A Method and System for Managing Clinical Nursing Data in Anesthesiology

Abstract

This invention relates to the technical field of anesthesia nursing data processing, specifically including a method and system for managing clinical nursing data in anesthesiology. The method includes: collecting clinical nursing data from anesthesiology; configuring first and second data mapping tables based on static and dynamic features according to the platform and application ends; generating a nursing knowledge graph; acquiring features of adjacent events; and conducting multi-scale risk assessment combined with dynamic threshold alerts. This invention solves the technical problem of insufficient consideration of the temporal correlation and synergistic effects of events during anesthesia, making it impossible to dynamically adjust risk assessment standards for different anesthesia stages and easily leading to misjudgments. By configuring first and second data mapping tables and combining the time entropy corresponding to the time interval of adjacent events and the support degree of association rules, it achieves hierarchical management and precise mapping of static and dynamic features in anesthesiology clinical nursing data. It dynamically adjusts risk thresholds based on the temporal features of different anesthesia stages, significantly improving the accuracy and timeliness of risk assessment.

Description

Technical Field

[0001] This invention relates to the technical field of anesthesia nursing data processing, specifically to a method and system for managing clinical nursing data in anesthesiology. Background Technology

[0002] The anesthesiology department generates heterogeneous data from multiple sources, including basic patient information, anesthesia protocol data, and anesthesia monitoring data. These data contain key information such as the patient's individualized anesthesia response patterns and risk evolution characteristics. However, the anesthesia process is highly dynamic and strongly time-dependent. From the induction and maintenance of anesthesia to the recovery, even small data fluctuations can trigger risks.

[0003] Currently, the management of clinical nursing data in anesthesiology lacks a unified mapping and fusion mechanism, resulting in prominent data silos. In addition, risk assessment often relies on single indicator thresholds, failing to consider the temporal correlation and synergistic effects between anesthetic events. Furthermore, risk thresholds cannot be dynamically adjusted with the stage of anesthesia, which can easily lead to misjudgments or delayed responses. This makes it difficult to adapt to individual patient differences and changes in clinical scenarios, thus hindering the intelligent development and risk control capabilities of clinical nursing in anesthesiology.

[0004] In summary, existing technologies have technical problems such as not fully considering the temporal correlation and synergistic effects of various events during anesthesia, being unable to dynamically adjust risk assessment criteria for different stages of anesthesia, and being prone to misjudgment of risks. Summary of the Invention

[0005] This application provides a method and system for managing clinical nursing data in anesthesiology, aiming to solve the technical problems in the prior art that do not fully consider the temporal correlation and synergistic effects of various events during anesthesia, cannot dynamically adjust risk assessment standards for different stages of anesthesia, and are prone to risk misjudgment.

[0006] In view of the above problems, the technical solution to achieve the present application is as follows:

[0007] This application provides a method for managing clinical nursing data in anesthesiology. The method includes: collecting clinical nursing data in anesthesiology, including patient basic information, anesthesia protocol data, and anesthesia monitoring data; configuring a first data mapping table based on static and dynamic features according to the edge computing node corresponding to the platform; configuring a second data mapping table based on static and dynamic features according to the visualization unit corresponding to the application; generating a nursing knowledge graph based on the clinical nursing data in anesthesiology, combined with the first and second data mapping tables; simultaneously, obtaining the time entropy and association rule support corresponding to the time intervals of adjacent events defined by the nursing knowledge graph through a sliding time window; performing partial least squares discriminant analysis using the time entropy and association rule support corresponding to the time intervals of adjacent events defined by the nursing knowledge graph, using the dose change rate as an input feature, to perform multi-scale assessment of potential risks, and providing risk alerts in conjunction with dynamic risk thresholds.

[0008] Preferably, temporal feature analysis is performed using event tags, which include the anesthesia induction phase, anesthesia maintenance phase, and awakening phase; dynamic risk thresholds are configured based on the temporal relationships corresponding to the event tags.

[0009] Preferably, an attention mechanism is used to process outliers in the anesthesiology clinical nursing data, and federated learning alignment is performed on heterogeneous data with inconsistent device identifiers; the anesthesiology clinical nursing data is mapped into high-dimensional vectors through the first data mapping table and the second data mapping table to generate the nursing knowledge graph.

[0010] Preferably, an active learning-driven dynamic graph update mechanism is set up to filter high-conflict nodes; using a graph neural network, a conflict association feature set is obtained through the temporal association of the nursing knowledge graph; at the high-conflict nodes, the edge weights and node attributes of the nursing knowledge graph are dynamically adjusted in combination with the conflict association feature set.

[0011] Preferably, the node with the highest uncertainty in the nursing knowledge graph is selected as the sample to be labeled using Bayesian posterior probability; the sample to be labeled is grouped, and the group type is associated with the event label; the edge weights and node attributes of the nursing knowledge graph are updated using local parameters based on the grouped sample to be labeled.

[0012] Preferably, a multi-head attention mechanism is used to fuse the temporal entropy corresponding to the time interval with the support of the association rule to configure the temporal dependency vector of the key risk event; through the temporal dependency vector of the key risk event, a path search is performed in the nursing knowledge graph to formulate a risk propagation path; when there is a node with a sudden drop in the anesthesia depth index in the risk propagation path, a dynamic compensation mechanism is triggered.

[0013] Preferably, static features are mapped to the local storage area of ​​the edge computing node; a hierarchical index is established according to the time series based on dynamic features, and the data update frequency and data transmission strategy are configured; a data synchronization protocol between the edge computing node and the cloud server is established using the data transmission strategy, and bidirectional differential synchronization is used to transmit only the changed data blocks, while recording the data version number.

[0014] Preferably, a sliding time window is defined to traverse the event nodes in the nursing knowledge graph; the time interval between adjacent events under the constraints of the nursing knowledge graph is determined by time series, and the time entropy corresponding to the time interval is obtained; the co-occurrence probability of events is quantified by the association rule support of adjacent events under the constraints of the nursing knowledge graph, and an event association strength matrix is ​​formed.

[0015] Preferably, the structured data in the nursing knowledge graph is spatiotemporally aligned with the image and audio data during the anesthesia process, and a generative adversarial network is used to formulate virtual risk scenarios; based on the virtual risk scenarios, training data is collected, and the training data is used to optimize the nursing knowledge graph.

[0016] In another aspect, this application provides an anesthesiology clinical nursing data management system, wherein the system includes: a data collection module for collecting anesthesiology clinical nursing data including patient basic information, anesthesia protocol data, and anesthesia monitoring data; a data mapping module for configuring a first data mapping table based on static and dynamic features according to the edge computing node corresponding to the platform end; and configuring a second data mapping table based on static and dynamic features according to the visualization unit corresponding to the application end; a nursing knowledge graph generation module for generating a nursing knowledge graph based on the anesthesiology clinical nursing data and in combination with the first and second data mapping tables; a sliding analysis module for obtaining the time entropy and association rule support corresponding to the time interval of adjacent events under the limitation of the nursing knowledge graph through a sliding time window; and a risk alert module for performing partial least squares discriminant analysis based on the time entropy and association rule support corresponding to the time interval of adjacent events under the limitation of the nursing knowledge graph, using the dose change rate as an input feature to perform multi-scale assessment of potential risks, and providing risk alerts in combination with dynamic risk thresholds.

[0017] In summary, one or more technical solutions provided in this application, by configuring the first and second data mapping tables and combining the time entropy and association rule support corresponding to the time interval of adjacent events, can conduct multi-scale assessment of potential anesthesia risks, realize hierarchical management and precise mapping of static and dynamic features in anesthesiology clinical nursing data, and dynamically adjust risk thresholds according to the temporal characteristics of different anesthesia stages, significantly improving the accuracy and timeliness of risk assessment. Attached Figure Description

[0018] Figure 1 This application provides a flowchart illustrating a clinical nursing data management method for anesthesiology.

[0019] Figure 2 This application provides a schematic diagram of the structure of an anesthesiology clinical nursing data management system.

[0020] Explanation of reference numerals in the attached diagram: Data collection module M100, data mapping module M200, nursing knowledge graph generation module M300, sliding analysis module M400, risk alert module M500. Detailed Implementation

[0021] Example 1: The present application will be described in detail below with reference to the accompanying drawings, as follows... Figure 1 As shown, this application provides a method for managing clinical nursing data in anesthesiology, wherein the method includes:

[0022] S1: Collect anesthesiology clinical nursing data, including patient basic information, anesthesia protocol data, and anesthesia monitoring data; S2: Configure a first data mapping table based on static and dynamic features according to the corresponding edge computing node on the platform; Configure a second data mapping table based on static and dynamic features according to the corresponding visualization unit on the application side.

[0023] Specifically, basic patient information includes static data such as age, gender, weight, and medical history, which are relatively stable during anesthesia and form the basis for developing anesthesia plans. Anesthesia plan data includes the selection, dosage, and administration time of anesthetic drugs. Anesthesia monitoring data consists of dynamic data generated in real time during anesthesia, such as heart rate, blood pressure, blood oxygen saturation, and anesthesia depth index, which reflect the physiological changes of the patient under anesthesia and have high frequency and strong temporal sequence.

[0024] Edge computing nodes refer to computing devices deployed close to the data source, such as monitors and anesthesia machines in the operating room. They can perform preliminary processing on the collected data, reduce data transmission latency, and improve the real-time performance of data processing. Visualization units are terminal devices used to view and analyze nursing data, such as anesthesia monitoring screens and mobile nursing workstations. They need to present complex data in an intuitive and easy-to-understand way. Static features and dynamic features are two basic attributes of data. Static features, such as the patient's past medical history and allergy history, remain basically unchanged during anesthesia. Dynamic features, such as real-time physiological monitoring indicators, change continuously as anesthesia progresses.

[0025] Implementation steps: Collecting clinical nursing data in the anesthesiology department, including basic patient information, anesthesia protocol data, and anesthesia monitoring data, is the foundation of the entire data management method. Basic patient information provides the individualized context of the patient for subsequent anesthesia protocol development and risk assessment. For example, a 60-year-old male patient weighing 70 kg with a history of hypertension will directly affect the selection and dosage of anesthetic drugs.

[0026] The anesthesia protocol data clarifies the specific anesthesia methods and drug administration plans. For example, if general anesthesia is chosen, drugs such as propofol and remifentanil are used, with initial doses of X mg / kg and Y μg / kg, respectively. These data guide the implementation of anesthesia. Anesthesia monitoring data are key data generated in real time during anesthesia. Taking heart rate as an example, the normal heart rate of an adult is 60-100 beats / minute. During the anesthesia induction phase, the patient's heart rate decreases due to the effects of anesthetic drugs. Real-time monitoring of heart rate changes can promptly detect potential circulatory system risks.

[0027] Based on the corresponding edge computing nodes on the platform, a first data mapping table is configured with static and dynamic features. The raw data collected by the edge computing nodes is classified and processed according to static and dynamic features. For example, the patient's past medical history (static features) is stored in the local storage area of ​​the edge computing node. At the same time, a hierarchical index is established for real-time dynamic feature data such as heart rate and blood pressure according to time series. The data update frequency is configured to be updated once per minute, and a data transmission strategy is formulated. Data is only transmitted to the cloud server when the data change exceeds the set threshold, which reduces the amount of data transmission and transmission latency.

[0028] Based on the corresponding visualization unit on the application side, a second data mapping table is configured with static and dynamic features. This allows medical staff to clearly distinguish between static and dynamic information when viewing data, improving data readability and usability. For example, the visualization interface displays the patient's basic information (static features) in a table format, and displays dynamic monitoring data such as heart rate and blood pressure in real time in a line graph format, facilitating medical staff to quickly obtain key information and supporting clinical decision-making. The above steps provide a structured data foundation for subsequent risk assessment, ensuring the efficiency and accuracy of data processing.

[0029] S3: Based on the anesthesiology clinical nursing data, and combined with the first data mapping table and the second data mapping table, a nursing knowledge graph is generated; S4: Simultaneously, by using a sliding time window, the time entropy and association rule support corresponding to the time intervals of adjacent events under the constraints of the nursing knowledge graph are obtained; S5: Partial least squares discriminant analysis is performed using the time entropy and association rule support corresponding to the time intervals of adjacent events under the constraints of the nursing knowledge graph, with the dose change rate as the input feature, to conduct multi-scale assessment of potential risks, and risk alerts are provided in conjunction with dynamic risk thresholds.

[0030] Specifically, a nursing knowledge graph is a graph-based data model used to represent entities (such as patients, drugs, physiological indicators, etc.) and their relationships in anesthesiology clinical nursing data. It can integrate scattered data into a semantically related knowledge network, facilitating in-depth mining of the information contained in the data. The time entropy corresponding to the time interval is a concept in information theory, which is used here to quantify the uncertainty of the time interval between adjacent events in the nursing knowledge graph. The more random the time interval, the higher the entropy value, indicating that the time regularity of the event occurrence is weaker.

[0031] Association rule support measures the frequency of co-occurrence between different events. Higher support indicates a greater probability of two events occurring simultaneously and a stronger association between them. Partial least squares discriminant analysis can classify and predict potential risks when processing multivariate data by finding the optimal linear combination between independent variables (such as dose change rate) and dependent variables (such as risk level). It is particularly suitable for complex data scenarios where there is multicollinearity among variables. Dynamic risk thresholds refer to risk assessment criteria that are automatically adjusted according to different stages of the anesthesia process (such as induction, maintenance, and recovery stages) and the patient's real-time physiological state. Compared with fixed thresholds, they are more adaptable to the dynamic changes in the anesthesia process and improve the accuracy and timeliness of risk warnings.

[0032] Execution steps: The process of generating a nursing knowledge graph based on anesthesiology clinical nursing data and combining the first and second data mapping tables is a process of deep integration and semantic association of classified static and dynamic data. For example, in the first data mapping table, the patient's past medical history (static features) and real-time heart rate data (dynamic features) are associated as edges between "patient nodes" and "heart rate monitoring nodes" in the nursing knowledge graph after preliminary processing by edge computing nodes. The weight of the edges can be dynamically adjusted according to the relevance of the data.

[0033] Meanwhile, the second data mapping table combines the visualization needs of the application with data characteristics, so that the knowledge graph not only stores the raw data, but also stores the intermediate results that prepare for visualization. For example, it converts the patient's blood pressure change trend data into a format suitable for display as a line graph on the monitoring screen, thus transforming the raw data into a knowledge structure with semantic association and operability, which is a key function and provides a foundation for subsequent in-depth analysis and risk assessment.

[0034] By using a sliding time window, the temporal entropy and association rule support corresponding to the time intervals of adjacent events within a nursing knowledge graph are obtained. A sliding time window is a data processing technique that traverses the event sequence in a knowledge graph sequentially according to a set time length and step size. This captures the changes in time intervals and associations of events at different time scales. For example, setting the sliding time window to 5 minutes and the step size to 1 minute, when analyzing the event sequence during the anesthesia induction phase, the time intervals and their changing patterns between adjacent events (such as drug injection events and changes in patient physiological indicators) within each 5-minute window can be obtained. The temporal entropy corresponding to the time intervals can be calculated to assess the uncertainty of event occurrence time. Simultaneously, the co-occurrence frequency between events can be counted, and the association rule support can be calculated to quantify the association strength between events, forming an event association strength matrix.

[0035] For example, within 5 minutes after the injection of a certain anesthetic drug, the support of the association rule between the events of decreased heart rate and decreased blood pressure reached 0.8 (meaning that these two events occur simultaneously in 80% of cases). This indicates a strong correlation between the two events. Using the dose change rate as the input feature, the potential risk is assessed on a multi-scale basis through partial least squares discriminant analysis. The dose change rate refers to the magnitude of change in the dose of the anesthetic drug per unit time, which is an important indicator reflecting the depth of anesthesia and the patient's physiological response. Furthermore, in the partial least squares discriminant analysis model, the dose change rate is used as the independent variable, and the risk level (such as low risk, medium risk, and high risk) is used as the dependent variable. By training the model with historical data, a mapping relationship between the dose change rate and the risk level can be established.

[0036] By combining dynamic risk thresholds for risk alerts, which are adjusted according to different anesthesia stages and the patient's real-time condition, and further, during the anesthesia induction phase, the patient's physiological system is more sensitive to the drug, and the dynamic risk threshold will be relatively low at this time. When the dose change rate exceeds the threshold of the anesthesia induction phase, a risk alert is immediately issued to remind medical staff to pay attention to changes in the patient's physiological indicators. This achieves accurate and dynamic assessment and timely warning of potential risks during anesthesia, significantly improving the accuracy and timeliness of risk assessment.

[0037] Furthermore, by incorporating dynamic risk thresholds for risk alerts, the method described in this application includes:

[0038] Temporal feature analysis is performed using event tags, which include the anesthesia induction phase, anesthesia maintenance phase, and recovery phase; dynamic risk thresholds are configured based on the temporal relationships corresponding to the event tags.

[0039] Specifically, event tagging is an identifier used to mark and classify different stages of the anesthesia process, dividing it into the induction, maintenance, and recovery phases. Each phase has its unique physiological and pharmacological characteristics. The induction phase is from the time the patient begins receiving anesthetic drugs until they enter a state of unconsciousness; during this phase, the patient's physiological indicators change rapidly, and they are highly sensitive to drugs. The maintenance phase is the period during surgery where the patient's anesthetic state is maintained; this phase requires continuous monitoring and adjustment of drug dosages to maintain the depth of anesthesia. The recovery phase is the period when the patient regains consciousness from the anesthetic state; this phase requires close observation of the patient's recovery and timely management of any potential complications.

[0040] Temporal feature analysis is used to study the distribution and variation patterns of these event labels over time, revealing the temporal characteristics of different anesthesia stages; dynamic risk thresholds are risk assessment criteria that can be automatically adjusted according to temporal relationships. Unlike fixed risk thresholds, dynamic risk thresholds can be adaptively adjusted according to different anesthesia stages and the patient's real-time status, more accurately reflecting the risk level.

[0041] Execution steps: The process of performing time-series feature analysis using event tags involves dividing the anesthesia process into stages such as induction, maintenance, and recovery, and performing time-series statistics and analysis on the events in each stage. For example, by labeling each event with the corresponding stage tag in the nursing knowledge graph, the frequency of occurrence, duration, and time interval between events in each stage can be statistically analyzed. For instance, the average time interval between the event tags "drug injection" and "heart rate decrease" in the anesthesia induction stage is 3 minutes, while this interval extends to 5 minutes in the maintenance stage, indicating that there are significant differences in the time-series characteristics of events in different stages.

[0042] By analyzing temporal characteristics, we can understand the risk evolution patterns at each stage, providing data support for configuring dynamic risk thresholds. The process of configuring dynamic risk thresholds based on the temporal relationships corresponding to event tags involves setting differentiated risk assessment criteria according to different temporal characteristics. For example, during anesthesia induction, due to significant fluctuations in patient physiological indicators, the dynamic risk threshold for the risk event of a decreased heart rate can be set to trigger a risk alert when the heart rate drops below 50 beats per minute for more than 2 minutes. During maintenance, when physiological indicators are relatively stable, a risk alert is only triggered when the heart rate drops below 55 beats per minute for more than 3 minutes. This dynamic adjustment mechanism makes risk assessment more aligned with actual clinical scenarios, effectively reducing false alarm and false negative rates, and improving the timeliness and accuracy of risk warnings. In these steps, the results of temporal characteristic analysis are closely linked to risk assessment, enabling intelligent and dynamic adjustment of risk assessment standards and significantly improving the risk control capabilities of anesthesiology clinical nursing.

[0043] Furthermore, based on the aforementioned anesthesiology clinical nursing data, and in conjunction with the first data mapping table and the second data mapping table, a nursing knowledge graph is generated. The method of this application also includes:

[0044] An attention mechanism is used to process outliers in the anesthesiology clinical nursing data, and federated learning alignment is performed on heterogeneous data with inconsistent device identifiers. The anesthesiology clinical nursing data is mapped into high-dimensional vectors through the first data mapping table and the second data mapping table to generate the nursing knowledge graph.

[0045] Specifically, the attention mechanism is an algorithm that simulates the human attention allocation pattern, which can automatically focus on key features while ignoring unimportant information when processing data. In anesthesiology clinical nursing data, outliers are caused by equipment errors, individual patient differences, or sudden situations, such as momentary erroneous readings in heart rate monitoring (such as a normal heart rate of 90 beats / minute suddenly jumping to 150 beats / minute).

[0046] Heterogeneous data with inconsistent device identifiers refers to data from different devices or systems. Due to differences in device models and data formats, the data cannot be directly compared or fused. For example, blood pressure data from monitor A is in mmHg, while that from monitor B is in kPa. Federated learning is a distributed machine learning method that allows multiple devices or systems to jointly train a model by exchanging encrypted model parameters without sharing the original data, thereby protecting data privacy and security. High-dimensional vectors are a multidimensional data representation that can map complex data features to a high-dimensional space, making the inherent structure and relationships of the data easier to analyze and mine.

[0047] Execution steps: The process of using an attention mechanism to process outliers in anesthesiology clinical nursing data involves automatically identifying and focusing on key feature points in the data through algorithms, thereby effectively detecting outliers. For example, when processing heart rate monitoring data, the attention mechanism can automatically focus on the trend and pattern of heart rate changes. When an instantaneous reading that deviates significantly from the trend occurs (such as a sudden increase to 150 beats per minute when the normal trend is 70-80 beats per minute), it is marked as an outlier.

[0048] Meanwhile, performing federated learning alignment on heterogeneous data with inconsistent device identifiers is a process that, under the premise of protecting data privacy, uses federated learning algorithms to align features and train models for data from different devices. For example, for blood pressure data from monitor A (unit mmHg) and monitor B (unit kPa), the federated learning model can convert the data into a unified standard unit (such as mmHg) locally and train the model parameters locally. Then, only the encrypted model parameters are uploaded to the central server for aggregation to obtain a unified blood pressure analysis model applicable to all devices.

[0049] The process of mapping anesthesiology clinical nursing data into high-dimensional vectors using the first and second data mapping tables involves classifying and organizing the preprocessed and aligned data according to static and dynamic features, and then converting it into a high-dimensional vector representation. For example, static features such as age and gender, and dynamic features such as heart rate and blood pressure are mapped to different dimensions to form a high-dimensional vector that comprehensively reflects the patient's condition. If the static features include age (1 dimension) and gender (1 dimension), and the dynamic features include heart rate (1 dimension), blood pressure (2 dimensions, systolic and diastolic blood pressure), and blood oxygen saturation (1 dimension), then the resulting high-dimensional vector has 6 dimensions.

[0050] The process of generating a nursing knowledge graph involves using these high-dimensional vectors as nodes and constructing a graph structure by analyzing the relationships between nodes (such as causal relationships and temporal relationships). For example, the high-dimensional vector nodes of patients can be connected to drug dosage nodes and physiological indicator change nodes. The weight of the edges can be calculated based on the correlation. In the above steps, the accuracy and consistency of the data are improved through attention mechanisms and federated learning, and the analyzability and interpretability of the data are enhanced through high-dimensional vector mapping and knowledge graph generation.

[0051] Furthermore, the method of this application also includes:

[0052] An active learning-driven dynamic update mechanism for the knowledge graph is set up to filter high-conflict nodes; using a graph neural network, conflict association feature sets are obtained through the temporal association of the nursing knowledge graph; at the high-conflict nodes, the edge weights and node attributes of the nursing knowledge graph are dynamically adjusted in combination with the conflict association feature sets.

[0053] Specifically, active learning is a machine learning strategy that uses the model to actively select the most valuable data for annotation and learning in order to improve the model's performance. In the context of nursing knowledge graphs, the active learning-driven dynamic update mechanism of the graph will actively detect and select nodes that have high conflicts with the existing knowledge graph structure. High conflict nodes refer to nodes in the knowledge graph that have inconsistent or contradictory relationships with other nodes. For example, a node shows that the patient's blood pressure suddenly rises sharply during the maintenance of anesthesia, while other related nodes show that the blood pressure is stable. This abnormal situation may indicate data errors or potential clinical risks.

[0054] Graph neural networks can utilize the relationships between nodes and edges in a graph structure for information transmission and feature learning. In nursing knowledge graphs, graph neural networks capture complex temporal relationships between nodes. For example, by analyzing the temporal relationship between the anesthetic drug dosage adjustment node and the patient's physiological indicator change node, the effects of drugs on patients can be revealed. Conflict association feature sets refer to the set of features extracted by graph neural networks that are associated with high-conflict nodes. These features can reflect the nature and degree of conflict.

[0055] Execution steps: The process of setting up an active learning-driven dynamic update mechanism for the knowledge graph and screening high-conflict nodes involves real-time monitoring and analysis of the nursing knowledge graph using an active learning algorithm to identify nodes containing data errors or potential risks. For example, in the knowledge graph, if a node records that a patient's blood oxygen saturation suddenly drops to 80% (normal range is 95%-100%) during the maintenance of anesthesia, while adjacent nodes show that the dosage of anesthetic drugs and ventilator parameters have not been adjusted, this abnormal blood oxygen saturation node will be screened as a high-conflict node.

[0056] The process of obtaining a conflict-related feature set by using graph neural networks and the temporal association of nursing knowledge graphs involves deep analysis of the temporal relationships between high-conflict nodes and their surrounding nodes using graph neural networks. This process extracts features that reflect the conflict. Specifically, graph neural networks can analyze the temporal associations between the blood oxygen saturation node and its preceding nodes (such as drug dosage and ventilator settings) and subsequent nodes (such as heart rate changes and doctor interventions), extracting features such as the rate of blood oxygen decline and the accompanying heart rate change patterns, thus forming a conflict-related feature set.

[0057] At high-conflict nodes, the process of dynamically adjusting the edge weights and node attributes of the nursing knowledge graph by combining conflict association feature sets involves optimizing the structure of the knowledge graph based on information from the conflict association feature sets. For example, if analysis reveals that an abnormal decrease in blood oxygen saturation is related to an excessively low tidal volume setting on the ventilator, the system will increase the edge weights between the ventilator parameter setting node and the blood oxygen saturation node, while updating the attributes of the blood oxygen saturation node, marking it as "needs attention," and potentially triggering a risk alert. In the above steps, the knowledge graph is continuously optimized by actively learning to discover potential problems, using graph neural networks to deeply analyze the causes of problems, and dynamically adjusting the graph structure, thereby improving the accuracy and reliability of the nursing knowledge graph.

[0058] Furthermore, the method of this application also includes:

[0059] Using Bayesian posterior probability, the node with the highest uncertainty in the nursing knowledge graph is selected as the sample to be labeled; the sample to be labeled is grouped, and the group type is associated with the event label; using the grouped sample to be labeled, the edge weights and node attributes of the nursing knowledge graph are updated using local parameters.

[0060] Specifically, Bayesian posterior probability is used to calculate the probability that a hypothesis is true given certain observational data. In nursing knowledge graphs, the nodes with the highest uncertainty are those whose attributes or relationships are not clear enough, have low correlation with other nodes, or have a lot of missing data, representing potential risks or unknown factors in the data. Grouping nodes in nursing knowledge graphs is based on the characteristics and relationships of the nodes, grouping similar or related nodes together for batch processing and analysis.

[0061] Event labels are identifiers associated with key events during the anesthesia process, such as "drug injection," "abnormal heart rate," and "delayed awakening." These labels help to associate node groups with the actual stages and events of the anesthesia process. Local parameter updates involve adjusting the parameters of specific nodes and edges in the knowledge graph, rather than updating the entire graph globally. This allows for more efficient use of computational resources and reduces interference with the overall structure of the graph.

[0062] Execution steps: The process of selecting the node with the highest uncertainty in the nursing knowledge graph as the sample to be labeled using Bayesian posterior probability is to quantify the uncertainty of the node using a probability model and screen out the nodes that most need further confirmation and labeling. For example, in the nursing knowledge graph, a certain node represents a hypotension event that occurs in a patient during the maintenance of anesthesia, but the correlation between this node and other nodes (such as drug dosage, heart rate changes) is weak. Bayesian posterior probability calculation will show that this node has high uncertainty. Selecting such a node for labeling can help clarify its true relationship with other nodes and improve the accuracy of the knowledge graph.

[0063] The grouping of the samples to be labeled involves classifying nodes based on their characteristics and relationships. For example, nodes related to hypotension are grouped into one group, and nodes related to abnormal heart rate are grouped into another. Each group's type is associated with an event label. The process of updating the edge weights and node attributes of the nursing knowledge graph using local parameters based on the grouped samples involves targeted optimization of the knowledge graph. For example, for the hypotension event node group, analysis of the labeled samples reveals a strong correlation between it and the anesthetic drug dosage node. Therefore, the edge weights between these two nodes are increased, and the attributes of the hypotension node are updated, such as adding the label "caused by anesthetic drug dosage." In the above steps, key nodes are selected using Bayesian posterior probability, and local parameters are updated after grouping, enabling the knowledge graph to more realistically reflect the actual situation during anesthesia.

[0064] Furthermore, the method of this application includes:

[0065] Using a multi-head attention mechanism, the temporal entropy corresponding to the time interval and the support of the association rule are fused to configure the temporal dependency vector of key risk events; through the temporal dependency vector of the key risk events, path search is performed in the nursing knowledge graph to formulate risk propagation paths; when there is a node with a sudden drop in the anesthesia depth index in the risk propagation path, a dynamic compensation mechanism is triggered.

[0066] Specifically, the multi-head attention mechanism can simultaneously focus on multiple features in the data and assign different weights. In nursing knowledge graphs, the temporal entropy corresponding to the time interval and the support of association rules are two key features, reflecting the temporal uncertainty and association strength between events, respectively. By fusing these two features through the multi-head attention mechanism, a feature vector that comprehensively reflects the temporal dependency of events can be generated, namely the temporal dependency vector of key risk events.

[0067] Key risk events refer to events that may lead to serious consequences during the nursing process, such as "persistent hypotension" and "respiratory depression"; a time-series dependency vector is a mathematical vector that represents the dependency relationship of events over time; path search is an algorithm for finding specific paths in a graph structure, used to locate potential risk propagation paths in the nursing knowledge graph; a risk propagation path refers to the path in the knowledge graph from a risk event to other nodes through its associations; a dynamic compensation mechanism is a strategy that automatically adjusts system parameters to cope with sudden risks. When a serious risk is detected in the risk propagation path (such as a sudden drop in the anesthesia depth index), the dynamic compensation mechanism is triggered.

[0068] Execution steps: The process of configuring the temporal dependency vector of key risk events by fusing the temporal entropy and association rule support corresponding to the time interval using a multi-head attention mechanism is to comprehensively analyze these two features through the multi-head attention mechanism to generate a feature vector that reflects the temporal dependency relationship of events. For example, in a nursing knowledge graph, there is a strong correlation between key risk events (such as hypotension) and the preceding event "increased drug dosage", with a temporal entropy of 0.8 and an association rule support of 0.7 corresponding to the time interval. By fusing these two features into a temporal dependency vector through the multi-head attention mechanism, the weights respectively represent the contribution of the temporal entropy and association rule support corresponding to the time interval to the key risk event.

[0069] The process of searching for paths in a nursing knowledge graph using the temporal dependency vectors of key risk events involves using path search algorithms to find risk propagation paths. For example, starting from the node of increased drug dosage, it can be connected to the node of low blood pressure through its temporal dependency vector, and then connected to the node of decreased heart rate through the temporal dependency vector of the low blood pressure node, thus forming a risk propagation path.

[0070] When a sudden drop in the anesthesia depth index occurs in the risk transmission path, the process of triggering the dynamic compensation mechanism is a key node in monitoring the risk transmission path. Once a sudden drop in the anesthesia depth index is detected (e.g., from 60 to 40, while the normal range is 40-60), the system immediately triggers the compensation mechanism. The compensation mechanism will automatically adjust relevant parameters according to preset rules, such as reducing the dosage of anesthetic drugs, and issue an emergency alert. In the above steps, key features are integrated through a multi-head attention mechanism, the path search locates the risk transmission path, and the dynamic compensation mechanism responds to sudden risks in a timely manner, significantly improving the risk prevention and control capabilities of the nursing process.

[0071] Furthermore, based on the corresponding edge computing node on the platform, a first data mapping table is configured using static and dynamic features. The method of this application includes:

[0072] Static features are mapped to the local storage area of ​​the edge computing node; a hierarchical index is established according to the time series based on dynamic features, and the data update frequency and data transmission strategy are configured; a data synchronization protocol between the edge computing node and the cloud server is established using the data transmission strategy, and bidirectional differential synchronization is used to transmit only the data blocks that have changed, while recording the data version number.

[0073] Specifically, static features refer to patient information that remains relatively unchanged during anesthesia, such as the patient's age, gender, and medical history; dynamic features refer to physiological indicators that change in real time during anesthesia, such as heart rate, blood pressure, and blood oxygen saturation, which are constantly updated over time; edge computing nodes are computing devices located near the data source that can perform preliminary processing on the collected data to reduce data transmission latency and improve the real-time performance of data processing.

[0074] Local storage is the storage space on edge computing nodes used to temporarily store processed data; hierarchical indexing is a data indexing structure that indexes dynamic feature data in layers according to time series for fast querying and access; data update frequency refers to the update interval of dynamic feature data on edge computing nodes, such as once per second; data transmission strategy refers to the rules for transmitting data from edge computing nodes to cloud servers, including the time, conditions, and methods of transmission; bidirectional differential synchronization is a data synchronization technology that only transmits the changed parts of the data to reduce the amount of data transmission; data version number is an identifier used to record the number of data updates, which increments after each data update.

[0075] Execution steps: Static features are mapped to the local storage area of ​​the edge computing node. This ensures rapid access to the patient's static information during anesthesia. For example, static features such as the patient's age and medical history are stored in the local storage area of ​​the edge computing node and can be immediately retrieved when needed without having to obtain them from a remote server, reducing latency. Based on dynamic features, a hierarchical index is built according to time series, and the data update frequency and data transmission strategy are configured to ensure efficient management and timely updates of dynamic data. For example, heart rate monitoring data is updated once per second, and a hierarchical index is built according to time series, so that relevant data can be quickly located when querying heart rate changes within a specific time period.

[0076] The data transmission strategy stipulates that updated data blocks are only transmitted to the cloud server when the heart rate data changes by more than 3%, and the data version number is recorded to ensure data consistency and transmission efficiency. Based on this data transmission strategy, a data synchronization protocol is established between the edge computing node and the cloud server. Bidirectional differential synchronization is used to transmit only the changed data blocks, while recording the data version number. By transmitting only the changed data blocks, the amount of data transmission is significantly reduced, saving a considerable amount of bandwidth. Simultaneously, recording the data version number ensures that the data on the cloud server and the edge computing node remains synchronized, avoiding data conflicts. Through these steps, the real-time performance and consistency of the data are ensured, providing reliable data support for subsequent risk assessment and decision-making.

[0077] Furthermore, by using a sliding time window, the method obtains the time entropy and association rule support corresponding to the time intervals of adjacent events defined by the nursing knowledge graph. The method of this application includes:

[0078] Define a sliding time window and traverse the event nodes in the nursing knowledge graph; determine the time interval between adjacent events under the constraints of the nursing knowledge graph using time series, and obtain the time entropy corresponding to the time interval; quantify the event co-occurrence probability by using the association rule support of adjacent events under the constraints of the nursing knowledge graph, and form an event association strength matrix.

[0079] Specifically, a sliding time window iterates through time-series data by sliding a fixed-length window along the time axis; in a nursing knowledge graph, event nodes represent various events related to patient care, such as drug injections and changes in physiological indicators; the time entropy corresponding to the time interval is an indicator that measures the uncertainty of the time interval between events, and the more irregular the interval, the higher the entropy value; the support of association rules is used to quantify the probability of two events occurring simultaneously, and the higher the support, the stronger the association between the events; the event association strength matrix is ​​used to store the association strength between all event pairs for quick querying and analysis.

[0080] Execution steps: Define a sliding time window, traverse the event nodes in the nursing knowledge graph, and slide the window sequentially along the time axis of the knowledge graph by setting a fixed time window length (e.g., 5 minutes) and step size (e.g., 1 minute). Check the event nodes within the window. For example, during the anesthesia induction phase, the sliding time window can capture physiological indicator changes after drug injection. Determine the time interval between adjacent events using the time series and obtain the time entropy corresponding to the time interval. Within the sliding window, the time interval between drug injection and heart rate decrease varies greatly across multiple events. If the calculated time entropy corresponding to the time interval is 0.9, it indicates that the time interval has high uncertainty.

[0081] The process of quantifying the co-occurrence probability of events and forming an event association strength matrix by analyzing the support of association rules for adjacent events within a nursing knowledge graph involves statistically analyzing the frequency of co-occurrence of event pairs (such as drug injection and decreased heart rate) in the knowledge graph. For example, if drug injection and decreased heart rate occur simultaneously 80 times in 100 event sequences, the support of the association rule is 0.8. The support of all event pairs is then organized into a matrix. Furthermore, the temporal characteristics of events are captured through sliding time windows and entropy calculations. The association between events is quantified using the support of association rules and the strength matrix, providing important temporal and association information for risk assessment.

[0082] Furthermore, the method of this application includes:

[0083] The structured data in the nursing knowledge graph is spatiotemporally aligned with the image and audio data during the anesthesia process, and a virtual risk scenario is formulated using a generative adversarial network. Based on the virtual risk scenario, training data is collected and used to optimize the nursing knowledge graph.

[0084] Specifically, structured data refers to information stored in a fixed format in the nursing knowledge graph, such as the patient's age, drug dosage, blood pressure, and blood oxygen saturation; imaging data includes ultrasound and X-ray images during anesthesia, which usually have a timestamp; audio data includes voice recordings in the operating room, which also usually have a timestamp; spatiotemporal alignment matches different types of medical data in time and space dimensions for comprehensive analysis; generative adversarial networks are used to generate seemingly realistic virtual data, virtual risk scenarios are datasets that simulate possible medical risk situations, and feedback optimization uses training data to update the existing nursing knowledge graph.

[0085] Execution steps: The structured data in the nursing knowledge graph is spatiotemporally aligned with image and audio data. Synchronization of multi-source data is ensured through timestamp and spatial coordinate matching. For example, the patient's heart rate data (structured data) is aligned with simultaneous ultrasound images (image data) and voice recordings (audio data). A generative adversarial network (GAN) is used to formulate virtual risk scenarios. By training a generator and discriminator, data under risk scenarios such as hypotension is simulated. Furthermore, the generator is used to generate virtual nursing data, making it as close as possible to real data; the discriminator is used to distinguish between the virtual data generated by the generator and real data.

[0086] Through continuous iterative training, the virtual data generated by the generator becomes increasingly closer to real data until the discriminator can no longer distinguish between real and virtual data. Possible virtual risk scenarios are defined, such as hypotension and arrhythmia. Training data is collected based on the virtual risk scenarios, including virtual structured data, image data, and audio data, to optimize the nursing knowledge graph. The virtual and real data are then fused to update the nodes and edges in the nursing knowledge graph. The training data is then used to feed back into and optimize the nursing knowledge graph, adjusting its structure and parameters to enable it to more accurately identify and predict risk events.

[0087] In summary, the beneficial effects of the embodiments of this application are:

[0088] This approach utilizes clinical nursing data from the anesthesiology department, including patient basic information, anesthesia protocol data, and anesthesia monitoring data. A first data mapping table is configured based on static and dynamic features according to the corresponding edge computing nodes on the platform. A second data mapping table is configured based on static and dynamic features according to the corresponding visualization units on the application side. Based on the anesthesiology clinical nursing data, and combining the first and second data mapping tables, a nursing knowledge graph is generated. Simultaneously, by using a sliding time window, the temporal entropy and association rule support corresponding to the time intervals of adjacent events within the constraints of the nursing knowledge graph are obtained. Partial least squares discriminant analysis is then performed using the temporal entropy and association rule support corresponding to the time intervals of adjacent events within the constraints of the nursing knowledge graph, with dose change rate as the input feature, to conduct multi-scale assessments of potential risks and provide risk alerts based on dynamic risk thresholds. This application provides a method and system for managing clinical nursing data in anesthesiology. By configuring first and second data mapping tables and combining the time entropy and association rule support corresponding to the time interval of adjacent events, it enables multi-scale assessment of potential anesthesia risks. This achieves hierarchical management and precise mapping of static and dynamic features in clinical nursing data in anesthesiology, and dynamically adjusts risk thresholds according to the temporal characteristics of different anesthesia stages, significantly improving the accuracy and timeliness of risk assessment.

[0089] Example 2, based on the same inventive concept as the anesthesiology clinical nursing data management method in the foregoing examples, such as... Figure 2 As shown in the figure, this application provides an anesthesiology clinical nursing data management system, wherein the system includes:

[0090] Data collection module M100: Collects clinical nursing data in the anesthesiology department, including basic patient information, anesthesia protocol data, and anesthesia monitoring data.

[0091] Data mapping module M200: Configures the first data mapping table with static and dynamic features based on the corresponding edge computing node on the platform; configures the second data mapping table with static and dynamic features based on the corresponding visualization unit on the application side.

[0092] Nursing knowledge graph generation module M300: Based on the clinical nursing data of the anesthesiology department, and combined with the first data mapping table and the second data mapping table, a nursing knowledge graph is generated.

[0093] Sliding analysis module M400: Simultaneously, by sliding the time window, it obtains the time entropy and association rule support corresponding to the time interval of adjacent events under the constraints of the nursing knowledge graph.

[0094] Risk alert module M500: It performs partial least squares discriminant analysis based on the time entropy and association rule support corresponding to the time interval of adjacent events under the limitation of the nursing knowledge graph, uses the dose change rate as input feature, performs multi-scale assessment of potential risks, and provides risk alerts in combination with dynamic risk thresholds.

[0095] Furthermore, the risk alert module M500 is also used to perform the following methods:

[0096] Temporal feature analysis is performed using event tags, which include the anesthesia induction phase, anesthesia maintenance phase, and recovery phase; dynamic risk thresholds are configured based on the temporal relationships corresponding to the event tags.

[0097] Furthermore, the nursing knowledge graph generation module M300 is also used to perform the following methods:

[0098] An attention mechanism is used to process outliers in the anesthesiology clinical nursing data, and federated learning alignment is performed on heterogeneous data with inconsistent device identifiers. The anesthesiology clinical nursing data is mapped into high-dimensional vectors through the first data mapping table and the second data mapping table to generate the nursing knowledge graph.

[0099] Furthermore, the nursing knowledge graph generation module M300 is also used to perform the following methods:

[0100] An active learning-driven dynamic update mechanism for the knowledge graph is set up to filter high-conflict nodes; using a graph neural network, conflict association feature sets are obtained through the temporal association of the nursing knowledge graph; at the high-conflict nodes, the edge weights and node attributes of the nursing knowledge graph are dynamically adjusted in combination with the conflict association feature sets.

[0101] Furthermore, the nursing knowledge graph generation module M300 is also used to perform the following methods:

[0102] Using Bayesian posterior probability, the node with the highest uncertainty in the nursing knowledge graph is selected as the sample to be labeled; the sample to be labeled is grouped, and the group type is associated with the event label; using the grouped sample to be labeled, the edge weights and node attributes of the nursing knowledge graph are updated using local parameters.

[0103] Furthermore, the nursing knowledge graph generation module M300 is also used to perform the following methods:

[0104] Using a multi-head attention mechanism, the temporal entropy corresponding to the time interval and the support of the association rule are fused to configure the temporal dependency vector of key risk events; through the temporal dependency vector of the key risk events, path search is performed in the nursing knowledge graph to formulate risk propagation paths; when there is a node with a sudden drop in the anesthesia depth index in the risk propagation path, a dynamic compensation mechanism is triggered.

[0105] Furthermore, the data mapping module M200 is used to perform the following method:

[0106] Static features are mapped to the local storage area of ​​the edge computing node; a hierarchical index is established according to the time series based on dynamic features, and the data update frequency and data transmission strategy are configured; a data synchronization protocol between the edge computing node and the cloud server is established using the data transmission strategy, and bidirectional differential synchronization is used to transmit only the data blocks that have changed, while recording the data version number.

[0107] Furthermore, the sliding analysis module M400 is also used to perform the following methods:

[0108] Define a sliding time window and traverse the event nodes in the nursing knowledge graph; determine the time interval between adjacent events under the constraints of the nursing knowledge graph using time series, and obtain the time entropy corresponding to the time interval; quantify the event co-occurrence probability by using the association rule support of adjacent events under the constraints of the nursing knowledge graph, and form an event association strength matrix.

[0109] Furthermore, the sliding analysis module M400 is also used to perform the following methods:

[0110] The structured data in the nursing knowledge graph is spatiotemporally aligned with the image and audio data during the anesthesia process, and a virtual risk scenario is formulated using a generative adversarial network. Based on the virtual risk scenario, training data is collected and used to optimize the nursing knowledge graph.

[0111] In summary, any step can be stored as a computer instruction or program in an unrestricted computer memory and can be called and recognized by an unrestricted computer processor; no further restrictions are imposed here.

[0112] Furthermore, the above technical solutions only embody the preferred technical solutions of the embodiments of this application. Any changes that those skilled in the art may make to certain parts of these solutions embody the novel principles of the embodiments of this application. Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application.

Claims

1. A method for managing clinical nursing data in anesthesiology, characterized in that, The method includes: Collect clinical nursing data in the anesthesiology department, including basic patient information, anesthesia protocol data, and anesthesia monitoring data; Based on the edge computing nodes corresponding to the platform, a first data mapping table is configured with static and dynamic features; based on the visualization units corresponding to the application, a second data mapping table is configured with static and dynamic features. Based on the aforementioned clinical nursing data from the anesthesiology department, and in conjunction with the first data mapping table and the second data mapping table, a nursing knowledge graph is generated. Simultaneously, by sliding the time window, the temporal entropy and association rule support corresponding to the time interval of adjacent events under the constraints of the nursing knowledge graph are obtained; Partial least squares discriminant analysis is performed using the temporal entropy and association rule support corresponding to the time interval of adjacent events defined by the nursing knowledge graph. The dose change rate is used as the input feature to conduct multi-scale assessment of potential risks and risk alerts are provided in combination with dynamic risk thresholds. Among these, risk alerts are provided in conjunction with dynamic risk thresholds, including: Temporal feature analysis was performed using event tags, which included the anesthesia induction phase, anesthesia maintenance phase, and recovery phase. Configure dynamic risk thresholds based on the temporal relationships corresponding to the event tags; The generation of a nursing knowledge graph based on the aforementioned anesthesiology clinical nursing data, combined with the first data mapping table and the second data mapping table, also includes: An attention mechanism is used to process outliers in the anesthesiology clinical nursing data, and federated learning alignment is performed on heterogeneous data with inconsistent device identifiers. The anesthesiology clinical nursing data is mapped into high-dimensional vectors using the first data mapping table and the second data mapping table to generate the nursing knowledge graph. Among them, an active learning-driven dynamic graph update mechanism is set up to filter high-conflict nodes; Using graph neural networks, conflict association feature sets are obtained through the temporal association of the nursing knowledge graph; At the high-conflict nodes, the edge weights and node attributes of the nursing knowledge graph are dynamically adjusted in conjunction with the conflict association feature set. This also includes: Using Bayesian posterior probability, the node with the highest uncertainty in the nursing knowledge graph is selected as the sample to be labeled. The samples to be labeled are grouped, and the group type is associated with the event label. The edge weights and node attributes of the nursing knowledge graph are updated using local parameters based on the grouped samples to be labeled. Specifically, by using a sliding time window, the temporal entropy and association rule support corresponding to the time intervals of adjacent events defined by the nursing knowledge graph are obtained, including: Define a sliding time window and traverse the event nodes in the nursing knowledge graph; The time interval between adjacent events defined by the nursing knowledge graph is determined by time series analysis, and the time entropy corresponding to the time interval is obtained. By using the support of association rules for adjacent events defined by the nursing knowledge graph, the co-occurrence probability of events is quantified to form an event association strength matrix; This includes: The structured data in the nursing knowledge graph is spatiotemporally aligned with the image and audio data during the anesthesia process, and a generative adversarial network is used to formulate virtual risk scenarios. Based on the virtual risk scenario, training data is collected, and the training data is used to optimize the nursing knowledge graph.

2. The method for managing clinical nursing data in anesthesiology as described in claim 1, characterized in that, The method includes: Using a multi-head attention mechanism, feature fusion is performed on the temporal entropy corresponding to the time interval and the support of the association rule to configure the temporal dependency vector of key risk events; By using the temporal dependency vectors of the key risk events, a path search is performed in the nursing knowledge graph to determine the risk propagation path; When a sudden drop in the anesthesia depth index occurs in the risk propagation path, a dynamic compensation mechanism is triggered.

3. The method for managing clinical nursing data in anesthesiology as described in claim 2, characterized in that, Based on the corresponding edge computing node on the platform, a first data mapping table is configured using static and dynamic features. The method includes: Map static features to the local storage area of ​​edge computing nodes; Based on dynamic characteristics, a hierarchical index is established according to the time series, and the data update frequency and data transmission strategy are configured. Using the aforementioned data transmission strategy, a data synchronization protocol is established between the edge computing node and the cloud server. Bidirectional differential synchronization is used to transmit only the data blocks that have changed, while recording the data version number.

4. A clinical nursing data management system for anesthesiology, characterized in that, The system is used to implement the anesthesiology clinical nursing data management method according to any one of claims 1-3, the system comprising: Data collection module: Collects clinical nursing data in the anesthesiology department, including basic patient information, anesthesia protocol data, and anesthesia monitoring data; Data mapping module: Configures the first data mapping table based on the edge computing node corresponding to the platform using static and dynamic features; configures the second data mapping table based on the visualization unit corresponding to the application using static and dynamic features. Nursing knowledge graph generation module: Based on the aforementioned clinical nursing data from the anesthesiology department, and in conjunction with the first data mapping table and the second data mapping table, a nursing knowledge graph is generated; Sliding analysis module: Simultaneously, by sliding the time window, the time entropy and association rule support corresponding to the time interval of adjacent events under the constraints of the nursing knowledge graph are obtained; Risk alert module: Partial least squares discriminant analysis is performed using the time entropy and association rule support corresponding to the time interval of adjacent events under the constraints of the nursing knowledge graph. The dose change rate is used as the input feature to conduct multi-scale assessment of potential risks and to provide risk alerts in combination with dynamic risk thresholds. Among these, risk alerts are provided in conjunction with dynamic risk thresholds, including: Temporal feature analysis was performed using event tags, which included the anesthesia induction phase, anesthesia maintenance phase, and recovery phase. Configure dynamic risk thresholds based on the temporal relationships corresponding to the event tags; The generation of a nursing knowledge graph based on the aforementioned anesthesiology clinical nursing data, combined with the first data mapping table and the second data mapping table, also includes: An attention mechanism is used to process outliers in the anesthesiology clinical nursing data, and federated learning alignment is performed on heterogeneous data with inconsistent device identifiers. The anesthesiology clinical nursing data is mapped into high-dimensional vectors using the first data mapping table and the second data mapping table to generate the nursing knowledge graph. Among them, an active learning-driven dynamic graph update mechanism is set up to filter high-conflict nodes; Using graph neural networks, conflict association feature sets are obtained through the temporal association of the nursing knowledge graph; At the high-conflict nodes, the edge weights and node attributes of the nursing knowledge graph are dynamically adjusted in conjunction with the conflict association feature set. This also includes: Using Bayesian posterior probability, the node with the highest uncertainty in the nursing knowledge graph is selected as the sample to be labeled. The samples to be labeled are grouped, and the group type is associated with the event label. The edge weights and node attributes of the nursing knowledge graph are updated using local parameters based on the grouped samples to be labeled. Specifically, by using a sliding time window, the temporal entropy and association rule support corresponding to the time intervals of adjacent events defined by the nursing knowledge graph are obtained, including: Define a sliding time window and traverse the event nodes in the nursing knowledge graph; The time interval between adjacent events defined by the nursing knowledge graph is determined by time series analysis, and the time entropy corresponding to the time interval is obtained. By using the support of association rules for adjacent events defined by the nursing knowledge graph, the co-occurrence probability of events is quantified to form an event association strength matrix; This includes: The structured data in the nursing knowledge graph is spatiotemporally aligned with the image and audio data during the anesthesia process, and a generative adversarial network is used to formulate virtual risk scenarios. Based on the virtual risk scenario, training data is collected, and the training data is used to optimize the nursing knowledge graph.

Images