Prognostic scoring method and system for predicting advanced biliary tract malignant tumor immunotherapy

Abstract

The invention discloses a prognostic scoring method and system for predicting advanced biliary tract malignant tumor immunotherapy, and belongs to the technical field of advanced biliary tract malignant tumor immunotherapy. The method comprises the following steps: firstly, generating random replacement copies for real variables to obtain shadow variables, and classifying the real variables into related classes by comparing importance scores of the random replacement copies and the shadow variables; real variables in related classes are screened based on LASSO regression coefficients, and a variable set is formed; combining the variables in the set, and screening out a variable combination corresponding to a minimum AIC value model according to an AIC criterion to obtain a prognosis key variable set; calculating an index score according to the COX regression coefficient of each real variable; and finally, adding the index scores of the variables in the prognosis key variable set of each patient to obtain a prognosis BICCAPS score. Through three-step progressive screening of the random forest algorithm, the LASSO and the optimal subset method, the problems of variable screening, colinearity and model conciseness in high-dimensional clinical data modeling are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of immunotherapy for advanced biliary tract cancer, and in particular to a prognosis scoring method and system for predicting immunotherapy for advanced biliary tract cancer. BACKGROUND

[0002] Biliary tract cancer (BTC) is a highly malignant tumor with a very poor prognosis. Due to its concealed anatomical location and atypical early symptoms, about 60%-70% of patients have developed unresectable locally advanced or distant metastasis by the time of diagnosis. For patients with advanced BTC, treatment options are very limited, and the survival status is not optimistic. In recent years, the application of immune checkpoint inhibitors (ICIs) has brought a major breakthrough in the treatment of advanced BTC. Based on the positive results of TOPAZ-1 and KEYNOTE-966 phase III clinical studies, durvalumab or pembrolizumab combined with gemcitabine and cisplatin has become a new standard for first-line treatment of advanced BTC.

[0003] However, the application of immunotherapy in BTC still faces great challenges. First, the survival benefit brought by ICI combined with chemotherapy is still limited, with a median overall survival of only 1-2 months. Second, there is significant individual difference in treatment response, only a part of patients can benefit from long-term treatment, while most patients not only cannot benefit from treatment, but also need to bear high treatment costs and potential risks of immune-related adverse reactions. Therefore, how to accurately identify the dominant population that may benefit from immunotherapy before treatment and avoid unnecessary treatment for ineffective patients has become a key problem in clinical practice.

[0004] Currently, the biomarkers commonly used to predict the efficacy of immunotherapy in clinical practice mainly include PD-L1 expression, microsatellite instability (MSI) and tumor mutation burden (TMB). However, the application of these markers in BTC has obvious limitations: the predictive value of PD-L1 expression is controversial, and the detection results are affected by tumor heterogeneity, detection antibodies and interpretation criteria; MSI-H / dMMR has a very low incidence in BTC (only about 2-5%); TMB detection is costly and the critical value standard is not uniform. More importantly, the detection of these markers depends on tumor tissue samples, which need to be obtained through invasive biopsy, and has low applicability for patients with poor general condition or difficult to puncture tumor site. The above limitations seriously restrict the wide application of these markers in clinical practice.

[0005] In recent years, inflammation, nutrition and liver function indicators based on routine blood tests and biochemical tests have received extensive attention due to their non-invasive, convenient, low cost and dynamic monitoring characteristics. Previous studies have shown that neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), systemic immune inflammation index (SII), prognostic nutritional index (PNI) and albumin-bilirubin score (ALBI) are correlated with the immunotherapy effect of various tumors (such as lung cancer and liver cancer). In the field of BTC, existing studies are mostly limited to the prognostic value of a single indicator, and there are problems such as small sample size, single statistical analysis method, and lack of independent verification cohort. For example, the study by Yang et al. only verified the prognostic value of PNI, and although Wang et al. evaluated 13 inflammatory indicators, they did not integrate them into a practical clinical tool. The predictive performance of these single indicators is limited (AUC is mostly less than 0.7), and the stability is insufficient, which cannot meet the needs of clinical precision medicine.

[0006] Recently, Persano et al. tried to build a multi-indicator prediction model called MAGIC-D index, which represents a progress in this field. However, this study only showed the preliminary results of the model in the training set, and only showed the stratified survival difference of patients in the external validation cohort, lacking model performance evaluation indicators (such as AUC value, C index, calibration curve, etc.), which limits the reliability of the model. SUMMARY

[0007] In view of the above problems in the prior art, the present application provides a method and system for predicting the prognosis of immunotherapy for advanced biliary tract cancer, which solves the following problems: First, the prediction accuracy is limited: relying on a single indicator cannot fully reflect the complex immune inflammation, nutrition and liver function status of the host, the information dimension is insufficient, resulting in low discrimination and accuracy of the prediction model (AUC is usually <0.7), which makes it difficult to reliably guide clinical decision-making.

[0008] Second, the clinical practicability and accessibility are poor: existing histology-based marker detection requires invasive procedures, is high in cost and long in cycle, and some patients cannot obtain qualified tissue specimens, which limits its clinical application. Existing hematological index research mostly stays at the academic level and has not developed into a simple and clear clinical tool.

[0009] Third, the statistical method has limitations: many studies directly include multiple inflammation / nutrition indicators with collinearity in traditional Cox regression when building multi-factor models, which may lead to unstable models, incorrect exclusion of important variables, and unreliable conclusions.

[0010] To achieve the above application purposes, the technical scheme adopted by the present application is as follows: a method for predicting the prognosis of immunotherapy for advanced biliary tract cancer, comprising the following steps: S1, generating a random permutation copy of the real variable to obtain a shadow variable, comparing the importance scores of the real variable and the shadow variable, classifying the real variable to obtain a relevant class; S2, screening the real variables in the relevant class based on the LASSO regression coefficients to obtain a variable set; S3, combining the real variables in the variable set, screening the variable combination corresponding to the minimum AIC value of the model based on the AIC criterion to obtain a prognosis key variable set; S4, calculating the index score of each real variable according to the COX regression coefficient of each real variable; S5, adding the index scores of each real variable in the prognosis key variable set of each patient to obtain a prognosis BICCAPS score.

[0011] Further, S1 includes the following steps: S11, generating a random permutation copy of the real variable to obtain a shadow variable; S12, calculating the importance score of the real variable and the shadow variable for predicting the target by using the random forest algorithm; S13, screening the shadow variable with the highest importance score from all shadow variables to obtain the best shadow variable; S14, determining whether the importance score of the real variable is significantly higher than the importance score of the best shadow variable by Z-test, if yes, marking as a "relevant class", if no, marking as an "irrelevant class" or a "tentative class"; S15, repeating steps S11-S14 until all real variables are classified or the maximum number of iterations is reached, to obtain the relevant class.

[0012] Further, the standard for being significantly higher in S14 is: p<0.01, p is the probability value.

[0013] Further, S2 includes the following steps: S21, introducing a penalty coefficient in the LASSO regression, and compressing the LASSO regression coefficients of the real variables in the relevant class to 0; S22, taking each real variable with a LASSO regression coefficient not equal to 0 as an element to form a variable set.

[0014] Further, S3 includes the following steps: S31, combining each real variable in the variable set to obtain a plurality of variable combinations; S32, using the Akaike information criterion as the model evaluation standard to calculate the AIC value of the model constructed by each variable combination; S33, find the variable combination corresponding to the model with the minimum AIC value, and obtain the prognosis key variable set.

[0015] Further, the elements in the prognosis key variable set include: the number of ICI treatment lines, CA19-9, CEA, peritoneal metastasis, SIRI, PNI and total bilirubin.

[0016] Further, S4 includes the following sub-steps: S41, using the COX regression algorithm to analyze the COX regression coefficient of each real variable in the prognosis key variable set; S42, taking the COX regression coefficient corresponding to the total bilirubin unit quantization interval as the reference coefficient; S43, using the COX regression coefficient of each real variable divided by the reference coefficient of the corresponding real variable, and taking the integer part, to obtain the index score of the corresponding real variable.

[0017] Further, the value of the reference coefficient in S42 is 0.198.

[0018] A prognosis scoring system for predicting the immunotherapy of advanced biliary tract malignant tumors, comprising: a first variable screening unit, a second variable screening unit, a third variable screening unit, an index score calculation unit and a prognosis scoring calculation unit; The first variable screening unit is used for generating a random permutation copy of the real variable to obtain a shadow variable, comparing the importance scores of the real variable and the shadow variable, classifying the real variable to obtain a related class; The second variable screening unit is used for screening the real variables in the related class based on the LASSO regression coefficient to obtain a variable set; The third variable screening unit is used for combining the real variables in the variable set, screening the variable combination corresponding to the model with the minimum AIC value based on the AIC criterion, and obtaining a prognosis key variable set; The index score calculation unit is used for calculating the index score of each real variable according to the COX regression coefficient of each real variable; The prognosis scoring calculation unit is used for adding the index scores of each real variable in the prognosis key variable set of each patient to obtain a prognosis BICCAPS score.

[0019] The beneficial effects of the present application are: 1. This invention generates shadow variables by creating random permutations of real variables, classifies them by comparing their importance scores, and obtains related class variables, thus including as many variables as possible related to the prognosis of immunotherapy in the consideration. Subsequently, based on LASSO regression coefficients, related class real variables are screened to further remove redundant information and retain key variables. Finally, the model corresponding variable combination with the lowest AIC value is selected using the AIC criterion to obtain the set of key prognostic variables. This series of operations achieves the extraction of information from multiple key variables, improving the information dimensionality.

[0020] 2. The prognostic scoring method of the present invention is based entirely on inflammation, nutrition and liver function indicators from peripheral blood routine tests and biochemical examinations. It does not require invasive tissue biopsy, thus successfully avoiding the risks of puncture. Furthermore, the method has low cost and short cycle per test, making it convenient for clinical use.

[0021] 3. This invention employs a three-stage process to construct the CoX model: a random forest algorithm (for comprehensive initial screening to avoid the erroneous removal of important variables in traditional multifactor CoX regression analysis due to collinearity), followed by LASSO regression for dimensionality reduction and collinearity removal, and finally, the Akaike information criterion for selecting the optimal variable combination. This ensures the robustness and simplicity of the final CoX model.

[0022] 4. The traditional Cox regression model is an abstract and complex formula that is not convenient for clinical application. Even when converted into a nomogram, it remains too complex for clinical use. This invention transforms the complex and abstract model into simple and easy-to-use index scores, successfully converting the complex statistical model into a clinical scoring tool that doctors can directly use. Attached Figure Description

[0023] Figure 1 A flowchart of a prognostic scoring method for predicting the prognosis of immunotherapy for advanced biliary tract malignancies; Figure 2 Kaplan-Meier curves for progression-free survival (PFS) in patients from different risk groups based on BICCAPS scores; Figure 3 Kaplan-Meier curves for overall survival (OS) of patients in different risk groups based on BICCAPS scores. Detailed Implementation

[0024] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0025] The prognosis of advanced biliary tract cancer (BTC) is extremely poor. Immune checkpoint inhibitors (ICIs) combined with chemotherapy have become the standard first-line treatment, but patients have limited survival benefits, heterogeneous response rates, and risks of immune-related adverse reactions and economic burdens. Currently, there is a lack of tools that can effectively and conveniently predict patient prognosis in clinical practice. Existing predictive markers (such as PD-L1 expression and microsatellite instability) have defects such as invasive detection, high cost, low positive rate, or limited predictive value, and cannot be widely used in clinical practice. Therefore, the present application aims to solve the technical problem of how to non-invasively, quickly and accurately stratify the risk and predict the prognosis of advanced biliary tract cancer (BTC) patients receiving immune checkpoint inhibitors (ICIs) treatment, and provide key evidence for treatment decisions.

[0026] The purposes of the present application include: providing a prognosis scoring method and system with strong comprehensiveness and high prediction accuracy, constructing a stable and reliable prediction model by integrating easily accessible indicators in multiple dimensions such as inflammation, nutrition, liver function, and tumor burden; converting the model into an integer-based risk stratification tool that can be quickly calculated at the bedside, greatly improving clinical practicability and accessibility; using advanced machine learning algorithms (random forest algorithm, LASSO regression algorithm) combined with traditional statistical models for variable selection and model construction to ensure that the selected variables are both important and stable, overcoming the limitations of traditional methods.

[0027] In the selection of characteristic variables of the prediction model, since each analysis indicator actually represents the body's inflammation / nutrition / liver function to varying degrees, there is some overlap in the calculation formula, and therefore there is a certain degree of multicollinearity between these inflammation / nutrition / liver function indicators. Therefore, traditional methods (single factor analysis screening followed by multivariate COX regression) may miss key variables or produce unstable hazard ratio estimates when multicollinearity exists, and the results obtained are not reliable. Therefore, to solve the problem of multicollinearity between multiple inflammation, nutrition, and liver function indicators and ensure that the most stable and important prognostic factors are selected, the present application uses a multi-stage, progressive variable selection and model optimization strategy to ensure the robustness of the model while minimizing its complexity in clinical application.

[0028] English abbreviations, English full names and Chinese corresponding relationships of the present application: AIC (Akaike Information Criterion, Akaike information criterion), ALBI (Albumin-Bilirubin Score, albumin-bilirubin score), ALI (Advanced Lung Cancer Inflammation Index, advanced lung cancer inflammation index), AUC (Area Under the Curve, area under the curve), BTC (Biliary Tract Cancer, biliary tract cancer), CA19-9 (Carbohydrate Antigen 19-9, carbohydrate antigen 19-9), CEA (Carcinoembryonic Antigen, carcinoembryonic antigen), COX (Cox Proportional Hazards Model, Cox proportional hazards model), dMMR (Mismatch Repair Deficiency, mismatch repair deficiency), HR (Hazard Ratio, hazard ratio), ICI (Immune Checkpoint Inhibitor, immune checkpoint inhibitor), LASSO (Least Absolute Shrinkage and Selection Operator, least absolute shrinkage and selection operator), LIPI (Lung Immune Prognostic Index, lung immune prognostic index), LP (Linear Predictor, linear predictor), MSI (Microsatellite Instability, microsatellite instability), MSI-H (Microsatellite Instability-High, microsatellite instability-high), NLR (Neutrophil-to-Lymphocyte Ratio, neutrophil-to-lymphocyte ratio), OS (Overall Survival, overall survival), PD-L1 (Programmed Death-Ligand 1, programmed death-ligand 1), PFS (Progression-Free Survival, progression-free survival), PLR (Platelet-to-Lymphocyte Ratio, platelet-to-lymphocyte ratio), PNI (Prognostic Nutritional Index, prognostic nutritional index), SII (Systemic Immune-inflammation Index,SIRI (Systemic Immune-inflammation Response Index), TMB (Tumor Mutational Burden).

[0029] As shown in Embodiment 1, a method for predicting a prognosis score of immunotherapy of advanced biliary tract malignant tumor, comprising the following steps: Figure 1 S1, generating a random permutation copy of the real variable to obtain a shadow variable, comparing the importance scores of the real variable and the shadow variable, classifying the real variable to obtain a relevant class; S2, screening the real variable in the relevant class based on the LASSO regression coefficient to obtain a variable set; S3, combining the real variable in the variable set, and screening the variable combination corresponding to the minimum AIC value based on the AIC criterion to obtain a prognosis key variable set; S4, calculating the index score of each real variable according to the COX regression coefficient of each real variable; S5, adding the index scores of each real variable in the prognosis key variable set of each patient to obtain a prognosis BICCAPS score.

[0030] In this embodiment, S1 comprises the following steps: S11, generating a random permutation copy of the real variable to obtain a shadow variable; S12, calculating the importance score of the real variable and the shadow variable for predicting the target by using the random forest algorithm; S13, screening the shadow variable with the highest importance score from all shadow variables to obtain the best shadow variable; S14, judging whether the importance score of the real variable is significantly higher than that of the best shadow variable by Z test, if yes, marking as a “relevant class”, if not, marking as an “irrelevant class” or a “tentative class”; S15, repeating steps S11-S14 until all real variables are classified or the maximum number of iterations is reached, to obtain a relevant class.

[0031] In this embodiment, the standard for being significantly higher in S14 is: p<0.01, p is the probability value.

[0032] ​The specific implementation process of S1 is as follows: in this embodiment, a random permutation copy (named as a “shadow variable”) of the original real variable is created, and the importance scores of the real variable and the shadow variable on the prediction target (total survival time OS in this case) are compared. First, the random forest importance scores of all real variables and shadow variables are calculated, for each real variable, a statistical test (Z test) is performed, if the importance is significantly higher than that of the best-performing shadow variable (default p<0.01), it is marked as “relevant”; otherwise, it is marked as “irrelevant” or “tentative”, and iteration is performed until all real variables are classified or the maximum number of iterations is reached. This method can avoid the important variables being mistakenly excluded in the traditional multivariate COX regression analysis due to collinearity.

[0033] In this embodiment, S2 includes the following steps: S21, introducing a penalty coefficient in LASSO regression, and compressing the LASSO regression coefficient of the real variable in the relevant class to 0; S22, each real variable with a LASSO regression coefficient not equal to 0 is taken as an element to form a variable set.

[0034] The specific implementation process of S2 is as follows: LASSO regression is used for fine screening. LASSO regression is a linear regression with an L1 penalty term. By introducing a penalty coefficient λ, the coefficients of unimportant variables are compressed to 0, realizing automatic variable selection and collinearity control, solving the collinearity problem while retaining a subset of variables, and improving the model generalization ability. LASSO regression is performed on the variables preliminarily screened by the random forest algorithm, and a more compact and stable variable set is obtained.

[0035] In this embodiment, S3 includes the following steps: S31, combining each real variable in the variable set to obtain a plurality of variable combinations; S32, using Akaike information criterion as the model evaluation standard, calculating the AIC value of the model constructed by each variable combination; S33, finding the variable combination corresponding to the model with the minimum AIC value to obtain a set of prognostic key variables.

[0036] In this embodiment, the elements in the set of prognostic key variables include: the number of ICI treatment lines, CA19-9, CEA, peritoneal metastasis, SIRI, PNI and total bilirubin.

[0037] The specific implementation process of S3 is: optimal model determination based on AIC criterion (i.e. optimal subset method). This is a key improvement step introduced by the present application to pursue the optimal balance between model simplicity and prediction performance. After the first two rounds of screening, the number of variables has been greatly reduced, but to ensure the bedside application friendliness of the final model, the "optimal subset method" is used to traverse all possible combinations of the remaining variables. The Akaike information criterion is used as the evaluation standard, and the model with the smallest AIC value is automatically selected. The AIC criterion balances the complexity of the model (the number of variables) and the goodness of fit, and selecting the model with the smallest AIC means finding a "best" model that is both simple and effective.

[0038] In the present embodiment, S4 includes the following steps: S41, using COX regression algorithm to analyze the COX regression coefficient of each real variable in the prognosis key variable set; S42, taking the COX regression coefficient corresponding to the total bilirubin unit quantization interval as the reference coefficient; S43, dividing the COX regression coefficient of each real variable by the reference coefficient corresponding to the real variable, and rounding, to obtain the index score of the corresponding real variable.

[0039] In the present embodiment, the value of the reference coefficient in S42 is 0.198, and the total bilirubin unit quantization interval in S42 is 10 μmol / L The specific implementation process of S4 is: converting continuous regression coefficients into integer scores convenient for clinical use. The specific conversion formula is: the index score is equal to the regression coefficient of the index divided by the selected reference coefficient and rounded. In the present embodiment, the regression coefficient 0.198 corresponding to "total bilirubin increased by 10 μmol / L per liter" is selected as the reference coefficient for obtaining 1 point. According to this algorithm, the specific scores of each variable in the BICCAPS scoring table are finally determined: 4 points for ICI treatment line number (0.875 divided by 0.198); 3 points for CA19-9 higher than 181 U / mL (0.647 divided by 0.198); 2 points for CEA higher than 4.69 ng / mL (0.405 divided by 0.198); 3 points for peritoneal metastasis (0.548 divided by 0.198); 1.52×10 91 point for every 10 pmol / L of total bilirubin. With the mean value of total bilirubin in this cohort being 15 pmol / L, 0 points were given for ≤15 pmol / L, 1 point for 15-25 pmol / L, 2 points for 25-35 pmol / L, 3 points for 35-45 pmol / L, 4 points for 45-55 pmol / L, and 5 points for >55 pmol / L.

[0040] The scores of each predictor are shown in Table 1.

[0041] Table 1

[0042] Then, a risk stratification and clinical application system was established. The total BICCAPS score was obtained by adding the scores of the seven indicators of each patient. The relationship between the total BICCAPS score of the patient and the linear predictor LP was approximately LP = BICCAPS total score.

[0043] Based on the COX baseline survival function S0(t), the absolute survival probability of the individual can be obtained: S(t|X) = S0(t)exp(LP) = S0(t)exp(BICCAPS total score), where S(t|X) is the absolute survival probability of the individual, exp is the exponential function, and the baseline survival function S0(t) is obtained from the reference patients with all risk factors at low risk level and total bilirubin ≤15 pmol / L. According to the total score, patients were divided into three risk levels: low risk group (0-6 points, corresponding to a relative risk of about 1.00-2.65 times), with a median overall survival of 17.9 months; medium risk group (7-10 points, about 2.97-6.10 times), with a median overall survival of 10.2 months; high risk group (≥11 points, ≥7.42 times), with a median overall survival of 6.6 months. The survival difference between different risk groups was statistically significant, with a p value less than 0.001. Before the patient starts immunotherapy, the clinician can stratify the risk according to the baseline clinical data of the patient, calculate the BICCAPS total score by referring to the scoring table, and stratify the risk. This result can be used to predict the prognosis of the patient and provide an objective reference for treatment decision-making, such as strengthening doctor-patient communication, actively providing supportive treatment, or considering participating in new drug clinical trials for high-risk patients.

[0044] Table 2 presents the results of the multi-factor COX regression analysis, showing the association of different predictive factors (including the number of ICI treatment lines, CA19-9 level, CEA level, peritoneal metastasis, SIRI, PNI, total bilirubin) with the prognosis of patients. Through the hazard ratio (HR), the COX regression coefficient β and the corresponding p value, the influence degree and statistical significance of each predictive factor on the survival outcome of patients are reflected. For example, multi-line ICI treatment, CA19-9 level > 181 U / mL, CEA level > 4.69 ng / mL, presence of peritoneal metastasis, SIRI > 1.52 x 10 9 / L, and high total bilirubin increase the risk of poor prognosis of patients, while PNI > 45.6 reduces the risk of poor prognosis, and the association of each factor has corresponding statistical significance (p value reflects).

[0045] Table 2

[0046] In this embodiment, on the basis of the core seven indicators (number of ICI treatment lines, CA19-9, CEA, peritoneal metastasis, SIRI, PNI and total bilirubin), other verified inflammatory, nutritional or liver function indicators (such as ALI, ALBI, NLR, LIPI, etc.) with prognostic value can be added to fine-tune and optimize the model, or the weight of the indicators can be adjusted for different populations (such as different regions, races) to form variants or derivative versions of the scoring system. In addition, the present application does not distinguish between first-line and subsequent line treatment populations, and it may have better results to construct separate prognostic models according to the number of treatment lines. In terms of calculation method, the risk score can not strictly use "total bilirubin per 10 μmol / L" as the reference benchmark, but use the coefficient of other variables as the reference coefficient; the critical value of risk stratification can also be adjusted according to different clinical practice needs or larger data sets, such as redefining the cut-off points of low, medium and high risk using machine learning clustering methods, or subdividing into more risk levels. In terms of application scenarios, the construction ideas and methodologies of the scoring system can be extended to the prognosis prediction of other treatment methods (such as targeted therapy, chemotherapy) for biliary malignant tumors, or after adaptive adjustment, applied to the prognosis prediction of other immune therapy sensitive cancer types (such as liver cancer, gastric cancer, lung cancer).

[0047] Embodiment 2, a prognostic scoring system for predicting the prognosis of immunotherapy for advanced biliary malignant tumors, comprising: a first variable screening unit, a second variable screening unit, a third variable screening unit, an indicator score calculation unit and a prognostic score calculation unit; The first variable screening unit is used to generate a random permutation copy of the real variable to obtain a shadow variable, compare the importance scores of the real variable and the shadow variable, classify the real variable, and obtain related classes. The second variable screening unit is configured to screen real variables in the relevant class based on LASSO regression coefficients to obtain a variable set; The third variable screening unit is configured to combine real variables in the variable set, screen a variable combination corresponding to a model with a minimum AIC value based on an AIC criterion, and obtain a prognosis key variable set. The index score calculation unit is configured to calculate an index score of each real variable according to a COX regression coefficient of the real variable. The prognosis score calculation unit is configured to add the index scores of the real variables in the prognosis key variable set of each patient to obtain a prognosis BICCAPS score.

[0048] The specific implementation scheme of the embodiment 2 is consistent with that of the embodiment 1.

[0049] As shown in Figure 2 and 3 , both figures show the survival probability of patients in different risk groups (low, medium and high risk groups) changing with time (months) through Kaplan-Meier curves, and both show that the survival difference between different risk groups is highly statistically significant (p<0.0001), and the risk number table below presents the number of patients still in the "at risk" state in each risk group at different time points, which helps to explain the sample size change of the survival analysis. In Figure 2 and Figure 3 , L belongs to the low risk group, M belongs to the medium risk group, and H belongs to the high risk group.

[0050] The scheme is based on real world data of 273 patients with advanced biliary tract malignant tumors receiving immunotherapy as a training cohort to construct a model, and is externally verified in two prospective clinical trial cohorts with a total of 94 patients. The verification results show that the scheme has good prediction performance: in the prospective external verification cohort, the AUC of predicting 12, 15 and 18 months of OS rate is 0.737, 0.832 and 0.804 respectively. The calibration curve shows that the predicted survival rate is highly consistent with the actual observed survival rate.

[0051] The present application first establishes a prognosis evaluation system comprising seven specific clinical indicators, which cover multiple dimensions such as treatment background (number of treatment lines), tumor burden (CA19-9, CEA, peritoneal metastasis), inflammatory state (SIRI), nutritional status (PNI) and liver function (total bilirubin), and develops a unique integer scoring algorithm. By dividing the regression coefficient by the selected reference coefficient (regression coefficient of total bilirubin per 10 μmol / L 0.198) and rounding, the complex statistical results are converted into simple integer scores, greatly improving the clinical practicability. Finally, a clear risk stratification standard is established to divide patients into low, medium and high risk groups, corresponding to different prognoses and treatment strategies.

[0052] The present application has high prediction accuracy and strong reliability. By integrating information of five dimensions of inflammation, nutrition, liver function, tumor burden and treatment background, the shortcomings of single or small number of indicators are overcome, and the prediction accuracy is greatly improved.

[0053] The present application has high clinical practicability and accessibility. The final product is a simple integer scoring table, all indicators come from routine clinical examination, doctors do not need complex calculation or additional expensive detection at the bedside, and risk assessment can be completed within a few minutes, which is easy to popularize and apply in hospitals at all levels, and solves the pain point of high-end biomarkers "applauding but not attending".

[0054] The present application is non-invasive and has very low cost, completely avoiding invasive tissue biopsy, and all required indicators are routine test items, greatly reducing the economic and physical burden of patients, in line with the principles of health economics.

[0055] The present application provides a complete and reusable model optimization methodology process. Through the three-step progressive screening of random forest algorithm, LASSO and optimal subset method, the problems of variable screening, collinearity and model simplicity in high-dimensional clinical data modeling are systematically solved. The present application has universality and can be popularized to the development of prediction models for other diseases.

[0056] The above is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A prognostic scoring method for predicting the prognosis of immunotherapy for advanced biliary tract malignancies, characterized in that, Includes the following steps: S1. Generate random permutation copies of the real variables to obtain shadow variables. Compare the importance scores of the real variables and shadow variables, classify the real variables, and obtain the relevant classes. S2. Based on the LASSO regression coefficients, the true variables in the relevant classes are screened to obtain the variable set; S3. Combine the real variables in the variable set, and select the model corresponding variable combination with the smallest AIC value based on the AIC criterion to obtain the set of key prognostic variables; S4. Calculate the index score for each real variable based on the Cox regression coefficients of each real variable. S5. Sum the scores of each real variable in the key prognostic variables set for each patient to obtain the prognostic BICCAPS score.

2. The prognostic scoring method for predicting the prognosis of immunotherapy for advanced biliary tract malignancies according to claim 1, characterized in that, S1 includes the following steps: S11. Generate random permutation copies of the real variables to obtain the shadow variables; S12. Use the random forest algorithm to calculate the importance scores of the real variable and the shadow variable to the prediction target; S13. Select the shadow variable with the highest importance score from all shadow variables to obtain the best shadow variable; S14. Use the Z-test to determine whether the importance score of the true variable is significantly higher than the importance score of the best shaded variable. If so, mark it as "relevant"; otherwise, mark it as "irrelevant" or "provisional". S15. Repeat steps S11 to S14 until all real variables have been classified or the preset maximum number of iterations has been reached to obtain the relevant classes.

3. The prognostic scoring method for predicting the prognosis of immunotherapy for advanced biliary tract malignancies according to claim 2, characterized in that, The standard that is significantly higher than S14 is: p < 0.01, where p is a probability value.

4. The prognostic scoring method for predicting the prognosis of immunotherapy for advanced biliary tract malignancies according to claim 1, characterized in that, S2 includes the following steps: S21. Introduce a penalty coefficient in LASSO regression to compress the LASSO regression coefficients of the true variables in the relevant classes to 0. S22. Take each real variable whose LASSO regression coefficient is not 0 as an element to form a variable set.

5. The prognostic scoring method for predicting the prognosis of immunotherapy for advanced biliary tract malignancies according to claim 1, characterized in that, S3 includes the following steps: S31. Combine the real variables in the variable set to obtain multiple variable combinations; S32. Using the Akaike Information Criterion as the model evaluation standard, calculate the AIC value of the model constructed for each combination of variables; S33. Find the variable combination corresponding to the model with the minimum AIC value to obtain the set of key prognostic variables.

6. The prognostic scoring method for predicting immunotherapy outcomes in advanced biliary tract malignancies according to claim 1 or 5, characterized in that, The key prognostic variables include: number of ICI treatment lines, CA19-9, CEA, peritoneal metastasis, SIRI, PNI, and total bilirubin.

7. The prognostic scoring method for predicting the prognosis of immunotherapy for advanced biliary tract malignancies according to claim 1, characterized in that, S4 includes the following steps: S41. Use the Cox regression algorithm to analyze the Cox regression coefficient of each true variable in the set of key prognostic variables; S42. Use the Cox regression coefficient corresponding to the total bilirubin unit quantification interval as the reference coefficient. S43. Divide the Cox regression coefficient of each real variable by the reference coefficient of the corresponding real variable and round down to obtain the index score of the corresponding real variable.

8. The prognostic scoring method for predicting immunotherapy outcomes in advanced biliary tract malignancies according to claim 7, characterized in that, The reference coefficient in S42 is 0.

198.

9. A prognostic scoring system for predicting the prognostic outcome of immunotherapy for advanced biliary tract malignancies, implemented based on the prognostic scoring method for predicting the prognostic outcome of immunotherapy for advanced biliary tract malignancies according to any one of claims 1 to 8, characterized in that, include: The system includes a first variable screening unit, a second variable screening unit, a third variable screening unit, an indicator score calculation unit, and a prognostic score calculation unit. The first variable filtering unit is used to generate random permutation copies of the real variables to obtain shadow variables, compare the importance scores of the real variables and shadow variables, classify the real variables, and obtain the relevant classes; The second variable screening unit is used to screen the real variables in the relevant class based on the LASSO regression coefficients to obtain a variable set; The third variable screening unit is used to combine the real variables in the variable set, and select the model corresponding variable combination with the smallest AIC value based on the AIC criterion to obtain the prognostic key variable set; The indicator score calculation unit is used to calculate the indicator score of each real variable based on the Cox regression coefficient of each real variable. The prognostic scoring calculation unit is used to sum the index scores of each real variable in the prognostic key variable set for each patient to obtain the prognostic BICCAPS score.

Images