Cardiac death prediction model for acute myocardial infarction patient and working method of cardiac death prediction model

Abstract

The invention relates to the technical field of medical prediction, and discloses an acute myocardial infarction patient cardiogenic death prediction model and a working method thereof, and the model comprises a data fusion module, an initial prediction module, a dynamic monitoring module and a prediction correction module. The working method corresponds to the model. According to the application, the data fusion module integrates multi-source data of clinical basis, inspection, treatment, environment, follow-up visit and the like, and the prediction comprehensiveness is improved; the dynamic monitoring module synchronously collects multi-source fusion data changes and risk prediction value changes, and the problem that a traditional static model is difficult to adapt to rapid fluctuation of the illness state is solved by combining multi-dimensional adjustment of the prediction correction module; based on an initial prediction module and a prediction correction module, under the condition of considering prediction efficiency and precision, the clinical real-time decision-making requirement is better met, and a reliable basis is provided for early intervention of cardiac death after acute myocardial infarction.

Description

Technical Field

[0001] This application relates to the field of medical prediction technology, specifically a predictive model for cardiac death in patients with acute myocardial infarction and its working method. Background Technology

[0002] Several predictive tools are currently available for assessing the risk of adverse cardiovascular events following acute myocardial infarction (AMI), such as the GRACE score and TIMI risk score. These tools are based on traditional statistical methods and rely primarily on a limited set of clinical variables, such as age, blood pressure, and electrocardiogram changes, for risk assessment.

[0003] However, the data sources of existing tools are relatively limited, usually only including basic clinical indicators (such as troponin levels and electrocardiogram results) and some demographic information (such as age and gender), lacking the integration of multi-dimensional data, such as environmental factors (temperature, PM2.5), detailed test results (such as color Doppler ultrasound reports), and treatment variables (such as angiography records and surgical scores).

[0004] Furthermore, traditional risk prediction models often employ linear models such as Logistic regression, which struggle to capture the high-dimensional nonlinear relationships between clinical characteristics, resulting in insufficient prediction accuracy and poor generalization ability to external cohorts.

[0005] Chinese invention patent application CN120032877A discloses a method, system, electronic device and storage medium for predicting postoperative cardiac death risk, but it does not analyze the dynamic correlation between postoperative multidimensional data changes and cardiac death risk, and cannot achieve real-time prediction optimization.

[0006] In conclusion, there is an urgent need for a new technical solution for predicting cardiac death in patients with acute myocardial infarction. Summary of the Invention

[0007] The purpose of this application is to provide a predictive model for cardiac death in patients with acute myocardial infarction and its working method, so as to solve the technical problems mentioned in the background art.

[0008] To achieve the above objectives, this application discloses the following technical solutions: In a first aspect, this application discloses a predictive model for cardiac death in patients with acute myocardial infarction, the model comprising: The data fusion module is used to acquire and fuse medical multi-source data of patients with acute myocardial infarction to obtain multi-source fused data; wherein the medical multi-source data includes at least clinical basic data, laboratory examination data, treatment data, environmental data, and follow-up data; The initial prediction module is used to obtain the corresponding risk prediction value of cardiac death based on the multi-source fusion data; The dynamic monitoring module is used to continuously collect the first change information of the multi-source fusion data and the second change information of the risk prediction value. The first change information includes the difference, change rate and trend characteristics calculated for at least one type of medical multi-source data and the significant change information is marked. The second change information includes the numerical difference, fluctuation frequency and change direction of the risk prediction value, and the second change information is associated with the corresponding first change information. The prediction correction module is used to adjust the prediction logic of the initial prediction module based on the first change information and the second change information to obtain the risk prediction value of the next cardiac death.

[0009] Preferably, the first change information is obtained in the following way: For at least one type of medical multi-source data, the difference, rate of change, and trend characteristics of the multi-source fusion data at the current moment and the multi-source fusion data at the previous acquisition moment are calculated in real time; wherein, the trend characteristics include an upward trend, a downward trend, or a fluctuation amplitude within multiple acquisition cycles, and the difference greater than a preset first threshold or the rate of change greater than a preset second threshold is marked as significant change information to obtain the first change information.

[0010] Preferably, the second change information is obtained in the following manner: The system records the numerical differences, fluctuation frequency, and direction of change of multiple consecutive risk prediction values. The numerical difference is the ratio of the absolute difference between two adjacent risk prediction values ​​to the previous risk prediction value. The fluctuation frequency is the number of times the risk prediction value crosses a preset risk level threshold within a preset unit of time. The direction of change includes risk increase, decrease, or stabilization. When the numerical difference is greater than a preset third threshold or the number of consecutive risk prediction values ​​crossing the risk level threshold is greater than a preset fourth threshold, it is determined that the second change information exists, and the corresponding first change information is associated with it.

[0011] Preferably, adjusting the prediction logic of the initial prediction module includes: Based on the medical multi-source data type corresponding to the significant change information marked in the first change information, the feature weight of this type of data in the initial prediction module is increased. Based on the numerical differences and fluctuation frequency of the second change information, the sensitivity coefficient of the initial prediction module is adjusted; By using the correlation between the first change information and the second change information, the mapping relationship between the corresponding data type and the risk prediction value in the initial prediction module is corrected.

[0012] Preferably, the correlation between the first change information and the second change information includes the probability of risk change, the degree of change matching, and the degree of change correlation; wherein: The probability of risk change is the probability of the change in the risk prediction value in the second change information corresponding to the medical multi-source data type marked in the first change information. The change matching degree is the degree of deviation between the time interval between the moment when the multi-source fused data undergoes a significant change and the time interval between the corresponding change in the risk prediction value and a preset time window. The correlation degree of change is a quantitative correlation between the magnitude of change of the first change information and the numerical difference of the second change information.

[0013] Preferably, correcting the mapping relationship between the corresponding data type and the risk prediction value in the initial prediction module includes: When the probability of risk change meets the preset probability condition, the mapping weight of the medical multi-source data type with the corresponding risk prediction value change direction in the initial prediction module is increased, and the increase in weight is positively correlated with the probability of risk change. When the change matching degree satisfies that the time interval is within the time window or its deviation degree satisfies the preset deviation condition, the dynamic mapping relationship between the multi-source fusion data and the risk prediction value corresponding to the time series is strengthened. The mapping function parameters are adjusted based on the preset correlation interval in which the change correlation degree is located. The correlation interval includes a high correlation interval, a medium correlation interval, and a low correlation interval.

[0014] Preferably, adjusting the mapping function parameters based on the preset correlation interval where the change in correlation degree is located includes: When the correlation of the change is in the high correlation range, the impact of the data change on the risk prediction value is amplified. When the degree of correlation of the change is in the medium correlation range, the original influence strength is maintained; When the correlation of the change is in the low correlation range, a preset attenuation coefficient is introduced to reduce the interference of this data type on the risk prediction value.

[0015] Preferably, the basic clinical data includes at least the age, sex, BMI, smoking history, and history of heart disease of patients with acute myocardial infarction; The test data include at least coronary angiography, myocardial enzyme profile, electrocardiogram, echocardiography, and hemodynamic parameters; The treatment data includes at least the timing of reperfusion therapy, medication records, and information on coronary interventional therapy. The environmental data includes at least the temperature at the time of onset, PM2.5 concentration, and diurnal rhythm; The follow-up data includes at least records of cardiac death events and malignant arrhythmia episodes.

[0016] Preferably, the initial prediction module obtains the risk prediction value of cardiac death based on a lightweight gradient booster.

[0017] Secondly, this application discloses a working method for a predictive model of cardiac death in patients with acute myocardial infarction, applied to the predictive model of cardiac death in patients with acute myocardial infarction as described above. The working method includes the following steps: Step 1: Acquire and fuse medical data from multiple sources for patients with acute myocardial infarction to obtain multi-source fused data; Step 2: Based on the multi-source fusion data, obtain the corresponding risk prediction value for cardiac death; Step 3: Continuously collect the first change information of the multi-source fusion data and the second change information of the risk prediction value; the first change information includes the difference, change rate and trend characteristics calculated for at least one type of medical multi-source data and marked with significant change information; the second change information includes the numerical difference, fluctuation frequency and change direction of the risk prediction value, and the second change information is associated with the corresponding first change information. Step 4: Based on the first and second change information, adjust the prediction logic of Step 2 to obtain the risk prediction value of the next cardiac death.

[0018] Beneficial Effects: The cardiac mortality prediction model and its working method for patients with acute myocardial infarction in this application integrate multiple sources of data, including clinical basic data, laboratory examinations, treatment, environment, and follow-up data, through a data fusion module, breaking through the limitations of traditional single data sources and improving the comprehensiveness of predictions. The dynamic monitoring module simultaneously collects changes in multi-source fused data and changes in risk prediction values, and combined with the multi-dimensional adjustments of the prediction correction module, it solves the problem that traditional static models are difficult to adapt to rapid fluctuations in the condition. Through feature weight enhancement, sensitivity coefficient adjustment, and dynamic correction of mapping relationships, especially the fine optimization based on the probability of risk changes, the degree of matching of changes, and the degree of correlation of changes, the prediction sensitivity and accuracy are significantly improved. The initial prediction uses a lightweight gradient boosting machine, which, while taking into account prediction efficiency and accuracy, is more in line with the needs of real-time clinical decision-making, providing a reliable basis for early intervention in cardiac mortality after acute myocardial infarction. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A structural block diagram of the cardiac death prediction model for patients with acute myocardial infarction provided in the embodiments of this application; Figure 2 A flowchart illustrating the working method of the cardiac death prediction model for acute myocardial infarction patients provided in this application embodiment. Detailed Implementation

[0021] The technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0022] In this document, the term "comprising" is intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0023] Existing models for predicting cardiac death in patients with acute myocardial infarction mostly rely on static data or a single data source, lacking closed-loop feedback on dynamic changes in data and prediction results, making them difficult to adapt to the rapidly fluctuating clinical needs. Therefore, this application provides a model that includes data fusion, initial prediction, dynamic monitoring, and prediction correction. Through multi-module collaboration, it achieves a complete process from multi-source data fusion to dynamic correction of prediction logic, solving the problems of staticity and lack of feedback in traditional models, and significantly improving the timeliness and accuracy of predictions.

[0024] Example 1 like Figure 1 As shown in this embodiment, a model for predicting cardiac death in patients with acute myocardial infarction is disclosed. The model includes: The data fusion module is used to acquire and fuse medical multi-source data of patients with acute myocardial infarction to obtain multi-source fused data; wherein the medical multi-source data includes at least clinical basic data, laboratory examination data, treatment data, environmental data, and follow-up data; The initial prediction module is used to obtain the corresponding risk prediction value of cardiac death based on multi-source fusion data; The dynamic monitoring module is used to continuously collect first change information of multi-source fusion data and second change information of risk prediction values; it is used to continuously collect the first change information of the multi-source fusion data and the second change information of the risk prediction values, wherein the first change information includes the difference, change rate and trend characteristics calculated for at least one type of medical multi-source data and marking significant change information, and the second change information includes the numerical difference, fluctuation frequency and change direction of the risk prediction values, and the second change information is associated with the corresponding first change information; The prediction correction module is used to adjust the prediction logic of the initial prediction module based on the first change information and the second change information to obtain the risk prediction value of the next cardiac death.

[0025] Traditional models often limit their monitoring of changes in multi-source data to a single dimension (such as focusing only on numerical differences), making it difficult to accurately identify key changes. In this embodiment, the collection of first change information involves calculating differences, rates of change, and trend characteristics, and marking significant changes. This achieves multi-dimensional quantification of changes in multi-source fused data, providing accurate data change information for subsequent model corrections and avoiding prediction bias caused by missing key changes.

[0026] Specifically, the first change information is obtained through the following methods: For at least one type of medical multi-source data, the difference, rate of change, and trend characteristics of the multi-source fusion data at the current moment and the multi-source fusion data at the previous acquisition moment are calculated in real time. Among them, the trend characteristics include the upward trend, downward trend, or fluctuation amplitude within multiple acquisition cycles, and the difference greater than a preset first threshold or the rate of change greater than a preset second threshold is marked as significant change information to obtain the first change information.

[0027] As a prediction result, the changes in risk prediction values ​​are often overlooked or simply recorded without being correlated with data changes, resulting in a lack of targeted model correction. In this embodiment, the collection of second change information records the numerical differences, fluctuation frequency, and direction of change of the prediction values, and correlates them with the corresponding first change information. This makes the dynamic characteristics of the prediction values ​​quantifiable and traceable, providing clear predictive feedback for model correction and improving the targeted nature of the correction logic.

[0028] Specifically, the second change information is obtained through the following methods: Record the numerical differences, fluctuation frequency, and direction of change of multiple consecutive risk prediction values; wherein, the numerical difference is the ratio of the absolute difference between two adjacent risk prediction values ​​to the previous risk prediction value, the fluctuation frequency is the number of times the risk prediction value crosses a preset risk level threshold within a preset unit of time, and the direction of change includes risk increase, decrease, or stabilization; when the numerical difference is greater than a preset third threshold or the number of consecutive risk prediction values ​​crossing the risk level threshold is greater than a preset fourth threshold, it is determined that there is second change information, and the corresponding first change information is associated with it.

[0029] Traditional models often adjust the prediction logic in a single dimension, such as adjusting only weights, without considering the dual changes in data and predicted values, resulting in limited adjustment effects. In this embodiment, the prediction logic is adjusted by increasing feature weights, adjusting sensitivity coefficients, and correcting mapping relationships. By synergistically utilizing both first and second change information, multi-dimensional dynamic optimization of the model is achieved, solving the problem of the one-sidedness of traditional adjustment methods and significantly improving the model's adaptability.

[0030] Specifically, the prediction logic of the initial prediction module is adjusted, including: Based on the medical multi-source data types corresponding to the significant change information marked in the first change information, the feature weight of this type of data in the initial prediction module is increased; Based on the numerical differences and fluctuation frequency of the second change information, the sensitivity coefficient of the initial prediction module is adjusted; By analyzing the correlation between the first and second change information, the mapping relationship between the corresponding data type and the risk prediction value in the initial prediction module is corrected.

[0031] Furthermore, the lack of a clear analysis of the correlation between data changes and predicted value changes leads to a lack of quantitative basis for model adjustments. This embodiment defines three parameters—risk change probability, change matching degree, and change correlation degree—to transform the abstract correlation into quantifiable indicators, making the relationship between data changes and predicted value changes clearer. This provides a precise quantitative basis for subsequent mapping relationship adjustments and avoids blind adjustments.

[0032] Specifically, the relationship between the first and second changes includes the probability of risk change, the degree of matching of changes, and the degree of correlation of changes; among which: The probability of risk change is the probability of the change in the risk prediction value corresponding to the significant change information marked in the first change information and the corresponding medical multi-source data type. The change matching degree is the degree of deviation between the time interval between the moment when significant changes occur in the multi-source fusion data and the corresponding changes in the risk prediction value and the preset time window. The correlation degree of change is the quantitative correlation between the magnitude of change in the first change information and the numerical difference in the second change information.

[0033] In practical applications, the mapping relationship between data and predicted values ​​is often neglected during model correction, leading to a disconnect between the two and the actual correlation. This embodiment addresses this by adjusting mapping weights, strengthening dynamic mapping, or optimizing function parameters based on the probability of risk change, the degree of matching change, and the correlation interval. This makes the mapping relationship more closely reflect the actual correlation between data and predicted values, improving the model's adaptability to complex clinical scenarios.

[0034] Specifically, the mapping relationship between the corresponding data types and risk prediction values ​​in the initial prediction module is corrected, including: When the probability of risk change meets the preset probability conditions, the mapping weight of this medical multi-source data type with the corresponding risk prediction value change direction in the initial prediction module is increased, and the increase in weight is positively correlated with the probability of risk change. When the degree of change matches the time interval within the time window or the degree of deviation meets the preset deviation conditions, the dynamic mapping relationship between the multi-source fusion data and the risk prediction value corresponding to the time series is strengthened. The mapping function parameters are adjusted based on the preset correlation interval where the correlation degree changes. The correlation interval includes high correlation interval, medium correlation interval and low correlation interval.

[0035] Furthermore, applying a uniform processing method to data types with different correlation strengths can lead to the underestimation of the impact of strongly correlated data or the amplification of interference from weakly correlated data, affecting the accuracy of risk prediction. This embodiment employs processing strategies such as amplifying the impact, maintaining the strength, or introducing a decay coefficient, based on the differences in the range of varying correlation strengths, to match the data's influence on the predicted value with the correlation strength, reduce invalid interference, and improve the stability of the prediction results.

[0036] Specifically, the mapping function parameters are adjusted based on the preset correlation interval where the correlation degree changes, including: When the correlation of changes is in the high correlation range, the impact of data changes on risk prediction values ​​is amplified. When the degree of correlation of the change is in the medium correlation range, the original influence strength is maintained; When the correlation of change is in the low correlation range, a preset attenuation coefficient is introduced to reduce the interference of this data type on the risk prediction value.

[0037] Traditional predictive models often rely on data limited to clinical laboratory indicators, neglecting key factors such as environment, treatment, and follow-up, resulting in insufficient predictive comprehensiveness. This embodiment utilizes multi-source medical data encompassing clinical baseline data, laboratory examinations, treatment, environment, and follow-up data, achieving comprehensive coverage of multi-dimensional factors influencing heart failure risk and laying a more complete data foundation for accurate prediction.

[0038] Specifically, basic clinical data should include at least the patient's age, sex, BMI, smoking history, and history of heart disease. The laboratory test data should include at least the following: coronary angiography (number of diseased vessels, degree of stenosis), myocardial enzyme profile (peak troponin, dynamic changes in CK-MB), electrocardiogram (malignant arrhythmias, ST segment elevation), echocardiography (left ventricular ejection fraction, ventricular aneurysm formation), and hemodynamic parameters; Treatment data should include at least the timing of reperfusion therapy, medication records (antiplatelet drugs, beta-blockers, etc.), and information on coronary intervention. Environmental data should include at least the temperature at the time of onset, PM2.5 concentration, and diurnal rhythm; Follow-up data should include at least records of cardiac death events and malignant arrhythmia episodes.

[0039] It should be noted that the medical multi-source data in this embodiment was collected from a multicenter prospective cohort study (NCT03297164) and the AMI supplemental cohort of the Department of Cardiology, the coordinating medical center of the study, Harbin Medical University Second Affiliated Hospital. The patient enrollment period was from January 2017 to July 2022, and the data covered the full cycle information of patients from admission to one year of follow-up after discharge.

[0040] This embodiment employs six machine learning algorithms, including logistic regression, XGBoost, Lightweight Gradient Boosting Machine (LightGBM), Random Forest, Gradient Boosting, and Decision Tree. Patients from the Second Affiliated Hospital of Harbin Medical University were used as the exploratory cohort (n=11752), randomly divided into a training set and an internal test set at an 8:2 ratio; patients from other medical centers served as the external validation cohort (n=2593). Grid search and five-fold cross-validation were used to select hyperparameters. Five-fold cross-validation was used to construct prediction models for different machine learning algorithms on the training set. After performance evaluation metrics analysis and performance evaluation on the internal test set, Lightweight Gradient Boosting Machine was selected as the basis for the initial prediction module. In terms of performance evaluation metrics, AUC and F1 score were calculated on the test set (20% of the data). The AUC results for the six machine learning algorithms are as follows: Random Forest 0.85 (95% CI, 0.81-0.90), XGBoost 0.82 (95% CI, 0.77-0.87), Logistic Regression 0.86 (95% CI, 0.81-0.90), Gradient Boosting 0.86 (95% CI, 0.81-0.90), LightGBM 0.85 (95% CI, 0.80-0.90), and Decision Tree 0.82 (95% CI, 0.77-0.87). Based on the distribution of the calibration curves, LightGBM and Gradient Boosting showed better fit, and LightGBM's Brier score of 0.089 was superior to Gradient Boosting's 0.100. In a sensitivity analysis of 9230 AMI patients with no missing data, LightGBM maintained superior performance with an AUC of 0.83 (95% CI, 0.77–0.89). Therefore, the LightGBM algorithm was selected to construct the initial prediction module for external validation. The external validation AUC of the initial prediction module constructed using the LightGBM algorithm was 0.80 (95% CI, 0.79–0.91). Based on these results, LightGBM was chosen as the optimal algorithm for constructing the initial prediction module.

[0041] Specifically, the initial prediction module uses a lightweight gradient booster to obtain risk predictions for cardiac death.

[0042] In one specific application of this embodiment, a 30-year-old patient with a body mass index of 22.0, an onset temperature of 36.6℃, PM2.5 concentration of 10.0, Gensini score of 60, left ventricular end-diastolic diameter of 50mm, ejection fraction of 60%, slightly elevated cTnI, severely elevated BNP, and normal creatinine, D-dimer, and fasting blood glucose levels was analyzed for risk prediction based on the input data. The output probability of cardiac death was 0.1714, classified as intermediate risk, suggesting the need to change unhealthy lifestyle habits, improve medication adherence, and undergo periodic follow-up examinations.

[0043] Example 2 like Figure 2 As shown, this embodiment discloses a working method for a predictive model of cardiac death in patients with acute myocardial infarction, applied to the predictive model of cardiac death in patients with acute myocardial infarction as described above. The working method includes the following steps: Step 1: Acquire and fuse medical data from multiple sources for patients with acute myocardial infarction to obtain multi-source fused data; wherein the medical multi-source data includes at least clinical basic data, laboratory examination data, treatment data, environmental data, and follow-up data; Step 2: Based on multi-source fusion data, obtain the corresponding risk prediction value for cardiac death; Step 3: Continuously collect the first change information of the multi-source fusion data and the second change information of the risk prediction value; the first change information includes the difference, change rate and trend characteristics calculated for at least one type of medical multi-source data and marked with significant change information; the second change information includes the numerical difference, fluctuation frequency and change direction of the risk prediction value, and the second change information is associated with the corresponding first change information. Step 4: Based on the first and second change information, adjust the prediction logic of Step 2 to obtain the risk prediction value of the next cardiac death.

[0044] It should be noted that the working method of the acute myocardial infarction patient cardiac death prediction model in this embodiment corresponds to the aforementioned acute myocardial infarction patient cardiac death prediction model. Therefore, any content not specifically described in the working method of the acute myocardial infarction patient cardiac death prediction model in this embodiment, including but not limited to functional definitions, working principles, and technical effects, can be referred to the description in the aforementioned acute myocardial infarction patient cardiac death prediction model, and will not be repeated here.

[0045] In summary, the cardiac mortality prediction model and its working method for acute myocardial infarction patients in this embodiment integrates multi-source data, including clinical baseline data, laboratory tests, treatment data, environmental data, and follow-up data, overcoming the limitations of traditional single data sources and improving the comprehensiveness of predictions. The dynamic monitoring module simultaneously collects changes in multi-source fused data and risk prediction values, and combined with the multi-dimensional adjustments of the prediction correction module, it solves the problem that traditional static models cannot adapt to rapid fluctuations in the patient's condition. Through feature weight enhancement, sensitivity coefficient adjustment, and dynamic correction of mapping relationships, especially the fine optimization based on the probability of risk changes, the degree of matching of changes, and the degree of correlation of changes, the prediction sensitivity and accuracy are significantly improved. The initial prediction uses a lightweight gradient boosting machine, which, while balancing prediction efficiency and accuracy, better meets the needs of real-time clinical decision-making and provides a reliable basis for early intervention in cardiac mortality after acute myocardial infarction.

[0046] In the embodiments provided in this application, it should be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any suitable combination thereof. For hardware implementation, the processor may be implemented in one or more of the following: application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to implement the functions described herein, or combinations thereof. For software implementation, some or all of the processes of the embodiments may be performed by a computer program instructing the associated hardware. During implementation, the program may be stored in a computer-readable storage medium or transmitted as one or more instructions or code on a computer-readable storage medium. Computer-readable storage media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media may be any available medium accessible to a computer. Computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible to a computer.

[0047] Finally, it should be noted that the above description is only a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A predictive model for cardiac death in patients with acute myocardial infarction, characterized in that, The model includes: The data fusion module is used to acquire and fuse medical multi-source data of patients with acute myocardial infarction to obtain multi-source fused data; wherein the medical multi-source data includes at least clinical basic data, laboratory examination data, treatment data, environmental data, and follow-up data; The initial prediction module is used to obtain the corresponding risk prediction value of cardiac death based on the multi-source fusion data; The dynamic monitoring module is used to continuously collect the first change information of the multi-source fusion data and the second change information of the risk prediction value. The first change information includes the difference, change rate and trend characteristics calculated for at least one type of medical multi-source data and the significant change information is marked. The second change information includes the numerical difference, fluctuation frequency and change direction of the risk prediction value, and the second change information is associated with the corresponding first change information. The prediction correction module is used to adjust the prediction logic of the initial prediction module based on the first change information and the second change information to obtain the risk prediction value of the next cardiac death.

2. The predictive model for cardiac death in patients with acute myocardial infarction according to claim 1, characterized in that, The first change information is obtained through the following methods: For at least one type of medical multi-source data, the difference, rate of change, and trend characteristics of the multi-source fusion data at the current moment and the multi-source fusion data at the previous acquisition moment are calculated in real time. The trend features include upward trends, downward trends, or fluctuation amplitudes within multiple collection periods, and the difference greater than a preset first threshold or the rate of change greater than a preset second threshold is marked as significant change information to obtain the first change information.

3. The predictive model for cardiac death in patients with acute myocardial infarction according to claim 2, characterized in that, The second change information is obtained in the following way: Record the numerical differences, fluctuation frequency, and direction of change of multiple consecutive risk prediction values; wherein, the numerical difference is the ratio of the absolute difference between two adjacent risk prediction values ​​to the previous risk prediction value, the fluctuation frequency is the number of times the risk prediction value crosses a preset risk level threshold within a unit of time, and the direction of change includes risk increase, decrease, or stability. When the numerical difference is greater than a preset third threshold or the number of times the consecutive risk prediction value crosses the risk level threshold is greater than a preset fourth threshold, it is determined that the second change information exists, and the corresponding first change information is associated with it.

4. The predictive model for cardiac death in patients with acute myocardial infarction according to claim 3, characterized in that, The adjustment of the prediction logic of the initial prediction module includes: Based on the medical multi-source data type corresponding to the significant change information marked in the first change information, the feature weight of this type of data in the initial prediction module is increased. Based on the numerical differences and fluctuation frequency of the second change information, the sensitivity coefficient of the initial prediction module is adjusted; By using the correlation between the first change information and the second change information, the mapping relationship between the corresponding data type and the risk prediction value in the initial prediction module is corrected.

5. The predictive model for cardiac death in patients with acute myocardial infarction according to claim 4, characterized in that, The correlation between the first change information and the second change information includes the probability of risk change, the degree of change matching, and the degree of change correlation; wherein: The probability of risk change is the probability of the change in the risk prediction value in the second change information corresponding to the medical multi-source data type marked in the first change information. The change matching degree is the degree of deviation between the time interval between the moment when the multi-source fused data undergoes a significant change and the time interval between the corresponding change in the risk prediction value and a preset time window. The correlation degree of change is the quantitative correlation between the magnitude of change of the first change information and the numerical difference of the second change information.

6. The predictive model for cardiac death in patients with acute myocardial infarction according to claim 5, characterized in that, The step of correcting the mapping relationship between the corresponding data type and the risk prediction value in the initial prediction module includes: When the probability of risk change meets the preset probability condition, the mapping weight of the medical multi-source data type with the corresponding risk prediction value change direction in the initial prediction module is increased, and the increase in weight is positively correlated with the probability of risk change. When the change matching degree satisfies that the time interval is within the time window or its deviation degree satisfies the preset deviation condition, the dynamic mapping relationship between the multi-source fusion data and the risk prediction value corresponding to the time series is strengthened. The mapping function parameters are adjusted based on the preset correlation interval in which the change correlation degree is located. The correlation interval includes a high correlation interval, a medium correlation interval, and a low correlation interval.

7. The predictive model for cardiac death in patients with acute myocardial infarction according to claim 6, characterized in that, The adjustment of the mapping function parameters based on the preset correlation interval where the change correlation degree is located includes: When the correlation of the change is in the high correlation range, the impact of the data change on the risk prediction value is amplified. When the degree of correlation of the change is in the medium correlation range, the original influence strength is maintained; When the correlation of the change is in the low correlation range, a preset attenuation coefficient is introduced to reduce the interference of this data type on the risk prediction value.

8. The predictive model for cardiac death in patients with acute myocardial infarction according to claim 1, characterized in that, The basic clinical data should include at least the age, sex, BMI, smoking history, and history of heart disease of patients with acute myocardial infarction. The test data include at least coronary angiography, myocardial enzyme profile, electrocardiogram, echocardiography, and hemodynamic parameters; The treatment data includes at least the timing of reperfusion therapy, medication records, and information on coronary interventional therapy. The environmental data includes at least the temperature at the time of onset, PM2.5 concentration, and diurnal rhythm; The follow-up data includes at least records of cardiac death events and malignant arrhythmia episodes.

9. The predictive model for cardiac death in patients with acute myocardial infarction according to claim 1, characterized in that, The initial prediction module obtains the risk prediction value of cardiac death based on a lightweight gradient booster.

10. A method for operating a predictive model of cardiac death in patients with acute myocardial infarction, applied to the predictive model of cardiac death in patients with acute myocardial infarction as described in any one of claims 1-9, characterized in that, This working method includes the following steps: Step 1: Acquire and fuse medical data from multiple sources for patients with acute myocardial infarction to obtain multi-source fused data; Step 2: Based on the multi-source fusion data, obtain the corresponding risk prediction value for cardiac death; Step 3: Continuously collect the first change information of the multi-source fusion data and the second change information of the risk prediction value; the first change information includes the difference, change rate and trend characteristics calculated for at least one type of medical multi-source data and marked with significant change information; the second change information includes the numerical difference, fluctuation frequency and change direction of the risk prediction value, and the second change information is associated with the corresponding first change information. Step 4: Based on the first and second change information, adjust the prediction logic of Step 2 to obtain the risk prediction value of the next cardiac death.

Images