics review, and no informed consent was required. All procedures performed in studies involving human participants were in accordance with the 1964 Helsinki Declaration.

Study population selection
We limited the data to the following selection criteria: EC patients of all ages diagnosed with liver metastasis between 2010 and 2016. The patients with ECLM were selected for further research. Patients diagnosed with other tumors were exempt from the selected population. EC patients with brain, bone and lung metastases were also selected in the same manner for comparison to those with liver metastases.
Patients who were diagnosed via autopsy or a death certi cate or whose detailed information was unknown or blank were excluded.

Data elements
The detailed demographic and clinicopathological features for each patient are shown in Table 1.
Demographic variables included age at diagnosis, gender, race, vital status and survival (in months). Cancer characteristics included the primary tumor location, grade, T stage, N stage, histologic type, and sites of DM. Treatment information, including chemotherapy and local treatment (surgery and/or radiation), was also collected from the database. Local treatment was de ned as surgery at the primary lesion and/or radiotherapy. Histological types were based on the International Classi cation of Diseases

Statistical analysis
Patients were divided into three groups based on age by using the X-tile program (Yale University, New Haven, CT, USA) [13] . The optimal cutoff values were 57 and 77 years. For convenience, we divided the patients into 3 age groups (0-57, 58-77, and 78+ years) for analysis. Kaplan-Meier analysis and log-rank tests were used to compare OS.
We examined the incidence for patients diagnosed between 2010 and 2016 through SEER*Stat statistical software for Windows. We also compared the incidence according to the sites of DM (liver, brain, bone, and lung). The 1-, 3-, 6-, 9-and 12-month relative cancer survival rates for ECLM patients were also examined according to the sites of DM. Detailed differences between different age subgroups were also explored. Moreover, survival curves were described by using Kaplan-Meier methods after adjusting for age, gender, race, histological type, primary site, tumor stage, lymph node stage, sites of DM, chemotherapy and local treatment. Differences between the OS curves were compared using the log-rank test. In all statistical analyses, a p value <0.05 was considered signi cant.
Prognostic nomogram for OS Cox proportional hazards regression was used to analyze the associations between OS and relevant prognostic factors, including age, gender, race, primary tumor site, histological type, grade, tumor stage, lymph node stage, bone metastasis, lung metastasis, brain metastasis, local treatment and chemotherapy, along with treatment information, based on the SEER database. A visual nomogram based on possible prognostic factors associated with OS was established using R software on the basis of the Cox regression results. The discriminative ability of the prognostic model was evaluated with Harrell's concordance index (C-index). The C-index estimates the probability of concordance between predicted OS and observed OS. Calibration was performed by comparing predicted OS with observed OS against the nomogram's 6-month and 1-year predicted OS outcomes. Decision curve analysis (DCA) was also employed to evaluate the clinical usefulness of the model. All statistical analyses were completed with SPSS software (version 23; IBM Corp., USA) and R software (version 3.3.0; http://www.r-project.org/).

Results
Characteristics of the study population 4080 patients were diagnosed with ECLM between 2010 and 2016 in the United States and included in our initial analysis. After excluding patients with missing follow-up or unknown data, a total of 1892 patients were included in the present study. The demographic data of the patients and the characteristics of the tumors are shown in Table 1

Incidence
The incidence of ECLM increased with age ( Figure 1A), but survival rates showed the opposite trend ( Figure 1D). It is notable that the survival rate of adults continued to decrease with increasing age. We then analyzed the annual incidence of patients in different age groups from 2010 to 2016. As age increased, the incidence increased annually, peaking at age 70 to 79 years and then decreasing. We then compared the incidence of four common metastatic sites of EC (bone, brain, liver, and lung) from 2010 to 2016 and found that the most common site of metastasis was the liver ( Figure 1B), followed by the lung, bone and brain.

Survival analysis
The OS of patients with EC with/without metastasis is shown in Figure 1C. All patients with metastases at the four most common sites had a worse prognosis than those without metastasis, EC patients with bone metastasis having the worst prognosis. Liver metastases were associated with better OS than bone metastases. The survival rate of patients with ECLM is shown in Figure 1C. The 1-, 3-, 6-, 9-and 12-month relative cancer survival rates for ECLM patients were 84.6%, 61.3%, 44.0%, 31.4%, and 21.8%, respectively. Kaplan-Meier survival curves also con rmed the effect of prognostic factors, including age, gender, race, histological type, primary site, tumor stage, lymph node stage, grade, site of metastasis and treatment information. The log-rank tests for all Kaplan-Meier survival curves except for those on gender (p>0.05) and race (p>0.05) were statistically signi cant (p<0.05), but the curves were crossed (Figure 2). We also found that the median survival time was similar and not optimistic.

Prognostic factors of OS
The prognostic factors of OS were calculated with univariate and multivariate Cox proportional hazards regression analyses ( and local treatment (p=0.002) were independent prognostic factors for OS, while gender (p=0.078) and race (p=0.126) were not ( Table 2). The indicators with p value<0.1 were included in the multivariate cox proportional hazards regression analysis. The results revealed that age, gender, grade, histological type, bone metastasis, lung metastasis, brain metastasis, chemotherapy and local treatment were independent prognostic factors for OS. The corresponding survival curves are shown in Figure 2. Nomogram A visual nomogram was constructed to predict 6-month and 1-year OS based on the multivariate cox proportional hazard regression analysis. Each variable was given a score on a points scale (Figure 3). The sum of these numbers was located on the total points axis. By summing the total scores projected on the bottom scale, we estimated the probability of 6-month and 1-year OS. In our newly visualized nomogram, based on internal validation, the corrected C-index for predicted 6-month OS and 1-year OS was 0.735 (95% CI: 0.723-0.748). The calibration plot for the probability of 6-month and 1-year OS showed good correlation between the nomogram prediction and actual observation (Figure 4). DCA was employed to further verify the clinical validity of the model, and it demonstrated good clinical applicability of the nomogram ( Figure 5).

Discussion
EC patients with metastatic diseases differ from patients without metastasis in epidemiology, biology, and prognosis. Approximately 50% of patients present with metastases to distant lymph nodes or organs at initial diagnosis [4] . The prognosis of metastatic EC is poor, and the ve-year survival rate is less than 5% [7,9] . We conducted an updated population-based study to investigate the epidemiological characteristics for ECLM. Moreover, a visual nomogram was established to predict individualized survival and to make treatment recommendations for patients with ECLM.
Previous epidemiological studies on ECLM have shown that the prognosis is poor [9,11] . Haiqi He et al reported that the age-adjusted incidence of EC increased from 5.55 to 7.44 per 100,000 person-years between 1973 and 2004. Later, the rate decreased at an annual percentage change of 1.23% [2] .
Studies have shown that the liver is one of the most common sites of metastases, occurring in 35-40% of patients at the time of EC diagnosis [6,14] , and it is the rst site of recurrence in 6-25% of patients after esophagectomy [15] . Similar retrospective studies of patients with DM have reported that the most common sites of DM are the liver, followed by the distant lymph nodes, lung, bone, and brain [15,16] . In the present study, the incidence of ECLM increased with age ( Figure 1A), but survival rates showed the opposite trend ( Figure 1D). We analyzed the annual incidence of patients in different age groups from 2010 to 2016. As age increased, the incidence increased annually, peaking at age 70 to 79 years and then decreasing. The highest incidence of metastasis was in the liver, followed by the lung, bone and brain. All patients with metastases at the four most common sites had a poorer prognosis than those without metastasis, and patients with bone metastasis of EC had the worst prognosis. Moreover, liver metastases were associated with better OS than bone metastases.
There are limited studies on the effect of the site of DM on survival in patients with metastatic EC. Chen et al found that DM was not associated with OS in patients with metastatic squamous cell esophageal cancer [16] . Tanaka et al also observed no signi cant difference in the median survival time based on the site of DM (liver, bone, and lung; p=0.8786) [7] . Blank et al found that the localization and number of metastases were not signi cant prognostic factors for survival [17] . However, they did not further analyze the effect of site-speci c DM on survival. San-gang Wu et al reported that bone metastases were associated with poorer OS than liver metastases, especially in patients with esophageal AC [4] . They also observed similar OS rates for esophageal SCC patients with liver, bone, or lung metastases. As shown in Figure 1C, all patients with metastases at the four most common sites had a worse prognosis than those without metastasis, with EC patients with bone metastasis having the worst outcome ( Figure 1C). The four survival curves demonstrated similar OS rates for EC patients with liver, bone, brain or lung metastasis. We further screened that liver metastases, bone metastasis, lung metastasis, and brain metastasis were independent prognostic factors for OS. Therefore, the classi cation of patients with metastatic EC by site-speci c metastasis may help physicians more accurately assess prognosis and generate treatment strategies.
We chose a series of signi cant factors to generate a nomogram. In our study, we found that gender was an independent predictive factor for liver metastasis in patients with EC. The incidence of liver metastasis was 86.2% (1631/1892) in male patients and 13.8% (261/1892) in female patients. A previous study demonstrated that the in uence of sex hormones may explain the sex difference in EC incidence [18] . Our results also indicate that EAC is the most prevalent type of EC in the U.S. EAC patients accounted for 80% of the patient population and had a worse prognosis than squamous cell cancer patients. A related study reported that the incidence of EAC has increased since 1973 and surpassed that of squamous cell cancer [2] . The sharp increase in EAC incidence in the U.S. can be partly explained by the increasing numbers of overweight individuals or individuals with gastroesophageal re ux disease (GERD), which is associated with obesity [19] .
Previous studies have demonstrated that T stage and N stage are predictors for the DM of EC. Zhang et al showed that T stage (T3-4 vs. T1-2) and N stage (N1 vs. N0) were important prognostic factors for metastasis-free survival [20] . Sakanaka et al. revealed that metastatic lymph nodes were related to the occurrence of DM, and patients with larger metastatic lymph nodes had a higher risk of DM [21] . In our univariate analysis, T stage (p=0.004) and N stage (p≤0.001) were signi cant prognostic factors for ECLM. In contrast to a previous study, the multivariate logistic regression analysis showed that T stage (p=0.316) and N stage (p=0.403) were not signi cant indicators for survival.
Chemotherapy is the standard treatment for metastatic EC. With the current chemotherapy regimen, the duration of the response is short, and survival is usually 6 to 8 months [22] . Compared to the survival rate of patients who receive best supportive care, the survival rate of patients who receive chemotherapy alone is signi cantly higher (p=0.061) [16] . Chemotherapy increases the number of patients who are able to undergo resection or radiotherapy [23] . However, there is no consensus on whether palliative radiotherapy or surgery is of value in metastatic EC. Several studies have suggested that palliative radiotherapy and surgery can improve survival in patients with metastatic EC [9,11,24] . Moreover, several retrospective studies have found that primary tumor resection may be considered a strategy for selected patients with stage IV EC who achieve a favorable response to systemic chemotherapy [17,25] . Another study of the SEER database also found that surgery and preoperative radiotherapy were associated with improved survival in patients with metastatic EC [9] . In the present study, most patients received chemotherapy, 39 (1.60%) patients received primary surgery, and 632 (33.40%) patients received radiotherapy. The survival analyses suggested that local treatment and chemotherapy improve survival in ECLM patients. Local treatment (p=0.004) and chemotherapy (p≤0.001) were identi ed as independent favorable prognostic factors for OS, similar to results in metastatic breast cancer and colorectal cancer [26,27] . However, the number of patients who were suitable for local treatment of the primary tumor was not enough to reach de nite conclusions. Therefore, further studies are required to explore the subgroups of patients who may bene t from local therapy.
The prognostic study of EC will be useful for clinical practice. Therefore, there are a number of nomograms for metastatic EC.  [29] , and the OS nomogram had a C-index of 0.633. However, previous studies did not focus on less common sites of metastasis. Therefore, we conducted the present study to focus on liver metastasis. For this reason, we included a scope of signi cant variables to achieve the widest applicability. The prediction model presented good discrimination, with a C-index of 0.735 (95% CI: 0.723-0.748), and the calibration plot showed satisfactory calibration. The nomogram also demonstrated good clinical applicability.
Our study also has some shortcomings, such as the lack of some key details, especially the use of chemotherapy or radiotherapy, and surgery details are not described in the SEER database. Second, because of a lack of data support from another database, we did not perform an external validation.
Therefore, the presented nomogram needs to be replicated and then prospectively validated before it can be implemented in clinical practice.

Conclusions
This study provides new insights into the epidemiology of patients with ECLM. Age, gender, tumor characteristics, and therapy should be considered when predicting the prognosis of patients with ECLM. The nomogram described herein may improve clinicians' abilities to predict individualized survival and make treatment recommendations for patients with ECLM. Availability of data and materials The datasets used and/or analyzed during the current study are available from the corresponding author.  Calibration plot showing nomogram-predicted 6-month and 1-year OS ECLM probabilities.

Figure 5
Decision curve analysis (DCA) for the nomogram and conventional forecasting indicators, including age, gender, histological type, grade, bone metastasis, lung metastasis, brain metastasis, chemotherapy and local treatment.