Progress tracking and early warning method and system for clinical trial projects

Abstract

The present invention relates to a progress tracking and early warning method and system for clinical trial projects. The method: collects and scores the subject recruitment progress data, follow-up completion data, test process data and project milestone completion data of the clinical trial center, obtains a data quality scoring matrix and data node connection status; performs network mapping to obtain a project progress dependency graph; performs spatial feature and temporal feature extraction to obtain a risk feature vector; performs dual-target dynamic evaluation to obtain a risk assessment report; dynamically adjusts early warning indicators to obtain an early warning rule configuration file; performs early warning trigger judgment, and performs hierarchical push processing to generate early warning information push records. The implementation of the present invention reduces system resource consumption and improves the processing capacity of large-scale multi-center clinical trial projects.

Description

Technical Field

[0001] The present invention relates to the technical field of project early warning, and in particular to a progress tracking and early warning method and system for a clinical trial project. Background Art

[0002] The traditional milestone-based progress tracking method is difficult to meet the project management requirements in terms of accuracy and timeliness when faced with distributed data collection, multi-center collaboration, and real-time monitoring. In particular, it becomes more difficult to accurately grasp the project progress status when data quality varies and the test centers are widely distributed.

[0003] At present, the progress management of clinical trial projects mainly relies on centralized data processing methods, which face the problem of excessive consumption of storage and computing resources when processing large-scale multi-center clinical trial data. At the same time, due to the lack of an effective early warning mechanism, project managers are often unable to detect potential risks in a timely manner, resulting in frequent project delays and waste of resources. Summary of the invention

[0004] The main purpose of the present invention is to provide a progress tracking and early warning method and system for clinical trial projects, which reduces system resource consumption and improves the processing capacity of large-scale multi-center clinical trial projects.

[0005] To achieve the above object, the present invention provides a progress tracking and early warning method for a clinical trial project, comprising the following steps:

[0006] Collect and score the quality of subject recruitment progress data, follow-up completion data, trial process data, and project milestone completion data of the clinical trial center to obtain the data quality score matrix and data node connection status;

[0007] Inputting the data quality scoring matrix and the data node connection status into the progress network construction model for network mapping to obtain a project progress dependency graph;

[0008] Extracting spatial features and temporal features from the project schedule dependency graph to obtain a risk feature vector;

[0009] Performing a dual-objective dynamic assessment on the risk feature vector to obtain a risk assessment report;

[0010] According to the risk assessment report, the early warning indicators are dynamically adjusted to obtain an early warning rule configuration file;

[0011] Perform warning trigger judgment on the risk assessment report and the warning rule configuration file, perform hierarchical push processing, and generate a warning information push record.

[0012] The present invention also provides a progress tracking and early warning system for a clinical trial project, comprising:

[0013] The collection module is used to collect and score the quality of the subject recruitment progress data, follow-up completion data, trial process data and project milestone completion data of the clinical trial center, and obtain the data quality score matrix and data node connection status;

[0014] A mapping module, used for inputting the data quality scoring matrix and the data node connection status into the progress network construction model for network mapping to obtain a project progress dependency graph;

[0015] An extraction module, used to extract spatial features and temporal features from the project schedule dependency graph to obtain a risk feature vector;

[0016] An evaluation module, used for performing a dual-objective dynamic evaluation on the risk feature vector to obtain a risk evaluation report;

[0017] An adjustment module, used to dynamically adjust the early warning indicators according to the risk assessment report to obtain an early warning rule configuration file;

[0018] The generation module is used to perform warning trigger judgment on the risk assessment report and the warning rule configuration file, and perform hierarchical push processing to generate a warning information push record.

[0019] In summary, the technical solution provided by the present invention solves the problem of inconsistent data collection quality in multi-center clinical trials by constructing a data quality scoring system and a data node connection mechanism, and adopts a time graph convolutional neural network for feature extraction, which effectively captures the spatial correlation and temporal dependency between the test centers and improves the accuracy of risk identification; introduces a Markov decision process framework and a mixed integer linear programming model to achieve dual-objective dynamic optimization of progress achievement rate and resource utilization rate, and reduces the risk of project delay; establishes a dynamic early warning indicator system by introducing algorithms and sensitivity analysis methods, so that early warning rules can be adaptively adjusted according to actual conditions; designs a hierarchical early warning push mechanism, combined with multi-level cache and heartbeat detection technology, to ensure the timely delivery and processing tracking of early warning information; adopts a distributed computing framework for data processing, reduces system resource consumption, and improves the processing capacity of large-scale multi-center clinical trial projects; effectively reduces data redundancy and optimizes system performance through network structure modeling and feature dimensionality reduction processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a schematic diagram of the steps of a method for tracking and warning the progress of a clinical trial project in one embodiment of the present invention;

[0021] Figure 2It is a structural block diagram of a progress tracking and early warning system for a clinical trial project in one embodiment of the present invention.

[0022] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0024] Reference Figure 1 This embodiment provides a progress tracking and early warning method for a clinical trial project, comprising the following steps:

[0025] S1, collect and score the quality of the subject recruitment progress data, follow-up completion data, trial process data and project milestone completion data of the clinical trial center, and obtain the data quality score matrix and data node connection status;

[0026] Among them, the subject recruitment progress data is calculated, focusing on the enrollment rate and screening failure rate. These indicators can reflect the speed of recruitment and the qualified rate of subject screening. Through these data, it is judged whether the subject recruitment is progressing as expected, which in turn affects the overall progress of the project. The follow-up completion data is counted, focusing on the timely follow-up rate and the dropout rate. The timely follow-up rate reflects whether the subjects complete the follow-up on time during the trial, while the dropout rate reveals the situation of the subjects withdrawing midway during the trial, which directly affects the integrity and quality of the trial data. At the same time, the trial process data is analyzed, focusing on the incidence of adverse events and protocol compliance. The incidence of adverse events reflects the frequency of adverse reactions encountered by the subjects during the trial, while protocol compliance reflects whether the subjects strictly abide by the experimental protocol during the trial. The project milestone completion data is calculated, including the planned completion rate and the extension rate. These two indicators respectively reflect whether the milestones of the project are completed on time and whether there are extensions, thereby affecting the overall progress of the trial and the realization of project goals. By calculating these raw data indicators, the raw data indicator set is obtained. Based on the original data indicator set, an evaluation standard system is established, and each data is quantitatively scored and weighted according to this standard system to obtain the data quality score of each clinical trial center. The evaluation standard system will set weights for the importance of each indicator so that the relative importance of different data can be reflected in the scoring. The frequency of data interaction and the timeliness of information transmission between clinical trial centers are statistically analyzed to obtain a data interaction intensity matrix, which reflects the frequency of data exchange and the speed of information transmission between centers, and then reveals which centers have high data sharing and communication efficiency and which centers have the risk of untimely information transmission. The data quality scores of each clinical trial center are fused with the data interaction intensity matrix to obtain a comprehensive data quality score matrix, which reflects the operating status of the entire trial project. The data quality score matrix is ​​normalized. Data of different dimensions and scales are converted into a unified standard to facilitate subsequent comparison and analysis. The normalized scores are used to classify and divide the connection relationships between clinical trial centers, which are usually divided into three levels: high priority, medium priority and low priority. High-priority connection relationships mean that these centers are more important in terms of data quality or information interaction, and special attention should be paid to their data flows and risk points. In contrast, low-priority connections mean that these centers have less impact on the overall project, but their data transmission still needs to be timely and accurate. Through division, the priority of each connection relationship is determined. Path optimization and dynamic monitoring are performed on the initial node connection status. The purpose of path optimization is to ensure the efficiency of data flow and the selection of the optimal path, while dynamic monitoring can track the connection status between centers in real time, discover potential risks and bottlenecks, take corrective measures in time, and ensure that the project proceeds as planned.By continuously optimizing and adjusting the node connection status, the overall execution and data quality of clinical trials can be effectively improved.

[0027] S2, input the data quality score matrix and data node connection status into the progress network construction model for network mapping to obtain the project progress dependency graph;

[0028] Specifically, the score values ​​in the data quality score matrix are normalized and mapped, and the score values ​​of different dimensions and scales are unified into the same standard range. The normalized data reflects the relative data quality of each clinical trial center, and the node weight matrix of the clinical trial center is obtained. Different weight values ​​are assigned to the connections of different priorities in the data node connection status. According to the importance of the connection, the high priority connection is given a weight of 0.9, the medium priority connection is given a weight of 0.6, and the low priority connection is given a weight of 0.3, and the weight matrix of the connection edge is obtained. The clinical trial center node weight matrix and the connection edge weight matrix are input into the ClinicalDispatch calculation framework. The framework calculates the degree centrality, betweenness centrality and closeness centrality of each clinical trial center based on the node weight and edge weight. These three centrality indicators can comprehensively evaluate the importance and influence of each node in the entire project network. Degree centrality reflects the number of direct connections between each node and other nodes, betweenness centrality indicates the ability of a node to act as a bridge in the network, and closeness centrality indicates the average distance between a node and other nodes. After calculating these three indicators, a set of node centrality indicators is obtained, which is used for the comprehensive evaluation of each node. The weighted summation method is used to comprehensively score the node centrality index set. Considering the weights of each centrality index, they are integrated into a single comprehensive score, which can reflect the relative importance and risk of each clinical trial center in the project as a whole. Based on the comprehensive score of the node, a network adjacency matrix is ​​constructed, which shows the connection relationship and strength between nodes. By analyzing the score difference, the connection strength between nodes is determined according to the size of the score difference. If the comprehensive score difference between two nodes is less than 0.2, it is considered that the connection relationship between the two nodes is strong, and the system will establish a strong connection for them in the network; if the score difference is between 0.2 and 0.5, a weak connection relationship is established, and nodes with a score difference greater than 0.5 are not connected, and a preliminary network structure is obtained. The initial network structure is input into the Latin hypercube sampling model for feature dimensionality reduction. Latin hypercube sampling is an efficient multivariate analysis method that reduces redundant information by dimensionality reduction, making the final network structure more concise and representative, and obtaining the target network structure. The progress indicator is calculated for each clinical trial center node in the target network structure. These progress indicators reflect the progress of each center in the project, including the progress of subject recruitment, the progress of the trial process, the completion of follow-up, etc. The progress status of each node is converted into a progress state vector, which shows the specific progress status of each clinical trial center in the project. The progress state vector is mapped to the target network structure to form the final project progress dependency graph, which not only shows the dependency between the clinical trial centers, but also helps the system understand the overall progress of the project, identify potential progress delay nodes, and predict possible future progress risks.

[0029] S3, extract spatial features and temporal features from the project schedule dependency graph to obtain a risk feature vector;

[0030] It should be noted that the project schedule dependency graph is input into the input layer of the spatial convolutional network, which can extract spatial features from image or graph structure data. Three convolution kernels of different sizes are set, namely 3×3, 5×5 and 7×7. These convolution kernels can capture information of different scales when extracting spatial features. Through the convolution operation, a multi-scale spatial feature matrix is ​​obtained, which shows the spatial relationship and dependency pattern between each node in the project. The multi-scale spatial feature matrix is ​​subjected to the maximum pooling operation. The window size of the pooling operation is 2×2, and the step size is set to 2. The complexity of the data is reduced by dimension reduction, while the most important spatial information is retained to obtain the reduced dimension spatial features. The reduced dimension spatial features are reorganized according to the time series to form a time series data sequence. The project schedule has not only spatial dependencies, but also obvious time dependencies. The tasks or activities in the experiment usually have a time order and a before-after relationship. The spatial features are arranged according to the time dimension. The time series data is input into the bidirectional gated recurrent unit (Bi-GRU) for processing. The bidirectional gated recurrent unit is a special recurrent neural network that can process the forward and reverse information in the sequence at the same time and extract important features in the time series. Through the bidirectional GRU model, two feature sequences are obtained: the forward feature sequence and the reverse feature sequence. The forward sequence captures the time progression from the starting point to the current node, while the reverse sequence captures the time backward information from the end point to the current node. Bidirectional processing helps to understand the temporal dynamics in the project schedule more comprehensively and make up for the shortcomings of unidirectional sequence analysis. The forward feature sequence and the reverse feature sequence are concatenated to merge the two sequences into a complete temporal feature vector, which reflects the temporal evolution in the project schedule. The temporal feature vector and the spatial feature vector are input into the attention layer for weighted calculation. The attention mechanism can automatically assign weights according to the importance of different features to emphasize the information that is more critical to risk prediction. The weight of the spatial feature is set to 0.6, and the weight of the temporal feature is set to 0.4, balancing the contribution of spatial information and temporal information to risk assessment. The weighted features calculated by the attention mechanism help highlight the key risk points in the project schedule. The weighted features are standardized to ensure that the eigenvalues ​​are within the same dimensional range to avoid affecting subsequent processing due to large differences in eigenvalues. The standardized feature matrix is ​​input into the fully connected layer for nonlinear transformation and activated using the ReLU activation function to increase the nonlinear ability of the model, and the final risk feature vector is obtained, which contains risk information at the spatial and temporal levels of the project schedule.

[0031] S4, perform dual-objective dynamic evaluation on the risk feature vector to obtain a risk assessment report;

[0032] Specifically, the risk feature vector is mapped into a state space, and the risk information of the project is mapped into a more structured state vector. The state vector includes three key indicators: progress completion rate, resource utilization rate and risk level. These indicators can fully reflect the progress of the project, resource allocation efficiency and potential risk level, and constitute the basic data for assessing project risks. Based on the state vector, a Markov state transition model is constructed to describe the possible state changes of the project at different time points, and thus the initial state space is obtained. The Markov model helps the system understand the evolution path of the project progress at different stages by analyzing the transition probability between each state. The initial state space is divided into different subspaces according to the risk level. The risk level is divided into high-risk subspace, medium-risk subspace and low-risk subspace to identify risks of different degrees in more detail. In each subspace, the probability of each state transition is calculated according to the actual situation to obtain a state transition matrix, which reflects the possibility of the project transitioning from one state to another, and helps the system understand the changing trend of the project progress under different risk levels. Dynamic response processing and state transition probability update are performed based on the state transition matrix to obtain a dynamic state transition matrix. The dynamic matrix can reflect the latest situation of risk level and state transfer according to the real-time changes of project progress, so as to better guide subsequent decision-making. Based on the dynamic state transfer matrix, a mixed integer programming objective function is established, which contains two main objectives: progress achievement rate and resource utilization rate. The progress achievement rate objective function aims to ensure that the project completes various tasks on time, while the resource utilization rate objective function helps optimize the efficiency of resource use. They are given weight coefficients of 0.6 and 0.4 respectively to obtain the weighted objective function. Constraints are set, including project budget, time limit, resource limit, etc., to ensure that the actual operation requirements can be met during the optimization process. After the weighted objective function and constraint set are input into the linear programming solver, the branch and bound method is used to solve the problem to obtain the optimal decision sequence. This process can help the clinical trial team find the optimal risk control plan under different risk scenarios and ensure that the project is smoothly promoted under controllable risks. Risk quantitative assessment is performed on each decision node, and the product of the risk probability and the impact degree is calculated to obtain the risk assessment score. By analyzing the assessment score, the risk is graded according to the score. Risks with high scores represent critical issues in the project and need to be dealt with first, while risks with low scores can be tolerated to a certain extent. A risk assessment report is generated based on the risk grading results.

[0033] S5, dynamically adjust the early warning indicators according to the risk assessment report to obtain the early warning rule configuration file;

[0034] Among them, data extraction is performed on the progress delay index, resource overrun index, quality abnormality index and risk accumulation index in the risk assessment report, and a preliminary early warning index matrix is ​​established in combination with the historical data of the clinical trial project. The matrix can cover multiple dimensions such as project progress, resource usage, quality control and overall risk level. The initial early warning indicator set is input into the sensitivity analysis model for processing, and the degree of influence of each indicator on the project risk, that is, the sensitivity coefficient, is calculated. Through calculation, it is evaluated which early warning indicators are most sensitive to changes in project risks and which have less impact. Based on the sensitivity assessment results, the early warning indicator priority sequence analysis is performed to obtain a hierarchical early warning indicator set. The hierarchical early warning indicator set is input into the launch algorithm model. The model can generate a set of test sample sets based on the input indicator set. Through the analysis of these samples, a preliminary early warning rule set is generated according to the dynamic combination of task allocation and early warning plan. The early warning effect of the initial early warning rule set is evaluated to obtain an early warning rule evaluation matrix, which reflects the effects of different early warning rules in actual operation, including their accuracy, response time and adaptability. The early warning rule evaluation matrix is ​​input into the recursive boundary search module, and the early warning threshold and trigger conditions are adjusted by iterative optimization. This process can continuously optimize the early warning rules to make them more accurate and efficient until the evaluation indicators of the early warning rules reach the preset goals. In this process, the threshold of the rules is continuously adjusted according to the actual feedback to ensure that each triggering of the early warning can accurately reflect the risk situation of the project and take effective measures to deal with potential problems in a timely manner. The optimized early warning rules obtain the target early warning rule set. On this basis, the rule set is divided into risk levels, and the corresponding early warning configuration plan is constructed according to different risk levels. The early warning configuration plan is integrated into a configuration file in a standard format to facilitate subsequent application and management in different systems or scenarios, ensuring that the early warning mechanism can run stably in different environments.

[0035] S6, make warning trigger judgment on the risk assessment report and warning rule configuration file, perform hierarchical push processing, and generate warning information push records.

[0036] Specifically, the risk assessment score in the risk assessment report and the warning threshold in the warning rule configuration file are monitored to determine whether the warning is triggered. The relationship between the risk assessment score and the warning threshold is monitored in real time to obtain the warning trigger state matrix, which reflects whether the warning trigger conditions are met in different risk scenarios, and helps the system determine whether there are risks that need to be dealt with immediately. Through these monitoring, potential risk events are captured in time. Based on the warning trigger state matrix, graded warning processing is carried out to obtain the response strategy matrix. According to different warning trigger states and risk levels, corresponding response strategies are formulated. Different warning levels require different response strategies. For example, for high-risk situations, emergency plans need to be activated immediately; for low-risk situations, continuous monitoring is selected and measures are taken at the appropriate time. The response strategy matrix helps the system clarify what actions should be taken in different warning scenarios, so as to ensure that the strategy can be flexibly adjusted according to the severity of the risk in actual operation. Based on the response strategy matrix, a warning information template is constructed. In the template, different types of risk information are sorted, including risk type, impact range, probability of occurrence, and disposal suggestions. According to different warning levels, this information is classified and sorted, so as to provide targeted processing suggestions for relevant personnel according to different risk levels. Match the recipients of the warning information and generate a list of recipients of the warning information. According to the different roles and responsibilities of the project, the warning information will be sent to the relevant personnel or departments. By matching the appropriate recipients, it is ensured that the relevant personnel in each link can receive the warning information that needs attention in a timely manner. The information content and the receiving list are input into the message distribution system, and the multi-level cache mechanism is used to push the information. This mechanism can ensure the stability and efficiency of the information during the transmission process, avoid information delay or loss due to network or system problems, and ensure that the warning information can be delivered to the relevant personnel on time. After the information is pushed, the information push status is monitored in real time, and the heartbeat detection mechanism is used to track the delivery of the information. Through the heartbeat detection, it is confirmed whether the warning information is successfully delivered to the intended recipient, and the time and confirmation status of the information reception are recorded. These feedback data help the system understand the potential problems in the information transmission process, such as whether the information is not delivered in time, so as to facilitate the adoption of measures for adjustment and optimization. After obtaining the push feedback data, these data are quantitatively analyzed to calculate key indicators such as delivery rate, response timeliness and processing progress under different warning levels. Through quantitative analysis, the overall effect of the early warning mechanism is evaluated, including whether the early warning information is delivered in a timely manner, whether the response is efficient, and whether the processing progress meets expectations. Generate early warning information push records, including the warning classification, push object, response status, and processing results.

[0037] In one example, the subject recruitment progress data, follow-up completion data, trial process data, and project milestone completion data of the clinical trial center are collected and scored for quality, and a data quality scoring matrix and data node connection status are obtained, including:

[0038] The enrollment rate and screening failure rate in the subject recruitment progress data were calculated, the timely follow-up rate and dropout rate in the follow-up completion data were counted, the adverse event rate and protocol compliance in the trial process data were analyzed, and the planned completion rate and extension rate in the project milestone completion data were calculated to obtain the original data indicator set;

[0039] An evaluation standard system is established based on the original data indicator set, and based on the evaluation standard system, the original data indicator set is quantitatively scored and weighted to obtain the data quality score of each clinical trial center;

[0040] Conduct statistical analysis on the frequency of data interaction and the timeliness of information transmission between clinical trial centers to obtain a data interaction intensity matrix, and perform data fusion processing on the data quality scores and data interaction intensity matrix of each clinical trial center to obtain a data quality score matrix;

[0041] The data quality score matrix is ​​normalized to obtain a normalized score. The connection relationship between clinical trial centers is divided into high priority, medium priority and low priority according to the normalized score, and the data transmission path of each priority is determined to obtain the initial node connection state;

[0042] The path optimization and dynamic monitoring of the initial node connection status are performed to obtain the data node connection status.

[0043] In this example, the enrollment rate and screening failure rate in the subject recruitment progress data are calculated, the timely follow-up rate and dropout rate in the follow-up completion data are counted, the adverse event rate and protocol compliance in the trial process data are analyzed, and the planned completion rate and delay rate in the project milestone completion data are calculated to obtain the original data indicator set. Based on the original data indicator set, an evaluation standard system is established. Each indicator is weighted according to its importance, impact on project progress, and risk level, and a weighted coefficient vector is used. , where each weight value represents the importance of the indicator to the overall score. Quantitative scoring is achieved by standardizing each indicator, and the score obtained after standardization is usually between 0 and 1. For example, for the enrollment rate, the standardized score is calculated using the following formula:

[0044] ;

[0045] This standardized method ensures that all indicators have the same dimension, which is convenient for weighted calculation. According to the evaluation standard system, the comprehensive data quality score of each clinical trial center is calculated by weight. The higher the score, the higher the data quality of the center, and vice versa. On this basis, the data interaction intensity is analyzed, and the data interaction frequency and information transmission timeliness between clinical trial centers are counted to obtain the data interaction intensity matrix. The elements of this matrix represent the interaction frequency and information transmission timeliness between different centers, which are calculated by the following formula:

[0046] ;

[0047] in, Indicates and The interaction strength between centers, and They represent the frequency of data exchange and the timeliness of information transmission between the two centers, respectively. By integrating these data, a data quality score matrix is ​​obtained, which integrates the data quality scores of each clinical trial center and the information of the data interaction intensity matrix. The data quality score matrix is ​​normalized to obtain a normalized score, which represents the overall performance of each center after comprehensively considering data quality and interaction intensity. The normalized score is calculated using the following formula:

[0048] ;

[0049] in, Centered Rating, and are the minimum and maximum scores of all centers, respectively. Based on the normalized scores, the connection relationships between clinical trial centers are divided into high priority, medium priority and low priority. The priority division is achieved by setting thresholds. For example, the high priority threshold is above 0.8, the medium priority is 0.5 to 0.8, and the low priority is below 0.5. Each node in the network (i.e., each center) is assigned a different priority according to its score, and the data transmission path between the priorities is determined. The initial node connection status is path optimized and dynamically monitored to ensure efficient data transmission between different centers. Path optimization is achieved through graph theory algorithms, such as using the shortest path algorithm or the maximum flow algorithm to optimize the efficiency of data transmission. At the same time, dynamic monitoring is carried out to adjust the connection status between nodes in real time to cope with network delays or data transmission anomalies, ensuring that the project can proceed smoothly.

[0050] In one example, the data quality score matrix and the data node connection status are input into the schedule network construction model for network mapping to obtain a project schedule dependency graph, including:

[0051] Normalize and map the score values ​​in the data quality score matrix to obtain the clinical trial center node weight matrix;

[0052] The high priority connection in the data node connection status is assigned a weight of 0.9, the medium priority connection is assigned a weight of 0.6, and the low priority connection is assigned a weight of 0.3, and the connection edge weight matrix is ​​obtained;

[0053] Input the clinical trial center node weight matrix and the connection edge weight matrix into the ClinicalDispatch calculation framework, calculate the degree centrality, betweenness centrality and closeness centrality of each clinical trial center, and obtain the node centrality index set;

[0054] The node centrality index set is comprehensively scored using the weighted summation method to obtain the node comprehensive score;

[0055] A network adjacency matrix is ​​constructed based on the comprehensive node scores. Strong connections are established for nodes with a score difference of less than 0.2, and weak connections are established for nodes with a score difference between 0.2 and 0.5, to obtain the initial network structure.

[0056] The initial network structure is input into the Latin hypercube sampling model for feature dimension reduction to obtain the target network structure;

[0057] The progress index is calculated for each clinical trial center node in the target network structure to obtain a progress state vector, and the progress state vector is mapped to the target network structure to obtain a project progress dependency graph.

[0058] In this example, the data quality score matrix is ​​normalized and mapped to obtain the weight matrix of the clinical trial center node. Assume that the data quality score matrix is , where each element Indicates Clinical trial centers in Each score is normalized to convert it to a standard range (usually between 0 and 1) for subsequent weighted calculation. The normalization formula is:

[0059] ;

[0060] in, represents the normalized score, and They are Through the normalization process, the scores of each clinical trial center are adjusted to a unified scale, making each indicator comparable and able to be reasonably weighted in subsequent calculations. The resulting normalized score matrix Used to calculate the node weight of each clinical trial center. Assume that each indicator has a different weight coefficient , and the final weight of each center is obtained by weighted calculation of its score on each indicator. The calculation formula of the node weight matrix W is:

[0061] ;

[0062] in, It is The node weight of each clinical trial center, It is The weight coefficient of each indicator is It is Clinical trial centers in The normalized scores on the indicators. The obtained node weight matrix is a column vector, representing the weighted scores of all clinical trial centers. When building network connections, the connection relationships between data nodes are assigned different weights according to their priorities. Suppose there is a matrix representing node connections , where the elements Indicates and The connection relationship between two clinical trial centers. According to the priority of the connection, the high priority connection is given a weight of 0.9, the medium priority connection is given a weight of 0.6, and the low priority connection is given a weight of 0.3. These weights are expressed as a connection edge weight matrix , and its calculation formula is:

[0063] ;

[0064] The node weights and edge weights are input into the ClinicalDispatch computational framework for network analysis. In this framework, three centrality indicators are calculated for each clinical trial center: degree centrality, betweenness centrality, and closeness centrality. These three centrality indicators represent the degree of connection of each node in the network, the frequency of information flow, and the proximity to other nodes. Degree centrality Represented as a node The number of connections is the number of edges that directly connect this node to other nodes. The formula is:

[0065] ;

[0066] in, Indicates and The edge weights between nodes. The higher the degree centrality, the more connections the node has with other nodes, and the higher its importance. It measures the role of the node as an intermediary in the network, indicating that The ratio of the number of shortest paths to the total number of all shortest paths. The calculation formula is:

[0067] ;

[0068] in, Represents a slave node To Node The number of shortest paths, Represents a slave node To Node The shortest path passes through the node The number of paths. Nodes with high betweenness centrality play an important intermediary role in information transmission in the network. It measures the average distance between a node and other nodes, and represents the inverse of the distance from the node to all other nodes in the network. The calculation formula is:

[0069] ;

[0070] in, Representation Node and nodes The shortest path distance between is the total number of nodes in the network. The higher the closeness centrality, the closer the node is to other nodes, and the more efficient the information dissemination is. By calculating these centrality indicators, we get the node centrality indicator set. In order to comprehensively consider each centrality indicator, we use the weighted summation method to score. Assume that the weights of degree centrality, betweenness centrality and closeness centrality are , the comprehensive score of the node Calculated by the following formula:

[0071] ;

[0072] in, Make sure that the total weight of the weighted sum is 1. In this way, the comprehensive score of each clinical trial center takes into account its degree of connectivity, its role in information flow, and its proximity to other nodes. Based on the comprehensive scores of the nodes, a network adjacency matrix is ​​constructed. In this process, it is determined which nodes are strongly connected, weakly connected, or not connected. If the difference in the comprehensive scores of two nodes is less than 0.2, a strong connection is established; if the difference is between 0.2 and 0.5, a weak connection is established; if the difference is greater than 0.5, no connection is established. It is expressed by the following rules:

[0073] ;

[0074] in, Representation Node and nodes The adjacency matrix value between . The initial network structure will be subjected to feature dimensionality reduction through the Latin hypercube sampling model to reduce the dimension and eliminate noise to obtain the target network structure. In the target network structure, the progress state vector is calculated according to the progress indicator of each clinical trial center. The progress state vector is constructed in the following way:

[0075] ;

[0076] in, Representation Node The progress completion rate, Representation Node Resource utilization, Representation Node The progress state vector is mapped to the target network structure to obtain the project progress dependency graph, thereby visualizing the progress of each node (clinical trial center) in the project and their mutual dependencies.

[0077] In one example, spatial features and temporal features are extracted from the project schedule dependency graph to obtain a risk feature vector, including:

[0078] The project schedule dependency graph is input into the input layer of the spatial convolutional network, and three convolution kernels are set to extract spatial features. The sizes of the convolution kernels are 3×3, 5×5, and 7×7, respectively, to obtain a multi-scale spatial feature matrix.

[0079] Perform the maximum pooling operation on the multi-scale spatial feature matrix, with a pooling window size of 2×2 and a step size of 2 to obtain the reduced-dimensional spatial features;

[0080] The reduced-dimensional spatial features are reorganized according to the time series to construct a time series data sequence, and the time series data sequence is input into the bidirectional gated recurrent unit to obtain a forward feature sequence and a reverse feature sequence;

[0081] The forward feature sequence and the reverse feature sequence are concatenated to obtain a complete temporal feature vector, and the spatial feature and temporal feature vectors are input into the attention layer to calculate the feature weight coefficients. The spatial feature weight is 0.6 and the temporal feature weight is 0.4 to obtain the weighted feature.

[0082] The weighted features are standardized to obtain a standardized feature matrix, which is then input into the fully connected layer and transformed nonlinearly using the ReLU activation function to obtain a risk feature vector.

[0083] In this example, the project schedule dependency graph is input into the input layer of the spatial convolutional network. The project schedule dependency graph is essentially a multidimensional matrix, where each node represents a clinical trial center, and the connections between nodes represent the dependencies between different nodes. In this graph, the spatial features reflect the clinical trial centers and the relationships between them, while the temporal features involve the progress changes between nodes over time. The input project schedule dependency graph is a two-dimensional matrix that represents the relationship between each clinical trial center. In order to extract the spatial features, this matrix is ​​convolved at the input layer of the convolutional network. Let the input matrix be ,in is the number of nodes, is the number of features of each node. In the spatial convolution operation, three convolution kernels (filters) are set, with sizes of 3×3, 5×5, and 7×7 respectively. These convolution kernels are used to extract spatial features of different scales, and multiple spatial feature matrices are obtained through convolution operations. The calculation formula for the convolution operation is:

[0084] ;

[0085] in, is the feature map after convolution, is the convolution kernel, is the size of the convolution kernel (for example, a 3×3 convolution kernel corresponds to , * indicates convolution operation, and is the index of the convolution kernel sliding in the input matrix. When performing convolution operations, the spatial features of each position in the input image are extracted through the convolution kernel. These features can reflect the association information between clinical trial centers. Through the convolution operation, spatial feature matrices of different scales are obtained, assuming that they are , where the size of each matrix is ​​slightly different from the input matrix, depending on the size of the convolution kernel. The multi-scale spatial feature matrix is ​​subjected to maximum pooling to achieve dimensionality reduction and extract the most important features. The pooling operation reduces the dimension of the feature matrix by selecting the maximum value in each pooling window and improves the computational efficiency and noise resistance of the model. The size of the pooling window is set to 2×2, the step size is 2, and the maximum pooling calculation formula is:

[0086] ;

[0087] in, is the feature matrix after pooling, is an element in the convolution feature matrix. Through this maximum pooling operation, the reduced-dimensional spatial feature matrix is ​​obtained, assuming that The spatial feature matrix after dimensionality reduction is reorganized into a time series to capture the trend of progress over time. In this step, the spatial features of each clinical trial center are organized into a time series data sequence in chronological order. Let the reorganized time series data sequence be ,in is the number of time steps, is the number of features at each moment. The time series data sequence is input into the Bidirectional Gated Recurrent Unit (Bi-GRU) model. Bi-GRU is a recurrent neural network structure used to process time series data. It improves the ability of feature extraction by considering both forward and reverse time series information. For each time step in the time series data sequence , Bi-GRU calculates the forward feature sequence and reverse feature sequence For the forward GRU calculation, the following formula is used:

[0088] ;

[0089] in, Indicates time The input features of is the hidden state of the previous moment. The calculation of the reverse GRU is similar, but it starts from the end of the time series and propagates information backward. After passing through the Bi-GRU model, the forward feature sequence and the reverse feature sequence are obtained. The two sequences are concatenated to obtain the complete time series feature vector The formula for the splicing operation is:

[0090] ;

[0091] The spatial feature and temporal feature vectors are input into an attention layer. The goal of the attention mechanism is to assign a weight coefficient to each feature in order to better emphasize the features that contribute to the final prediction result. The weight coefficient of the spatial feature is set to 0.6, and the weight coefficient of the temporal feature is set to 0.4. The formula for the weighting process is:

[0092] ;

[0093] In this formula, represents the spatial eigenvector, represents the time series feature vector, is the weighted feature vector. Through weighting, different importance is given to spatial features and temporal features, thereby improving the model's ability to process different types of information. The weighted feature vector is standardized to adjust the feature value to a uniform scale so that different features have similar ranges. The formula for the standardization operation is:

[0094] ;

[0095] in, and are the mean and standard deviation of the weighted eigenvectors, is the standardized feature matrix. The standardized feature matrix is ​​input into the fully connected layer and nonlinearly transformed using the ReLU activation function to extract features and generate the final risk feature vector. The calculation formula of the fully connected layer is:

[0096] ;

[0097] in, and are the weight matrix and bias term of the fully connected layer, respectively. is the normalized feature vector, It is the output feature after being processed by the ReLU activation function, representing the final risk feature vector.

[0098] In one example, a dual-objective dynamic assessment is performed on the risk feature vector to obtain a risk assessment report, including:

[0099] Map the risk feature vector to the state space, construct a state vector including the progress completion rate, resource utilization rate, and risk level, and build a Markov state transition model based on the state vector to obtain the initial state space;

[0100] The initial state space is divided into subspaces according to the risk level, and the risk level is divided into high-risk subspace, medium-risk subspace and low-risk subspace, and the state transition probability of each subspace is calculated to obtain the state transition matrix;

[0101] Based on the state transfer matrix, dynamic response processing and state transfer probability update are performed to obtain the dynamic state transfer matrix. Based on the dynamic state transfer matrix, the mixed integer programming objective function is established, the progress achievement rate objective function and the resource utilization objective function are set, and the weight coefficients are assigned to 0.6 and 0.4 respectively to obtain the weighted objective function;

[0102] Set constraints on the weighted objective function to obtain a constraint set, and input the weighted objective function and the constraint set into a linear programming solver, use the branch and bound method to solve the optimal decision sequence, and obtain a risk control plan;

[0103] Conduct a quantitative risk assessment on each decision node in the risk control plan, calculate the product of the risk probability and the degree of impact, obtain a risk assessment score, and grade the risks based on the risk assessment score to generate a risk assessment report.

[0104] In this example, the risk feature vector is mapped to the state space to capture the key dynamic characteristics of project operation related to schedule, resources and risks. Let the risk feature vector be ,in is the dimension of the feature. It is transformed into a state vector through mapping operation , the state vector consists of three key indicators: progress completion rate , Resource Utilization and risk level The calculation of the state vector is expressed as follows:

[0105] ;

[0106] Among them, the progress completion rate Indicates the completion ratio of the current task, calculated as:

[0107] ;

[0108] Resource Utilization Indicates the ratio of actual resource usage to allocated resources:

[0109] ;

[0110] Risk level It is a weighted score that takes into account various risk characteristics. The formula is:

[0111] ;

[0112] in, is the weight of each risk characteristic, For the The normalized value of the risk feature is constructed by , establish an initial state space that reflects the running status of the project. In order to capture the dynamic changes of the state over time, a Markov state transition model is constructed based on the state vector. Let the set of state space be ,in is the total number of possible states. In the Markov model, state transitions are represented by the transition probability matrix Description, its elements Indicates from the state Transfer to state Probability:

[0113] ;

[0114] Each state in the initial state space is divided into high-risk, medium-risk, and low-risk subspaces according to its risk level. Assume that the threshold of risk level is , the division rules are:

[0115] ;

[0116] By analyzing the states in these subspaces, the state transition probability of each subspace is calculated. State transition probability matrix The construction of is based on statistical observations or historical data analysis, such as the frequency of each state transitioning to other states within a certain time window. The state transition matrix is ​​dynamically responded to and updated probabilistically to reflect the impact of real-time data. For example, real-time monitoring data indicates that the transition probability of some states needs to be adjusted, generating a dynamic state transition matrix The update formula is:

[0117] ;

[0118] in, is the smoothing coefficient, It is the transition probability recalculated based on real-time data. On this basis, a mixed integer programming objective function is constructed to optimize the progress achievement rate and resource utilization rate while balancing the risk level. The objective function consists of two parts: progress achievement rate objective function and resource utilization objective function , respectively defined as:

[0119] ;

[0120] Weighted objective function for:

[0121] ;

[0122] Among them, the weight coefficients 0.6 and 0.4 represent the importance weights of progress and resource goals. In order to ensure that the optimization process meets actual needs, constraints are set. Assume that the total resource budget is , the total time is , the constraints are defined as:

[0123] ;

[0124] Input the weighted objective function and constraints into a linear programming solver, such as using the branch-and-bound method, to obtain the optimal decision sequence. The branch and bound method recursively decomposes the problem, gradually narrows the solution space, and finally finds the optimal solution that meets the constraints. After generating the risk control plan, a risk quantitative assessment is performed on each decision node. The key to risk assessment is to calculate the risk score of each node. , the formula is:

[0125] ;

[0126] in, is the probability of risk occurrence, The risk level is the impact of the risk (e.g., negative impact on project schedule or resource allocation). By sorting and grading the risk scores, nodes are divided into high, medium, and low risk levels. Based on the risk assessment scores of all nodes, a risk assessment report is generated. The report includes the risk level, key influencing factors, and recommended countermeasures for each node.

[0127] In one example, according to the risk assessment report, the early warning indicators are dynamically adjusted to obtain an early warning rule configuration file, including:

[0128] Extract data on progress delay indicators, resource overrun indicators, quality abnormality indicators, and risk accumulation indicators in the risk assessment report, and establish an early warning indicator matrix based on historical data of clinical trial projects to obtain an initial early warning indicator set;

[0129] The initial warning indicator set is input into the sensitivity analysis model to calculate the sensitivity coefficient, and the indicator sensitivity evaluation results are obtained. Based on the indicator sensitivity evaluation results, the warning indicator priority sequence analysis is performed to obtain the graded warning indicator set;

[0130] Input the hierarchical warning indicator set into the algorithm model to generate a test sample set, and generate an initial warning rule set through a dynamic combination of task allocation and warning plan;

[0131] The warning effect of the initial warning rule set is evaluated to obtain the warning rule evaluation matrix, and the warning rule evaluation matrix is ​​input into the recursive boundary search module. The warning threshold and trigger conditions are adjusted through iterative optimization until the evaluation index reaches the preset target, and the target warning rule set is obtained;

[0132] The target warning rule set is divided into risk levels to obtain a graded warning configuration plan, and the graded warning configuration plan is integrated into a configuration file in a standard format to obtain a warning rule configuration file.

[0133] In this example, data is extracted from the progress delay index, resource overrun index, quality abnormality index, and risk accumulation index in the risk assessment report, and an early warning index matrix is ​​established in combination with the historical data of the clinical trial project to obtain the initial early warning indicator set. The progress delay index is calculated as the difference between the actual progress of the project and the planned progress, and the formula is as follows:

[0134] ;

[0135] The resource overrun indicator measures the ratio of actual resource usage to allocated resources, and the formula is:

[0136] ;

[0137] The quality anomaly index measures the frequency of quality problems that occur during the project, and the formula is:

[0138] ;

[0139] The risk accumulation index measures the comprehensive degree of various risks, and the formula is:

[0140] ;

[0141] in is the weight of each risk item, It is After extracting these four indicators, combined with the historical data of the clinical trial project, these indicators are calculated for each clinical trial center to form an initial early warning indicator set. Clinical trial centers, Early warning indicators. Then the early warning indicator matrix It is expressed as:

[0142] ;

[0143] The initial warning indicator set is input into the sensitivity analysis model to calculate the sensitivity coefficient of each indicator. Sensitivity analysis uses partial derivatives to measure the impact of each indicator on the warning result. For example, for a certain warning indicator The sensitivity coefficient , and its calculation formula is:

[0144] ;

[0145] By calculating the sensitivity coefficient, we can identify which indicators are most sensitive to changes in project risks and which indicators have less impact on the adjustment of early warning rules. Based on the sensitivity assessment results, we prioritize the early warning indicators. Assume that the sensitivity coefficient set is , arrange in descending order and generate a hierarchical warning indicator set, indicating the relative importance of each warning indicator. Input the hierarchical warning indicator set into the algorithm model to generate a test sample set. At this stage, build a model and train the model using the warning indicators in historical data and their corresponding response states (such as whether the project is delayed, resource overruns, etc.). Infer the warning rules through classification algorithms in machine learning (such as decision trees, SVM, etc.). Generate an initial warning rule set through the dynamic combination of task allocation and warning plans. The setting of the rules is adjusted according to the feedback of the warning indicators and historical data obtained previously to ensure the adaptability and effectiveness of the rules. After the initial warning rule set is generated, the warning effect is evaluated. Evaluate the accuracy and effectiveness of the current rules by analyzing historical warning data. For example, measure the performance of the warning rules by calculating indicators such as the accuracy, recall rate, and F1-score of the rules. Warning rule evaluation matrix It is expressed as:

[0146] ;

[0147] The warning rule evaluation matrix is ​​input into the recursive boundary search module, and the warning threshold and trigger conditions are adjusted through iterative optimization until the evaluation index reaches the preset target (such as accuracy exceeding 90%). Complete, assuming there is a threshold adjustment function , adjust the warning rules based on historical feedback:

[0148] ;

[0149] Through repeated iterations, the optimized target warning rule set is obtained. The target warning rule set is divided into risk levels to determine different levels of warning response strategies. For example, based on historical data, if certain indicators exceed a certain threshold, the clinical trial center is considered to be at a high risk level, otherwise it is considered to be at a medium or low risk level. These risk levels are divided by the following formula:

[0150] Risk Level

[0151] According to these classification rules, a hierarchical warning configuration plan is generated. The warning rule configuration plan is integrated according to the standard format to generate a warning rule configuration file that can be used by the system This profile can be integrated into the project management system to achieve automated early warning and risk management.

[0152] In one example, the risk assessment report and the warning rule configuration file are used to determine the warning trigger, and a hierarchical push process is performed to generate a warning information push record, including:

[0153] Perform threshold monitoring on the risk assessment score in the risk assessment report and the warning threshold in the warning rule configuration file to obtain the warning trigger state matrix;

[0154] Perform hierarchical warning processing based on the warning trigger state matrix to obtain the response strategy matrix;

[0155] Build an early warning information template based on the response strategy matrix, classify and sort the risk type, impact scope, probability of occurrence and disposal suggestions according to the early warning level, and obtain the early warning information content;

[0156] Match the recipients of the warning information content to obtain a list of recipients of the warning information;

[0157] Input the warning information content and the warning information receiving list into the message distribution system, use the multi-level cache mechanism to execute information push, and obtain the information push status;

[0158] Monitor the status of information push in real time, use the heartbeat detection mechanism to track the delivery of information, record the information reception time and confirmation status, and obtain push feedback data;

[0159] Quantitative analysis is performed on push feedback data to calculate the delivery rate, response time and processing progress of different warning levels to obtain the warning effect evaluation results. Based on the warning effect evaluation results, a warning information push record is generated that includes the warning grade, push object, response status and processing result.

[0160] In this example, the risk assessment score in the risk assessment report and the warning threshold in the warning rule configuration file are monitored to obtain the warning trigger state matrix. For example, suppose there is a risk assessment score matrix Risk assessment scores representing different clinical trial centers:

[0161] ;

[0162] in Indicates The risk assessment score of each clinical trial center. Each score is calculated based on multiple indicators, such as schedule delay, resource overrun, quality abnormality, etc. In order to decide whether to trigger an alert, it is compared with the alert threshold in the alert rule configuration file. Assume that the alert threshold is , then the warning trigger state matrix It is represented as a matrix, where each element indicates whether an early warning is triggered in a specific clinical trial center. The specific calculation formula is:

[0163] ;

[0164] According to the above formula, if the risk assessment score of a trial center is Exceeding the set warning threshold , then the warning trigger status of the test center is 1, indicating that the warning is triggered; if the score is lower than the threshold, the status is 0, indicating that the warning is not triggered. Based on the warning trigger status matrix, hierarchical warning processing is performed to obtain the response strategy matrix Each element of the response strategy matrix represents the measures to be taken in the corresponding warning state. For example, when the risk assessment score triggers an early warning, different response strategies are set for different risk levels according to the score. Specifically, the classification is carried out in the following way:

[0165] ;

[0166] in and Represents the thresholds of high risk and medium risk respectively. Different response strategies are adopted for different levels of risk, such as adding resources, adjusting schedules or reallocating tasks. Based on the response strategy matrix , build an early warning information template. The template will include information such as risk type, impact scope, probability of occurrence and disposal suggestions, and classify and organize this information according to the warning level. Match the recipient of the early warning information content to obtain the early warning information receiving list The receiving list contains all the people or teams that need to receive the warning information. The receiving list is represented as a matrix, where each row represents a recipient and the column represents the type of warning information received:

[0167] ;

[0168] Input the warning information content and the receiving list into the message distribution system for push. Assume that a multi-level cache mechanism is used to implement the push and distribution of information. The multi-level cache mechanism can effectively ensure the efficient transmission of information while avoiding system overload. The message distribution system pushes the warning information to the corresponding receiving object and records the status of each push. The push status matrix It is expressed as:

[0169] ;

[0170] The system also needs to perform real-time monitoring to ensure the delivery of information. Through the heartbeat detection mechanism, it regularly checks whether the information has been delivered and records the receiving time and confirmation status. Assume that the information delivery status matrix Represents the delivery status of each receiver, where each element Representative Whether the recipient confirms receipt of the message within the specified time:

[0171] ;

[0172] After the information is delivered and confirmed, the collected push feedback data will be quantitatively analyzed. The key indicators of quantitative analysis include the delivery rate, response time and processing progress of different warning levels. The delivery rate refers to the proportion of warning information successfully delivered, the response time refers to the average time for the recipient to respond to the warning, and the processing progress indicates the completion of the response measures taken. An early warning effect evaluation report is generated based on the results of quantitative analysis. The report records the warning level, push object, response status and processing results, and provides a basis for optimizing the early warning strategy. Through a continuous feedback loop, the early warning system can gradually improve its accuracy and response efficiency, ensuring that clinical trial projects can identify and respond to potential risks in a timely manner, thereby improving the success rate of the project and data quality.

[0173] Reference Figure 2 This embodiment provides a progress tracking and early warning system for a clinical trial project, including:

[0174] Collection module 1 is used to collect and score the quality of the subject recruitment progress data, follow-up completion data, trial process data and project milestone completion data of the clinical trial center, and obtain the data quality scoring matrix and data node connection status;

[0175] Mapping module 2, used for inputting the data quality scoring matrix and the data node connection status into the progress network construction model for network mapping, and obtaining a project progress dependency diagram;

[0176] Extraction module 3, used to extract spatial features and temporal features from the project schedule dependency graph to obtain a risk feature vector;

[0177] Evaluation module 4, used for performing dual-objective dynamic evaluation on the risk feature vector to obtain a risk evaluation report;

[0178] Adjustment module 5, used to dynamically adjust the early warning indicators according to the risk assessment report to obtain the early warning rule configuration file;

[0179] The generating module 6 is used to make warning trigger judgment on the risk assessment report and the warning rule configuration file, and perform hierarchical push processing to generate a warning information push record.

[0180] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0181] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and 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 provided by the present invention and used in the embodiments 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. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0182] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0183] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A progress tracking and early warning method for a clinical trial project, characterized in that: The following steps are involved: The subject recruitment progress data, follow-up completion data, trial process data and project milestone completion data of the clinical trial center are collected and scored for quality, and the data quality scoring matrix and data node connection status are obtained; specifically, the following are included: the enrollment rate and screening failure rate in the subject recruitment progress data are calculated, the timely follow-up rate and dropout rate in the follow-up completion data are counted, the adverse event incidence rate and protocol compliance in the trial process data are analyzed, and the planned completion rate and extension rate in the project milestone completion data are calculated to obtain the original data indicator set; an evaluation standard system is established based on the original data indicator set, and based on the evaluation standard system, the original data indicator set is quantitatively scored and added. weight calculation to obtain the data quality score of each clinical trial center; statistical analysis is performed on the data interaction frequency and information transmission timeliness between the clinical trial centers to obtain a data interaction intensity matrix, and data fusion processing is performed on the data quality scores of the clinical trial centers and the data interaction intensity matrix to obtain a data quality scoring matrix; the data quality scoring matrix is ​​normalized to obtain a normalized score, and the connection relationship between the clinical trial centers is divided into high priority, medium priority and low priority according to the normalized score, and the data transmission path of each priority is determined to obtain the initial node connection state; the initial node connection state is path optimized and dynamically monitored to obtain the data node connection state; Inputting the data quality scoring matrix and the data node connection status into the progress network construction model for network mapping to obtain a project progress dependency graph; Extracting spatial features and temporal features from the project schedule dependency graph to obtain a risk feature vector; Performing a dual-objective dynamic assessment on the risk feature vector to obtain a risk assessment report; According to the risk assessment report, the early warning indicators are dynamically adjusted to obtain an early warning rule configuration file; Perform warning trigger judgment on the risk assessment report and the warning rule configuration file, perform hierarchical push processing, and generate a warning information push record.

2. The progress tracking and early warning method of a clinical trial project according to claim 1, characterized in that: The step of inputting the data quality scoring matrix and the data node connection status into the progress network construction model for network mapping to obtain a project progress dependency graph includes: Normalizing and mapping the score values ​​in the data quality score matrix to obtain a clinical trial center node weight matrix; Assign a weight of 0.9 to the high priority connection in the data node connection state, a weight of 0.6 to the medium priority connection, and a weight of 0.3 to the low priority connection, to obtain a connection edge weight matrix; Input the clinical trial center node weight matrix and the connection edge weight matrix into the ClinicalDispatch calculation framework, calculate the degree centrality, betweenness centrality and closeness centrality of each clinical trial center, and obtain a node centrality index set; The node centrality index set is comprehensively scored using a weighted summation method to obtain a node comprehensive score; A network adjacency matrix is ​​constructed based on the comprehensive scores of the nodes, strong connections are established for nodes with a score difference of less than 0.2, and weak connections are established for nodes with a score difference between 0.2 and 0.5, to obtain an initial network structure; Inputting the initial network structure into a Latin hypercube sampling model to perform feature dimensionality reduction to obtain a target network structure; The progress index is calculated for each clinical trial center node in the target network structure to obtain a progress state vector, and the progress state vector is mapped to the target network structure to obtain a project progress dependency graph.

3. The progress tracking and early warning method of a clinical trial project according to claim 2, characterized in that: The step of extracting spatial features and temporal features from the project schedule dependency graph to obtain a risk feature vector includes: Input the project schedule dependency graph into the input layer of the spatial convolutional network, set three convolution kernels for spatial feature extraction, and the convolution kernel sizes are 3×3, 5×5, and 7×7, respectively, to obtain a multi-scale spatial feature matrix; Performing a maximum pooling operation on the multi-scale spatial feature matrix, with a pooling window size of 2×2 and a step size of 2, to obtain a spatial feature after dimensionality reduction; Reorganize the reduced-dimensional spatial features according to the time series to construct a time series data sequence, and input the time series data sequence into a bidirectional gated recurrent unit to obtain a forward feature sequence and a reverse feature sequence; The forward feature sequence and the reverse feature sequence are concatenated to obtain a complete temporal feature vector, and the spatial feature and the temporal feature vector are input into the attention layer to calculate the feature weight coefficient, where the spatial feature weight is 0.6 and the temporal feature weight is 0.4, to obtain the weighted feature; The weighted features are standardized to obtain a standardized feature matrix, and the standardized feature matrix is ​​input into a fully connected layer, and a nonlinear transformation is performed using a ReLU activation function to obtain a risk feature vector.

4. The progress tracking and early warning method of a clinical trial project according to claim 3, characterized in that: The dual-objective dynamic evaluation of the risk feature vector is performed to obtain a risk evaluation report, including: Performing state space mapping on the risk feature vector, constructing a state vector including progress completion rate, resource utilization rate, and risk degree, and constructing a Markov state transition model based on the state vector to obtain an initial state space; Dividing the initial state space into subspaces according to the risk level, dividing the risk level into a high-risk subspace, a medium-risk subspace and a low-risk subspace, and calculating the state transition probability of each subspace to obtain a state transition matrix; Based on the state transfer matrix, dynamic response processing and state transfer probability update are performed to obtain a dynamic state transfer matrix, a mixed integer programming objective function is established based on the dynamic state transfer matrix, a progress achievement rate objective function and a resource utilization rate objective function are set, and weight coefficients of 0.6 and 0.4 are assigned respectively to obtain a weighted objective function; Setting constraints on the weighted objective function to obtain a constraint set, and inputting the weighted objective function and the constraint set into a linear programming solver, using a branch and bound method to solve an optimal decision sequence to obtain a risk control solution; A risk quantitative assessment is performed on each decision node in the risk control scheme, the product of the risk occurrence probability and the impact degree is calculated to obtain a risk assessment score, and the risks are graded according to the risk assessment score to generate a risk assessment report.

5. The progress tracking and early warning method of a clinical trial project according to claim 4, characterized in that: According to the risk assessment report, the early warning indicators are dynamically adjusted to obtain an early warning rule configuration file, including: Extract data on the progress delay index, resource overrun index, quality abnormality index and risk accumulation index in the risk assessment report, and establish an early warning index matrix in combination with the historical data of the clinical trial project to obtain an initial early warning indicator set; Inputting the initial warning indicator set into the sensitivity analysis model to calculate the sensitivity coefficient, obtain the indicator sensitivity evaluation result, and perform warning indicator priority sequence analysis based on the indicator sensitivity evaluation result to obtain the graded warning indicator set; Input the hierarchical warning indicator set into the algorithm model to generate a test sample set, and generate an initial warning rule set through a dynamic combination of task allocation and warning plan; Performing a warning effect evaluation on the initial warning rule set to obtain a warning rule evaluation matrix, and inputting the warning rule evaluation matrix into a recursive boundary search module, adjusting the warning threshold and triggering condition by an iterative optimization method until the evaluation index reaches a preset target, thereby obtaining a target warning rule set; The target warning rule set is divided into risk levels to obtain a graded warning configuration scheme, and the graded warning configuration scheme is integrated into a configuration file in a standard format to obtain a warning rule configuration file.

6. The progress tracking and early warning method of a clinical trial project according to claim 5, characterized in that: The step of performing warning triggering judgment on the risk assessment report and the warning rule configuration file, and performing hierarchical push processing to generate a warning information push record includes: Perform threshold monitoring on the risk assessment score in the risk assessment report and the warning threshold in the warning rule configuration file to obtain a warning trigger state matrix; Perform hierarchical warning processing based on the warning trigger state matrix to obtain a response strategy matrix; Building an early warning information template based on the response strategy matrix, classifying and arranging the risk type, impact scope, occurrence probability and disposal suggestions according to the early warning level, and obtaining the early warning information content; Match the warning information content with the recipient to obtain a warning information recipient list; Input the warning information content and the warning information receiving list into the message distribution system, use a multi-level cache mechanism to execute information push, and obtain the information push status; Monitor the information push status in real time, use a heartbeat detection mechanism to track the information delivery, record the information reception time and confirmation status, and obtain push feedback data; The push feedback data is quantitatively analyzed to calculate the delivery rate, response time and processing progress of different warning levels to obtain the warning effect evaluation result, and generate a warning information push record including the warning level, push object, response status and processing result based on the warning effect evaluation result.

7. A progress tracking and early warning system for clinical trial projects, characterized in that: For implementing the steps of the method according to any one of claims 1 to 6, the system comprises: The collection module is used to collect and score the quality of the subject recruitment progress data, follow-up completion data, trial process data and project milestone completion data of the clinical trial center, and obtain the data quality score matrix and data node connection status; A mapping module, used for inputting the data quality scoring matrix and the data node connection status into the progress network construction model for network mapping to obtain a project progress dependency graph; An extraction module, used to extract spatial features and temporal features from the project schedule dependency graph to obtain a risk feature vector; An evaluation module, used for performing a dual-objective dynamic evaluation on the risk feature vector to obtain a risk evaluation report; An adjustment module, used to dynamically adjust the early warning indicators according to the risk assessment report to obtain an early warning rule configuration file; The generation module is used to perform warning trigger judgment on the risk assessment report and the warning rule configuration file, and perform hierarchical push processing to generate a warning information push record.

Images