A nomogram was developed using clinicopathological features to predict postoperative liver metastasis in patients with colorectal cancer

The objective of this study is to examine the risk factors that contribute to the development of liver metastasis (LM) in patients who have suffered radical resection for colorectal cancer (CRC), and to establish a nomogram model that can be used to predict the occurrence of the LM. The present study enrolled 1377 patients diagnosed with CRC between January 2010 and July 2021. The datasets were allocated to training (n = 965) and validation (n = 412) sets in a randomly stratified manner. The study utilized univariate and multivariate logistic regression analyses to establish a nomogram for predicting LM in patients with CRC. Multivariate analysis revealed that T stage, N stage, number of harvested lymph nodes (LNH), mismatch repair (MMR) status, neutrophil count, monocyte count, postoperative carcinoembryonic antigen (CEA) levels, postoperative cancer antigen 125 (CA125) levels, and postoperative carbohydrate antigen 19–9 (CA19-9) levels were independent predictive factors for LM after radical resection. These factors were then utilized to construct a comprehensive nomogram for predicting LM. The nomogram demonstrated great discrimination, with an area under the curve (AUC) of 0.782 for the training set and 0.768 for the validation set. Additionally, the nomogram exhibited excellent calibration and significant clinical benefit as confirmed by the calibration curves and the decision curve analysis, respectively. This nomogram has the potential to support clinicians in identifying high-risk patients who may develop LM post-surgery. Clinicians can devise personalized treatment and follow-up plans, ultimately leading to an improved prognosis for patients.


Introduction
Colorectal cancer (CRC) is the third leading cancer type and the second most common cause of cancer-related deaths worldwide (Sung et al. 2021).According to estimates, there will be 52,550 CRC deaths in 2023, with 7% (3750 individuals) of the decedents being younger than 50 years of age (Siegel et al. 2023).Due to its anatomical position relative to the portal circulation, colorectal cancer cells most often metastasize to the liver (Wang et al. 2021).Despite successful surgical resection of the primary tumor, approximately 15-25% of CRC cases occur liver metastasis (LM) during follow-up, leading to a reduction in five-year survival rates from 80 to 90% for localized disease to 10-20% for patients with LM (Giannis et al. 2020;Chandra et al. 2021;Zeng et al. 2021;Liu et al. 2021).Liver metastasis, with its high incidence and mortality, is the main cause of poor prognosis in colorectal cancer patients (Wu et al. 2020;Siegel et al. 2020).
The early identification of high-risk LM patients is critical for improving clinical outcomes.To date, clinical Xinyu Dou, Jiaona Xi, and Gaozan Zheng contributed equally to this work.parameters, such as age, preoperative levels of carcinoembryonic antigen (CEA) and carbohydrate antigen 19-9 (CA19-9), genetic mutations, lymphatic vascular invasion (LVI), and neural invasion, have been identified as potential biomarkers for identifying patients with an elevated risk of developing LM (Y.Li et al. 2022;Hao et al. 2022).Additionally, several predictive models have been developed based on independent prognostic factors for LM (Hao et al. 2022;Mo et al. 2020).However, prior studies have been limited by shortcomings, such as small sample sizes and incomplete analysis of predictive factors, resulting in low predictive value.
Although the use of preoperative tumor markers in predicting prognosis for CRC has been widely studied, predictive power using postoperative tumor markers has received less attention.However, recent studies have demonstrated that postoperative tumor markers have a stronger prognostic value than preoperative tumor markers (Konishi et al. 2018;S. Zhou et al. 2021;C. Li et al. 2023).Unfortunately, only a limited number of studies have thoroughly analyzed postoperative tumor markers with regard to LM of CRC following surgery, resulting in a lack of consensus in this area (Khan et al. 2019).
Furthermore, previous studies indicated that the survival and distant metastasis of circulating tumor cells (CTCs) in postoperative tumor patients are closely related to the levels of immune cells in peripheral blood (Hamilton et al. 2016;Hamilton and Rath 2017;Spicer et al. 2012;Osmulski et al. 2021).Hamilton et al. reported that CTCs are capable of differentiating monocytes into tumor-associated macrophages (TAMs) (Hamilton et al. 2016;Hamilton and Rath 2017).Osmulski et al. suggested that TAMs might enhance CTCs' mechanical adhesiveness and endurance, facilitating the formation of protective clusters of cells and conferring resistance to shear stress, thereby promoting distant metastasis (Osmulski et al. 2021).Additionally, the fusion of macrophages and tumor cells is believed to increase the probability of cell survival and induce metastasis.Furthermore, Spicer et al. demonstrated that neutrophils can directly cling to CTCs via the Mac-1/ICAM-1 interaction and serve as a bridge between liver parenchyma and tumor cells, thus facilitating extravasation and liver metastasis (Spicer et al. 2012).These findings suggest that peripheral blood immune cells may play a crucial role in predicting the occurrence of postoperative LM in patients with colorectal cancer.
Given this situation, we comprehensively analyzed the relationship between clinicopathological features, including peripheral blood immune cells and postoperative tumor markers, and postoperative LM of colorectal cancer patients.In addition, we attempted to build a prediction model upon these clinicopathological features, to predict postoperative LM of colorectal cancer more precisely.

Patients
The study retrospectively recruited patients who underwent curative resection for CRC in our department from January 2010 to July 2021.Participants fulfilled the following inclusion criteria: (1) the presence of a confirmed pathological diagnosis of colorectal cancer; (2) underwent radical resection; (3) no current or previous background of other malignant tumors; (4) all non-LM patients were monitored for a minimum of 3 years; (5) fully complete clinical data available.Exclusion criteria consisted of: (1) synchronous distant metastasis at the time of initial diagnosis; (2) liver metastasis developing within 3 months post-surgery; (3) distant metastasis in other locations following surgery; (4) loss to follow-up.Ultimately, 1377 patients were enrolled, categorized into two groups: (1) the LM group, defined as the liver metastasis that occurred more than 3 months subsequent to the primary tumor resection, and (2) the non-LM group, defined as patients without distant metastasis within the initial 36 months after surgery.

Clinical information
Clinical data, inclusive of age, gender, T stage, N stage, TNM stage, tumor size, tumor location, differentiation status, lymphatic vascular invasion, neural invasion, neoadjuvant and adjuvant chemoradiotherapy, number of lymph nodes harvested (LNH) during surgery, mismatch repair (MMR) status, preoperative fibrin degradation products (FDP), neutrophil, monocyte, lymphocyte, and platelet counts, as well as CEA, CA19-9, cancer antigen 125 (CA125), and alpha-fetoprotein (AFP) levels (determined three months following the surgery), were obtained via the retrospective examination of medical records.The optimal cutoff values for CEA, CA125, CA19-9, AFP, FDP, LNH, tumor size, neutrophil, monocyte, lymphocyte, and platelet counts were determined utilizing receiver operating characteristic (ROC) curves.Median values for age were used as the cutoff values for this data.

Clinical follow-up
Regular clinical follow-up examinations were conducted at intervals of 3 months during the initial 3 years and subsequently every 6 months.During the follow-up, detection of LM was confirmed via Contrast-Enhanced CT scan.Patients without metastasis were monitored for a minimum of 3 years after undergoing surgery.The follow-up time was defined as the period commencing from the day following CRC surgery until the appearance of LM or the endpoint of the follow-up.

Statistical analysis
Statistical analyses were performed with R version 4.2.1 software and SPSS version 25.0.To facilitate the analyses, all continuous variables were converted to categorical variables.Categorical variables were presented as frequencies and percentages, and statistical tests such as chisquare or Fisher's exact test were conducted.Risk factors for LM in patients were first evaluated through univariate analysis.Subsequently, a backward regression multivariable analysis was used to identify independent predictors for LM when p values were less than 0.1.A prediction nomogram was constructed based on the final regression analysis, and its predictive accuracy was evaluated utilizing discrimination and calibration.Specifically, the overall discriminatory ability of the model was assessed using the area under the curve (AUC).Calibration was conducted using bootstrapping, using 1000 research resamples.The calibration plots, which illustrate the relationship between observed proportions and predicted probabilities, were utilized to evaluate calibration.The clinical utility of the nomogram was evaluated using decision curve analysis (DCA).All statistical tests were two-tailed, and statistical significance was set at P < 0.05.

Results
A total of 1377 patients were included in this study, comprising 824 males and 553 females with a median age of 58 years (range 21-91 years).The primary tumor in 636 patients located in the colon and in the other 741 patients located in the rectum.The median duration of follow-up was 79 months (range 21-149 months) for the LM group, and 79 months (range 38-149 months) for the non-LM group.In total, 165 patients (12.0%) experienced LM.Among them, 89 cases occurred within 3-12 months, 48 cases within 13-24 months, 16 cases within 25-36 months, and 12 cases after 36 months.The overall cohort of participants was randomly partitioned into a training set (n = 965) and a validation set (n = 412) using a 7:3 ratio.In the training cohort, the incidence of LM was found to be 12.5% (121/965), whereas in the validation cohort the incidence was 10.7% (44/412).However, this difference was found to be statistically insignificant (P = 0.331).Furthermore, patients' characteristics were found to be comparable between the two cohorts (Table 1).
In the training set, a uni-variable analysis was conducted to identify the potential predictors for LM in patients with CRC, where age, tumor size, T stage, N stage, TNM stage, LVI, neural invasion, LNH, MMR, neutrophil count, monocyte count, lymphocyte count, postoperative CEA level, postoperative CA125 level, and postoperative CA19-9 level were considered as potential variables.The backward elimination method was used for the multivariable logistic regression analysis, which revealed that T stage, N stage, LNH, MMR, neutrophil count, monocyte count, postoperative CEA level, postoperative CA125 level, and postoperative CA19-9 level were all independent predictors for LM (Table 2).
A nomogram was developed according to the multivariate analysis results to predict the probability of LM occurrence in patients with CRC (Fig. 1).The prediction model displayed a high predictive accuracy for both the training and validation sets, with an AUC of 0.782 and 0.768 in the training and validation sets, respectively (Fig. 2).The calibration curve also revealed that the predicted values of the model in both sets were in excellent agreement with the actual values (Fig. 3).Additionally, the DCA demonstrated significant net clinical benefit in both the training and validation sets, indicating the model's practical value in clinical applications (Fig. 4).Moreover, the accuracy, sensitivity, and specificity of the model were analyzed and are presented, which exhibited favorable discrimination performance (Table 3).Taken together, these findings suggest that the nomogram we developed has a significant potential for clinical decision-making in predicting LM occurrence among patients with CRC.

Discussion
Colorectal cancer is considered one of the primary causes of cancer-related mortality across the globe (Siegel et al. 2022).Liver metastasis is the major prognostic factor for CRC, and thus it becomes crucial to predict its occurrence in advance to devise appropriate strategies for improving patient survival.Our present study identified T stage, N stage, LNH, MMR, neutrophil counts, monocyte counts, postoperative CEA level, postoperative CA125 level, and postoperative CA19-9 level as independent predictors for liver metastasis in CRC patients.According to these risk factors, we developed a nomogram to predict the risk of LM in postoperative patients with CRC.The nomogram demonstrated a high level of accuracy, calibration, discrimination, and clinical application value.Furthermore, our study has a significant sample size and incorporates a comprehensive set of clinicopathological features.Importantly, all of these variables can be easily obtained during routine treatment, indicating the potential costeffectiveness of our nomogram.
According to NCCN guidelines, proper staging of CRC requires the sampling of at least 12 lymph nodes.Previous studies have suggested that CRC patients with less than 12 The MMR status is a significant tumor suppressor pathway that serves as one of the prognostic markers for CRC.Numerous previous studies have established the prognostic significance of MMR protein expression in CRC (Lanza et al. 2006;Guastadisegni et al. 2010).Notably, Zhu et al. demonstrated that CRC patients with dMMR status had a superior outcome than those with pMMR tumors (Zhu et al. 2018).In our study, we observed a substantially lower incidence of LM in patients with dMMR status compared to those with pMMR status, thereby implying that dMMR status may represent a robust protective factor against LM.
An increasing number of studies have indicated that elevated neutrophils and monocytes in peripheral blood are associated with unfavorable prognosis in CRC (Ocana et al.Neutrophils play a critical role in various stages of the carcinogenic process, encompassing tumorigenesis, growth, proliferation, and metastatic spread (Swierczak et al. 2015;Coffelt et al. 2016).Specifically, they promote tumorigenesis by releasing reactive oxygen species (ROS), reactive nitrogen (RNS), or proteases, impair the immune system to facilitate tumor proliferation, and further the dissemination of metastatic cells by impeding natural killing activity and promoting extravasation (Spiegel et al. 2016;Antonio et al. 2015).In addition, prior research has demonstrated that neutrophils promote liver metastasis of colorectal cancer (Spicer et al. 2012;Gordon-Weeks et al. 2017;Tang et al. 2022).
Consistent with these previous studies, our finding revealed that patients with a preoperative peripheral blood neutrophil count ≥ 8.6 × 10 9 cells/L exhibited a significantly greater risk of liver metastasis than those with a count < 8.6 × 10 9 cells/L.Monocytes are known to have a crucial role in tumor initiation and metastasis (Guilliams et al. 2018;Qian et al. 2011).Furthermore, myeloid-derived suppressor cells (MDSCs) with monocyte-like or granulocyte-like characteristics have been shown to enhance radiation resistance, impede T cell function, and facilitate tumor progression and metastasis (Liang et al. 2017;Tesi 2019).Despite their significance in cancer development, only one study has investigated the relationship between monocyte count and LM.Specifically, Hu et al. found that a preoperative peripheral blood monocyte count of > 0.505 × 10 9 cells/L was an independent risk factor for preoperative LM (Hu et al. 2016).Our results revealed that the incidence of postoperative LM was 17.4% in patients with a monocyte count ≥ 0.5 × 10 9 cells/L, compared to 11.1% in patients with a count < 0.5 × 10 9    The use of preoperative serum molecular markers in predicting CRC prognosis has garnered significant attention, but prediction using postoperative markers is less (Park et al. 2009).Postoperative tumor markers are strong indicators for patient prognosis (Konishi et al. 2018;S. Zhou et al. 2021;C. Li et al. 2023;Dai et al. 2021;C. Li et al. 2021).One of the most broadly used tumor markers for CRC is CEA (Primrose et al. 2014).Preoperative serum CEA levels have been found to be significantly associated with postoperative LM (Chuang et al. 2011).Conversely, a retrospective cohort study involving 434 patients with rectal cancer found that only postoperative CEA levels were a risk factor for postoperative LM, with preoperative levels showing no association (Khan et al. 2019).Dai et al. also highlighted the significance of postoperative CEA levels in predicting relapse (Dai et al. 2021).Accordingly, in our study, we did not analyze preoperative serum tumor markers and found that the incidence of LM was 28.6% in patients with postoperative CEA levels ≥ 3.4 ng/mL, compared to 9.3% in those with levels < 3.4 ng/mL.
On the other hand, the potential of postoperative CA125 as a prognostic factor in CRC has rarely been discussed, but it has great potential for predicting CRC recurrence (Tokodai et al. 2016).Elevated levels of CA125 have been closely linked to worse prognosis and metastasis development, which could be attributed to the promotion of tumor cell proliferation and the inhibition of anti-tumor immune responses (Giannakouros et al. 2015;Schultes and Whiteside 2003).Several studies have demonstrated that CA125 levels were significantly related to LM in other cancers (Shi et al. 2016;Chen et al. 2015).Zhang et al. found that preoperative CA125 levels are important in predicting LM for CRC patients (D.Zhang et al. 2013).However, no previous reports have explored the association between postoperative CA125 levels and LM.For the first time, our analysis supported the value of postoperative serum CA125 in predicting LM for colorectal cancer.In our study, the incidence of LM was higher in patients with a postoperative CA125 level greater than or equal to 23.8 U/mL.CA19-9 is a cell surface glycoprotein involved in cellular adhesion and its expression in cancer cells can indicate a higher potential for metastasis and invasion (Scarà et al. 2015;Swords et al. 2016).The CA19-9 monosialoganglioside is also implicated in tumor cell-induced platelet aggregation, which facilitates distant metastases in CRC (Martini et al. 2000).Prior studies have demonstrated that CRC patients with elevated preoperative CA19-9 levels have a high risk of postoperative metastasis, leading to a worse survival rate (Z.Lu et al. 2016;Sato et al. 2016;W. Zhou et al. 2019).However, the relationship between postoperative CA19-9 levels and LM has not been studied previously.Our research revealed that patients with a postoperative CA19-9 level ≥ 14.3 U/mL had a significantly higher risk of LM than those with a postoperative CA19-9 level < 14.3 U/mL (20.5% vs. 9.6%).Our findings suggest that even if postoperative CA19-9 levels are within normal limits, they still possess prognostic prediction ability.
There are some limitations in this study which must be acknowledged.First, the use of a retrospective method inevitably generates selection bias.Second, the training and validation sets were obtained retrospectively from a sole center, thereby limiting the generalizability of the model.Consequently, further validation is required through multicenter prospective clinical studies.Lastly, a larger sample size is necessary for a multicenter analysis in order to verify the conclusions of this study.
In conclusion, this study has successfully developed and validated a novel and user-friendly nomogram for the prediction of LM risk in CRC patients.This nomogram could assist clinicians in screening patients at high risk of LM after surgery.Clinicians could make individualized followup strategies and treatment protocols for these patients to improve the long-term prognosis.
Author contributions All authors contributed to the study conception and design.XD, FF, and JZ designed the study.XD, JX, YT, LD, and LN collected the data.XD, GZ, GR, and HD analyzed the data.XD, JX, ZX, HW, and RL visualized the data.XD drafted the manuscript.FF and JZ revised the manuscript.All authors read and approved the final manuscript.

Funding
The authors declare that no funds, grants, or other support was received during the preparation of this manuscript.

Fig. 1 Fig. 3
Fig. 1 Nomogram for predicting liver metastasis in patients with colorectal cancer

Fig. 4
Fig. 4 DCA of the nomogram model for predicting LM in the training (A) and validation sets (B)

Table 1
Characteristics of the patients in the training and validation cohorts

Table 2
Univariate and multivariate logistic regression in the training cohort Zhang et al. 201916), while elevated lymphocytes suggest a more favorable clinical outcome (Y.Y.Zhang et al. 2019).

Table 3
The predictive performance of the training cohort and the validation cohort AUC area under the curve; CI confidence interval