An operating room nursing key node decision support system

Abstract

The present application relates to the technical field of medical intelligent decision-making, in particular to a key node decision support system for operating room nursing, comprising: a nursing operation benchmark containing standard execution sequence and parameter interval is established based on historical cases. By integrating discrete operations and continuous monitoring data in real-time surgery, a unified semantic nursing process event chain is formed. The system compares the event chain with the execution benchmark to identify potential risk segments, and automatically extracts complete operating room multi-dimensional state snapshots within a specific time window before and after each segment. According to the state snapshot, a cause deduction network is dynamically constructed, and the cause chain that may lead to the risk is speculated by traversing the network. All deviation information and cause analysis results are integrated to generate a structured decision intervention prompt. The present application can realize the deepening from automatic risk identification to intelligent root cause diagnosis, and provide precise and efficient decision support for operating room nursing.

Description

Technical Field

[0001] This invention relates to the field of medical intelligent decision-making technology, and in particular to a decision support system for key nodes in operating room nursing. Background Technology

[0002] Currently, the monitoring and management of operating room nursing processes mainly rely on electronic medical record systems and independent alarm mechanisms of medical equipment. Existing technical solutions generally focus on threshold monitoring of single vital sign parameters of patients or post-event integrity verification of nursing operation records. These technologies decompose the continuous and collaborative nursing process into discrete data points and static records, lacking real-time and holistic tracking and analysis of the logical sequence of nursing operations, the dynamic correlation between multiple parameters, and the entire operating room operational context.

[0003] Existing technologies have limitations. When a system alerts to an anomaly, it typically only presents superficial phenomena such as excessive parameters or operational omissions, failing to automatically elucidate the underlying causes. Operating room nursing risks are often the result of multiple factors intertwined and evolving over time, involving interactions across multiple dimensions such as personnel, equipment, processes, and the environment. Traditional technologies cannot proactively capture and integrate panoramic state data of the operating room before and after a given moment during risk identification, nor can they perform effective causal reasoning based on this complex integrated context. This leads to inefficient tracing of the root causes of risks, with decision support remaining at the level of superficial alerts, making it difficult to formulate intervention recommendations with in-depth insights and precise targeting.

[0004] A technological solution is needed to overcome the aforementioned limitations. This solution must be able to automatically identify deviations from standard procedures in real-time operational workflows and automatically reconstruct the complete multi-dimensional operating room context before and after the identified risks occur. Furthermore, based on this panoramic context, the solution should be able to intelligently analyze and deduce the potential causal chains leading to the risks. The ultimate goal is to achieve a substantial leap in operating room nursing decision support from "phenomenon alerts" to "root cause diagnosis." Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a decision support system for key nodes in operating room nursing.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a decision support system for key nodes in operating room nursing, comprising:

[0007] The benchmark establishment module establishes the execution benchmarks for operating room nursing operations. These benchmarks are extracted from nursing process data of historical successful surgical cases and include the expected execution sequence and allowable parameter fluctuation range for each standard nursing step.

[0008] The data integration module connects to the real-time nursing operation flow of the ongoing surgery and integrates the discrete operation actions and continuous monitoring data in the real-time nursing operation flow into a nursing process event chain with unified semantic tags according to the time axis.

[0009] The risk identification module compares the nursing process event chain with the execution benchmark at multiple levels to identify potential risk segments in the nursing process event chain that deviate from the execution benchmark.

[0010] The scenario backtracking module triggers a deep scenario backtracking analysis for each identified potential risk segment. The deep scenario backtracking analysis is used to extract a complete operating room state snapshot within a specific time window before and after the occurrence of the potential risk segment.

[0011] The cause deduction module constructs a dynamically evolving cause deduction network based on the complete operating room state snapshot, and infers the possible causal chains that lead to the occurrence of the potential risk segment by traversing the cause deduction network.

[0012] The decision generation module integrates deviation information of all the potential risk segments with the corresponding possible causal chains to generate a structured set of decision intervention prompts.

[0013] Preferably, the establishment of the operating room nursing operation execution standards is specifically implemented as follows:

[0014] Collect and clean the complete records of historical successful surgical cases, and extract the action instruction sequences, physiological parameter response curves and event trigger logs related to nursing operations;

[0015] Clustering methods were used to group the historical successful surgical cases according to surgical type and complexity. The nursing operation data of each group were summarized to form typical nursing pathways for the corresponding groups.

[0016] Key decision points are marked on the typical nursing pathway, and for each key decision point, a set of preceding events, a set of succeeding events, and a normal reference range for associated monitoring parameters are defined, which together constitute the execution benchmark.

[0017] Preferably, the access to the real-time nursing operation flow of the ongoing surgery integrates the discrete operation actions and continuous monitoring data in the real-time nursing operation flow into a nursing process event chain with unified semantic tags according to the time axis, specifically implemented as follows:

[0018] Simultaneously receive raw data packets from different operating room information systems and sensor networks;

[0019] The raw data messages are parsed to convert instrument operation signals, personnel voice commands, and drug infusion records into standardized event objects, and vital sign waveforms and gas concentration readings into timestamped data point sequences.

[0020] The standardized event objects are interwoven with the timestamped data point sequence in chronological order, and each arrangement unit is assigned a semantic label based on the medical knowledge base to form the nursing process event chain.

[0021] Preferably, the step of performing multi-level comparisons between the nursing process event chain and the execution benchmark to identify potential risk segments in the nursing process event chain that deviate from the execution benchmark is specifically implemented as follows:

[0022] The multi-level comparison includes the integrity check of the event sequence, the calculation of the offset of the execution time sequence, and the consistency test of the changing trends of key physiological parameters.

[0023] The event sequence of the nursing process event chain is matched with the typical nursing path of the corresponding surgical type in the execution benchmark, missing or redundant events are marked, and event integrity comparison results are generated.

[0024] Calculate the time difference between the actual occurrence time of key operations in the nursing process event chain and the preset time in the execution benchmark. When the time difference exceeds the tolerance threshold, generate a time sequence offset warning.

[0025] Analyze the data point sequence of specific physiological parameters in the nursing process event chain, determine whether the change curve of the parameter continuously exceeds the normal reference range defined by the execution benchmark or deviates from the expected trend, and generate parameter abnormality alarm;

[0026] Based on the combined results of the event integrity comparison, the time sequence offset warning, and the parameter anomaly alarm, the time periods with continuous anomalies in the nursing process event chain are identified and defined as the potential risk segments.

[0027] Preferably, for each identified potential risk segment, a deep contextual retrospective analysis is triggered. This deep contextual retrospective analysis is used to extract a complete operating room state snapshot within a specific time window before and after the occurrence of the potential risk segment. Specifically, this is implemented as follows:

[0028] The complete operating room status snapshot covers personnel activities, equipment operation, medication records, and environmental indicators;

[0029] Based on the start time of the potential risk segment, a preset duration is extended forward and backward to define the scenario retrospective time window;

[0030] Retrieve and extract all relevant records that fall within the aforementioned scenario recall time window from the full log database and video storage system of the operating room;

[0031] The extracted records are categorized and integrated according to personnel, equipment, medication, and environment dimensions to form a multi-dimensional, time-aligned snapshot of the complete operating room status.

[0032] Preferably, the step of constructing a dynamically evolving cause-effect inference network based on the complete operating room state snapshot, and inferring the possible causal chains leading to the occurrence of the potential risk segment by traversing the cause-effect inference network, is specifically implemented as follows:

[0033] The nodes of the causal deduction network represent state elements, and the edges represent the interaction relationships between elements;

[0034] Each individual state element in the complete operating room state snapshot is mapped to a node of the cause-and-effect network;

[0035] Based on a medical rule base and a causal knowledge graph, causal relationship edges between nodes are established that change over time, and these causal relationship edges have directionality and confidence weights.

[0036] Taking the state element corresponding to the potential risk segment as the starting point, perform a reverse breadth-first search along the causal edge in the cause inference network to enumerate all upstream node paths that can reach the starting point.

[0037] Based on the confidence weights of each edge on the path and the temporal proximity of the nodes, all upstream node paths found are sorted, and the top-ranked paths are selected as the possible causal chains.

[0038] Preferably, the step of integrating the deviation information of all the potential risk segments with the corresponding possible causal chains to generate a structured set of decision intervention prompts is specifically implemented as follows:

[0039] Create a decision intervention prompt record for each of the aforementioned potential risk segments;

[0040] In the decision intervention prompt record, the location index of the potential risk segment in the nursing process event chain, the deviation type description, and the summary of the main causes extracted from the possible cause chain are stored in association;

[0041] Based on the summary of the main causes and the preset coping strategy mapping table, a set of corrective measures associated with the decision intervention prompt records is generated;

[0042] All decision intervention prompts are organized in chronological order according to their associated potential risk segments to form the decision intervention prompt set. Each item in the decision intervention prompt set is associated with a specific position in the nursing process event chain and contains the set of corrective actions.

[0043] Preferably, the step of grouping the historical successful surgical cases according to surgical type and complexity using a clustering method is specifically implemented as follows:

[0044] Extract the surgical type code and surgical complexity score for each historical successful surgical case to construct a case feature vector;

[0045] The density clustering algorithm is used to perform unsupervised clustering analysis on the feature vectors of the cases to identify groups of historical successful surgical cases with similar features;

[0046] The silhouette coefficient of each cluster group is evaluated, and the target groups that meet the preset conditions for cluster quality are selected.

[0047] Pattern summarization was performed on the nursing operation data of historical successful surgical cases within each target group, and statistical characteristics of typical nursing pathways within the target group were calculated.

[0048] Preferably, the step of calculating the time difference between the actual occurrence time of the key operation in the nursing process event chain and the preset time in the execution benchmark, and generating a time sequence offset warning when the time difference exceeds a tolerance threshold, is specifically implemented as follows:

[0049] Extract the preset expected execution time and corresponding tolerance threshold range for each key operation from the execution benchmark;

[0050] Locate the actual timestamp of each critical operation in the nursing process event chain;

[0051] Calculate the absolute time difference between the actual occurrence time and the preset expected execution time for each critical operation;

[0052] The absolute time difference is compared with the corresponding tolerance threshold range. When the absolute time difference exceeds the upper or lower limit of the tolerance threshold range, a timing offset warning containing specific time difference information and key operation identifiers is generated.

[0053] Preferably, the step of using the state element corresponding to the potential risk segment as the starting point, and performing a reverse breadth-first search along the causal edges in the cause-effect network to enumerate all upstream node paths that can reach the starting point, is specifically implemented as follows:

[0054] Mark the state elements corresponding to the potential risk segments as the search starting node set;

[0055] Initialize the search queue by adding all nodes from the starting node set to the search queue;

[0056] Remove the current node from the search queue and traverse all upstream nodes pointing to the current node along the incoming edge direction;

[0057] For each upstream node, record the path information from the starting node to the upstream node, and add the upstream node to the search queue;

[0058] Repeat the node retrieval and upstream node traversal operations until the search queue is empty or the preset search depth limit is reached;

[0059] Collect all reverse search paths starting from the starting node to generate a set of possible causal paths.

[0060] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0061] For in-depth contextual retrospective analysis automatically triggered by risk segments, a complete snapshot of the operating room status within a specific time window before and after the risk occurs is extracted. This process is not simply retrieving historical logs, but rather performing real-time synchronization, semantic alignment, and structured integration of heterogeneous data from different sources with their own timestamps. It systematically integrates personnel operation events, equipment operating parameters, environmental monitoring indicators, medication usage records, and key timing nodes, fusing previously discrete and isolated data streams into a multi-dimensional contextual dataset under a unified time benchmark. This establishes an objective and comprehensive analytical foundation for each identified risk, enabling subsequent causal analysis to be conducted based on a highly realistic digital scene. This overcomes the problem of insufficient judgment basis caused by information fragmentation in traditional systems, providing direct support for understanding the true context of risk occurrence.

[0062] Based on complete state snapshots, a causal inference network is constructed and traversed through dynamic evolution. This approach abandons direct matching of static rules and instead constructs a network model in real-time, where nodes represent potential causal factors and edges represent the relationships between these factors, based on the current specific state snapshot data. The network construction relies on an embedded medical knowledge graph and historical data statistical patterns. An algorithm traverses possible paths within the network, calculating and evaluating the probabilities of different causal chains from initial perturbations to the final risk event. This achieves context-adaptive diagnosis of risk root causes. Its inference conclusions are closely dependent on the specific state of the current surgery, revealing how complex dynamic interactions between multiple factors ultimately lead to specific deviations. Therefore, the output decision suggestions are highly targeted and root-cause-oriented, effectively assisting interventions to directly address the essence of the problem. Attached Figure Description

[0063] Figure 1 This is a sequence diagram of the decision support system for key nodes in operating room nursing as described in this invention;

[0064] Figure 2 A flowchart for establishing the benchmark;

[0065] Figure 3 A flowchart for integrating the event chain of the nursing process;

[0066] Figure 4 A bar chart comparing the number of potential risk segments for different surgical types;

[0067] Figure 5 A heatmap showing the correlation strength between inducements and measures in the operating room nursing decision-making intervention system. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0069] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0070] See Figure 1 The baseline establishment module establishes execution baselines by analyzing nursing process data from historical successful surgical cases. These baselines define the expected execution sequence of standard nursing procedures and the allowable parameter fluctuation range. The data integration module connects to the nursing operation flow during surgery in real time, integrating discrete operational actions with continuous monitoring data into a nursing process event chain with unified semantic tags. The risk identification module performs multi-level comparisons between the nursing process event chain and the execution baselines to identify potential risk segments that deviate from the baselines. The scenario retrospective module triggers deep scenario retrospective analysis for each potential risk segment, extracting a complete operating room state snapshot within a specific time window before and after the occurrence of the potential risk segment. The cause deduction module constructs a dynamically evolving cause deduction network based on the complete operating room state snapshots, and infers the possible causal chains leading to the occurrence of potential risk segments by traversing this network. The decision generation module integrates the deviation information of all potential risk segments with the corresponding possible causal chains to generate a structured set of decision intervention prompts.

[0071] In one embodiment of the present invention, see [reference] Figure 2 To establish execution benchmarks for operating room nursing procedures, the following steps were taken: First, the complete records of historical successful surgical cases were collected and cleaned, extracting action command sequences, physiological parameter response curves, and event trigger logs related to nursing procedures. Second, a clustering method was used to group historical successful surgical cases according to surgical type and complexity. Patterns were summarized from the nursing operation data of each group to form typical nursing pathways for that group. Third, key decision points were marked on these typical nursing pathways, and for each key decision point, a set of preceding events, a set of succeeding events, and normal reference ranges for associated monitoring parameters were defined, collectively forming the execution benchmarks. The clustering method involved extracting the surgical type code and surgical complexity score for each historical successful surgical case, constructing a case feature vector, and using a density clustering algorithm to perform unsupervised clustering analysis on the case feature vectors to identify groups of historical successful surgical cases with similar characteristics. Finally, a silhouette coefficient was evaluated for each cluster group, and target groups that met preset clustering quality criteria were selected. Patterns were summarized from the nursing operation data of historical successful surgical cases within each target group, and the statistical characteristics of typical nursing pathways within the target group were calculated.

[0072] In practice, establishing execution standards for operating room nursing procedures begins with collecting and cleaning complete records of historical successful surgical cases. These records originate from the hospital's electronic medical record system, anesthesia information management system, and operating room equipment logs. From these records, sequences of action instructions related to nursing procedures are extracted. These sequences include timestamps and operational details of steps performed by nurses, such as instrument transfer, patient positioning, disinfection, and draping. Physiological parameter response curves are also extracted, representing continuous monitoring data of the patient's vital signs during surgery. Event trigger logs are further extracted, recording discrete time-point information such as medication use and the occurrence of special events.

[0073] In some embodiments, before pattern summarization of the nursing operation data of extracted historical successful surgical cases, clustering methods are used to group the historical successful surgical cases according to surgical type and complexity. This grouping operation specifically involves extracting the surgical type code and surgical complexity score for each historical successful surgical case. The surgical type code uses the International Classification of Diseases (ICD) surgical and operational codes, and the surgical complexity score is calculated comprehensively based on dimensions such as surgical duration, estimated blood loss, and the number of organ systems involved. A case feature vector is constructed based on these features. A density-based clustering algorithm is then used to perform unsupervised clustering analysis on the case feature vectors. The parameters, neighborhood radius and minimum sample size, are pre-set according to the historical data distribution. After the algorithm runs, it identifies groups of historical successful surgical cases with similar characteristics.

[0074] It is understandable that after completing cluster analysis, the quality of the groupings needs to be evaluated. This involves assessing the silhouette coefficient for each cluster. The silhouette coefficient measures the similarity between feature vectors of cases within the same cluster and the difference between feature vectors of cases in different clusters. The formula for calculating the silhouette coefficient is:

[0075]

[0076] in: Represents the feature vector of a case The profile coefficient, Represents the feature vector of a case The average distance to the feature vectors of all other cases within its cluster group. Represents the feature vector of a case The minimum average distance to the feature vectors of all cases in any other cluster group. Cluster groups with an average silhouette coefficient greater than a preset threshold are selected as target groups whose clustering quality meets the preset conditions.

[0077] In practice, pattern summarization is performed on nursing operation data from historical successful surgical cases within each target group. This process calculates the statistical characteristics of typical nursing pathways within the target group. These characteristics include the frequency of each nursing operation step in the group's cases, the median and quartile intervals of execution time, and the mean range of changes in related physiological parameters before and after the operation. Based on these statistical characteristics, typical nursing pathways for the corresponding groups are formed. A typical nursing pathway is a standard sequence consisting of key nursing operation nodes, their temporal relationships, and parameter contexts. Key decision points are marked on these typical nursing pathways. These are the steps whose execution, timing, or parameters significantly affect subsequent nursing progress or patient status; examples include administering antibiotics before a specific surgical step or initiating vasopressor infusion when the patient's blood pressure drops to a certain threshold. A set of preceding events is defined for each key decision point; these are events that must be completed before the key decision point occurs. A set of subsequent events is also defined; these are events expected to occur after the key decision point. Finally, normal reference ranges for related monitoring parameters are defined.

[0078] In one embodiment of the present invention, see [reference] Figure 3The system integrates the real-time nursing workflow of ongoing surgeries, combining discrete actions and continuous monitoring data along a timeline into a nursing process event chain with unified semantic tags. Specifically, it synchronously receives raw data packets from different operating room information systems and sensor networks. The raw data packets are parsed, converting instrument operation signals, personnel voice commands, and medication infusion records into standardized event objects, and vital sign waveforms and gas concentration readings into timestamped data point sequences. The standardized event objects and timestamped data point sequences are interleaved and arranged chronologically, and each arrangement unit is assigned a semantic tag based on a medical knowledge base, forming the nursing process event chain.

[0079] In practice, the process involves accessing the real-time nursing workflow of an ongoing surgery. This begins with the synchronous reception of raw data packets from different operating room information systems and sensor networks. The operating room information systems include anesthesia clinical information systems, surgical instrument tracking systems, and nursing documentation systems. The sensor networks include patient vital signs monitors, anesthesia machine breathing circuit sensors, and operating room environmental monitoring probes. The raw data packets follow different communication protocols and data formats. For example, HL7 messages are used for patient event transmission, the IEEE 11073 standard is used for streaming physiological parameters, and a custom serial port protocol is used for status reporting of specific surgical equipment.

[0080] In some embodiments, the received raw data packets are parsed according to a predefined packet structure dictionary and data conversion rules. Instrument operation signals are converted into standardized event objects. These signals originate from the status switches of devices such as electrosurgical units and ultrasonic scalpels. The standardized event objects include the event type, occurrence time, instrument code, and energy setting parameters. Personnel voice commands are also converted into standardized event objects. These commands are captured by a voice recognition system in the operating room. The standardized event objects include the event type, occurrence time, the role of the person issuing the command, and the command text. Finally, drug infusion records are converted into standardized event objects. These records originate from the recording interface of the intelligent infusion pump. The standardized event objects include the event type, occurrence time, drug name, infusion rate, and cumulative dose. Vital sign waveforms are converted into timestamped data point sequences. Vital sign waveforms, such as electrocardiograms and arterial blood pressure waveforms, are digitized using a fixed sampling frequency. Each sampling point is appended with its precise timestamp of being collected, forming a timestamped data point sequence. Gas concentration readings, such as end-tidal carbon dioxide concentration and inhaled oxygen concentration, are also converted into timestamped data point sequences. Gas concentration readings, such as end-tidal carbon dioxide concentration and inhaled oxygen concentration, are collected and recorded at certain time intervals. Each reading point is also appended with a timestamp, forming a timestamped data point sequence.

[0081] Understandably, after data parsing, timeline integration is required, interleaving standardized event objects with timestamped data point sequences according to chronological order. The logic of this interleaving is to merge and sort the time points of all standardized event objects and timestamped data point sequences on a global, high-precision timeline. This sorting process needs to handle time synchronization issues from different data sources. The synchronization mechanism uses network time protocols to calibrate the clocks of each data source and compensate for transmission delays. The arrangement produces an ordered time-series stream that mixes discrete event points and continuous data points.

[0082] In practical implementation, each permutation unit generated by the interleaving arrangement is assigned a semantic label based on a medical knowledge base to form a nursing process event chain with unified semantic tags. The medical knowledge base contains a standard set of medical terms and logical definitions of clinical events, such as the SNOMEDCT terminology set and custom logical rules for surgical stages. When the permutation unit is a standardized event object, the semantic label identifies its clinical meaning; when the permutation unit is a data point in a time-stamped data point sequence, the semantic label identifies its parameter type and physiological state context. The time synchronization compensation process needs to consider the latency of the data processing pipeline and introduces a compensation value based on historical latency statistics to ensure timing accuracy. Optionally, the real-time integration process continuously performs data quality checks. Data points from specific sensors that have completely unchanged values ​​over a period of time, or data points that seriously conflict with the physiological range of the same type of parameter, are marked as suspicious data points and given special attention in subsequent processing. The process of assigning semantic labels uses a combination of rule matching and lightweight reasoning. For example, if the value of a "blood oxygen saturation" data point sequence drops rapidly from 98% to below 90% within 30 seconds, it will be assigned the semantic label "rapid drop in blood oxygen saturation event" and inserted as a special event into the nursing process event chain. The final generated nursing process event chain is a time-ordered and semantically clear sequence, where each element contains the original data, timestamp, data source identifier, and parsed and assigned semantic information.

[0083] In one embodiment of the present invention, a multi-level comparison is performed between the nursing process event chain and the execution benchmark to identify potential risk segments in the nursing process event chain that deviate from the execution benchmark. Specifically, the multi-level comparison includes checking the completeness of the event sequence, calculating the deviation of the execution time sequence, and verifying the consistency of the changing trends of key physiological parameters. The event sequence of the nursing process event chain is matched with the typical nursing path of the corresponding surgical type in the execution benchmark, and missing or redundant events are marked to generate an event completeness comparison result. The time difference between the actual occurrence time of the key operation in the nursing process event chain and the preset time in the execution benchmark is calculated. When the time difference exceeds the tolerance threshold, a time sequence deviation warning is generated. The data point sequence of specific physiological parameters in the nursing process event chain is analyzed to determine whether its change curve continuously exceeds the normal reference range defined by the execution benchmark or deviates from the expected trend, generating a parameter anomaly warning. By combining the event completeness comparison result, the time sequence deviation warning, and the parameter anomaly warning, the period of continuous abnormality in the nursing process event chain is identified and defined as a potential risk segment. The specific implementation of calculating time difference involves extracting the preset expected execution time and corresponding tolerance threshold range for each key operation from the execution benchmark, locating the actual occurrence timestamp of each key operation in the nursing process event chain, calculating the absolute time difference between the actual occurrence time of each key operation and the preset expected execution time, comparing the absolute time difference with the corresponding tolerance threshold range, and generating a time offset warning containing specific time difference information and key operation identifiers when the absolute time difference exceeds the upper or lower limit of the tolerance threshold range.

[0084] In practice, the nursing process event chain is compared with the execution benchmark at multiple levels. These multiple levels include verifying the completeness of the event sequence, calculating the offset of the execution timeline, and checking the consistency of key physiological parameter trends. The completeness verification of the event sequence involves extracting all discrete events with semantic tags from the nursing process event chain in chronological order, forming a list of actual event sequences. This list is then matched against the standard event sequences contained in the typical nursing pathway for the corresponding surgical type in the execution benchmark. The standard event sequences are derived from pattern summaries of historical successful surgical cases. The matching process employs an alignment algorithm based on event semantic tags and contextual relationships. The algorithm outputs the correspondence between the actual event sequence and the standard event sequence, marking events present in the actual event sequence but missing in the standard event sequence (defined as redundant events) and events present in the standard event sequence but missing in the actual event sequence (defined as missing events). Based on the alignment results, an event completeness comparison result is generated. This result is a structured record listing all marked missing and redundant events and their approximate positions on the timeline.

[0085] In some embodiments, the offset calculation of the execution time sequence focuses on critical operations in the nursing process event chain. Critical operations are the specific operational steps corresponding to key decision points on a typical nursing path defined in the execution baseline. The time difference between the actual occurrence time of a critical operation in the nursing process event chain and a preset time in the execution baseline is calculated. Specifically, this involves extracting the preset expected execution time and corresponding tolerance threshold range for each critical operation from the execution baseline. The preset expected execution time is defined with the surgical critical milestone as the relative time zero point, and the tolerance threshold range defines a time interval that allows the actual execution time to be earlier or later. The actual timestamp of each critical operation is located in the nursing process event chain. This location process is accomplished by matching the semantic tags of the critical operations with the semantic tags of events in the nursing process event chain. For each successfully matched critical operation, its absolute time difference is calculated. The formula for calculating the absolute time difference is:

[0086]

[0087] in: Indicates the absolute time difference. Indicates the actual time when the critical operation occurs. This represents the preset expected execution time of the critical operation. The calculated absolute time difference is compared to the corresponding tolerance threshold range, which is typically represented as a lower limit. and an upper limit value The interval formed When the absolute time difference exceeds the upper limit of the tolerance threshold, i.e. Or, although the absolute time difference falls within When a time-series offset warning is generated when the actual occurrence time within the time interval deviates from the preset expected execution time by a sign (earlier or later) that violates clinical requirements. The generated time-series offset warning is a message containing specific time difference information, key operation identifiers, and the direction of offset (earlier or later).

[0088] The consistency test for key physiological parameter trends analyzes the sequence of data points for specific physiological parameters within the nursing process event chain, such as mean arterial pressure, heart rate, and end-tidal carbon dioxide partial pressure. The consistency test determines whether the change curve consistently exceeds the normal reference range defined by the execution baseline or deviates from the expected trend. The normal reference range defined by the execution baseline may vary depending on the surgical stage and the patient's baseline condition. The analysis employs a sliding time window mechanism, sliding a time window with a fixed step size across the data point sequence. Within each window, the proportion of data points whose physiological parameter values ​​fall within the normal reference range is counted, and the local trend of the parameter change curve within that window is calculated. If the proportion of data points within several consecutive sliding time windows is lower than a preset threshold, it is considered to be consistently exceeding the normal reference range, generating a parameter anomaly alarm. If the local trend of the parameter change curve is opposite to the trend expected based on the typical nursing pathway in the execution baseline, and the difference exceeds the allowable range, it is considered to be deviating from the expected trend, also generating a parameter anomaly alarm. The parameter anomaly alarm includes the parameter type, abnormal time period, abnormal nature, and specific abnormal value.

[0089] In practical implementation, the identification of potential risk segments requires a comprehensive analysis of event integrity comparison results, time-series deviation warnings, and parameter anomaly alarms. The system maintains a risk marker line parallel to the timeline of the nursing process event chain. The time intervals covered by missing or redundant events marked in the event integrity comparison results, the occurrence times of key operations associated with time-series deviation warnings, and the abnormal time periods corresponding to parameter anomaly alarms are all mapped onto this risk marker line and then overlaid and merged. Continuously abnormal time periods in the nursing process event chain are delineated; that is, those periods with densely packed or consecutively occurring risk markers are identified on this risk marker line. These periods may contain multiple types of deviations simultaneously. These delineated, continuously abnormal time periods are defined as potential risk segments. Each potential risk segment is described by a start timestamp, an end timestamp, and all specific deviation information contained within this time period.

[0090] In one embodiment of the present invention, a deep contextual backtracking analysis is triggered for each identified potential risk segment. This analysis extracts a complete operating room state snapshot within a specific time window before and after the occurrence of the potential risk segment. Specifically, the complete operating room state snapshot covers personnel activities, equipment operation, medication records, and environmental indicators. A contextual backtracking time window is defined by extending forward and backward by a preset duration based on the start time of the potential risk segment. All relevant records falling within the contextual backtracking time window are retrieved and extracted from the operating room's full log database and video storage system. The extracted records are categorized and merged according to personnel, equipment, medication, and environmental dimensions to form a multi-dimensional, time-aligned complete operating room state snapshot. A dynamically evolving causal inference network is constructed based on the complete operating room state snapshot. By traversing the causal inference network, possible causal chains leading to the occurrence of potential risk segments are inferred. Specifically, nodes in the causal inference network represent state elements, and edges represent the interaction relationships between elements. Each independent state element in the complete operating room state snapshot is mapped to a node in the causal inference network. Causal relationship edges between nodes are established based on a medical rule base and a causal knowledge graph, varying over time. These edges have directionality and confidence weights. Starting with the state element corresponding to a potential risk segment, a reverse breadth-first search is performed along the causal edges in the causal inference network, enumerating all upstream node paths that can reach the starting point. All upstream node paths are sorted according to the confidence weights of the edges on the paths and the temporal proximity of the nodes. The top-ranked paths are selected as possible causal chains. Specifically, the reverse breadth-first search involves marking the state element corresponding to the potential risk segment as the search starting node set, initializing the search queue by adding all nodes from the starting node set, and then retrieving the current node from the search queue and traversing all upstream nodes pointing to the current node along the incoming edge direction. For each upstream node, record the path information from the starting node to the upstream node and add the upstream node to the search queue. Repeat the node retrieval and upstream node traversal operations until the search queue is empty or the preset search depth limit is reached. Collect all reverse search paths starting from the starting node to generate a set of possible causal paths.

[0091] In practice, a deep contextual retrospective analysis is triggered for each identified potential risk segment. This analysis extracts a complete snapshot of the operating room's status within a specific time window before and after the occurrence of the potential risk segment. This snapshot includes personnel activities, equipment operation, medication records, and environmental indicators. Based on the start time of the potential risk segment, a preset time window is extended forward and backward. This preset time window is configured according to the type of surgery and the nature of the potential risk. For example, a longer time window may be used for drug-related risks, while a shorter time window may be used for sudden abnormal physiological parameters. This is how the contextual retrospective time window is defined. All relevant records falling within the contextual retrospective time window are retrieved and extracted from the operating room's full log database and video storage system. The full log database integrates the operation logs and alarm records of all networked devices such as anesthesia machines, monitors, and infusion pumps. The video storage system provides a de-anonymized panoramic video stream of the operating room and close-up video streams of the key equipment's control panels. The extracted records were categorized and fused according to personnel, equipment, medication, and environmental dimensions. Personnel records included entry and exit records of medical staff, role location data, and voice command text. Equipment records included equipment power-on / off status, operating modes, key parameter settings, and alarm history. Medication records included all records of medication dispensing, preparation, infusion initiation, and rate changes. Environmental records included monitored values ​​of operating room temperature, humidity, differential pressure, and anesthetic gas concentration in the air. The fusion process used a unified, high-precision timestamp to align and correlate records from all dimensions, forming a multi-dimensional, time-aligned, complete operating room status snapshot. This snapshot, at any given moment, contains a status description of all monitored elements within the operating room.

[0092] In some embodiments, a dynamically evolving causal deduction network is constructed based on a complete operating room state snapshot. Nodes in the causal deduction network represent state elements, and edges represent interactions between elements. Each independent state element in the complete operating room state snapshot is mapped to a node in the causal deduction network. Examples of independent state elements include "Nurse A is located on the left side of the patient's head," "The anesthesia machine is in volume-controlled ventilation mode," "The remifentanil infusion rate is 0.1 μg / kg / min," and "The operating room temperature is 21.5 degrees Celsius." Causal relationship edges between nodes are established over time based on a medical rule base and a causal knowledge graph. The medical rule base contains causal rules extracted from clinical guidelines and expert experience, while the causal knowledge graph describes broader causal and correlational relationships between medical concepts. Causal relationship edges have directionality and confidence weights. Directionality indicates the direction of influence of the causal relationship, and confidence weights indicate the strength or degree of confirmation of the causal relationship, with values ​​derived from the confidence level of the rule or the statistical strength of the relationship in the knowledge graph. Starting with the state elements corresponding to potential risk segments, a reverse breadth-first search is performed along the causal edges in the causal deduction network, enumerating all upstream node paths that can reach the starting point. Specifically, the reverse breadth-first search is implemented by marking the state elements corresponding to potential risk segments as the search starting node set. This set may contain multiple nodes; for example, if the potential risk segment is "hypotension," the corresponding state element nodes might include "mean arterial pressure below 65 mmHg" and "increased heart rate." A search queue is initialized by adding all nodes from the search starting node set. The current node is then removed from the search queue, and all upstream nodes pointing to the current node are traversed along the incoming edges. For each upstream node, the path information from the starting node to the upstream node is recorded, and the upstream node is added to the search queue. This node removal and upstream node traversal operation is repeated until the search queue is empty or a preset search depth limit is reached. The search depth limit prevents the search from proceeding indefinitely in an excessively large network. All reverse search paths originating from the starting node are collected, generating a set of possible causal paths.

[0093] Understandably, the set of possible causal paths obtained through searching typically contains multiple paths. It's necessary to sort all upstream node paths found based on the confidence weights of each edge on the path and the temporal proximity of the nodes. The overall confidence score of a path is calculated using the confidence weights of all edges on the path and the node temporal proximity factor. The node temporal proximity factor reflects the temporal proximity between the cause node and the result node; the shorter the time interval, the higher the factor value. The confidence score of a possible causal path is shown below. It can be calculated using the following formula:

[0094]

[0095] in: Representing a path Confidence score, Indicates the formation of a path The set of all causal edges, Representing an edge Confidence weights Indicates the timestamp of the result node (the node in the current search layer). The function represents the timestamp of the cause node (upstream node). It is a time proximity decay function, for example , The attenuation coefficient is... It is a path The quantities above are used for normalization. The calculated confidence scores are then used. All possible causal paths are sorted in descending order, and the top-ranked paths are selected as possible causal chains. These chains are presented in order from root cause to direct cause to risky outcome. See Table 1.

[0096] Table 1: Example clips of snapshots showing the complete operating room status

[0097]

[0098] In practice, the constructed causal deduction network is dynamically evolving. As the surgical procedure and snapshots of the complete operating room state are updated, nodes and edges in the network are added or removed. The timestamp attribute of nodes is used to determine the temporal rationality of causal relationships; only causal edges where the timestamp of the cause node is earlier than the timestamp of the result node are included in the search scope. After selecting several top-ranked paths as possible causal chains, each possible causal chain is output in a structured manner, including the sequence of nodes involved in the chain, the relationships between nodes, and the overall confidence score of the chain.

[0099] See Figure 4 This is a bar chart comparing the number of potential risk segments across different surgical types. It visually displays the number of potential risks identified at key nursing points in five common surgical procedures, serving as a crucial basis for assessing surgical nursing risks. The low-risk performance of orthopedic surgery can be used as a benchmark to analyze standardized experiences in its nursing processes and extend them to higher-risk surgical types. Nursing training can be tailored to the risk distribution shown in this chart, allowing for differentiated training content for nurses in different departments. For example, cardiothoracic surgery nurses could receive focused training on managing risks related to extracorporeal circulation. This chart validates the effectiveness of the operating room nursing decision support system, demonstrating its ability to accurately identify risk differences across different surgical types and providing data support for subsequent intelligent interventions.

[0100] In one embodiment of the present invention, deviation information of all potential risk segments is integrated with the corresponding possible causal chains to generate a structured set of decision intervention prompts. Specifically, a decision intervention prompt record is created for each potential risk segment. The decision intervention prompt record stores the position index of the potential risk segment in the nursing process event chain, a description of the deviation type, and a summary of the main contributing factors extracted from the possible causal chains. Based on the summary of the main contributing factors and a preset coping strategy mapping table, a set of corrective measures associated with the decision intervention prompt record is generated. All decision intervention prompt records are organized according to the timeline order of their associated potential risk segments to form the decision intervention prompt set. Each item in the decision intervention prompt set is associated with a specific position in the nursing process event chain and contains a set of corrective measures.

[0101] In practice, the process of integrating deviation information from all potential risk segments with their corresponding possible causal chains to generate a structured set of decision intervention prompts begins with creating a decision intervention prompt record for each potential risk segment. This record is a data object whose data structure is designed to encapsulate all analytical conclusions and intervention recommendations related to a specific potential risk segment. When creating a decision intervention prompt record, the system obtains the start and end timestamps and specific deviation details of the potential risk segment in the nursing process event chain from the output of the risk identification module, and obtains a sorted list of possible causal chains associated with that potential risk segment from the output of the cause deduction module.

[0102] In practice, the decision intervention prompt record stores the location index of the potential risk segment in the nursing process event chain, the deviation type description, and a summary of the main triggering factors extracted from the possible causal chain. The location index is determined by mapping the start and end timestamps of the potential risk segment to the global timeline of the nursing process event chain, providing a precise association between the decision intervention prompt record and the original data stream. The deviation type description is a field combining classification and text, comprehensively describing the specific type and degree of deviation from event integrity, temporal shift, or parameter anomaly involved in the potential risk segment, such as "the critical event 'administration of a loading dose of antibiotics' is missing, accompanied by a subsequent sustained increase in heart rate exceeding the threshold." The extraction process of the main triggering factor summary involves analyzing high-confidence paths in the possible causal chain ranking list, extracting the upstream, most fundamental cause node or the most influential intermediate node from these paths, and refining and summarizing the state element descriptions of these nodes to form a concise text summary, such as "Triggering factor summary: Delayed preoperative antibiotic infusion (root cause) may lead to an increased risk of intraoperative infection, resulting in inflammatory response and increased heart rate."

[0103] In some embodiments, a set of corrective actions associated with a decision intervention prompt record is generated based on a summary of the primary triggering factor and a pre-defined coping strategy mapping table. The pre-defined coping strategy mapping table is a knowledge base storing standardized coping action sets corresponding to typical deviations or triggering factors in various clinical scenarios. This mapping table is constructed based on clinical guidelines, expert consensus, and validated best practices. The process of generating the set of corrective actions involves using the summary of the primary triggering factor as the query key to perform a matching search within the coping strategy mapping table. Matching can be based on keyword matching or semantic similarity calculation. After finding a matching standardized coping action, the system instantiates these actions, that is, it fills in and specifies the variables in the action description according to the specific context of the current surgery. For example, for the triggering factor summary "remifentanil infusion rate too high," the matched standardized coping action might be "consider reducing the remifentanil infusion rate to range X and observe blood pressure and heart rate responses." The instantiation process replaces "range X" with a specific numerical range calculated based on the patient's weight and the intensity of the current surgical stimulus. Finally, a set of one or more specific, actionable corrective actions is generated and bound to the decision intervention prompt record.

[0104] It is understandable that organizing all decision intervention prompt records according to the timeline of their associated potential risk segments is the final step in forming the decision intervention prompt set. The system sorts all created decision intervention prompt records in ascending order based on the start timestamp of their associated potential risk segments. The sorted sequence of decision intervention prompt records constitutes the decision intervention prompt set. Each item in the decision intervention prompt set, i.e., each decision intervention prompt record, is associated with a specific time position in the nursing process event chain through its internally stored location index. Each record in the decision intervention prompt set fully includes the location index, deviation type description, summary of the main triggers, and a set of associated corrective actions. This organization allows decision support information to be presented in a timeline format, consistent with the actual progress of the surgery. For multiple potential risk segments that exist simultaneously or are very close in time, the order of their corresponding decision intervention prompt records in the set also reflects this temporal relationship, which helps nurses understand the development of risk events.

[0105] Optionally, the process of generating the decision intervention suggestion set may also include a prioritization step. This step, based on chronological organization, re-sorts or labels the decision intervention suggestion records according to the severity of potential risk segments and the confidence level of possible causal chains. Severity can be calculated comprehensively based on factors such as the category of deviation and the magnitude of abnormal parameter deviations from the baseline. A decision intervention suggestion record is assigned a comprehensive urgency score. It can be calculated as follows:

[0106]

[0107] in: Representing records The overall urgency score, This indicates the severity score of the potential risk segment corresponding to the record. This indicates the confidence score of the possible causal chain adopted by the record. and This is a coefficient used to balance the weights of severity and confidence. The final set of decision intervention suggestions can be arranged according to a comprehensive urgency score. Sort in descending order, or while sorting by time, append a key based on... Priority labels are used to more intuitively guide caregivers in the order of their attention.

[0108] See Figure 5 This is a heatmap showing the correlation strength between triggers and corrective actions in an operating room nursing decision-making intervention system. It visually displays the correlation strength (0.4–0.9) between different "primary trigger types" and "corrective action types" using color depth, with redder colors indicating stronger correlations. When the system identifies a trigger, nurses can quickly find the most relevant corrective action using this map, shortening decision-making time. This map reflects the "trigger-action" mapping relationship based on clinical data and expert experience, serving as a core basis for the system's response strategy mapping table. For mappings with low correlation strength but still existing, their clinical rationality can be further verified, optimizing the knowledge base. Analyzing weakly correlated areas can reveal potential loopholes in the existing intervention logic. This may indicate that the system needs to reassess the clinical rationality of these correlations or supplement any missing key measures.

[0109] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An operating room nursing key node decision support system, characterized by, The system includes: The benchmark establishment module establishes the execution benchmarks for operating room nursing operations. These benchmarks are extracted from nursing process data of historical successful surgical cases and include the expected execution sequence and allowable parameter fluctuation range for each standard nursing step. The data integration module connects to the real-time nursing operation flow of the ongoing surgery and integrates the discrete operation actions and continuous monitoring data in the real-time nursing operation flow into a nursing process event chain with unified semantic tags according to the time axis. The risk identification module compares the nursing process event chain with the execution benchmark at multiple levels to identify potential risk segments in the nursing process event chain that deviate from the execution benchmark. The scenario backtracking module triggers a deep scenario backtracking analysis for each identified potential risk segment. The deep scenario backtracking analysis is used to extract a complete operating room state snapshot within a specific time window before and after the occurrence of the potential risk segment. The cause deduction module constructs a dynamically evolving cause deduction network based on the complete operating room state snapshot, and infers the possible causal chains that lead to the occurrence of the potential risk segment by traversing the cause deduction network. The decision generation module integrates the deviation information of all the potential risk segments with the corresponding possible causal chains to generate a structured set of decision intervention prompts; For each identified potential risk segment, a deep contextual retrospective analysis is triggered. This analysis is used to extract a complete snapshot of the operating room status within a specific time window before and after the occurrence of the potential risk segment. Specifically, this is implemented as follows: The complete operating room status snapshot covers personnel activities, equipment operation, medication records, and environmental indicators; Based on the start time of the potential risk segment, a preset duration is extended forward and backward to define the scenario retrospective time window; Retrieve and extract all relevant records that fall within the aforementioned scenario recall time window from the full log database and video storage system of the operating room; The extracted records are categorized and integrated according to personnel, equipment, medication, and environment dimensions to form a multi-dimensional, time-aligned snapshot of the complete operating room status. The step of constructing a dynamically evolving cause-effect inference network based on the complete operating room state snapshot, and inferring the possible causal chains leading to the occurrence of the potential risk segment by traversing the cause-effect inference network, is specifically implemented as follows: The nodes of the causal deduction network represent state elements, and the edges represent the interaction relationships between elements; Each individual state element in the complete operating room state snapshot is mapped to a node of the cause-and-effect network; Based on a medical rule base and a causal knowledge graph, causal relationship edges between nodes are established that change over time, and these causal relationship edges have directionality and confidence weights. Taking the state element corresponding to the potential risk segment as the starting point, perform a reverse breadth-first search along the causal edge in the cause inference network to enumerate all upstream node paths that can reach the starting point. Based on the confidence weights of each edge on the path and the temporal proximity of the nodes, all upstream node paths found are sorted, and the top-ranked paths are selected as the possible causal chains.

2. The decision support system for key nodes in operating room nursing according to claim 1, characterized in that, The establishment of the operating room nursing procedure execution standards is specifically implemented as follows: Collect and clean the complete records of historical successful surgical cases, and extract the action instruction sequences, physiological parameter response curves and event trigger logs related to nursing operations; Clustering methods were used to group the historical successful surgical cases according to surgical type and complexity. The nursing operation data of each group were summarized to form typical nursing pathways for the corresponding groups. Key decision points are marked on the typical nursing pathway, and for each key decision point, a set of preceding events, a set of succeeding events, and a normal reference range for associated monitoring parameters are defined, which together constitute the execution benchmark.

3. An operating room nursing key node decision support system according to claim 2, characterized in that, The access to the real-time nursing operation flow during surgery involves integrating discrete operational actions and continuous monitoring data within the real-time nursing operation flow into a nursing process event chain with unified semantic tags, specifically implemented as follows: Simultaneously receive raw data packets from different operating room information systems and sensor networks; The raw data messages are parsed to convert instrument operation signals, personnel voice commands, and drug infusion records into standardized event objects, and vital sign waveforms and gas concentration readings into timestamped data point sequences. The standardized event objects are interwoven with the timestamped data point sequence in chronological order, and each arrangement unit is assigned a semantic label based on the medical knowledge base to form the nursing process event chain.

4. An operating room nursing key node decision support system according to claim 3, characterized in that, The step of performing a multi-level comparison between the nursing process event chain and the execution benchmark to identify potential risk segments in the nursing process event chain that deviate from the execution benchmark is specifically implemented as follows: The multi-level comparison includes the integrity check of the event sequence, the calculation of the offset of the execution time sequence, and the consistency test of the changing trends of key physiological parameters. The event sequence of the nursing process event chain is matched with the typical nursing path of the corresponding surgical type in the execution benchmark, missing or redundant events are marked, and event integrity comparison results are generated. Calculate the time difference between the actual occurrence time of key operations in the nursing process event chain and the preset time in the execution benchmark. When the time difference exceeds the tolerance threshold, generate a time sequence offset warning. Analyze the data point sequence of specific physiological parameters in the nursing process event chain, determine whether the change curve continuously exceeds the normal reference range defined by the execution benchmark or deviates from the expected trend, and generate parameter abnormality alarm; Based on the combined results of the event integrity comparison, the time sequence offset warning, and the parameter anomaly alarm, the time periods with continuous anomalies in the nursing process event chain are identified and defined as the potential risk segments.

5. An operating room nursing key node decision support system according to claim 4, characterized in that, The process of integrating deviation information from all potential risk segments with the corresponding possible causal chains to generate a structured set of decision intervention prompts is specifically implemented as follows: Create a decision intervention prompt record for each of the aforementioned potential risk segments; The decision intervention prompt record is associated with the location index of the potential risk segment in the nursing process event chain, the deviation type description, and the summary of the main causes extracted from the possible causal chain; Based on the summary of the main causes and the preset coping strategy mapping table, a set of corrective measures associated with the decision intervention prompt records is generated; All decision intervention prompts are organized in chronological order according to their associated potential risk segments to form the decision intervention prompt set. Each item in the decision intervention prompt set is associated with a specific position in the nursing process event chain and contains the set of corrective actions.

6. An operating room nursing key node decision support system according to claim 2, characterized in that, The clustering method is used to group the historical successful surgical cases according to surgical type and complexity, specifically as follows: Extract the surgical type code and surgical complexity score for each historical successful surgical case to construct a case feature vector; The density clustering algorithm is used to perform unsupervised clustering analysis on the feature vectors of the cases to identify groups of historical successful surgical cases with similar features; The silhouette coefficient of each cluster group is evaluated, and the target groups that meet the preset conditions for cluster quality are selected. Pattern summarization was performed on the nursing operation data of historical successful surgical cases within each target group, and statistical characteristics of typical nursing pathways within the target group were calculated.

7. An operating room nursing key node decision support system according to claim 4, characterized in that, The calculation of the time difference between the actual occurrence time of key operations in the nursing process event chain and the preset time in the execution benchmark, and the generation of a time sequence offset warning when the time difference exceeds the tolerance threshold, is specifically implemented as follows: Extract the preset expected execution time and corresponding tolerance threshold range for each key operation from the execution benchmark; Locate the actual timestamp of each critical operation in the nursing process event chain; Calculate the absolute time difference between the actual occurrence time and the preset expected execution time for each critical operation; The absolute time difference is compared with the corresponding tolerance threshold range. When the absolute time difference exceeds the upper or lower limit of the tolerance threshold range, a timing offset warning containing specific time difference information and key operation identifiers is generated.

8. An operating room nursing key node decision support system according to claim 7, characterized in that, The step of taking the state element corresponding to the potential risk segment as the starting point and performing a reverse breadth-first search along the causal edges in the cause-effect inference network to enumerate all upstream node paths that can reach the starting point is specifically implemented as follows: Mark the state elements corresponding to the potential risk segments as the search starting node set; Initialize the search queue by adding all nodes from the starting node set to the search queue; Remove the current node from the search queue and traverse all upstream nodes pointing to the current node along the incoming edge direction; For each upstream node, record the path information from the starting node to the upstream node, and add the upstream node to the search queue; Repeat the node retrieval and upstream node traversal operations until the search queue is empty or the preset search depth limit is reached; Collect all reverse search paths starting from the starting node to generate a set of possible causal paths.

Images