Background

Liver metastasis, as the most commonly involved organ by colorectal cancer, has been recognized as the leading causes of death. The WHO announced more than 1.9 million new cases worldwide in 2020 [1], of which nearly half of patients develop liver metastasis during the course of the disease. Liver metastasis, with high incidence and mortality, has become the primary determinant of poor prognosis and frequent recurrence of colorectal cancer [2,3,4]. Although primary tumor and liver metastasis can be detected by preoperative thoraco-abdominal computed tomography (CT) in time, and these patients can be treated with surgical intervention, neoadjuvant chemoradiotherapy, or adjuvant chemoradiotherapy, a significant proportion of colorectal cancer cases, approximately 15–25%, would inevitably develop liver metastasis during follow-up after primary tumor resection [5,6,7]. Metachronous liver metastasis (MLM) is defined when liver involvement occurs after diagnosis/operation of primary colorectal cancer (cut-off point). Both prognosis and quality of life of patients with colorectal cancer who have undergone MLM are inferior to those with colorectal localized tumors, regardless of secondary resection, or adjuvant chemotherapy or targeted therapy. However, molecular mechanism of MLM is not yet clear, its pathogenesis can be affected by clinicopathological features, such as histological patterns, preoperative tumor markers, as well as genetics/epigenetics.

Nomograms are mainly used for risk prediction and prognostic evaluation. Currently, nomograms are widely applied in clinical studies on cancer patients. By assigning scores to various predictive factors, calculating and evaluating the probability of dependent variables, complex regression analysis can be converted into visual graphics [26, 27]. In view of the high morbidity and mortality characteristics of colorectal cancer, various predictive models focusing on the prognosis of colorectal cancer have been developed in recent years [28,29,30], including a prognostic model related to liver metastasis [31,32,33]. Time-dependent factors can effectively predict patient survival. In addition, there has been an increasing interest in exploring the risk of develo** liver metastasis. Ding et al. [34] applied the nomogram model to show us the risk factors for liver metastasis from colorectal neuroendocrine neoplasm. Mo et al. [35] analyzed the specific distant metastatic sites of stage I–IV colorectal cancer by univariate and multivariate logistic regression analysis, supporting the application of the nomogram model based on clinicopathological features to predict the metastatic sites of colorectal cancer, while confirming to us that sex, tumor site, grade, age, histological type, tumor size, T stage, N stage, and lymph node harvested were important risk factors for liver metastasis from colorectal cancer. Tang et al. [36] analyzed clinical data from the SEER database of 203,998 colorectal cancer patients to establish a nomogram to predict synchronous liver metastasis from colorectal cancer and concluded that male, black, uninsured status, left colon, T4/T1, bone, and lung metastasis were positively associated with the risk of synchronous liver metastasis.

To our knowledge, more reports have focused on the prognosis of liver metastasis from colorectal cancer and the impact of surgical treatment on the survival of patients with synchronous liver metastasis from colorectal cancer, and few studies have focused on MLM from colorectal cancer and no corresponding nomogram models have been developed. Therefore, this study focused on the risk and prognostic factors of MLM and developed and validated a nomogram model to predict the likelihood and risk factors of MLM in colorectal cancer during the high-risk time period for the development of MLM, i.e., 2 years after surgery.

By screening variables and assigning scores to those variables, nomogram visualizes data from multivariable regression analysis and individually predicts susceptibility to clinical events. In this study, LASSO regression analysis was adopted for variable selection. The LASSO regression model can not only combine selected features into radiomic features, but also check correlation between predicted factors, reduce selection bias, and optimize prediction [12, 37, 38]. Of 21 clinical factors, 7 variables were selected by LASSO regression analysis. Based on multiple logistic regression model, CEA level, vascular invasion, pT4, pN+, and KRAS mutation were independent risk factors for MLM of colorectal cancer. Combining the above two models, we established and verified a nomogram model for predicting potential risk of MLM within two years after diagnosis/operation.

As is known to all, compared with the young, the elderly is more likely to be diagnosed as malignant tumor, colorectal cancer is no exception. Studies have considered the mean age of the MLM group was younger than that of the synchronous liver metastasis group [39]. The latter is outside this study, we therefore compared the MLM cohort and non-MLM cohort of age, 18.67% of the patients in the MLM cohort are younger than 60 years old, and 21.56% were non-elderly patients in the non-MLM cohort. Age did not differ, but met the criteria for inclusion in the nomogram model, and we included it in the prediction model, with a score of 25 for the risk factor of > 60 years (Fig. 3), which is significant for predicting MLM.

CEA is mainly cleared in the liver [40], and abnormal liver function caused by tumor implantation may lead to the increase of serum CEA. Similar to previous studies, CEA is recognized as an important tumor marker for colorectal cancer. Pre-operative CEA and post-operative CEA suggest an association with systemic disease [41], increased pre-operative CEA accelerating metastasis, and spread of tumors after surgery [42]. Generally, increase in serum CEA level may be associated with liver metastasis of colorectal cancer [43,44,45]. Chuang et al. [46] retrospectively analyzed 1099 patients who underwent curative resection MLM of colorectal cancer by conducting univariate and multivariate analyses. Interestingly, preoperative serum CEA level, positive tumor depth, lymph node metastasis, and vascular invasion predicted MLM after curative resection. In addition, Mohr et al. [47] observed consistent trends. Although previous studies have suggested that postoperative serum CEA is a risk factor for liver metastasis of colorectal cancer [48], controversy remains inconclusive. In our study, patients with high or borderline levels of preoperative serum CEA are more likely to develop MLM within 2 years than those with normal levels, which is not contrary to actual clinical experience.

Genotypic differences of the primary tumor lead to differences in tumor behavior, causing MLM or synchronous liver metastasis [49]. Among the many colorectal cancer genes, RAS genetic alteration is the only recognized prognostic indicator of colorectal cancer. The KRAS mutation rate can reach 25–52% [50,51,52]. Previous studies [53] have shown that KRAS codon 13 mutation is an independent factor for metachronous distant metastasis of colorectal cancer, but there is no conclusive evidence for MLM currently, so this study focuses on the effect of RAS genes on MLM. Of the 293 patients enrolled in our cohort, 72 carried KRAS mutations (24.57%). 40% of patients with MLM within 2 years harbored KRAS mutations, which is consistent with other centres. Nras mutation occurred in 34.81%, and of the 75 patients with metachronous liver metastasis, 64% were Nras wild type, and 36% were Nras mutation, with a lower probability of Nras mutation in patients with metachronous liver metastasis compared to Kras. The absence of statistically significant Nras mutation in our study cannot be ruled out as a limitation of the limited sample size. LASSO regression screened out KRAS gene as a predictor of MLM. Multivariate logistic regression verified KRAS mutation as an independent risk factor for liver metastasis of colorectal cancer (p value < 0.001) and was included in the nomogram prediction model, with Kras positivity scoring 58 points in the model, effectively predicting metachronous liver metastasis.

Currently, tumor-node-metastasis (TNM) is a well-accepted staging system for colorectal cancer, with invasion depth and lymph node involvement closely related to liver metastasis [54,55,56]. Khan et al. [48] retrospectively analyzed the clinicopathological data of 434 patients with rectal cancer, and concluded that T staging and lymph node metastasis were related to the MLM of rectal cancer. This is consistent with the opinion of Chuang et al. [46]. A recent Italian study highlighted that lymph node ratios (ratio of positive lymph nodes to the total number of lymph nodes retrieved) can be a predictor of MLM after surgery when lymph nodes are sampled in sufficient numbers [57]. In addition, lymph nodes are considered to be independent risk factors for vascular invasion [58], and the combined action of the three factors can accelerate the progression of postoperative MLM. There are, of course, still a few opposing views that support the different subtypes of lymph node metastasis and distant metastasis, and lymph node status should not be treated as a precursor of distant metastasis [39, 56]. Due to the limitations of the study subjects and the diverse molecular subtypes of colorectal cancer patients, it is difficult to independently confirm whether T stage, N stage, and vascular invasion promote or inhibit liver metastasis. By Lasso regression analysis and logistic regression analysis, pT, pN, and vascular invasion were considered as the more important predictors in candidate nomogram model, with pT4, pN+, and positive vascular invasion being independent risk factors for MLM from colorectal cancer (p < 0.05), a view that would be supported by the majority of studies.

Imaging evaluation of liver metastasis is the mainstay to assess progression of colorectal cancer in clinical practice, especially CT and MRI, which are the most commonly used auxiliary methods for colorectal cancer patients. Thoraco-abdominal CT is mainly used to evaluate the depth of local invasion and distant staging. Although MRI can make up for the limited accuracy of CT scan and further stage distant metastasis, due to the limitations of objective factors such as cost and time cost, no matter preoperative diagnosis or postoperative review, thoraco-abdominal CT is still the most commonly used imaging examination for the diagnosis of distant metastasis of colorectal cancer [59, 60]. Therefore, liver metastasis with thoraco-abdominal CT was regarded as an outcome event in this study. If tumors have reached pT4 and involved lymph nodes when undergoing curative resection, small liver metastatic lesions cannot be ruled out. Thus, accurate assessment can help identify potential risk of MLM in patients with colorectal cancer, and specify individual follow-up plan. Simultaneously, high-risk patients can receive more effective treatment. This prediction model can be used as an auxiliary method for imaging to jointly predict MLM of colorectal cancer.

There are some limitations in the present study. First, only patients admitted to Bei**g Shijitan hospital were recruited. Second, it is difficult to include all risk factors affecting liver metastasis, so our results may be biased to some extent. In addition, patients with colorectal cancer generally receive chemotherapy after surgery. Due to individual differences in sensitivity to chemotherapy, development of liver metastasis may be affected by different drugs. However, there is currently no definite evidence that chemotherapy has an impact on our observation. Third, due to the limitation of follow-up time, we only predicted the risk factors for MLM within two years, although this is the most common time for the occurrence of MLM, it is still necessary to further study the risk factors for MLM at different times in the additional study, which will provide greater help for doctors to predict liver metastasis from colorectal cancer. Finally, although bootstrap test was used for internal validation of candidate model, external validation was not performed. Therefore, its applicability to colorectal cancer in other regions and countries remains unknown, and more extensive external verification should be carried out.

Conclusion

We have established a nomogram model for predicting potential risk of MLM from colorectal cancer, which has good discrimination and high accuracy. This model may help assess susceptibility to MLM in patients with colorectal cancer after surgery and develop individualized treatment and follow-up plans. This model predicts clinically liver metastasis, and thus provides an important reference for screening.