Prognostic Nomograms based on Homogeneous and Heterogeneous Associated Factors for Predicting the Overall Survival of Colorectal Cancer Patients with Distant Metastases

Background: To identify the heterogeneous and homogeneous prognostic factors associated with distant metastases in colorectal cancer (CRC) patients and then construct nomograms to predict prognosis. Methods: CRC patients registered in the Surveillance, Epidemiology, and End Results (SEER) database between 2010 and 2016 were included. Cox regression was used to analyse homogeneous and heterogeneous prognostic factors, and Kaplan-Meier analysis was used to estimate overall survival (OS). Predictive nomograms were constructed, and their performance was evaluated with C-indexes and calibration curves. Results: A total of 34933 patients with distant metastases were included. The median survival time of patients with liver metastases, lung metastases, bone metastases, and brain metastases were 12.00 months (95% CI: 11.71-12.29 months), 10.00 months (95% CI: 9.57-10.43 months), 5.00 months (95% CI: 4.47-5.53 months), and 3.00 months (95% CI: 2.31-3.70 months), respectively. Older age and no surgery were identied as homogeneous prognostic factors of the four types of metastases. Male sex, black race, unmarried status, uninsured status, primary CRC site, poor differentiation/grade, advanced N stage, T stage, high carcinoembryonic antigen (CEA) level and metastatic organ were heterogeneously associated with the prognosis of patients with distant metastases. The calibration curves and C-indexes exhibited good performance for predicting the OS of patients with distant metastases. Conclusion: CRC patients with distant metastases exhibited homogeneous and heterogeneous prognostic factors, all of which were associated with poor survival. The nomograms showed good accuracy and can be used as tools for clinicians to predict the prognosis of CRC patients with distant metastases.


Background
Colorectal cancer (CRC) is the third most commonly diagnosed cancer and the second leading cause of cancer-related death worldwide [1]. Approximately 20% of patients with CRC have distant metastases at initial diagnosis [2], and these patients have a lower 1-year and 5-year cancer-speci c survival rate than patients without distant metastases [3][4][5]. However, due to the heterogeneity of metastatic patterns, CRC patients with speci c organ metastases experience different prognoses; for instance, CRC patients with brain metastases have worse 5-year survival than those with liver metastases (0.4% vs. 16.9%) [6,7]. The early detection of prognostic factors for CRC with distant metastases may help predict the probability of survival. However, the survival of CRC patients with distant metastases and the prognostic factors for their speci c organ metastases are still unclear.
To our knowledge, some studies have investigated the prognostic factors of distant metastases in CRC, but majority of them have focused on patients who have already received treatments [8][9][10]. In addition, only a few studies described models to predict the survival of CRC patients with distant metastases [11,12], and the results remain controversial due to the limited sample size. Moreover, homogeneous and heterogeneous prognostic risk factors for CRC patients with speci c organ metastases have not yet been explored.
Therefore, this study aimed to characterize the differences in prognostic factors based on a large population cohort of CRC patients with distant metastases. Furthermore, a nomogram was developed to predict the prognosis of CRC patients with speci c organ metastases and provide a reference tool for clinicians to guide individualized treatment for CRC patients.

Population
In the present study, CRC patient data were acquired from the US National Cancer Institute (

Statistical analysis
Quantitative data are presented as the mean ± standard deviation (SD), and categorical data are described as the number and percentage (N, %). Univariable and multivariable Cox regression analyses were conducted to identify potentially associated prognostic factors. Based on the results of the Cox regression analysis, the intersection of the prognostic factors for the four types of metastases was used to identify homogenous or heterogeneous factors. The Kaplan-Meier method was used to estimate overall survival (OS). Based on the results of the multivariable Cox regression analysis, predictive nomograms for liver, lung, bone, and brain metastases were formulated. C-indexes and calibration curves were used to evaluate their distinguishing and predictive abilities. Statistically signi cant levels were twotailed and set at p < 0.05. Statistical analyses were conducted with the IBM Statistical Package for the Social Sciences (SPSS) version 23.0 software package for Windows (SPSS, Inc., Chicago, IL, USA). The nomogram was plotted using the "rms", "Hmisc", and "survival" packages in R version 3.4.1 (R Foundation for Statistical Computing, Vienna, Austria; www.r-project.org).

Demographic and clinical characteristics
A total of 34933 patients with distant metastases and 31288 patients with liver metastases, 10598 patients with lung metastases, 2553 patients with bone metastases, and 587 patients with brain metastases were included in this study. The detailed work ow for patient selection is shown in Fig. 1. The mean age of all 34933 CRC patients with distant metastases was 64.19 ± 14.13 years (range 14 to 108), 55.2% of patients (N = 19273) were male, and 44.8% of patients (N = 15660) were female. Most of the patients were white (75.1%, N = 26234), followed by black (15.5%, N = 5400). Forty-eight percent (N = 16944) of patients were married, and 77.7% (N = 27152) were insured. Regarding the primary tumour sites, most were localized in the left colon (35.70%, N = 12470) and right colon (35.17%, N = 12285), followed by the rectum (20.31%, N = 7098). A total of 30.38% (N = 10612) of patients were initially diagnosed with a stage T3 tumour, and 47.23% (N = 16498) were diagnosed with a grade II tumour. Detailed demographic and clinical characteristics are shown in Table 1.

Survival Outcomes In Patients With Distant Metastases
The 1-, 3-, and 5-year OS rates were 49.5%, 18.9%, and 9.5% for all patients, respectively. The mean survival time was 20.94 months. The 1-, 3-, and 5-year OS rates rapidly decreased with increasing age. However, the 1-, 3-, and 5-year OS rates in patients who underwent surgery at the primary site were much higher than those in patients who did not (p < 0.001). The detailed data of the patients' 1-, 3-, and 5-year OS rates are listed in Table 1, and the median survival times are listed in Table 2. Abbreviations: CEA = carcinoembryonic antigen, NA = not available, Surg(pri) = surgical treatments of primary site.
a Includes American Indian/Alaska Native and Asian or Paci c Islander.
b Includes single, separated, widowed, and divorced.

Prognostic Factors In Crc Patients With Distant Metastases
Age, sex, race, marital status, insurance status, primary tumour site, histological grade, lymphatic metastasis, T stage, carcinoembryonic antigen (CEA) level and surgical treatment at the primary site were correlated with the prognosis of patients with distant metastases in the univariate analysis. The results of the multivariable model indicated that older age, male sex, black race, unmarried status, uninsured status, poor histological grade, stage T4/T1, high CEA level and no surgical treatment all showed positive associations with the prognosis of patients with metastases (see Table 2). The associated prognostic factors and their HRs with 95% CIs according to the different sites of metastases are presented in Tables S5-S8.
The prognostic factors associated with CRC with metastases at different organs were homogeneous and heterogeneous. Both older age and no surgery at the primary site showed positive associations with the prognosis of CRC patients with distant metastases. Male sex, black race, unmarried status, uninsured status, poor histological grade, lymphatic metastasis, stage T4/T1, high CEA level, and metastatic organ showed positive associations with the prognosis of CRC patients with liver metastases. Unmarried status, uninsured status, poor histological grade, lymphatic metastasis, high CEA level, and metastatic organ showed positive associations with the prognosis of CRC patients with lung metastases. Unmarried status, poor histological grade, lymphatic metastasis, and metastatic organ showed positive associations with the prognosis of CRC patients with bone metastases, and high CEA level showed a positive association with the OS of CRC patients with brain metastases (see Fig. 3).

Discussion
Recently, majority of studies on the prognosis of CRC patients with distant metastases have shown different results. The results of our study were not entirely consistent with those of some previous studies [5,[13][14][15][16][17][18]12]. The 5-year survival rates for CRC patients with liver metastases and lung metastases were 16.9% [7] and 8.0% [12], respectively, which were higher than those found in our study. Compared to CRC patients with liver or lung metastases, the prognosis of CRC patients with brain or bone metastases was worse, with a 5-year survival rate of < 5%. Previous studies reported that the median survival time for CRC patients with bone metastases ranged from 5.0 to 7 months [16, 13,19], consistent with our results.
However, we found that the median survival time for patients with brain metastases was only three months, which was worse than that reported (from 4.2 to 7 months) [14,16,20,21]. Overall, the OS rates for patients with distant metastases were poor, and the prognosis of CRC patients with brain metastases was the worst, followed by those with bone metastases, lung metastases and liver metastases. This nding is similar to that of a previous study [3].
The present study showed that different sites of metastasis exhibited homogeneity and heterogeneity regarding the prognostic factors associated with distant metastases from CRC. To our knowledge, a previous study has not reported these homogeneous prognostic factors (older age, no surgery at the primary site) for the prognosis of CRC patients with distant metastases. This nding will help clinicians tailor targeted preventive treatment strategies, thereby further improving patient survival. Previous studies have mostly focused on the prognosis of CRC patients, and whether age is an independent factor remains controversial. Some studies reported that young patients with CRC had a worse prognosis than older patients [22][23][24]. Some studies reported the opposite results [25,26], and another study demonstrated that age was not associated with the prognosis of CRC [27]. This controversy may be due to the limited sample sizes. Our study, based on a large population cohort, showed that older age was a prognostic factor for the four types of metastases. Thus, compared with older patients, younger patients have a better prognosis, consistent with previous reports [28,29]. This nding is presumably due to their better physiological reserve and lower incidence of comorbidities. The present study also demonstrated that surgery at the primary site showed a positive association with the four types of metastases (liver, lung, bone, and brain). Many studies on different samples all showed that surgical resection may improve survival. Therefore, when CRC patients are initially diagnosed with distant metastases, surgery at both the primary site and distant lesions may be an appropriate and effective treatment. However, this potential treatment approach still needs to be con rmed through more research.
In addition to these homogeneous factors, some other heterogeneous associated prognostic factors were also found. However, these heterogeneous factors are not completely consistent with those described in previous studies. For instance, both unmarried status and uninsured status showed positive associations with the prognosis of CRC patients with liver metastases, lung metastases and bone metastases. However, these factors have seldom been reported in other studies [12,30]. One possible reason for the poor prognosis of unmarried or uninsured patients is the lack of mental or nancial support from their family members. In addition, the primary site of CRC, a poor differentiation grade, lymphatic metastasis and the metastatic organ were found to be signi cantly associated with the development of the four types of distant metastases [31]. However, in the present study, these four factors were identi ed only as homogeneous factors of the prognosis of patients with three types of distant metastases (all except brain metastases). A previous study reported that the tumour site and extracranial metastases were signi cantly associated with the poor survival of CRC patients with brain metastases [14]. Whether these factors are associated with the prognosis of CRC patients with brain metastases is still controversial, and more studies are needed to con rm this hypothesis.
Additionally, CEA is a widely used biomarker for CRC, and routine monitoring of CEA levels is commonly used to monitor postoperative recurrence and distant metastases [32][33][34]. A previous study reported that CEA can also be used to predict the prognosis of CRC patients with distant metastases [12,35,36].
However, this study showed that CRC patients with bone metastases did not tend to have a signi cantly poor prognosis when they had a high CEA level. For patients with bone metastases, whether CEA levels affect survival remains to be studied.
Although some studies have reported prognostic factors for metastatic CRC, the homogeneity and heterogeneity of the prognostic factors of the four types of metastases from CRC have not been comprehensively studied. The homogeneous and heterogeneous associated prognostic factors mentioned in our study may help clinicians guide individualized treatment for patients with metastatic CRC. Based on these prognostic factors, four predictive nomograms were constructed, and all were validated and showed good prediction performance. Similar to the model used to predict distant metastases previously constructed by our team [31], the prognostic and predictive model described herein is also a rapid and cost-effective tool. After screening using our model, patients with metastatic CRC could receive timely targeted therapy to improve their prognosis and avoid other unnecessary expenses.

Consent for publication
Not applicable.

Availability of data and materials
The data that support the ndings of this study are available from the corresponding author upon reasonable request.

Figure 1
Flowchart of the colorectal cancer patient selection.

Figure 2
Kaplan-Meier analysis of overall survival in colorectal cancer patients with different distant metastases.  Nomogram for predicting the 1-, 3-, and 5-year overall survival rates of colorectal cancer patients with liver metastases (A), lung metastases (B), bone metastases (C), and brain metastases (D), respectively.