Prognosis evaluation method and system for II-III stage colorectal cancer patient

Abstract

The invention discloses a prognosis evaluation method and system for II-III stage colorectal cancer patients, and relates to the technical field of wisdom medicines.The prognosis evaluation method comprises the steps that variables related to survival prognosis of the patients are screened, potential variables influencing RFS and OS of the patients are obtained, and prognosis outcomes of the colorectal cancer patients are evaluated by evaluating relevance between various indexes and prognosis outcomes of the colorectal cancer patients and correcting confounding factors; screening out indexes having important value for the model; the screened indexes and machine learning algorithms are utilized to construct RFS and OS prognosis models of the CRC patients, the RFS and OS prognosis models are used for performing prognosis evaluation on the CRC patients in the II-III stages, and the machine learning algorithms comprise six types of LR, RF, XGB, SVC, MLP and GNB. By integrating clinical data and a machine learning technology, an efficient and accurate tool is provided for prognosis evaluation of the II-III stage colorectal cancer patient, and a scientific basis is provided for making clinical decision support and an individualized treatment scheme.

Description

Technical Field

[0001] The present invention relates to the field of smart medical technology, and in particular to a method and system for prognostic assessment of patients with stage II-III colorectal cancer. Background Art

[0002] In recent years, the incidence of colorectal cancer (CRC) has been increasing annually. According to the latest statistics from the International Agency for Research on Cancer, approximately 1.9 million new cases of CRC (including anal cancer) will be diagnosed worldwide in 2022, resulting in 904,000 deaths. Its incidence ranks third among cancers worldwide, and its mortality rate ranks second only to lung cancer. Radical surgery is the only effective cure for early-stage CRC patients, with a 5-year survival rate of approximately 90% for stage I patients after radical surgery. However, for stage II and stage III CRC patients, the 3-year recurrence or distant metastasis rates are approximately 22% and 43%, respectively, due to the presence of minimal residual disease. Currently, there is no cure for stage IV patients. In clinical practice, CEA and CA19-9 are commonly used biomarkers for CRC prognosis. However, some patients with negative CEA or CA19-9 levels still have high recurrence rates, and even with the same tumor marker concentration, survival outcomes vary significantly. Therefore, there is an urgent need to identify more effective biomarkers to accurately assess patient prognosis and guide clinicians in developing personalized treatment plans.

[0003] CRC is a multifactorial gastrointestinal malignancy whose development results from a complex interaction between genetic susceptibility and environmental exposures (such as dietary pattern and gut microbiome dysbiosis). Notably, a chronic inflammatory microenvironment is a core pathological feature throughout the carcinogenesis process. Epidemiological studies have shown that unhealthy dietary habits, such as processed foods, red meat consumption, and excessive alcohol consumption, as well as inappropriate lifestyles, such as sedentary lifestyles and lack of exercise, are modifiable environmental risk factors for CRC. These factors can lead to gut microbial dysbiosis, which in turn affects the stability of the gut microenvironment. Studies have shown that CRC precancerous lesions are closely associated with early-life antibiotic exposure, which can reshape the composition of the gut microbiota and disrupt the balance of the gut microbiome. Furthermore, certain specialized intestinal bacteria can promote the formation of an intestinal inflammatory environment by adhering to epithelial cells or suppressing cellular immune tolerance. This, in turn, activates the WNT / β-catenin signaling pathway, inducing rapid proliferation of intestinal epithelial cells, thereby promoting the development and progression of CRC. Furthermore, individuals with long-term ulcerative colitis or Crohn's disease have a significantly increased risk of CRC, approximately two to three times that of the general population. The risk of CRC increases with longer disease duration. However, even among people with unhealthy lifestyles, aspirin use has been shown to reduce the risk of CRC. A meta-analysis of 27 studies showed that aspirin use was an independent prognostic factor for postoperative cancer-specific survival (CSS) and overall survival (OS) in CRC patients, further suggesting that chronic inflammation plays a key role in the induction and progression of CRC. Furthermore, changes in chronic inflammation levels can be reflected in inflammatory factors and cells in the tumor microenvironment and peripheral circulation.

[0004] In early-stage CRC patients, chronic inflammation levels are typically low, making it difficult to detect statistically significant changes using a single marker. However, in advanced or advanced stages, tumor-related inflammation levels rise significantly, primarily manifested by low albumin, low prealbumin, and low lymphocyte counts, as well as elevated fibrinogen, high white blood cell counts, high neutrophil counts, and high monocyte counts. Previous studies by the research team have found that single inflammatory markers (such as white blood cell count, lymphocyte count, and neutrophil count) are not significantly correlated with colorectal cancer prognosis. In contrast, inflammatory ratios, such as the fibrinogen-to-prealbumin ratio (FPR) and the fibrinogen-to-lymphocyte ratio (FLR), are more sensitive and can more effectively reflect cancer patient prognosis. Based on this, the research team further collected clinical data and inflammatory markers, including fibrinogen and the platelet-to-prealbumin ratio (FPAR), from 489 patients undergoing gastric cancer surgery. The results showed that a machine learning prognostic model constructed based on FPAR provided more accurate prognostic predictions than when it was not included. In this study, the research team included a total of 61 laboratory test indicators (two tumor markers and 59 inflammatory indicators). Among them, indicators such as FPR and albumin-prealbumin ratio (AFR) have been reported to be associated with CRC prognosis. However, some new chronic inflammatory ratios or scores are reported for the first time, and their correlation with CRC patient prognosis requires further analysis.

[0005] Machine learning, a key branch of artificial intelligence, focuses on deeply analyzing existing datasets to uncover inherent patterns and regularities, thereby predicting future data trends based on these learned patterns. Leveraging its powerful data processing capabilities, machine learning has been widely applied in the medical field, primarily in assisting disease diagnosis, predicting prognosis, and optimizing personalized treatment plans, effectively improving the rational allocation and efficient utilization of medical resources. Accurately predicting the prognosis of CRC patients is currently a focus of intense research. Existing CRC prognostic models have been established based on logistic regression (LR), random forest (RF), support vector machine classifiers (SVC), k-nearest neighbor (KNN), multilayer perceptron (MLP), and extreme gradient boosting (XGBoost). These machine learning algorithms, through deep feature mining, can accurately identify key variables and effectively predict patient prognosis. In related research, a random forest model developed based on proteomics was used to predict the effectiveness of the CRC chemotherapy drug FOLFOX, achieving an accuracy rate exceeding 90%. Ahn et al. trained an algorithm for predicting CRC lymph node metastasis using the SEER database, achieving an accuracy rate of 96%. In addition, the least absolute shrinkage and selection operator (LASSO) was used to screen variables and found that the inflammatory marker-derived NLR (dNLR) and ALB were important predictors of postoperative infection in CRC patients. These findings suggest that machine learning algorithms based on inflammatory markers have significant potential in predicting the occurrence, progression, and postoperative outcome of CRC. However, further studies have not reported the construction of a prognostic prediction model for stage II-III CRC based on 59 new chronic inflammatory ratios or scores and the use of machine learning algorithms. Summary of the Invention

[0006] In order to solve the above problems, the purpose of the present invention is to provide a prognostic assessment technology for patients with stage II-III colorectal cancer, aiming to systematically analyze the relationship between inflammatory indicators and CRC prognosis, construct a prognostic model for recurrence-free survival and overall survival of patients with stage II-III CRC based on a machine learning algorithm, and comprehensively evaluate its performance to verify its effectiveness and practicality in clinical applications.

[0007] In order to achieve the above technical objectives, the present application provides a method for prognostic assessment of patients with stage II-III colorectal cancer, comprising the following steps:

[0008] Screen variables related to patient survival prognosis, obtain potential variables affecting patients' RFS and OS, evaluate the correlation between various indicators and the prognosis of colorectal cancer patients, and correct for confounding factors to screen indicators with important value for the model;

[0009] The above-screened indicators were used to construct RFS and OS prognostic models for CRC patients using machine learning algorithms, which were used to evaluate the prognosis of stage II-III CRC patients. Among them, the machine learning algorithms included LR, RF, XGB, SVC, MLP and GNB.

[0010] Preferably, when screening variables related to the patient's survival prognosis, the variables related to the patient's survival prognosis are screened out by combining Kaplan-Meier curve analysis with Log-rank test analysis.

[0011] Preferably, when obtaining potential variables that affect the patient's RFS and OS, univariate Cox regression analysis is used to determine the potential variables that affect the patient's RFS and OS.

[0012] Preferably, when evaluating the correlation between each indicator and the prognostic outcome of colorectal cancer patients, Cox multivariate regression analysis is used to correct the confounding factors and use the backward method to evaluate the correlation between each indicator and the prognostic outcomes RFS and OS of colorectal cancer patients.

[0013] Preferably, the hazard ratio (HR) and its 95% confidence interval (CI) are used to quantify the strength of the association between variables in both univariate and multivariate Cox regression analyses.

[0014] Preferably, when screening out indicators of great value to the model, the Pearson correlation coefficient is used to evaluate the strength and direction of the correlation between variables. The variables with low AUC values ​​were eliminated step by step.

[0015] Preferably, when screening out indicators of great value to the model, the LASSO method is used to compress the coefficients of variables with little impact on prognosis to 0, thereby screening out key variables with non-zero coefficients and constructing a new indicator, the chronic inflammation score CIS;

[0016] Preferably, when screening out indicators of great value to the model, based on the new indicator CIS and the basic clinical parameters and tumor marker indicators screened by the multivariate Cox regression model, SVM-RFE is used to further recursively eliminate unimportant variables, and finally screen out indicators of great value to the model.

[0017] Preferably, when screening out indicators of great value to the model, the RFS and OS prognostic models of CRC patients were constructed by six machine learning algorithms, and the area under the receiver operating characteristic curve (AUROC) and decision curve analysis (DCA) were used to evaluate and compare the effectiveness of each model in predicting prognosis. Among them, the LR model was used as the best machine learning algorithm when constructing the RFS and OS prognostic models of CRC patients.

[0018] The present invention discloses a prognostic evaluation system for patients with stage II-III colorectal cancer, which is used in the above-mentioned prognostic evaluation method for patients with stage II-III colorectal cancer, comprising:

[0019] The indicator screening module is used to screen variables related to patient survival prognosis, obtain potential variables affecting patients' RFS and OS, evaluate the correlation between various indicators and the prognosis of colorectal cancer patients, and correct for confounding factors to screen indicators with important value for the model;

[0020] The prognostic evaluation module is used to use the screened indicators and machine learning algorithms to construct RFS and OS prognostic models for CRC patients, which are used to evaluate the prognosis of stage II-III CRC patients. Among them, the machine learning algorithms include LR, RF, XGB, SVC, MLP and GNB.

[0021] The present invention discloses the following technical effects:

[0022] By integrating clinical data with machine learning technology, this invention provides an efficient and accurate tool for the prognosis assessment of patients with stage II-III colorectal cancer, providing a scientific basis for clinical decision support and the formulation of individualized treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 It is a schematic diagram of the research process of the present invention;

[0025] Figure 2 is a heat map of the correlation of the seven chronic inflammatory variables described in the present invention;

[0026] Figure 3 This is the LASSO result analysis of the inflammatory variables of the training cohort described in the present invention, wherein A: RFS cross-validation curve; B: coefficient path diagram of each RFS variable; C: OS cross-validation curve; D: coefficient path diagram of each OS variable;

[0027] Figure 4 It is the training cohort RFS analysis LASSO regression screening analysis described in the present invention;

[0028] Figure 5 is the LASSO regression screening analysis for OS analysis of the training cohort described in the present invention;

[0029] Figure 6 It is the feature importance ranking of CRC in RFS analysis described in the present invention;

[0030] Figure 7 It is the feature importance ranking of CRC in OS analysis described in the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application.

[0032] like Figure 1-Figure 7 As shown, the present invention provides a prognostic assessment technology for patients with stage II-III colorectal cancer. Based on the theoretical basis of the relationship between chronic inflammation and the occurrence and progression of CRC, the present invention combines machine learning technology to construct and evaluate a prognostic model for patients with stage II-III colorectal cancer. The present invention enrolled 1,829 patients. First, patients were screened according to inclusion and exclusion criteria. Peripheral blood samples were collected from patients before the initial intervention visit, and peripheral blood cell count, albumin, prealbumin, fibrinogen, and CEA and CA19-9 levels were promptly tested. Second, clinical parameters were collected, and three-year RFS and OS data were obtained through follow-up. Finally, the optimal cutoff value for each variable was calculated using X-tile software to determine the classification criteria for continuous variables. Univariate and multivariate Cox regression analyses were performed to assess the association between each variable and prognostic outcome, and potential prognostic factors were screened. Variables were further screened using LASSO regression analysis and support vector machine recursive feature elimination (SVM-RFE) to identify key prognostic indicators. Finally, based on the selected key variables, a prognostic model was constructed using a machine learning algorithm, and its performance was comprehensively evaluated. Specifically, the following aspects were included:

[0033] like Figure 1 As shown, the present invention adopts a prospective study design, aiming to systematically analyze the relationship between inflammatory indicators and CRC prognosis, construct a prognostic model for recurrence-free survival and overall survival in stage II-III CRC patients based on a machine learning algorithm, and comprehensively evaluate its performance to verify its effectiveness and practicality in clinical applications.

[0034] This invention included patients who underwent radical surgery for CRC for the first time from January 2015 to December 2020. The pathology reports showed that there were 2235 patients in stage II-III, of which 20 had other tumor diseases, 57 had trauma, renal failure, 202 had fever, and 45 had liver lesions. A total of 324 cases were excluded, and 82 cases were lost to follow-up within 3 months after surgery. The final study population was 1829 cases.

[0035] Based on the results of peripheral blood leukocyte, neutrophil, lymphocyte, monocyte, and platelet counts, and the results of fibrinogen, albumin, and prealbumin, we established 47 derived inflammatory ratio indices (neutrophil-lymphocyte ratio (NLR), lymphocyte-monocyte ratio (LMR), neutrophil-platelet ratio (NPR), platelet-leukocyte ratio (PWR), neutrophil-monocyte ratio (NMR), FPR-LMR ratio (FPMLR), and fibrinogen-lymphocyte ratio (FLR)) and four inflammatory scores (albumin-LMR score (SIS), modified albumin-LMR score (mSIS), neutrophil-platelet score (NPS), and fibrinogen-NLR score (F-NLR)) using calculation formulas. X-tile 3.6 software was used to calculate the cutoff value for each continuous newly detected variable based on patient survival data. The specific calculation formulas and cutoff values ​​for continuous variables are shown in Table 1. Peripheral blood lymphocyte count, albumin, prealbumin, LMR, albumin-lymphocyte sum (PNI), and albumin-fibrinogen ratio (AFR) values ​​above or below the cutoff values ​​were assigned a score of 0 and 1, respectively. SIS, mSIS, NPS, and F-NLR were scored as 0, 1, and 2, respectively, based on the cutoff values ​​for each parameter. The remaining markers were scored as 1 and 0, respectively, when they were above or below the cutoff values. Specific definitions of these indicator scores are shown in Table 1.

[0036] Table 1

[0037]

[0038]

[0039]

[0040] The optimal cutoff value was calculated according to the patient's RFS using X-tile software; WBC: white blood cell; Neu: neutrophil; Lym: lymphocyte; Mon: monocyte; PLT: platelet; Alb: albumin; pAlb: prealbumin; Fib: fibrinogen.

[0041] The present invention first screened for variables associated with patient survival prognosis using Kaplan-Meier curve analysis combined with the Log-rank test. Univariate Cox regression analysis was then used to identify potential variables influencing patients' RFS and OS. Furthermore, multivariate Cox regression analysis (using the backward method) was used to assess the association of various variables with prognostic outcomes (RFS and OS) in colorectal cancer patients. Confounding factors adjusted for included 12 factors, including gender, smoking history, history of diabetes, and lymph node metastasis. Both univariate and multivariate Cox regression analyses quantified the strength of association using the hazard ratio (HR) and its 95% confidence interval (CI). To identify variables with significant prognostic impact, the present invention employed the LASSO method to compress the coefficients of variables with minimal prognostic influence to zero, thereby identifying key variables with non-zero coefficients. Furthermore, support vector machine-recurrent entrapment analysis (SVM-RFE) was used to recursively eliminate insignificant variables, ultimately identifying indicators with significant model value. The strength and direction of correlations between variables were assessed using the Pearson correlation coefficient. The area under the receiver operating characteristic curve (AUROC) and decision curve analysis (DCA) were used to evaluate and compare the prognostic effectiveness of each model. For statistical description, dichotomous variables were presented as frequencies and percentages, and differences between groups were compared using the chi-square test or Fisher's exact test. Continuous variables were presented as mean ± standard deviation (SD), and comparisons between groups were performed using the Mann-Whitney U test and the Kruskal-Wallis H test. A p-value of <0.05 was used as the threshold for determining statistical significance.

[0042] Kaplan-Meier curves and Cox analysis: Based on the cutoff values ​​of each inflammatory indicator, ratio-based inflammatory indicators were divided into 0 and 1 points, while score-based inflammatory indicators were divided into 0, 1, and 2 points. A total of 12 clinical characteristic indicators and 61 laboratory test indicators (tumor markers and inflammatory indicators) were included in the study. The relationship between various indicators and the prognosis of CRC patients was systematically evaluated through Kaplan-Meier survival curve analysis and univariate and multivariate Cox regression analysis. Specifically, the RFS and OS of CRC patients were analyzed based on clinical characteristics analysis and experimental detection indicators. The Kaplan-Meier curve log-rank test and univariate and multivariate Cox analysis were used. Finally, the indicators related to patient prognosis were clinical characteristic parameters (lymph node metastasis, cell differentiation and tumor invasion depth), tumor markers (CEA, CA19-9) and 53 inflammatory parameters (platelets, albumin, LMR, MAR, FLR, FPR, FPSIRI, etc., including 6 original detection indicators, 3 SIS, mSIS and F-NLR scoring indicators, and 44 ratio indicators) which were associated with RFS and OS of stage II-III CRC patients.

[0043] Time-dependent ROC curve analysis: In addition to the above three scoring indicators, the 3-year RFS and OS of CRC patients were evaluated based on the AUC values ​​of the remaining 50 inflammation ratio indicators, and the indicators were ranked from high to low according to the AUC values; the Pearson correlation analysis method was used to systematically evaluate the relationship between the above 50 inflammation ratio variables: the indicators with high correlation (Pearson correlation coefficient ) were gradually eliminated. The operation screening steps are as follows: First, for the indicator FPR with the highest AUC value, the inflammatory ratios such as FPMLR (r=0.71), FPALR (r=0.72), PLPAR (r=0.70), FPPLR (r=0.76), NPAR (r=0.73), and PPAR (r=0.80) with high correlation with FPR were eliminated according to the preset threshold. Then, among the remaining variables, the indicator FPSIRI with the best diagnostic efficacy was screened out, and then the variables such as MPAR (r=0.59), MLPAR (r=0.75), FPNLR (r=0.77), and FPPLR (r=0.54) with extremely strong correlation with it were eliminated. Among the remaining indicators, the one with the highest AUC value was pAlb, and the indicators with high correlation coefficients with this indicator continued to be deleted. By iteratively executing this screening strategy, 7 independent inflammatory variables were finally obtained, and the correlation coefficients between these variables were , indicating that their correlations with each other are low, such as Figure 2 As shown, FPSIRI=FPR*SIRI=the product of the fibrinogen-prealbumin ratio and the systemic inflammatory response index, FPR represents the fibrinogen-prealbumin ratio, FLR represents the fibrinogen-lymphocyte ratio (equivalent to the FPR ratio multiplied by the SIRI index), MAR represents mixed antiglobulin reaction, LMR represents lymphocyte-monocyte ratio, Alb represents albumin, and Platelet represents platelet.

[0044] LASSO regression screening: Among the indicators eliminated in the above screening process, the indicator with the highest AUC value for clinical outcome (FPMLR), CRC prognosis independent scoring variables (SIS, mSIS and F-NLR) and 7 independent inflammatory indicators were selected as input variables. With clinical outcomes (RFS and OS) as dependent variables, the LASSO regression method was used for variable screening and optimization. As the penalty parameter gradually increased, the coefficient of the independent variable continued to decrease, and the number of independent variables screened also decreased accordingly, such as Figure 3 shown.

[0045] In the recurrence-free survival analysis of CRC patients, the LASSO regression screening results showed that the variables with greater correlation with RFS and their regression coefficients (CVs) were: 、 、 and ;like Figure 4 shown.

[0046] Based on the variables screened by LASSO regression, the present invention constructed a new indicator, chronic inflammation score (CIS):

[0047]

[0048] The formula for calculating FPR is Fib ÷ pAlb, the formula for FPMLR is (Fib ÷ pAlb) × (Mon ÷ Lym), and F-NLR is a scoring indicator composed of Fib and Neu / Lym. In a training population, the present invention divided CRC patients into a recurrence group and a non-recurrence group based on whether they relapsed. The recurrence group consisted of 253 patients and the non-recurrence group consisted of 736 patients. Clinical laboratory results were compared between the two groups. The results showed that no statistically significant differences were found in the single indicators neutrophils (p=0.15) and lymphocytes (p=0.13) between the two groups. In contrast, levels of monocytes (p<0.001), fibrinogen (p<0.001), FPR (p<0.001), and FPMLR (p<0.001) were significantly higher in the recurrence group than in the non-recurrence group, while the proportion of patients with high F-NLR scores was significantly lower in the recurrence group (p<0.001). In addition, the albumin (p < 0.001) and prealbumin (p < 0.001) levels in the relapse group were significantly lower than those in the non-relapse group (Table 2).

[0049] Table 2

[0050]

[0051] In the study of overall survival of colorectal cancer, after Lasso regression variable screening, the variables with more significant relationship with OS and their CVs are: 、 、 and The four features and their corresponding coefficient values ​​show that FPR has the largest coefficient and is most important to the OS model. Figure 5 shown.

[0052] The LASSO variables were selected and constructed:

[0053]

[0054] in, , A total of 989 patients were included in the training cohort of this invention. After three years of follow-up, 148 patients died and 841 patients survived. Patients were divided into a death group and a survival group based on whether they died. The results showed that the death group had significantly higher levels of circulating serum neutrophils (p < 0.05), monocytes (p < 0.05), fibrinogen (p < 0.001), FLR (p < 0.001), FPR (p < 0.001), FPMLR (p < 0.001), and FPSIRI (p < 0.001) than the survival group. Concomitantly, the serum lymphocyte count (p = 0.03) and prealbumin concentration (p < 0.001) of patients in the death group were significantly lower than those in the survival group (Table 3).

[0055] Table 3

[0056]

[0057] SVM-RFE analysis:

[0058] Based on the new inflammation index CIS constructed above, this risk score was then incorporated into the SVM-RFE analysis along with pathological parameters that showed significant differences in multivariate Cox regression analysis [tumor invasion depth (TID), lymph node metastasis (LNM), cell differentiation (CellDiff)] and tumor markers (CEA, CA19-9). The importance of the features was ranked, and the final selected indicators were incorporated into the construction of the machine learning model. In the model for predicting RFS of CRC patients, the feature importance ranking was: LNM (1.10) > CIS (1.01) > > > > like Figure 6 shown.

[0059] In the model for predicting OS in CRC patients, the feature importance (absolute value) is ranked as follows: > > > > > like Figure 7 shown.

[0060] Analysis of the prognostic relationship between CIS and CRC: X-tile 3.6 software was used to determine the optimal cutoff values ​​for CIS in stage II-III CRC patients for RFS and OS analysis, which were -0.51 (RFS) and -1.40 (OS), respectively. Kaplan-Meier curves and log-rank tests were used to analyze CIS (RFS: p < 0.001; OS: p < 0.001).

[0061] Univariate Cox analysis showed CIS (RFS: adjusted HR=4.52, 95%CI=3.45-5.93; OS: adjusted HR=5.00, 95%CI=3.53-7.09). The study group further adjusted for clinical parameters such as gender, age, smoking, drinking, and diabetes by Cox multivariate correction, CIS (RFS: adjusted HR=4.24, 95%CI=3.19-5.64; OS: adjusted HR=4.36, 95%CI=3.04-6.25).

[0062] Construction of RFS model for CRC patients: Based on the above-mentioned screened variables, the present invention includes CIS, CEA, CA19-9, LNM, CellDiff and TID in the training cohort, and uses six machine learning algorithms, LR, RF, XGB, SVC, MLP and GNB, to construct the RFS prognosis model for CRC patients. In order to avoid the impact of uneven data division on model performance, the present invention adopts a 10-fold cross-validation method. Specifically, the data is divided into 10 subsets. In each cross-validation, one subset is used as validation data, and the remaining nine subsets are used for model training. The process is cross-validated a total of 10 times to ensure that each subset has the opportunity to be used as a validation set. In this way, the constructed model will not be overly dependent on part of the data set, thereby effectively avoiding overfitting and improving the generalization ability and stability of the model.

[0063] CRC OS model construction: Based on the six variables ultimately screened—CIS, CEA, CA19-9, LNM, CellDiff, and TID—the present invention employed six machine learning algorithms to construct prognostic models for CRC OS. To avoid overfitting due to uneven data partitioning, the present invention employed a 10-fold cross-validation method to optimize the generalization ability of the six models in the validation cohort.

[0064] Determination of the Optimal Model: A comprehensive evaluation of the six models for predicting RFS and OS in CRC patients in both internal and external validation cohorts revealed that the LR and GNB models performed particularly well. In the internal validation cohort analysis of RFS, the DCA curves for the LR and GNB models demonstrated optimal predictive performance within the ranges of 0.10-0.80 and 0.10-0.70, respectively. In the external validation cohort analysis, the DCA curves for the two models demonstrated optimal performance within the ranges of 0.25-0.65 and 0.2-0.55, respectively. In the internal validation cohort analysis of OS, the DCA curves for the LR and GNB models demonstrated robust performance within the ranges of 0.10-0.50 and 0.10-0.35, respectively. In the external validation cohort analysis, the DCA curves for the LR model demonstrated optimal performance within the range of 0.10-0.80, while the DCA curves for the GNB model demonstrated satisfactory performance within the range of 0.10-0.50. In summary, the LR model demonstrated excellent predictive performance and clinical applicability, demonstrating high clinical utility. Therefore, the present invention determines that the LR model is the optimal model.

[0065] A bar chart visually illustrates the contribution of each feature to the model. Specifically, the absolute coefficient value of each feature in the LR model was calculated and ranked accordingly. The results showed that lymph node metastasis contributed the most to the LR model in predicting RFS and OS, while the CIS fit index ranked second.

[0066] In addition, the present invention constructed a prognostic LR model that included chemoradiotherapy (CRT) regimen as an independent variable. In the RFS analysis, the AUC values ​​of the LR model for the internal validation cohort and the external validation cohort were 0.79 (0.79-0.80) and 0.69 (0.68-0.69), respectively. In the OS analysis, the AUC values ​​for the internal validation cohort and the external validation cohort were 0.79 (0.78-0.80) and 0.73 (0.72-0.73), respectively. Comparative analysis with RFS and OS prediction models that did not include CRT regimen showed that the inclusion of CRT regimen had no significant impact on the predictive efficacy of the prognostic model.

[0067] Evaluation of machine learning models: Evaluation is carried out through accuracy, precision, recall, and specificity. Accuracy is also the ratio of the number of correctly predicted patient cases to the total number of cases. It is used to evaluate the closeness of the model's prediction of the correct result. The higher the accuracy, the closer it is to the true value, avoiding overdiagnosis of patients. The higher the precision, the better the accuracy of the model in predicting the proportion of patients who are actually sick, and the lower the proportion of healthy people being misdiagnosed as patients. The higher the recall rate, the more complete the coverage of the machine learning model, and the lower the probability of missed diagnosis of patients. The higher the specificity, the lower the probability of the model misdiagnosing healthy people as patients.

[0068] The CIS designed by the present invention has a significantly better efficacy in predicting patient prognosis than any other single indicator, and this indicator is an independent prognostic marker for CRC patients; the present invention successfully constructed six machine learning CRC prognosis prediction models based on chronic inflammation and clinical characteristic parameters, among which the LR prediction model constructed based on indicators such as chronic inflammation has a better efficacy in evaluating patient RFS and OS, and can effectively predict patient prognosis.

[0069] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0070] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0071] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for prognostic assessment of patients with stage II-III colorectal cancer, characterized in that: The following steps are involved: Screen variables related to patient survival prognosis, obtain potential variables affecting patients' RFS and OS, evaluate the correlation between various indicators and the prognosis of colorectal cancer patients, and correct for confounding factors to screen indicators with important value for the model; The above-screened indicators are used to construct RFS and OS prognostic models for CRC patients using machine learning algorithms, which are used to evaluate the prognosis of stage II-III CRC patients. The machine learning algorithms include LR, RF, XGB, SVC, MLP and GNB.

2. The method for prognosis assessment of patients with stage II-III colorectal cancer according to claim 1, characterized in that: When screening variables related to patient survival prognosis, Kaplan-Meier curve combined with Log-rank test analysis was used to screen out variables related to patient survival prognosis.

3. A method for prognosis assessment for patients with stage II-III colorectal cancer according to claim 2, characterized in that: When obtaining the potential variables affecting patients' RFS and OS, univariate Cox regression analysis was used to determine the potential variables affecting patients' RFS and OS.

4. A method for prognosis assessment for patients with stage II-III colorectal cancer according to claim 3, characterized in that: When evaluating the correlation between various indicators and the prognostic outcomes of colorectal cancer patients, Cox multivariate regression analysis was used to adjust the confounding factors and use the backward method to evaluate the correlation between various indicators and the prognostic outcomes RFS and OS of colorectal cancer patients.

5. A method for prognosis assessment for patients with stage II-III colorectal cancer according to claim 4, characterized in that: In both univariate and multivariate Cox regression analyses, the hazard ratio (HR) and its 95% confidence interval (CI) were used to quantify the strength of the association between the variables.

6. A method for prognosis assessment for patients with stage II-III colorectal cancer according to claim 5, characterized in that: When screening out indicators of great value to the model, the Pearson correlation coefficient was used to evaluate the strength and direction of the correlation between variables, and variables with a Pearson correlation coefficient |r| > 0.5 were gradually eliminated, and variables with low AUC values ​​were eliminated.

7. A method for prognosis assessment for patients with stage II-III colorectal cancer according to claim 6, characterized in that: When screening out indicators of great value to the model, the LASSO method was used to compress the coefficients of variables with little impact on prognosis to 0, thereby screening out key variables with non-zero coefficients and constructing a new indicator, the chronic inflammation score CIS.

8. A method for prognosis assessment of patients with stage II-III colorectal cancer according to claim 7, characterized in that: When screening out indicators of great value to the model, based on the new indicator CIS and the basic clinical parameters and tumor marker indicators screened by multivariate Cox, SVM-RFE was used to further recursively eliminate unimportant variables, and finally screen out indicators of great value to the model.

9. A method for prognosis assessment for patients with stage II-III colorectal cancer according to claim 8, characterized in that: When screening out indicators of great value to the model, six machine learning algorithms were used to construct RFS and OS prognostic models for CRC patients. The area under the receiver operating characteristic curve (AUROC) and decision curve analysis (DCA) were used to evaluate and compare the effectiveness of each model in predicting prognosis. Among them, the LR model was used as the best machine learning algorithm when constructing the RFS and OS prognostic models for CRC patients.

10. A prognostic evaluation system for patients with stage II-III colorectal cancer, used to implement a prognostic evaluation method for patients with stage II-III colorectal cancer according to any one of claims 1 to 9, characterized in that: include: The indicator screening module is used to screen variables related to patient survival prognosis, obtain potential variables affecting patients' RFS and OS, evaluate the correlation between various indicators and the prognosis of colorectal cancer patients, and correct for confounding factors to screen indicators with important value for the model; The prognostic evaluation module is used to use the screened indicators and machine learning algorithms to construct RFS and OS prognostic models for CRC patients, which are used to perform prognostic evaluation on stage II-III CRC patients. The machine learning algorithms include LR, RF, XGB, SVC, MLP and GNB.

Images