Liver cancer prognosis model and construction method and application thereof

Abstract

The invention belongs to the field of gene detection, relates to a liver cancer prognosis model and a construction method and application thereof, discovers that SLAMF7 high-expression macrophages are mainly enriched in cancer tissues, and defines that the SLAMF7 high-expression macrophages have an obvious space co-localization phenomenon with tumor reactive CD8T cells. Further through alternating current analysis, it is found that the SLAMF7 high-expression macrophage can regulate and control cytotoxicity related genes of CD8T cells, and the immune stimulation function is achieved; on the basis of liver cancer transcriptome data, SLAMF7 high-expression macrophages and tumor reactive CD8T cell alternating current ligand receptors are utilized to construct a liver cancer risk assessment model, the result shows that the person with high risk score has shorter total lifetime than the person with low risk score, and the person with high risk score often has drug resistance to immunotherapy. Evaluating the immunotherapy response condition of the patient based on the quantified model score, wherein the model can be used for assisting in evaluating prognosis and individualized guidance of the liver cancer patient.

Description

Technical Field

[0001] The present invention belongs to the field of gene detection, and relates to a liver cancer prognosis model, a construction method thereof, and an application thereof. Background Art

[0002] With the discovery of immune checkpoints and the advent of immune checkpoint inhibitors, immunotherapy has been included in the primary liver cancer diagnosis and treatment guidelines. Recently, the results of multiple clinical trials around anti-PD1 monoclonal antibodies have demonstrated their potential in the treatment of advanced liver cancer. However, some patients still cannot benefit from it. The advent of single-cell sequencing technology has made it possible to analyze the cell heterogeneity within the tumor microenvironment, and the emergence of spatial transcriptome sequencing technology can provide spatial information of different cell populations. A recent study integrated single-cell transcriptome and spatial transcriptome analysis and found that SPP1-highly expressed macrophages strengthen the function of tumor-associated fibroblasts through cell communication, forming an "immune barrier" that blocks the infiltration of anti-tumor immune cells at the cancer focus junction, which is related to shorter survival and immune therapy resistance in liver cancer patients. The above research provides a potential prognosis assessment model, that is, constructing a risk assessment model based on the communication ligand-receptor of key cell populations in disease progression, which is of great significance for the formulation of individualized treatment plans for liver cancer patients.

[0003] The wide application of next-generation sequencing technology has further deepened people's understanding of liver cancer, and different research teams have also developed a large number of prognosis models for patient prognosis assessment. However, the past liver cancer prognosis models were usually constructed based on differentially expressed genes screened from tissue RNA sequencing expression profiles, but this may lose the key features of cancer development and incorporate interference signals unrelated to the disease process, resulting in insufficient model efficacy and limited application. Through the integrated analysis of single-cell transcriptome and spatial transcriptome, identifying key cell populations in the microenvironment that affect disease development, clarifying the downstream effects of cell communication, and using the communication ligand-receptor of target cell populations, a more efficient and interpretable liver cancer prognosis model can be constructed. Summary of the Invention

[0004] Aiming at the deficiencies of the existing liver cancer prognosis model in the above-mentioned prior art, the present invention provides a method for constructing a liver cancer prognosis model and its application.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] On the one hand, the present invention provides a liver cancer prognosis model, which is constructed based on the communication ligand-receptor genes of SLAMF7-highly expressed macrophages and tumor-reactive CD8 T cells; the specific model is Riskscore = ∑ i Expression(mRNA) i *Coefficent(mRNA)i , where i is the key gene screened.

[0007] In a second aspect, the present invention provides an application of the above liver cancer prognosis model, and risk assessment is performed on liver cancer patients in each cohort through this model. First, the risk score of the patient is calculated through the prognosis model, and then the optimal cut-off value of the risk score is determined based on the surv_cutpoint function of the survival package. When the risk score of the patient is equal to or higher than the optimal cut-off value, the patient is included in the high-risk group, and when the risk score of the patient is lower than the optimal cut-off value, the patient is included in the low-risk group.

[0008] Furthermore, the optimal cut-off value of the above risk score is determined through the R packages survival and survminer.

[0009] In a third aspect, the present invention provides a method for constructing a liver cancer prognosis model, including the following steps:

[0010] 1) Collect single-cell data from the same public cohort containing cancer single-cell data to identify tumor-reactive CD8 T cells and macrophages;

[0011] 2) Clarify the spatial position between the tumor-reactive CD8 T cells obtained in step 1) from the perspective of spatial transcriptome data, and find that there is a high co-localization between the tumor-reactive CD8 T cells and macrophages with high expression of SLAMF7, so as to obtain macrophages with high expression of SLAMF7 having a high spatial correlation with tumor-reactive CD8 T cells;

[0012] 3) Use Nichenetr to infer the signals sent by the macrophages with high expression of SLAMF7 obtained in step 2) to tumor-reactive CD8 T cells, and identify the ligand-receptor pairs with the top active scores; then use univariate Cox regression analysis to screen the ligand-receptor pairs with the top scores identified by Nichenetr to obtain ligand-receptor genes that have a suggestive effect on prognosis;

[0013] 4) Construct a liver cancer prognosis model using multivariate cox regression based on the ligand-receptor genes that have a suggestive effect on prognosis obtained in step 3).

[0014] Furthermore, the specific implementation method for identifying tumor-reactive CD8 T cells in step 1) adopted by the present invention is:

[0015] 1.1) Collect various cancer single-cell data from a public cohort containing single-cell data of primary liver cancer, such as single-cell data of hepatocellular carcinoma, mixed liver cancer, intrahepatic cholangiocarcinoma, and adjacent cancer cells, etc.; obtain single-cell data of primary liver cancer;

[0016] 1.2) Use the R package Seurat to annotate the cell types of the primary liver cancer single-cell data obtained in step 1.1), and extract the CD8 T cell population with high expression of CD8A;

[0017] 1.3) Use the FindClusters function to further subdivide the CD8 T cells with high expression of CD8A obtained in step 1.2); screen out the CD8 T cell subsets enriched in cancer tissues through OR analysis;

[0018] 1.4) Use the fgsea function to perform GSEA analysis on the CD8 T cells enriched in cancer tissues obtained in step 1.3), and screen out the CD8 T cell subsets with high expression of the T cell activation pathway, that is, the tumor-reactive CD8 T cell subsets.

[0019] Furthermore, the specific implementation method for identifying macrophages with high expression of SLAMF7 in step 2) adopted in the present invention is:

[0020] Use the R package Seurat to annotate the cell types of the primary liver cancer single-cell data obtained in step 1.1), and extract macrophages with high expression of CD68; use the FindClusters function to further subdivide the obtained macrophages with high expression of CD68, and identify the SLAMF7-high-expressing macrophage subsets with high spatial correlation with tumor-reactive CD8 T cells.

[0021] Furthermore, the specific implementation method for step 3) adopted in the present invention is:

[0022] Use Nichenetr to infer the signals sent by the SLAMF7-high-expressing macrophages obtained in step 2) to tumor-reactive CD8 T cells, and retain the top thirty ligand-receptor pairs and the genes regulated by the ligands in terms of activity scores, that is, obtain the ligand-receptor genes to be screened.

[0023] Collect the GSE76427 cohort with prognostic information from the GEO database, and use univariate Cox regression analysis based on this cohort to screen the ligand-receptor genes obtained as described above; retain the genes with p < 0.05 as the ligand-receptor genes with prognostic implications for model construction.

[0024] Furthermore, the specific implementation method for step 4) adopted in the present invention is:

[0025] Collect the TCGA cohort with prognostic information from the UCSCXena database, the GSE76427 cohort, GSE144269 cohort, and GSE14520 cohort with prognostic information from the GEO database, and the CHCC liver cancer cohort with prognostic information from the NODE database. Use the ligand-receptor genes identified in step 3) that are predictive of prognosis in the above cohorts to construct a multivariate Cox model. The calculation formula of the model is as follows:

[0026] Riskscore = ∑ i Expression(mRNA) i *Coefficent(mRNA) i

[0027] Riskscore is the patient risk score, i is each gene selected, Expression is the expression level of gene i, and Coefficent is the regression coefficient of gene i.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] The present invention provides a liver cancer prognosis model, its construction method and application. After identification, it is found that SLAMF7-highly expressed macrophages are mainly enriched in cancer tissues, and it is clarified that there is an obvious spatial co-localization phenomenon with tumor-reactive CD8 T cells. Further communication analysis reveals that SLAMF7-highly expressed macrophages can regulate the cytotoxicity-related genes of CD8 T cells and exert an immune-stimulating function. Therefore, based on liver cancer transcriptome data, the present invention develops a liver cancer risk assessment model using the ligand-receptor for communication between SLAMF7-highly expressed macrophages and tumor-reactive CD8 T cells. The results show that patients with a high risk score have a shorter overall survival compared to those with a low score, and patients with a high risk score are often resistant to immunotherapy. Based on the quantitative model score, the immune therapy response of patients is evaluated. This model can be used to assist in assessing the prognosis of liver cancer patients and providing individualized guidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a diagram showing the identification and functional analysis of tumor-reactive CD8 T cells used in the present invention. Among them, Figure 1-A : The UMAP diagram shows 9 CD8 T cell subsets, Figure 1-B : The stacked bar chart shows the proportions of each CD8 subset in different samples, Figure 1-C : The heat map shows the expression of the classical functional gene set of T cells in each CD8 T cell subset, Figure 1-D : The heat map shows the expression of the T cell activation pathway in each CD8 T cell subset, Figure 1-E : The heat map shows the correlation between the CD8 T cell subsets of our own cohort and the tumor-reactive CD8 T cell subsets of Eberhardt et al.Figure 1-F : The UMAP plot shows the CD8 T cell subsets in our own cohort and those in Liu et al.'s study. Figure 1-G : The box plot shows the expansion fractions of CD8 T cell subsets in Liu et al.'s study.

[0031] Figure 2 is a diagram for identifying potential communication partners of tumor-reactive CD8 T cells used in the present invention. Among them, Figure 2-A : The Venn diagram shows the intersecting genes of the top 200 genes correlated with the SLAMF7 gene in 4 cohorts. Figure 2-B : The violin plot shows the scores of the intersecting genes in each cell population. Figure 2-C : The scatter plot shows the correlation between the degree of myeloid cell infiltration and the degree of infiltration of CXCL13-high-expressing CD8 T cells in 4 cohorts.

[0032] Figure 3 is a diagram for identifying SLAMF7-high-expressing macrophages and analyzing the prognostic association used in the present invention. Among them, Figure 3-A : The UMAP plot shows 8 macrophage subsets. Figure 3-B : The stacked bar plot shows the proportions of each macrophage subset in different samples. Figure 3-C : The histogram shows the functional pathways enriched from the reactome database. Figure 3-D : The heat map shows the expression of classical functional pathways of macrophages in different macrophage subsets. Figure 3-E : The Kaplan-meier curve shows the clinical characteristics of Scissor+ macrophages and Scissor- macrophages. Figure 3-F : The histogram shows the proportions of different macrophage subsets in Scissor+ macrophages and Scissor- macrophages.

[0033] Figure 4 is a diagram for analyzing the spatial co-localization of SLAMF7-high-expressing macrophages and tumor-reactive CD8 T cells used in the present invention. Among them, Figure 4-A : The CellTrek algorithm performs spatial mapping on SLAMF7-high-expressing macrophages and CXCL13-high-expressing CD8 T cells. Figure 4-B : The box plot shows the spatial distances between CXCL13-high-expressing CD8 T cells and each macrophage subset.

[0034] Figure 5 is a diagram for constructing a risk assessment model based on communication ligands and receptors and evaluating cohorts used in the present invention. Among them, Figure 5-A : Ligands of macrophages predicted by Nichenetr. Figure 5-B : Target genes regulated by CXCL13-high-expressing CD8 T cells predicted by Nichenetr. Figure 5-C : The forest plot shows ligand-receptor genes related to prognosis. Figure 5-D: The Kaplan - meier curve shows the prognostic efficacy of the screened genes, Figure 5-E : The Kaplan - meier curve shows the prognostic efficacy of the constructed model, Figure 5-F : The box plot shows that the constructed model can be used for predicting immunotherapy response, Figure 5-G : The Kaplan - meier curve shows the prognostic efficacy of the constructed model in the immunotherapy cohort. Specific implementation manners

[0035] The present invention provides a method for constructing a liver cancer prognosis model, which mainly includes the following steps:

[0036] Step 1: Identification and functional analysis of tumor - reactive CD8 T cells. Specifically: 7 cases of hepatocellular carcinoma, mixed hepatocellular carcinoma, intrahepatic cholangiocarcinoma and adjacent cancer samples were collected from the public cohort of Xue et al. (National Genomics Data Center, project ID number is PRJCA007744). The single - cell data of 28 cases of primary liver cancer tissues collected above were annotated using the R package seurat. CD8 T cells with high expression of CD8A in the above - mentioned samples were extracted. Subsequently, the extracted CD8 T cells were further subdivided, and the obtained CD8 T cell data are as follows: As Figure 1-A shown, the UMAP plot shows that CD8 T cells are divided into 9 subsets. The cell proportion plot shows the distribution proportions of different CD8 T cell subsets in different types of samples. As Figure 1-B shown, CD8 T cells with high expression of CXCL13 mainly infiltrate into cancer samples of various liver cancers, especially cancer samples of CHC. The 9 different CD8 T cell subsets were scored for classical T cell gene sets using UCELL. As Figure 1-C shown, CD8 T cells with high expression of CXCL13 show higher co - stimulation scores and cytotoxicity scores. The gsea analysis of 5 pathways of T cell activation was performed on 9 different CD8 T cell subsets using the fgsea function. As Figure 1-D shown, CD8 T cells with high expression of CXCL13 up - regulate 5 activation pathways. After integrating the obtained CD8 T cell (target CD8 T cell) data with the CD8 T cell (external CD8 T cell) data verified by in vitro experiments of Eberhardt et al. (GEO database, project number is GSE180268), a correlation analysis was performed. As Figure 1-E shown, the target CD8 T cells and the external CD8 T cells have the highest correlation. After integrating the obtained CD8 T cell data with the CD8 T cell data with paired TCR sequencing of Liu et al. (GEO database, project number is GSE179994), umap clustering was performed. As Figure 1-FAs shown, there is also a subset of CD8 T cells with high expression of CXCL13 in CD8 T cells of Liu et al. The TCR data was analyzed by the STARTRAC algorithm. As Figure 1-G shown, CXCL13-high-expressing CD8 T cells have the highest amplification fraction.

[0037] The above results together indicate that the subset of CD8 T cells with high expression of CXCL13 is tumor-reactive CD8 T cells.

[0038] Step 2: Identification of potential communication partners of tumor-reactive CD8 T cells, specifically:

[0039] Collect the TCGA-LIHC cohort from the UCSCXena database, and collect the GSE25097 cohort, GSE14520 cohort, and GSE36376 cohort from the GEO database. Around these four liver cancer tissue sequencing cohorts with more than 200 samples, perform spearman correlation analysis of SLAMF7 with other genes in turn, sort according to the correlation coefficient from high to low, take the top 200 genes for each cohort and then take the intersection. As shown in Figure 2A, finally 32 genes are obtained. Then use the addmodulescore function to score the 32 genes for different cell types. As shown in Figure 2B, the scores show that these 32 genes are highly expressed on myeloid cells such as monocytes, neutrophils, dendritic cells, and macrophages. Finally, perform Cibersortx deconvolution on the four cohorts to evaluate cell type abundance and perform spearman correlation analysis. As shown in Figure 2C, macrophages and tumor-reactive CD8 T cells have a stable positive correlation. The above results indicate that macrophages are potential communication partners of tumor-reactive CD8 T cells.

[0040] Step 3: Identification of SLAMF7-high-expressing macrophages and prognostic association analysis, specifically:

[0041] Use the R package seurat to subset macrophages expressing CD68. As Figure 3-A shown, the UMAP plot shows that macrophages are divided into 8 subsets, and the cell proportion plot shows the distribution proportions of different macrophage subsets in different types of samples. As Figure 3-B shown, among them, the macrophage subset with high expression of SLAMF7 is mainly infiltrated in cancer samples of various liver cancers, especially cancer samples of CHC. Then perform spearman correlation analysis of SLAMF7 with other genes in the TCGA liver cancer cohort, sort according to the correlation coefficient from high to low, and take the top 200 genes for reactome enrichment analysis. As Figure 3-C shown, the SLAMF7 gene is closely associated with the T cell activation pathway. Perform gene set scoring on different macrophage populations through UCELL. As Figure 3-DAs shown, macrophages with high SLAMF7 expression are M1 macrophages, and their antigen presentation function is upregulated. Finally, the Scissor algorithm was used to evaluate the prognostic value of the target macrophages. As Figure 3-E shown, Scissor+ is associated with better survival prognosis, and Scissor- is associated with worse survival prognosis. Finally, the relationship between different macrophages and prognosis was evaluated. As Figure 3-F shown, Scissor+ macrophages are mainly macrophages with high SLAMF7 expression, indicating that the target macrophages are associated with better prognosis.

[0042] Step 4: Spatial co-localization analysis of macrophages with high SLAMF7 expression and tumor-reactive CD8 T cells, specifically:

[0043] The downsample function was used to resample macrophages and CD8 T cells, with 500 cells taken from each cell subset. Then, the CellTrek algorithm was used to map the single-cell data to the spatial transcriptome data. As Figure 4-A shown, on three spatial transcriptome slices of liver cancer, macrophages with high SLAMF7 expression co-localize with tumor-reactive CD8 T cells. Then, the kdist function was used to calculate the spatial distance between different macrophage subsets and tumor-reactive CD8 T cells. As Figure 4-B shown, macrophages with high SLAMF7 expression have the closest spatial distance to tumor-reactive CD8 T cells compared to other macrophage subsets.

[0044] Step 5: Construction of a risk assessment model based on ligand-receptor interactions and cohort evaluation graph, specifically:

[0045] Nichenetr was used to infer the signals sent by macrophages with high SLAMF7 expression to tumor-reactive CD8 T cells. As Figure 5-A and Figure 5-B shown, the top thirty ligand-receptor pairs with active scores and the genes regulated by the ligands were retained.

[0046] The GSE76427 cohort with prognostic information was collected from the GEO database. Based on this cohort, univariate Cox regression analysis was used to screen the ligand-receptor pairs identified by Nichenetr, and 5 ligand-receptor genes were obtained. As Figure 5-C shown, the forest plot shows these 5 ligand-receptor genes that are significantly associated with the prognosis of liver cancer patients (Cox p < 0.05), and these genes are all protective factors (HR < 1). As Figure 5-D shown, the Kaplan-meier curve indicates that these genes are associated with better prognosis of patients.

[0047] Based on the 5 ligand-receptor genes screened out, a multivariate Cox model was constructed in the GSE76427 cohort, and the model formula is as follows:

[0048] Riskscore = ∑ i Expression(mRNA) i *Coefficent(mRNA) i

[0049] Riskscore is the patient's risk score, i is each of the 5 ligand-receptor genes screened out, Expression is the expression level of gene i, and Coefficent is the regression coefficient of gene i.

[0050] The risk score of the patient was calculated through the constructed prognostic model, and then the optimal cut-off value of the risk score was determined based on the surv_cutpoint function of the survival package. When the patient's risk score is equal to or higher than the optimal cut-off value, the patient is included in the high-risk group; when the patient's risk score is lower than the optimal cut-off value, the patient is included in the low-risk group. The Kaplan-Meier curve was used to evaluate the survival prediction performance of the model. As Figure 5-E shown, excellent prediction performance was shown in the GSE76427, TCGA, CHCC, GSE14520, and GSE144269 liver cancer cohorts. Patients with a high risk score had a shorter overall survival. The model also showed excellent survival prediction performance in the liver cancer cohort receiving immunotherapy. As Figure 5-G shown, those with a high risk score had a worse prognosis. The model can also be used to evaluate the response of patients to immunotherapy. As Figure 5-F shown, non-responders to immunotherapy showed a higher risk score.

[0051] The above results indicate that this model can be used to assist in evaluating the prognosis of liver cancer patients and guiding individualized treatment.

[0052] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for constructing a liver cancer prognosis model, characterized in that: The following steps are involved: 1) Collecting single-cell data from the same public cohort containing cancer single-cell data to identify tumor-reactive CD8 T cell subsets and macrophages; 2) The spatial position between the tumor-reactive CD8T cell subpopulation and the macrophages obtained in step 1) was clarified from the perspective of spatial transcriptome data, and it was found that the tumor-reactive CD8T cell subpopulation and the macrophage subpopulation with high expression of SLAMF7 were highly co-localized, thereby obtaining the macrophage subpopulation with high expression of SLAMF7 that had a high spatial correlation with the tumor-reactive CD8T cell subpopulation; 3) using Nichenetr to infer the signals sent by the SLAMF7-high-expressing macrophages to the tumor-reactive CD8 T cells obtained in step 2), and identifying the ligand receptor pairs with the highest activity scores; then using univariate Cox regression analysis to screen the ligand receptor pairs with the highest scores identified by Nichenetr, and obtaining ligand receptor genes that have a prognostic effect; 4) Use multivariate Cox regression to construct a liver cancer prognostic model based on the ligand receptor genes that have a prognostic effect obtained in step 3).

2. The method for constructing a liver cancer prognosis model according to claim 1, characterized in that: The step of identifying tumor-reactive CD8 T cells in step 1) comprises: 1.1) Use the R package seurat to annotate the cell types of the primary liver cancer single-cell data in the public cohort containing the primary liver cancer single-cell data, and extract CD8 T cells with high expression of CD8A; 1.2) using the FindClusters function to subdivide the CD8 T cells obtained in step 1.1); preferably, screening out the CD8 T cell subpopulation enriched in cancer tissues by OR analysis; 1.3) Use the fgsea function to perform GSEA analysis on the CD8 T cells enriched in cancer tissues obtained in step 1.2) to screen out CD8 T cell subpopulations that highly express T cell activation pathways, that is, to obtain tumor-reactive CD8 T cell subpopulations.

3. The method for constructing a liver cancer prognosis model according to claim 2, characterized in that: The step of identifying SLAMF7 highly expressing macrophages in the step 2) includes: using the R package seurat to annotate the primary liver cancer single cell data obtained in the step 1.1) with cell types, annotating the cell types, and extracting macrophages that highly express CD68; using the FindClusters function to further subdivide the obtained macrophages that highly express CD68, and identifying a subpopulation of SLAMF7 highly expressing macrophages that has a high spatial correlation with tumor-reactive CD8T cells.

4. The method for constructing a liver cancer prognosis model according to claim 3, characterized in that: The step 3) includes: using Nichenetr to infer the signals sent by SLAMF7-high-expressing macrophages to tumor-reactive CD8T cells, retaining the ligand receptor pairs with the highest activity scores and the genes regulated by the ligands, that is, obtaining the exchange ligand receptor genes to be screened; collecting the GSE76427 cohort with prognostic information from the GEO database, and screening the obtained exchange ligand receptor genes based on the cohort using univariate Cox regression analysis; retaining genes with p<0.05 as ligand receptor genes that have a prognostic effect, and the ligand receptor genes that have a prognostic effect are used to construct a liver cancer prognosis model.

5. The method for constructing a liver cancer prognosis model according to claim 4, characterized in that: The step 4) comprises: The TCGA-LIHC cohort with prognostic information was collected from the UCSCXena database, the GSE76427 cohort, the GSE144269 cohort and the GSE14520 cohort with prognostic information were collected from the GEO database, and the CHCC liver cancer cohort with prognostic information was collected from the NODE database. The above cohorts were used to construct a multivariate Cox model based on the ligand receptor genes that have a prognostic effect determined in step 3). The model calculation formula is as follows: Riskscore=∑ i Expression(mRNA) i *Coefficent(mRNA) i in: i is the key gene for screening.

6. A liver cancer prognosis model constructed based on the liver cancer prognosis model construction method according to any one of claims 1 to 5, characterized in that: The expression of the liver cancer prognosis model is as follows: Riskscore=∑ i Expression(mRNA) i *Coefficent(mRNA) i in: Riskscore is the patient risk score; i represents each gene screened; Expression is the expression level of gene i; Coefficent is the regression coefficient of gene i.

7. Use of the liver cancer prognosis model according to claim 6 in risk assessment of liver cancer samples.

8. The use according to claim 7, characterized in that: The implementation method of the application is: first, the patient's risk score is calculated through the prognostic model, and then the optimal critical value of the risk score is determined based on the surv_cutpoint function of the survival package. When the patient's risk score is equal to or higher than the optimal critical value, the patient is included in the high-risk group, and when the patient's risk score is lower than the optimal critical value, the patient is included in the low-risk group.

Images