A VTE intelligent monitoring system supporting multi-role requirements

Abstract

The application relates to the technical field of medical data processing, and particularly relates to a VTE intelligent monitoring system supporting multi-role requirements. The following scheme is provided: a doctor role end constructs an evidence hypergraph based on multi-source clinical data, generates an evidence correlation vector through partial order conflict grid merging, searches in a guideline version dictionary tree to obtain a current effective node, outputs prevention and treatment suggestions and version pointing information to form a traceable evidence chain. A nursing role end receives and analyzes the traceable chain, screens matched prevention and treatment schemes from an intervention measure set in combination with real-time monitoring data, generates nursing execution instructions and dynamically adjusts and feeds back according to intervention effects, and realizes closed-loop cooperation of risk assessment, intervention execution and result back-checking among the multi-roles, so that the timeliness, traceability and execution consistency of the prevention and treatment suggestions are guaranteed.

Description

Technical Field

[0001] This application relates to the field of medical data processing technology, and in particular to a VTE intelligent monitoring system that supports the needs of multiple roles. Background Technology

[0002] Currently, VTE risk monitoring and prevention in hospitals largely rely on manual assessment or static rule engines. The monitoring process is fixed and lacks continuous tracking of patients' dynamic risks. Existing technologies cannot achieve real-time collaboration and information consistency among different roles when faced with multi-source, heterogeneous clinical data, leading to a disconnect between assessment results and actual changes in the patient's condition. Physician-side prevention recommendations lack traceable evidence corresponding to guideline versions, making assessment logic difficult to verify and audit. Nursing staff often experience frequent assessment triggers with low effective intervention rates, redundant reminders, and incomplete closed-loop processes, failing to establish a unified risk management path. Due to data update delays and fragmented assessment processes, existing solutions are insufficient in terms of assessment accuracy, execution continuity, and feedback verifiability, making it difficult to meet the needs of accurate VTE risk identification and dynamic prevention in a multi-role collaborative environment within the hospital.

[0003] To address the above issues, this application presents a VTE intelligent monitoring system that supports the needs of multiple roles. Summary of the Invention

[0004] The technical problem this application aims to solve is to address the shortcomings of existing technologies by providing a VTE intelligent monitoring system that supports the needs of multiple roles. On the physician side, an evidence hypergraph is constructed based on multi-source clinical data. Evidence association vectors are generated through partial-order conflict lattice merging, and the currently effective node is retrieved from the guideline version trie. Prevention and treatment recommendations and version-specific information are output to form a traceable evidence chain. On the nursing side, the traceability chain is received and parsed. Combined with real-time monitoring data, matching prevention and treatment plans are selected from the intervention set, generating nursing execution instructions and dynamically adjusting and providing feedback based on the intervention effects. By introducing cross-version mapping and temporal rule constraints, a closed-loop collaboration of risk assessment, intervention execution, and result verification among multiple roles is achieved, ensuring the timeliness, traceability, and consistency of implementation of prevention and treatment recommendations.

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] A VTE intelligent monitoring system supporting multiple roles includes a physician role terminal configured with a guideline version dictionary tree, and a nursing role terminal used for evaluation orchestration and closed-loop quality control based on the output of the physician role terminal. The physician role terminal includes:

[0007] A multi-source data processing module is used to acquire multi-source data, construct an evidence hypergraph based on the multi-source data, and merge the partial-order conflict lattices of the evidence hypergraph to obtain an evidence association vector.

[0008] The retrieval module is used to search the guide version trie using the evidence association vector to obtain the currently active node. The retrieval includes generating triggered connection edges and virtual connection edges in the guide version trie based on the evidence association vector. The triggered connection edges are used to locate candidate entry nodes that match the evidence association vector.

[0009] The evidence witness set generation module is used to generate a traceability chain including the evidence witness set based on the VTE prevention recommendations and version pointing information output by the currently effective node.

[0010] The communication module is used to establish two-way communication, synchronize the traceability chain to the nursing role terminal, and receive result feedback from the nursing role terminal.

[0011] The multi-source data processing module includes:

[0012] The data receiving unit is used to receive multi-source data and convert the multi-source data into a standardized format. The multi-source data includes at least one of the following: diagnostic information, test results, imaging reports, surgical records, medical order data, medical records, and medication information.

[0013] The semantic parsing unit is used to perform semantic parsing on multi-source data in a standardized format;

[0014] The time-series annotation unit is used to annotate the time information in multi-source data in a standardized format.

[0015] The data analysis unit is used to analyze the outputs of the semantic parsing unit and the temporal annotation unit to obtain the evidence association vector;

[0016] The multi-source data processing module further includes processing logic, which is used to construct an evidence hypergraph based on the semantic entity category, contextual relationship and effective time interval of each data, and generate an evidence association vector based on the evidence hypergraph. The processing logic is configured within the data analysis unit.

[0017] The processing logic is configured as follows:

[0018] Based on the semantic entity categories, semantic entities in multi-source data are nodeized to construct an initial evidence hypergraph with scale factors, diagnostic elements, test indicators, and disease course events as nodes.

[0019] The initial evidence hypergraph is layered and associated with enhanced processing based on the contextual relationships and the effective time intervals, including establishing dependency edges between nodes within the same time layer and establishing evolutionary edges across time layers, in order to generate a time-layered evidence hypergraph.

[0020] For multi-source data pointing to the same semantic node, a partial order conflict lattice is constructed in each time layer of the evidence hypergraph, wherein the elements of the partial order conflict lattice are characterized by a quadruple of value range, source level, sampling timestamp and record type.

[0021] Within each time layer, the top element of the partially ordered conflicting lattices of the same semantic node is merged to obtain a unified representation within the layer. Then, the unified representations within the layer are converged layer by layer along the inter-layer evolution edge according to the order of near to far to obtain the merging result.

[0022] The evidence vector of each semantic node is calculated based on the merging results, and the semantic nodes are combined according to the dependency edges to obtain the evidence association vector.

[0023] The retrieval module includes:

[0024] A triggerable connection edge generation unit is used to generate triggerable connection edges between nodes of the guide version trie based on the semantic factors and time attributes of the evidence association vector, to obtain a triggerable pre-index, wherein the pre-index is used to locate and identify candidate entry nodes that match the evidence association vector during retrieval.

[0025] A virtual connection edge generation unit is used to generate virtual connection edges based on cross-version entry mapping relationships when there is a difference between the version information of the candidate entry node and the evidence association vector, so as to perform conditional redirection between different version entries;

[0026] The retrieval path optimization unit is used to calculate the matching degree of the candidate entry nodes by combining the triggered connection edge and the virtual connection edge, so as to determine the currently effective node;

[0027] The retrieval module further includes a retrieval strategy, which is used to optimize the retrieval path. The retrieval strategy includes trigger matching logic and version adaptation logic. The trigger matching logic is configured in the trigger-based connection edge generation unit, and the version adaptation logic is configured in the virtual connection edge generation unit.

[0028] The trigger matching logic is configured as follows:

[0029] Based on the semantic factors, time attributes and confidence in the evidence association vector, a time-series rule matrix is ​​constructed. The time-series rule matrix analyzes the evidence in different time windows, records the time relationship and change pattern between the evidence, and obtains the optimized path in the time dimension during the retrieval process.

[0030] Based on the time-series rule matrix, the nodes of the guide version trie are filtered to obtain multiple pre-indexes;

[0031] In the preceding index, candidate entry nodes corresponding to the preceding index are connected by triggered connection edges according to the time priority information corresponding to the time sequence rule matrix.

[0032] Based on the time-series rule matrix, the nodes of the guide version trie are filtered to obtain multiple preceding indices, including:

[0033] Based on the temporal dependency relationship of the temporal rule matrix, the corresponding flow pattern is obtained, and an evidence flow path diagram is generated. The nodes of the evidence flow path diagram represent evidence items, the edges represent causal relationships between evidence, and the evidence flow path diagram represents the reasoning and dependency relationship of evidence in the time dimension.

[0034] Identify the corresponding jump nodes based on the evidence flow path diagram, where a jump node represents an evidence node that spans multiple time windows in the time series;

[0035] Based on the temporal position, timeliness, and clinical relevance of the skip nodes, the skip nodes are prioritized to generate multiple pre-indexes.

[0036] The version adaptation logic is configured as follows:

[0037] The entries in the guide version dictionary are compared to identify the changes between the current version and historical versions, including added entries, modified entries, and deleted entries.

[0038] When the semantic similarity between the version information of the candidate entry node and the evidence association vector is less than a preset matching threshold, a virtual connection edge is generated based on the changed content and the cross-version entry mapping relationship.

[0039] In the process of generating the virtual connection edge, the redirected version entries are prioritized based on the time span between versions and the evidence change pattern using a time window discount factor. The time window discount factor is calculated by combining the time difference between versions and the evidence change pattern with a preset decay function. When redirecting across versions, the candidate version entries are weighted and matched by combining the confidence and conflict levels in the evidence association vector, and the corresponding version entry is selected based on the weighted matching result.

[0040] The nursing role includes:

[0041] The data acquisition module is used to collect monitoring data, receive a traceability chain containing a set of evidence witnesses from the doctor's role, and parse the traceability chain to obtain VTE prevention and treatment recommendations and the corresponding evidence chain of the VTE prevention and treatment recommendations.

[0042] The instruction generation module, based on the VTE prevention and control recommendations and the monitoring data, filters matching prevention and treatment measures from the intervention set and generates nursing execution instructions;

[0043] The adjustment module is used to collect real-time physiological data during nursing intervention, determine whether the intervention effect matches the expected response conditions in the evidence chain based on the real-time physiological data, and if not, reconstruct the nursing execution instructions based on the intervention effect and feed the reconstruction result back to the doctor role to update the guideline version dictionary tree.

[0044] The instruction generation module includes:

[0045] The preliminary screening unit is used to analyze the current physiological state based on the VTE prevention and control recommendations and the monitoring data, and generate initial screening conditions.

[0046] A refined screening unit is used to generate final screening conditions based on the initial screening conditions and the object's past history, wherein the past history includes past treatment responses, pathological examination results and the object's individual characteristics.

[0047] The screening unit, based on the final screening criteria, selects matching prevention and treatment measures from the set of interventions.

[0048] The adjustment module includes:

[0049] Real-time data acquisition unit, used to monitor the patient's physiological data in real time;

[0050] The intervention effect evaluation unit is used to compare real-time collected physiological data with expected response conditions, calculate the intervention effect matching degree, and determine whether the current intervention measures have achieved the expected effect based on the matching degree.

[0051] The dynamic adjustment unit is used to adjust the nursing execution instructions based on the output of the intervention effect evaluation unit when the intervention effect does not meet expectations, and synchronously feed back to the doctor's role.

[0052] The adjustment module further includes evaluation logic, which is configured within the intervention effect evaluation unit. The evaluation logic is configured as follows:

[0053] A matching degree is generated by comparing the real-time physiological data with the expected response conditions;

[0054] The intervention effect matching degree is calculated based on the matching degree. If the intervention effect matching degree is less than the preset response threshold, it means that the intervention effect corresponding to the current prevention and treatment measures does not match the expected response conditions.

[0055] Compared with the prior art, the beneficial effects of this application are:

[0056] This application achieves a collaborative closed loop between physician-level risk assessment and nurse-level intervention execution by constructing a version dictionary tree based on medical guidelines and a multi-source evidence association model. When faced with heterogeneous and asynchronous clinical data, the solution can dynamically generate prevention and control recommendations while maintaining consistency with guideline versions, ensuring the traceability and replayability of assessment results. Nursing staff use the traceability chain to filter instructions and provide real-time feedback, enabling prevention and control measures to adaptively adjust according to changes in the patient's condition. Attached Figure Description

[0057] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0058] Figure 1 An exemplary application scenario diagram provided for an embodiment of this application;

[0059] Figure 2 A schematic diagram of a VTE intelligent monitoring system supporting multiple role requirements provided in an embodiment of this application;

[0060] Figure 3 A flowchart illustrating a VTE intelligent monitoring method supporting multiple role requirements provided in an embodiment of this application;

[0061] Figure 4 A schematic diagram of the structure of the trie provided in the embodiments of this application;

[0062] Figure 5 This is a flowchart illustrating another VTE intelligent monitoring method that supports multiple role requirements, provided in an embodiment of this application. Detailed Implementation

[0063] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0064] The term "embodiment" as used herein means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0065] In the management of hospitalized patients, risk assessment and prevention of venous thromboembolism (VTE) are carried out across multiple nursing and medical order trigger points, including the perioperative period, long-term bed rest in internal medicine, high-risk obstetrics, cancer treatment cycles, and interdepartmental transfers.

[0066] Understandably, in a real hospital environment, medical procedures do not unfold linearly:

[0067] Diagnostic, laboratory, imaging, surgical, and medical order information is often written to information systems asynchronously. Furthermore, the time granularity of medical records and consultation opinions is inconsistent, leading to a disconnect between different roles regarding when to assess, under what criteria, how recommendations are implemented, and how to track progress. A long-standing common practice has been to trigger corresponding actions within fixed time windows (such as 24 hours after admission, 48 hours after surgery, or 24 hours after transfer to another department), or to use a static rule engine in electronic medical records to compare several key fields and display prompts.

[0068] Understandably, conventional practices often lead to a dilemma when evidence is incomplete or contradictory, when guideline provisions evolve across versions, and when there is an uneven workload in wards:

[0069] Doctors are unable to provide sufficient explanations regarding the traceability of recommendations and evidence, while nurses are subjected to frequent but inefficient reminders, and closed-loop confirmation relies on manual record-keeping and is difficult to audit.

[0070] The VTE intelligent monitoring system provided in this embodiment does not assume a unified, fixed assessment rhythm within the hospital, nor does it require prior division of proprietary procedures for a certain department / operative. Instead, it revolves around the goals of traceable evidence, version-identifiable data, and closed-loop execution, forming a consistent and replayable assessment logic for the same patient from different perspectives. The core processing logic of this application can be summarized as follows:

[0071] On the physician's side, heterogeneous clinical data is mapped into chains of evidence. Each piece of evidence in the chain carries source, timeliness, and consistency markers. When there are value conflicts or time decay, uncertainty is characterized according to standardized merging rules. When linking risk and evidence to clinical guidelines, the system does not simply match clause numbers, but enhances the search path within the guideline version trie.

[0072] On the one hand, triggerable connection edges are generated in real time based on the current evidence pattern, enabling the rapid location of similar semantics and entries with the same time window;

[0073] On the other hand, when there are differences in thresholds or applicable conditions between historical and current guidelines for an entry, virtual connection edges are dynamically generated to complete condition redirection, thereby ensuring the consistency between the currently effective node and the evidence and providing a clear version indication.

[0074] Meanwhile, the nursing role does not replicate the reasoning of the physician's side. Instead, it parses the traceability chain containing the evidence witness set and generates executable nursing instructions under constraints such as the actual workload of the ward, postoperative window, transfer time, and equipment occupancy. If a vital sign trajectory that is inconsistent with the expected response of the traceability chain is detected during execution, the system triggers a lightweight re-evidence and only performs differential updates to the relevant time window, ensuring a balance between reminder density and closed-loop quality. All key actions and parameter snapshots are recorded in a verifiable event structure to meet subsequent medical quality control and record-keeping requirements.

[0075] This embodiment is applicable to a variety of typical scenarios within a hospital, not limited to a single department: such as the coordinated process of cancer patients receiving catheterization and chemotherapy, the multi-node transfer from the operating room to the anesthesia recovery room to the ward after orthopedic surgery, the short-window assessment of the hypercoagulable state in obstetric patients during the peripartum period, and the reassessment of critically ill internal medicine patients when they are transferred from the ICU to a general ward.

[0076] The commonalities of the aforementioned scenarios are: asynchronous evidence writing, alternating versions of guideline clauses, assessment timing constrained by nursing workload and clinical window, and the need to form an auditable closed loop within the quality control cycle.

[0077] Those skilled in the art will understand that this embodiment achieves dynamic assessment and closed-loop control of VTE risk by establishing an intelligent assessment and collaboration mechanism for both doctors and nurses. The doctor's role focuses on risk identification and suggestion generation, integrating and analyzing multi-source data from the electronic medical record system through a structured rule model to automatically provide prevention and treatment suggestions consistent with the latest clinical guidelines. The nurse's role focuses on implementation and feedback, generating individualized nursing measures based on the prevention and treatment suggestions output by the doctor and combined with real-time patient monitoring data, and continuously tracking and dynamically adjusting patient responses during implementation. Both roles maintain data and decision consistency through a traceability chain, enabling the system to maintain continuous and auditable assessment logic even with multiple assessments, interdepartmental transfers, and version updates, thereby improving the timeliness of risk identification and the efficiency of nursing intervention.

[0078] refer to Figure 1 , Figure 1 This is an exemplary application scenario diagram provided for an embodiment of this application.

[0079] Figure 1 The application scenario illustrates a multi-role collaborative environment within a hospital information system, where doctors and nurses are each connected to the same VTE intelligent prevention and control system (not shown in the figure). The doctor's end integrates and analyzes multi-source clinical data from patients to generate a traceability chain. The nurse's end executes corresponding prevention and control measures and nursing interventions in the ward based on the traceability chain output by the doctor's end. The system achieves bidirectional data interaction through the hospital network. The doctor's end can receive real-time feedback on nursing execution and update the guideline version dictionary tree, while the nurse's end adjusts subsequent assessments and prevention and control operations based on the updated guidance information, thus forming a closed-loop collaborative dynamic assessment scenario between clinical diagnosis and treatment and nursing execution.

[0080] Figure 1 The client-side interface is also shown, which is understood to be specifically the patient's client-side interface.

[0081] refer to Figure 2 , Figure 2 This is a schematic diagram of a VTE intelligent monitoring system that supports multiple role requirements, provided as an embodiment of this application.

[0082] In one example, the VTE intelligent monitoring system includes a physician role terminal, which is configured with a guideline version dictionary tree built based on medical guidelines. The VTE intelligent monitoring system also includes a nursing role terminal, which is used for evaluation orchestration and closed-loop quality control based on the output of the physician role terminal. The physician role terminal includes:

[0083] A multi-source data processing module is used to acquire multi-source data, construct an evidence hypergraph based on the multi-source data, and merge the partial-order conflict lattices of the evidence hypergraph to obtain an evidence association vector.

[0084] The retrieval module is used to search the guide version trie using the evidence association vector to obtain the currently active node. The retrieval includes generating triggered connection edges and virtual connection edges in the guide version trie based on the evidence association vector. The triggered connection edges are used to locate candidate entry nodes that match the evidence association vector.

[0085] The evidence witness set generation module is used to generate a traceability chain including the evidence witness set based on the VTE prevention recommendations and version pointing information output by the currently effective node.

[0086] The communication module is used to establish two-way communication, synchronize the traceability chain to the nursing role terminal, and receive result feedback from the nursing role terminal.

[0087] The multi-source data processing module includes:

[0088] The data receiving unit is used to receive multi-source data and convert the multi-source data into a standardized format. The multi-source data includes at least one of the following: diagnostic information, test results, imaging reports, surgical records, medical order data, medical records, and medication information.

[0089] The semantic parsing unit is used to perform semantic parsing on multi-source data in a standardized format;

[0090] The time-series annotation unit is used to annotate the time information in multi-source data in a standardized format.

[0091] The data analysis unit is used to analyze the outputs of the semantic parsing unit and the temporal annotation unit to obtain the evidence association vector;

[0092] The multi-source data processing module further includes processing logic, which is used to construct an evidence hypergraph based on the semantic entity category, contextual relationship and effective time interval of each data, and generate an evidence association vector based on the evidence hypergraph. The processing logic is configured within the data analysis unit.

[0093] The retrieval module includes:

[0094] A triggerable connection edge generation unit is used to generate triggerable connection edges between nodes of the guide version trie based on the semantic factors and time attributes of the evidence association vector, to obtain a triggerable pre-index, wherein the pre-index is used to locate and identify candidate entry nodes that match the evidence association vector during retrieval.

[0095] A virtual connection edge generation unit is used to generate virtual connection edges based on cross-version entry mapping relationships when there is a difference between the version information of the candidate entry node and the evidence association vector, so as to perform conditional redirection between different version entries;

[0096] The retrieval path optimization unit is used to calculate the matching degree of the candidate entry nodes by combining the triggered connection edge and the virtual connection edge, so as to determine the currently effective node;

[0097] The retrieval module further includes a retrieval strategy, which is used to optimize the retrieval path. The retrieval strategy includes trigger matching logic and version adaptation logic. The trigger matching logic is configured in the trigger-based connection edge generation unit, and the version adaptation logic is configured in the virtual connection edge generation unit.

[0098] In another example, the VTE intelligent monitoring system is applied to a nursing role, which is configured with a set of intervention measures. The VTE intelligent monitoring system also includes a physician role, which generates a traceability chain for prevention and control decision tracing and implementation feedback, and synchronizes the traceability chain to the nursing role. The nursing role includes:

[0099] The data acquisition module is used to collect monitoring data, receive a traceability chain containing a set of evidence witnesses from the doctor's role, and parse the traceability chain to obtain VTE prevention and treatment recommendations and the corresponding evidence chain of the VTE prevention and treatment recommendations.

[0100] The instruction generation module, based on the VTE prevention and control recommendations and the monitoring data, filters matching prevention and treatment measures from the intervention set and generates nursing execution instructions;

[0101] The adjustment module is used to collect real-time physiological data during nursing intervention, determine whether the intervention effect matches the expected response conditions in the evidence chain based on the real-time physiological data, and if not, reconstruct the nursing execution instructions based on the intervention effect and feed the reconstruction result back to the doctor role to update the guideline version dictionary tree.

[0102] In some optional implementations, the nursing role also includes a communication module for two-way communication with the doctor role.

[0103] The instruction generation module includes:

[0104] The preliminary screening unit is used to analyze the current physiological state based on the VTE prevention and control recommendations and the monitoring data, and generate initial screening conditions.

[0105] A refined screening unit is used to generate final screening conditions based on the initial screening conditions and the object's past history, wherein the past history includes past treatment responses, pathological examination results and the object's individual characteristics.

[0106] The screening unit, based on the final screening criteria, selects matching prevention and treatment measures from the set of interventions.

[0107] The adjustment module includes:

[0108] Real-time data acquisition unit, used to monitor the patient's physiological data in real time;

[0109] The intervention effect evaluation unit is used to compare real-time collected physiological data with expected response conditions, calculate the intervention effect matching degree, and determine whether the current intervention measures have achieved the expected effect based on the matching degree.

[0110] The dynamic adjustment unit is used to adjust the nursing execution instructions based on the output of the intervention effect evaluation unit when the intervention effect does not meet expectations, and synchronously feed back to the doctor's role.

[0111] The adjustment module also includes evaluation logic, which is configured within the intervention effect evaluation unit.

[0112] In some optional implementations, embodiments of this application provide a VTE intelligent monitoring method that supports multiple role requirements.

[0113] Next, with reference to the accompanying drawings, a VTE intelligent monitoring method supporting multi-role requirements provided by an embodiment of this application will be further described. Figure 3 The method shown is applied to the physician role in a VTE intelligent prevention and control system. The physician role includes a guideline version dictionary tree built based on medical guidelines. The VTE intelligent prevention and control system also includes a nursing role, which is used for evaluation, orchestration, and closed-loop quality control based on the output of the physician role. The method includes:

[0114] S1: Obtain multi-source data, construct an evidence hypergraph based on the multi-source data, merge the partial-order conflict lattices of the evidence hypergraph, and obtain the evidence association vector;

[0115] In this embodiment, multi-source data may include one or more of the following: diagnostic information, test results, imaging reports, surgical records, medical orders, medical records, and medication information. Through semantic parsing and temporal attribute annotation, various types of data undergo unified structural processing, enabling data from different sources and at different time granularities to be mapped at the same semantic level. Subsequently, using semantic entity categories as nodes, dependency edges and evolutionary edges are established on the time series to form an evidence hypergraph. For conflicting data appearing at the same node, a partial-order conflict lattice merging method is used, with data source level and time decay weight as the sorting criteria, to form a set of nodes with consistency and confidence representation, thereby generating an evidence association vector.

[0116] Those skilled in the art will understand that multi-source data can be determined according to the configuration of the actual acquisition system, as long as it is at a minimum sufficient to be able to construct evidence nodes and time dependencies based on the data, and this application does not impose any limitations on this.

[0117] S2: Use the evidence association vector to search the guide version trie to obtain the currently effective node;

[0118] In this embodiment, the doctor's role dynamically generates triggered connection edges between nodes in the guideline version trie based on the semantic factors and time attributes of the evidence association vector, enabling the system to quickly locate candidate entry nodes that match the current evidence. When the guideline version to which a candidate entry belongs is inconsistent with the version currently being evaluated, virtual connection edges are further generated based on the cross-version entry mapping relationship to achieve conditional redirection and cross-version jumps. During the retrieval process, the system comprehensively evaluates the semantic similarity, version timeliness, and threshold differences of candidate nodes based on the matching degree calculation logic, thereby obtaining the currently effective node and its corresponding guidance clause.

[0119] S3: Based on the VTE prevention recommendations and version information output by the currently active node, generate a traceability chain including an evidence witness set, and synchronize the traceability chain to the nursing role terminal;

[0120] In this embodiment, the physician role determines the type and intensity level of VTE prevention recommendations based on the current effective node, and establishes a traceable evidence chain for each recommendation by combining the generated evidence association vector. Each evidence chain includes the corresponding value source, time interval, and uncertainty information, forming an evidence witness set. The physician role generates a traceability chain based on this evidence witness set to fully record the entire process of risk identification, recommendation generation, and version pointing. The traceability chain is synchronized to the nursing role through a communication channel, and the nursing role performs assessment scheduling and closed-loop management accordingly.

[0121] Those skilled in the art will understand that the traceability chain can be implemented as a structured database record, a verifiable event chain, or other forms that can be recorded by the system, as long as the requirements of traceability and verifiability of information are met.

[0122] For details on how to generate the traceability chain, please refer to existing methods for constructing structured medical knowledge graphs or event chain tracing; this application will not elaborate on these methods here.

[0123] Before delving into the specific technical details of the steps, the embodiments of this application need to be emphasized again.

[0124] The complexity of VTE risk assessment lies not in the accuracy of the computational model, but in the asynchronous nature of clinical data and the dynamic consistency of the evidence chain. In real-world inpatient management processes, the evolution of a patient's condition exhibits non-linearity and the superposition of multiple pathways, meaning that the same clinical indicator has different significance in different time windows.

[0125] For example, the risk interpretation of elevated D-dimer levels preoperatively and postoperatively is completely different, and the discontinuation or delay of certain medication orders may change the reference boundaries of the aforementioned indicators. This situation of superficial similarity but semantic difference makes traditional assessment methods based on static rule matching prone to generating false positive risk points or assessment gaps in continuous monitoring, similar to the mirror artifact phenomenon in imaging detection. It seems abnormal, but it originates from the reflection and repetition of information rather than real risk.

[0126] Therefore, the assessment logic used in this embodiment does not directly determine the level of risk, but rather identifies the directionality of risk disturbances by calculating the trajectory of risk factors across different time segments and their convergence trend in multi-source data. Specifically, when the same risk factor shows a consistent evolution direction and temporal continuity in records from different sources, it can be considered as genuine disease progression; conversely, when changes in indicators cancel each other out or alternate repeatedly over time, they are marked as information interference.

[0127] Next, we will further elaborate on the technical aspects of the evidence association vector in this application.

[0128] In one example, obtaining the evidence association vector includes:

[0129] S1.1: Based on the semantic entity categories, the semantic entities in the multi-source data are nodeized to construct an initial evidence hypergraph with scale factors, diagnostic elements, test indicators and disease events as nodes;

[0130] Specifically, differences in field naming, value definitions, and time granularity among multi-source data can easily lead to ambiguity when directly used in subsequent reasoning. To ensure that information from different sources is processed uniformly at the same abstract level, semantic normalization and unique identifier mapping are first performed on core objects such as scale factors, diagnostic elements, test indicators, and disease course events. Semantic normalization includes:

[0131] The system maps custom fields within the hospital to a controlled terminology table; merges synonyms, aliases, and abbreviations; performs equivalent conversions on units and dimensions to form standard value ranges; and extracts key noun phrases from free text using entity recognition and sentence segmentation rules, then disambiguates them to standard entities. The unique identifier mapping uses a combination key of entity type, standard terminology code, and contextual label. The contextual label records the source system category, department domain, and record genre to prevent cross-domain confusion.

[0132] In this embodiment, after entity normalization is completed, an initial evidence hypergraph is constructed using each normalized entity as a node. The hypergraph allows a hyperedge to connect two or more nodes, used to represent situations where the same clinical fact is supported by multiple pieces of evidence.

[0133] For example, the scale factor "activity restriction ≥ 3 days" can be jointly pointed to by nursing records, exercise assessment sheets, and doctor's orders to discontinue activities, all connected to the factor node by a single hyperedge. To ensure subsequent computability, each node stores basic attributes: a list of sources, a list of timestamps, a value description, a confidence level, and a pointer to the evidence path; each hyperedge stores the generation rule number and a summary of triggering conditions for replay.

[0134] S1.2: Based on the contextual relationships and the effective time interval, the initial evidence hypergraph is subjected to layering and association strengthening processing, including establishing dependency edges between nodes within the same time layer and establishing evolutionary edges across time layers, so as to generate a time-layered evidence hypergraph;

[0135] Specifically, the interpretation of clinical evidence depends on its temporal context; the same value can have different meanings at different stages. To avoid the mixing of data from different time slices, the initial evidence hypergraph is layered according to its effective time interval:

[0136] For instantaneous records, the timestamp is extended to a very short effective interval; for continuous records, the effective interval is formed by the start and end times; for phased events, the boundary is marked to form a relative time coordinate.

[0137] After stratification, dependency edges are established within the same time layer to represent the conditional dependency and scale affiliation of entities in the same layer; evolutionary edges are established between time layers to represent the temporal evolution and semantic transfer of the same entity or the same clinical concept.

[0138] In this embodiment, association strengthening is achieved through two types of rules:

[0139] One type is the same-layer coupling rule, which establishes constraints based on the clinical dependencies between guideline scale items, such as activity limitation, lower limb swelling and pain forming complementary evidence clusters within the same layer;

[0140] Another type is cross-layer connection rules, which are used to bind the evolutionary relationship of stage nodes with the nursing window, so as to ensure that relevant evidence can be quickly located by using the stage window as the main axis during subsequent searches.

[0141] Furthermore, to control the graph size, intra-layer deduplication and cross-layer folding strategies are used for low-value duplicate evidence: intra-layer deduplication merges synonymous nodes based on criteria of value equivalence and similar sources; cross-layer folding aggregates small fluctuations in adjacent time layers to reduce meaningless oscillations.

[0142] S1.3: For multi-source data pointing to the same semantic node, construct a partial order conflict lattice in each time layer of the evidence hypergraph, wherein the elements of the partial order conflict lattice are characterized by a quadruple of value range, source level, sampling timestamp and record type.

[0143] Specifically, the same semantic node often receives values ​​from multiple sources at different times, which may contradict each other. To avoid information loss caused by the crude approach of overwriting old values ​​with new ones, all observations at the same level and node are organized into a set of evidence elements. Each evidence element is recorded as a quadruple: value range, source level, sampling timestamp, and record type. Based on this, comparable and incomparable relationships are defined: if two evidence elements have different source levels and their value ranges are inclusive or adjacent, the former is considered to have priority over the latter; if the value ranges are mutually exclusive or the source levels are incomparable, they are marked as incomparable and conflict is retained. This forms a partial order structure at the node, based on more reliable information and closer to the current time, providing order for subsequent merging.

[0144] In this embodiment, to reduce interference from text extraction noise, contextual verification entries are set for evidence elements of the record type (text). These entries store the original text extract, negation / degree adverb markers, and syntactic positions, and are used for manual or rule-based verification when conflicts arise with structured evidence. The source level setting is combined with hospital data governance practices: for example, the test results are higher than the medical record text summary, higher than manual nursing input, higher than third-party imports, etc. When the source levels are the same but the value ranges are inconsistent, time proximity is introduced as a decision aid, prioritizing the group closer to the evaluation time point.

[0145] S1.4: Merge the top elements of the partial order conflict lattices of the same semantic node in each time layer to obtain a unified representation within the layer. Then, converge the unified representations of each layer along the inter-layer evolution edge according to the order of near to far to obtain the merging result.

[0146] Specifically, within each time layer, the partial order structure of the same semantic node is merged. The merging strategy adopts a unified representation approach within each layer:

[0147] Based on source level and temporal proximity, comparable chains are compressed from bottom to top into a representative evidence element. For incomparable but compatible evidence, a juxtaposed representation is generated, and the value intervals are minimized. For incomparable and contradictory evidence, conflict markers are retained, and the triggered mutual exclusion rule numbers are recorded. The merged output not only provides a representative value but also carries a confidence summary and a conflict summary for subsequent calculation of the evidence vector.

[0148] In this embodiment, after intra-layer unification is completed, convergence is performed layer by layer along the evolution edge from the evaluation time point to past time layers. To avoid early strong fluctuations from excessively influencing the current judgment, a near-layer-first convergence order is adopted, and trend consistency is checked at each convergence:

[0149] If the change direction of the representative value of the current layer is consistent with that of the previous layer and the fluctuation does not exceed a preset threshold, the confidence summary of the previous layer is accumulated; if the direction is opposite or the fluctuation exceeds the threshold, the information of that layer is included in the deviation record and reflected separately in the final vector. For textual evidence that appears repeatedly across layers, if the trend is opposite to that of structured evidence, only its existence is retained and no more confidence is accumulated.

[0150] S1.5: Calculate the evidence vector of each semantic node based on the merging result, and combine the vectors of each semantic node according to the dependency edges to obtain the evidence association vector;

[0151] Specifically, based on the results of intra-layer unification and cross-layer convergence, an evidence vector is generated for each semantic node. The evidence vector includes, but is not limited to, the following elements: comprehensive confidence (calculated from the cumulative source level and time coverage), conflict degree (reflected by the ratio of mutually exclusive trigger counts to unresolved conflicts), time validity (derived from the combination of near-layer weights and stage threshold consistency), consistency (converted from the cross-layer trend consistency ratio and divergence ratio), sparsity penalty factor (used to characterize the uncertainty caused by excessively long evidence intervals), and value summary (numerical nodes provide the interval center and dispersion, qualitative nodes provide the ratio of primary and secondary labels). Each component of the vector can be directly derived from the aforementioned confidence / conflict / trend / coverage summaries without the need for an external learning process.

[0152] In this embodiment, the evidence vectors of each semantic node are combined according to dependency edges to form the evidence association vectors of the face vector table and the clause retrieval. The combination order follows the scale structure:

[0153] Factor aggregation is first completed within the same scale, and then evidence cluster aggregation is completed within the same clause. During the aggregation process, if the consistency of a key node in the dependency chain is low or the degree of conflict is high, the chain is marked as weakened to prevent a single weak chain from dominating the overall judgment. For nodes shared across clauses, reference copies are used to enter the aggregation of different clauses, rather than physically copying the nodes, to avoid difficulties in subsequent maintenance.

[0154] Next, we will further elaborate on the technical content of the method in this application regarding the trie of the guide version.

[0155] It is understood that the guideline version dictionary tree in this application is constructed based on medical guidelines. The specific construction process can refer to the construction methods of existing medical knowledge graphs or hierarchical clause trees. For example, the guideline chapter number, clause number and table item are used as hierarchical node indexes, and the clause content, applicable conditions and version identifier are used as node attributes to establish parent-child relationships and cross-version mapping relationships between nodes. This application will not elaborate on these details.

[0156] It is important to note that the guideline version dictionary in this application specifically includes multiple hierarchical nodes for storing and organizing information on medical guideline clauses. Each node corresponds to a chapter, clause, or table item in the medical guideline, forming a complete tree structure through hierarchical relationships. Each node stores structured information related to the clause, including attributes such as clause number, title, clause summary, applicable conditions, contraindications, recommendation level, source of evidence, and effective version identifier.

[0157] refer to Figure 4 , Figure 4 This is a schematic diagram of the structure of the trie provided in the embodiments of this application.

[0158] like Figure 4 As shown, the guideline version's trie consists of multiple hierarchical nodes, forming a top-down tree structure. Each node corresponds to a clause unit in the medical guideline, and the hierarchical nodes also include multiple pieces of structured information, which are stored in the form of branch nodes.

[0159] Those skilled in the art will understand that Figure 4 The nodes and connections shown are only abstract representations of structural relationships, used to illustrate the corresponding logic between versions, and do not limit the actual number of levels or the display method.

[0160] Next, we will further elaborate on the technical content of the method in this application regarding the currently effective node.

[0161] Before the search begins, the guide version trie only provides the hierarchical and version inheritance relationships of the clauses. Once the evidence association vector enters the search engine, two types of computable path constraints are locally constructed on the tree, and a joint semantic-temporal-version determination is performed within a finite candidate domain. This joint determination is not simply a matter of choosing the shortest path or the highest similarity; rather, it uses the matching relationship between the evidence's valid temporal interval and the clause's applicable conditions as a hard constraint, and the tolerance zone for changes in evidence uncertainty and clause thresholds as a soft constraint. Candidates that do not meet the conditions are eliminated layer by layer, ultimately converging to a clause node that can be uniquely identified and whose formation process can be replayed. Subsequent suggestion generation and traceability chain construction are both anchored to this node, avoiding suggestion drift and replay ambiguity caused by guide version switching or the coexistence of semantically nearest clauses.

[0162] It is understandable that the trigger-based connection edge in this application can be understood as a kind of pre-index edge dynamically generated during the retrieval period. This edge uses semantic factors and temporal attributes in the evidence association vector as trigger words, temporarily expanding the original tree structure with only parent-child levels into a short-circuit path that directly reaches the candidate entry if the trigger predicate is satisfied. The trigger predicate is accompanied by three types of information during generation:

[0163] One is the semantic mapping template, which is used to perform one-to-many matching between synonymous entities of evidence and key factors of clauses;

[0164] The second is the time gating condition, which limits the edge to be activated only within a specific stage window;

[0165] Thirdly, there is the witness fragment pointer, which is used to directly locate the triggered evidence entry during playback.

[0166] The presence of triggered connection edges transforms full tree traversal into a jump to a small number of hit candidates, reducing the reliance of retrieval on data completeness. At the same time, time gating enables automatic diversion of clauses pointed to by the same evidence at different stages, reducing mismatches between clauses with similar semantics.

[0167] It is understood that the virtual connection edge in this application can be specifically understood as a cross-version redirection edge, used to provide controlled migration when there is a version inconsistency between the candidate entry and the evidence. The virtual connection edge is not permanent, but is generated immediately after a candidate entry is matched, based on the version evolution relationship of the clause, and carries redirection rules:

[0168] When a threshold is adjusted in a clause, the edge records the threshold conversion and the difference in applicable scope; when a clause is merged / split, the edge records the merge mapping or split assignment conditions; when the applicable population or scenario of a clause is rewritten, the edge records the compensation rules for the addition or deletion of conditions.

[0169] To avoid incorrectly selecting outdated clauses, virtual link edges are generated using time span reduction and evidence change pattern verification. Mappings from older versions or those incompatible with evidence trends are automatically downgraded or blocked. For cases with one-to-many relationships between versions, conditional assignment ensures that redirection only occurs when new applicable conditions are met. Through the intervention of virtual link edges, even during the evolution of guideline clauses, the continuity of the search path and the clarity of the suggested source are maintained, and the redirection process can be replayed exactly as it was when traced back.

[0170] It should be noted that the currently effective nodes in this application are those obtained by combining the cross-version redirection of virtual connection edges within the candidate domain defined by the triggered connection edges, and are determined by three constraints: semantic consistency, temporal validity, and version adaptability. The number of nodes is not unique, and each node has at least two direct effects:

[0171] On the one hand, it provides a stable basis for the generation of recommendations, reducing fluctuations in results caused by data increments or version replacements; on the other hand, it provides a clear connection point for the execution and verification of nursing roles, ensuring consistency between the source of recommendations and the version direction when only differential updates are made to relevant time windows and factors in the future.

[0172] Those skilled in the art will understand that the two types of connection edges mentioned above can be generated and revoked using memory indexing or persistent mapping as needed for deployment. As long as the triggering predicate and redirection rules are replayable and verifiable, the same technical effect can be achieved.

[0173] In one example, the currently active node is obtained by retrieving the evidence association vector in the guide version trie, including:

[0174] S2.1: Based on the semantic factors and time attributes of the evidence association vector, triggerable connection edges are generated between the nodes of the guide version trie to obtain a triggerable pre-index, wherein the pre-index is used to locate and identify candidate entry nodes that match the evidence association vector during retrieval.

[0175] Specifically, when the trie in the guide version only has parent-child levels, direct root traversal is affected by tree depth, branching degree, and semantic proximity of clauses, resulting in excessive retrieval span. To transform the full traversal into a controlled candidate domain, it is necessary to construct a triggering predicate based on the semantic factors and temporal attributes in the evidence association vector. This predicate temporarily generates short-path edges that can be activated on the tree, forming a pre-index.

[0176] In one example, trigger-based connection edges are generated between nodes in the guide version of the trie, including:

[0177] Based on the semantic factors, time attributes and confidence in the evidence association vector, a time-series rule matrix is ​​constructed. The time-series rule matrix analyzes the evidence in different time windows, records the time relationship and change pattern between the evidence, and obtains the optimized path in the time dimension during the retrieval process.

[0178] Based on the time-series rule matrix, the nodes of the guide version trie are filtered to obtain multiple pre-indexes;

[0179] In the preceding index, candidate entry nodes corresponding to the preceding index are connected by triggered connection edges according to the time priority information corresponding to the time sequence rule matrix.

[0180] Specifically, before generating trigger-based connection edges, to avoid a single piece of evidence triggering multiple invalid nodes in different time slices and causing disordered retrieval paths, this embodiment constrains the connection logic in the time dimension by constructing a temporal rule matrix. The temporal rule matrix uses semantic factors in the evidence association vector as rows and discretized time windows as columns. Matrix elements record the direction, rate of change, and confidence level of the factor's value change between adjacent windows. The confidence level is derived from a comprehensive calculation of the evidence source level and time interval weight. After the matrix is ​​initialized, the factor state changes of any two adjacent time windows can form a temporal trajectory, which globally describes the continuous evolution of risk factors over time. When analyzing this matrix, the system aggregates trajectories with the same trend and identifies their inflection points, transforming continuous fluctuations in the time dimension into a finite number of key windows. These key windows represent the substantial stages of dynamic risk changes, such as the stable zone before risk escalation, the abrupt change period, and the recovery period.

[0181] In this embodiment, the nodes of the guide version trie contain the applicable time sequence information of the clauses. Therefore, the key windows of the time sequence rule matrix can be aligned with the time sequence labels of the clause nodes to determine which nodes are in the same time phase as the current risk trajectory. The alignment process is accomplished by calculating the overlap between the window and the clause time label: when the overlap exceeds a threshold and the semantic factors match, the node is included in the candidate set of the pre-index. Furthermore, the matrix also records the correlation relationships of factor changes. These correlations are extracted through the covariance sub-blocks of the matrix and used to filter the joint conditions of clause nodes when generating the pre-index. This allows nodes consistent with the multi-factor change pattern to be activated preferentially during retrieval, forming a time sequence optimization path.

[0182] Furthermore, after the preliminary indexes are generated, these indexes are sorted and connected using the time priority information in the time-series rule matrix. The time priority information is obtained by multiplying the rate of change by the confidence level of each key window in the matrix, representing the urgency of risk evolution. By establishing trigger-based connection edges between nodes corresponding to high-priority windows, a retrieval path conforming to clinical time logic can be formed: for example, high-risk nodes in the postoperative acute phase will be activated before chronic phase clauses. If different clause nodes overlap in time but do not conflict, a bidirectional trigger edge is established between them, with conditional labels attached for subsequent judgment. For clauses with temporal inclusion relationships, the weight of the trigger-based connection edge is set according to the inclusion decay setting, causing the retrieval path to tend to access more specific nodes with smaller time granularity first. Through this time-constrained trigger connection, not only is local optimal path selection in the time dimension achieved, but branches irrelevant to the current risk status can also be effectively filtered out in the version trie, thus maintaining retrieval efficiency and decision consistency even with large-scale guideline data.

[0183] In one example, based on the time-series rule matrix, the nodes of the guide version trie are filtered to obtain multiple preceding indices, including:

[0184] Based on the temporal dependency relationship of the temporal rule matrix, the corresponding flow pattern is obtained, and an evidence flow path diagram is generated. The nodes of the evidence flow path diagram represent evidence items, the edges represent causal relationships between evidence, and the evidence flow path diagram represents the reasoning and dependency relationship of evidence in the time dimension.

[0185] Identify the corresponding jump nodes based on the evidence flow path diagram, where a jump node represents an evidence node that spans multiple time windows in the time series;

[0186] Based on the temporal position, timeliness, and clinical relevance of the skip nodes, the skip nodes are prioritized to generate multiple pre-indexes.

[0187] S2.2: When there is a difference between the version information of the candidate entry node and the evidence association vector, a virtual connection edge is generated according to the cross-version entry mapping relationship to perform conditional redirection between different version entries;

[0188] Specifically, the guidelines may undergo threshold adjustments, additions or deletions of applicable conditions, merging or splitting of clauses across different versions. Direct matching based solely on the current version can easily lead to mismatches, such as evidence corresponding to older logic falling under newer clauses, or candidate dispersion due to clause splitting. To ensure the continuity of the search path across version evolutions, virtual connection edges need to be established between candidate entry nodes and their historical or future corresponding entries. These edges are generated in real-time by cross-version mapping relationships. The mapping relationship records three types of information based on the clause evolution type: change type, condition differences, and time stamp. Virtual connection edges are not permanently fixed; they are generated only as needed when a candidate matches and version inconsistencies exist, and are released immediately after the judgment is completed or the evidence becomes invalid, avoiding the formation of too many persistent cross-edges in the tree.

[0189] In one example, generating virtual connection edges based on cross-version entry mapping relationships includes:

[0190] The entries in the guide version dictionary are compared to identify the changes between the current version and historical versions, including added entries, modified entries, and deleted entries.

[0191] When the semantic similarity between the version information of the candidate entry node and the evidence association vector is less than a preset matching threshold, a virtual connection edge is generated based on the changed content and the cross-version entry mapping relationship.

[0192] In the process of generating the virtual connection edge, the redirected version entries are prioritized based on the time span between versions and the evidence change pattern using a time window discount factor. The time window discount factor is calculated by combining the time difference between versions and the evidence change pattern with a preset decay function. When redirecting across versions, the candidate version entries are weighted and matched by combining the confidence and conflict levels in the evidence association vector, and the corresponding version entry is selected based on the weighted matching result.

[0193] Specifically, establishing cross-version entry mapping relationships is the foundation for generating virtual connection edges. The version evolution of medical guidelines typically involves three types of changes:

[0194] First, content can be added or deleted, such as adding or removing specific risk factors;

[0195] Second, adjustments to the clause structure, such as splitting the original clause into multiple sub-clauses or merging multiple related clauses;

[0196] Third, changes in thresholds or conditions, such as adjustments to the upper and lower limits of risk scoring intervals or revisions to medication recommendation levels.

[0197] To ensure computational comparability between different versions of the clauses, this embodiment first performs a version comparison of the structured guide text across versions. A node-level content difference detection algorithm is employed, comparing clause numbers, semantic backbone similarity, and consistency of key parameters within the clauses to identify corresponding relationships. Semantic backbone similarity is calculated using a hybrid approach combining syntactic tree edit distance and medical concept vector similarity, while key parameter consistency checks the retention rate of numerical thresholds, drug names, or conditional logic within the clauses. After comparison, a version evolution mapping unit is generated for each pair of clauses with an evolutionary relationship, containing information such as change type, change direction, impact parameters, and effective date.

[0198] In this embodiment, when the semantic similarity between a candidate entry node and the current evidence association vector is lower than a preset threshold, a cross-version check mechanism is triggered. The system reads the version evolution mapping unit to determine whether there are any historical or future entries associated with the candidate node. If so, a virtual connection edge is generated based on the entry's evolution type.

[0199] For modified entries, edges carry conversion rules for parameter differences to map old parameters to new ones during redirection. For split entries, conditional assignment tables are added to edges to determine which sub-entry to jump to based on population, stage, or disease labels in the evidence association vector. For merged entries, source assignment markers are added to edges to record the old versions of the merged entries and their weights, allowing for source restoration during tracing. The generation of connecting edges is also affected by the time span between versions; the larger the time span and the more frequent the clause modifications, the higher the uncertainty of redirection. Therefore, a time window discount factor is introduced during edge generation to attenuate the priority of entries from more distant versions.

[0200] Furthermore, the calculation of the time window discount factor comprehensively considers the version effective time difference and the evidence change pattern. The version effective time difference is based on the difference in version release dates, standardized by time scale, and input into a preset decay function to calculate the basic decay weight. The evidence change pattern reflects the trend direction of each risk factor in the evidence association vector; for example, the semantic consistency of continuously rising and falling risk factor values ​​differs in version migration. By comparing the degree of matching between the evidence change pattern and the direction of clause evolution, it can be determined whether the decay factor has been amplified or reduced, thereby achieving adaptive adjustment of "high priority can be maintained even with a large time distance but consistent trend". The final discount value is stored as the weight parameter of the edge in the virtual connection edge attribute table for weighted calculation in the subsequent matching stage.

[0201] S2.3: Based on the triggered connection edge and the virtual connection edge, calculate the matching degree of the candidate entry nodes to obtain at least one currently active node;

[0202] Specifically, a joint matching evaluation needs to be performed between the candidate domains given by triggered connection edges and the possible redirections given by virtual connection edges to determine the currently effective node. Joint matching is not a simple similarity ranking, but rather considers three types of constraints simultaneously at the candidate node level:

[0203] Semantic consistency, temporal validity, and version compatibility. Each candidate entry node forms a decision summary, which includes which triggering edges it was hit by, whether it underwent redirection, a list of conditions satisfied or not satisfied during the redirection process, and a consistent relationship with the evidence evolution trend.

[0204] Joint matching performs pruning within the candidate set:

[0205] If a candidate fails to meet any hard constraint, it is directly eliminated; the remaining candidates are sorted according to the richness of constraint satisfaction and the completeness of difference description, and the top one is taken as the current effective node; if there are multiple parallel nodes that do not contradict each other, they are retained as parallel effective nodes and recorded in the traceability chain in a parallel branch manner.

[0206] In this embodiment, semantic consistency is directly given by the hit coverage summary of the triggering edge, including the hit status of required factors, the coverage ratio of optional factors, and the processing result of rejected factors; temporal validity is generated by comparing time gating with stage identifiers. If the evidence spans multiple stages, it is independently pruned in the candidate set of each stage to avoid cross-stage interference; version compatibility is given by the rule evaluation result of virtual connection edges. If there is a one-to-many redirection, only the path that completes all verifications of the condition stack is retained.

[0207] Next, with reference to the accompanying drawings, we will further describe another VTE intelligent monitoring method that supports multiple role requirements provided by the embodiments of this application. Figure 5 The method shown is applied to the nursing role end of a VTE intelligent prevention and control system, which is configured with a set of intervention measures. The VTE intelligent prevention and control system also includes a physician role end, which generates a traceability chain for prevention and control decision tracing and execution feedback, and synchronizes the traceability chain to the nursing role end. The method includes:

[0208] A1: Receive the traceability chain containing the evidence witness set from the doctor's role, and parse the traceability chain to obtain VTE prevention and treatment recommendations and the evidence chain corresponding to the VTE prevention and treatment recommendations;

[0209] Specifically, the traceability chain is used to record the process of forming evidence into recommendations from the physician's perspective, and it needs to be stably parsed into actionable information on the nursing side.

[0210] In this embodiment, the nursing-side analysis process consists of three steps:

[0211] Perform integrity verification on the traceability chain (signature verification, hash matching, and time freshness check).

[0212] The suggestion summary is mapped to nursing actionable items. The mapping is completed by a one-to-many binding from suggestion to nursing action, including frequency of execution, dosage / intensity, assessment timing and necessary monitoring items.

[0213] The witness sequence is loaded as execution preconditions and persisted locally in the form of a Boolean expression or a threshold comparison. If there are prohibitions in the condition section, a list of prohibitions is generated locally along with a condition for their removal.

[0214] A2: Obtain monitoring data, and based on the VTE prevention and control recommendations and the monitoring data, screen matching prevention and treatment measures from the set of intervention measures, and generate nursing execution instructions;

[0215] In one example, matching prevention and treatment measures are screened from the set of interventions, including:

[0216] Based on the VTE prevention recommendations and the monitoring data, the current physiological state is analyzed to generate initial screening criteria;

[0217] Based on the initial screening criteria, the final screening criteria are generated by combining the object's past history, wherein the past history includes past treatment responses, pathological examination results, and the object's individual characteristics.

[0218] Based on the final screening criteria, matching prevention and treatment measures are selected from the set of interventions.

[0219] A3: The controller executes the nursing execution instructions, collects real-time physiological data during the nursing intervention, and judges whether the intervention effect matches the expected response conditions in the evidence chain based on the real-time physiological data;

[0220] In one example, determining whether the intervention effect matches the expected response condition in the chain of evidence based on the real-time physiological data includes:

[0221] A matching degree is generated by comparing the real-time physiological data with the expected response conditions;

[0222] The intervention effect matching degree is calculated based on the matching degree. If the intervention effect matching degree is less than the preset response threshold, it means that the intervention effect corresponding to the current prevention and treatment measures does not match the expected response conditions.

[0223] A4: If there is no match, the nursing execution instruction is reconstructed based on the intervention effect, and the reconstruction result is fed back to the doctor role to update the guideline version dictionary tree;

[0224] Specifically, when assessment results consistently fail to meet expected response conditions, the nursing staff needs to perform differential reconstruction of previous instructions, rather than simply starting from scratch. To this end, differential reconstruction units are introduced to locate the source of the problem at the smallest granularity of the traceable chain.

[0225] Is it due to improper parameter settings, incorrect action type selection, or loss of applicable conditions? Location criteria include consecutive mismatched response pairs, abnormal equipment modes, changes in monitoring integrity and the taboo list.

[0226] In this embodiment, differential reconstruction follows a strategy that prioritizes minimal changes:

[0227] Prioritize adjusting parameters and review timing within the same action type; if still unsatisfactory, try similar alternatives (same type of medication, same level of mechanical device); only switch action types when necessary. Each refactoring generates a "Change Note," listing the differences before and after the change, the triggering reason, related event pairs, and the expected observation window. To ensure cross-role consistency, the refactoring results are sent back to the physician's end via communication channels, along with a feedback summary: including mismatch statistics, trend graph summary, equipment status summary, and changes in contraindications. The physician triggers partial playback based on this, re-examining clauses only in the relevant branches of the guideline version's dictionary tree to avoid duplicate searches of irrelevant entries.

[0228] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A VTE intelligent monitoring system supporting multi-role requirements, characterized in that, The VTE intelligent monitoring system includes a physician role terminal, which is configured with a guideline version dictionary tree. The system also includes a nursing role terminal, which is used for evaluation, orchestration, and closed-loop quality control based on the output of the physician role terminal. The physician role terminal includes: A multi-source data processing module is used to acquire multi-source data, construct an evidence hypergraph based on the multi-source data, and merge the partial-order conflict lattices of the evidence hypergraph to obtain an evidence association vector. The retrieval module is used to search the guide version trie using the evidence association vector to obtain the currently active node. The retrieval includes generating triggered connection edges and virtual connection edges in the guide version trie based on the evidence association vector. The triggered connection edges are used to locate candidate entry nodes that match the evidence association vector. The evidence witness set generation module is used to generate a traceability chain including the evidence witness set based on the VTE prevention recommendations and version pointing information output by the currently effective node. A communication module is used to establish two-way communication, synchronize the traceability chain to the nursing role terminal, and receive result feedback from the nursing role terminal; The retrieval module includes: A triggerable connection edge generation unit is used to generate triggerable connection edges between nodes of the guide version trie based on the semantic factors and time attributes of the evidence association vector, to obtain a triggerable pre-index, wherein the pre-index is used to locate and identify candidate entry nodes that match the evidence association vector during retrieval. A virtual connection edge generation unit is used to generate virtual connection edges based on cross-version entry mapping relationships when there is a difference between the version information of the candidate entry node and the evidence association vector, so as to perform conditional redirection between different version entries; The retrieval path optimization unit is used to calculate the matching degree of the candidate entry nodes by combining the triggered connection edge and the virtual connection edge, so as to determine the currently effective node; The retrieval module further includes a retrieval strategy, which is used to optimize the retrieval path. The retrieval strategy includes trigger matching logic and version adaptation logic. The trigger matching logic is configured in the trigger-based connection edge generation unit, and the version adaptation logic is configured in the virtual connection edge generation unit.

2. The VTE intelligent monitoring system supporting multi-role requirements according to claim 1, wherein, The multi-source data processing module includes: The data receiving unit is used to receive multi-source data and convert the multi-source data into a standardized format. The multi-source data includes at least one of the following: diagnostic information, test results, imaging reports, surgical records, medical order data, medical records, and medication information. The semantic parsing unit is used to perform semantic parsing on multi-source data in a standardized format; The time-series annotation unit is used to annotate the time information in multi-source data in a standardized format. The data analysis unit is used to analyze the outputs of the semantic parsing unit and the temporal annotation unit to obtain the evidence association vector; The multi-source data processing module further includes processing logic, which is used to construct an evidence hypergraph based on the semantic entity category, contextual relationship and effective time interval of each data, and generate an evidence association vector based on the evidence hypergraph. The processing logic is configured within the data analysis unit.

3. The VTE intelligent monitoring system supporting multi-role requirements according to claim 2, characterized in that, The processing logic is configured as follows: Based on the semantic entity categories, semantic entities in multi-source data are nodeized to construct an initial evidence hypergraph with scale factors, diagnostic elements, test indicators, and disease course events as nodes. The initial evidence hypergraph is layered and associated with enhanced processing based on the contextual relationships and the effective time intervals, including establishing dependency edges between nodes within the same time layer and establishing evolutionary edges across time layers, in order to generate a time-layered evidence hypergraph. For multi-source data pointing to the same semantic node, a partial order conflict lattice is constructed in each time layer of the evidence hypergraph, wherein the elements of the partial order conflict lattice are characterized by a quadruple of value range, source level, sampling timestamp and record type. Within each time layer, the top element of the partially ordered conflicting lattices of the same semantic node is merged to obtain a unified representation within the layer. Then, the unified representations within the layer are converged layer by layer along the inter-layer evolution edge according to the order of near to far to obtain the merging result. The evidence vector of each semantic node is calculated based on the merging results, and the semantic nodes are combined according to the dependency edges to obtain the evidence association vector.

4. The VTE intelligent monitoring system supporting multi-role requirements according to claim 1, characterized in that, The trigger matching logic is configured as follows: Based on the semantic factors, time attributes and confidence in the evidence association vector, a time-series rule matrix is ​​constructed. The time-series rule matrix analyzes the evidence in different time windows, records the time relationship and change pattern between the evidence, and obtains the optimized path in the time dimension during the retrieval process. Based on the time-series rule matrix, the nodes of the guide version trie are filtered to obtain multiple pre-indexes; In the preceding index, candidate entry nodes corresponding to the preceding index are connected by triggered connection edges according to the time priority information corresponding to the time sequence rule matrix.

5. The VTE intelligent monitoring system supporting multi-role requirements according to claim 4, characterized in that, Based on the time-series rule matrix, the nodes of the guide version trie are filtered to obtain multiple preceding indices, including: Based on the temporal dependency relationship of the temporal rule matrix, the corresponding flow pattern is obtained, and an evidence flow path diagram is generated. The nodes of the evidence flow path diagram represent evidence items, the edges represent causal relationships between evidence, and the evidence flow path diagram represents the reasoning and dependency relationship of evidence in the time dimension. Identify the corresponding jump nodes based on the evidence flow path diagram, where a jump node represents an evidence node that spans multiple time windows in the time series; Based on the temporal position, timeliness, and clinical relevance of the skip nodes, the skip nodes are prioritized to generate multiple pre-indexes.

6. The VTE intelligent monitoring system supporting multi-role requirements according to claim 1, characterized in that, The version adaptation logic is configured as follows: The entries in the guide version dictionary are compared to identify the changes between the current version and historical versions, including added entries, modified entries, and deleted entries. When the semantic similarity between the version information of the candidate entry node and the evidence association vector is less than a preset matching threshold, a virtual connection edge is generated based on the changed content and the cross-version entry mapping relationship. In the process of generating the virtual connection edge, the redirected version entries are prioritized based on the time span between versions and the evidence change pattern using a time window discount factor. The time window discount factor is calculated by combining the time difference between versions and the evidence change pattern with a preset decay function. When redirecting across versions, the candidate version entries are weighted and matched by combining the confidence and conflict levels in the evidence association vector, and the corresponding version entry is selected based on the weighted matching result.

7. The VTE intelligent monitoring system supporting multi-role requirements according to claim 1, characterized in that, The nursing role terminal is configured with a set of intervention measures, and the nursing role terminal includes: The data acquisition module is used to collect monitoring data, receive a traceability chain containing a set of evidence witnesses from the doctor's role, and parse the traceability chain to obtain VTE prevention and treatment recommendations and the corresponding evidence chain of the VTE prevention and treatment recommendations. The instruction generation module, based on the VTE prevention and control recommendations and the monitoring data, filters matching prevention and treatment measures from the intervention set and generates nursing execution instructions; The adjustment module is used to collect real-time physiological data during nursing intervention, determine whether the intervention effect matches the expected response conditions in the evidence chain based on the real-time physiological data, and if not, reconstruct the nursing execution instructions based on the intervention effect and feed the reconstruction result back to the doctor role to update the guideline version dictionary tree.

8. The VTE intelligent monitoring system supporting multi-role requirements according to claim 7, characterized in that, The instruction generation module includes: The preliminary screening unit is used to analyze the current physiological state based on the VTE prevention and control recommendations and the monitoring data, and generate initial screening conditions. A refined screening unit is used to generate final screening conditions based on the initial screening conditions and the object's past history, wherein the past history includes past treatment responses, pathological examination results and the object's individual characteristics. The screening unit, based on the final screening criteria, selects matching prevention and treatment measures from the set of interventions.

9. The VTE intelligent monitoring system supporting multi-role requirements according to claim 7, characterized in that, The adjustment module includes: Real-time data acquisition unit, used to monitor the patient's physiological data in real time; The intervention effect evaluation unit is used to compare real-time collected physiological data with expected response conditions, calculate the intervention effect matching degree, and determine whether the current intervention measures have achieved the expected effect based on the matching degree. The dynamic adjustment unit is used to adjust the nursing execution instructions based on the output of the intervention effect evaluation unit when the intervention effect does not meet expectations, and synchronously feed back to the doctor's role. The adjustment module further includes evaluation logic, which is configured within the intervention effect evaluation unit. The evaluation logic is configured as follows: A matching degree is generated by comparing the real-time physiological data with the expected response conditions; The intervention effect matching degree is calculated based on the matching degree. If the intervention effect matching degree is less than the preset response threshold, it means that the intervention effect corresponding to the current prevention and treatment measures does not match the expected response conditions.

Images