Intelligent management method and system for drug clinical items

Abstract

The invention discloses an intelligent management method and system for a drug clinical project, and relates to the field of medical informatics, and the method comprises the steps: constructing a dynamic data co-evolution diagram, and dividing the dynamic data co-evolution diagram into a plurality of collaborative small groups; identifying risk collaborative small groups, and judging and identifying high-risk nodes; performing high-risk rechecking on the data in the high-risk node, and if the rechecking result is misjudgment, returning to the main processing channel; and if the re-checking result is that misjudgment is not carried out, adjusting the configuration weight of the high-risk node, and applying the configuration weight to a linkage calculation module. According to the method, a closed-loop mechanism for dynamically identifying and finely controlling high-risk nodes is constructed, preposed discovery, differentiated response and resource coordinated regulation and control of sensitive risks can be realized in drug clinical data management, and the method has good system expandability, safety protection capability and intelligent level.

Description

Technical Field

[0001] The present invention relates to the field of medical informatics, and in particular to an intelligent management method and system for drug clinical projects. Background Art

[0002] The clinical data collection, processing, and management process involves a large amount of high-frequency, cross-departmental, and cross-platform data interactions. Traditional clinical trial management systems (such as EDC, CDMS, etc.) are mostly based on static flowcharts and rule engines, and perform linear management of operations such as collection, review, transmission, and storage of trial data.

[0003] Existing technologies, such as the invention patent with announcement number CN114999663B, are data management methods for clinical research projects. The method includes: in response to a node event initiated by a workflow management middle platform, a call request is triggered, and the workflow management middle platform communicates with multiple clinical research tools; in response to the call request, a specific research tool among the multiple clinical research tools is called, and at least one target parameter provided by the specific research tool is collected; a customized format copy of the target parameter is created and stored in the underlying database.

[0004] Existing technologies, such as the invention patent with announcement number CN112885487B, are a drug genetic testing project management system, including: a drug basic testing project creation module, which is used to obtain patient information, historical medication parameters and corresponding blood test parameters after each medication, to identify the patient's effective drug factors and allergic drug factors, and to create drug basic testing projects based on the identification results; a drug basic testing plan compilation module, which is used to dynamically compile drug basic testing steps based on the created drug basic testing projects and the corresponding testing parameters after each medication of the sample; a drug basic testing report generation module, which is used to realize the mining and storage of genetic testing information, and based on the identification of genetic testing information, obtain recommended medication and generate a genetic testing report.

[0005] Based on the above solutions, it can be seen that in the field of medical informatics, existing technologies usually use nodes as units for authority allocation and task scheduling, rely on workflow middleware to drive data collection and tool calls, are mostly based on linear processes, focus on result output and tool calls, lack risk modeling mechanisms and dynamic control strategies for data ontology, and are unable to meet the higher requirements of increasingly complex drug clinical trials for sensitive data security, flow controllability, and flexible authority configuration. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the present invention provides an intelligent management method and system for drug clinical projects. To achieve the above objectives, the present invention is implemented through the following technical solutions: An intelligent management method for drug clinical projects, comprising:

[0007] The data units uploaded by the drug clinical project are mapped into structured nodes, a dynamic data co-evolution graph is constructed, and the dynamic data co-evolution graph is divided into several collaborative small groups.

[0008] Identify small risk collaborative groups, construct perturbed data copies of the original data of each structured node in the risk collaborative small group, inject the original data and perturbed data copies of each structured node into the simulation processing channel in parallel, collect the contextual collaboration factors output by the simulation processing channel, obtain the initial sensitivity score of each structured node in the risk collaborative small group, and identify high-risk nodes.

[0009] Generate a temporary freeze instruction for the project, separate the execution path of the high-risk node from the main processing channel, build a local execution environment, and perform a high-risk review on the data in the high-risk node. If the review result is a misjudgment, the execution path of the high-risk node will be returned to the main processing channel.

[0010] If the review result is no misjudgment, the configuration weight of the high-risk node is adjusted and applied to the linkage calculation module.

[0011] As a preferred technical solution, the data units uploaded by the drug clinical project are mapped into structured nodes, a dynamic data co-evolution graph is constructed, and the dynamic data co-evolution graph is divided into several collaborative small groups. The specific process is as follows:

[0012] The raw data units uploaded during the collection, transmission and processing of drug clinical projects are structured and mapped to form structured nodes.

[0013] A dynamic data co-evolution graph is constructed based on the historical data interaction records of structured nodes, wherein the edges between the structured nodes in the dynamic data co-evolution graph represent the co-evolution relationship.

[0014] Based on the graph structure partitioning algorithm, highly co-evolving structured nodes are clustered into small collaborative groups. Small collaborative groups are sets of data nodes formed based on the co-evolution relationship between nodes in the dynamic data co-evolution graph.

[0015] As a preferred technical solution, identify risk-cooperative small groups. The specific process is as follows:

[0016] Track the average sensitivity score of each collaborative small group in the historical clinical project process, extract the average score threshold from the database, and if the average sensitivity score of a collaborative small group in the historical clinical project process exceeds the preset average score threshold, it will be identified as a collaborative small group with potential risks and recorded as a risky collaborative small group.

[0017] If the average sensitivity score of a collaborative subgroup in the historical clinical project process is less than or equal to the average score threshold, it will not be identified as a collaborative subgroup with potential risks.

[0018] As an optimal technical solution, a perturbed data copy of the original data of each structured node in the risk collaboration small group is constructed, and the original data and perturbed data copy of each structured node are injected into the simulation processing channel in parallel, specifically including:

[0019] The average score difference between the average sensitivity score of each risk group in the historical clinical project process and the average score threshold is obtained, and mapping matching is performed based on the average score difference to obtain the preset perturbation parameters of each risk group.

[0020] The preset disturbance parameters include weakening copy parameters and enhancing copy parameters. The weakening copy parameters are specifically the context field hiding ratio, and the enhancing copy parameters are specifically the abnormal combination field supplement ratio.

[0021] For the original data of each structured node in a risk collaboration small group, based on the preset perturbation parameters, a perturbed data copy is constructed, including a weakened copy and an enhanced copy;

[0022] The original data, weakened copies and enhanced copies of each structured node in the risk collaborative small group are injected into the simulation processing channel in parallel. The simulation processing channel includes a normal simulation processing channel, an extracted simulation processing channel and an intervention simulation processing channel. Among them, the original data is injected into the normal simulation processing channel as a benchmark reference, the weakened copy is injected into the extracted simulation processing channel to simulate the channel processing behavior when the context field is missing, and the enhanced copy is injected into the intervention simulation processing channel to simulate the channel processing behavior when the abnormal combination field exists.

[0023] As a preferred technical solution, the contextual synergy factor output by the simulation processing channel is collected to obtain the initial sensitivity score of each structured node in the risk synergy small group, and to identify high-risk nodes, specifically including:

[0024] The contextual synergy factors output by each simulation processing channel are collected. The contextual synergy factors specifically include the structural co-occurrence index, the path conflict factor, and the sensitive node influence distribution factor.

[0025] The difference between the contextual synergy factor output by the intervention channel and the contextual synergy factor output by the normal channel is obtained. At the same time, the difference between the contextual synergy factor output by the extraction channel and the contextual synergy factor output by the normal channel is obtained. The obtained two-layer difference is weighted and corrected to obtain the initial sensitivity score of each structured node in the risk synergy small group.

[0026] If the initial sensitivity score of a structured node is greater than or equal to the preset sensitivity score threshold, the structured node is judged to be a high-risk node.

[0027] If the initial sensitivity score of a structured node is less than the preset sensitivity score threshold, the structured node is judged to be a low-risk node.

[0028] As a preferred technical solution, high-risk review is conducted on the data in high-risk nodes, specifically including:

[0029] Generate a temporary freeze instruction for the project, separate the execution path of the high-risk node from the main processing channel, call the containerized execution unit, and build a local execution environment. The local execution environment is an execution environment without downstream write permissions. Based on the initial sensitivity score difference between the initial sensitivity score of the high-risk node and the sensitivity score threshold, the initial sensitivity score difference pre-stored in the database - the local execution environment resource configuration index mapping is concentratedly mapped to obtain the local execution environment resource configuration index of the high-risk node, and the local execution environment is constructed according to the local execution environment resource configuration index.

[0030] In the local execution environment, the contextual synergy factor differences of the data in the high-risk nodes are reanalyzed to obtain the review sensitivity score.

[0031] If the reviewed sensitivity score of the high-risk node is still greater than or equal to the sensitivity score threshold, the review result is no misjudgment.

[0032] If the review sensitivity score of the high-risk node is less than the sensitivity score threshold, the review result is a misjudgment, and the execution path of the high-risk node is returned to the main processing channel.

[0033] As a preferred technical solution, adjust the configuration weight of the high-risk node, specifically including:

[0034] The initial sensitivity score and the reviewed sensitivity score of the high-risk node are extracted, and the average is taken as the comprehensive sensitivity score of the high-risk node.

[0035] Based on the comprehensive sensitivity score of the high-risk node, the mapping set of comprehensive sensitivity score-configuration weight adjustment factor pre-stored in the database is input, and mapping matching is performed to obtain the configuration weight adjustment factor of the high-risk node. The configuration weight adjustment factor is a parameter indicator used to quantify the authority adjustment range of the high-risk node.

[0036] As a preferred technical solution, adjusting the configuration weight of the high-risk node also includes adjusting the configuration weights of each risk-related node corresponding to the high-risk node. The specific processing conditions are:

[0037] Obtain other structured nodes in the risk collaboration subgroup where the high-risk node is located, extract metadata, and parse to obtain the call fields of other structured nodes.

[0038] Cross-compare the call fields of other structured nodes with the fields in high-risk nodes to identify structured nodes that have associated call relationships with high-risk nodes. These are recorded as risk-associated nodes. The associated call relationships include field reading, downstream jumps, and permission writing.

[0039] Obtain the number of associated call relationships between each risk-associated node and the high-risk node, and input the mapping of the number of associated call relationships-configuration weight adjustment factor attenuation ratio pre-stored in the database to perform mapping matching to obtain the configuration weight adjustment factor attenuation ratio of each risk-associated node, and interact the configuration weight adjustment factor attenuation ratio of each risk-associated node with the configuration weight adjustment factor of the high-risk node to obtain the associated configuration weight adjustment factor of each risk-associated node.

[0040] As an optimal technical solution, it is applied to the linkage calculation module, specifically including:

[0041] Extract the linkage computing module numbers of high-risk nodes and each risk-related node, and send the configuration weight adjustment factor of the high-risk node and the associated configuration weight adjustment factor of each risk-related node to the corresponding linkage computing module in the form of configuration instruction packets. After receiving the configuration instruction packet, the linkage computing module maps the configuration instruction to the resource scheduling control parameter configuration table, and matches the resource scheduling control parameters of the linkage computing module, including the disk read rate limit value, the disk write rate limit value, the upper limit of available memory, and the CPU time quota limit value.

[0042] The linked computing module writes resource scheduling control parameters and calls various management APIs to execute resource scheduling control parameters.

[0043] In addition, an intelligent management system for drug clinical projects is also provided, including:

[0044] The collaborative small group division module is used to map the data units uploaded by the drug clinical project into structured nodes, construct a dynamic data co-evolution graph, and divide the dynamic data co-evolution graph into several collaborative small groups.

[0045] The high-risk node identification module is used to identify small risk collaborative groups, construct perturbed data copies of the original data of each structured node in the risk collaborative small group, inject the original data and perturbed data copies of each structured node into the simulation processing channel in parallel, collect the contextual collaboration factors output by the simulation processing channel, obtain the initial sensitivity score of each structured node in the risk collaborative small group, and judge and identify high-risk nodes.

[0046] The high-risk node review module is used to generate temporary freeze instructions for the project, separate the execution path of the high-risk node from the main processing channel, build a local execution environment, and perform high-risk review on the data in the high-risk node. If the review result is a misjudgment, the execution path of the high-risk node will be returned to the main processing channel.

[0047] The high-risk node configuration weight module is used to adjust the configuration weight of the high-risk node when the review result is not a misjudgment, and apply it to the linkage calculation module.

[0048] Compared with the prior art, the embodiments of the present invention have at least the following beneficial effects:

[0049] (1) The present invention provides an intelligent management method for drug clinical projects. By mapping data units in drug clinical projects into structured nodes and constructing a dynamic data co-evolution graph, the co-evolution characteristics of clinical data in the time series processing process are accurately captured, effectively revealing the linkage between potential data nodes. Based on the graph structure partitioning algorithm, small synergistic groups are extracted, and the historical sensitivity scores in the co-evolution process are further compared with the threshold to identify risky synergistic groups.

[0050] (2) The present invention introduces an interference copy mechanism to construct weakened and enhanced copies of the original data based on preset disturbance parameters, and injects them into the extraction channel and intervention channel respectively to simulate the data behavior pattern under the condition of missing context information or abnormal enhancement. By calculating the difference between the structural co-occurrence index, path conflict factor and sensitive node influence distribution factor output by the normal channel, the initial sensitivity score of the node is formed, and the quantitative modeling of the node risk level is achieved. At the same time, the system also constructs a local execution environment to review high-risk nodes, and judges whether there is a misjudgment based on the review score, making the risk identification mechanism more accurate and fault-tolerant, and avoiding false freezing and miscontrol.

[0051] (3) The present invention uses the mapping relationship and mechanism of score difference-configuration index-resource parameters to enable high-risk nodes to fluctuate according to their sensitivity scores, and dynamically generate the most suitable local execution environment resource configuration plan. Compared with the fixed resource allocation mechanism, it better balances performance assurance and system security, and improves the response efficiency and processing flexibility of clinical data processing.

[0052] (4) This invention uses field cross-comparison to identify the strength of association with high-risk nodes, and then calculates the association configuration weight adjustment factor. This factor, along with the high-risk node factor, is transmitted to the corresponding linkage calculation module to dynamically adjust its resource control parameters. This effectively prevents the spread of data anomalies in the pathway, enhancing the system's collaborative defense capabilities and closed-loop processing of highly sensitive events.

[0053] Of course, any product implementing the present invention does not necessarily need to achieve all of the above advantages at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Schematic diagram of the method of the present invention.

[0055] Figure 2 Schematic diagram of the system module of the present invention.

[0056] Figure 3 It is a logical flow diagram of the present invention. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] In the description of the present invention, it should be understood that the terms "opening", "upper", "lower", "thickness", "top", "middle", "length", "inside", "around" and the like indicating orientation or positional relationship are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0059] See also Figure 1 As shown, an embodiment of the present invention provides an intelligent management method for drug clinical projects, including:

[0060] like Figure 3 FIG. 1 is a logic flow chart involved in an embodiment of the present invention.

[0061] The data units uploaded by the drug clinical project are mapped into structured nodes, a dynamic data co-evolution graph is constructed, and the dynamic data co-evolution graph is divided into several collaborative small groups.

[0062] The raw data units uploaded during the collection, transmission and processing of drug clinical projects are structured and mapped to form structured nodes.

[0063] In this embodiment of the present invention, raw data units include clinical test records, drug administration records, project operation logs, and task flow execution records. Raw data units are mapped based on semantic independence. For example, in drug administration records, they can be mapped to structured nodes such as drug name, drug dosage, and administration time.

[0064] A dynamic data co-evolution graph is constructed based on the historical data interaction records of structured nodes, wherein the edges between the structured nodes in the dynamic data co-evolution graph represent the co-evolution relationship.

[0065] In an embodiment of the present invention, the co-evolution relationship specifically includes a same-node call relationship, a same-role contact relationship, a decision chain correlation relationship, and a simultaneous occurrence time sequence relationship.

[0066] Historical data interaction records refer to the systematic recording of the time, space, and context-related interactive behaviors that occur in the collection, transmission, processing, and call stages of each data unit (i.e., structured node) throughout the entire experiment.

[0067] Among them, the same-node call relationship refers to two structured nodes being called, processed or dependent on the same functional node (such as a processing module, algorithm task, interface service, etc.) at the same time in the historical process.

[0068] A same-role access relationship means that two structured nodes are accessed, modified, or read by the same type of user during the data life cycle.

[0069] The decision chain correlation relationship refers to the existence of a clear conditional dependency relationship between two structured nodes, that is, the value or state of one node affects the judgment, selection or branch decision of the other node.

[0070] A simultaneous temporal relationship means that two structured nodes have a stable time synchronization relationship in the historical process, that is, their data generation or processing time is frequently synchronized or at fixed intervals.

[0071] When there is a co-evolution relationship between a structured node and another structured node, a directed edge is established to construct a dynamic data evolution graph.

[0072] In an embodiment of the present invention, based on the Louvain graph structure partitioning algorithm, highly co-evolving structured nodes are clustered into small collaborative groups. The small collaborative group is a set of data nodes formed based on the co-evolution relationship between nodes in the dynamic data co-evolution graph.

[0073] In the embodiment of the present invention, the identification process of the collaborative small group is as follows:

[0074] The edge weight values between structured nodes are obtained from the dynamic data evolution graph. The Louvain graph structure partitioning algorithm is used to identify the structured node groups whose sum of edge weight values is higher than the sum of preset weight values, and record them as collaborative small groups.

[0075] Identify small risk collaborative groups, construct perturbed data copies of the original data of each structured node in the risk collaborative small group, inject the original data and perturbed data copies of each structured node into the simulation processing channel in parallel, collect the contextual collaboration factors output by the simulation processing channel, obtain the initial sensitivity score of each structured node in the risk collaborative small group, and identify high-risk nodes.

[0076] Identify risk-cooperating small groups. The specific process is as follows:

[0077] Track the average sensitivity score of each collaborative small group in the historical clinical project process, extract the average score threshold from the database, and if the average sensitivity score of a collaborative small group in the historical clinical project process exceeds the preset average score threshold, it will be identified as a collaborative small group with potential risks and recorded as a risky collaborative small group.

[0078] It should be noted that the average sensitivity score is obtained by averaging the sensitivity scores of each structured node in the collaborative small group in the historical clinical project process. If a structured node is a newly emerged structured node, the average score threshold is used as the sensitivity score of the newly emerged structured node.

[0079] If the average sensitivity score of a collaborative subgroup in the historical clinical project process is less than or equal to the average score threshold, it will not be identified as a collaborative subgroup with potential risks.

[0080] Construct a perturbed data copy of the original data of each structured node in the risk collaboration small group, and inject the original data and perturbed data copy of each structured node into the simulation processing channel in parallel, specifically including:

[0081] The average sensitivity score of each risk group in the historical clinical project process is subtracted from the average score threshold to obtain the average score difference of each risk group. Mapping and matching are performed based on the average score difference to obtain the preset perturbation parameters of each risk group.

[0082] The preset disturbance parameters include weakening copy parameters and enhancing copy parameters. The weakening copy parameters are specifically the context field hiding ratio, and the enhancing copy parameters are specifically the abnormal combination field supplement ratio.

[0083] For the original data of each structured node in a risk collaborative small group, based on the preset perturbation parameters, a perturbed data copy is constructed, including a weakened copy and an enhanced copy. The specific process is as follows:

[0084] For the original data of a structured node, find the fields that appear in both the upstream and downstream structured nodes to form a context field set. Based on the context field hiding ratio of the structured node, perform field value hiding operations in the context field set, such as setting to zero, masking, and replacing placeholders, to form a weakened copy.

[0085] For the original data of a structured node, based on the supplement ratio of the abnormal combination fields of the structured node, the abnormal combination fields are retrieved from the abnormal field combination library pre-established in the database, and attached to the data structure of the structured node to construct an enhanced copy.

[0086] The original data, weakened copies and enhanced copies of each structured node in the risk collaborative small group are injected into the simulation processing channel in parallel. The simulation processing channel includes a normal simulation processing channel, an extracted simulation processing channel and an intervention simulation processing channel. Among them, the original data is injected into the normal simulation processing channel as a benchmark reference, the weakened copy is injected into the extracted simulation processing channel to simulate the channel processing behavior when the context field is missing, and the enhanced copy is injected into the intervention simulation processing channel to simulate the channel processing behavior when the abnormal combination field exists.

[0087] The original data, weakened copies, and enhanced copies of each structured node in a risk collaboration cluster are injected in parallel into three different simulation processing channels: a normal simulation channel, an extracted simulation channel, and an intervention simulation channel. The primary purpose is to comprehensively assess the sensitivity and potential risks of structured nodes by comparing data processing performance under different perturbation conditions. The normal channel injects the original data as a baseline, reflecting processing behavior in an undisturbed environment. The extracted channel injects weakened copies with missing context fields to simulate processing behavior when context information is incomplete. The intervention channel injects enhanced copies containing unusual combinations of fields to simulate scenarios of unusual coupling and potential risks. This triple simulation mechanism captures the differences in node behavior under different contexts and perturbation conditions.

[0088] Collect the contextual synergy factors output by the simulation processing channel to obtain the initial sensitivity scores of each structured node in the risk synergy small group, and identify high-risk nodes, including:

[0089] The contextual synergy factors output by each simulation processing channel are collected. The contextual synergy factors specifically include the structural co-occurrence index, the path conflict factor, and the sensitive node influence distribution factor.

[0090] The structural co-occurrence index measures the frequency and proportion of co-occurrence of structured nodes with contextual fields during processing. During collection, the number of times the structured node and its associated fields co-exist in the current channel output is counted, reflecting the integrity and stability of the contextual information.

[0091] Path conflict reflects the transitions and conflicts of structured nodes within a process path. This is done by tracking whether a structured node experiences abnormal transitions, branch changes, or conflicts. Data collection includes process log analysis, status code comparison, and transition node matching. The path deviation rate is calculated as a conflict indicator.

[0092] The sensitive node impact distribution factor measures the scope and intensity of a structured node's impact on other structured nodes. By analyzing the state changes, anomaly triggering times, and data transmission changes caused by the node in the processing channel, the number of nodes affected and the magnitude of the impact are quantified. Data collection methods include association rule mining, causal path tracing, and anomaly propagation analysis.

[0093] The collection process of contextual synergy factors needs to be combined with process tracing technology and completed through automated analysis tools.

[0094] The difference between the context synergy factor output by the intervention channel and the context synergy factor output by the normal channel is obtained. At the same time, the difference between the context synergy factor output by the extraction channel and the context synergy factor output by the normal channel is obtained. The obtained two-layer difference is weighted and corrected to obtain the initial sensitivity score of each structured node of the risk synergy small group, including:

[0095] ;

[0096] Among them, MGC i is the initial sensitivity score of the i-th structured node, SCI 1i is the structural co-occurrence index of the structured nodes output by the i-th intervention channel, PCT 1i is the path conflict factor of the structured node output by the i-th intervention channel, SNIS 1i is the sensitive node influence distribution factor of the structured node output by the i-th intervention channel, SCI 2i is the structural co-occurrence index of the structured node output by the i-th extraction channel, PCT 2i is the path conflict factor of the structured node output by the i-th extraction channel, SNIS 2i is the sensitive node influence distribution factor of the structured node output by the i-th extraction channel, SCI 0i is the structural co-occurrence index of the structured node output by the i-th normal channel, PCT 0i is the path conflict factor of the structured node output by the i-th normal channel, SNIS 0iis the sensitive node influence distribution factor of the structured node output by the i-th normal channel, α1 is the weighting factor of the structural co-occurrence index, α2 is the weighting factor of the path conflict factor, α3 is the weighting factor of the sensitive node influence distribution factor, i is the number of the structured node in the risk collaboration small group, i=1,2,3,...,n, n is the total number of structured nodes in the risk collaboration small group.

[0097] It should be noted that the weighting factors for the structural co-occurrence index, the path conflict factor, and the sensitive node impact distribution factor are used to dynamically adjust the weight of different types of contextual synergy factors in the initial sensitivity score. The structural co-occurrence index weighting factor controls the contribution of field co-occurrence strength in the initial sensitivity score, the path conflict factor weighting factor adjusts the contribution of process deviation behavior in the initial sensitivity score, and the sensitive node impact distribution factor weighting factor adjusts the contribution of risk spillover in the initial sensitivity score. By adjusting the values of these three weighting factors, the recognition sensitivity of different risk dimensions can be precisely controlled.

[0098] It should also be noted that there is a certain correlation between the structural co-occurrence index, path conflict factor, and sensitive node influence distribution factor, as follows:

[0099] Fluctuations in the structural co-occurrence index often precede changes in other factors. When the context field of a structured node is missing, obscured, or structurally mutated, the node's contextual information is incomplete during runtime. This structural incompleteness directly interferes with process logic, leading to path errors, process jump deviations, or rollbacks, manifesting as an increase in the path conflict factor. These structural and path perturbations form a typical causal chain relationship. An increase in the path conflict factor can have a wider impact. When a structured node behaves abnormally along a process path, it can disrupt input judgment, data transmission, or task execution at multiple downstream collaborative nodes, causing the impact of the anomaly to spread more widely. In this case, if the structured node is detected to have significantly linked perturbations to surrounding structured nodes, the score of its sensitive node impact distribution factor will increase. Furthermore, when all three factors show a simultaneous upward trend—i.e., significant structural loss, frequent path anomalies, and widespread impact—this indicates that the structured node is not only inherently anomalous, but its behavior has also materially impacted system stability and data security, indicating a closed-loop high-risk profile.

[0100] If the initial sensitivity score of a structured node is greater than or equal to the preset sensitivity score threshold, the structured node is judged to be a high-risk node.

[0101] If the initial sensitivity score of a structured node is less than the preset sensitivity score threshold, the structured node is judged to be a low-risk node.

[0102] Generate a temporary freeze instruction for the project, separate the execution path of the high-risk node from the main processing channel, build a local execution environment, and perform a high-risk review on the data in the high-risk node. If the review result is a misjudgment, the execution path of the high-risk node will be returned to the main processing channel.

[0103] Conduct high-risk review on data in high-risk nodes, including:

[0104] Generate a temporary freeze instruction for the project, separate the execution path of the high-risk node from the main processing channel, call the containerized execution unit, build a local execution environment, which is an execution environment without downstream write permission, and subtract the initial sensitivity score of the high-risk node from the sensitivity score threshold to obtain the initial sensitivity score difference of the high-risk node. Input the mapping of the initial sensitivity score difference and the local execution environment resource configuration index pre-stored in the database to obtain the local execution environment resource configuration index of the high-risk node, and then build the local execution environment based on the local execution environment resource configuration index. The specific construction process includes:

[0105] The containerized execution unit is called, which is a Docker container in an embodiment of the present invention. Based on the local execution environment resource configuration index, a corresponding resource configuration template is matched. The resource configuration template defines basic resource quotas such as CPU, memory, bandwidth, and I / O priority, and the basic resource quotas in the local execution environment are dynamically adjusted based on the resource configuration template.

[0106] The local execution environment resource allocation index is a numerical value. The larger the value, the higher the resource allocation required for the high-risk node in order to accurately and efficiently complete the review task. By matching the corresponding resource allocation template through the mapping relationship, the containerized execution unit (such as Docker) is guided to dynamically allocate the appropriate amount of resources, realizing on-demand elastic allocation of resources and avoiding resource waste or shortage.

[0107] It should be noted that, in the embodiment of the present invention, a mapping set of local execution environment resource configuration index-resource configuration template is pre-stored in the database.

[0108] For example, a matching resource configuration template might include 1 CPU core, 512MB of memory, 10Mbps of bandwidth, and low I / O priority. During deployment, the containerized execution unit is invoked and, through the docker update API, an isolated local execution environment is created based on this matching resource configuration template. During container creation, resource parameters such as the container's CPU quota, memory limit, network bandwidth limit, and I / O priority are set.

[0109] In the local execution environment, the contextual synergy factor differences of the data in the high-risk nodes are reanalyzed to obtain the review sensitivity score.

[0110] If the reviewed sensitivity score of the high-risk node is still greater than or equal to the sensitivity score threshold, the review result is no misjudgment.

[0111] If the review sensitivity score of the high-risk node is less than the sensitivity score threshold, the review result is a misjudgment, and the execution path of the high-risk node is returned to the main processing channel.

[0112] If the review result is no misjudgment, the configuration weight of the high-risk node is adjusted and applied to the linkage calculation module.

[0113] Adjust the configuration weight of the high-risk node, including:

[0114] The initial sensitivity score and the reviewed sensitivity score of the high-risk node are extracted, and the average is taken as the comprehensive sensitivity score of the high-risk node.

[0115] Based on the comprehensive sensitivity score of the high-risk node, the mapping set of comprehensive sensitivity score-configuration weight adjustment factor pre-stored in the database is input, and mapping matching is performed to obtain the configuration weight adjustment factor of the high-risk node. The configuration weight adjustment factor is a parameter indicator used to quantify the authority adjustment range of the high-risk node.

[0116] Adjusting the configuration weight of the high-risk node also includes adjusting the configuration weights of each risk-related node corresponding to the high-risk node. The specific processing conditions are:

[0117] Obtain other structured nodes in the risk collaboration subgroup where the high-risk node is located, extract metadata, and parse to obtain the call fields of other structured nodes.

[0118] Cross-compare the call fields of other structured nodes with the fields in high-risk nodes to identify structured nodes that have associated call relationships with high-risk nodes. These are recorded as risk-associated nodes. The associated call relationships include field reading, downstream jumps, and permission writing.

[0119] The number of associated call relationships between each risk-associated node and the high-risk node is obtained, and the mapping of the number of associated call relationships-configuration weight adjustment factor attenuation ratio pre-stored in the database is input for centralized mapping and matching to obtain the configuration weight adjustment factor attenuation ratio of each risk-associated node, and the configuration weight adjustment factor attenuation ratio of each risk-associated node is interacted with the configuration weight adjustment factor of the high-risk node. In an embodiment of the present invention, the interaction is specifically multiplication to obtain the associated configuration weight adjustment factor of each risk-associated node.

[0120] Applied in the linkage calculation module, specifically including:

[0121] Extract the linkage computing module numbers of high-risk nodes and each risk-related node, and send the configuration weight adjustment factor of the high-risk node and the associated configuration weight adjustment factor of each risk-related node to the corresponding linkage computing module in the form of configuration instruction packets. After receiving the configuration instruction packet, the linkage computing module maps the configuration instruction to the resource scheduling control parameter configuration table, and matches the resource scheduling control parameters of the linkage computing module, including the disk read rate limit value, the disk write rate limit value, the upper limit of available memory, and the CPU time quota limit value.

[0122] The disk read rate limit is used to limit the maximum rate at which the linked computing module reads data from the storage device. This prevents high-risk nodes from overloading the storage system due to frequent or large-scale reads, ensures stable overall system IO performance, and prevents bottlenecks and response delays.

[0123] The disk write rate limit controls the speed at which linked computing modules write data to storage devices. This prevents abnormal write operations from causing excessive disk write pressure, ensures data integrity and storage performance, and avoids write congestion that may cause performance degradation in other modules.

[0124] The memory cap can be used to limit the maximum allocatable memory resources of linked computing modules, preventing a single high-risk node from occupying too much memory and causing overall system memory shortages. This ensures the stable operation of other computing modules and processes and avoids system crashes or sharp performance degradation caused by insufficient memory.

[0125] The CPU time quota limit is used to limit the CPU time that the linked computing modules are allowed to use per unit time. This prevents high-risk nodes from occupying the CPU for a long time, ensures fair competition for CPU resources among multiple modules, and prevents single-point overload from causing overall computing performance bottlenecks and slowing system response.

[0126] The linked computing module writes resource scheduling control parameters and calls various management APIs to execute resource scheduling control parameters.

[0127] For example, there is a high-risk node N17 with a comprehensive sensitivity score of 0.87. The configuration weight adjustment factor obtained by mapping and matching is 1.5. At the same time, the high-risk node N17 has two risk-associated nodes N16 and N18, and the calculated associated configuration weight adjustment factors are 0.8 and 0.6 respectively.

[0128] Generate three configuration instruction packages, including:

[0129] Configuration instruction packet 1 is sent to N17's linkage computing module C7. After receiving it, C7 maps and matches the configuration weight adjustment factors to obtain resource scheduling control parameters. The disk read rate limit is 120 MB / s, the disk write rate limit is 90 MB / s, the available memory limit is 1024 MB, and the CPU time quota limit is 400 ms / s.

[0130] Configuration instruction packet 2 is sent to the linkage computing module C9 of N16. After receiving it, C9 maps and matches the configuration weight adjustment factors to obtain resource scheduling control parameters. The disk read rate limit is 100 MB / s, the disk write rate limit is 70 MB / s, the available memory limit is 768 MB, and the CPU time quota limit is 300 ms / s.

[0131] Configuration instruction packet 3 is sent to the linkage computing module C5 of N18. After receiving it, C5 maps and matches the configuration weight adjustment factor to obtain resource scheduling control parameters. The disk read rate limit is 80MB / s, the disk write rate limit is 50MB / s, the available memory limit is 640MB, and the CPU time quota limit is 250ms / s.

[0132] In this embodiment, if Figure 2 As shown, the present invention provides an intelligent management system for drug clinical projects, comprising:

[0133] The collaborative small group division module is used to map the data units uploaded by the drug clinical project into structured nodes, construct a dynamic data co-evolution graph, and divide the dynamic data co-evolution graph into several collaborative small groups.

[0134] The high-risk node identification module is used to identify small risk collaborative groups, construct perturbed data copies of the original data of each structured node in the risk collaborative small group, inject the original data and perturbed data copies of each structured node into the simulation processing channel in parallel, collect the contextual collaboration factors output by the simulation processing channel, obtain the initial sensitivity score of each structured node in the risk collaborative small group, and judge and identify high-risk nodes.

[0135] The high-risk node review module is used to generate temporary freeze instructions for the project, separate the execution path of the high-risk node from the main processing channel, build a local execution environment, and perform high-risk review on the data in the high-risk node. If the review result is a misjudgment, the execution path of the high-risk node will be returned to the main processing channel.

[0136] The high-risk node configuration weight module is used to adjust the configuration weight of the high-risk node when the review result is not a misjudgment, and apply it to the linkage calculation module.

[0137] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0138] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all details in detail, nor do they limit the invention to specific implementation methods. Obviously, many modifications and changes can be made based on the contents of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can better understand and utilize the present invention. As long as they do not deviate from the structure of the present invention or exceed the scope defined by the present invention, they should fall within the scope of protection of the present invention.

Claims

1. An intelligent management method for drug clinical projects, characterized in that: include: Map the data units uploaded by the drug clinical project into structured nodes, construct a dynamic data co-evolution graph, and divide the dynamic data co-evolution graph into several collaborative small groups; Identify small risk collaboration groups, construct perturbed data copies of the original data of each structured node in the risk collaboration group, inject the original data and perturbed data copies of each structured node into the simulation processing channel in parallel, collect the contextual collaboration factors output by the simulation processing channel, obtain the initial sensitivity score of each structured node in the risk collaboration group, and identify high-risk nodes; Generate a temporary freeze instruction for the project, separate the execution path of the high-risk node from the main processing channel, build a local execution environment, and conduct a high-risk review of the data in the high-risk node. If the review result is a false positive, return the execution path of the high-risk node to the main processing channel; If the review result is no misjudgment, the configuration weight of the high-risk node is adjusted and applied to the linkage calculation module.

2. The intelligent management method for drug clinical projects according to claim 1, characterized in that: The data units uploaded by the drug clinical project are mapped into structured nodes, a dynamic data co-evolution graph is constructed, and the dynamic data co-evolution graph is divided into several collaborative small groups. The specific process is as follows: The raw data units uploaded during the collection, transmission and processing of drug clinical projects are structured and mapped into structured nodes; A dynamic data co-evolution graph is constructed based on the historical data interaction records of structured nodes, where the edges between structured nodes in the dynamic data co-evolution graph represent the co-evolution relationship; Based on the graph structure partitioning algorithm, highly co-evolved structured nodes are clustered into small co-evolving groups. The small co-evolving group is a set of data nodes formed based on the co-evolution relationship between nodes in the dynamic data co-evolution graph.

3. The intelligent management method for drug clinical projects according to claim 1, characterized in that: The specific process of identifying risk collaborative small groups is as follows: Track the average sensitivity score of each collaborative group in the historical clinical project process, extract the average score threshold from the database, and if the average sensitivity score of a collaborative group in the historical clinical project process exceeds the preset average score threshold, it will be identified as a collaborative group with potential risks and recorded as a risky collaborative group; If the average sensitivity score of a collaborative subgroup in the historical clinical project process is less than or equal to the average score threshold, it will not be identified as a collaborative subgroup with potential risks.

4. The intelligent management method for drug clinical projects according to claim 1, characterized in that: The method of constructing a perturbed data copy of the original data of each structured node in the risk collaborative small group and injecting the original data of each structured node and the perturbed data copy into the simulation processing channel in parallel specifically includes: Obtain the average score difference between the average sensitivity score of each risk group in the historical clinical project process and the average score threshold, and perform mapping matching based on the average score difference to obtain the preset perturbation parameters of each risk group; The preset disturbance parameters include a weakened copy parameter and an enhanced copy parameter. The weakened copy parameter is specifically a context field hiding ratio, and the enhanced copy parameter is specifically a supplementary ratio of abnormal combination fields. For the original data of each structured node in a risk collaboration small group, based on the preset perturbation parameters, a perturbed data copy is constructed, including a weakened copy and an enhanced copy; The original data, weakened copies and enhanced copies of each structured node in the risk collaborative small group are injected into the simulation processing channel in parallel. The simulation processing channel includes a normal simulation processing channel, an extracted simulation processing channel and an intervention simulation processing channel. The original data is injected into the normal simulation processing channel for use as a benchmark reference, the weakened copy is injected into the extracted simulation processing channel for simulating the channel processing behavior when the context field is missing, and the enhanced copy is injected into the intervention simulation processing channel for simulating the channel processing behavior when the abnormal combination field exists.

5. The intelligent management method for drug clinical projects according to claim 1, characterized in that: The contextual synergy factor output by the acquisition simulation processing channel is obtained to obtain the initial sensitivity score of each structured node in the risk synergy small group, and to identify high-risk nodes, specifically including: Collecting contextual synergy factors output by each simulation processing channel, wherein the contextual synergy factors specifically include a structural co-occurrence index, a path conflict factor, and a sensitive node influence distribution factor; The difference between the contextual synergy factor output by the intervention channel and the contextual synergy factor output by the normal channel is obtained. At the same time, the difference between the contextual synergy factor output by the extraction channel and the contextual synergy factor output by the normal channel is obtained. The obtained two-layer difference is weighted and corrected to obtain the initial sensitivity score of each structured node of the risk synergy small group after coupling. If the initial sensitivity score of a structured node is greater than or equal to the preset sensitivity score threshold, the structured node is judged to be a high-risk node; If the initial sensitivity score of a structured node is less than the preset sensitivity score threshold, the structured node is judged to be a low-risk node.

6. The intelligent management method for drug clinical projects according to claim 1, characterized in that: The high-risk review of the data in the high-risk nodes specifically includes: Generate a temporary freeze instruction for the project, separate the execution path of the high-risk node from the main processing channel, call the containerized execution unit, and build a local execution environment, where the local execution environment is an execution environment without downstream write permission. Based on the initial sensitivity score difference between the initial sensitivity score of the high-risk node and the sensitivity score threshold, input the initial sensitivity score difference pre-stored in the database - the local execution environment resource configuration index mapping to obtain the local execution environment resource configuration index of the high-risk node, and build the local execution environment according to the local execution environment resource configuration index; In the local execution environment, the contextual synergy factor differences of the data in the high-risk nodes are reanalyzed to obtain the review sensitivity score; If the reviewed sensitivity score of the high-risk node is still greater than or equal to the sensitivity score threshold, the review result is no misjudgment; If the review sensitivity score of the high-risk node is less than the sensitivity score threshold, the review result is a misjudgment, and the execution path of the high-risk node is returned to the main processing channel.

7. The intelligent management method for drug clinical projects according to claim 1, characterized in that: The adjusting of the configuration weight of the high-risk node specifically includes: Extract the initial sensitivity score and the reviewed sensitivity score of the high-risk node, and take the average as the comprehensive sensitivity score of the high-risk node; Based on the comprehensive sensitivity score of the high-risk node, the mapping set of comprehensive sensitivity score-configuration weight adjustment factor pre-stored in the database is input, and mapping matching is performed to obtain the configuration weight adjustment factor of the high-risk node. The configuration weight adjustment factor is a parameter indicator used to quantify the authority adjustment range of the high-risk node.

8. The intelligent management method for drug clinical projects according to claim 7, characterized in that: The adjustment of the configuration weight of the high-risk node also includes adjusting the configuration weights of each risk-related node corresponding to the high-risk node. The specific processing conditions are: Obtain other structured nodes in the risk collaboration subgroup where the high-risk node is located, extract metadata, and parse to obtain the call fields of other structured nodes; Cross-compare the call fields of other structured nodes with the fields in the high-risk nodes to identify structured nodes that have associated call relationships with the high-risk nodes and record them as risk-associated nodes. The associated call relationships include field reading, downstream jumps, and permission writing. Obtain the number of associated call relationships between each risk-associated node and the high-risk node, and input the mapping of the number of associated call relationships-configuration weight adjustment factor attenuation ratio pre-stored in the database to perform mapping matching to obtain the configuration weight adjustment factor attenuation ratio of each risk-associated node, and interact the configuration weight adjustment factor attenuation ratio of each risk-associated node with the configuration weight adjustment factor of the high-risk node to obtain the associated configuration weight adjustment factor of each risk-associated node.

9. The intelligent management method for drug clinical projects according to claim 1, characterized in that: The application in the linkage calculation module specifically includes: Extract the linkage calculation module numbers of the high-risk node and each risk-related node, and send the configuration weight adjustment factor of the high-risk node and the associated configuration weight adjustment factor of each risk-related node to the corresponding linkage calculation module in the form of a configuration instruction packet. After receiving the configuration instruction packet, the linkage calculation module maps the configuration instruction to the resource scheduling control parameter configuration table, and matches the resource scheduling control parameters of the linkage calculation module, including the disk read rate limit value, the disk write rate limit value, the upper limit of available memory, and the CPU time quota limit value; The linked computing module writes resource scheduling control parameters and calls various management APIs to execute resource scheduling control parameters.

10. A system using the intelligent management method for drug clinical projects according to any one of claims 1 to 9, characterized in that: The collaborative small group division module is used to map the data units uploaded by the drug clinical project into structured nodes, build a dynamic data co-evolution graph, and divide the dynamic data co-evolution graph into several collaborative small groups; The high-risk node identification module is used to identify risk-cooperating small groups, construct perturbed data copies of the original data of each structured node in the risk-cooperating small group, inject the original data and perturbed data copies of each structured node into the simulation processing channel in parallel, collect the contextual coordination factors output by the simulation processing channel, obtain the initial sensitivity score of each structured node in the risk-cooperating small group, and identify high-risk nodes; The high-risk node review module is used to generate a temporary freeze instruction for the project, separate the execution path of the high-risk node from the main processing channel, build a local execution environment, and perform a high-risk review on the data in the high-risk node. If the review result is a false positive, the execution path of the high-risk node is returned to the main processing channel; The high-risk node configuration weight module is used to adjust the configuration weight of the high-risk node when the review result is not a misjudgment, and apply it to the linkage calculation module.

Images