Construction method of vaccine dynamic evaluation early warning model based on multi-source data

Abstract

The invention relates to the technical field of vaccine dynamic evaluation and early warning, in particular to a method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data. The method comprises the following steps: acquiring multi-source clinical medical data of a vaccinator through a medical management system, and carrying out time sequence sorting processing on the data to form a sorted data set; and then, in combination with vaccine adverse reaction prompt data, analyzing the delay influence of the vaccination abnormal reaction, evaluating the abnormal reaction delay difference of individuals, and identifying susceptible population. And based on a support vector machine algorithm, constructing a vaccine risk dynamic early warning model for identifying vaccination reaction risk factors, and feeding back a result to the terminal. Finally, vaccine risk dynamic early warning firmware is designed and embedded into a medical management system, and dynamic evaluation and early warning after vaccination are achieved. According to the method, the vaccine dynamic evaluation and early warning technology is optimized, so that the vaccine dynamic evaluation and early warning technology is more perfect.

Description

Technical Field

[0001] The present invention relates to the technical field of vaccine dynamic evaluation and early warning, and in particular to a method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data. Background Art

[0002] Although vaccines play a vital role in preventing diseases, different individuals will experience different degrees of adverse reactions during the vaccination process. In addition, the occurrence of adverse reactions usually has a certain delay, and some abnormal reactions may not appear until weeks or months after vaccination, which makes the traditional static assessment method have great limitations in timely discovering potential risks. With the development of big data and artificial intelligence technology, dynamic risk assessment methods based on multi-source data have become a hot topic of research. By integrating multiple data sources such as data from different medical systems, health monitoring equipment, laboratory tests, and self-reported symptom data from patients, different reactions after vaccination can be captured more comprehensively. The fusion of multi-source data can not only provide richer personal health information, but also provide stronger support for evaluating vaccine safety and risk management. However, the traditional method of constructing a dynamic vaccine assessment and early warning model based on multi-source data has the problems of inaccurate vaccine safety early warning and low automation. Summary of the invention

[0003] Based on this, it is necessary to provide a method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above purpose, a method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data includes the following steps: Step S1: obtaining a multi-source clinical medical data set of vaccine recipients through a medical management system; performing time-series sorting on the multi-source clinical medical cleansing data set to obtain a multi-source clinical medical sorting data set; Step S2: Acquire adverse reaction prompt data of vaccination; perform an impact analysis of delayed abnormal reaction of vaccination on a multi-source clinical medical ranking data set based on the adverse reaction prompt data of vaccination to obtain delayed abnormal reaction impact data of vaccination; perform individual abnormal reaction delay difference evaluation based on the delayed abnormal reaction impact data of vaccination to obtain individual abnormal reaction delay difference data; Step S3: classify the susceptible population based on the individual abnormal reaction delay difference data to obtain the individual physical sign susceptible population data; construct a vaccine risk dynamic early warning model based on the vaccination abnormal reaction delay impact data based on the support vector machine algorithm to obtain the vaccine risk dynamic early warning model; identify the vaccination reaction risk factors based on the individual physical sign susceptible population data according to the vaccine risk dynamic early warning model to obtain the vaccination risk factor data, and feed the vaccination risk factor data back to the terminal; Step S4: Automated firmware design for the vaccine risk dynamic warning model to obtain vaccine risk dynamic warning firmware; embed the vaccine risk dynamic warning firmware into the medical management system to execute vaccine dynamic assessment and warning.

[0005] The present invention obtains a multi-source clinical medical data set of vaccine recipients through a medical management system, which includes patient health information from different medical institutions and at different time points. At this stage, these multi-source data are first cleaned to remove noise and redundant information, and then sorted in chronological order to ensure the timeliness and continuity of the data, thereby providing an orderly and complete basic data set for subsequent analysis. The beneficial effect of this step is to ensure the accuracy, integrity and consistency of the data, and provide reliable data support for the subsequent analysis of abnormal reactions to vaccination. After obtaining the adverse reaction prompt data of vaccination, it is compared and analyzed with the multi-source clinical medical sorting data set, and the impact analysis of delayed abnormal reactions to vaccination is performed to identify delayed abnormal reactions. Further, through individual difference evaluation, the individual differences of each vaccine recipient's reaction after vaccination are analyzed. The beneficial effect of this step is that by accurately analyzing the time differences of adverse reactions after vaccination, the reaction characteristics of different individuals to vaccines can be revealed, providing an in-depth basis for the safety evaluation of vaccination. Based on the individual abnormal reaction delay difference data, the susceptible population is classified. Through cluster analysis, feature extraction and other technologies, the susceptible population with high risk is identified. Subsequently, a dynamic early warning model for vaccine risk is constructed using a support vector machine algorithm, and risk factors after vaccination are monitored in real time by analyzing individual characteristics. The beneficial effect of this step is to achieve dynamic early warning of potential risks by accurately identifying susceptible populations and vaccine risk factors, prevent adverse reactions in advance, and make timely interventions based on the early warning results. After the dynamic early warning model for vaccine risk is constructed, an automated firmware supporting the model is designed and implemented to ensure that the system can run in real time in the medical management system and perform dynamic risk assessment after vaccination. The early warning firmware is embedded in the medical management system, which can automatically perform risk assessment after the vaccine recipient receives the vaccine and feedback risk warning information to the doctor or terminal. The beneficial effect of this step is that through the combination of hardware and software, the risk assessment after vaccination is more automated and accurate, thereby improving the medical system's ability to respond to adverse reactions to vaccines and enhancing the safety of vaccination. Therefore, the present invention is an optimization process for a traditional method for constructing a dynamic evaluation early warning model for vaccines based on multi-source data, which solves the problems of inaccurate early warning of vaccine safety and low automation in a traditional method for constructing a dynamic evaluation early warning model for vaccines based on multi-source data, improves the accuracy of early warning of vaccine safety, and improves the degree of automation.

[0006] Preferably, step S1 comprises the following steps: Step S11: Acquire a multi-source clinical medical data set of vaccine recipients through a medical management system; Step S12: performing data standardization processing on the multi-source clinical medical data set of vaccine recipients to obtain a multi-source clinical medical optimized data set; Step S13: Perform time-series sorting processing on the multi-source clinical medical optimization data set to obtain a multi-source clinical medical sorting data set; wherein the multi-source clinical medical sorting data set includes: vaccine recipient vital sign data, vaccination cycle data and diagnosis report.

[0007] At this stage, the present invention obtains a multi-source clinical medical data set of vaccine recipients through the medical management system. The data set comes from different medical institutions, laboratory tests, diagnostic reports, vaccination records, etc., covering multi-dimensional data such as basic health information, vaccination history, and disease diagnosis records of vaccine recipients. The beneficial effect of this step is to collect comprehensive and diverse patient information, ensure the diversity and comprehensiveness of the data, and provide multi-level basic data support for subsequent risk assessment and analysis. Standardization of multi-source clinical medical data sets aims to unify the format and scale of data from different sources and eliminate data heterogeneity. Standardization includes converting data provided by different medical institutions into the same unit, format, and dimension to ensure that all types of data are consistent in subsequent analysis. The beneficial effect of this step is to eliminate the deviation or inconsistency between the data, so that the data has better comparability and operability, and provides standardized input data for subsequent precise analysis and modeling, thereby improving the accuracy of the analysis results. After completing data standardization, the optimized multi-source clinical medical data set is sorted in time series. The core of this step is to sort the information such as the vaccine recipient's physical sign data, vaccination cycle data, and diagnostic reports according to the time sequence, so as to ensure the timeliness and consistency of the data. The beneficial effect of this step is that through time series sorting, it can accurately reflect the health status of the vaccine recipients at different time points and the changes that occur after vaccination, making subsequent dynamic analysis and predictions more timely and scientific, and helping to identify potential vaccine reaction delays and adverse reaction trends.

[0008] Preferably, step S12 comprises the following steps: Step S121: performing data cleaning on the multi-source clinical medical data set of vaccine recipients to obtain a multi-source clinical medical cleansing data set, wherein the data cleaning includes: filling in missing values, removing duplicate values, and checking the consistency of data types; Step S122: performing date alignment processing on the multi-source clinical medical cleansing data set to obtain multi-source clinical medical date aligned data; Step S123: performing encoding processing on the multi-source clinical medical date alignment data to obtain multi-source clinical medical coding alignment data; Step S124: Perform data standardization processing on the multi-source clinical medical coding alignment data to obtain a multi-source clinical medical optimization data set.

[0009] Data cleaning in the present invention is the core step of data preprocessing, which aims to eliminate noise and inconsistency in the data. In this process, missing value filling, duplicate value removal and data type consistency check are performed on the multi-source clinical medical data set of vaccine recipients. Missing value filling ensures the integrity of the data and avoids analysis bias due to missing data; duplicate value removal eliminates redundant data and reduces model interference; data type consistency check ensures that the formats of different data fields are consistent, which is helpful for subsequent data processing and analysis. The beneficial effect of this step is to ensure the high quality, availability and reliability of the data set, laying a solid foundation for further processing. Date alignment processing is to ensure the consistency and synchronization of data from different sources in the time dimension. At this stage, data from different medical institutions or equipment are aligned according to a unified time axis to ensure that all data records have the same time scale. This alignment process enables data such as various signs of vaccine recipients, vaccination cycles and diagnostic reports to be accurately matched in time. The beneficial effect of this step is to make subsequent analysis more accurate by solving the problem of inconsistency in time, reflect the timeliness changes in the vaccination process, and enhance the timeliness and comparability of data. The encoding process converts data from different sources into a unified format so that it can be further analyzed in a machine-readable manner. In this process, all categorical, textual or identifier data (such as diagnosis type, vaccine type, etc.) will be converted into numerical codes to ensure data consistency and processing efficiency. The beneficial effect of this step is that by uniformly encoding the data, the integration of multi-source data becomes more efficient and convenient, reducing the processing difficulty caused by different data types, and facilitating model construction and feature extraction in subsequent analysis. Data standardization is to make data with different features comparable on the same scale to avoid unfair impact of certain features on the analysis results due to different dimensions. At this stage, by normalizing or standardizing the numerical data, all data are processed at the same scale, thereby improving the training effect and analysis accuracy of the model. The beneficial effect of this step is that by eliminating the dimensional differences between different features, the influence of various features is more balanced, reducing the deviation in the data processing process, and improving the stability and effectiveness of subsequent analysis and models.

[0010] Preferably, step S2 comprises the following steps: Step S21: Obtaining adverse reaction prompt data of vaccination; Step S22: coupling the diagnosis report with abnormal reactions according to the adverse reaction prompt data of vaccination to obtain the vaccination visit abnormal reaction coupling data; Step S23: analyzing the delayed impact of abnormal reactions to vaccination on the coupled data of abnormal reactions to vaccination visits based on the vaccine recipient's physical sign data and the vaccination cycle data, to obtain delayed impact data of abnormal reactions to vaccination; Step S24: Based on the vaccination abnormal reaction delay impact data, the individual abnormal reaction delay difference evaluation is performed on the vaccinee's physical sign data and the vaccination cycle data to obtain individual abnormal reaction delay difference data.

[0011] This step of the present invention provides key early warning information for further analysis by collecting adverse reaction prompt data related to vaccination. The acquired adverse reaction prompt data comes from multiple channels such as clinical records, patient feedback, and medical monitoring systems. The data usually contains information such as common side effects after vaccination and their severity and occurrence time. The beneficial effect of this step is to provide early prompts of potential risks after vaccination for subsequent analysis, which can help identify risk groups and provide data support for further individualized risk assessment and preventive measures. At this stage, the adverse reaction prompt data obtained in step S21 is used to couple abnormal reactions to the diagnostic report of the vaccinee. This means that the occurrence of adverse reactions is associated with the specific diagnostic data of the vaccinee (such as symptoms, signs, laboratory test results, etc.) to accurately identify abnormal reactions related to vaccination. Through this coupling, the specific manifestations of adverse reactions and their relationship with vaccination can be analyzed in multiple dimensions. The beneficial effect of this step is that it can systematically identify abnormal reactions related to vaccination, improve the detection accuracy of abnormal reactions, and provide more specific and detailed clinical data for subsequent analysis. This step analyzes the impact of abnormal reaction delays after vaccination by combining the physical sign data of the vaccinee and the vaccination cycle data. The physical sign data and vaccination cycle data of the vaccine recipient can help analyze whether there is a delayed effect of the adverse reaction, that is, whether the reaction is manifested within a period of time after vaccination. By analyzing these data, it is possible to identify which abnormal reactions are delayed and evaluate the possibility and impact of the delayed occurrence. The beneficial effect of this step is that it can capture those adverse reactions that failed to appear in the early stage and provide a scientific basis for early intervention and management, thereby improving the safety of vaccination. At this stage, according to the delayed impact data of abnormal reactions to vaccination obtained in step S23, individualized differences are evaluated for the physical sign data and vaccination cycle data of different vaccine recipients. By analyzing the individual characteristics of the vaccine recipients, such as age, gender, previous medical history, etc., combined with the vaccination cycle, the differences in the responses of different individuals to the vaccine, especially the differences in the delay time of abnormal reactions, are evaluated. This helps to deeply understand the reaction patterns of different individuals after vaccination, identify high-risk individuals, and provide support for personalized medical intervention. The beneficial effect of this step is that it can provide each vaccine recipient with a personalized vaccination safety assessment, which helps to optimize the vaccination strategy and improve the level of health monitoring after vaccination.

[0012] Preferably, step S23 includes the following steps: Step S231: extracting the vaccine recipient's weight, height and age parameters from the vaccine recipient's physical sign data to obtain the vaccine recipient's weight / height / age parameters; Step S232: performing a multivariate linear regression analysis on the coupled data of abnormal reactions to vaccination visits according to the vaccine recipient's weight / height / age parameters and vaccination cycle data, to obtain linear regression data of physical signs-abnormal reactions to vaccination cycles; Step S233: performing independent variable long-acting reaction analysis on the linear regression data of physical sign-vaccination cycle abnormal reaction to obtain independent variable abnormal long-acting reaction data; Step S234: performing autocorrelation abnormal trend coupling on the linear regression data of the abnormal reaction of the sign-vaccination period according to the abnormal long-acting reaction data of the independent variable to obtain autocorrelation abnormal trend coupling data; Step S235: Based on the autocorrelation abnormal trend coupling data and the independent variable abnormal long-term reaction data, the delayed impact of abnormal reaction to vaccination is analyzed on the abnormal reaction coupling data of vaccination and medical treatment to obtain the delayed impact data of abnormal reaction to vaccination.

[0013] The present invention first extracts important physiological parameters such as weight, height and age from the physical sign data of the vaccinee. These parameters are the basis for analyzing the individual's response to the vaccine, because weight, height and age are key factors affecting the response of drugs and vaccines, especially related to the responsiveness and metabolic rate of the immune system. By extracting these physiological parameters, the necessary basic data can be provided for subsequent multivariate analysis. The beneficial effect of this step is that the core physiological indicators of the individual can be accurately extracted, providing an important basis for further analyzing the reaction differences after vaccination, thereby ensuring the comprehensiveness of the data and the reliability of the analysis. Using the parameters such as the weight, height, age of the vaccinee, and the vaccination cycle data, a mathematical model is established through multivariate linear regression analysis. This model can reveal the relationship between the physiological characteristics of the vaccinee and the vaccination cycle, and further analyze the impact of these factors on abnormal reactions after vaccination. Through this regression analysis, it can be identified which physiological characteristics have a strong correlation with the occurrence of abnormal reactions. The beneficial effect of this step is that it can systematically evaluate how the physical signs of the vaccinee and the vaccination cycle work together to affect the occurrence of adverse reactions, and provide a scientific basis for risk prediction and intervention after vaccination. The long-term response of the independent variables (such as physiological parameters such as weight, height, and age) in the multivariate linear regression model and the vaccination cycle data was analyzed. Through the analysis of long-term response, it is possible to evaluate whether certain reactions after vaccination continue to affect the health of the vaccinee for a long time, especially some delayed adverse reactions. The results of the analysis help determine which physiological characteristics are more closely related to long-term responses. The beneficial effect of this step is that it can identify and quantify those delayed long-term reactions, making the later risk monitoring of vaccination more comprehensive and ensuring long-term observation and management of adverse reactions. Using the data of independent variable abnormal long-term reactions, autocorrelation analysis was performed on the results of the linear regression model of physical signs and abnormal reactions of the vaccination cycle to identify potential time trends and change patterns in the data. Through autocorrelation analysis, the trend of abnormal reactions over time can be found, whether there is a periodic or delayed effect, and further determine the persistence or regularity of abnormal reactions. The beneficial effect of this step is that it can reveal the time pattern of post-vaccination reactions, thereby helping medical personnel accurately predict when continuous monitoring or intervention of vaccinees is needed to ensure timely treatment of delayed health problems. Combine the previous autocorrelation abnormal trend coupling data with the independent variable abnormal long-term reaction data to perform delayed impact analysis on the abnormal reactions of the vaccine recipients. By combining the autocorrelation trend and long-term reaction data, analyze and evaluate whether there is a trend of delayed occurrence of adverse reactions and their long-term impact on the health of the vaccine recipients. The beneficial effect of this step is that it can accurately identify delayed abnormal reactions after vaccination and the extent of their impact, provide more personalized and refined intervention strategies for health management after vaccination, and ensure the long-term safety of vaccine recipients after vaccination.

[0014] Preferably, step S234 includes the following steps: Perform multi-scale decomposition of the trend evolution of the abnormal long-term response data of the independent variable to obtain the scale decomposition data of the abnormal trend evolution; According to the abnormal trend evolution scale decomposition data, the independent variable abnormal long-term reaction data is assigned independent variable weights to obtain the abnormal reaction trend independent variable weight data; Based on the abnormal reaction trend independent variable weight data, the autocorrelation abnormal trend coupling data of the abnormal reaction linear regression data of the sign-vaccination period was performed to obtain the autocorrelation abnormal trend coupling data.

[0015] The present invention extracts the trend characteristics of different levels hidden in the abnormal long-term reaction data of the independent variable through a multi-scale decomposition method. By decomposing the data in a multi-scale manner, the changing trend of the data can be understood from multiple levels from macro to micro, thereby effectively removing noise and analyzing its long-term change pattern. This method helps to reveal the evolution of various trends such as long-term, periodic or sudden changes in the data, and enhances the accuracy and predictability of the analysis. Based on the obtained abnormal trend evolution scale decomposition data, the independent variables are further weighted to effectively evaluate the relative importance of different independent variables in the abnormal reaction trend. This step helps to assign different weights to different independent variables by analyzing the contribution of each variable, making the subsequent analysis more accurate. Through this weight assignment method, it is possible to avoid some independent variables from affecting the overall analysis results due to ignoring their importance, and improve the reliability and accuracy of the prediction model. By combining the weight data of the independent variables with the linear regression results of the abnormal reaction of the sign-vaccination cycle, autocorrelation analysis is performed to capture the coupling relationship between abnormal trends. Autocorrelation analysis helps to discover periodic or long-term correlations in the data. This coupling can reveal the potential influence patterns and relationships between variables, thereby improving the model's comprehensive understanding and prediction capabilities of the impact of physical sign changes and vaccination cycles on abnormal reactions.

[0016] Preferably, step S3 comprises the following steps: Step S31: classifying susceptible populations based on the individual abnormal reaction delay difference data to obtain individual physical sign susceptible population data; Step S32: normalizing the delayed impact data of abnormal reactions to vaccination to obtain normalized delayed impact data of abnormal reactions; Step S33: constructing a vaccine risk dynamic early warning model based on the support vector machine algorithm for the normalized data of abnormal reaction delay impact, to obtain a vaccine risk dynamic early warning model; Step S34: Identify vaccination reaction risk factors based on the individual physical sign susceptible population data according to the vaccine risk dynamic early warning model to obtain vaccination risk factor data; the vaccination risk factor data includes: the characteristics and ingredients of the vaccine itself, and the individual differences and characteristics of the vaccination subjects.

[0017] The present invention analyzes the individual abnormal reaction delay differences and accurately classifies susceptible populations based on statistical and machine learning methods. By comparing the immune response characteristics of different populations, individuals who are more likely to have delayed reactions after vaccination are identified. This classification process helps to better understand the impact of different physiological characteristics (such as age, gender, health status, etc.) on vaccine reactions, ensuring that personalized vaccination plans are provided for different groups. The beneficial effect of this step is that it can accurately identify susceptible populations, improve the pertinence of safety monitoring after vaccination, and ensure that high-risk groups receive more detailed and timely attention. All collected data on the impact of delayed abnormal reactions to vaccination will be normalized. By uniformly standardizing the data, the influence of different dimensions and scales is eliminated, so that various types of data can be compared and analyzed at the same scale. This step can effectively unify the reaction differences of different individuals within a comparable range, avoid deviations caused by data differences, and improve the accuracy of subsequent analysis and modeling. Its beneficial effect is that it provides a unified and standard data set, which lays the foundation for subsequent modeling and risk prediction, making data analysis and decision-making more scientific and efficient. The support vector machine (SVM) algorithm was used to construct a dynamic early warning model for vaccine risk using the normalized delayed impact data of abnormal reactions. The support vector machine is an effective supervised learning algorithm that can identify potential patterns in data and make classification predictions. By establishing this model, the vaccination risk of the vaccine recipient can be dynamically predicted and warned, and the potential abnormal reaction risk can be identified in advance. The beneficial effect of this step is that it can achieve real-time monitoring and early warning of the reaction after vaccination, thereby helping medical institutions to take timely response measures, effectively reduce the risk of vaccination, and enhance the safety of vaccination. Based on the previously constructed dynamic early warning model for vaccine risk, the data of the susceptible population with individual physical signs were further analyzed to identify specific risk factors for vaccination reactions. These risk factors include the composition and characteristics of the vaccine itself, as well as individual differences in the recipients, such as age, gender, and allergy history. By fully identifying these risk factors, doctors and recipients can understand the risks in advance, so as to take corresponding preventive measures or choose appropriate vaccines. The beneficial effect of this analysis is that it can provide a detailed individualized risk assessment to ensure that a health management plan can be tailored according to the characteristics of the recipient during the vaccination process to minimize the occurrence of adverse reactions.

[0018] Preferably, step S33 includes the following steps: Step S331: dividing the abnormal reaction delay influence normalized data into a data set to obtain an abnormal reaction delay influence training set and an abnormal reaction delay influence test set; Step S332: performing parameter optimization training on the abnormal reaction delay impact training set based on the neural network model to obtain abnormal reaction delay optimization parameters; Step S333: constructing an initial vaccine risk dynamic early warning model based on the abnormal reaction delay optimization parameters based on the support vector machine algorithm to obtain a vaccine risk dynamic initial early warning model; Step S334: Perform a logic test on the vaccine risk dynamic initial warning model according to the abnormal reaction delay impact test set to obtain a vaccine risk dynamic warning optimization model.

[0019] The present invention divides the normalized abnormal reaction delay impact data to obtain a training set and a test set respectively. The training set is used for model training and optimization, while the test set is used for subsequent model verification and performance evaluation. The division of the data set can ensure that the model training will not be disturbed by the test data, improve the generalization ability of the model, and reduce the risk of overfitting. The divided training set and test set can ensure the effectiveness of the model under different data conditions, making subsequent modeling and verification more reliable, and laying a solid foundation for further early warning model construction. The neural network model is used to train the data in the training set, and the weights and biases of the network are continuously adjusted by parameter optimization methods (such as back propagation, gradient descent, etc.) to improve the model's prediction ability for abnormal reaction delay. Through this optimization training, the neural network can identify the complex nonlinear relationship in the data, and then obtain a set of optimal parameters, so that the model has a higher degree of fit to the data. The beneficial effect of this process is that it can enhance the model's prediction accuracy for delayed reaction after vaccination, improve the performance of the model, and ensure that the potential risks after vaccination can be better identified in practical applications. Using the support vector machine (SVM) algorithm, a preliminary vaccine risk dynamic early warning model is constructed based on the parameters optimized in step S332. Support vector machine is a powerful classification algorithm, especially suitable for high-dimensional data analysis. Through this algorithm, potential risks after vaccination can be effectively identified and dynamic warnings can be issued. By combining the optimization parameters with the SVM model, the model can quickly and accurately identify high-risk groups and potential abnormal reactions when facing complex vaccination risk data. The beneficial effect of this stage is that through the powerful classification ability of SVM, the model can handle complex risk prediction tasks, providing a strong foundation for subsequent risk warnings. The preliminary constructed vaccine risk dynamic warning model is logically tested using the test set to verify its performance on unknown data. The test set does not participate in model training and can provide the model with evaluation data in real scenarios. Through logical testing, the performance indicators of the model such as accuracy, sensitivity, and specificity can be detected, and the model can be adjusted and optimized as necessary according to the test results to ensure the stability and reliability of its predictive ability in practical applications. The beneficial effect of this step is that through continuous verification and optimization, a more perfect and robust warning model is finally obtained, which can accurately predict and identify risks after vaccination, thereby improving the safety of vaccination.

[0020] Preferably, step S332 includes the following steps: Performing weighted mapping of the training set on the abnormal reaction delay influence training set based on the neurons in the neural network model to obtain the influence weighted mapping training set; Perform network learning rate adaptation matching on the influence weighted mapping training set to obtain influence weighted learning rate adaptation data; The abnormal reaction delay impact training set is trained with parameter optimization according to the impact weighted learning rate adaptation data to obtain the abnormal reaction delay optimization parameters.

[0021] The neural network model of the present invention distinguishes the importance of different feature data by weighted mapping the data in the training set. The purpose of weighted mapping is to dynamically adjust the weight of each input feature in the training process according to the contribution of each input feature to the prediction result. This can help the neural network better focus on those data features that have a greater impact on abnormal reaction delay, thereby improving the learning efficiency and prediction accuracy of the model. Through this method, it is possible to avoid some unimportant features from having too much impact on the model training process, effectively improve the quality of the training set, and ensure that the model can more accurately capture the key patterns in the data. For the training set data after weighted mapping, an adaptive learning rate strategy is used to adjust the learning rate of the network. The traditional fixed learning rate causes the convergence speed to be too slow or too fast in the neural network training, while the adaptive learning rate dynamically adjusts the learning rate according to the feedback of the model to ensure that each round of parameter updates in the training process is more accurate. For each data sample, according to its weighted importance, a suitable learning rate is selected for adjustment to avoid training fluctuations caused by improper learning rate. The beneficial effect of this strategy is that it can improve the efficiency of model training, avoid overfitting or underfitting, and accelerate the network's learning process for complex data features. By combining the influence weighted learning rate adaptation data obtained in the previous steps, the training set is further trained for parameter optimization. By carefully adjusting the parameters in each layer of the neural network, it is ensured that the model can fit the data in the training set most effectively. The optimization process usually uses methods such as gradient descent so that the parameters in the network are adjusted in a more appropriate direction in each round of training, gradually reducing the prediction error. Through the dual mechanism of weighted mapping and adaptive learning rate, the model can optimize parameters more accurately and maximize its ability to predict abnormal reaction delays. The beneficial effect of this process is that it enables the model to achieve optimal performance on complex abnormal reaction data, improving prediction accuracy and generalization ability.

[0022] Preferably, the characteristics and components of the vaccine itself include: vaccine formulation, vaccine strain, adjuvants and media.

[0023] The beneficial effect of the present invention is that a multi-source clinical medical data set of vaccine recipients is obtained through a medical management system, which includes patient health information from different medical institutions and different time points. At this stage, these multi-source data are first cleaned to remove noise and redundant information, and then sorted in chronological order to ensure the timeliness and continuity of the data, thereby providing an orderly and complete basic data set for subsequent analysis. The beneficial effect of this step is to ensure the accuracy, integrity and consistency of the data, and provide reliable data support for the subsequent analysis of abnormal reactions to vaccination. After obtaining the adverse reaction prompt data of vaccination, it is compared and analyzed with the multi-source clinical medical sorting data set, and the impact analysis of delayed abnormal reactions to vaccination is performed to identify delayed abnormal reactions. Further, through individual difference evaluation, the individual differences of each vaccine recipient's reaction after vaccination are analyzed. The beneficial effect of this step is that by accurately analyzing the time differences of adverse reactions after vaccination, the reaction characteristics of different individuals to vaccines can be revealed, providing an in-depth basis for the safety assessment of vaccination. Based on the individual abnormal reaction delay difference data, the susceptible population is classified. Through cluster analysis, feature extraction and other technologies, the susceptible population with high risk is identified. Subsequently, a dynamic early warning model for vaccine risk is constructed using a support vector machine algorithm, and risk factors after vaccination are monitored in real time by analyzing individual characteristics. The beneficial effect of this step is to achieve dynamic early warning of potential risks by accurately identifying susceptible populations and vaccine risk factors, prevent adverse reactions in advance, and make timely interventions based on the early warning results. After the dynamic early warning model for vaccine risk is constructed, an automated firmware supporting the model is designed and implemented to ensure that the system can run in real time in the medical management system and perform dynamic risk assessment after vaccination. The early warning firmware is embedded in the medical management system, which can automatically perform risk assessment after the vaccine recipient receives the vaccine and feedback risk warning information to the doctor or terminal. The beneficial effect of this step is that through the combination of hardware and software, the risk assessment after vaccination is more automated and accurate, thereby improving the medical system's ability to respond to adverse reactions to vaccines and enhancing the safety of vaccination. Therefore, the present invention is an optimization process for a traditional method for constructing a dynamic evaluation early warning model for vaccines based on multi-source data, which solves the problems of inaccurate early warning of vaccine safety and low automation in a traditional method for constructing a dynamic evaluation early warning model for vaccines based on multi-source data, improves the accuracy of early warning of vaccine safety, and improves the degree of automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of the steps of a method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data; Figure 2 for Figure 1 Detailed implementation steps of step S2 in FIG.

[0025] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0026] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0027] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0028] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0029] To achieve this, please refer to Figure 1 to Figure 2 A method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data, the method comprising the following steps: Step S1: obtaining a multi-source clinical medical data set of vaccine recipients through a medical management system; performing time-series sorting on the multi-source clinical medical cleansing data set to obtain a multi-source clinical medical sorting data set; Step S2: Acquire adverse reaction prompt data of vaccination; perform an impact analysis of delayed abnormal reaction of vaccination on a multi-source clinical medical ranking data set based on the adverse reaction prompt data of vaccination to obtain delayed abnormal reaction impact data of vaccination; perform individual abnormal reaction delay difference evaluation based on the delayed abnormal reaction impact data of vaccination to obtain individual abnormal reaction delay difference data; Step S3: classify the susceptible population based on the individual abnormal reaction delay difference data to obtain the individual physical sign susceptible population data; construct a vaccine risk dynamic early warning model based on the vaccination abnormal reaction delay impact data based on the support vector machine algorithm to obtain the vaccine risk dynamic early warning model; identify the vaccination reaction risk factors based on the individual physical sign susceptible population data according to the vaccine risk dynamic early warning model to obtain the vaccination risk factor data, and feed the vaccination risk factor data back to the terminal; Step S4: Automated firmware design for the vaccine risk dynamic warning model to obtain vaccine risk dynamic warning firmware; embed the vaccine risk dynamic warning firmware into the medical management system to execute vaccine dynamic assessment and warning.

[0030] In the embodiment of the present invention, reference Figure 1 As shown, it is a schematic diagram of the steps of a method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data of the present invention. In this example, the method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data includes the following steps: Step S1: obtaining a multi-source clinical medical data set of vaccine recipients through a medical management system; performing time-series sorting on the multi-source clinical medical cleansing data set to obtain a multi-source clinical medical sorting data set; In an embodiment of the present invention, a multi-source clinical medical data set of vaccine recipients is obtained through a medical management system. The data set should cover the basic vital signs data of the vaccine recipients (such as weight, height, age, gender, etc.), vaccination record data (type of vaccine, dosage, vaccination date, etc.), diagnosis report (including detailed medical diagnosis description, examination report and test results), etc. The medical management system needs to have a comprehensive query function for vaccination records and related medical records to ensure that the extracted data covers all clinical medical information related to vaccination. The multi-source clinical medical data set is cleaned. The cleaning process includes filling missing values, analyzing the correlation between vaccination records and vital signs data, and reasonably filling missing items using mean or interpolation methods; removing duplicate values, matching different medical records with the unique identifier of a specific vaccine recipient, and deleting redundant entries; performing data type consistency check, and ensuring data standardization by defining field types (such as unified format of date fields, unified units of numerical fields, etc.). After cleaning, a multi-source clinical medical cleaned data set is formed. The cleaned data is sorted in time series. First, the vaccination records in the dataset are sorted according to the vaccination time to ensure the continuity of the timeline; second, each vaccination record is matched with the post-vaccination diagnosis report and arranged in the logical order before and after vaccination; finally, for the same vaccinee, all historical medical records are associated with the current vaccination records to form a complete time series data. The sorted dataset is defined as a multi-source clinical medical sorting dataset.

[0031] Step S2: Acquire adverse reaction prompt data of vaccination; perform an impact analysis of delayed abnormal reaction of vaccination on a multi-source clinical medical ranking data set based on the adverse reaction prompt data of vaccination to obtain delayed abnormal reaction impact data of vaccination; perform individual abnormal reaction delay difference evaluation based on the delayed abnormal reaction impact data of vaccination to obtain individual abnormal reaction delay difference data; In an embodiment of the present invention, adverse reaction prompt data of vaccination is obtained. The data is derived from the adverse reaction reporting module in the public health database or the medical management system, and common or rare adverse reaction prompt items related to specific vaccines are screened out by statistical methods (such as event frequency statistics or association rule mining). According to the adverse reaction prompt data of vaccination, the multi-source clinical medical sorting data set is analyzed for the impact of delayed abnormal reaction of vaccination. The analysis process adopts a delay correlation evaluation method combined with a time window division technology. In the specific operation, a period of time after vaccination is defined as an observation period (such as 1 day, 3 days, 7 days, etc.), and the delay distribution between the vaccination time point and the abnormal reaction occurrence time is calculated, and the Pearson correlation coefficient is used to quantify the association strength between the delay time and the abnormal reaction incidence rate. Based on the above delay analysis data, an individual abnormal reaction delay difference evaluation is carried out. Cluster analysis is performed on the abnormal reaction records of each vaccinee to identify the differences in delay patterns among different vaccinees. A hierarchical clustering algorithm is used to define the distance metric of the cluster as the Euclidean distance of the vaccination reaction delay time, and an appropriate threshold is set to distinguish the differences in delay patterns. Finally, the individual abnormal reaction delay difference data is output.

[0032] Step S3: classify the susceptible population based on the individual abnormal reaction delay difference data to obtain the individual physical sign susceptible population data; construct a vaccine risk dynamic early warning model based on the vaccination abnormal reaction delay impact data based on the support vector machine algorithm to obtain the vaccine risk dynamic early warning model; identify the vaccination reaction risk factors based on the individual physical sign susceptible population data according to the vaccine risk dynamic early warning model to obtain the vaccination risk factor data, and feed the vaccination risk factor data back to the terminal; In the embodiment of the present invention, the susceptible population is classified based on the individual abnormal reaction delay difference data. A classification method based on statistical characteristics is used to calculate the susceptibility probability of each type of population by interactive analysis of the physical sign data (such as age, gender, weight, etc.) of the vaccine recipients and the abnormal reaction delay data. The specific method is to analyze the distribution of abnormal reactions in different physical sign groups by chi-square test, and optimize the classification results by using the eigenvalue significance to generate the individual physical sign susceptible population data. Based on the support vector machine algorithm, a vaccine risk dynamic early warning model is constructed for the vaccination abnormal reaction delay impact data. First, the kernel function mapping technology is used to convert the vaccination abnormal reaction delay impact data into a high-dimensional space; secondly, the support vector boundary is calculated by the quadratic programming method to obtain the optimal decision surface between the classifications; finally, the parameters are optimized based on the training data to generate a vaccine risk dynamic early warning model. According to the vaccine risk dynamic early warning model, the vaccination reaction risk factors are identified for the individual physical sign susceptible population data. Using the classification results output by the model, combined with the specific physical signs of the susceptible population (such as age, genetic background, basic health status, etc.), the association rule mining technology is used to extract risk factors. Risk factors include the characteristics of the vaccine itself (formulation, excipients, etc.) and the specific physical signs of the vaccine recipient. The vaccination risk factor data is output and fed back to the terminal.

[0033] Step S4: Automated firmware design for the vaccine risk dynamic warning model to obtain vaccine risk dynamic warning firmware; embed the vaccine risk dynamic warning firmware into the medical management system to execute vaccine dynamic assessment and warning.

[0034] In an embodiment of the present invention, an automated firmware design is performed on the vaccine risk dynamic early warning model. The design process adopts an object-oriented programming method to encapsulate the risk assessment logic into modular decision-making units. Each unit includes an input interface (receiving real-time clinical data), a core processing module (risk assessment and prediction), and an output interface (generating assessment results and risk warnings). The integrity and accuracy of the module functions are verified by integrated testing, and finally the vaccine risk dynamic early warning firmware is obtained. The vaccine risk dynamic early warning firmware is embedded in the medical management system. First, the data transmission channel is designed in combination with the firmware's input and output interfaces and the database access protocol of the medical system to ensure that the firmware can call the vaccination data in real time; secondly, the output results of the firmware are connected to the alarm module of the system to realize real-time prompts and displays of abnormal risks; finally, the continuous testing method is used to verify the operating performance of the firmware under different data loads to ensure the stability and efficiency of the early warning function, thereby realizing the dynamic evaluation and early warning function of vaccines.

[0035] Preferably, step S1 comprises the following steps: Step S11: Acquire a multi-source clinical medical data set of vaccine recipients through a medical management system; Step S12: performing data standardization processing on the multi-source clinical medical data set of vaccine recipients to obtain a multi-source clinical medical optimized data set; Step S13: Perform time-series sorting processing on the multi-source clinical medical optimization data set to obtain a multi-source clinical medical sorting data set; wherein the multi-source clinical medical sorting data set includes: vaccine recipient vital sign data, vaccination cycle data and diagnosis report.

[0036] In an embodiment of the present invention, the specific implementation operations of obtaining a multi-source clinical medical data set of vaccine recipients through a medical management system include the following. First, retrieve the personal identification data of the vaccine recipient, such as the ID number or medical card number, to ensure that each vaccine recipient can be uniquely identified. Then, extract the clinical data related to the vaccine recipient from the multi-module database in the medical management system, including basic physical sign data (such as gender, age, weight, height, allergy history, etc.), vaccination records (vaccine type, batch number, vaccination date and dosage), and diagnostic reports (laboratory test results, radiological image descriptions, drug use records, etc.). During the extraction process, it is necessary to use structured query language (SQL) to query the database and construct a multi-table join query statement to achieve complete data extraction. In order to ensure the comprehensiveness of the data, the data field coverage should be checked to ensure that all clinical information related to vaccination is extracted into the intermediate storage table to form a multi-source clinical medical data set for vaccine recipients. The specific operations for data standardization of the multi-source clinical medical data set for vaccine recipients are as follows. First, the extracted multi-source data is checked for field type consistency, including time field formatting (unified to ISO 8601 standard time format, such as YYYY-MM-DD) and numerical field unit conversion (such as weight unified to kilograms and height unified to centimeters). Secondly, the outliers in the data are processed. For continuous data, the interquartile range method is used to screen outliers that exceed the normal range and mark them; for categorical data, statistical methods are used to detect abnormal categories and delete or reclassify them. Thirdly, for fields with missing values, a reasonable filling strategy is selected according to the data characteristics, such as using the mean filling method to complete the vital sign data, or improving the vaccination records based on the interpolation of the previous and next time series. Finally, the normalization method is used to transform the numerical fields in the vital sign data to eliminate the impact of dimensional differences on subsequent analysis. Through the above steps, a multi-source clinical medical optimization data set is generated, which is consistent, complete and standardized. The specific implementation process of time series sorting of multi-source clinical medical optimization data sets includes the following. First, extract the time-related fields in the data set (such as vaccination date, diagnosis date, etc.), sort and preprocess them, arrange all records of each vaccinee in chronological order, and generate a new timeline index. Secondly, combined with the characteristics of the vaccination cycle, take the vaccination record as the main line, match the diagnosis report and physical sign data to the vaccination event according to the vaccination time, and mark them as pre-vaccination and post-vaccination records. In specific operations, the Pandas library in the Python programming language can be used to achieve dynamic grouping and timestamp alignment of data through "groupby" and "merge" operations. For multiple vaccination records of the same vaccinee, the event is segmented based on the vaccination date to form an independent vaccination cycle data block.Finally, the sorted data is divided into three logical parts: vaccine recipients’ physical data (such as weight, height, allergy history, etc.), vaccination cycle data (such as vaccination type, vaccination time and dosage), and diagnostic reports (such as laboratory tests and clinical imaging descriptions). They are saved in separate data tables according to vaccination time and data type, and finally a multi-source clinical medical sorting data set is generated, providing a high-quality data foundation for subsequent vaccine evaluation and early warning.

[0037] Preferably, step S12 comprises the following steps: Step S121: performing data cleaning on the multi-source clinical medical data set of vaccine recipients to obtain a multi-source clinical medical cleansing data set, wherein the data cleaning includes: filling in missing values, removing duplicate values, and checking the consistency of data types; Step S122: performing date alignment processing on the multi-source clinical medical cleansing data set to obtain multi-source clinical medical date aligned data; Step S123: performing encoding processing on the multi-source clinical medical date alignment data to obtain multi-source clinical medical coding alignment data; Step S124: Perform data standardization processing on the multi-source clinical medical coding alignment data to obtain a multi-source clinical medical optimization data set.

[0038] In an embodiment of the present invention, the specific operations for data cleaning of a multi-source clinical medical data set of vaccine recipients include the following. First, the missing value filling operation needs to check whether there are missing values ​​in each data field, especially key fields such as the vaccine recipient's vital signs data and vaccination records. For the filling of missing values, statistical methods are used, such as filling with mean or median for numerical data (such as body temperature, blood pressure, etc.), and for categorical data (such as gender, vaccine type, etc.), filling by mode or using domain knowledge for reasonable inference. For missing values ​​that cannot be determined, a deletion strategy is used for cleaning. Secondly, the duplicate value removal operation identifies duplicate entries by comparing each record in the data set. For duplicate records, the most representative records are retained based on fields such as vaccination date and vaccination type, and other redundant data are deleted. Furthermore, the data type consistency check ensures that the numerical data does not contain text data by checking the data type of each field, and the timestamp field is formatted in a unified standard format (such as ISO 8601). The categorical data ensures that each category has a clear label and the value range meets the specification. Finally, through these operations, a multi-source clinical medical cleaned data set is generated to ensure that the data set has a certain quality and lay a solid foundation for subsequent processing. The implementation process of date alignment processing of multi-source clinical medical cleaned data sets includes the following steps. First, identify all time fields in the data set, mainly including vaccination date, diagnosis report date, and timestamps of various vital signs records. According to different time ranges (for example, the vaccination date is earlier than the diagnosis report date), all date fields need to be uniformly converted into a standard time format (such as year-month-day). Next, sort all relevant records in the order of the time fields to ensure that all events and records are arranged in chronological order. Then, the key operation of date alignment is to pair the data of the same vaccinee at different time points. With the vaccination date as the benchmark, all relevant records before and after (such as diagnosis reports, vital signs data, etc.) are aligned with the vaccination date to ensure that these data have an appropriate logical relationship in time and avoid time sequence disorder. For those diagnosis records with vaccination time too early or too late, the time difference threshold is used to filter or mark them. Through these processes, multi-source clinical medical date alignment data is obtained, providing a consistent time structure for subsequent analysis. The specific operations for encoding multi-source clinical medical date alignment data include the following. First, all categorical data fields are encoded. For example, the type of vaccination field can be encoded using integer value mapping, such as mapping "inactivated vaccine" to 1, "live attenuated vaccine" to 2, etc. Similarly, for some categorical variables in the physical sign data, such as gender, allergy history, etc., similar integer encoding methods are also used to convert categories into numbers to facilitate subsequent data analysis and calculations.Then, for date fields, such as vaccination date, diagnosis date, etc., date difference coding is used, that is, by calculating the time difference between each event, the date field is converted into the time interval (in days) after or before vaccination. In addition, all continuous variables, such as age and weight, need to be binned or discretized to further reduce the complexity of the data and ensure that they meet the requirements of subsequent algorithms. After the coding is completed, multi-source clinical medical coding alignment data is obtained, and all data have been converted into numerical or computable data for subsequent processing and analysis. The operation of data standardization of multi-source clinical medical coding alignment data includes the following steps. First, identify and select all numerical fields (such as weight, height, blood pressure, etc.), and then use the standardization method to process them. The standardization process uses the Z-score standardization method, that is, the mean and standard deviation of each numerical field are calculated to convert it into a distribution with zero mean and unit variance.

[0039] Preferably, step S2 comprises the following steps: Step S21: Obtaining adverse reaction prompt data of vaccination; Step S22: coupling the diagnosis report with abnormal reactions according to the adverse reaction prompt data of vaccination to obtain the vaccination visit abnormal reaction coupling data; Step S23: analyzing the delayed impact of abnormal reactions to vaccination on the coupled data of abnormal reactions to vaccination visits based on the vaccine recipient's physical sign data and the vaccination cycle data, to obtain delayed impact data of abnormal reactions to vaccination; Step S24: Based on the vaccination abnormal reaction delay impact data, the individual abnormal reaction delay difference evaluation is performed on the vaccinee's physical sign data and the vaccination cycle data to obtain individual abnormal reaction delay difference data.

[0040] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes: Step S21: Obtaining adverse reaction prompt data of vaccination; In an embodiment of the present invention, obtaining the adverse reaction prompt data of vaccination requires a strict data collection and integration process. First, the vaccine recipient feedback information in the vaccination record is screened, and the sources of adverse reaction data involved are collected, including but not limited to the medical consultation record after vaccination, the adverse event reporting system (such as AEFI report), and the monitoring data during the observation period after vaccination, as well as the precautions for the adverse reactions of the vaccine. For these data, it is necessary to clarify the time range and related fields, such as the type, time of occurrence, duration and severity of the adverse reaction, and extract the description content with standardized coding, such as the diagnostic code in ICD-10. Secondly, check and eliminate duplicate or incomplete adverse reaction data entries to ensure that only high-confidence records related to vaccination are retained. At the same time, the records are matched and integrated across tables using key fields (such as vaccine recipient ID, vaccination time and vaccine type) to avoid data loss and redundancy. Finally, an adverse reaction prompt data set containing clear classification and coding is output to provide a basis for subsequent abnormal reaction analysis.

[0041] Step S22: coupling the diagnosis report with abnormal reactions according to the adverse reaction prompt data of vaccination to obtain the vaccination visit abnormal reaction coupling data; In an embodiment of the present invention, coupling of abnormal reactions to the diagnostic report based on the adverse reaction prompt data of vaccination requires precise matching and analysis across data sets. First, extract the key fields in the diagnostic report, including the diagnosis time, diagnosis conclusion, physical sign record and its descriptive data (such as fever, headache and other symptoms). Then, based on the time information in the adverse reaction prompt data of vaccination, filter the data in the diagnostic report that has a temporal association with the vaccination time. Compare the vaccination time with the diagnosis time, and filter the qualified diagnostic records based on the set time window (for example, within 7 days). Then, based on the symptom similarity matching algorithm, semantically match the description in the diagnostic conclusion with the known reaction type in the adverse reaction prompt data. For example, the similarity between the diagnostic text and the reaction prompt text is calculated by the TF-IDF weighted method to determine the coupling relationship. Finally, the successfully matched data is merged to generate the abnormal reaction coupling data of vaccination visits containing time association and symptom similarity, which is convenient for further delay analysis.

[0042] Step S23: analyzing the delayed impact of abnormal reactions to vaccination on the coupled data of abnormal reactions to vaccination visits based on the vaccine recipient's physical sign data and the vaccination cycle data, to obtain delayed impact data of abnormal reactions to vaccination; In an embodiment of the present invention, the specific operation of analyzing the delayed impact of abnormal reactions to vaccination based on the vaccinee's vital sign data and the vaccination cycle data on the vaccination visit abnormal reaction coupling data is as follows. First, extract the vaccinee's vital sign data, including the records of changes in vital signs before and after vaccination, such as body temperature, heart rate, blood pressure, etc., to ensure that its timestamp can be aligned with the vaccination cycle data and the abnormal reaction coupling data. Then, for each abnormal reaction record, analyze the difference between the reaction occurrence time and the vaccination time, calculate the delay time parameter, and perform a correlation analysis with the fluctuation amplitude and frequency in the vital sign data. The Pearson correlation coefficient or the Spearman rank correlation coefficient is used to quantify the correlation between different variables and evaluate the degree of influence of reaction delay on the changes in vital signs. For reactions with strong delayed effects, further analyze the correlation between the reaction type (such as systemic reactions and local reactions) and the vaccination cycle to determine the characteristic pattern of delayed effects. Through these steps, the delayed impact data of abnormal reactions to vaccination are generated, and the timing and intensity distribution of the delayed effect are clarified.

[0043] Step S24: Based on the vaccination abnormal reaction delay impact data, the individual abnormal reaction delay difference evaluation is performed on the vaccinee's physical sign data and the vaccination cycle data to obtain individual abnormal reaction delay difference data.

[0044] In an embodiment of the present invention, the specific operation of evaluating the individual abnormal reaction delay difference of the vaccinee's physical sign data and the vaccination cycle data according to the delayed impact data of the abnormal reaction to vaccination includes the following. First, the delayed impact data and the corresponding physical sign fluctuation records of each vaccinee are extracted, and grouped according to individual difference characteristics, such as stratified analysis according to age, gender or previous medical history. Secondly, statistical methods, such as one-way analysis of variance (ANOVA) or multiple regression analysis, are used to quantify the differences in delayed reactions among different vaccinee groups. For individuals with significant changes in physical signs, the dynamic characteristics of their physical sign change curves are evaluated by time series analysis tools (such as ARIMA models or sliding window analysis methods) to clarify the individual differences in delayed reactions in the time dimension. Finally, combined with the information such as dosage and interval in the vaccination cycle data, the abnormal reaction characteristic patterns of specific individuals are identified, and the individual abnormal reaction delay difference data are integrated to generate data support for subsequent personalized evaluation and decision-making.

[0045] Preferably, step S23 includes the following steps: Step S231: extracting the vaccine recipient's weight, height and age parameters from the vaccine recipient's physical sign data to obtain the vaccine recipient's weight / height / age parameters; Step S232: performing a multivariate linear regression analysis on the coupled data of abnormal reactions to vaccination visits according to the vaccine recipient's weight / height / age parameters and vaccination cycle data, to obtain linear regression data of physical signs-abnormal reactions to vaccination cycles; Step S233: performing independent variable long-acting reaction analysis on the linear regression data of physical sign-vaccination cycle abnormal reaction to obtain independent variable abnormal long-acting reaction data; Step S234: performing autocorrelation abnormal trend coupling on the linear regression data of the abnormal reaction of the sign-vaccination period according to the abnormal long-acting reaction data of the independent variable to obtain autocorrelation abnormal trend coupling data; Step S235: Based on the autocorrelation abnormal trend coupling data and the independent variable abnormal long-term reaction data, the delayed impact of abnormal reaction to vaccination is analyzed on the abnormal reaction coupling data of vaccination and medical treatment to obtain the delayed impact data of abnormal reaction to vaccination.

[0046] In an embodiment of the present invention, the specific operation process for extracting the vaccinee's weight, height and age parameters from the vaccinee's physical sign data is as follows. First, the fields related to weight, height and age in the vaccinee's physical sign data set are screened, and their unit formats (for example, weight in kilograms, height in centimeters, and age in integer years) are clarified, and records with abnormal values ​​or missing values ​​are eliminated. Next, the complete data rows containing weight, height and age information are extracted into a separate temporary data table through conditional screening. In view of the problem of data integrity, individual missing values ​​can be filled by multiple interpolation methods to ensure the representativeness of the number of analysis samples. Then, a unified data format check is performed on each record, such as converting age data from date calculation to standard age expression. Finally, a structured data set containing the vaccinee's weight, height and age parameters is generated, and a unique vaccinee identifier is marked to match subsequent data. The specific process of performing multiple linear regression analysis on the coupled data of abnormal reactions to vaccination visits based on the vaccinee's weight, height and age parameters and the vaccination cycle data includes the following operations. First, a data input matrix was constructed, in which the independent variables included weight, height, age, and vaccination cycle data (such as vaccination time interval and dosage size), and the dependent variable was the quantitative description in the abnormal reaction record of vaccination visit (such as the degree of temperature increase, the number of abnormal heart rate, etc.). The input data was normalized to ensure that the dimensions of each variable were consistent to improve the model fitting accuracy. Secondly, the ordinary least squares (OLS) method was used to calculate the coefficients of multivariate linear regression, and the hypothesis test was used to analyze the significance of the impact of each independent variable on the dependent variable. The fitting effect of the regression model was evaluated by adjusting the coefficient of determination (R²), and the significant variables and their regression coefficients were recorded to generate the linear regression data of physical signs-abnormal reactions of vaccination cycle. The final output data contains the regression equation, variable weights, and statistical significance marks, which lays the foundation for the subsequent long-term reaction analysis. The specific operation process of the linear regression data of physical signs-abnormal reactions of vaccination cycle for independent variable long-term reaction analysis is as follows. First, the significant independent variables in the regression analysis were extracted, and the physical signs and vaccination cycle data with large variable weights were screened as candidate indicators. For these variables, time series analysis methods are used to evaluate their long-term impact on the dependent variables, such as calculating the duration and change pattern of the variable impact through segmented regression or sliding window analysis techniques. Secondly, draw the trend curve of each variable's impact on the dependent variable, and mark the time nodes of significant changes. Combined with statistical test methods (such as t-test or F-test), judge the continuity of the variable's impact and quantify its cumulative effect on abnormal reactions. Finally, generate independent variable abnormal long-acting reaction data, including the long-acting reaction curve, duration, and cumulative impact value of each variable, to provide a reference for the next step of trend coupling. The specific operations for autocorrelation abnormal trend coupling of the linear regression data of abnormal reactions of signs-vaccination cycle based on the independent variable abnormal long-acting reaction data are as follows.First, the trend curve in the independent variable long-term reaction data is feature extracted, such as analyzing the frequency characteristics of variable changes through Fourier transform, or capturing the abnormal points of local changes in the curve through wavelet transform. Secondly, the trend of the dependent variable in the linear regression data of the sign-vaccination cycle abnormal reaction is feature extracted by the same method to ensure that the characteristic vectors of the two types of data are comparable in the same space. The matching degree between the trend of the long-term variable and the trend of the dependent variable is evaluated by autocorrelation analysis technology (such as lag correlation analysis or cross-correlation function calculation), and the coupling relationship matrix is ​​generated based on the results to mark the main contribution points of the highly correlated variables to the abnormal trend. Finally, the autocorrelation abnormal trend coupling data, including the coupling coefficient matrix and the set of related variables, are output to provide input support for the comprehensive delay impact analysis. The specific steps for analyzing the delayed impact of vaccination abnormal reaction on the vaccination visit abnormal reaction coupling data based on the autocorrelation abnormal trend coupling data and the independent variable abnormal long-term reaction data are as follows. First, the highly correlated variables in the autocorrelation abnormal trend coupling data are screened as key input items, and the delay analysis data set is constructed in combination with the abnormal records in the vaccination visit abnormal reaction coupling data. Then, the distributed lag model (DLM) was used to calculate the cumulative impact curve of the vaccination cycle on abnormal reactions, and to quantify the functional relationship between the reaction delay time and the abnormal incidence rate. Furthermore, based on statistical regression analysis, the peak and recovery time of the abnormal reaction intensity within the delay time window were calculated to identify the key time points that had the greatest impact on the vaccine recipients. Finally, the delayed impact data of abnormal reactions to vaccination were generated, including the detailed time distribution of the delayed effect and the results of the individual characteristic association analysis, providing basic support for the subsequent personalized vaccine evaluation and early warning model optimization.

[0047] Preferably, step S234 includes the following steps: Perform multi-scale decomposition of the trend evolution of the abnormal long-term response data of the independent variable to obtain the scale decomposition data of the abnormal trend evolution; According to the abnormal trend evolution scale decomposition data, the independent variable abnormal long-term reaction data is assigned independent variable weights to obtain the abnormal reaction trend independent variable weight data; Based on the abnormal reaction trend independent variable weight data, the autocorrelation abnormal trend coupling data of the abnormal reaction linear regression data of the sign-vaccination period was performed to obtain the autocorrelation abnormal trend coupling data.

[0048] In an embodiment of the present invention, the specific implementation method of multi-scale decomposition of trend evolution of abnormal long-term reaction data of independent variables is as follows. First, the time series in the abnormal long-term reaction data of the independent variable is extracted into a standard form to ensure that each variable data has a consistent time interval and measurement unit. Secondly, the time series is decomposed by the empirical mode decomposition (EMD) method, and the intrinsic mode function (IMF) on different time scales is extracted, and the fluctuation and trend components in the original sequence are decomposed from high frequency to low frequency. For each IMF component obtained by decomposition, the significance of each component is calculated by the energy proportion method, and the high-frequency noise component and the low-frequency redundant component are eliminated. Further, the remaining components are feature aggregated, and the trend evolution curves of different time scales are obtained by weighted average calculation. Finally, the abnormal trend evolution scale decomposition data is generated, including the eigenvalues ​​of multi-scale decomposition, the trend change curves of each scale and their corresponding time windows, which provide basic data support for the subsequent weight assignment step. The implementation process of assigning independent variable weights to the abnormal long-term reaction data of the independent variable according to the abnormal trend evolution scale decomposition data is as follows. First, based on the trend curves of each scale in the abnormal trend evolution scale decomposition data, the contribution of each independent variable at different scales is calculated. Through the covariance matrix or correlation coefficient method, the explanatory power of each variable on the abnormal trend is quantified, and the corresponding contribution matrix is ​​generated. Secondly, the multi-scale contribution is normalized by the entropy weight method or the Delphi method to ensure that the sum of the variable weights at different scales is 1. On the basis of the normalized weight, combined with the cumulative effect index in the abnormal long-term reaction data of the independent variable, the comprehensive weight of each independent variable is further calculated to generate a variable weight distribution table. Finally, the abnormal reaction trend independent variable weight data is output, including the multi-scale contribution, normalized weight and comprehensive weight of each independent variable, to provide weighting parameters for the subsequent autocorrelated abnormal trend coupling. The implementation steps of coupling the autocorrelated abnormal trend of the linear regression data of the abnormal reaction of the sign-vaccination cycle based on the abnormal reaction trend independent variable weight data are as follows. First, the dependent variables and independent variables in the linear regression data of the abnormal reaction of the sign-vaccination cycle are extracted respectively, and the variables are matched with the abnormal reaction trend independent variable weight data to ensure that the weight information of each independent variable is consistent with the coefficient of its corresponding regression relationship. Secondly, the time series lagged correlation analysis method is used to perform autocorrelation analysis on the dependent variable of the regression data and the respective variables to evaluate the lagged influence characteristics of each variable on the dependent variable. Combined with the weights of the independent variables, the results of the lagged correlation analysis are weighted and summarized to form a comprehensive coupling matrix. Finally, the main influencing factors in the coupling matrix are extracted through matrix decomposition or principal component analysis methods to obtain autocorrelation abnormal trend coupling data, including coupling coefficients, the influence intensity of key variables, and the lagged time window, which provide input data for the delayed impact analysis of abnormal vaccination reactions.

[0049] Preferably, step S3 comprises the following steps: Step S31: classifying susceptible populations based on the individual abnormal reaction delay difference data to obtain individual physical sign susceptible population data; Step S32: normalizing the delayed impact data of abnormal reactions to vaccination to obtain normalized delayed impact data of abnormal reactions; Step S33: constructing a vaccine risk dynamic early warning model based on the support vector machine algorithm for the normalized data of abnormal reaction delay impact, to obtain a vaccine risk dynamic early warning model; Step S34: Identify vaccination reaction risk factors based on the individual physical sign susceptible population data according to the vaccine risk dynamic early warning model to obtain vaccination risk factor data; the vaccination risk factor data includes: the characteristics and ingredients of the vaccine itself, and the individual differences and characteristics of the vaccination subjects.

[0050] In an embodiment of the present invention, the implementation method of classifying susceptible populations based on individual abnormal reaction delay difference data is as follows. First, the core variables in the individual abnormal reaction delay difference data are extracted, including parameters such as physical sign characteristics (such as age, weight, height, etc.), previous medical history, genetic factors, and the type of vaccination. Then, the distance matrix between variables is constructed using the hierarchical clustering method with Euclidean distance as the metric. Based on the agglomerative clustering algorithm, starting from the group with higher similarity of individual physical signs, it is gradually aggregated to form a multi-level clustering tree. By determining the optimal number of clusters (which can be calculated based on the silhouette coefficient method), the specific categories of susceptible populations of individual physical signs are divided, such as high-risk group, medium-risk group and low-risk group. After the classification is completed, the data of susceptible populations of individual physical signs are generated, including the key feature description, distribution and correlation with delayed abnormal reactions of vaccination for each category of susceptible population. The implementation method of normalizing the data on delayed abnormal reactions to vaccination is as follows. First, for the variables in the delayed impact data of abnormal reactions to vaccination (including the time characteristics of different vaccination events, impact intensity indicators, etc.), the minimum and maximum values ​​of each variable were determined by analyzing their numerical range. The linear normalization method was used to map the original data to the [0,1] interval. During the normalization process, the data containing outliers were screened and eliminated to ensure the accuracy of normalization. After normalization, the data was uniformly dimensioned, which was convenient for subsequent model construction and multivariate analysis. Finally, the normalized data of delayed abnormal reaction effects were generated, including the normalized values ​​of each variable and its associated weight matrix with other variables. The implementation method of constructing a dynamic early warning model for vaccine risk based on the normalized data of delayed abnormal reaction effects based on the support vector machine (SVM) algorithm is as follows. First, the normalized data of delayed abnormal reaction effects were divided into training sets and test sets to ensure balanced data distribution. The radial basis function (RBF) kernel function in the SVM algorithm was selected as the classification kernel, and the kernel function was used to map the normalized data to a high-dimensional feature space to facilitate the identification of complex nonlinear relationships. During the model training process, the initial values ​​of the penalty parameters and kernel parameters are set, and the parameter combination is optimized by the grid search method to maximize the classification accuracy. After the training is completed, the prediction performance of the model is evaluated using the test set data, including indicators such as accuracy, recall, and F1 score. Finally, a dynamic early warning model for vaccine risk is generated, which contains decision boundaries and classification rules for real-time assessment of the risk of abnormal reactions to vaccination. The implementation method for identifying risk factors for vaccination reactions based on the data of susceptible populations with individual physical signs according to the dynamic early warning model for vaccine risk is as follows.First, each group characteristic in the data of susceptible populations with individual physical signs is input into the dynamic early warning model for vaccine risks, and the risk level of each group of people is judged based on the decision boundary of the model. Then, combined with the classification results of the model, the main influencing factors are extracted for the groups predicted to be high-risk and medium-risk, including vaccine components (such as adjuvants, antigen types, etc.) and individual differences of vaccination subjects (such as genetic susceptibility, immune status, etc.). The significance of each influencing factor is evaluated using the stepwise regression analysis method, and a priority ranking of risk factors is generated. Finally, the vaccination risk factor data is output, including the quantitative description of the risk factor, its manifestation in different susceptible populations, and its specific correlation with the risk of vaccination reaction, providing a scientific basis for safety prevention and control during vaccination.

[0051] Preferably, step S33 includes the following steps: Step S331: dividing the abnormal reaction delay influence normalized data into a data set to obtain an abnormal reaction delay influence training set and an abnormal reaction delay influence test set; Step S332: performing parameter optimization training on the abnormal reaction delay impact training set based on the neural network model to obtain abnormal reaction delay optimization parameters; Step S333: constructing an initial vaccine risk dynamic early warning model based on the abnormal reaction delay optimization parameters based on the support vector machine algorithm to obtain a vaccine risk dynamic initial early warning model; Step S334: Perform a logic test on the vaccine risk dynamic initial warning model according to the abnormal reaction delay impact test set to obtain a vaccine risk dynamic warning optimization model.

[0052] In an embodiment of the present invention, the implementation method of dividing the data set for the normalized data affected by abnormal reaction delay is as follows. First, the normalized data affected by abnormal reaction delay is comprehensively analyzed, including statistics of information such as the number of variables, distribution characteristics, and numerical intervals. A stratified sampling method is used to ensure that the divided data set maintains the distribution characteristics of the original data on key variables. The data set division ratio is set, and the data is divided into a training set and a test set, wherein the training set accounts for 80% of the total data and the test set accounts for 20%. In the specific division operation, a unique identifier is first assigned to each data record, and the division boundary of the training set and the test set is determined by a random number generation function. During the division process, duplicate records or outliers are excluded to ensure that the training set and the test set data are independent and do not overlap. After the division is completed, an abnormal reaction delay impact training set and an abnormal reaction delay impact test set are generated, and detailed statistical information of the division results is recorded for subsequent operation verification. The implementation method of performing parameter optimization training on the abnormal reaction delay impact training set based on the neural network model is as follows. A multi-layer feedforward neural network (Multi-Layer Perceptron, MLP) is selected as the basic structure, and a three-layer network architecture including an input layer, a hidden layer, and an output layer is designed. Among them, the number of neurons in the input layer is consistent with the number of variables in the abnormal reaction delay impact training set, the hidden layer uses a fully connected layer with a ReLU activation function, and the output layer contains a single neuron and uses a Sigmoid activation function to support the classification task. Initialize the network parameters, including weights and bias terms, and use normal distribution to generate initial values. The Adam optimization algorithm is used to update the network weights, and the learning rate is set to 0.001 and the batch size is 64. The difference between the model output and the actual label is measured by the cross entropy loss function. The loss value and accuracy curve of each iteration are recorded during the training process to evaluate the optimization effect. After several rounds of iterations, the optimization parameters obtained by training are extracted and output as abnormal reaction delay optimization parameter data, and the relevant log files are saved for subsequent use. The implementation method of constructing an initial vaccine risk dynamic early warning model for abnormal reaction delay optimization parameters based on the support vector machine algorithm is as follows. First, the abnormal reaction delay optimization parameters are mapped to the input space of the support vector machine algorithm, and the radial basis function (RBF) is selected as the kernel function according to the distribution characteristics of the optimization parameters. In the model construction stage, the initial values ​​of the penalty coefficient and kernel parameter of the support vector machine algorithm were set to 1.0 and 0.1, respectively. By using quadratic programming to solve the optimal decision boundary of the support vector machine, data classification segmentation is achieved in the high-dimensional feature space. In order to enhance the generalization ability of the model, the 5-fold cross-validation technology is used to conduct a preliminary evaluation of the model, and the range of parameter values ​​is adjusted according to the classification accuracy to gradually optimize the model performance. The final model constructed is a dynamic initial warning model for vaccine risks, which contains information such as the number of support vectors, the equation of the classification boundary, and the classification accuracy.The implementation method of logical testing of the vaccine risk dynamic initial warning model based on the abnormal reaction delay impact test set is as follows. Input the abnormal reaction delay impact test set into the vaccine risk dynamic initial warning model one by one, and output the risk prediction label according to the classification rules of the model. At the same time, compare the prediction results with the actual labels of the test set, and calculate the classification accuracy, recall rate and F1 score of the model on the test set. For the existing misclassified samples, analyze their feature distribution and the relationship with the decision boundary, and identify specific factors that affect the classification accuracy of the model. Further adjust the penalty coefficient and kernel parameters of the model, and optimize the model logic rules based on cross-validation repeated training and testing processes. Finally, combined with the analysis results of the logic test, generate a vaccine risk dynamic warning optimization model, and record the optimized model parameters and its performance on the test set, so as to perform risk dynamic warning operations in subsequent applications.

[0053] Preferably, step S332 includes the following steps: Performing weighted mapping of the training set on the abnormal reaction delay influence training set based on the neurons in the neural network model to obtain the influence weighted mapping training set; Perform network learning rate adaptation matching on the influence weighted mapping training set to obtain influence weighted learning rate adaptation data; The abnormal reaction delay impact training set is trained with parameter optimization according to the impact weighted learning rate adaptation data to obtain the abnormal reaction delay optimization parameters.

[0054] In an embodiment of the present invention, a specific implementation method for weighted mapping of a training set for an abnormal reaction delay impact training set is as follows. First, a neural network model comprising an input layer, a hidden layer, and an output layer is constructed, wherein the number of neurons in the input layer is consistent with the characteristic dimension of the training set, and the number of neurons in the hidden layer is determined by a grid search method to ensure that the model has appropriate expression capabilities. Initialize the network weights and bias terms, and generate random values ​​using a normal distribution. Input the training set data into the neural network model one by one, and calculate the weighted sum in each layer of neurons. The specific implementation method for adaptive matching of the network learning rate for the weighted mapping training set is as follows. Based on the optimization principle of the gradient descent algorithm, the learning rate is gradually adjusted to improve the stability and convergence speed of the training. First, the initial learning rate is set, and a dynamic adjustment strategy is adopted to adapt to the weighted characteristics of the training set. The dynamic adjustment strategy includes two schemes: cosine annealing and learning rate decay. In the implementation process, the gradient value is calculated by batch training (mini-batch), and the learning rate is dynamically adjusted according to the gradient change rate. The specific implementation method for parameter optimization training of the abnormal reaction delay impact training set according to the weighted learning rate adaptation data is as follows. Based on the influence of weighted learning rate adaptation data, the backpropagation algorithm is used to optimize the network parameters. First, the loss function is defined, and the cross entropy loss function is selected to measure the difference between the network output and the true label. In each round of iteration, the chain rule is used to calculate the gradient value of the loss function to the network weights and bias terms, and the network parameters are updated according to the dynamically adjusted learning rate. To avoid overfitting, the L2 regularization term is added and the loss trend is regularly monitored on the validation set. When the training reaches the preset rounds or the loss value tends to converge, the final optimized network weights and bias terms are extracted and output as the abnormal reaction delay optimization parameters, and the training records are archived in detail for subsequent analysis and reproduction.

[0055] Preferably, the characteristics and components of the vaccine itself include: vaccine formulation, vaccine strain, adjuvants and media.

[0056] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0057] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data, characterized in that: The following steps are involved: Step S1: obtaining a multi-source clinical medical data set of vaccine recipients through a medical management system; performing time-series sorting on the multi-source clinical medical cleansing data set to obtain a multi-source clinical medical sorting data set; Step S2: Acquire adverse reaction prompt data of vaccination; perform an impact analysis of delayed abnormal reaction of vaccination on a multi-source clinical medical ranking data set based on the adverse reaction prompt data of vaccination to obtain delayed abnormal reaction impact data of vaccination; perform individual abnormal reaction delay difference evaluation based on the delayed abnormal reaction impact data of vaccination to obtain individual abnormal reaction delay difference data; Step S3: classify the susceptible population based on the individual abnormal reaction delay difference data to obtain the individual physical sign susceptible population data; construct a vaccine risk dynamic early warning model based on the vaccination abnormal reaction delay impact data based on the support vector machine algorithm to obtain the vaccine risk dynamic early warning model; identify the vaccination reaction risk factors based on the individual physical sign susceptible population data according to the vaccine risk dynamic early warning model to obtain the vaccination risk factor data, and feed the vaccination risk factor data back to the terminal; Step S4: Automating the firmware design of the vaccine risk dynamic early warning model to obtain the vaccine risk dynamic early warning firmware; The vaccine risk dynamic warning firmware is embedded into the medical management system to perform dynamic vaccine assessment and warning.

2. The method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: Acquire a multi-source clinical medical data set of vaccine recipients through a medical management system; Step S12: performing data standardization processing on the multi-source clinical medical data set of vaccine recipients to obtain a multi-source clinical medical optimized data set; Step S13: Perform time-series sorting processing on the multi-source clinical medical optimization data set to obtain a multi-source clinical medical sorting data set; wherein the multi-source clinical medical sorting data set includes: vaccine recipient vital sign data, vaccination cycle data and diagnosis report.

3. The method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data according to claim 2, characterized in that: Step S12 includes the following steps: Step S121: performing data cleaning on the multi-source clinical medical data set of vaccine recipients to obtain a multi-source clinical medical cleansing data set, wherein the data cleaning includes: filling in missing values, removing duplicate values, and checking the consistency of data types; Step S122: performing date alignment processing on the multi-source clinical medical cleansing data set to obtain multi-source clinical medical date aligned data; Step S123: performing encoding processing on the multi-source clinical medical date alignment data to obtain multi-source clinical medical coding alignment data; Step S124: Perform data standardization processing on the multi-source clinical medical coding alignment data to obtain a multi-source clinical medical optimization data set.

4. The method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data according to claim 2, characterized in that: Step S2 includes the following steps: Step S21: Acquire adverse reaction prompt data of vaccination; Step S22: performing abnormal reaction coupling on the diagnosis report according to the adverse reaction prompt data of vaccination to obtain the abnormal reaction coupling data of vaccination and medical treatment; Step S23: analyzing the delayed impact of abnormal reactions to vaccination on the coupled data of abnormal reactions to vaccination visits based on the vaccine recipient's physical sign data and the vaccination cycle data, to obtain delayed impact data of abnormal reactions to vaccination; Step S24: Based on the vaccination abnormal reaction delay impact data, the individual abnormal reaction delay difference evaluation is performed on the vaccinee's physical sign data and the vaccination cycle data to obtain individual abnormal reaction delay difference data.

5. The method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data according to claim 4, characterized in that: Step S23 includes the following steps: Step S231: extracting the vaccine recipient's weight, height and age parameters from the vaccine recipient's physical sign data to obtain the vaccine recipient's weight / height / age parameters; Step S232: performing a multivariate linear regression analysis on the coupled data of abnormal reactions to vaccination visits according to the vaccine recipient's weight / height / age parameters and vaccination cycle data, to obtain linear regression data of physical signs-abnormal reactions to vaccination cycles; Step S233: performing independent variable long-acting reaction analysis on the linear regression data of physical sign-vaccination cycle abnormal reaction to obtain independent variable abnormal long-acting reaction data; Step S234: performing autocorrelation abnormal trend coupling on the linear regression data of the abnormal reaction of the sign-vaccination period according to the abnormal long-acting reaction data of the independent variable to obtain autocorrelation abnormal trend coupling data; Step S235: Based on the autocorrelation abnormal trend coupling data and the independent variable abnormal long-term reaction data, the delayed impact of abnormal reaction to vaccination is analyzed on the abnormal reaction coupling data of vaccination and medical treatment to obtain the delayed impact data of abnormal reaction to vaccination.

6. The method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data according to claim 4, characterized in that: Step S234 includes the following steps: Perform multi-scale decomposition of the trend evolution of the abnormal long-term response data of the independent variable to obtain the scale decomposition data of the abnormal trend evolution; According to the abnormal trend evolution scale decomposition data, the independent variable abnormal long-term reaction data is assigned independent variable weights to obtain the abnormal reaction trend independent variable weight data; Based on the abnormal reaction trend independent variable weight data, the autocorrelation abnormal trend coupling was performed on the abnormal reaction linear regression data of the sign-vaccination period to obtain the autocorrelation abnormal trend coupling data.

7. The method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: classifying susceptible populations based on the individual abnormal reaction delay difference data to obtain individual physical sign susceptible population data; Step S32: normalizing the delayed impact data of abnormal reaction to vaccination to obtain normalized data of delayed impact of abnormal reaction; Step S33: constructing a vaccine risk dynamic early warning model based on the support vector machine algorithm for the normalized data of abnormal reaction delay impact, to obtain a vaccine risk dynamic early warning model; Step S34: Identify vaccination reaction risk factors based on the individual physical sign susceptible population data according to the vaccine risk dynamic early warning model to obtain vaccination risk factor data; the vaccination risk factor data includes: the characteristics and ingredients of the vaccine itself, and the individual differences and characteristics of the vaccination subjects.

8. The method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data according to claim 7, characterized in that: Step S33 includes the following steps: Step S331: dividing the abnormal reaction delay influence normalized data into a data set to obtain an abnormal reaction delay influence training set and an abnormal reaction delay influence test set; Step S332: performing parameter optimization training on the abnormal reaction delay impact training set based on the neural network model to obtain abnormal reaction delay optimization parameters; Step S333: constructing an initial vaccine risk dynamic early warning model based on the abnormal reaction delay optimization parameters based on the support vector machine algorithm to obtain a vaccine risk dynamic initial early warning model; Step S334: Perform a logic test on the vaccine risk dynamic initial warning model according to the abnormal reaction delay impact test set to obtain a vaccine risk dynamic warning optimization model.

9. The method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data according to claim 8, characterized in that: Step S332 includes the following steps: Performing weighted mapping of the training set on the abnormal reaction delay influence training set based on the neurons in the neural network model to obtain the influence weighted mapping training set; Perform network learning rate adaptation matching on the influence weighted mapping training set to obtain influence weighted learning rate adaptation data; The abnormal reaction delay impact training set is trained with parameter optimization according to the impact weighted learning rate adaptation data to obtain the abnormal reaction delay optimization parameters.

10. The method for constructing a vaccine dynamic evaluation and early warning model based on multi-source data according to claim 7, characterized in that: The characteristics and components of the vaccine itself include: vaccine formulation, vaccine strain, adjuvants and media.

Images