Heart valve postoperative death risk prediction method based on physiological index interaction

Abstract

The invention discloses a heart valve postoperative death risk prediction method based on physiological index interaction, and relates to the technical field of medical engineering. The method specifically comprises the steps that S1, variable importance analysis is carried out to reduce interference of unimportant feature physiological indexes; s2, analyzing a survival curve for determining the influence of the key feature physiological indexes on the postoperative survival state of the patient; s3, constructing an intelligent discrimination model, and constructing the model by adopting a classification machine learning algorithm; and S4, intelligent discrimination and prediction: screening the postoperative death risk prediction model of the heart valve patient for predicting the postoperative death risk of the heart valve patient, and reducing the discrimination difficulty. According to the method, by analyzing the synergistic effect and antagonism among different characteristic physiological indexes, the heart valve postoperative death risk prediction model with high accuracy and good robustness is screened, and physiological data is provided for postoperative risk analysis and evaluation of heart valve patients.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical engineering, in particular to a method for predicting the risk of death after heart valve surgery based on the interaction of physiological indicators. BACKGROUND

[0002] Heart valve surgery is the main medical treatment for treating aortic valve, mitral valve, tricuspid valve and pulmonary valve stenosis and insufficiency. However, it has the problems of high surgical risk, large trauma and high postoperative mortality. The traditional postoperative death risk of heart valve patients is mainly judged by medical staff according to the patient's characteristic physiological indicators. There are more than 40 characteristic physiological indicators for heart valve patients after surgery, which has the defects of high technical difficulty, high experience requirement and unstable subjective judgment results. Moreover, there is a certain correlation between the characteristic physiological indicators. When medical staff assess the postoperative death risk of heart valve patients through characteristic physiological indicators, the combination of characteristic physiological indicators will have a synergistic or antagonistic effect on the postoperative death risk judgment, and the interaction between the characteristic physiological indicators increases the difficulty of medical staff in judging the postoperative death risk of patients. SUMMARY

[0003] The present application aims to provide a method for predicting the risk of death after heart valve surgery based on the interaction of physiological indicators to solve the problems raised in the background.

[0004] In order to solve the above technical problems, the present application provides the following technical scheme: a method for predicting the risk of death after heart valve surgery based on the interaction of physiological indicators, comprising the following specific steps:

[0005] Step S1: variable importance analysis: processing the characteristic physiological indicators of heart valve patients after surgery, calculating the importance of the characteristic physiological indicators by using SHAP (Shapley Additive Explanation), and selecting the first key characteristic physiological indicator set according to the importance ranking, which is used to reduce the interference of non-important characteristic physiological indicators;

[0006] Step S2: survival curve analysis: making a survival curve for each key characteristic physiological indicator in step S1, screening the survival curves with sudden drop trend, and arranging the key characteristic physiological indicators according to the sudden drop value of the survival curve to obtain the second key characteristic physiological indicator set, which is used to determine the importance of the key characteristic physiological indicators on the survival state of patients after surgery;

[0007] Step S3: intelligent discrimination model construction: the second key feature physiological indicators in step S2 are divided into X groups, the interaction of the key feature physiological indicators in the combination is analyzed by SHAP algorithm, the key feature physiological indicators with synergistic effect are screened as the third key feature physiological indicator set, and the third key feature physiological indicator set and the postoperative survival state of the patient constitute a full data set, and the full data is divided into a training set and a test set in proportion, the third key feature physiological indicator is taken as input data, the postoperative survival state of the patient is taken as output data, and different classification machine learning algorithms are used for model construction;

[0008] Step S4: intelligent discrimination prediction: the different classification machine learning models constructed in step S3 are evaluated by model performance indicators, the model with small difference between the training set model and the test set model is selected as the postoperative death risk prediction model of the heart valve patient, which is used for postoperative death risk prediction of the heart valve patient.

[0009] Preferably, the processing of the postoperative characteristic physiological indicators of the heart valve patient in step S1 specifically includes:

[0010] Obtaining the data of the postoperative characteristic physiological indicators of the heart valve patient;

[0011] The postoperative characteristic physiological indicator data of the heart valve patient is arranged, non-data information is removed, and a structured heart valve patient postoperative characteristic physiological indicator data set is formed;

[0012] The structured heart valve patient postoperative characteristic physiological indicator data set is repaired to obtain a complete heart valve patient postoperative characteristic physiological indicator data set;

[0013] The importance proportion of each feature physiological indicator is calculated by SHAP algorithm, and the top K important feature physiological indicators are selected to form a first key feature physiological indicator set.

[0014] Preferably, the calculation of the importance proportion of each feature physiological indicator by SHAP algorithm is as follows:

[0015]

[0016] Wherein, i is a feature physiological indicator, S is a feature physiological indicator set not containing i, N is the total number of feature physiological indicators, f(S) is the death risk prediction value of the feature physiological indicator set S, and f(S∪{i}) is the death risk prediction value of the feature physiological indicator set containing i.

[0017] Preferably, step S2 specifically includes:

[0018] The survival curve is drawn with the postoperative survival rate of the heart valve patient as the vertical coordinate and the survival time as the horizontal coordinate;

[0019] determine the trend change of the influence of each key feature physiological indicator on the postoperative survival state of the patient by the survival curve, and obtain the change of the survival rate of the key feature physiological indicator at different time nodes;

[0020] screen the survival curve with a sudden drop trend, arrange the key feature physiological indicators according to the size of the sudden drop value, obtain a second set of key feature physiological indicators, and determine the importance of the influence of the key feature physiological indicators on the postoperative survival state of the patient.

[0021] As preferred, the interaction of the key feature physiological indicators in the combination is analyzed by the SHAP algorithm in step S3, specifically:

[0022] The interaction value between two key feature physiological indicators is calculated, and the formula is:

[0023]

[0024] Wherein, M is the total number of key feature physiological indicators in the combination, i and j are the key feature physiological indicators in the combination, A is the set of key feature physiological indicators not containing i and j, f(A) is the death risk prediction value of the set of key feature physiological indicators A, f(A∪{i}) is the death risk prediction value of the set of key feature physiological indicators containing i, f(A∪{j}) is the death risk prediction value of the set of key feature physiological indicators containing j, f(A∪{i,j}) is the death risk prediction value of the set of key feature physiological indicators containing i and j, and φ i,j > 0, determined as synergistic effect, and φ i,j < 0, determined as antagonistic effect.

[0025] As preferred, the classification machine learning algorithm in step S3 specifically includes Gaussian naive Bayes GNB, gradient boosting classifier GBC, linear discriminant analysis LDA, support vector classifier SVC, quadratic discriminant analysis QLDA and extreme gradient boosting XGB.

[0026] As preferred, the model performance indicators specifically include specificity, accuracy, recall rate and precision rate.

[0027] The application has the beneficial effects that: the application analyzes the influence degree of each characteristic physiological index on the death risk, determines the importance proportion, eliminates non-key characteristic physiological indexes, and helps to solve the complexity of risk discrimination caused by a large number of indexes; the survival curve of the key characteristic physiological index is established to analyze the influence and change trend of the key characteristic physiological index at different time nodes on the survival rate of the patient, and the importance of the key characteristic physiological index on the postoperative survival state of the patient is determined; the key characteristic physiological indexes are grouped and the interaction analysis is performed, the key characteristic physiological indexes with antagonistic effect are eliminated, the key characteristic physiological indexes with synergistic effect are optimized, and the complexity of the interaction between the key characteristic physiological indexes is reduced, so that the complexity of the death risk assessment of the medical staff is reduced; different machine learning algorithms are used to construct, evaluate and screen the model, and the model with high accuracy and good robustness is selected as the risk prediction model, which is applied to the related department and used for the prediction of the death risk of the heart valve patient after the operation, so as to assist the medical staff in judging the death risk of the heart valve patient after the operation, avoid the instability of subjective judgment, reduce the deviation caused by human factors, and reduce the difficulty of the medical staff in judging the death risk of the patient. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is a specific step schematic diagram of the application.

[0029] Figure 2 is a whole characteristic physiological index survival curve change diagram of the heart valve patient after the operation.

[0030] Figure 3 is a characteristic physiological index interaction diagram of the heart valve patient after the operation.

[0031] Figure 4 is a classification machine learning model training set model training result matrix diagram.

[0032] Figure 5 is a classification machine learning model test set model training result matrix diagram. DETAILED DESCRIPTION

[0033] In the following description, a large number of specific details are given to provide a more thorough understanding of the application. However, it is obvious to those skilled in the art that the application can be implemented without one or more of these details. In other examples, some technical features known in the art are not described in order not to obscure the application.

[0034] REFERENCE Figure 1The application provides a heart valve postoperative death risk prediction method based on physiological index interaction, and specific steps are as follows: step S1, variable importance analysis; step S2, survival curve analysis; step S3, intelligent discrimination training; and step S4, intelligent discrimination prediction.

[0035] The postoperative characteristic physiological index and postoperative survival state data of the heart valve patient are regression data, which are mainly data obtained by medical staff when arranging physiological examination and postoperative survival state in the postoperative rehabilitation process of the heart valve patient, the characteristic physiological index of each specific patient is extracted as machine learning model construction data, and the postoperative survival state is 0 or 1, wherein 0 represents that the patient is in a postoperative survival state when discharged from hospital, and 1 represents a postoperative death state.

[0036] In the embodiment, in the heart valve postoperative death risk discrimination, the characteristic physiological index mainly includes 34 indexes, and the indexes are specifically gender, age (Age), body mass index (BMI), hemoglobin (HB), New York heart function classification (NYHA), smoking history (SH), aortic valve disease (AVD), aortic valve insufficiency (AI), aortic valve stenosis (AS), mitral valve disease (MVD), mitral valve insufficiency (MVI), mitral valve stenosis (MS), tricuspid valve disease (TVD), tricuspid valve insufficiency (TI), myocardial ischemia (MI), hypertension (HTN), hyperlipidemia (HL), diabetes (DM), acute kidney injury (AKI), mechanical ventilation time (DMV), hypoxemia (HPX), infective endocarditis (IE), carotid artery stenosis (CAS), cerebrovascular disease (CVD), chronic obstructive pulmonary disease (COPD), aortic aneurysm (AA), dialysis, cardiopulmonary resuscitation (CPR), endotracheal intubation (ET), atrial fibrillation (AF), gastrointestinal hemorrhage (GH), liver failure (HI), acute gastrointestinal injury (AGI) and reoperation (REO). The collected 34 characteristic physiological index data sets are analyzed, redundant data or non-data information is removed, and structural data is formed; in the case that the patient physiological index data is missing, the missing data is modified by using a k-nearest neighbor algorithm KNN, a random forest RF, a linear regression Linear, a chained multiple imputation MICE and a mean imputation Mean algorithm, and a complete characteristic index data set is generated.

[0037] In the embodiment, the mean imputation Mean is used to repair the missing data in the case that the postoperative physiological index of the heart valve patient is missing, and the repairing specifically includes the following steps.

[0038] The average number is calculated for the data of the non-missing characteristic physiological index, and the calculation formula is as follows:

[0039]

[0040] Nvalid the number of feature physiological indicators for the non-missing value of the feature physiological indicator X, x i is the feature value corresponding to the ith feature physiological indicator, and μx is the mean value of each feature physiological indicator x mising the filled feature physiological indicator, the formula is:

[0041] x mising = μx.

[0042] The importance proportion of the feature physiological indicators of the heart valve patients after the operation is analyzed by the SHAP algorithm, the feature physiological indicators of the heart valve patients after the operation are combined according to the importance ranking, and the key feature physiological indicators are obtained. The importance of each feature physiological indicator is calculated, and the calculation formula is:

[0043]

[0044] wherein i is a feature physiological indicator, S is a feature physiological indicator set not containing i, N is the total number of feature physiological indicators, f(S) is the death risk prediction value of the feature physiological indicator set S, and f(S∪{i}) is the death risk prediction value containing i. According to the importance calculation result of each feature physiological indicator, the feature physiological indicators with the top K importance rankings are combined to form a first key feature physiological indicator set. In this embodiment, K is 16, and the feature physiological indicators specifically include: HTN, COPD, AKI, DMV, MI, BMI, Age, HB, AS, TVD, AI, MS, MVI, HI, GH, and AF. The feature physiological indicators ranked after 18 are removed to reduce the influence on the postoperative state discrimination of the patients.

[0045] The survival curve is drawn with the postoperative survival rate of the heart valve patients as the vertical coordinate and the survival time as the horizontal coordinate, the influence trend change of each key feature physiological indicator on the postoperative survival state of the patients is judged, and the change of the survival rate of the key feature physiological indicators at different time nodes is obtained. Referring to Figure 2 is the survival curve of the overall 16 key feature physiological indicators, and the shadow represents the influence of different key feature physiological indicators on the survival rate at the same time node. The survival time mainly refers to the time from the postoperative state to the death of the patient, and the unit is day. The survival curve is calculated and plotted by the kaplan-meier algorithm for a single feature physiological indicator, wherein the survival rate calculation formula is:

[0046]

[0047] wherein t is the survival time, t θ is the death event time point of the heart valve patients after the operation, d θ is the survival time of the heart valve patients after the operation, and d θThe number of heart valve patient deaths at time node t, n θi The number of deaths related to the risk at time node t θ , ∏ is the multiplication symbol, and the product is mainly multiplied by the time points t θ ≤t.

[0048] The first 7 days of the rehabilitation stage after heart valve surgery is a critical risk period, and the patient's mortality rate is higher during this period. If the survival curve shows a cliff-like drop between two adjacent time points, it indicates that the critical characteristic physiological indicator has a sudden increase in mortality risk at that postoperative time.

[0049] The survival curve of heart valve patients within 7 days after surgery is screened out from 1.0 to 0.6 and below, and then arranged according to the size of the drop value to obtain the second critical characteristic physiological indicator set. According to the ranking of the critical characteristic physiological indicators, the importance of the critical characteristic physiological indicators on the survival status of heart valve patients after surgery is obtained, and the ones arranged in the front are considered as the key characteristic physiological indicators. In this embodiment, the ranking order of the critical characteristics is: HTN, BMI, AS, TVD, COPD, Age, HB, AKI, DMV, AI, MS, MVI, HI, GH, MI, AF. Different arrangement methods will have different effects on the model performance, and arranging by the size of the drop value can obtain better model prediction results.

[0050] In this embodiment, the 16 critical characteristic physiological indicators are divided into four groups, specifically: combination 1: HTN, BMI, AS, TVD; combination 2: COPD, Age, HB, AKI; combination 3: DMV, AI, MS, MVI; combination 4: HI, GH, MI, AF. They can also be divided into other combinations, and the interaction between the critical characteristic physiological indicators in the combination is calculated by the SHAP algorithm.

[0051] The interaction value between two critical characteristic physiological indicators is calculated by the formula:

[0052]

[0053] Where M is the total number of critical characteristic physiological indicators in the combination, i and j are the critical characteristic physiological indicators in the combination, A is the set of critical characteristic physiological indicators excluding i and j, f(A) is the mortality risk prediction value of the critical characteristic physiological indicator set A, f(A∪{i}) is the mortality risk prediction value of the critical characteristic physiological indicator set containing i, f(A∪{j}) is the mortality risk prediction value of the critical characteristic physiological indicator set containing j, f(A∪{i,j}) is the mortality risk prediction value of the critical characteristic physiological indicator set containing i and j, and φ i,j >0, indicating synergistic effect, and φ i,j <0, indicating antagonistic effect.i,j Key physiological indicators with a value <0 were used to screen for φ. i,j The key physiological indicators with a value >0 are used as the third set of key physiological indicators.

[0054] Reference Figure 3 The seven key physiological indicators with the best synergistic effects, selected through screening, are ranked in descending order of synergy as follows: HTN, BMI, AS, TVD, COPD, Age, and HB. The diagonal lines represent the main effects, while the areas on either side of the diagonal represent the interaction effects. A larger distance between the blue and red scatter points indicates a stronger interaction, demonstrating the synergistic effects among the key physiological indicators. The selected key physiological indicators with synergistic effects and good results are combined to form the third set of key physiological indicators. This set is then used to reconstruct the data with patient postoperative survival status, resulting in the full dataset.

[0055] The entire dataset is divided into training and testing sets according to a set ratio. In this embodiment, the dataset of 1200 patient physiological characteristic indicators is divided into training and testing sets using a method sampling method. The dataset is divided into training and testing sets at a ratio of 3:1, resulting in 900 training set samples and 300 testing set samples. The training set samples are used to build the training model, while the testing set samples are mainly used to build the testing model.

[0056] Using the third key feature, physiological indicators, as input data and the postoperative survival status of patients as output data, models were constructed using Gaussian Naive Bayes (GNB), Gradient Boosting Classifier (GBC), Linear Discriminant Analysis (LDA), Support Vector Classifier (SVC), Quadratic Discriminant Analysis (QLDA), and Ultimate Gradient Boosting (XGB) machine learning algorithms, respectively.

[0057] In this embodiment, SVC is used for model construction, and the main process includes:

[0058] The mathematical formula for choosing the RBF kernel function is:

[0059] K(x n ,x m )=exp(-γ||x n -x m || 2 );

[0060] Where K(x) n ,x m (x) is the third key physiological characteristic indicator. n and x m The kernel function similarity value between them, ||x n -x m|| is the Euclidean distance between two key feature physiological indicators, and γ is a regulating parameter related to the RBF kernel function, where the greater γ is, the more complex the training model is and the more likely overfitting phenomenon occurs, the smaller γ is, the more likely underfitting phenomenon occurs, and the exponential function exp() represents the exponential function e x .

[0061] Two important parameters of the SVC, the gamma parameter and the C parameter, are set, where the SVC boundary becomes more complex and overfitting phenomenon is more likely to occur when the gamma parameter increases, and the boundary performs more smoothly but underfitting results are more likely to occur when the gamma parameter decreases. At the same time, when the C parameter increases, the SVC boundary becomes more complex and the classification is more strict, and when the C parameter decreases, the boundary is smooth and the error tolerance increases. The gamma parameter and the C parameter are adjusted by the genetic algorithm and the grid search method to obtain a better model construction result.

[0062] The specificity Spec, the accuracy Acc, the recall rate Rec and the precision rate Prec are used as the model performance indicators to evaluate the constructed SVC model and compare the differences in model training performance between the training model and the test model. The accuracy is the proportion of correctly classified samples in the total samples, and thus the model performance is evaluated, and the formula is:

[0063]

[0064] Among them, TP is the number of true cases, that is, the number of correctly predicted positive cases, TN is the number of true negative cases, FP is the number of false positive cases, that is, the number of incorrectly predicted positive cases, and FN is the number of false negative cases, that is, the number of missed predictions.

[0065] In this embodiment, for the training set, the number of correctly predicted survival patients TP = 459, the number of correctly predicted death patients TN = 270, the number of death patients predicted as survival patients FP = 81, and the number of missed patients FN = 90, and thus the model classification accuracy is

[0066] For the test set, the number of correctly predicted survival patients TP = 157, the number of correctly predicted death patients TN = 80, the number of death patients predicted as survival patients FP = 30, and the number of missed patients FN = 330, and thus the model classification accuracy is

[0067] In this embodiment, the model's accuracy on the training set is 0.81, and its accuracy on the test set is 0.79. The higher accuracy on the training set compared to the test set indicates overfitting, suggesting model instability and requiring optimization. Model performance can be optimized by adjusting parameters. If the test set classification accuracy is higher than the training set accuracy, it indicates underfitting, and parameter adjustments can also be used to optimize model performance.

[0068] Reference Figure 4 and Figure 5 The constructed GNB, GBC, LDA, SVC, QLDA, and XGB models were evaluated using specificity, accuracy, recall, and precision, respectively, yielding the model performance matrix results. The matrix results show that the GBC model achieved specificity of 0.93, accuracy of 0.93, recall of 0.93, and precision of 0.94 on the training set, and the same on the test set. Compared to other models, the GBC model showed a classification accuracy difference of less than 0.1 between the training and test sets, and its accuracy was closest to 1, making it the model with the highest accuracy and best robustness. Therefore, the GBC model was selected as a model for predicting the postoperative mortality risk in patients with heart valve disease. The predictive ability of the selected postoperative mortality risk prediction model for heart valve patients was analyzed using clinical data from heart valve surgery. This model was then applied in clinical practice to predict the postoperative mortality risk of heart valve surgery patients. This assists medical staff in assessing the postoperative mortality risk of heart valve surgery patients, helps them gain a more comprehensive understanding of the survival status and risk level of heart valve surgery patients, avoids the instability of subjective judgment, reduces bias caused by human factors, and lowers the difficulty for medical staff in assessing the mortality risk of patients.

[0069] As described above, although the invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the invention itself. Various changes in form and detail may be made without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. A method for predicting the risk of postoperative death of a heart valve based on the interaction of physiological indicators, characterized by, The specific steps include: Step S1: variable importance analysis: the postoperative characteristic physiological indicators of heart valve patients are processed, the importance of the characteristic physiological indicators is calculated by using SHAP (Shapley Additive Explanation), and according to the importance ranking, the first key characteristic physiological indicator set is screened from the first K characteristic physiological indicators, which is used to reduce the interference of non-important characteristic physiological indicators; Step S2: survival curve analysis: the survival curve of each key characteristic physiological indicator in step S1 is drawn, and the survival curve with a sudden drop trend is screened, and the key characteristic physiological indicators are arranged according to the sudden drop value of the survival curve, to obtain the second key characteristic physiological indicator set, which is used to determine the importance of the key characteristic physiological indicators on the postoperative survival state of the patient; Step S3: intelligent discrimination model construction: the second key characteristic physiological indicators in step S2 are divided into X groups, the interaction of the key characteristic physiological indicators in each group is analyzed by using the SHAP algorithm, the key characteristic physiological indicators with synergistic effect are screened as the third key characteristic physiological indicator set, and the third key characteristic physiological indicator set and the postoperative survival state of the patient form a full data set, the full data set is divided into a training set and a test set according to a proportion, the third key characteristic physiological indicator is used as input data, and the postoperative survival state of the patient is used as output data, different classification machine learning algorithms are used to construct models; Step S4: intelligent discrimination prediction: the different classification machine learning models constructed in step S3 are evaluated according to the model performance indicators, the model with small difference between the training set model and the test set model is selected as the postoperative death risk prediction model of the heart valve patient, which is used for postoperative death risk prediction of the heart valve patient. 2.The method of predicting the risk of death after heart valve surgery based on the interaction of physiological indicators according to claim 1, characterized in that, In step S1, the postoperative characteristic physiological indicators of heart valve patients are processed, specifically including: obtaining the data of the postoperative characteristic physiological indicators of heart valve patients; organizing the postoperative characteristic physiological indicator data of heart valve patients, eliminating non-data information, and forming a structured postoperative characteristic physiological indicator data set of heart valve patients; performing data repair on the structured postoperative characteristic physiological indicator data set of heart valve patients to obtain a complete postoperative characteristic physiological indicator data set of heart valve patients; the importance proportion of each characteristic physiological indicator is calculated by using the SHAP algorithm, and the characteristic physiological indicators with the top K importance are screened to form the first key characteristic physiological indicator set. 3.The method of predicting the risk of death after heart valve surgery based on the interaction of physiological indicators according to claim 2, characterized in that, The importance proportion of each characteristic physiological indicator is calculated by using the SHAP algorithm, and the characteristic physiological indicators with the top K importance are screened to form the first key characteristic physiological indicator set. The formula is: 4.The method of claim 1, wherein, wherein i is a characteristic physiological indicator, S is a characteristic physiological indicator set not containing i, N is the total number of characteristic physiological indicators, f(S) is the death risk prediction value of the characteristic physiological indicator set S, and f(S∪{i}) is the death risk prediction value of the characteristic physiological indicator set containing i. Step S2 specifically includes: drawing a survival curve with the postoperative survival rate of heart valve patients as the vertical coordinate and the survival time as the horizontal coordinate; determine the influence trend change of each key characteristic physiological indicator on the postoperative survival state of the patient from the survival curve, and obtain the change of the survival rate of the key characteristic physiological indicator at different time nodes; Screening the survival curve with sudden drop trend, and arranging the key characteristic physiological indexes according to the size of the sudden drop value to obtain a second key characteristic physiological index set, and determining the importance of the key characteristic physiological indexes on the postoperative survival state of the patient. 5.The method of predicting the risk of death after heart valve surgery based on the interaction of physiological indicators according to claim 1, characterized in that, The interaction of the key characteristic physiological indexes in the combination is analyzed by the SHAP algorithm in step S3, specifically: The interaction value between two key characteristic physiological indexes is calculated, and the formula is: Wherein, M is the total number of key characteristic physiological indicators in the combination, i and j are the key characteristic physiological indicators in the combination, A is the set of key characteristic physiological indicators not containing i and j, f(A) is the death risk prediction value of the key characteristic physiological indicator set A, f(A∪{i}) is the death risk prediction value of the key characteristic physiological indicator set containing i, f(A∪{j}) is the death risk prediction value of the key characteristic physiological indicator set containing j, f(A∪{i,j}) is the death risk prediction value of the key characteristic physiological indicator set containing i and j, φ i,j > 0, it is determined that there is synergistic effect, φ i,j < 0, it is determined that there is antagonistic effect. 6.The method of predicting the risk of death after heart valve surgery based on the interaction of physiological indicators according to claim 1, characterized in that, The classification machine learning algorithm specifically includes Gaussian naive Bayes GNB, gradient boosting classifier GBC, linear discriminant analysis LDA, support vector classifier SVC, quadratic discriminant analysis QLDA and extreme gradient boosting XGB. 7.The method of claim 1, wherein the method is based on interaction of physiological indicators. The model performance indicators specifically include specificity, accuracy, recall rate and precision rate.

Images