Chronic disease patient clinical data analysis method and system based on knowledge graph

Abstract

The invention relates to the technical field of medical health, and discloses a chronic disease patient clinical data analysis method and system based on a knowledge graph, and the method comprises the steps: obtaining a clinical data set; performing semantic extraction to obtain structured event data; constructing a personalized knowledge graph; carrying out conceptual mapping to obtain a conceptual knowledge graph; performing association path mining to obtain a multi-order time sequence association path; and matching with historical data to determine missing items and updating an acquisition strategy. According to the method, through two-stage knowledge graph construction and two-dimensional deletion detection, the depth and data integrity of chronic disease clinical data analysis are improved.

Description

Technical Field

[0001] This invention relates to the field of medical and health technology, and in particular to a method and system for analyzing clinical data of patients with chronic diseases based on knowledge graphs. Background Technology

[0002] Existing technologies only construct a unified knowledge graph without distinguishing between individual patient instances and standard medical concepts, resulting in the graph failing to balance individual specificity and medical standardization. The edge construction of knowledge graphs only considers semantic associations and does not consider temporal proximity relationships, thus failing to capture the temporal causal chain in the development of chronic diseases; The missing data detection only targets node entities and does not detect relationship edges, which makes it impossible to detect the missing logical connections between clinical events; The lack of a closed-loop mechanism to dynamically optimize the data collection strategy based on missing detection results makes it impossible to continuously improve data quality. Summary of the Invention

[0003] This invention provides a method and system for analyzing clinical data of patients with chronic diseases based on knowledge graphs, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, this invention provides a method for analyzing clinical data of chronic disease patients based on knowledge graphs, comprising: A1: Obtain the clinical dataset of the target patient, which includes the target patient's identification information, clinical examination indicator data, and clinical record text; A2: Semantic extraction is performed on the clinical record text to obtain the structured event data of the target patient; A3: Using the identity information as nodes and the clinical examination indicator data and the structured event data as connecting edges, construct a personalized knowledge graph for the target patient; A4: Based on the preset medical knowledge ontology, perform conceptual mapping on the nodes and attribute edges in the personalized knowledge graph to obtain the conceptual knowledge graph of the target patient; A5: Perform path mining on the conceptual knowledge graph to obtain the multi-order temporal path of the target patient; A6: Match the multi-level temporal correlation path with the historical clinical event data of the target patient, and determine the missing data items in the clinical dataset based on the matching results, so as to update the data collection strategy of the clinical dataset.

[0005] In a preferred embodiment, obtaining the clinical dataset of the target patient includes: Retrieve original clinical data records that match the target patient's identification information from the electronic medical record database in the hospital information terminal; The original clinical data records are cleaned to generate standardized clinical records for the target patient; Simultaneously extract clinical examination indicator data and clinical record text from the standardized clinical records.

[0006] In a preferred embodiment, the step of semantically extracting the clinical record text to obtain the structured event data of the target patient includes: The clinical record text is segmented into words to obtain the text segmentation sequence of the clinical record text; Named entity recognition is performed on the text segmentation sequence to identify medical entities in the text segmentation sequence, which are then used as a candidate entity set for the clinical record text. The semantic relationships of the candidate entity set are determined by classifying the relationships of any medical entity in the candidate entity set. Based on the semantic relationship, the candidate entity set is instantiated to obtain the original relation triplet of the target patient; The time information of the original relation triples is extracted from the clinical record text, and the time information is combined with the original relation triples as time tags to obtain the structured event data of the target patient.

[0007] In a preferred embodiment, constructing a personalized knowledge graph for the target patient, using the identity information as nodes and the clinical examination indicator data and the structured event data as connection edges, includes: Using the identity information as the central node, an initial graph structure for the target patient is constructed; The event subject, event action, and event object in the structured event data are used as event nodes. The event nodes are instantiated and associated with the central node to obtain the first type of connection edge of the initial graph structure. Using the examination indicators in the clinical examination indicator data as indicator nodes, attribute mapping is performed between the indicator nodes and the central node to obtain the second type of connection edges of the initial graph structure. In the time dimension, a third type of connection edge is constructed for the initial graph structure based on the proximity relationship between the event node and the indicator node; The central node, the first type of connecting edge, the second type of connecting edge, and the third type of connecting edge are injected into the initial graph structure to obtain the personalized knowledge graph of the target patient.

[0008] In a preferred embodiment, constructing a third type of connection edge for the initial graph structure based on the proximity relationship between the event node and the indicator node in the time dimension includes: Synchronously acquire the first and second time-series labels corresponding to the event nodes and the indicator nodes; The time difference of the initial graph structure is obtained by parsing the difference between the first time tag and the second time tag. Based on the time difference, the association between the event node and the indicator node is established, and the third type of connection edge of the initial graph structure is obtained.

[0009] In a preferred embodiment, the step of performing conceptual mapping on the nodes and attribute edges in the personalized knowledge graph according to a preset medical knowledge ontology to obtain the conceptual knowledge graph of the target patient includes: Obtain a preset medical knowledge ontology, which includes concept nodes and concept relationship edges; Traverse the nodes to be mapped in the personalized knowledge graph and calculate the initial matching degree between the attribute information of the nodes to be mapped and the concept nodes in the medical knowledge ontology; Based on the initial matching degree, the concept node with the highest matching degree is selected from the medical knowledge ontology as the mapping target for the node to be mapped, thereby obtaining the node mapping relationship of the personalized knowledge graph; Traverse the unmapped attribute edges in the personalized knowledge graph, and match the unmapped attribute edges with the concept relationship edges to obtain the attribute edge mapping relationship of the personalized knowledge graph; Based on the node mapping relationship and the attribute edge mapping relationship, the personalized knowledge graph is conceptualized to obtain the conceptual knowledge graph of the target patient.

[0010] In a preferred embodiment, calculating the initial matching degree between the attribute information of the node to be mapped and the concept nodes in the medical knowledge ontology includes: Obtain the node type identifier, associated nodes, and attribute information of the node to be mapped, and obtain the concept type identifier, concept association pattern, and definition features of the concept node; Based on the consistency between the node type identifier and the concept type identifier, determine the type matching factor between the node to be mapped and the concept node; Based on the attribute information and the defined features, the semantic matching degree between the node to be mapped and the concept node is evaluated; Based on the association pattern between the associated nodes and the concept nodes, the structural matching degree between the node to be mapped and the concept node is quantified; Based on the clinical expert experience database, a first weight factor for the type matching factor, a second weight factor for the semantic matching degree, and a third weight factor for the structural matching degree are determined, wherein the clinical expert experience database includes historical case samples and expert-annotated matching results; Based on the type matching factor, the semantic matching degree, and the structural matching degree, the initial matching degree between the attribute information of the node to be mapped and the concept nodes in the medical knowledge ontology is calculated, wherein the formula for calculating the initial matching degree is as follows: ; In the formula, The initial matching degree between the node to be mapped and the concept node. For the type matching factor, The semantic matching degree, The structural matching degree, The first weighting factor, This is the second weighting factor. This is the third weighting factor.

[0011] In a preferred embodiment, association path mining is performed on the conceptual knowledge graph to obtain the multi-order temporal association paths of the target patient, including: Nodes with time attribute information in the conceptual knowledge graph are designated as time-sensitive nodes. Starting from the time-sensitive node, the path is expanded in the conceptual knowledge graph in chronological order to obtain the initial set of associated paths for the target patient; A temporal consistency check is performed on the initial associated paths in the initial associated path set to obtain the temporally consistent path set for the target patient; Based on preset time window parameters, the temporal proximity filtering of the temporal consistent path set is performed to obtain the temporal proximity path set of the target patient; The path length of the temporal proximity path set is statistically analyzed, and the paths whose statistical results reach a preset order threshold are marked as multi-order temporal association paths of the target patient.

[0012] In a preferred embodiment, the step of matching the multi-order temporal association path with the target patient's historical clinical event data, and determining the missing data items in the clinical dataset based on the matching results to update the data collection strategy for the clinical dataset, includes: Obtain the historical clinical event data of the target patient; The nodes in the multi-order temporal association path are matched with the historical clinical event data at the node level to obtain the first type of missing data items in the clinical dataset. By performing edge-level matching between the conceptual relationship edges in the multi-order temporal association path and the historical clinical event data, the second type of missing data items in the clinical dataset is obtained. Based on the first type of missing data items and the second type of missing data items, the data acquisition strategy of the clinical dataset is updated to optimize the subsequent data acquisition of the target patients.

[0013] To address the aforementioned problems, the present invention also provides a knowledge graph-based clinical data analysis system for patients with chronic diseases, the system comprising: The data acquisition module is used to acquire the clinical dataset of the target patient, which includes the target patient's identification information, clinical examination indicator data, and clinical record text. The semantic extraction module is used to extract semantics from the clinical record text to obtain the structured event data of the target patient. The knowledge graph construction module is used to construct a personalized knowledge graph of the target patient, using the identity information as nodes and the clinical examination indicator data and the structured event data as connection edges. The concept mapping module is used to perform concept mapping on the nodes and attribute edges in the personalized knowledge graph according to the preset medical knowledge ontology, so as to obtain the conceptual knowledge graph of the target patient. The path mining module is used to perform association path mining on the conceptual knowledge graph to obtain the multi-order temporal association path of the target patient. The synchronous update module is used to match the multi-level temporal association path with the historical clinical event data of the target patient, and determine the missing data items in the clinical dataset based on the matching results, so as to update the data collection strategy of the clinical dataset.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention transforms unstructured clinical text into structured event data with time-series labels through standardized clinical data collection and fine-grained semantic extraction. It then constructs a personalized knowledge graph by combining patient identity information and examination indicators, thus fully characterizing the clinical status and temporal correlation features of patients with chronic diseases and providing a high-quality data foundation for in-depth analysis.

[0015] 2. This invention relies on medical knowledge ontology to complete the conceptual mapping of knowledge graph, accurately mine multi-level temporal correlation paths, automatically identify missing items in clinical data through dual-dimensional matching of nodes and relationships, dynamically optimize data collection strategies, significantly improve the depth, completeness and relevance of chronic disease clinical data analysis, and realize data-driven precision diagnosis and treatment management. Attached Figure Description

[0016] Figure 1This is a flowchart illustrating a method for analyzing clinical data of chronic disease patients based on knowledge graphs, as provided in an embodiment of the present invention. Figure 2 This is a functional module diagram of a knowledge graph-based clinical data analysis system for chronic disease patients, provided in one embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] This application provides a knowledge graph-based method for analyzing clinical data of patients with chronic diseases. The executing entity of this knowledge graph-based method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the knowledge graph-based method for analyzing clinical data of patients with chronic diseases can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0019] Reference Figure 1 The diagram shown is a flowchart illustrating a knowledge graph-based clinical data analysis method for chronic disease patients according to an embodiment of the present invention. In this embodiment, the knowledge graph-based clinical data analysis method for chronic disease patients includes: A1: Obtain the clinical dataset of the target patient, which includes the target patient's identification information, clinical examination indicator data, and clinical record text; In this embodiment of the invention, obtaining the clinical dataset of the target patient includes: Retrieve original clinical data records that match the target patient's identification information from the electronic medical record database in the hospital information terminal; The original clinical data records are cleaned to generate standardized clinical records for the target patient; Simultaneously extract clinical examination indicator data and clinical record text from the standardized clinical records.

[0020] Using the unique identification information of the target patient, a full-field matching search is performed in the electronic medical record database of the hospital information terminal. The electronic medical record database stores the entire process data of diagnosis and treatment of all patients in the hospital that have been collected and archived in compliance with regulations. After retrieving all diagnosis and treatment related records that completely match the identification information, the original clinical data record of the target patient is completely retrieved and summarized.

[0021] The retrieved original clinical data records were cleaned item by item, redundant data entries were removed, abnormal data with obvious errors were corrected, invalid characters and blank content that had no diagnostic or therapeutic significance were removed, and patient privacy information was desensitized. After processing, standardized clinical records of the target patients with uniform format and complete content were obtained.

[0022] The standardized clinical records that have been generated are classified and extracted according to data type. Numerical and result-type content reflecting physical examination results are extracted separately as clinical examination indicator data, and textual content describing the diagnosis and treatment process and medical orders is extracted separately as clinical record text. Both types of data are derived from the actual diagnosis and treatment content of the standardized clinical records.

[0023] The beneficial effects are that electronic medical record data can be accurately retrieved based on identity identification, and patient's full-cycle diagnosis and treatment information can be completely collected, providing comprehensive, authentic and complete original data support for subsequent analysis and ensuring the reliability of the data source.

[0024] Data cleaning generates standardized records, which can eliminate redundancy, correct errors, desensitize privacy, unify data format specifications, improve data quality, and lay a solid foundation for structured extraction and graph construction.

[0025] Simultaneously extract indicators and text data to achieve classification, efficient splitting, avoid duplication of processing, improve data utilization efficiency, and adapt to the needs of personalized knowledge graphs for accessing multiple types of data.

[0026] A2: Semantic extraction is performed on the clinical record text to obtain the structured event data of the target patient; In this embodiment of the invention, the step of semantically extracting the clinical record text to obtain the structured event data of the target patient includes: The clinical record text is segmented into words to obtain the text segmentation sequence of the clinical record text; Named entity recognition is performed on the text segmentation sequence to identify medical entities in the text segmentation sequence, which are then used as a candidate entity set for the clinical record text. The semantic relationships of the candidate entity set are determined by classifying the relationships of any medical entity in the candidate entity set. Based on the semantic relationship, the candidate entity set is instantiated to obtain the original relation triplet of the target patient; The time information of the original relation triples is extracted from the clinical record text, and the time information is combined with the original relation triples as time tags to obtain the structured event data of the target patient.

[0027] The Chinese medical word segmenter based on the BERT pre-trained language model splits clinical record text word by word according to the lexical boundaries of medical text. It is incrementally trained on a dedicated medical word segmentation dataset containing 500,000 Chinese medical texts with a training epoch of 3 rounds and a learning rate of 2e-5. The segmentation process strictly follows the usage norms of Chinese medical vocabulary and the expression habits of clinical text, splitting continuous text content into independent and semantically complete lexical units. All lexical units are arranged in the original text order to form the word segmentation sequence of the clinical record text. The clinical record text is derived from the diagnostic and treatment text descriptions in the standardized clinical records of the target patients.

[0028] For each word in the text segmentation sequence, a BiLSTM-CRF named entity recognition model is used to determine entities in a medical scenario. The number of hidden units in the BiLSTM layer of this model is set to 256, and the CRF layer is used to constrain the global optimality of the label sequence. The model is trained on a dataset containing 100,000 labeled medical entities, with the training set, validation set, and test set divided in an 8:1:1 ratio. The Adam optimizer is used for training, with an initial learning rate of 0.001 and 50 training epochs. The early stopping mechanism is set to stop training when the validation set loss does not decrease for 5 consecutive epochs. Medical words belonging to disease names, symptoms, examination items, drug names, treatment methods, and body parts are identified one by one. All identified medical words are aggregated to form a candidate entity set for the clinical record text. The identification is based on the standard medical entity categories defined in the medical domain knowledge graph.

[0029] Each group of different medical entities in the candidate entity set is selected for pairing analysis. A graph convolutional neural network relationship classification model based on attention mechanism is adopted. The number of graph convolutional layers in this model is set to 3, the number of attention heads is set to 8, and the hidden layer dimension is set to 256. Based on the clinical diagnosis and treatment logic and medical semantic rules, the relationship between the two groups of entities is determined. It is determined that the entities belong to one of the following relationships: causal relationship, accompanying relationship, examination correspondence relationship, and treatment applicability relationship. All the relationship types determined together constitute the semantic relationship of the candidate entity set.

[0030] According to the established semantic relationships, the corresponding medical entities are combined in a fixed format. The two related medical entities and the semantic relationship connecting them are integrated into a unified structure of associated instances, which fully restores the real diagnosis and treatment events described in the clinical record text. All the integrated associated instances together constitute the original relation triplet of the target patient.

[0031] The entire clinical record text is traversed, and a rule-based time information extraction method is adopted. Regular expressions are used to match time information such as consultation time, examination time, onset time, and medication time. The extracted time information is defined as time tags, and then the time tags are bound and combined with the corresponding original relation triples to finally form the target patient structured event data containing time information.

[0032] The beneficial effects are that professional medical word segmentation processing of clinical record text can accurately split medical terms into standardized word segments, providing a reliable text foundation for subsequent entity recognition and relation extraction, and improving the accuracy of semantic processing.

[0033] Named entity recognition based on medical scenarios can efficiently extract key medical entities such as diseases, symptoms, and examinations, forming a complete candidate entity set, and locking in core semantic elements for the structured transformation of clinical events.

[0034] Classifying medical entities by relationship can clarify the diagnostic and treatment logic and semantic connections between entities, restore the inherent business relationships in clinical texts, and provide clear relational basis and semantic support for the construction of triples.

[0035] Entity relations are instantiated based on semantic relationships, generating standard original relation triples, transforming unstructured text into structured representations, and significantly improving the machine readability and processing efficiency of clinical data.

[0036] Extract time information and bind it as triple time tags to form structured event data with time sequence, fully preserve the time sequence characteristics of diagnosis and treatment, and adapt to the needs of knowledge graph time sequence association and path mining.

[0037] A3: Using the identity information as nodes and the clinical examination indicator data and the structured event data as connecting edges, construct a personalized knowledge graph for the target patient; In this embodiment of the invention, the step of constructing a personalized knowledge graph for the target patient, using the identity information as nodes and the clinical examination indicator data and the structured event data as connection edges, includes: Using the identity information as the central node, an initial graph structure for the target patient is constructed; The event subject, event action, and event object in the structured event data are used as event nodes. The event nodes are instantiated and associated with the central node to obtain the first type of connection edge of the initial graph structure. Using the examination indicators in the clinical examination indicator data as indicator nodes, attribute mapping is performed between the indicator nodes and the central node to obtain the second type of connection edges of the initial graph structure. In the time dimension, a third type of connection edge is constructed for the initial graph structure based on the proximity relationship between the event node and the indicator node; The central node, the first type of connecting edge, the second type of connecting edge, and the third type of connecting edge are injected into the initial graph structure to obtain the personalized knowledge graph of the target patient.

[0038] The third type of connection edge construction of the initial graph structure in the time dimension, based on the proximity relationship between the event node and the indicator node, includes: Synchronously acquire the first and second time-series labels corresponding to the event nodes and the indicator nodes; The time difference of the initial graph structure is obtained by parsing the difference between the first time tag and the second time tag. Based on the time difference, the association between the event node and the indicator node is established, and the third type of connection edge of the initial graph structure is obtained.

[0039] The unique identification information of the target patient is used as the core center point of the entire graph. Based on this, a blank graph framework containing only the central node is built to form the initial graph structure of the target patient. The identification information comes from the patient's unique identity registration content in the electronic medical record database of the hospital information terminal.

[0040] Independent event subjects, event actions, and event objects are extracted from structured event data. These three types of content are added as independent event nodes to the initial graph structure. Each event node is then connected to the central node using direct association to form the first type of connection edge used to describe the diagnosis and treatment event. The structured event data comes from the semantic extraction and triple combination results of clinical record text. The semantic extraction F1 score is 0.89 for the experimental group and 0.72 for the control group.

[0041] Each independent examination indicator is extracted from the clinical examination indicator data. Each examination indicator is added as an independent indicator node to the initial graph structure. Then, according to the attribute correspondence of the examination indicator to the patient, the indicator nodes are mapped and connected to the central node to form a second type of connection edge used to describe the examination results. The clinical examination indicator data comes from the examination results in the standardized clinical records of the target patients.

[0042] Based on the time tags carried by the event nodes and indicator nodes, the time difference threshold method is used to determine the time proximity between the two. This time difference threshold is set according to the recommended time interval for "event-examination correlation analysis" in the clinical diagnosis and treatment guidelines for chronic diseases. For common chronic diseases such as hypertension and diabetes, the relevant clinical guidelines recommend a follow-up interval of 1 week. Therefore, this embodiment sets the time difference threshold to 7 days. At the same time, different threshold parameters can be configured for different types of chronic diseases: for patients with stable chronic diseases, the threshold can be relaxed to 14 days; for patients in the acute exacerbation phase, the threshold can be narrowed to 3 days.

[0043] When the absolute value of the time difference between an event node and an indicator node is less than or equal to a set threshold, they are determined to be temporally adjacent. This adjacency is used as the basis for association, forming a third type of connection edge to reflect temporal correlation. The time information comes from the diagnosis and treatment time records in the structured event data and clinical examination indicator data. For nodes with missing time labels, a time interpolation method is used: if only one of the event node and indicator node is missing a time label, the time information is inferred from the context of the associated clinical record text, or the patient's most recent visit time is used as an approximation, and this time is marked as the inferred value in the graph.

[0044] The first, second, and third types of connecting edges of the already determined central node are all integrated and added to the initial graph structure, so that the nodes and connecting edges form a complete and interconnected graph structure system, and finally generate a personalized knowledge graph that can comprehensively display the diagnosis and treatment information of the target patient.

[0045] The first time-series label is extracted and determined from the time information inherent in the event node, and the second time-series label is extracted and determined from the time information inherent in the indicator node. The first time-series label comes from the structured event data obtained after semantic extraction of clinical record text, and the second time-series label comes from the clinical examination indicator data extracted from standardized clinical records.

[0046] The first and second time-series labels are converted to Unix timestamp format according to a unified time scale. The absolute difference between the two timestamps is calculated to obtain the time difference value of the initial graph structure that can reflect the time proximity of the nodes. The time scale adopts the unified time measurement standard commonly used in medical diagnosis and treatment data.

[0047] The time difference is used to determine the proximity of event nodes and indicator nodes in time. When the time difference is less than or equal to the preset 7-day threshold, an undirected connection edge is established between the event node and the indicator node. This connection is the third type of connection edge in the initial graph structure, which is used to reflect the inherent connection between the diagnosis and treatment event and the examination indicator in the time dimension. When the time difference is greater than 7 days, no connection is established.

[0048] The beneficial effects are as follows: by building the initial graph structure with the patient's identity as the central node, the core anchor point of the graph can be established, allowing all clinical data to be organized around the patient in a unified manner, ensuring a clear graph structure and well-defined attribution. By setting diagnostic and treatment event elements as event nodes and associating them with the central node, the patient's diagnosis and treatment process can be completely reconstructed, clearly presenting the full picture of clinical events and providing intuitive evidence for disease tracking.

[0049] By setting examination indicators as indicator nodes and completing attribute mapping, examination data can be standardized and integrated into the data graph, enabling precise binding of indicators to patients and improving the structured utilization of clinical data. Constructing time-dimensional connection edges based on temporal proximity relationships establishes dynamic associations between events and indicators, preserving the temporal characteristics of diagnosis and treatment, and adapting to the needs of long-term temporal analysis of chronic diseases.

[0050] By integrating the three types of connecting edges with the central node to form a complete graph, it is possible to fully integrate patient identity, examination, and event information, providing a high-quality data carrier for subsequent concept mapping and path mining.

[0051] By synchronously acquiring the time series labels of event nodes and indicator nodes, the time information of both can be collected in a unified manner, ensuring that the time series data source is consistent and complete, and providing accurate and reliable basic data for subsequent time difference calculation.

[0052] By performing difference analysis on the two types of time series labels, the time interval between events and indicators can be accurately quantified, objectively reflecting the temporal proximity of diagnosis and treatment behaviors, and providing a quantitative basis for determining time correlation.

[0053] By establishing node associations based on time differences and forming third-type connection edges, temporal features can be integrated into the graph structure, strengthening the intrinsic relationship between events and indicators, and supporting multi-dimensional clinical data analysis.

[0054] A4: Based on the preset medical knowledge ontology, perform conceptual mapping on the nodes and attribute edges in the personalized knowledge graph to obtain the conceptual knowledge graph of the target patient; In this embodiment of the invention, the step of performing conceptual mapping on the nodes and attribute edges in the personalized knowledge graph according to a preset medical knowledge ontology to obtain the conceptual knowledge graph of the target patient includes: Obtain a preset medical knowledge ontology, which includes concept nodes and concept relationship edges; Traverse the nodes to be mapped in the personalized knowledge graph and calculate the initial matching degree between the attribute information of the nodes to be mapped and the concept nodes in the medical knowledge ontology; Based on the initial matching degree, the concept node with the highest matching degree is selected from the medical knowledge ontology as the mapping target for the node to be mapped, thereby obtaining the node mapping relationship of the personalized knowledge graph; Traverse the unmapped attribute edges in the personalized knowledge graph, and match the unmapped attribute edges with the concept relationship edges to obtain the attribute edge mapping relationship of the personalized knowledge graph; Based on the node mapping relationship and the attribute edge mapping relationship, the personalized knowledge graph is conceptualized to obtain the conceptual knowledge graph of the target patient.

[0055] The calculation of the initial matching degree between the attribute information of the node to be mapped and the concept nodes in the medical knowledge ontology includes: Obtain the node type identifier, associated nodes, and attribute information of the node to be mapped, and obtain the concept type identifier, concept association pattern, and definition features of the concept node; Based on the consistency between the node type identifier and the concept type identifier, determine the type matching factor between the node to be mapped and the concept node; Based on the attribute information and the defined features, the semantic matching degree between the node to be mapped and the concept node is evaluated; Based on the association pattern between the associated nodes and the concept nodes, the structural matching degree between the node to be mapped and the concept node is quantified; Based on the clinical expert experience database, a first weight factor for the type matching factor, a second weight factor for the semantic matching degree, and a third weight factor for the structural matching degree are determined, wherein the clinical expert experience database includes historical case samples and expert-annotated matching results; Based on the type matching factor, the semantic matching degree, and the structural matching degree, the initial matching degree between the attribute information of the node to be mapped and the concept nodes in the medical knowledge ontology is calculated, wherein the formula for calculating the initial matching degree is as follows: ; In the formula, The initial matching degree between the node to be mapped and the concept node. For the type matching factor, The semantic matching degree, The structural matching degree, The first weighting factor, This is the second weighting factor. This is the third weighting factor.

[0056] The medical knowledge ontology is retrieved from the standard knowledge base in the medical field. This medical knowledge ontology consists of professional clinical terminology and standard diagnostic and treatment logic. It contains concept nodes for representing medical content such as disease symptom examination, as well as concept relationship edges for representing the relationship between medical concepts.

[0057] One by one, all nodes in the personalized knowledge graph that need to be standardized and mapped are accessed. The name feature, type feature and clinical meaning feature of each node are extracted. These features are compared with the features of each concept node in the medical knowledge ontology, and the degree of feature overlap is statistically analyzed to obtain the initial matching degree of each node.

[0058] The nodes are sorted according to their initial matching scores. For each node to be mapped, the concept node with the highest initial matching score and completely consistent clinical meaning is selected as the final mapping target. The correspondence between all nodes to be mapped and their corresponding concept nodes is compiled and summarized to form the node mapping relationship of the personalized knowledge graph.

[0059] One by one, all attribute edges in the personalized knowledge graph that need to be standardized are accessed, the clinical association meaning represented by each attribute edge is clarified, the meaning is semantically matched with the conceptual relationship edges in the medical knowledge ontology, the standard conceptual relationship corresponding to the attribute edge is determined, and all matching results are summarized to form the attribute edge mapping relationship of the personalized knowledge graph.

[0060] Based on the established node mapping and attribute edge mapping relationships, all patient-specific nodes and attribute edges in the personalized knowledge graph are replaced with corresponding standard concept nodes and concept relationship edges in the medical knowledge ontology. After completing all replacements and structural integration, a conceptual knowledge graph of the target patient is generated.

[0061] Extract the node type identifier, associated nodes, and attribute information of the node to be mapped. The node type identifier is used to distinguish node categories such as disease, symptom, examination, drug, and surgery. Associated nodes are other nodes directly connected to this node in the personalized knowledge graph. The attribute information is the clinical description, value range, and status characteristics of this node. At the same time, extract the concept type identifier, concept association pattern, and definition features of concept nodes in the medical knowledge ontology. The concept association pattern is the fixed connection method between standard medical concepts, and the definition features are the standard explanation content of authoritative medical terms.

[0062] The node type identifier of the node to be mapped is compared with the concept type identifier of the concept node for category consistency. When the two are completely consistent, the type matching factor is 1; when the two belong to the same major category but have sub-category differences, the value is 0.5; when the two categories are completely unrelated, the value is 0.

[0063] The attribute information of the node to be mapped and the definition features of the concept node are converted into 768-dimensional BERT semantic vectors respectively. The cosine similarity between the two vectors is calculated, and the calculated cosine similarity value is used as the semantic matching degree, with a value range from 0 to 1.

[0064] The structural similarity is calculated between the set of associated nodes of the node to be mapped and the set of associated nodes defined by the conceptual association pattern of the concept node. The Jaccard similarity coefficient is used, which is the size of the intersection of the two sets divided by the size of the union. The calculated Jaccard similarity coefficient is used as the value of the structural matching degree, and the value ranges from 0 to 1.

[0065] The weighting rules, pre-defined by clinical experts, were retrieved from a clinical expert experience database. This database was constructed as follows: 10 senior experts with over 10 years of clinical experience from tertiary hospitals provided 1000 historical case samples with complete clinical records. Each sample included a patient's personalized knowledge graph fragment and the corresponding standard medical knowledge ontology mapping result. Experts independently performed manual annotation to determine the optimal match. The annotation results underwent consistency testing, with a Kappa coefficient exceeding 0.85. Based on the annotation results, a multiple linear regression method was used to learn the weighting factors. The matching result annotated by experts was used as the dependent variable, and type matching factor, semantic matching degree, and structural matching degree were used as independent variables for fitting. Finally, the first weighting factor corresponding to the type matching factor was determined to be 0.3, the second weighting factor corresponding to the semantic matching degree was 0.5, and the third weighting factor corresponding to the structural matching degree was 0.2. These weighting factors remained consistent across different node types, but for specific node types, they could be fine-tuned based on clinical expert experience, with a fine-tuning range of ±0.05.

[0066] The type matching factor, semantic matching degree, and structural matching degree are weighted and fused according to their respective weight factors. The degree of fit of the three dimensions of category, semantics, and structure is combined to calculate the initial matching degree between the attribute information of the node to be mapped and the concept node in the medical knowledge ontology.

[0067] The type matching factor is derived from the consistency determination result between the node type identifier of the node to be mapped and the concept type identifier of the concept node in the medical knowledge ontology. The higher the consistency, the larger the value of the type matching factor.

[0068] Semantic matching degree is derived from the evaluation result of the semantic overlap between the attribute information carried by the node to be mapped and the standard definition features corresponding to the concept node. The closer the semantics, the higher the semantic matching degree.

[0069] The structural matching degree is derived from the quantitative result of the fit between the structure of the associated nodes of the node to be mapped in the personalized knowledge graph and the concept association pattern of the concept node in the medical knowledge ontology. The more consistent the structure, the greater the value of the structural matching degree.

[0070] The first weighting factor is derived from the importance configuration results of node type matching in the clinical expert experience database, and is used to control the proportion of the type matching factor in the overall calculation.

[0071] The second weighting factor is derived from the importance configuration results of node semantic matching in the clinical expert experience database, and is used to control the proportion of semantic matching degree in the overall calculation.

[0072] The third weighting factor is derived from the importance configuration results of node structure matching in the clinical expert experience database, and is used to control the proportion of structure matching degree in the overall calculation.

[0073] The initial matching degree is calculated by first multiplying the type matching factor by the first weight factor, then multiplying the semantic matching degree by the second weight factor, then multiplying the structural matching degree by the third weight factor, and finally adding the three sets of multiplication results to obtain the final value.

[0074] The initial matching degree is used to comprehensively measure the overall matching degree between the node to be mapped and the concept node, while taking into account the three key dimensions of category, semantics, and structure, to ensure that the mapping results comply with clinical medical standards.

[0075] The initial matching degree can provide an objective quantitative basis for the conceptual mapping of personalized knowledge graphs to standard medical knowledge ontology, avoid the bias caused by single-dimensional matching, and improve the accuracy and reliability of conceptual transformation. Among them, the concept mapping accuracy rate is 91.6% in the experimental group vs. 82.3% in the control group.

[0076] The initial matching degree calculation method fully integrates the experience of clinical experts, making the matching results more in line with the actual use needs of medical scenarios, laying the foundation for subsequent conceptual knowledge graph construction and intelligent diagnosis and treatment reasoning.

[0077] The beneficial effects are that by acquiring a standard medical knowledge ontology, a standardized mapping framework can be constructed based on authoritative clinical terminology and diagnostic and treatment logic, providing a unified medical semantic foundation for the standardized conversion of personalized knowledge maps.

[0078] By traversing the nodes to be mapped and calculating the initial matching degree, the node fit can be quantified from multiple dimensions such as category, semantics, and structure, providing an objective and reliable quantitative basis for accurate node mapping.

[0079] Selecting the mapping target based on the highest matching degree to generate node mapping relationships enables precise alignment between personalized patient entities and standard medical concepts, ensuring the accuracy of graph conceptualization transformation. Traversing attribute edges and matching them with conceptual relationship edges transforms personalized associations into standard clinical relationships, unifying the expression of graph relationships and improving the universality and reusability of clinical data.

[0080] Conceptual transformation is completed based on node and attribute edge mapping, which can generate standardized conceptual knowledge graphs, laying a standardized foundation for subsequent time-series path mining and data missing detection.

[0081] Simultaneously acquiring multi-dimensional feature information of nodes to be mapped and concept nodes allows for the comprehensive extraction of key content such as type, association, attributes, and definitions, providing a complete and accurate data foundation for multi-dimensional matching calculations. Determining the type matching factor based on the consistency of node type identifiers quantifies the degree of node matching from a category perspective, quickly eliminating irrelevant concepts and improving the initial screening efficiency of the mapping process.

[0082] By assessing semantic matching based on attribute information and defined features, the degree of node fit can be accurately measured from the perspective of clinical meaning, ensuring the accuracy and rationality of concept mapping in medical semantics. Quantifying structural matching by relying on associated nodes and concept association patterns allows for judgment of matching rationality from the perspective of graph topology, avoiding mapping bias caused by single-dimensional matching.

[0083] By combining a clinical expert experience database with three types of weighting factors, the matching calculation can be aligned with clinical diagnosis and treatment logic, making the mapping results more consistent with actual medical application scenarios and expert cognition. An initial matching degree is calculated using a weighted fusion formula, integrating three-dimensional features of type, semantics, and structure to output an objective and unified matching score, providing a scientific basis for selecting the optimal concept node.

[0084] A5: Perform path mining on the conceptual knowledge graph to obtain the multi-order temporal path of the target patient; In this embodiment of the invention, association path mining is performed on the conceptual knowledge graph to obtain the multi-order temporal association paths of the target patient, including: Nodes with time attribute information in the conceptual knowledge graph are designated as time-sensitive nodes. Starting from the time-sensitive node, the path is expanded in the conceptual knowledge graph in chronological order to obtain the initial set of associated paths for the target patient; A temporal consistency check is performed on the initial associated paths in the initial associated path set to obtain the temporally consistent path set for the target patient; Based on preset time window parameters, the temporal proximity filtering of the temporal consistent path set is performed to obtain the temporal proximity path set of the target patient; The path length of the temporal proximity path set is statistically analyzed, and the paths whose statistical results reach a preset order threshold are marked as multi-order temporal association paths of the target patient.

[0085] Traverse all nodes within the conceptual knowledge graph, filter out nodes carrying time attribute information such as consultation time, examination time, onset time, and medication time, and uniformly mark these nodes as time-sensitive nodes. The conceptual knowledge graph is derived from the standardized graph structure of the target patient's personalized knowledge graph after being mapped by the standard medical knowledge ontology.

[0086] Using each time-sensitive node as the starting point for path extension, a breadth-first search algorithm is used to extend the path in chronological order from early to late. The path is extended step by step along the existing connecting edges within the conceptual knowledge graph. Each time the path is extended, only adjacent nodes with time tags later than the current node are selected. The maximum extension depth of each path is limited to 10 hops. Each continuous path formed by the extension is fully recorded. All paths are summarized to form the initial associated path set of the target patient.

[0087] For each initial associated path in the initial associated path set, a temporal consistency check is performed. The time labels of adjacent nodes on the path are checked one by one to ensure that the time labels are strictly increasing, that is, the time label of the later node must be later than the time label of the earlier node. Paths with the same or decreasing time labels are removed, and paths with completely reasonable time order are retained. All retained paths are combined to form the temporally consistent path set of the target patient.

[0088] The time window parameter is set according to a pre-defined time span based on empirical values ​​for the correlation of clinical diagnosis and treatment events. It references the recommended time range for "event chain" analysis in chronic disease clinical guidelines. For most chronic diseases, relevant clinical studies have shown a significant correlation between clinical events within 90 days; therefore, this embodiment sets the time window parameter to 90 days. This parameter can be adaptively adjusted according to different types of chronic diseases: for chronic diseases with slower disease progression, the time window can be extended to 180 days; for chronic diseases with rapid disease changes, the time window can be narrowed to 30 days; for acute complication events, the time window can be further narrowed to 14 days. Users can dynamically adjust this parameter through the system configuration interface according to different diseases or analysis needs. Paths in the time-consistent path set are filtered one by one. The time label difference between the first and last nodes on the path is calculated. Only paths with a time difference less than or equal to the set time window are retained, while paths exceeding the time range are removed. After filtering, the time-proximity path set of the target patient is obtained.

[0089] The number of nodes in each path within the temporal proximity path set is counted. The path length is determined based on this count. Paths whose length reaches a preset order threshold are uniformly marked. This preset order threshold is set according to the actual needs of complex clinical association analysis. A 3rd-order path represents a temporal chain containing 3 clinical event nodes, reflecting typical chronic disease diagnosis and treatment chains such as "symptom-examination-medication" or "examination-diagnosis-treatment," meeting the basic needs of clinicians for multi-step association analysis. This threshold also supports dynamic configuration. For clinical research scenarios requiring more refined analysis, the threshold can be increased to 4 or 5; for basic follow-up scenarios, the threshold can be decreased to 2. The marked paths are the multi-order temporal association paths for the target patient.

[0090] The beneficial effect is that selecting nodes with time attributes as time-sensitive nodes can accurately pinpoint key elements of the diagnosis and treatment timeline, providing a clear starting point for subsequent timeline correlation path mining and ensuring the timeline effectiveness of path analysis.

[0091] Starting from time-sensitive nodes and expanding the path in chronological order, it can generate an initial path by following the sequence of occurrence of diagnosis and treatment behaviors, and restore the natural evolution logic and correlation of clinical events in chronic diseases.

[0092] Performing a temporal consistency check on the initial pathways can eliminate invalid pathways with out-of-order time, ensuring that all retained pathways conform to clinical time sequence rules and improving the standardization and reliability of pathway data.

[0093] By filtering based on temporal proximity within a preset time window, relevant paths with reasonable time spans can be retained, aligning with the patterns of chronic disease diagnosis and treatment, and enhancing the clinical reference value and practicality of the pathways.

[0094] By marking eligible paths by path length, we can obtain high-order association chains that meet the depth of analysis, supporting the mining of complex clinical logic and the accurate identification of missing data.

[0095] A6: Match the multi-level temporal correlation path with the historical clinical event data of the target patient, and determine the missing data items in the clinical dataset based on the matching results, so as to update the data collection strategy of the clinical dataset.

[0096] In this embodiment of the invention, the step of matching the multi-order temporal association path with the historical clinical event data of the target patient, and determining the missing data items in the clinical dataset based on the matching results, so as to update the data collection strategy of the clinical dataset, includes: Obtain the historical clinical event data of the target patient; The nodes in the multi-order temporal association path are matched with the historical clinical event data at the node level to obtain the first type of missing data items in the clinical dataset. By performing edge-level matching between the conceptual relationship edges in the multi-order temporal association path and the historical clinical event data, the second type of missing data items in the clinical dataset is obtained. Based on the first type of missing data items and the second type of missing data items, the data acquisition strategy of the clinical dataset is updated to optimize the subsequent data acquisition of the target patients.

[0097] Each node on the multi-order temporal association path is matched against entity information in historical clinical event data, using a combination of exact matching and semantic similarity matching: First, exact string matching of entity names is performed; if a match is successful, a corresponding record is considered to exist. If an exact match fails, the cosine similarity between the node name and the entity name in the historical event is calculated using the BERT semantic vector. The BERT model used here is the same as the BERT pre-trained language model described in step A2 to ensure consistency in semantic representation. A similarity greater than 0.85 is considered a successful semantic match. This similarity threshold was determined based on statistical analysis of 500 manually annotated matching results. At a threshold of 0.85, the matching precision reaches 94.2%, and the recall reaches 91.6%, which can effectively balance matching accuracy and completeness. Nodes that fail to match either exactly or semantically are classified as Category I missing data items in the clinical dataset.

[0098] Each conceptual relationship edge on the multi-level temporal association path is matched against the association information in historical clinical event data, employing a dual matching mechanism of relationship type and associated entity: first, it is determined whether the type of the conceptual relationship edge exists in the historical data as a relationship record of the same type; then, it is checked whether the two entities connected by the relationship both appear in the historical data and have the same type of relationship. A successful match is considered achieved when both conditions are met. For relationship edges that fail to match, the system further analyzes the reasons for failure: if the relationship type exists but the associated entity is missing, it is determined to be a missing entity item; if the relationship type does not exist, it is determined to be a missing relationship item. Unmatched relationship edges are classified as the second type of missing data item in the clinical dataset. The conceptual relationship edges originate from the standardized association relationships defined in the standard medical knowledge ontology.

[0099] For boundary situations that occur during the matching process, the system is configured with the following handling mechanism: When there are synonyms or abbreviations for the node name, a medical terminology thesaurus is pre-built, which contains the standard medical key terms in the latest IDC-11 guidelines, and synonym normalization is performed before precise matching; When there are multiple similar entities in historical clinical event data, the entity with the closest time tag to the target node is selected for matching first; When the matching confidence is between 0.75 and 0.85, the node is marked as an item to be confirmed and output to the manual review queue for final judgment by clinicians. The review results are used to continuously optimize the matching threshold.

[0100] Based on the missing content pointed to by the first and second types of missing data items, a data collection optimization strategy was generated: for missing node entities, corresponding data collection fields were added in subsequent data collection and set as mandatory fields in the electronic medical record form; for missing relationship edges, correlation collection items were added, requiring clinicians to simultaneously enter related examination indicators or clinical events when entering relevant diagnosis and treatment events. Among them, the accuracy rate of missing data item identification was 94.2% in the experimental group vs. 78.5% in the control group.

[0101] Simultaneously, the data collection frequency was adjusted based on the frequency and importance of missing items, changing the collection cycle for high-frequency missing items from quarterly to monthly. The data collection strategy for the clinical dataset was updated to optimize the completeness and relevance of subsequent clinical data collection for target patients.

[0102] The beneficial effects are that obtaining complete historical clinical event data of patients can build a comprehensive baseline database of diagnosis and treatment facts, providing a real and reliable basis for subsequent two-dimensional matching and verification of nodes and relationships, and ensuring accurate determination of missing data.

[0103] By conducting node-level matching to identify the first type of missing data items, we can accurately locate the missing content of diagnosis and treatment entities and clarify the location and type of missing entity information such as diseases, symptoms, and examinations in clinical data.

[0104] By performing edge-level matching to identify the second type of missing data items, it is possible to effectively discover omissions in the correlation between clinical events, fill in the gaps in the logical connections between data, and improve the completeness and relevance of the dataset.

[0105] By updating the data collection strategy based on the two types of missing items, we can specifically fill the data gaps, optimize the subsequent collection items and frequency, and achieve full-cycle, high-quality closed-loop management of chronic disease clinical data.

[0106] like Figure 2 The diagram shown is a functional block diagram of a knowledge graph-based clinical data analysis system for chronic disease patients provided in an embodiment of the present invention.

[0107] The knowledge graph-based clinical data analysis system for chronic disease patients described in this invention can be installed in an electronic device. Depending on the functions implemented, the knowledge graph-based clinical data analysis system for chronic disease patients may include a data acquisition module, a semantic extraction module, a graph construction module, a concept mapping module, a path mining module, and a synchronization update module. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, and are stored in the memory of the electronic device.

[0108] In this embodiment, the functions of each module / unit are as follows: The data acquisition module is used to acquire the clinical dataset of the target patient, which includes the target patient's identity information, clinical examination indicator data, and clinical record text. The semantic extraction module is used to extract semantics from the clinical record text to obtain the structured event data of the target patient. The graph construction module is used to construct a personalized knowledge graph of the target patient, using the identity information as nodes and the clinical examination indicator data and the structured event data as connection edges. The concept mapping module is used to perform concept mapping on the nodes and attribute edges in the personalized knowledge graph according to the preset medical knowledge ontology, so as to obtain the conceptual knowledge graph of the target patient. The path mining module is used to perform association path mining on the conceptual knowledge graph to obtain the multi-order temporal association path of the target patient; The synchronization update module is used to match the multi-level temporal association path with the historical clinical event data of the target patient, and based on the matching result, determine the missing data items in the clinical dataset to update the data collection strategy of the clinical dataset.

[0109] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0110] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0111] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0112] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0113] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for analyzing clinical data of chronic disease patients based on knowledge graphs, characterized in that, The method includes: A1: Obtain the clinical dataset of the target patient, which includes the target patient's identification information, clinical examination indicator data, and clinical record text; A2: Semantic extraction is performed on the clinical record text to obtain the structured event data of the target patient; A3: Using the identity information as nodes and the clinical examination indicator data and the structured event data as connecting edges, construct a personalized knowledge graph for the target patient; A4: Based on the preset medical knowledge ontology, perform conceptual mapping on the nodes and attribute edges in the personalized knowledge graph to obtain the conceptual knowledge graph of the target patient; A5: Perform path mining on the conceptual knowledge graph to obtain the multi-order temporal path of the target patient; A6: Match the multi-level temporal correlation path with the historical clinical event data of the target patient, and determine the missing data items in the clinical dataset based on the matching results, so as to update the data collection strategy of the clinical dataset.

2. The method for analyzing clinical data of chronic disease patients based on knowledge graphs as described in claim 1, characterized in that, The acquisition of the target patient's clinical dataset includes: Retrieve original clinical data records that match the target patient's identification information from the electronic medical record database in the hospital information terminal; The original clinical data records are cleaned to generate standardized clinical records for the target patient; Simultaneously extract clinical examination indicator data and clinical record text from the standardized clinical records.

3. The method for analyzing clinical data of chronic disease patients based on knowledge graphs as described in claim 1, characterized in that, The semantic extraction of the clinical record text to obtain the structured event data of the target patient includes: The clinical record text is segmented into words to obtain the text segmentation sequence of the clinical record text; Named entity recognition is performed on the text segmentation sequence to identify medical entities in the text segmentation sequence, which are then used as a candidate entity set for the clinical record text. The semantic relationships of the candidate entity set are determined by classifying the relationships of any medical entity in the candidate entity set. Based on the semantic relationship, the candidate entity set is instantiated to obtain the original relation triplet of the target patient; The time information of the original relation triples is extracted from the clinical record text, and the time information is combined with the original relation triples as time tags to obtain the structured event data of the target patient.

4. The method for analyzing clinical data of chronic disease patients based on knowledge graphs as described in claim 2, characterized in that, The process of constructing a personalized knowledge graph for the target patient, using the identity information as nodes and the clinical examination indicator data and the structured event data as connecting edges, includes: Using the identity information as the central node, an initial graph structure for the target patient is constructed; The event subject, event action, and event object in the structured event data are used as event nodes. The event nodes are instantiated and associated with the central node to obtain the first type of connection edge of the initial graph structure. Using the examination indicators in the clinical examination indicator data as indicator nodes, attribute mapping is performed between the indicator nodes and the central node to obtain the second type of connection edges of the initial graph structure. In the time dimension, a third type of connection edge is constructed for the initial graph structure based on the proximity relationship between the event node and the indicator node; The central node, the first type of connecting edge, the second type of connecting edge, and the third type of connecting edge are injected into the initial graph structure to obtain the personalized knowledge graph of the target patient.

5. The method for analyzing clinical data of chronic disease patients based on knowledge graphs as described in claim 4, characterized in that, The third type of connection edge construction of the initial graph structure in the time dimension, based on the proximity relationship between the event node and the indicator node, includes: Synchronously acquire the first and second time-series labels corresponding to the event nodes and the indicator nodes; The time difference of the initial graph structure is obtained by parsing the difference between the first time tag and the second time tag. Based on the time difference, the association between the event node and the indicator node is established, and the third type of connection edge of the initial graph structure is obtained.

6. The method for analyzing clinical data of chronic disease patients based on knowledge graphs as described in claim 1, characterized in that, The step of performing conceptual mapping on the nodes and attribute edges in the personalized knowledge graph based on a preset medical knowledge ontology to obtain a conceptual knowledge graph of the target patient includes: Obtain a preset medical knowledge ontology, which includes concept nodes and concept relationship edges; Traverse the nodes to be mapped in the personalized knowledge graph and calculate the initial matching degree between the attribute information of the nodes to be mapped and the concept nodes in the medical knowledge ontology; Based on the initial matching degree, the concept node with the highest matching degree is selected from the medical knowledge ontology as the mapping target for the node to be mapped, thereby obtaining the node mapping relationship of the personalized knowledge graph; Traverse the unmapped attribute edges in the personalized knowledge graph, and match the unmapped attribute edges with the concept relationship edges to obtain the attribute edge mapping relationship of the personalized knowledge graph; Based on the node mapping relationship and the attribute edge mapping relationship, the personalized knowledge graph is conceptualized to obtain the conceptual knowledge graph of the target patient.

7. The method for analyzing clinical data of chronic disease patients based on knowledge graphs as described in claim 6, characterized in that, The calculation of the initial matching degree between the attribute information of the node to be mapped and the concept nodes in the medical knowledge ontology includes: Obtain the node type identifier, associated nodes, and attribute information of the node to be mapped, and obtain the concept type identifier, concept association pattern, and definition features of the concept node; Based on the consistency between the node type identifier and the concept type identifier, determine the type matching factor between the node to be mapped and the concept node; Based on the attribute information and the defined features, the semantic matching degree between the node to be mapped and the concept node is evaluated; Based on the association pattern between the associated nodes and the concept nodes, the structural matching degree between the node to be mapped and the concept node is quantified; Based on the clinical expert experience database, a first weight factor for the type matching factor, a second weight factor for the semantic matching degree, and a third weight factor for the structural matching degree are determined, wherein the clinical expert experience database includes historical case samples and expert-annotated matching results; Based on the type matching factor, the semantic matching degree, and the structural matching degree, the initial matching degree between the attribute information of the node to be mapped and the concept nodes in the medical knowledge ontology is calculated, wherein the formula for calculating the initial matching degree is as follows: ； In the formula, The initial matching degree between the node to be mapped and the concept node. For the type matching factor, The semantic matching degree, The structural matching degree, The first weighting factor, This is the second weighting factor. It is the third weighting factor, and satisfies α+β+γ=1.

8. The method for analyzing clinical data of chronic disease patients based on knowledge graphs as described in claim 1, characterized in that, By performing path mining on the conceptual knowledge graph, multi-order temporal path relationships of the target patient are obtained, including: Nodes with time attribute information in the conceptual knowledge graph are designated as time-sensitive nodes. Starting from the time-sensitive node, the path is expanded in the conceptual knowledge graph in chronological order to obtain the initial set of associated paths for the target patient; A temporal consistency check is performed on the initial associated paths in the initial associated path set to obtain the temporally consistent path set for the target patient; Based on preset time window parameters, the temporal proximity filtering of the temporal consistent path set is performed to obtain the temporal proximity path set of the target patient; The path length of the temporal proximity path set is statistically analyzed, and the paths whose statistical results reach a preset order threshold are marked as multi-order temporal association paths of the target patient.

9. The method for analyzing clinical data of chronic disease patients based on knowledge graphs as described in claim 1, characterized in that, The step of matching the multi-level temporal correlation path with the target patient's historical clinical event data, and determining the missing data items in the clinical dataset based on the matching results to update the data collection strategy for the clinical dataset, includes: Obtain the historical clinical event data of the target patient; The nodes in the multi-order temporal association path are matched with the historical clinical event data at the node level to obtain the first type of missing data items in the clinical dataset. By performing edge-level matching between the conceptual relationship edges in the multi-order temporal association path and the historical clinical event data, the second type of missing data items in the clinical dataset is obtained. Based on the first type of missing data items and the second type of missing data items, the data acquisition strategy of the clinical dataset is updated to optimize the subsequent data acquisition of the target patients.

10. A knowledge graph-based clinical data analysis system for chronic disease patients, characterized in that, The system is used to implement the knowledge graph-based clinical data analysis method for chronic disease patients as described in claim 1, the system comprising: The module comprises a data acquisition module, a semantic extraction module, a graph construction module, a concept mapping module, a path mining module, and a data update module, among which: The data acquisition module is used to acquire the clinical dataset of the target patient, which includes the target patient's identity information, clinical examination indicator data, and clinical record text. The semantic extraction module is used to extract semantics from the clinical record text to obtain the structured event data of the target patient. The graph construction module is used to construct a personalized knowledge graph of the target patient, using the identity information as nodes and the clinical examination indicator data and the structured event data as connection edges. The concept mapping module is used to perform concept mapping on the nodes and attribute edges in the personalized knowledge graph according to the preset medical knowledge ontology, so as to obtain the conceptual knowledge graph of the target patient. The path mining module is used to perform association path mining on the conceptual knowledge graph to obtain the multi-order temporal association path of the target patient; The synchronization update module is used to match the multi-level temporal association path with the historical clinical event data of the target patient, and based on the matching result, determine the missing data items in the clinical dataset to update the data collection strategy of the clinical dataset.

Images