A vaccine clinical trial quality management system optimization method and system

Abstract

The present invention relates to the field of quality management optimization and discloses a method and system for optimizing the quality management system of vaccine clinical trials. The method comprises: obtaining vaccine clinical trial data of a target trial population, constructing an interactive information network based on the extracted case-sharing trial information, performing node graph convolution propagation on the interactive information network to obtain a structural information embedding vector, identifying the structural outlier centers of the vaccine clinical trial data, and performing an abnormal type hierarchical analysis on the structural outlier centers to obtain a quality deviation classification result, and identifying local structural abnormal clusters based on the obtained abnormal propagation paths, generating a local quality control strategy based on the quality deviation classification results and the local structural abnormal clusters, and optimizing the vaccine clinical trial quality management system. The present invention can effectively capture the structural outlier centers in vaccine clinical trial data and construct a test data interactive graph structure in vaccine clinical trial data that lacks a natural interactive network.

Description

Technical Field

[0001] The present invention relates to the field of quality management optimization, and in particular to a method and system for optimizing the quality management system of vaccine clinical trials. Background Art

[0002] In current vaccine clinical trial quality management practices, data anomaly detection primarily relies on traditional statistical analysis methods and rule-setting techniques. The core process focuses on reviewing the integrity, range boundaries, and change trends of single case data. These methods typically focus on identifying anomalies in data integrity, range deviations, and time series fluctuations for individual cases. However, existing methods have significant limitations:

[0003] On the one hand, traditional anomaly detection methods, such as Z-score analysis, box plot method or isolation forest, usually make judgments based on the numerical characteristics of each case itself or the statistical distribution characteristics of all samples. However, traditional methods cannot detect coordinated anomalies in a certain structure or behavior of multiple cases. For example, multiple cases in a research center may have abnormalities in specific indicators at the same time, but each case does not exceed the limit when viewed individually; it is difficult to capture "collective deviations" caused by operational errors of the research center, equipment batch abnormalities or human interference, namely the so-called "structural outlier centers"; there is a lack of modeling capabilities for the "contextual structure of anomalies", and it is impossible to use the potential correlation between cases to assist in anomaly judgment.

[0004] On the other hand, graph neural networks (GNNs) have demonstrated excellent structured anomaly detection capabilities in recent years in fields such as financial anti-fraud, social networks, and recommendation systems. These methods rely on clear interactions or network connections between nodes (such as transactions, attention, and shared history). However, in vaccine clinical trial data, cases are typically independent observations, lacking explicit interactions in the trial design, such as shared drugs, doctors, or direct behavioral associations. The lack of a natural graph structure prevents the direct definition of edges between nodes (cases), hindering the implementation of graph neural network modeling. Even when attempting to construct similarity graphs, the highly heterogeneous data distribution (e.g., from different populations and centers) and inconsistent graph construction standards result in poor graph quality, impacting subsequent model performance.

[0005] Therefore, existing technologies have problems in effectively capturing structural outlier centers in vaccine clinical trial data that lack natural interaction networks, as well as in constructing the interaction graph structure of trial data. Summary of the Invention

[0006] The present invention provides a method for optimizing the quality management system of vaccine clinical trials, the main purpose of which is to solve the problems of difficulty in effectively capturing structural outlier centers in vaccine clinical trial data and difficulty in constructing the interactive graph structure of trial data in the absence of natural interaction networks.

[0007] In a first aspect, to achieve the above-mentioned objectives, the present invention provides a method for optimizing a vaccine clinical trial quality management system, comprising:

[0008] Acquiring vaccine clinical trial data of a target trial population, extracting case-sharing trial information of the vaccine clinical trial data, and constructing an interactive information network of the target trial population based on the case-sharing trial information;

[0009] Performing node graph convolution propagation on the interactive information network to obtain a structural information embedding vector;

[0010] Identifying structural outlier centers of the vaccine clinical trial data based on the structural information embedding vector, and performing a hierarchical analysis of abnormality types on the structural outlier centers to obtain a quality deviation classification result;

[0011] Obtaining an abnormal propagation path of the structural outlier center, and identifying a local structural abnormal cluster according to the abnormal propagation path;

[0012] A local quality control strategy is generated based on the quality deviation classification results and the local structural abnormality clusters, and the local quality control strategy is used to optimize the vaccine clinical trial quality management system.

[0013] In a second aspect, the present invention further provides a vaccine clinical trial quality management system optimization system, the system comprising:

[0014] An interactive network construction module is used to obtain vaccine clinical trial data of a target trial population, extract case-sharing trial information of the vaccine clinical trial data, and construct an interactive information network of the target trial population based on the case-sharing trial information;

[0015] An interactive network convolution module, configured to perform node graph convolution propagation on the interactive information network to obtain a structural information embedding vector;

[0016] an outlier center identification module, configured to identify the structural outlier centers of the vaccine clinical trial data based on the structural information embedding vector, and perform an abnormality type hierarchical analysis on the structural outlier centers to obtain a quality deviation classification result;

[0017] an abnormal cluster identification module, configured to obtain an abnormal propagation path of the structural outlier center and identify a local structural abnormal cluster based on the abnormal propagation path;

[0018] A control strategy generation module is used to generate a local quality control strategy based on the quality deviation classification results and the local structural abnormality cluster, and use the local quality control strategy to optimize the vaccine clinical trial quality management system.

[0019] The present invention obtains vaccine clinical trial data of a target trial group, extracts case-sharing trial information of the vaccine clinical trial data, and constructs an interactive information network of the target trial group based on the case-sharing trial information. Each subject in the target trial group is used as a network node. Combined with the interactive information network constructed based on the case-sharing trial information, the correlation and mutual influence between subjects on key trial features can be intuitively and systematically revealed. Node graph convolution propagation is performed on the interactive information network to obtain a structural information embedding vector. A node degree matrix is ​​constructed through an adjacency matrix and a node feature matrix. In combination with a node weight matrix, layer-by-layer aggregation and propagation of information between nodes are achieved, so that nodes gradually obtain richer contextual semantics and structural perception capabilities in each layer. The structural outlier center of the vaccine clinical trial data is identified based on the structural information embedding vector, and the structural outlier center is subjected to an abnormal type hierarchical analysis. , obtain the quality deviation classification results, accurately identify the structural outlier centers in the vaccine clinical trial data, effectively reveal potential abnormal individuals or abnormal patterns, combine the abnormal behavior rule table and rule activation vector cluster analysis constructed by structural characteristics and behavioral characteristics, realize detailed hierarchical classification of abnormal types, obtain the abnormal propagation path of the structural outlier center, and identify local structural abnormal clusters based on the abnormal propagation path. Combined with propagation weight mapping and path search, it effectively captures the weak connection and abnormal diffusion of information in the network, and then through the multi-dimensional evaluation of embedded space deviation and data integrity anomalies, accurately screen out high-risk nodes to form a focused and representative abnormal cluster structure. According to the quality deviation classification results and the local structural abnormal cluster, a local quality control strategy is generated, and the local quality control strategy is used to optimize the vaccine clinical trial quality management system, which can realize the transformation from macro management to refined and structured management and control. By accurately formulating and matching control strategy templates and targets based on local anomaly patterns generated based on quality deviation classification, network distribution characteristics and propagation patterns, and effectively identifying and intervening in anomaly clusters with specific structural risks and propagation behaviors, targeted regulation of high-risk nodes can be achieved, reducing the spread of anomalies, improving data integrity and test consistency, thereby enhancing the real-time, sensitivity and stability of quality management, and achieving closed-loop control from risk discovery, feature identification to strategy execution. The present invention can solve the problems of difficulty in effectively capturing structural outlier centers in vaccine clinical trial data and difficulty in constructing an interactive graph structure for test data in vaccine clinical trial data that lacks a natural interactive network. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0021] Figure 1 A schematic diagram of a process for optimizing a quality management system for vaccine clinical trials according to one embodiment of the present invention;

[0022] Figure 2 A schematic diagram of a module of a vaccine clinical trial quality management system optimization system provided by one embodiment of the present invention;

[0023] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, and to fully understand and implement how the present disclosure applies technical means to solve technical problems and achieve the corresponding technical effects, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. The embodiments of the present disclosure and the various features in the embodiments can be combined with each other without conflict, and the technical solutions formed are all within the scope of protection of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present disclosure.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] The present application provides an embodiment of a method for optimizing a quality management system for a vaccine clinical trial, which can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0027] Reference Figure 1 FIG2 is a flow chart of a method for optimizing a quality management system for clinical trials of vaccines according to an embodiment of the present invention. In this embodiment, the method for optimizing a quality management system for clinical trials of vaccines includes:

[0028] S1. Acquire vaccine clinical trial data of a target trial population, extract case-sharing trial information of the vaccine clinical trial data, and construct an interactive information network of the target trial population based on the case-sharing trial information.

[0029] In an embodiment of the present invention, during the course of a vaccine clinical trial, the subject groups participating in the vaccine clinical trial (such as age group, gender, underlying disease, region, etc.) are determined, and the test data of the target test group is obtained, including the behavior, reaction, physiological indicators, etc. of each subject in the trial. From these individualized data, representative shared test information at the group level is extracted, such as common adverse reaction characteristics, similar immune response patterns, or common medication compliance performance. Based on these case-shared test information, an interactive information network of the target test group is further constructed, with each individual as a node in the network and the similarity or correlation between individuals in test behavior or results as edges, thereby revealing the potential correlation structure between individuals and providing structured support for subsequent data analysis and quality control.

[0030] In detail, the case sharing trial information of the vaccine clinical trial data is extracted, including:

[0031] Performing feature alignment on the vaccine clinical trial data of each subject in the target trial population to obtain aligned trial data;

[0032] Performing unified processing on the time axis of the alignment test data to obtain standard test data;

[0033] extracting key test features of the standard test data, and normalizing the key test features to obtain normalized test features;

[0034] Randomly selecting two of the normalized test features as two target test features;

[0035] Calculating the feature similarity between every two target test features, and constructing a feature matrix based on the feature similarities;

[0036] Constructing a Laplace matrix according to the feature similarity and the feature matrix;

[0037] Performing eigendecomposition on the Laplace matrix, and sorting the decomposed initial eigenvectors from small to large to obtain a sorted eigenvector sequence;

[0038] Starting from the smallest initial feature vector in the sorted feature vector sequence, screening out a target feature vector with a preset number of features;

[0039] Clustering the target feature vector to obtain a plurality of target feature vector clusters;

[0040] The number of vectors in each target feature vector cluster is counted, and the target feature vector cluster corresponding to the largest number of vectors is used as case sharing test information.

[0041] In detail, key feature dimensions in the test data are obtained, such as age, gender, vaccination time, dose, immune response indicators, etc., and standardization, normalization, interpolation filling or unified coding methods are used to deal with the problems of missing, heterogeneous or inconsistent formats in the data of different subjects. According to the preset alignment rules or reference templates, the characteristics of each subject are mapped to a unified feature space to ensure that the data are comparable and consistent in structure, and the aligned test data containing unified feature representation are output to provide a reliable basis for subsequent analysis.

[0042] Determine a unified time reference point, such as the date of the first vaccination or the start date of the trial, as the benchmark of the time axis, and reposition the various test data of each subject (such as vaccination records, physiological indicators, immune responses, etc.) relative to this reference point and convert them into standard time node representations (such as day 0, day 7, day 28, etc.). For time offsets, differences in recording frequency, or missing time points among different subjects, adjustments and corrections are made through interpolation, time alignment algorithms, or window smoothing to obtain standard test data with a unified time scale and synchronous observation points, which are convenient for horizontal comparison and trend analysis.

[0043] According to the objectives of the vaccine clinical trials, key trial features that are representative and have analytical value are selected, such as antibody levels, body temperature changes, white blood cell counts, adverse reaction scores, etc., and statistical analysis is performed on the feature values ​​of different subjects to identify the numerical range, distribution pattern and outliers of each feature. Appropriate normalization methods (such as minimum-maximum scaling, Z-score standardization or logarithmic transformation) are used to convert feature values ​​of different dimensions and scales into a unified interval or standard distribution, and the normalized trial feature data is output to eliminate the impact of dimensional differences on subsequent analysis models.

[0044] Two features are randomly selected from the normalized test features as target test features, such as antibody levels and body temperature changes. For each two target test features, normalized numerical sequences are extracted from all subject samples respectively, and similarity calculation methods (such as Pearson correlation coefficient, cosine similarity or the opposite of Euclidean distance) are used to measure their similarity in numerical change trends. The calculated similarity values ​​are filled in the corresponding feature pair positions to construct a symmetric feature matrix. Each element in the matrix represents the similarity relationship between the corresponding two target test features. The feature matrix obtained provides basic data support for subsequent feature clustering, feature selection or correlation analysis.

[0045] According to the feature similarity, a feature matrix is ​​constructed, in which each element represents the similarity between a pair of target test features. The similarity matrix is ​​calculated by adding all the similarity values ​​of each row of the feature matrix to obtain a diagonal matrix, which represents the total similarity of each feature with other features. According to the formula ,in, represents the degree matrix, Represents the characteristic matrix, and the non-normalized Laplace matrix is ​​calculated , reflecting the structural relationship between the features and laying the foundation for subsequent spectral decomposition and feature screening.

[0046] The constructed Laplace matrix is ​​subjected to eigenvalue decomposition to obtain a set of eigenvalues ​​and their corresponding eigenvectors, where each eigenvector represents a potential characteristic structural pattern. All eigenvectors are sorted from small to large according to the size of the corresponding eigenvalue to form a sorted eigenvector sequence. Since eigenvectors with smaller eigenvalues ​​usually carry more important structural information, starting from the eigenvector with the smallest eigenvalue, the first several eigenvectors are selected in sequence until the preset number of features is reached, and finally the target eigenvector set for subsequent analysis is obtained.

[0047] Use appropriate clustering algorithms (such as K-means, spectral clustering or DBSCAN) to perform cluster analysis on the target feature vectors obtained by screening, divide similar vectors into several target feature vector clusters, count the number of feature vectors contained in each vector cluster, and reflect its representativeness in the overall feature structure. Select the cluster with the largest number of vectors from all vector clusters. It is believed that the characteristics represented by the cluster with the largest number of vectors have the highest commonality or consistency in the experimental group. The feature vector cluster corresponding to the cluster with the largest number of vectors is used as the case sharing test information.

[0048] In an embodiment of the present invention, the constructing of the interactive information network of the target test group based on the case-sharing test information includes:

[0049] Taking each subject in the target trial population as a network node;

[0050] Calculating the vector similarity between every two target feature vectors in the case sharing test information;

[0051] Determining whether the vector similarity is greater than a preset similarity threshold;

[0052] If the vector similarity is greater than the similarity threshold, a network edge is established between the two target feature vectors corresponding to the vector similarity greater than the similarity threshold;

[0053] If the vector similarity is less than or equal to the similarity threshold, no network edge is established between the two target feature vectors corresponding to the vector similarity less than or equal to the similarity threshold;

[0054] An interactive information network of the target test group is constructed using the network nodes and the network edges.

[0055] In detail, treating each subject in the target trial population as an independent node in the network means mapping the data entity of each subject into a node in the network structure, which makes it easier to represent the relationships and interactions between individuals through a graph model.

[0056] For each pair of target feature vectors in the case sharing test information, the vector similarity between them is calculated. Common methods include cosine similarity or Euclidean distance transformed similarity, and the calculated similarity is compared with the similarity threshold: when the similarity of a pair of vectors is greater than the threshold, the two feature vectors are considered to have a high correlation, and a network edge is established between them to represent the connection relationship; when the similarity of a pair of vectors is less than or equal to the threshold, no edge is established to ensure that the network structure only retains significant feature associations, thereby improving the expression accuracy and interpretability of the network.

[0057] The node corresponding to each subject is used as the basic unit in the network. Based on the similarity relationship between the feature vectors calculated previously, network edges are added between node pairs that meet the similarity threshold conditions to represent the association or mutual influence between subjects on key experimental features, forming a network structure that reflects the interaction and similarity of subjects within the group.

[0058] By treating each subject in the target trial population as a network node and combining it with an interactive information network built based on case-sharing trial information, the correlation and mutual influence between subjects on key trial characteristics can be intuitively and systematically revealed. This graph structure not only facilitates the capture of complex multi-dimensional relationships within a population, but also supports in-depth exploration using graph theory and network analysis methods, such as identifying important nodes, discovering potential trial population substructures, and predicting information propagation paths. This improves the ability to interpret vaccine clinical trial data and the level of quality management, facilitating precision medical decision-making and personalized intervention.

[0059] S2. Perform node graph convolution propagation on the interactive information network to obtain a structural information embedding vector.

[0060] In an embodiment of the present invention, based on the constructed target trial group interaction information network, graph neural networks (such as GCN, GraphSAGE) or random walk-based embedding methods (such as DeepWalk, node2vec) are used to fully capture the local neighbor relationships, group association patterns and global network structure information of nodes, thereby obtaining an embedding vector containing structural information, which can be used as a feature input for subsequent analysis tasks such as individual behavior modeling, anomaly identification or quality assessment, thereby improving the depth and accuracy of vaccine clinical trial data mining.

[0061] In detail, performing node graph convolution propagation on the interactive information network to obtain a structural information embedding vector includes:

[0062] Obtaining an adjacency matrix of the interactive information network and a node feature matrix of each network node in the interactive information network;

[0063] Constructing a node degree matrix according to the adjacency matrix and the node feature matrix;

[0064] Obtaining a weight matrix for each network node, and updating each network node in the interactive information network layer by layer using the weight matrix, the adjacency matrix, and the node degree matrix to obtain a final updated node in the final layer;

[0065] Each of the final updated nodes is embedded as a structural information vector.

[0066] In detail, the adjacency matrix is ​​extracted from the constructed interactive information network. The adjacency matrix represents whether there is a connection relationship or the strength of the connection between nodes in the network in the form of binary values ​​or weights. The node feature matrix corresponding to each network node is obtained, and the key experimental feature values ​​or normalized feature representations of each subject are recorded. The degree of each node is calculated according to the adjacency matrix, that is, the number of edges connecting the node to other nodes or the total connection weight, and the degree values ​​of all nodes are filled in the diagonal position to form a node degree matrix, which reflects the connection activity of each subject in the network and provides basic support for subsequent structure-aware analysis such as graph convolution.

[0067] An initial weight matrix is ​​assigned to each node in the interactive information network, which is usually initialized by learnable parameters or feature transformation functions to characterize the representation ability of the node in the feature propagation process. Combining the weight matrix, adjacency matrix and node degree matrix, a graph neural network (such as graph convolutional network, GCN) is used to update the node features layer by layer: in each layer, the node receives feature information from the adjacent nodes through the adjacency relationship, and at the same time combines its own features and weight matrix for weighted aggregation, and normalizes it through the node degree matrix to ensure that the feature propagation process is stable and effective. After iterating multiple layers in this way, the updated representation of each node in the last layer is obtained, that is, the final updated node. The representation of these final updated nodes is used as the structural information embedding vector, which comprehensively reflects the position, associated features and group context information of each subject in the network structure, and provides a structurally aware deep representation for subsequent modeling and decision-making.

[0068] By performing node graph convolutional propagation on the interactive information network, it is possible to fuse each subject's own key experimental characteristics and the associated information of their neighboring nodes while fully utilizing the network structure information. The node degree matrix is ​​constructed through the adjacency matrix and the node feature matrix, and the node weight matrix is ​​combined to achieve layer-by-layer aggregation and propagation of information between nodes, so that the nodes gradually gain richer contextual semantics and structural perception capabilities in each layer. The resulting structural information embedding vector not only retains individual characteristics but also encodes the structural position and association pattern in the entire experimental group, providing a more comprehensive and expressive feature representation for subsequent vaccine trial analysis, individual difference modeling, and group behavior prediction.

[0069] S3. Identify the structural outlier centers of the vaccine clinical trial data based on the structural information embedding vector, and perform a hierarchical analysis of the abnormality types on the structural outlier centers to obtain a quality deviation classification result.

[0070] In this embodiment of the present invention, the structural features of each individual in the network are mapped into a comparable low-dimensional representation. Outlier detection methods (such as LOF, KNN distance methods, or clustering-based anomaly identification) are used to identify structural outlier centers that significantly deviate from the group distribution in the embedding space. These are individual subjects exhibiting abnormal network structure. Combined with the original experimental behavioral characteristics of the outliers (such as data missing patterns, abnormal reaction times, and isolated network locations), a hierarchical analysis of the outlier centers is performed to identify possible causes such as operational bias, data collection errors, or abnormal physiological reactions. This classification results in a classification of structural quality deviations in clinical trials, providing a basis for data review and risk control.

[0071] In detail, identifying the structural outlier center of the vaccine clinical trial data based on the structural information embedding vector includes:

[0072] Calculating the Euclidean distance between each of the structural information embedding vectors;

[0073] Sort the neighbor vectors of each structural information embedding vector in ascending order according to the Euclidean distance to obtain the number of sorting bits of each neighbor vector;

[0074] The neighbor vector having the number of sorted bits being the preset number of neighbors is used as the target neighbor vector;

[0075] Taking the Euclidean distance between the structural information embedding vector and the target adjacent vector as the target distance;

[0076] Calculating the reachable distance from each of the structural information embedding vectors to the neighbor vectors according to the target distance;

[0077] Calculating a local reachable density of each of the structural information embedding vectors according to the reachable distance, and generating a local outlier factor value using the local reachable density;

[0078] Determining whether the local outlier factor value is greater than a preset outlier threshold;

[0079] If the local outlier factor value is less than or equal to the outlier threshold, the structural information embedding vector corresponding to the local outlier factor value less than or equal to the outlier threshold is not a structural outlier center;

[0080] If the local outlier factor value is greater than the outlier threshold, the structural information embedding vector corresponding to the local outlier factor value greater than the outlier threshold is used as the structural outlier center.

[0081] Specifically, for each subject's structural information embedding vector in the interactive information network, the numerical representation in the high-dimensional embedding space is extracted. For any two embedding vectors, the distance between them is calculated according to the Euclidean distance formula to obtain their straight-line distance in the feature space. The Euclidean distance between all structural information embedding vectors is calculated pairwise, and finally a symmetric distance matrix is ​​formed. The Euclidean distance calculation formula is as follows:

[0082]

[0083] in, Indicates the structural information embedding vector, Indicates the structural information embedding vector, Indicates the The structural information embedding vector is The Euclidean distance between the structural information embedding vectors.

[0084] For each embedding vector, all calculated Euclidean distances are sorted in ascending order to represent the neighbor relationship with high to low similarity. The embedding vectors of the first several nearest neighbors are selected from the sorting results according to the preset number of neighbors as the target neighbor vectors of the structural information embedding vector, thereby constructing a local similarity relationship and providing a basis for subsequent clustering, anomaly detection or individual association modeling.

[0085] In the calculated Euclidean distance matrix between the structural information embedding vectors, for each embedding vector, a preset number of nearest neighbor vectors (i.e., target neighbor vectors) are extracted, and the corresponding Euclidean distance values ​​between these neighbor vectors and the structural information embedding vector are found. These distances are used as the target distances between the structural information embedding vector and the local neighborhood.

[0086] According to the target distance between each structural information embedding vector and the target adjacent vector, the reachable distance from the structural information embedding vector to each neighbor vector is calculated. It is defined as the maximum value between the target distance and the minimum reachable distance of the neighbor vector, that is, the distance is smoothed in the local range. The calculation formula is as follows:

[0087]

[0088] in, Indicates the The structural information embedding vector is The target distance of the target adjacent vectors, Indicates the The structural information embedding vector is combined with the neighbor The Euclidean distance between the structural information embedding vectors, Indicates the The structural information is embedded into the vector The reachability distance between neighbor vectors.

[0089] These reachable distances are used to calculate the local reachable density of each embedding vector, which indicates the density of the structural information embedding vector in its neighborhood. The calculation formula is as follows:

[0090]

[0091] in, Indicates the The structural information embedding vector neighbor vectors, express The neighbor vector structural information embedding vector, Indicates the The structural information is embedded into the vector The reachable distance between neighbor vectors, Indicates the The local reachable density of the structural information embedding vector.

[0092] The local outlier factor (LOF) is obtained by performing a ratio analysis on the local reachability density and the local reachability density of the adjacent vectors. This value is used to measure whether the vector deviates from its neighborhood density. The higher the outlier factor value, the more likely the node is to be a structural anomaly or a potential anomaly individual. The calculation formula is as follows:

[0093]

[0094] in, Indicates the The structural information embedding vector neighbor vectors, express The neighbor vector structural information embedding vector, Indicates the The local reachable density of the structural information embedding vector, Indicates the The local reachable density of the structural information embedding vector, Indicates the The local outlier factor value of the structural information embedding vector.

[0095] The local outlier factor value corresponding to each embedding vector is compared with the outlier threshold: if the outlier factor value of a structural information embedding vector is less than or equal to the outlier threshold, it means that the local density of the structural information embedding vector is similar to that of the surrounding neighbors and there is no significant abnormality. Therefore, the structural information embedding vector is not considered a structural outlier center; if the outlier factor value of the structural information embedding vector is greater than the outlier threshold, it indicates that the density of the structural information embedding vector in its neighborhood is significantly low, which may represent an individual or region with abnormal structural characteristics in the experimental population. In this case, the structural information embedding vector is marked as a structural outlier center.

[0096] In an embodiment of the present invention, performing an abnormality type hierarchical analysis on the structural outlier center to obtain a quality deviation classification result includes:

[0097] Extracting structural features and behavioral features of the structural outlier center;

[0098] Constructing a triggerable abnormal behavior rule table based on the structural features and the behavioral features;

[0099] Mapping each of the structural outlier centers into a rule activation vector according to the triggerable abnormal behavior rule table;

[0100] Calculating the overlap of each of the rule activation vectors, and dividing the rule activation vectors whose overlap is greater than a preset overlap threshold into a group of target activation vector clusters;

[0101] Identifying the activation rule set corresponding to the target activation vector cluster in the triggerable abnormal behavior rule table;

[0102] Analyzing the abnormality type of the target activation vector cluster according to the activation rule set to obtain an abnormality type classification;

[0103] A quality deviation classification result of each of the structural outlier centers is generated according to the anomaly type classification.

[0104] In detail, the corresponding structural features are extracted from the identified structural outlier centers, such as the connectivity of nodes in the network, the spatial distribution characteristics of the embedding vectors, the distance distribution to neighbors, etc., as well as the position attributes of the structural outlier centers in the interactive information network. At the same time, the behavioral characteristics of the structural outlier centers are extracted, including abnormal immune responses, adverse event frequencies, and fluctuation trends of key indicators shown in vaccine trials. These structural and behavioral characteristics are summarized and organized to identify reproducible abnormal patterns or feature combinations. Based on this, a rule table that can trigger abnormal behavior is constructed, defining abnormal situations that may occur when specific structural characteristics and behavioral characteristics are met at the same time, providing an explainable and executable basis for subsequent risk warnings, abnormal screening or decision-making interventions.

[0105] For each structural outlier center, the abnormal behavior rules in the rule table are matched one by one to determine whether the structural and behavioral characteristics of the structural outlier center meet the triggering conditions of the corresponding rules. According to the set of satisfied rules, the position of the corresponding rule is marked as activated, forming a multi-dimensional binary or weight vector, namely the rule activation vector, which intuitively reflects the activation mode of the outlier center in the abnormal behavior rule space, and provides a structured expression and basis for subsequent abnormal analysis, risk assessment and precise intervention.

[0106] The overlap between each pair of rule activation vectors is calculated. The similarity in their activation rules is usually measured by the ratio of intersection to union (such as the Jaccard similarity coefficient). Rule activation vectors with an overlap greater than a preset threshold are grouped together to form several target activation vector clusters, representing sets of structural outlier centers with similar abnormal behavior patterns. Combined with the table of triggerable abnormal behavior rules, the activation rule set corresponding to each activation vector cluster is identified, that is, the abnormal rules triggered by all activation vectors in the cluster. This provides a basis for in-depth analysis of anomaly types and their associated mechanisms.

[0107] Based on the abnormal rules triggered by each activation vector cluster, the abnormal behavior patterns and characteristics represented are summarized, and then the abnormal types of the structural outlier centers in the cluster are classified. Combined with the abnormal type classification results, the quality deviation performance of each structural outlier center in the vaccine clinical trial is evaluated, such as abnormal immune response, high adverse events or abnormal data fluctuations, etc., and the corresponding quality deviation classification results are generated, which helps to accurately identify and manage abnormal individuals of different categories and improve the reliability of test data and the depth of analysis.

[0108] By calculating Euclidean distance and using the local outlier factor (LOF) method based on structural information embedding vectors, we can accurately identify structural outlier centers in vaccine clinical trial data, effectively revealing potential abnormal individuals or patterns. By combining structural and behavioral features to construct an abnormal behavior rule table and rule activation vector cluster analysis, we can achieve detailed hierarchical classification of abnormal types and generate quality deviation classification results. This systematic process not only improves the accuracy and interpretability of anomaly detection, but also supports a deeper understanding of multi-dimensional abnormal phenomena in complex vaccine trial populations, helping to promptly identify abnormal risks in trial data, ensure trial quality and safety, and promote the implementation of precise risk control and personalized intervention.

[0109] S4. Obtain an abnormal propagation path of the structural outlier center, and identify a local structural abnormal cluster based on the abnormal propagation path.

[0110] In an embodiment of the present invention, by analyzing the connection relationship of the structural outlier center in the interactive information network, the abnormal propagation path of the structural outlier center is further obtained, that is, expanding along the graph structure to the adjacent nodes, tracing the potential abnormal diffusion chain, reflecting the range of nodes that the outlier center may affect, and also revealing the possible local concentrated distribution trend of the abnormality. Combining the structural characteristics and behavioral characteristics of the nodes on the propagation path, a group of nodes that are closely connected to the outlier center in the graph structure and have similar abnormal characteristics are further identified, forming a local structural abnormal cluster, which provides support for the traceability of quality deviations and local risk control in vaccine clinical trials.

[0111] In detail, obtaining the abnormal propagation path of the structural outlier center and identifying the local structural abnormal cluster according to the abnormal propagation path includes:

[0112] Obtaining propagation weights, and mapping the propagation weights to the interactive information network to obtain an updated interactive network;

[0113] Performing diffusion path search on the structural outlier center in the updated interaction network, and taking the path whose propagation weight is less than a preset weight threshold during the path search as an abnormal propagation path;

[0114] Calculating the embedding space deviation degree and data integrity anomaly degree of each network node on the abnormal propagation path;

[0115] Screening out high-risk nodes on the abnormal propagation path according to the embedding space deviation degree and the data integrity abnormality degree;

[0116] Each of the structural outlier centers and the corresponding high-risk nodes on the anomaly propagation path are aggregated into a local structural anomaly cluster.

[0117] In detail, the propagation weights that reflect the intensity of information transmission between nodes are obtained. These weights are calculated based on feature similarity, interaction frequency or historical correlation. These propagation weights are mapped to the corresponding network edges in the interactive information network, replacing or adjusting the original edge weights, thereby forming an updated interactive network that more accurately reflects the actual information flow and influence strength between nodes, providing a more representative network structure foundation for subsequent propagation analysis, node importance evaluation or dynamic simulation.

[0118] Taking the structural outlier center as the starting point, diffusion path search is carried out to track the propagation path of information or anomalies in the network. During the search process, the propagation weight on the path is monitored in real time. If the propagation weight in a certain path is lower than the weight threshold, it means that the path information transmission is weak or the risk of abnormal propagation is high, and the path is marked as an abnormal propagation path. By identifying these abnormal propagation paths, it helps to reveal potential abnormal diffusion mechanisms and key influencing nodes, providing a basis for risk control and abnormal intervention in vaccine clinical trials.

[0119] On the abnormal propagation path, for each network node, the degree of deviation of each network node in the embedding space is calculated, that is, the distance difference between the structural information embedding vector of each network node and the average embedding vector of the normal node group is measured, which reflects the degree of abnormality in the network structure. The degree of abnormality in the data integrity of each network node is evaluated, and the data quality deviation is quantified by detecting the proportion of missing values, the frequency of abnormal values ​​or abnormal data fluctuations in the vaccine clinical trial data of each network node. Combining these two indicators, the abnormal characteristics of each node on the path are fully revealed, providing strong support for the cause analysis and risk positioning of abnormal propagation.

[0120] The degree of embedding space deviation and the degree of data integrity anomaly are combined to form a unified risk scoring indicator, which reflects the risk level of the node in terms of structural anomaly and data quality. Based on the risk threshold, the node risk score is screened, and the nodes with scores exceeding the threshold are selected as high-risk nodes. This helps to accurately identify key nodes that exhibit both structural anomalies and data anomalies in the abnormal propagation path, providing targets for key monitoring and intervention.

[0121] Collect all high-risk nodes identified by each structural outlier center on the abnormal propagation path, and form a cluster with the structural outlier center and these associated nodes to reflect the structural anomalies and potential risk collection within the local network. The formed local structural anomaly cluster can centrally display abnormal nodes and propagation relationships, which is convenient for in-depth analysis of abnormal patterns, locating abnormal sources and formulating targeted risk control strategies.

[0122] By obtaining the abnormal propagation path of the structural outlier center and identifying local structural abnormal clusters, we can systematically reveal the propagation mechanism and key nodes of abnormal phenomena in the vaccine clinical trial group. Combined with propagation weight mapping and path search, we can effectively capture the weak connection and abnormal diffusion of information in the network. Then, through the multi-dimensional evaluation of embedded spatial deviation and data integrity anomalies, we can accurately screen out high-risk nodes and form a focused and representative abnormal cluster structure. This not only improves the accuracy and timeliness of anomaly detection, but also provides a scientific basis for risk positioning, abnormal intervention and quality control, which helps to ensure the authenticity of test data and the overall safety of the test.

[0123] S5. Generate a local quality control strategy based on the quality deviation classification results and the local structural abnormality clusters, and use the local quality control strategy to optimize the vaccine clinical trial quality management system.

[0124] In an embodiment of the present invention, based on the obtained quality deviation classification results and local structural abnormality clusters, combined with the node characteristics and propagation patterns in the abnormal propagation path, targeted differentiated intervention measures are formulated for each abnormality cluster, such as data review of key nodes, operation process review of specific centers, rectification templates matching abnormality types, etc., thereby forming a closed-loop control at the source and propagation link, and improving the overall quality and stability of clinical trial data. The local quality control strategy is embedded in the vaccine clinical trial quality management system, and precise intervention is implemented for relevant subjects, centers or process links. The management process is iteratively optimized based on the execution effect, thereby achieving intelligent, structured and dynamic improvement of the overall trial quality system.

[0125] In detail, the generating of a local quality control strategy based on the quality deviation classification result and the local structural abnormality cluster includes:

[0126] Obtaining the abnormality type classification in the quality deviation classification result, and counting the composition ratio of the abnormality type classification in the local structural abnormality cluster;

[0127] identifying distribution characteristics of the local structural anomaly clusters in the updated interaction network;

[0128] Performing feature analysis on the propagation path sensitivity of the abnormal propagation path, and generating a propagation pattern of the local structural abnormal cluster based on the sensitivity features obtained from the analysis;

[0129] generating a local anomaly pattern according to the composition ratio, the distribution characteristics, and the propagation pattern;

[0130] generating a control strategy template and a control target according to the local abnormality pattern;

[0131] The control strategy corresponding to each of the local structural abnormality clusters is selected according to the control strategy template and the control target to obtain a local quality control strategy.

[0132] In detail, the abnormality type classification information corresponding to each structural outlier center in the quality deviation classification results is obtained. For each local structural abnormality cluster, the abnormality type classification distribution of all members in the cluster is statistically analyzed, and the proportion of each abnormality type in the cluster is calculated. The composition ratio of different abnormality types in each abnormal cluster is clarified, revealing the diversity of local abnormality patterns and the dominant abnormality types, providing a quantitative basis for abnormality analysis and precise intervention.

[0133] The abnormal nodes in each abnormal cluster are mapped in the updated interaction network, and the spatial distribution of abnormal nodes in the overall network structure, the connection density between nodes, and the relative position relationship between abnormal nodes and normal nodes are analyzed. By calculating network indicators such as the aggregation, marginality, and centrality of the abnormal cluster, it is determined whether the abnormal nodes are concentrated in a local area, distributed discretely, or clustered around key nodes. This reveals the embedded position and structural role of the abnormal cluster in the network, providing key support for understanding the abnormal propagation path, the scope of abnormal impact, and the positioning of high-risk areas.

[0134] In detail, the feature analysis of the propagation path sensitivity of the abnormal propagation path and the generation of the propagation pattern of the local structural abnormal cluster according to the sensitivity features obtained by the analysis include:

[0135] Constructing an abnormal cluster intra-propagation path graph according to the abnormal propagation path;

[0136] Perturbing the propagation path graph within the abnormal cluster to obtain an initial node length and a length after path propagation of each propagation node in the propagation path graph within the abnormal cluster after the disturbance;

[0137] Obtaining a path length growth rate based on a quotient of the initial node length and the path length after propagation;

[0138] Obtaining the number of activated abnormal propagation nodes in the propagation path graph within the abnormal cluster after the disturbance, and generating a node propagation gain according to the number of abnormal propagation nodes;

[0139] generating a sensitivity feature of the abnormal propagation path according to the path length growth rate and the node propagation gain;

[0140] Acquire a propagation mode type, match the propagation mode type according to the sensitivity feature, and obtain a propagation mode of a local structural anomaly cluster corresponding to the abnormal propagation path.

[0141] In detail, a propagation path graph within the anomaly cluster is constructed based on the anomaly propagation path, and directed connections are established between the structural outlier centers and the corresponding high-risk nodes according to the propagation path and propagation direction to form a path graph structure that reflects the anomaly diffusion relationship. Perturbation operations are performed on the propagation path graph within the anomaly cluster, such as randomly adjusting edge weights, rearranging path order, or introducing local connection changes, to simulate the impact of network structure changes on the propagation process; in the perturbed path graph, the original path length of each propagation node when starting from the initial node and the new path length generated by the perturbation path propagation are calculated. By comparing the difference between the two, the impact of network structure changes on anomaly propagation efficiency and path stability is evaluated, providing a basis for anomaly propagation mechanism analysis and system robustness evaluation.

[0142] According to the propagation path diagram within the abnormal cluster before and after the disturbance, the initial node length and the path propagation length after the disturbance of each propagation node are obtained respectively, and the two are divided to calculate the path length growth rate of each node, which is used to measure the extension degree of the propagation path under the influence of the disturbance. The number of abnormal propagation nodes that are successfully activated (i.e., still participating in the abnormal propagation process) in the propagation path diagram after the disturbance is counted, and the node propagation gain is generated based on the change between the number of abnormal propagation nodes and the original number of activated nodes, reflecting the amplification or suppression effect of the disturbance on the abnormal diffusion range.

[0143] Based on the path length growth rate and node propagation gain of each abnormal propagation path, the sensitivity characteristics of the abnormal propagation path are jointly generated to describe the comprehensive response intensity of the propagation path graph within the abnormal cluster to the propagation efficiency and diffusion range under network disturbances, and obtain a library of propagation mode types. Each type represents a specific abnormal diffusion characteristic, such as "rapid diffusion type", "edge blocking type" or "center aggregation type". By similarity matching the sensitivity characteristics of the abnormal propagation path with the feature templates of each propagation mode type, the propagation mode that best meets the path propagation behavior characteristics is identified. Finally, it is used as the propagation mode corresponding to the local structural abnormal cluster to guide risk assessment, propagation prediction and intervention strategy formulation.

[0144] In detail, the dominant anomaly type in the anomaly cluster is determined based on the composition ratio, the abnormal behavior characteristics within the anomaly cluster are characterized, and combined with the distribution characteristics of the cluster in the network, the position attributes in the overall structure are clarified, such as centralized, discrete or marginalized distribution, and the propagation pattern information is integrated to identify the diffusion mechanism and potential risk range of the anomaly cluster; through the organic integration of these three core elements, a complete, concrete and discernible local anomaly pattern is formed.

[0145] Based on the identified local abnormal patterns, the dominant abnormality types, network distribution characteristics and propagation characteristics of the local abnormal patterns are extracted, and the control strategy template library is matched accordingly to generate targeted control strategy templates and corresponding control objectives, such as reducing the propagation range, repairing data integrity or stabilizing the network structure, etc. These strategy templates and control objectives are applied to specific local structural abnormality clusters. By matching the abnormal characteristics of the local structural abnormality clusters with the control requirements, the most suitable intervention method is selected, and finally a local quality control strategy for each local structural abnormality cluster is formed to achieve personalized and precise vaccine clinical trial quality management and risk prevention and control.

[0146] The control strategies formulated based on local abnormality patterns will be embedded in the trial quality management system as a refined intervention mechanism for structural abnormality clusters. These strategies will be called in real time during actual trial data monitoring and network analysis, and classified governance will be implemented for identified high-risk abnormality clusters, such as adjusting data collection processes, repairing abnormal propagation paths, and enhancing data integrity management of key nodes. By deeply integrating local control strategies with quality management processes, closed-loop control from risk discovery, feature identification to strategy execution will be achieved, thereby improving the response efficiency, intervention accuracy and system stability of the entire clinical trial system to quality deviations.

[0147] Optimizing the quality management system for vaccine clinical trials using localized quality control strategies can achieve a shift from macro-level management to refined, structured control. By accurately developing and matching control strategy templates and objectives based on localized anomaly patterns generated based on quality deviation classification, network distribution characteristics, and propagation patterns, anomaly clusters with specific structural risks and propagation behaviors can be effectively identified and intervened. This strategy enables targeted regulation of high-risk nodes, reduces the spread of anomalies, and improves data integrity and trial consistency, thereby enhancing the real-time, sensitivity, and stability of quality management, ultimately ensuring the scientific nature, safety, and compliance of vaccine clinical trials.

[0148] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0149] like Figure 2 1 is a functional module diagram of a vaccine clinical trial quality management system optimization system provided by one embodiment of the present invention.

[0150] In the embodiment of the present disclosure, a vaccine clinical trial quality management system optimization system is provided, which corresponds one-to-one to the vaccine clinical trial quality management system optimization method of the above embodiment. Figure 2 As shown, the vaccine clinical trial quality management system optimization system 100 includes an interactive network construction module 101, an interactive network convolution module 102, an outlier center identification module 103, an abnormal cluster identification module 104, and a control strategy generation module 105. The functional modules are described in detail as follows:

[0151] An interactive network construction module 101 is used to obtain vaccine clinical trial data of a target trial population, extract case-sharing trial information of the vaccine clinical trial data, and construct an interactive information network of the target trial population based on the case-sharing trial information;

[0152] An interactive network convolution module 102 is configured to perform node graph convolution propagation on the interactive information network to obtain a structural information embedding vector;

[0153] An outlier center identification module 103 is used to identify the structural outlier centers of the vaccine clinical trial data based on the structural information embedding vector, and perform abnormal type hierarchical analysis on the structural outlier centers to obtain a quality deviation classification result;

[0154] Anomaly cluster identification module 104 is used to obtain the abnormal propagation path of the structural outlier center and identify local structural abnormal clusters based on the abnormal propagation path;

[0155] The control strategy generation module 105 is used to generate a local quality control strategy based on the quality deviation classification results and the local structural abnormality clusters, and optimize the vaccine clinical trial quality management system using the local quality control strategy.

[0156] In one embodiment, when extracting case sharing trial information from the vaccine clinical trial data, the interactive network construction module 101 is used to:

[0157] Performing feature alignment on the vaccine clinical trial data of each subject in the target trial population to obtain aligned trial data;

[0158] Performing unified processing on the time axis of the alignment test data to obtain standard test data;

[0159] extracting key test features of the standard test data, and normalizing the key test features to obtain normalized test features;

[0160] Randomly selecting two of the normalized test features as two target test features;

[0161] Calculating the feature similarity between every two target test features, and constructing a feature matrix based on the feature similarities;

[0162] Constructing a Laplace matrix according to the feature similarity and the feature matrix;

[0163] Performing eigendecomposition on the Laplace matrix, and sorting the decomposed initial eigenvectors from small to large to obtain a sorted eigenvector sequence;

[0164] Starting from the smallest initial feature vector in the sorted feature vector sequence, screening out a target feature vector with a preset number of features;

[0165] Clustering the target feature vector to obtain a plurality of target feature vector clusters;

[0166] The number of vectors in each target feature vector cluster is counted, and the target feature vector cluster corresponding to the largest number of vectors is used as case sharing test information.

[0167] In one embodiment, when executing the construction of the interactive information network of the target test group based on the case-sharing test information, the interactive network construction module 101 is configured to:

[0168] Taking each subject in the target trial population as a network node;

[0169] Calculating the vector similarity between every two target feature vectors in the case sharing test information;

[0170] Determining whether the vector similarity is greater than a preset similarity threshold;

[0171] If the vector similarity is greater than the similarity threshold, a network edge is established between the two target feature vectors corresponding to the vector similarity greater than the similarity threshold;

[0172] If the vector similarity is less than or equal to the similarity threshold, no network edge is established between the two target feature vectors corresponding to the vector similarity less than or equal to the similarity threshold;

[0173] An interactive information network of the target test group is constructed using the network nodes and the network edges.

[0174] In one embodiment, when performing node graph convolution propagation on the interactive information network to obtain a structural information embedding vector, the interactive network convolution module 102 is configured to:

[0175] Obtaining an adjacency matrix of the interactive information network and a node feature matrix of each network node in the interactive information network;

[0176] Constructing a node degree matrix according to the adjacency matrix and the node feature matrix;

[0177] Obtaining a weight matrix for each network node, and updating each network node in the interactive information network layer by layer using the weight matrix, the adjacency matrix, and the node degree matrix to obtain a final updated node in the final layer;

[0178] Each of the final updated nodes is embedded as a structural information vector.

[0179] In one embodiment, when identifying the structural outlier centers of the vaccine clinical trial data based on the structural information embedding vector, the outlier center identification module 103 is configured to:

[0180] Calculating the Euclidean distance between each of the structural information embedding vectors;

[0181] Sort the neighbor vectors of each structural information embedding vector in ascending order according to the Euclidean distance to obtain the number of sorting bits of each neighbor vector;

[0182] The neighbor vector having the number of sorted bits being the preset number of neighbors is used as the target neighbor vector;

[0183] Taking the Euclidean distance between the structural information embedding vector and the target adjacent vector as the target distance;

[0184] Calculating the reachable distance from each of the structural information embedding vectors to the neighbor vectors according to the target distance;

[0185] Calculating a local reachable density of each of the structural information embedding vectors according to the reachable distance, and generating a local outlier factor value using the local reachable density;

[0186] Determining whether the local outlier factor value is greater than a preset outlier threshold;

[0187] If the local outlier factor value is less than or equal to the outlier threshold, the structural information embedding vector corresponding to the local outlier factor value less than or equal to the outlier threshold is not a structural outlier center;

[0188] If the local outlier factor value is greater than the outlier threshold, the structural information embedding vector corresponding to the local outlier factor value greater than the outlier threshold is used as the structural outlier center.

[0189] In one embodiment, when performing an abnormality type hierarchical analysis on the structural outlier center to obtain a quality deviation classification result, the outlier center identification module 103 is configured to:

[0190] Extracting structural features and behavioral features of the structural outlier center;

[0191] Constructing a triggerable abnormal behavior rule table based on the structural features and the behavioral features;

[0192] Mapping each of the structural outlier centers into a rule activation vector according to the triggerable abnormal behavior rule table;

[0193] Calculating the overlap of each of the rule activation vectors, and dividing the rule activation vectors whose overlap is greater than a preset overlap threshold into a group of target activation vector clusters;

[0194] Identifying the activation rule set corresponding to the target activation vector cluster in the triggerable abnormal behavior rule table;

[0195] Analyzing the abnormality type of the target activation vector cluster according to the activation rule set to obtain an abnormality type classification;

[0196] A quality deviation classification result of each of the structural outlier centers is generated according to the anomaly type classification.

[0197] In one embodiment, when executing the process of obtaining the abnormal propagation path of the structural outlier center and identifying the local structural abnormal cluster according to the abnormal propagation path, the abnormal cluster identification module 104 is configured to:

[0198] Obtaining propagation weights, and mapping the propagation weights to the interactive information network to obtain an updated interactive network;

[0199] Performing diffusion path search on the structural outlier center in the updated interaction network, and taking the path whose propagation weight is less than a preset weight threshold during the path search as an abnormal propagation path;

[0200] Calculating the embedding space deviation degree and data integrity anomaly degree of each network node on the abnormal propagation path;

[0201] Screening out high-risk nodes on the abnormal propagation path according to the embedding space deviation degree and the data integrity abnormality degree;

[0202] Each of the structural outlier centers and the corresponding high-risk nodes on the anomaly propagation path are aggregated into a local structural anomaly cluster.

[0203] In one embodiment, when generating a local quality control strategy based on the quality deviation classification result and the local structural abnormality cluster, the control strategy generation module 105 is configured to:

[0204] Obtaining the abnormality type classification in the quality deviation classification result, and counting the composition ratio of the abnormality type classification in the local structural abnormality cluster;

[0205] identifying distribution characteristics of the local structural anomaly clusters in the updated interaction network;

[0206] Performing feature analysis on the propagation path sensitivity of the abnormal propagation path, and generating a propagation pattern of the local structural abnormal cluster based on the sensitivity features obtained from the analysis;

[0207] generating a local anomaly pattern according to the composition ratio, the distribution characteristics, and the propagation pattern;

[0208] generating a control strategy template and a control target according to the local abnormality pattern;

[0209] The control strategy corresponding to each of the local structural abnormality clusters is selected according to the control strategy template and the control target to obtain a local quality control strategy.

[0210] In one embodiment, when performing feature analysis on the propagation path sensitivity of the abnormal propagation path and generating the propagation pattern of the local structural abnormality cluster according to the sensitivity features obtained by the analysis, the control strategy generation module 105 is configured to:

[0211] Constructing an abnormal cluster intra-propagation path graph according to the abnormal propagation path;

[0212] Perturbing the propagation path graph within the abnormal cluster to obtain an initial node length and a length after path propagation of each propagation node in the propagation path graph within the abnormal cluster after the disturbance;

[0213] Obtaining a path length growth rate based on a quotient of the initial node length and the path length after propagation;

[0214] Obtaining the number of activated abnormal propagation nodes in the propagation path graph within the abnormal cluster after the disturbance, and generating a node propagation gain according to the number of abnormal propagation nodes;

[0215] generating a sensitivity feature of the abnormal propagation path according to the path length growth rate and the node propagation gain;

[0216] Acquire a propagation mode type, match the propagation mode type according to the sensitivity feature, and obtain a propagation mode of a local structural anomaly cluster corresponding to the abnormal propagation path.

[0217] In the present invention, a method for optimizing the quality management system of vaccine clinical trials is provided. By obtaining the vaccine clinical trial data of a target trial group, case-sharing trial information of the vaccine clinical trial data is extracted, and an interactive information network of the target trial group is constructed based on the case-sharing trial information. Each subject in the target trial group is used as a network node. Combined with the interactive information network constructed based on the case-sharing trial information, the correlation and mutual influence between subjects on key trial features can be intuitively and systematically revealed. Node graph convolution propagation is performed on the interactive information network to obtain a structural information embedding vector. A node degree matrix is ​​constructed through the adjacency matrix and the node feature matrix. In combination with the node weight matrix, layer-by-layer aggregation and propagation of information between nodes are realized, so that nodes gradually obtain richer contextual semantics and structural perception capabilities in each layer. The structural outlier center of the vaccine clinical trial data is identified according to the structural information embedding vector, and the structural outlier is analyzed. The center conducts hierarchical analysis of abnormal types to obtain quality deviation classification results, accurately identifies structural outlier centers in vaccine clinical trial data, effectively reveals potential abnormal individuals or abnormal patterns, and combines structural characteristics with behavioral characteristics to construct abnormal behavior rule tables and rule activation vector cluster analysis to achieve detailed hierarchical classification of abnormal types, obtain the abnormal propagation path of the structural outlier center, and identify local structural abnormal clusters based on the abnormal propagation path. Combining propagation weight mapping and path search, it effectively captures weak connections and abnormal diffusion of information in the network, and then through multi-dimensional evaluation of embedded spatial deviation and data integrity anomalies, accurately screens out high-risk nodes to form a focused and representative abnormal cluster structure. Based on the quality deviation classification results and the local structural abnormal cluster, a local quality control strategy is generated, and the local quality control strategy is used to optimize the vaccine clinical trial quality management system, which can achieve a transition from macro management to refined and structured management and control. By accurately formulating and matching control strategy templates and targets based on local anomaly patterns generated by quality deviation classification, network distribution characteristics and propagation patterns, and effectively identifying and intervening in abnormal clusters with specific structural risks and propagation behaviors, targeted regulation of high-risk nodes can be achieved, abnormal spread can be reduced, and data integrity and test consistency can be improved, thereby enhancing the real-time, sensitivity and stability of quality management, and achieving closed-loop control from risk discovery, feature identification to strategy execution. The present invention can solve the problem of being difficult to effectively capture structural outlier centers in vaccine clinical trial data and difficult to construct an interactive graph structure for test data in vaccine clinical trial data that lacks a natural interactive network. The specific definition of a vaccine clinical trial quality management system optimization system can be found in the definition of a vaccine clinical trial quality management system optimization method above, which will not be repeated here. The various modules in the above-mentioned vaccine clinical trial quality management system optimization system can be implemented in whole or in part by software, hardware and a combination thereof.The above modules may be embedded in or independent of the processor in the computer device in the form of hardware, or may be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0218] In the embodiments provided by the present invention, it should be understood that the disclosed system can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the module division is only a logical function division, and other division methods may be used in actual implementation.

[0219] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0220] Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the claims are intended to be embraced therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0221] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0222] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0223] Those skilled in the art will clearly understand that for the sake of convenience and brevity in description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0224] In the embodiments provided in the present disclosure, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the systems, methods, and computer program products according to the various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment, or a portion of code, and the above-mentioned module, program segment, or a portion of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0225] It should be noted that, in this disclosure, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element limited by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0226] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A method for optimizing the quality management system of vaccine clinical trials, characterized in that: The method comprises: Acquiring vaccine clinical trial data of a target trial population, extracting case-sharing trial information of the vaccine clinical trial data, and constructing an interactive information network of the target trial population based on the case-sharing trial information; Performing node graph convolution propagation on the interactive information network to obtain a structural information embedding vector; Calculate the Euclidean distance between each of the structural information embedding vectors, sort the neighbor vectors of each of the structural information embedding vectors in ascending order according to the Euclidean distance, obtain the number of sorting bits of each of the neighbor vectors, use the neighbor vector whose sorting bit is the preset number of neighbors as the target neighbor vector, use the Euclidean distance between the structural information embedding vector and the target neighbor vector as the target distance, calculate the reachable distance from each of the structural information embedding vectors to the neighbor vector according to the target distance, calculate the local reachable density of each of the structural information embedding vectors according to the reachable distance, and use the local reachable density to generate a local outlier factor value, determine whether the local outlier factor value is greater than a preset outlier threshold, and if the local outlier factor value is less than or equal to the outlier threshold, then the structural information embedding vector corresponding to the local outlier factor value less than or equal to the outlier threshold is not structural Outlier center, if the local outlier factor value is greater than the outlier threshold, then the structural information embedding vector corresponding to the local outlier factor value greater than the outlier threshold is used as the structural outlier center, the structural features and behavioral features of the structural outlier center are extracted, and a triggerable abnormal behavior rule table is constructed based on the structural features and the behavioral features. According to the triggerable abnormal behavior rule table, each of the structural outlier centers is mapped to a rule activation vector, the overlap of each rule activation vector is calculated, and the rule activation vectors with an overlap greater than a preset overlap threshold are divided into a group of target activation vector clusters, the activation rule set corresponding to the target activation vector cluster is identified in the triggerable abnormal behavior rule table, the abnormal type of the target activation vector cluster is analyzed according to the activation rule set, and an abnormal type classification is obtained, and a quality deviation classification result of each structural outlier center is generated according to the abnormal type classification; Acquire a propagation weight, map the propagation weight to the interactive information network to obtain an updated interactive network, perform a diffusion path search on the structural outlier center in the updated interactive network, and use the path whose propagation weight is less than a preset weight threshold during the path search as an abnormal propagation path, calculate the embedding space deviation degree and data integrity abnormality degree of each network node on the abnormal propagation path, screen out high-risk nodes on the abnormal propagation path based on the embedding space deviation degree and the data integrity abnormality degree, and aggregate each structural outlier center and the corresponding high-risk node on the abnormal propagation path into a local structural abnormal cluster; A local quality control strategy is generated based on the quality deviation classification results and the local structural abnormality clusters, and the local quality control strategy is used to optimize the vaccine clinical trial quality management system.

2. The method for optimizing the quality management system of vaccine clinical trials according to claim 1, wherein: The extraction of case sharing trial information of the vaccine clinical trial data includes: Performing feature alignment on the vaccine clinical trial data of each subject in the target trial population to obtain aligned trial data; Performing unified processing on the time axis of the alignment test data to obtain standard test data; extracting key test features of the standard test data, and normalizing the key test features to obtain normalized test features; Randomly selecting two of the normalized test features as two target test features; Calculating the feature similarity between every two target test features, and constructing a feature matrix based on the feature similarities; Constructing a Laplace matrix according to the feature similarity and the feature matrix; Performing eigendecomposition on the Laplace matrix, and sorting the decomposed initial eigenvectors from small to large to obtain a sorted eigenvector sequence; Starting from the smallest initial feature vector in the sorted feature vector sequence, screening out a target feature vector with a preset number of features; Clustering the target feature vector to obtain a plurality of target feature vector clusters; The number of vectors in each target feature vector cluster is counted, and the target feature vector cluster corresponding to the largest number of vectors is used as case sharing test information.

3. The method for optimizing the quality management system of vaccine clinical trials according to claim 2, wherein: The constructing of the target test group's interactive information network based on the case-sharing test information includes: Taking each subject in the target trial population as a network node; Calculating the vector similarity between every two target feature vectors in the case sharing test information; Determining whether the vector similarity is greater than a preset similarity threshold; If the vector similarity is greater than the similarity threshold, a network edge is established between the two target feature vectors corresponding to the vector similarity greater than the similarity threshold; If the vector similarity is less than or equal to the similarity threshold, no network edge is established between the two target feature vectors corresponding to the vector similarity less than or equal to the similarity threshold; An interactive information network of the target test group is constructed using the network nodes and the network edges.

4. The method for optimizing the quality management system for vaccine clinical trials according to claim 1, wherein: The performing node graph convolution propagation on the interactive information network to obtain a structural information embedding vector includes: Obtaining an adjacency matrix of the interactive information network and a node feature matrix of each network node in the interactive information network; Constructing a node degree matrix according to the adjacency matrix and the node feature matrix; Obtaining a weight matrix for each network node, and updating each network node in the interactive information network layer by layer using the weight matrix, the adjacency matrix, and the node degree matrix to obtain a final updated node in the final layer; Each of the final updated nodes is embedded as a structural information vector.

5. The method for optimizing the quality management system for clinical trials of vaccines according to claim 1, wherein: Generating a local quality control strategy according to the quality deviation classification result and the local structural abnormality cluster includes: Obtaining the abnormality type classification in the quality deviation classification result, and counting the composition ratio of the abnormality type classification in the local structural abnormality cluster; identifying distribution characteristics of the local structural anomaly clusters in the updated interaction network; Performing feature analysis on the propagation path sensitivity of the abnormal propagation path, and generating a propagation pattern of the local structural abnormal cluster based on the sensitivity features obtained from the analysis; generating a local anomaly pattern according to the composition ratio, the distribution characteristics, and the propagation pattern; generating a control strategy template and a control target according to the local abnormality pattern; The control strategy corresponding to each of the local structural abnormality clusters is selected according to the control strategy template and the control target to obtain a local quality control strategy.

6. The method for optimizing the quality management system for vaccine clinical trials according to claim 5, wherein: The performing feature analysis on the propagation path sensitivity of the abnormal propagation path and generating the propagation pattern of the local structural abnormal cluster according to the sensitivity features obtained by the analysis includes: Constructing an abnormal cluster intra-propagation path graph according to the abnormal propagation path; Perturbing the propagation path graph within the abnormal cluster to obtain an initial node length and a length after path propagation of each propagation node in the propagation path graph within the abnormal cluster after the disturbance; Obtaining a path length growth rate based on a quotient of the initial node length and the path length after propagation; Obtaining the number of activated abnormal propagation nodes in the propagation path graph within the abnormal cluster after the disturbance, and generating a node propagation gain according to the number of abnormal propagation nodes; generating a sensitivity feature of the abnormal propagation path according to the path length growth rate and the node propagation gain; Acquire a propagation mode type, match the propagation mode type according to the sensitivity feature, and obtain a propagation mode of a local structural anomaly cluster corresponding to the abnormal propagation path.

7. A vaccine clinical trial quality management system optimization system, characterized by A method for optimizing a vaccine clinical trial quality management system according to any one of claims 1 to 6, wherein the system comprises: An interactive network construction module is used to obtain vaccine clinical trial data of a target trial population, extract case-sharing trial information of the vaccine clinical trial data, and construct an interactive information network of the target trial population based on the case-sharing trial information; An interactive network convolution module, configured to perform node graph convolution propagation on the interactive information network to obtain a structural information embedding vector; an outlier center identification module, configured to identify the structural outlier centers of the vaccine clinical trial data based on the structural information embedding vector, and perform an abnormality type hierarchical analysis on the structural outlier centers to obtain a quality deviation classification result; an abnormal cluster identification module, configured to obtain an abnormal propagation path of the structural outlier center and identify a local structural abnormal cluster based on the abnormal propagation path; A control strategy generation module is used to generate a local quality control strategy based on the quality deviation classification results and the local structural abnormality cluster, and use the local quality control strategy to optimize the vaccine clinical trial quality management system.

Images