4.1 Epidemiological trends
The incidence of EC declined from 2010 to 2016 (Fig.2a). For both male and female, the incidence showed a similar trend (Fig.2b). Among the histological types, the incidence of ESCC declined, but the incidence of EAC still rose (Fig.2c). For all EC patients with BM from 2010 to 2015, prevalence was 7.11% (1503/21128). It peaked in patients between 31 and 40 and then displayed a downward trend. The prevalence of BM in metastatic esophageal cancer (mEC) was 23.5% (1503/6395). It rose with age and held steady in patients between 31 and 70 (Fig.2d).
4.2 Patient characteristics
A total of 21128 patients with EC were initially identified from2010 to 2015, and 5314 patients were included in the study. From 2010 to 2015, there were 23.74% (1782/7505) ESCC at stage IV, and 33.88% (4615/13623) EAC were advanced (Figure 3). At the time of diagnosis, there were 2566 (12.15%) patients with liver metastasis, 1562 (7.39%) patients with lung metastasis, 1241 (5.87%) patients with bone metastasis, and 314 (1.49%) patients with brain metastasis. Of the above four metastatic organs, 4105 (19.43%) patients had at least one metastasis (Figure 4). Characteristics of patients were shown in Table1. For BM, 440 (35.46%) patients were aged 61-70 years, 1075 (86.62%) patients were males, 725 (58.42%) patients had tumors located in the lower third of the esophagus, 928 (74.78%) patients were EAC, and 613 (49.40%) were poorly differentiated or undifferentiated.
4.3 Risk factors for BM
Univariable analysis (Table 1) showed that there were eight factors with statistically significance (P-value<0.05): age, sex, primary tumor location, brain metastasis, lung metastasis, liver metastasis, surgery and chemotherapy. BM was more common among 51-60 years old. Male patients were more likely to have BM. For tumor location, middle esophagus had higher percentage of BM. Patients with brain metastasis had a higher risk of having BM. However, EC patients with lung or liver metastasis were less likely to develop BM. Considering the logic, we did not put surgery and chemotherapy into multivariable logistic regression.
Multivariable logistic regression (Table 2) showed that with brain metastasis (vs without brain metastasis; OR, 1.546; 95%CI, 1.206-1.981; P=0.001) were associated with significantly greater odds of having BM in EC patients at the time of diagnosis. However, age 71-80 years (vs age <51 years; OR, 0.766; 95%CI, 0.599-0.980; P=0.034), female (vs male; OR, 0.718; 95%CI, 0.595-0.866; P=0.001), lower esophagus (vs middle esophagus; OR, 0.689; 95%CI, 0.566-0.840; P<0.001), with lung metastasis (vs without lung metastasis; OR, 0.801; 95%CI, 0.692-0.928; P=0.003), with liver metastasis (vs without liver metastasis; OR, 0.649; 95%CI, 0.569-0.742; P<0.001) were associated with lower rates of BM.
4.4 Survival and prognostic factors for BM
For 5314 patients with EC, the median OS was 5 (95%CI: 4.751-5.249) months, and the median CSS (cancer-specific survival) was 6 (95%CI: 5.710-6.290) months. For BM, the median OS was 4 (95%CI: 3.685-4.315) months, the median CSS was 4 (95%CI: 3.617-4.383) months. And for patients without BM, the median OS was 6 (95%CI: 5.674-6.326) months, the median CSS was 6 (95%CI: 5.664-6.336) months. Among the four metastatic sites (bone, brain, liver and lung), there were 2795 patients had only one metastasis. Among them, 556 patients had only BM, 114 patients had only brain metastasis, 1472 patients had only liver metastasis, and 653 patients suffered only lung metastasis. For BM, the median OS was 4 (95%CI: 3.334-4.666) months, the median CSS was 4 (95%CI: 3.260-4.740) months. And for the others, the median OS was 5 (95%CI: 4.609-5.391) months, the median CSS was 6 (95%CI: 5.539-6.461) months. For the numbers of metastatic sites, the median OS of one metastatic site was 5 (95%CI: 4.654-5.346) months, of two sites was 4 (95%CI: 3.599-4.401) months, of three sites was 3 (95%CI: 2.397-3.603) months, and of four sites was 2 (95%CI: 1.285-2.715) months. The median CSS of one site was 6 (95%CI: 5.582-6.418) months, of two sites was 4 (95%CI: 3.581-4.419) months, of three sites was 3 (95%CI: 2.279-3.721) months, of four sites was 2 (95%CI: 1.298-2.702) months. Figure 5 showed Kaplan-Meier analysis among EC patients. To reduce the influence of bias and confounding variables, we performed the Propensity score matching (PSM). After PSM, there were 2072 patients, the clinical features were as follows (Table 4). For BM, the median OS was 4 (95%CI: 3.579-4.421) months, and the median CSS was 4 (95%CI: 3.547-4.453) months. For EC patients without BM, the median OS was 6 (95%CI: 5.327-6.673) months, and the median CSS was 6 (95%CI: 5.319-6.681) months (Figure 6).
The cumulative incidence function curve (Figure 7) showed, compared to EC patients without BM, the cause-specific mortality was higher in BM (P<0.05). And the mortality rates of other causes were not statistically significant between the two groups (P=0.61).
The forest plot (Figure 8) showed that, for age (<51, 51-60, 61-70, 71-80 and >80), sex (male and female), race(white, black and others), site(upper esophagus, middle esophagus, lower esophagus and NOS), histology (EAC and ESCC), grade (1-2, 3-4 and unknown), brain metastasis (with and without), lung metastasis (with and without), liver metastasis (with and without), surgery (with and without) and chemotherapy (with and without), BM would increase the risk of death. However, for tumor location with overlapping, there was no statistical significance.
Univariable analysis showed, there were ten factors including age, race, tumor location, histology, grade, brain metastasis, lung metastasis, liver metastasis, surgery and chemotherapy that were statistically significant with OS. Then we put the above ten factors into the Cox regression model. The results confirmed that only four factors that were significant associated with OS: tumor grade, chemotherapy, liver metastasis and lung metastasis (Table 3).
On the multivariable Cox regression among the EC patients with BM, poorly differentiated and undifferentiated (vs highly differentiated and moderately differentiated; HR, 1.193; 95%CI, 1.044-1.364; P=0.010), with liver metastasis (vs without liver metastasis; HR: 1.365; 95%CI, 1.208-1.543; P<0.001), with lung metastasis (vs without lung metastasis; HR, 1.165; 95%CI, 1.018-1.333; P=0.027) showed worse OS. However, those patients who received chemotherapy (vs without chemotherapy; HR, 0.308; 95%CI, 0.271-0.350; P<0.001) had better OS.