Bleeding prediction method and device

Abstract

The invention provides a bleeding prediction method and device. The method comprises the following steps: acquiring an electronic case of a target user; screening m target characteristic parameters from the electronic medical record, wherein the target characteristic parameters are physiological parameters influencing the bleeding result of the target user; and inputting the m target feature parameters into the target prediction model to obtain a bleeding result of the target user. According to the method, the feature parameters influencing the bleeding of the user after the ventricular assist device is used during the high-risk PCI treatment are screened out from the electronic medical record of the user, and the feature parameters are input into the pre-selected trained target prediction model, so that the bleeding result of the user is obtained; by means of the method, whether the patient has the bleeding problem after the ventricular assist device operation or not can be predicted before the operation, the postoperative bleeding risk is reduced, operation prognosis and recovery are improved, and therefore the safety of a user is improved.

Description

Technical Field

[0001] This application relates to the field of medical device technology, and in particular to a bleeding prediction method and device. Background Technology

[0002] Currently, for high-risk patients undergoing percutaneous coronary intervention, acute compensated heart failure, and cardiogenic shock, an interventional ventricular assist device is implanted before the procedure to continuously supply blood to vital organs such as the heart and brain during the surgery, maintaining stable blood pressure and circulation. This provides doctors with a safer and more comfortable surgical environment, enabling them to perform more complex revascularization procedures.

[0003] However, some patients may experience bleeding during the implantation of an interventional ventricular assist device (VAD), potentially leading to bleeding complications. Severe bleeding complications can increase patient mortality and hospital stay. Currently, there is no precise method for preoperative bleeding risk assessment specifically for VAD procedures. Summary of the Invention

[0004] This application provides a bleeding prediction method and device that can accurately predict bleeding risk before ventricular assist device surgery, thereby improving patient safety.

[0005] In a first aspect, embodiments of this application provide a method for predicting bleeding, the method comprising: Obtain the target user's electronic medical records; m target feature parameters are selected from the electronic medical record. The target feature parameters are physiological parameters that affect the bleeding outcome after the ventricular assist device is used on the target user. m is a positive integer. The m target feature parameters are input into the target prediction model to obtain the bleeding result of the target user.

[0006] Secondly, this application provides a bleeding prediction device, which includes one or more processors, the one or more processors being configured to perform the following steps: Obtain the target user's electronic medical records; m target feature parameters are selected from the electronic medical record. The target feature parameters are physiological parameters that affect the bleeding outcome after the ventricular assist device is used on the target user. m is a positive integer. The m target feature parameters are input into the target prediction model to obtain the bleeding result of the target user.

[0007] Thirdly, embodiments of this application provide a medical device, the medical device including a processor, a memory, a communication interface, and one or more programs, the one or more programs being stored in the memory and configured to be executed by the processor, the programs including instructions for performing some or all of the steps described in the method described in the first aspect above.

[0008] The technical solution provided in this application involves acquiring the electronic medical record of a target user; selecting m target feature parameters from the electronic medical record, which are physiological parameters affecting the bleeding outcome of the target user; and inputting the m target feature parameters into a target prediction model to obtain the bleeding outcome of the target user. This application, by selecting feature parameters affecting bleeding after ventricular assist device (VAD) surgery from the user's electronic medical record and inputting these feature parameters into a pre-trained target prediction model to obtain the user's bleeding outcome, can predict before surgery whether the patient will experience bleeding problems after VAD surgery, reducing the risk of postoperative bleeding, improving surgical prognosis and recovery, and thus enhancing user safety. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram of a ventricular assist system provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a ventricular assist device provided in an embodiment of this application; Figure 3 This is a schematic flowchart of a bleeding prediction method provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a medical device provided in an embodiment of this application. Detailed Implementation

[0011] To help those skilled in the art better understand the technical solutions of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the description of the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, software, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but also includes steps or units not listed, or other steps or units inherent to such processes, methods, products, or apparatus.

[0013] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0014] It should be noted that in this application, the terms "proximal" or "proximal" refer to the end or side closer to the surgeon; and "distal" or "distal" refer to the end or side farther from the surgeon.

[0015] Please see Figures 1-2 , Figure 1 This is a schematic diagram of the structure of a ventricular assist device 100 provided in an embodiment of this application. Figure 2 This is a schematic diagram illustrating a ventricular assist device 100 located in the normal position of a patient's heart 120, as provided in an embodiment of this application. The ventricular assist device 100 can operate within the patient's heart, partially within the heart, outside the heart, partially outside the heart, partially outside the vascular system, or in any other suitable location within the vascular system. The ventricular assist device 100 can be percutaneously inserted into the aorta 124 via the femoral artery 122 and passes through the aorta 124 into the left ventricle 128. For example, the ventricular assist device 100 can be percutaneously inserted into the aorta 124 via the axillary artery 123 and passes through the aorta 124 into the left ventricle. In other embodiments, the ventricular assist device 100 can also be directly inserted into the aorta 124 and passes through the aorta 124 into the left ventricle 128. During operation, the ventricular assist device 100 pumps blood from the left ventricle 128 into the aorta 124.

[0016] The ventricular assist device 100 includes a cannula 10. The cannula 10 has a proximal end and a distal end, the distal end of the cannula 10 having a fluid inlet 101 and the proximal end of the cannula 10 having a fluid outlet 102, through which blood flows in from the fluid inlet 101 and out from the fluid outlet 102 via the cannula 10.

[0017] The ventricular assist device 100 includes an impeller (not shown). The impeller is located at least partially at the proximal end of the cannula 10, such as at the fluid inlet 101 of the cannula 10, such that when the ventricular assist device 100 is in operation, it drives the impeller to rotate to pump blood from the left ventricle 128 to the aorta 124.

[0018] The ventricular assist device 100 may include a motor (not shown in the figure), which may be located inside or outside the ventricular assist device 100. This embodiment of the application illustrates the example where the motor is located inside the ventricular assist device 100. For example, the motor is located in a motor housing 201, with the distal end of the motor housing 201 connected to the proximal end of the sleeve 10. The motor drives the drive shaft to rotate, thereby driving the impeller to rotate, thus realizing the pumping function of the ventricular assist device 100.

[0019] The ventricular assist device 100 includes a catheter 30, the distal end of which is connected to the proximal end of a motor housing 201, through which a drive cable extends. As an example, the catheter 30 may accommodate an electrical connection between the ventricular assist device 100 and an external controller. As an example, the ventricular assist device 100 also includes a distal component 110, such as a pigtail cannula, extending distally away from the distal end of the cannula 10.

[0020] When the ventricular assist device 100 is used on a patient's left heart, the ventricular assist device 100 is considered to be in the desired position when it is positioned such that the cannula 10 extends across the patient's aortic valve 126, the distal end of the cannula 10 is located in the patient's left ventricle 128, the proximal end of the cannula 10 is located within the patient's aorta 124, and the distal component 110 is just abutting against or at a predetermined distance from the inner wall of the patient's left ventricle 128. When the ventricular assist device 100 is used on a patient's right heart, the ventricular assist device 100 is considered to be in the desired position when it is positioned such that the cannula 10 extends across the pulmonary valve 125, the distal end of the cannula 10 is located in the patient's right ventricle 127, the proximal end of the cannula 10 is located within the patient's pulmonary artery 121, and the distal component 110 is just abutting against the inner wall of the patient's right ventricle 127. That is, the ventricular assist device 100 is in the correct position. The following explanation uses the ventricular assist device 100 used in the left heart as an example.

[0021] The ventricular assist device 100 also includes a pressure sensor disposed on the outer surface of the proximal end of the cannula 10, i.e., the outer surface of the fluid outlet 102, for detecting pressure at the fluid outlet 102 of the ventricular assist device 100. When the ventricular assist device 100 is positioned correctly across the aorta 124, the top (outer surface) of the pressure sensor is exposed to the aorta 124, thereby enabling measurement of aortic pressure.

[0022] The ventricular assist device 100 also includes a control unit, which can be used to perform any of the embodiments, aspects, and methods of this application. The control unit may be located inside or outside the ventricular assist device 100. The control unit is used to detect relevant parameters of the ventricular assist device 100 and the patient, and to control the operation of the ventricular assist device 100. For example, the control unit supplies current to a motor through one or more wires and detects the current through a current detection circuit (such as a phase current detection circuit); controls the rotational speed of the ventricular assist device 100 according to received instructions; receives feedback signals from pressure sensors to identify the position of the ventricular assist device 100, and so on.

[0023] Anticoagulants are used during and after the implantation of the ventricular assist device 100, which may cause serious bleeding in patients and even lead to bleeding complications. Bleeding complications can cause acute circulatory failure, shock, and even death within a short period of time; the rapid progression of bleeding complications seriously endangers the patient's safety.

[0024] Based on this, this application targets interventional ventricular assist devices, inputting all of the patient's physiological parameters into a trained prediction model to obtain bleeding results, so as to predict the patient's bleeding outcome before the operation and avoid harm to the patient.

[0025] Based on the above description, this application will now be described from the perspective of method examples.

[0026] Please see Figure 3 , Figure 3 This is a schematic flowchart of a bleeding prediction method provided in an embodiment of this application, which is used to predict bleeding such as... Figure 1-2 Pre-operative bleeding prediction before implantation of the ventricular assist device 100. For example... Figure 3 As shown, the method includes the following steps.

[0027] S310. Obtain the target user's electronic medical record.

[0028] Electronic medical records integrate scattered information, including the user's basic information, physiological parameters, examination reports, and treatment records. To determine whether a user is at risk of bleeding after ventricular assist device (VAD) implantation, the user's electronic medical record can be accessed. This record contains the target user's basic information, physiological parameters, examination reports, treatment records, and other data. Based on the target user's preoperative data, the risk of bleeding after VAD implantation can be determined, ensuring user safety.

[0029] S320. Select m target feature parameters from the electronic medical record. The target feature parameters are physiological parameters that affect the bleeding outcome of the target user after the ventricular assist device procedure, and m is a positive integer.

[0030] Electronic medical records contain many characteristic parameters of the target user. To prevent interference from redundant parameters and parameters with no predictive value, the characteristic parameters obtained from the electronic medical records are pre-screened to retain the core features related to the user's bleeding risk, thereby reducing the complexity of the prediction model.

[0031] In one possible example, before selecting m target feature parameters from the electronic medical record, the method further includes: obtaining n initial feature parameter groups, each initial feature parameter group including one feature parameter for bleeding users and non-bleeding users, wherein the n are positive integers greater than m; and filtering the feature parameters according to the n initial feature parameter groups to obtain the m target feature parameters.

[0032] Before selecting target feature parameters from electronic medical records, it's essential to determine which user feature parameters should be used as target feature parameters to input into the model for predicting bleeding outcomes. First, combine clinical experience to obtain multiple initial feature parameters that may be related to post-ventricular assist device (VAM) bleeding. Then, filter these initial feature parameters to identify relevant risk prediction features.

[0033] The initial feature parameters of the user with bleeding and the user without bleeding are collected to form n initial feature parameter groups. Each group includes one feature parameter for the user with bleeding and one for the user without bleeding. These feature parameters can be user information from basic information, physiological parameters, examination reports, or treatment records. Among them, the user's physical characteristics directly affect the implantation operation of ventricular assist devices (such as the compatibility of large sheaths) and drug metabolism (anticoagulant dosage), and are fundamental feature parameters affecting the user's bleeding risk. Furthermore, interventional ventricular assist devices require vascular access for implantation, and the user's physical characteristics directly affect the choice of access (such as the difficulty of femoral artery puncture), which is directly related to the bleeding risk associated with large-diameter access. Therefore, features such as age, gender, body surface area, weight, height, and body mass index can be selected from the user's basic information as initial feature parameters.

[0034] The user's baseline circulatory status, such as heart rate, systolic blood pressure, and diastolic blood pressure, are routine preoperative monitoring indicators. Unstable baseline circulation (such as hypotension) may increase the risk of anticoagulation-related bleeding, and it is directly related to the user's overall tolerance. Furthermore, ventricular assist devices are mainly used for users with heart failure (such as reduced left ventricular ejection fraction). Poor heart function leads to insufficient tissue perfusion, indirectly affecting coagulation function. Therefore, characteristic parameters such as heart rate, systolic blood pressure, diastolic blood pressure, left ventricular ejection fraction, and left ventricular end-diastolic diameter can be selected from the user's physiological parameters as initial characteristic parameters.

[0035] A user's physiological parameters directly affect their bleeding risk. For example, parameters related to coagulation directly impact bleeding risk; parameters related to liver and kidney function directly affect coagulation and drug metabolism; and parameters related to infection may exacerbate inflammatory responses and coagulation disorders, increasing the risk of bleeding. Therefore, characteristic parameters such as white blood cell count, hematocrit, platelet count, hemoglobin count, activated partial thromboplastin time, creatinine, creatinine clearance rate, and transaminase levels were selected from the test reports as initial characteristic parameters.

[0036] Known bleeding risk factors, such as the user's preoperative medical history, abnormalities, and treatment outcomes, significantly influence the probability of postoperative bleeding. For example, hypertension, diabetes, and chronic kidney disease are underlying conditions of vascular disease and coagulation dysfunction, and are positively correlated with the bleeding risk after ventricular assist device (VAD) implantation. Heart failure, arrhythmias, and other cardiac dysfunctions, accompanied by circulatory disturbances and increased anticoagulation requirements, further increase the bleeding risk. VADs require placement through a large-diameter arterial access; abnormal access (such as vascular stenosis or calcification) directly increases the risk of puncture injury and bleeding, and postoperative bleeding may exacerbate complications. High-risk percutaneous coronary interventions supported by VADs are often caused by complex lesions (such as multivessel disease, left main coronary artery disease, etc.). The number and location of lesions directly reflect the operation time and difficulty (e.g., left coronary lesions require more delicate procedures), and complex surgeries increase vascular injury and anticoagulation exposure time, thus increasing the bleeding risk. Therefore, the following characteristic parameters were selected from the treatment records as initial characteristic parameters: hypertension, diabetes, heart failure, arrhythmia, history of stroke, chronic kidney disease, infection, abnormal access vessels, number of left coronary lesions, number of right coronary lesions, and total number of lesions.

[0037] After obtaining the above-mentioned initial feature parameters of bleeding users and non-bleeding users, the initial feature parameters may have features with the same properties or features with no predictive value. Therefore, further analysis is needed to identify highly relevant risk prediction feature parameters.

[0038] Optionally, the step of filtering the feature parameters based on the n initial feature parameter groups to obtain the m target feature parameters includes: performing inter-group difference analysis on the n initial feature parameter groups to filter out k key feature parameters, where k is a positive integer less than n and greater than m, and the key feature parameters are statistically significant. p The value is less than 0.05; correlation analysis is performed on the k key feature parameters to filter out t unrelated feature parameters, the correlation coefficient of the t unrelated feature parameters is greater than 10, and t is a positive integer less than n and greater than m; the variance inflation factor of the t unrelated feature parameters is calculated, and the unrelated feature parameters whose variance inflation factor is greater than or equal to a preset third value are excluded to obtain the m target feature parameters.

[0039] Initial feature selection can identify characteristic parameters highly associated with bleeding risk through inter-group difference analysis and correlation analysis. First, key predictive characteristic parameters are identified by comparing the inter-group differences in initial characteristic parameters between bleeding and non-bleeding users, resulting in k key characteristic parameters. Then, correlation analysis is performed on these k key characteristic parameters to identify strong correlations among them, avoiding instability in the prediction model due to high correlations among key characteristic parameters.

[0040] The step of performing inter-group difference analysis on the n initial feature parameters to select k key feature parameters includes: classifying the n initial feature parameter groups according to data type to obtain p first feature parameter groups and q second feature parameter groups, where p and q are positive integers; and performing a normality test on each of the p first feature parameter groups to obtain the first feature parameter group. p Value; Perform expected frequency checks on each of the q second feature parameter groups to obtain the expected frequency of each second feature parameter group; if the expected frequency is greater than or equal to 5, apply the Fisher exact test to the q second feature parameter groups; if the expected frequency is less than 5, apply the chi-square test to the q second feature parameter groups to obtain the expected frequency of each second feature parameter group. p Value; if the stated p If the value is greater than or equal to 0.05, then the first feature parameter group or the second feature parameter group is deleted to obtain the k key feature parameters.

[0041] Different user parameters have different data types, and the methods for differential analysis differ for different data types. Therefore, we first classify the initial feature parameters according to their data types, dividing them into numeric (including continuous numeric, integer numeric, etc.) and categorical (including factor, character, logical, etc.). The first set of feature parameters consists of numeric initial feature parameters, and the second set consists of categorical initial feature parameters.

[0042] For numerical initial characteristic parameters, the distribution test method can be used for analysis. Specifically, a normality test is performed on the initial characteristic parameters in each of the p first characteristic parameter groups to obtain the initial characteristic parameters in each first characteristic parameter group. p Value, if p A value less than 0.05 indicates that the initial characteristic parameter follows a normal distribution, and a homogeneity of variance test can be performed. If the variances are homogeneous, a t-test is used to analyze the differences between the parameters of bleeding users and non-bleeding users in the initial characteristic group of the first characteristic parameter group, i.e., to analyze the differences between the initial characteristic groups, thus obtaining the first characteristic parameter group. pValues. If the variances are unequal, the Wilcoxon rank-sum test is used to analyze the differences between the parameters of bleeding users and non-bleeding users in the initial feature group of the first feature parameter group, thus obtaining the value of the first feature parameter group. p Value. Furthermore, if the initial characteristic parameters obtained by the normality test method... p When the value is greater than or equal to 0.05, it indicates that the initial feature parameter does not follow a normal distribution. Therefore, the Wilcoxon rank-sum test is used to analyze the difference between the parameters of bleeding users and non-bleeding users in the initial feature group of the first feature parameter group, thus obtaining the first feature parameter group. p Value. If p A value less than 0.05 indicates that the initial feature parameter has a significant difference between bleeding and non-bleeding users, and that this initial feature parameter will create a bleeding risk for users; therefore, this initial feature parameter can be retained. p A value greater than or equal to 0.05 indicates that the initial feature parameter has no significant difference between bleeding and non-bleeding users, and is therefore irrelevant to the user's bleeding risk; this initial feature parameter can be deleted. Through analysis of... p The magnitude of the value can be determined by filtering the initial feature parameters that affect the user's bleeding from the first feature parameter group and the second feature parameter group, thus obtaining the key feature parameters.

[0043] For the initial feature parameters of the classification, an observational approach is used for analysis. Specifically, the expected frequency check method is performed on the initial feature parameters in each of the q second feature parameter groups to obtain the expected frequency of the initial feature parameter in each cell of the second feature parameter group. If the expected frequency is less than 5, the Fisher exact test method is used to analyze the difference of the initial feature parameter between the bleeding user group and the non-bleeding user group, thus obtaining the initial feature parameter's... p If the expected frequency is greater than or equal to 5, then the chi-square test is used directly to analyze whether there is a difference in the classification distribution of this initial feature parameter between the bleeding user group and the non-bleeding user group, thus obtaining the initial feature parameter. p value.

[0044] Furthermore, through the analysis of p The magnitude of the value can be determined by filtering the initial feature parameters that affect user bleeding from the first and second feature parameter groups, thus obtaining the key feature parameters. If p A value less than 0.05 indicates a significant difference in the distribution of this initial feature parameter between the bleeding user group and the non-bleeding user group. This initial feature parameter poses a bleeding risk to users and can be retained. p If the value is greater than or equal to 0.05, it means that the initial feature parameter has no significant difference between the bleeding user group and the non-bleeding user group. The initial feature parameter is not related to the user's bleeding risk and can be deleted.

[0045] After obtaining k key feature parameters by analyzing the differences in the distribution of initial feature parameters between the bleeding user group and the non-bleeding user group, in order to avoid the decrease in the accuracy of the prediction model due to the high correlation between feature parameters, correlation analysis can be performed on the k key feature parameters to select one key feature parameter from multiple strongly correlated key feature parameters, thereby reducing the interference of redundant feature parameters on the prediction model.

[0046] Calculate the correlation coefficient matrix among k key feature parameters to determine the degree of linear correlation between them. Specifically, calculate the correlation coefficient between each pair of key feature parameters to obtain a symmetric matrix. Each value in the matrix represents the correlation coefficient between the corresponding two key feature parameters, where the value of the correlation coefficient ranges from [-1, 1]. The correlation coefficient between each pair of key feature parameters can be analyzed using either the Pearson correlation coefficient method or the Spearman rank correlation coefficient method. If both key feature parameters are quantitative parameters and follow a normal distribution, the Pearson correlation coefficient method is used to calculate the correlation coefficient; if the key feature parameters are ordinal categorical parameters or do not follow a normal distribution, the Spearman rank correlation coefficient method is used. If the correlation coefficient is greater than 0.7, it indicates that the two key feature parameters may be strongly correlated, meaning there is redundancy among these two key feature parameters, and one of the key feature parameters can be deleted. If the correlation coefficient is less than 0.7, it indicates that the two key feature parameters are not considered to have a strong correlation, and both key feature parameters can be retained.

[0047] By calculating the correlation coefficients between each pair of key feature parameters, redundant feature parameters can be removed from the k key feature parameters, reducing their interference with the prediction model and obtaining t feature parameters that are nonlinearly correlated with each other. Then, the variance inflation factor of the uncorrelated feature parameters is calculated to reflect the severity of multicollinearity between a certain uncorrelated feature parameter and the others, thereby improving the stability and interpretability of the prediction model.

[0048] The larger the variance inflation factor, the greater the degree to which the unrelated feature parameter is linearly explained by other unrelated feature parameters. This indicates more overlap (redundancy) between the feature and other features, and more severe multicollinearity, directly affecting the stability and interpretability of the regression model coefficients. Therefore, unrelated feature parameters with a variance inflation factor less than 10 are retained to obtain the final target feature parameters.

[0049] In this application, the feature parameters with significant inter-group differences are first retained through inter-group difference analysis to obtain key prediction parameters related to bleeding events; then, the feature parameters with strong correlations are excluded through correlation analysis and calculation of variance inflation factor, and the final target feature parameters are selected from n initial feature parameters. This allows the prediction model to be trained using the selected feature parameters, thereby reducing the complexity of the model and ensuring its robustness and generalization ability.

[0050] For example, the initial 31 characteristic parameters obtained are: age, sex, body surface area, weight, height, body mass index, heart rate, systolic blood pressure, diastolic blood pressure, left ventricular ejection fraction, left ventricular end-diastolic diameter, white blood cell count, hematocrit, platelet count, hemoglobin count, activated partial thromboplastin time, creatinine, creatinine clearance rate, aspartate aminotransferase (AST), alanine aminotransferase (ALT), hypertension, diabetes, heart failure, arrhythmia, history of stroke, chronic kidney disease, infection, abnormal access vessel, number of left coronary lesion sites, number of right coronary lesion sites, and total number of lesions. After intergroup difference analysis, correlation analysis, and calculation of variance inflation factor, the target characteristic parameters selected are: age, sex, body surface area, arrhythmia, history of stroke, chronic kidney disease, platelet count, hemoglobin, activated partial thromboplastin time, creatinine clearance rate, left ventricular ejection fraction, left ventricular end-diastolic diameter, total number of lesions, and number of left coronary lesions.

[0051] S330. Input the m target feature parameters into the target prediction model to obtain the bleeding result of the target user.

[0052] After obtaining m target feature parameters from the electronic medical record, these parameters are input into a pre-trained target prediction model, which outputs the predicted bleeding result. This target prediction model can be a logistic regression model.

[0053] In one possible example, the training method of the target prediction model includes: obtaining a j-th training dataset, the j-th training dataset comprising r training data groups, each training data group comprising m-j+1 target feature parameters of multiple users, where j is a positive integer; training the prediction model to be trained using the r-1 training data groups to obtain a target prediction model, and using a validation data group to determine whether the target prediction model has passed validation, the validation data group being the remaining training data groups excluding the r-1 training data groups; if the target prediction model fails validation, deleting the target prediction model from the training data group. p The target feature parameter with the largest value; let j = j + 1, repeat the above steps until the target prediction model is validated or the m-j+1 target feature parameters are found to have the largest value. p All values ​​are less than 0.05.

[0054] In this application, the selected target feature parameters are used to train the prediction model to be trained. By quantifying the linear correlation between the feature parameters and the logarithmic odds ratio of the probability of bleeding after ventricular assist device surgery, the interpretability of the bleeding risk outcome is achieved.

[0055] To improve the prediction accuracy of the target prediction model, this application employs a stratified sampling cross-validation strategy for model training and validation. First, the training dataset is stratified and grouped based on the ratio of bleeding to non-bleeding users, ensuring that each subset maintains the same bleeding / non-bleeding sample ratio as the training dataset. The model is then trained and validated sequentially across these subsets through multiple rounds of cross-validation to ensure its stability under different data distribution scenarios, avoid overfitting, and verify its generalization ability. During model training and cross-validation, the performance of the prediction model is evaluated by calculating dimensions such as discrimination, sensitivity, and specificity, verifying its ability to predict postoperative bleeding risk. Furthermore, when the performance of the prediction model does not meet the requirements, the target feature parameters are further optimized, eliminating those with low contribution to the prediction model and retaining features with core value for bleeding risk prediction, ultimately resulting in a high-precision and highly reliable target prediction model.

[0056] The training dataset includes m target feature parameters for multiple bleeding users and m target feature parameters for non-bleeding users. Based on the ratio of bleeding to non-bleeding users in the training dataset, a hierarchical partitioning method is used to divide the training dataset into r training data groups. The ratio of bleeding to non-bleeding users in both these r training data groups and the validation data groups is consistent with the bleeding / non-bleeding ratio in the training dataset. This ensures the training stability of the prediction model across different training data groups, avoids validation bias caused by sample distribution imbalance, and provides a reliable basis for model performance evaluation. The training model is trained using r-1 sets of training data. The remaining training data set is used as the validation data set to validate the performance of the trained model. The validation data set consists of the 1st, 2nd, 3rd, ..., rth sets of the r training data sets. The r-1 training data sets consist of the 2nd to rth sets, the 1st + 3rd to rth sets, the 1st + 2nd + 4th to rth sets, ..., the 1st to r-1th sets of the r training data sets. The model is trained r times and validated r times. If the trained prediction model passes the validation, it indicates that the currently trained prediction model has the ability to accurately predict the bleeding risk of the user after ventricular assist device surgery. If the trained prediction model fails the validation, it indicates that there may be redundant features or interference terms (i.e. features that contribute little to the bleeding risk prediction or even introduce noise) among the selected m target feature parameters. In this case, it is necessary to further optimize the m target feature parameters, reconstruct the training dataset using the optimized target features, and repeat the above "training-validation" process until the prediction model passes the validation. Finally, the validated prediction model is determined as the target prediction model.

[0057] Optionally, the step of training the prediction model to be trained using r-1 sets of training data to obtain a target prediction model, and using a validation set to determine whether the target prediction model is validated successfully, includes: determining the i-th training data set as the i-th validation data set, where i is a positive integer; training the prediction model to be trained using the remaining r-1 sets of training data excluding the i-th validation data set to obtain the target prediction model; inputting the i-th validation data set into the target prediction model and outputting the i-th estimated bleeding result; calculating the i-th discrimination, i-th sensitivity, and i-th specificity based on the i-th estimated bleeding result and the actual bleeding result; setting i=i+1 and repeating the above steps until i=r; calculating the average values ​​of the i-th discrimination, i-th sensitivity, and i-th specificity to obtain the average discrimination, average sensitivity, and average specificity; if the following conditions are not met: the average discrimination is greater than a first threshold, the average sensitivity is greater than a second threshold, and the average specificity is greater than a third threshold, then the target prediction model is determined to be validated successfully; otherwise, the target prediction model is determined to be unqualified.

[0058] This application uses an r-fold cross-validation loop to train and validate the target prediction model. First, the first training data set is used as the validation data set, and the remaining r-1 training data sets are used as the training data. The prediction model to be trained is then trained using these r-1 training data sets. After training, the trained prediction model is validated using the validation data sets. This validation data set is input into the trained prediction model to obtain the predicted bleeding result for each user. Based on the predicted bleeding result and the actual bleeding result, the discrimination, sensitivity, and specificity of the validation data set are calculated. Discrimination indicates the current prediction model's ability to distinguish between bleeding users and non-bleeding users, with a value between 0.5 and 1; the closer the discrimination is to 1, the stronger the discrimination ability of the current prediction model. Sensitivity indicates the current prediction model's ability to identify bleeding users, with a value between 0 and 1; the closer the sensitivity is to 1, the stronger the identification ability of the current prediction model and the fewer missed cases. Specificity represents the ability of the current prediction model to exclude users without bleeding risk; its value ranges from 0 to 1. The closer the specificity is to 1, the more accurately the current prediction model can identify users without bleeding risk, and the fewer false positives it has. Then, the second training data set is used as the validation data set, and the remaining r-1 training data sets are used as training data. The prediction model is trained using these r-1 training data sets. After training, the trained prediction model is validated using the validation data sets to obtain the predicted bleeding result for each user, and the discrimination, sensitivity, and specificity of the validation data set are calculated. Following this method, the i-th training data set is selected from the r training data sets for validation. The remaining training data sets are used to train the prediction model, and the i-th training data set is used as the validation data set to validate the trained prediction model, calculating the discrimination, sensitivity, and specificity of the validation data set, until i=r. This yields the discrimination, sensitivity, and specificity of the r validation data sets.

[0059] This application ensures that each training data set is validated using this method. A logistic regression model is fitted to the training data set, and its performance is evaluated on the validation data set. The average of the r validation results is then used as the generalization performance metric of the prediction model. Specifically, the average of the discrimination, sensitivity, and specificity of the r validation data sets is calculated to obtain the average discrimination, average sensitivity, and average specificity. If the average discrimination is greater than or equal to a first threshold, the average sensitivity is greater than or equal to a second threshold, and the average specificity is greater than or equal to a third threshold (i.e., all three requirements are met), then the current prediction model is considered optimal and can be used as the target prediction model. If the average discrimination, average sensitivity, and average specificity all meet the requirements (i.e., the average discrimination is less than the first threshold, the average sensitivity is less than the second threshold, or the average specificity is less than the third threshold), then the prediction accuracy of the current prediction model has not yet met the requirements and further optimization is needed.

[0060] The first, second, and third thresholds can be set according to the application scenario. For example, for general application scenarios, the first threshold can be set to 0.75, the second threshold to 0.75, and the third threshold to 0.75; for scenarios that need to prioritize the identification of bleeding users, the first threshold can be set to 0.80, the second threshold to 0.85, and the third threshold to 0.65; for scenarios that need to prioritize the prevention of false positives, the first threshold can be set to 0.80, the second threshold to 0.70, and the third threshold to 0.85.

[0061] In logistic regression models, the significance of the model coefficients can be tested (i.e., p Value tests are used to determine whether a significant statistical association exists between a single target feature parameter (predictor variable) and the bleeding outcome (dependent variable). p The larger the value, the higher the probability that the target feature parameter is not associated with the bleeding outcome, that is, the weaker its correlation with bleeding risk prediction and the lower its contribution; conversely, p The smaller the value, the more significant the statistical association between the target feature parameter and the bleeding outcome, and the higher its contribution to prediction. Based on the above... p The value test results are used to optimize the m target feature parameters, prioritizing the removal of parameters from the target prediction model. p The target feature parameter with the largest value is selected to eliminate redundant information, reduce model noise, and improve the model's prediction accuracy and interpretability.

[0062] After optimizing the target feature parameters, a training dataset is acquired to train the prediction model. The training dataset includes m-1 target feature parameters for multiple bleeding users and m-1 target feature parameters for non-bleeding users. The training dataset is divided into r training data groups, where the ratio of bleeding users to non-bleeding users is equal in each of the r training data groups. The validation data group is the i-th training data group among the r training data groups. Hierarchical cross-validation is performed on the prediction model using the r-1 training data groups and the validation data groups. As mentioned above, the first training data group is used as the validation data group, and the remaining r-1 training data groups are used as the training data. The prediction model to be trained is trained using the r-1 training data groups. After training, the trained prediction model is validated using the validation data groups to obtain the predicted bleeding result for each user. The discriminancy, sensitivity, and specificity of the validation data group are calculated. Following this method, the discriminancy, sensitivity, and specificity of the r validation data groups are calculated respectively. Next, calculate the average of the discrimination, sensitivity, and specificity of the r validation data sets. Based on these averages, determine if the current prediction model is optimal. If all three values ​​meet the requirements (i.e., the trained prediction model is validated), the loop ends, and the currently trained prediction model is designated as the target prediction model. If any one of these values ​​fails to meet the requirements, the target feature parameters need to be optimized further. Following the above method, the dataset is reconstructed using the optimized target feature parameters, and the prediction model is trained. This iterative process continues until all three values ​​meet the requirements, or until all remaining target feature parameters are optimized. p All values ​​are less than 0.05.

[0063] Furthermore, to improve the efficiency of predictive model training and avoid ineffective feature optimization iterations, the accuracy of the target predictive model needs to be better after optimizing the target feature parameters. This application adds a feature optimization effectiveness judgment rule: after deleting a target feature parameter, if the average discrimination of the predictive model trained on the training dataset is less than or equal to the average discrimination of the original model before deleting the feature, it indicates that deleting a target feature parameter does not improve the accuracy of the predictive model, that is, the predictive model before deletion has reached the optimal state based on the current feature combination. On this basis, if the average discrimination, average sensitivity, and average specificity of the optimal predictive model still cannot meet all preset threshold requirements, the predictive model is considered unsuccessful and the training of the predictive model can be terminated; if the model in the optimal state has met all preset threshold requirements, the current target feature set is directly determined as the optimal feature combination, and then the predictive model is retrained using the entire original training dataset based on the optimal feature combination to fully mine the effective information in the full sample, and finally obtain a final target predictive model with more stable generalization performance and higher prediction accuracy.

[0064] For example, a first training dataset is obtained, consisting of 50 users with bleeding episodes and 200 users without bleeding episodes. This dataset is divided into 10 training data groups at a 1:4 ratio. Each training data group contains 14 target feature parameters for 5 users with bleeding episodes and 14 target feature parameters for 20 users without bleeding episodes. The first training data group is used as the validation data group. The second to tenth training data groups are then input into the prediction model to be trained, resulting in a trained prediction model. The first training data group is then used as the validation data group and input into the trained prediction model, outputting 25 predicted bleeding results. Based on these 25 predicted bleeding results and the actual bleeding results of the 25 users, the discriminant, sensitivity, and specificity of the first validation data group are calculated. The second training data set is then used as the validation data set. The first, third, through tenth training data sets are input into the previously trained prediction model for further training, resulting in a trained prediction model. The second training data set is then used as the validation data set and input into the trained prediction model, outputting 25 predicted bleeding results. Based on these 25 predicted bleeding results and the actual bleeding results of 25 users, the discrimination, sensitivity, and specificity of the second validation data set are calculated. Following this method, the prediction model is trained separately, and the discrimination, sensitivity, and specificity of each validation data set are calculated, resulting in 10 values ​​for discrimination, sensitivity, and specificity. The average of these 10 values ​​is then calculated. If the average discrimination, sensitivity, and specificity are ≥0.75, ≥0.75, and ≥0.75, respectively, the currently trained prediction model is considered to have achieved the required prediction accuracy and can be used as the target prediction model. Otherwise, interference is considered to exist in the current 14 target feature parameters, and the corresponding values ​​of these 14 target feature parameters in the current prediction model are adjusted accordingly. pThe target feature parameter with the largest value is removed, leaving 13 target feature parameters. A second training dataset is obtained, containing 50 bleeding users and 200 non-bleeding users. The first training dataset is divided into 10 training data groups at a 1:4 ratio. Each training data group contains the 13 target feature parameters from 5 bleeding users and the 13 target feature parameters from 20 non-bleeding users. The first training data group is used as the validation data group. The second to tenth training data groups are input into the previously trained prediction model to obtain a trained prediction model. The first training data group is then used as the validation data group and input into the trained prediction model, outputting 25 predicted bleeding results. Based on these 25 predicted bleeding results and the actual bleeding results of the 25 users, the discrimination, sensitivity, and specificity of the first validation data group are calculated. The second training data set is then used as the validation data set. The first, third, through tenth training data sets are input into the previously trained prediction model for further training, resulting in a well-trained prediction model. This second training data set is then used as the validation data set and input into the trained prediction model, outputting 25 predicted bleeding results. Based on these 25 predicted bleeding results and the actual bleeding results of 25 users, the discrimination, sensitivity, and specificity of the second validation data set are calculated. This method is repeated to train the prediction model and calculate the discrimination, sensitivity, and specificity for each validation data set, resulting in 10 values ​​for discrimination, sensitivity, and specificity. The average of these 10 values ​​is then calculated. If the average discrimination, sensitivity, and specificity are ≥0.75, ≥0.75, and ≥0.75, respectively, the currently trained prediction model is considered to have achieved the required prediction accuracy and can be used as the target prediction model. Otherwise, interference is considered to exist among the 13 target feature parameters, and the corresponding values ​​of these 13 target feature parameters in the current prediction model are adjusted accordingly. p The target feature parameter with the largest value is removed, leaving 12 target feature parameters. The prediction model is trained using the method described above until the average discrimination ≥ 0.75, the average sensitivity ≥ 0.75, and the average specificity ≥ 0.75, or the target feature parameters... p All values ​​are less than 0.05. The prediction model after the cycle ends is used as the target prediction model. This target prediction model is then used to predict the postoperative bleeding risk for users who need ventricular assist devices implanted, thereby improving user safety.

[0065] As can be seen, this application proposes a bleeding prediction method, which involves obtaining the target user's electronic medical record; selecting m target feature parameters from the electronic medical record, which are physiological parameters affecting the bleeding outcome of the target user; and inputting the m target feature parameters into a target prediction model to obtain the bleeding outcome of the target user. This application, by selecting feature parameters affecting bleeding after ventricular assist device (VAD) surgery from the user's electronic medical record and inputting these feature parameters into a pre-trained target prediction model to obtain the user's bleeding outcome, can predict before surgery whether the patient will have bleeding problems after VAD surgery, reducing the risk of postoperative bleeding, improving surgical prognosis and recovery, and thus enhancing user safety.

[0066] The above primarily describes the solutions of the embodiments of this application from the perspective of the method execution process. It is understood that, in order to achieve the above functions, the network device includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments provided herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0067] For example, this application provides a bleeding prediction device that can be connected to a hospital's medical system and can acquire various patient information. The bleeding prediction device also includes one or more processors, which are used for: Obtain the target user's electronic medical records; m target feature parameters are selected from the electronic medical record. The target feature parameters are physiological parameters that affect the bleeding outcome after the ventricular assist device is used on the target user. m is a positive integer. The m target feature parameters are input into the target prediction model to obtain the bleeding result of the target user.

[0068] For example, this application also provides a medical device that includes the bleeding prediction device described above.

[0069] The control unit of each of the above solutions has the function of implementing the corresponding steps performed by the medical device in the above method; the function can be implemented by hardware or by hardware executing corresponding software.

[0070] In embodiments of this application, the processor may also be a chip or a chip system, such as a system-on-a-chip (SoC).

[0071] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a medical device provided in an embodiment of this application. The medical device includes: one or more processors, one or more memories, one or more communication interfaces, and one or more programs; the one or more programs are stored in the memories and configured to be executed by the one or more processors.

[0072] The above procedure includes instructions for performing the following steps: Obtain the target user's electronic medical records; m target feature parameters are selected from the electronic medical record. The target feature parameters are physiological parameters that affect the bleeding outcome after the ventricular assist device is used on the target user. m is a positive integer. The m target feature parameters are input into the target prediction model to obtain the bleeding result of the target user.

[0073] All relevant content in each scenario involved in the above method embodiments can be referenced from the functional description of the corresponding functional module, and will not be repeated here.

[0074] It should be understood that the aforementioned memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory. For example, the memory may also store information about the device type.

[0075] In the embodiments of this application, the processor of the above-described device may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0076] It should be understood that "at least one" in the embodiments of this application refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0077] Furthermore, unless otherwise stated, the ordinal numbers such as "first" and "second" mentioned in the embodiments of this application are used to distinguish multiple objects and are not used to limit the order, timing, priority, or importance of multiple objects. For example, "first information" and "second information" are only used to distinguish different information and do not indicate differences in the content, priority, sending order, or importance of these two types of information.

[0078] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly manifested as execution by a hardware processor, or as a combination of hardware and software units within the processor. The software units can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor executes the instructions in the memory, combining them with its hardware to complete the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0079] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the methods described in the above method embodiments.

[0080] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the methods described in the above method embodiments. This computer program product can be a software installation package.

[0081] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0082] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0083] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0084] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0085] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0086] If the aforementioned integrated units are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or TRP, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0087] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include flash drives, ROM, RAM, magnetic disks, or optical disks, etc.

[0088] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting bleeding, characterized in that, The method includes: Obtain the target user's electronic medical records; m target feature parameters are selected from the electronic medical record. The target feature parameters are physiological parameters that affect the bleeding outcome after the ventricular assist device is used on the target user. m is a positive integer. The m target feature parameters are input into the target prediction model to obtain the bleeding result of the target user.

2. The method according to claim 1, characterized in that, Before selecting m target feature parameters from the electronic medical record, the method further includes: Obtain n initial feature parameter groups, each initial feature parameter group including one feature parameter of bleeding users and non-bleeding users, wherein the n are positive integers greater than m; The feature parameters are filtered based on the n initial feature parameter groups to obtain the m target feature parameters.

3. The method according to claim 2, characterized in that, The step of filtering the feature parameters according to the n initial feature parameter groups to obtain the m target feature parameters includes: Inter-group difference analysis was performed on the n initial feature parameter groups to select k key feature parameters, where k is a positive integer less than n and greater than m. These key feature parameters are statistically significant. p The value is less than 0.05; Correlation analysis is performed on the k key feature parameters to filter out t irrelevant feature parameters. The correlation coefficient of the t irrelevant feature parameters is greater than 10, and t is a positive integer less than n and greater than m. Calculate the variance inflation factor of the t unrelated feature parameters, and exclude the unrelated feature parameters whose variance inflation factor is greater than or equal to a preset third value to obtain the m target feature parameters.

4. The method according to claim 2, characterized in that, The step involves performing inter-group difference analysis on the n initial feature parameters to select k key feature parameters, including: The n initial feature parameter groups are classified according to their data types to obtain p first feature parameter groups and q second feature parameter groups, where p and q are positive integers; Perform a normality test on each of the p sets of first feature parameters to obtain the normality of each set of first feature parameters. p value; The expected frequency of each of the q second feature parameter groups is checked to obtain the expected frequency of each second feature parameter group; If the expected frequency is greater than or equal to 5, the Fisher exact test is applied to the q sets of second characteristic parameters; if the expected frequency is less than 5, the chi-square test is applied to the q sets of second characteristic parameters to obtain the results for each set of second characteristic parameters. p value; If the above p If the value is greater than or equal to 0.05, then the first feature parameter group or the second feature parameter group is deleted to obtain the k key feature parameters.

5. The method according to claim 1, characterized in that, The training method for the target prediction model includes: Obtain the j-th training dataset, which includes r training data groups, each of which includes the m-j+1 target feature parameters of multiple users, where j is a positive integer; The target prediction model is obtained by training the prediction model to be trained using r-1 sets of training data, and the target prediction model is qualified by using a set of validation data. The set of validation data consists of the remaining training data sets other than the r-1 sets of training data. If the target prediction model fails the validation, then delete the target prediction model. p The target feature parameter with the largest value; Let j = j + 1, and repeat the above steps until the target prediction model is validated or the m - j + 1 target feature parameters are satisfied. p All values ​​are less than 0.

05.

6. The method according to claim 5, characterized in that, The step of training the prediction model to be trained using r-1 sets of training data to obtain the target prediction model includes: The i-th training data group is determined as the i-th validation data group, where i is a positive integer; The target prediction model is obtained by training the prediction model using the remaining r-1 training data sets excluding the i-th validation data set. The target prediction model is input using the i-th validation data set, and the i-th estimated bleeding result is output. The i-th discrimination, i-th sensitivity, and i-th specificity are calculated based on the i-th estimated bleeding result and the actual bleeding result. Let i = i + 1, and repeat the above steps until i = r.

7. The method according to claim 6, characterized in that, The step of using the validation data set to determine whether the target prediction model has passed validation includes: Calculate the average values ​​of the i-th discrimination, the i-th sensitivity, and the i-th specificity to obtain the average discrimination, average sensitivity, and average specificity; If the following conditions are not met: the average discrimination is greater than the first threshold, the average sensitivity is greater than the second threshold, and the average specificity is greater than the third threshold, then the target prediction model is determined to be qualified; otherwise, the target prediction model is determined to be unqualified.

8. The method according to any one of claims 5-7, characterized in that, The proportion of bleeding users and non-bleeding users is equal in all r training data sets.

9. A bleeding prediction device, characterized in that, The bleeding prediction device includes one or more processors, which are configured to perform the following steps: Obtain the target user's electronic medical records; m target feature parameters are selected from the electronic medical record. The target feature parameters are physiological parameters that affect the bleeding outcome after the ventricular assist device is used on the target user. m is a positive integer. The m target feature parameters are input into the target prediction model to obtain the bleeding result of the target user.

10. A medical device, characterized in that, The device includes a processor, a memory, and a communication interface. The memory stores one or more programs, and the one or more programs are executed by the processor. The one or more programs include instructions for performing the steps of the method as described in any one of claims 1-8.

Images