A method and system for quality control of electronic medical records based on temporal causal conflict propagation

Abstract

This invention discloses an electronic medical record quality control method and system based on temporal causal conflict propagation, comprising: S1: collecting multi-source medical record data generated by patients during diagnosis and treatment, and preprocessing it to generate field-level record units; S2: constructing a temporal dependency graph of medical records using field-level record units as nodes and the dependencies between nodes as edges; S3: performing verification based on a preset data quality verification method, identifying abnormal nodes, defining the conflict intensity for each abnormal node, and identifying conflicting nodes; S4: using a conflict propagation analysis method based on path constraints to identify the set of affected nodes corresponding to the conflicting nodes; S5: assigning error source scores to each conflicting node and the affected nodes corresponding to the conflicting node to determine the error source node; S6: generating quality control results containing the error source node. This invention improves the completeness and accuracy of medical record anomaly identification, as well as the efficiency and traceability of quality control processing.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare, specifically to an electronic medical record quality control method and system based on temporal causal conflict propagation. Background Technology

[0002] With the continuous advancement of hospital informatization, electronic medical record (EMR) systems have been widely applied in various medical service scenarios, including outpatient and emergency departments, inpatient departments, laboratory tests, examinations, surgeries, nursing care, medical orders, and medication administration. EMRs not only record the entire patient treatment process but are also closely related to tasks such as medical record front page completion, medical insurance settlement, medical quality management, clinical research, and hospital operational analysis. Therefore, the completeness, accuracy, consistency, and timeliness of EMR data directly impact the quality of medical services and the reliability of subsequent medical management.

[0003] Most existing electronic medical record (EMR) quality control methods are based on preset rules, field validation, or document consistency comparison. For example, they check for missing key fields, inconsistencies between diagnosis and surgery names, exceeding time limits for writing, and matching of the first page data with discharge records. These methods can detect missing, incorrect, and inconsistent information in medical records to a certain extent, and are relatively straightforward and easy to deploy, thus becoming a common technical approach in current EMR quality control systems. However, in practical applications, EMR data comes from multiple sources, involves multiple stages, and has strong correlations. The same error message often gets referenced, inherited, or reused in subsequent progress notes, surgical records, discharge records, and first page data, creating a chain of errors. Existing technologies mostly focus on independent validation of individual fields or documents, typically only providing scattered anomaly alerts. They struggle to reveal the correlation between multiple anomalies and determine whether a particular error is the source of subsequent anomalies. Therefore, when multiple conflicts exist simultaneously in a medical record, manual inspection is still often required, resulting in low quality control efficiency. Furthermore, existing technologies in the medical record repair process typically only provide error alerts, lacking a mechanism for coordinated review and closed-loop updates targeting the root cause of errors. Even if a critical field has been modified, related subsequent fields may not be reviewed synchronously, easily leading to residual inconsistencies after local corrections, affecting the overall quality of medical records and the effectiveness of subsequent data use.

[0004] In the prior art, Chinese invention patent with publication number CN 119293194 A discloses a medical record quality management system and method, including: collecting medical record data and preprocessing the collected medical record data; using NLP to parse the preprocessed medical record data and extracting core fields of the medical records through NER; constructing a dynamic medical knowledge graph and matching the dynamic medical knowledge graph with the core fields of the medical records; after the matching is completed, using a graph reasoning algorithm to perform logical consistency verification on the medical record records, marking logical contradictions, and generating verification results; and generating a medical record quality audit report based on the verification results. This method only verifies and marks contradictions, making it difficult to reveal the correlation between multiple anomalies, and also difficult to determine whether a certain error is the source of multiple subsequent anomalies, thus failing to comprehensively assess the overall quality of the medical records.

[0005] Therefore, how to establish the dependencies between fields in multi-source heterogeneous data in electronic medical records, analyze the propagation impact of errors in the medical record formation process, further locate the initial source of errors, and generate a linkage repair path accordingly has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art, improve the concentration of electronic medical record quality control objects and the degree of abnormal correlation identification, achieve accurate location of error sources and closed-loop repair, and provide an electronic medical record quality control method and system based on temporal causal conflict propagation.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] An electronic medical record quality control method based on temporal causal conflict propagation includes:

[0009] S1: Collect multi-source medical record data generated by patients during the diagnosis and treatment process, and preprocess the collected multi-source medical record data to generate field-level record units;

[0010] S2: Construct a medical record time-series dependency graph using the field-level record units as nodes and the dependencies between nodes as edges; preset the dependency degree between nodes according to the dependency relationship type, and define the dependency strength for each edge according to the preset dependency degree between nodes;

[0011] S3: Based on a preset data quality verification method, the nodes in the medical record time-series dependency graph are verified, abnormal nodes are identified, and a conflict intensity is defined for each abnormal node. Nodes whose conflict intensity reaches a first preset threshold are identified as conflict nodes.

[0012] S4: A path constraint-based conflict propagation analysis method, taking the conflict node as the propagation starting point, searches downstream nodes along its dependency relationship, calculates the inter-node propagation strength of the conflict node on the propagation path according to the dependency strength using a preset calculation rule, identifies nodes whose inter-node propagation strength reaches a second preset threshold as affected nodes, and obtains the set of affected nodes corresponding to the conflict node.

[0013] S5: Assign an error source score to each conflicting node and the affected node corresponding to the conflicting node based on the propagation strength and conflict strength between the nodes, and determine the error source node based on the error source score;

[0014] S6: Generate quality control results containing the error source nodes.

[0015] By preprocessing multi-source medical record data in electronic medical records, field-level record units are generated and treated as individual nodes for centralized processing of medical record data. A time-series dependency graph reflecting dependencies is constructed, defining the dependency strength between nodes. Nodes are validated according to data quality verification methods to identify abnormal nodes, and conflict intensity is defined to quantify the degree of conflict between abnormal nodes, thus identifying conflicting nodes. By acquiring the inter-node propagation intensity of conflicting nodes and their downstream nodes, affected nodes are identified, enabling propagation analysis of conflicting nodes to analyze the propagation impact of errors in the medical record formation process. By defining error source scores to quantify the potential for conflicting nodes and their corresponding affected nodes to cause anomalies, the source of errors is located, thereby achieving precise error source localization. This improves the degree of anomaly correlation identification in electronic medical record quality control.

[0016] Preferably, the method for preprocessing the collected multi-source medical record data in step S1 includes:

[0017] Perform field mapping on the collected multi-source medical record data, mapping fields with different names but the same meaning to a unified field category;

[0018] After the field mapping is performed, the field value is formatted or mapped to a standard encoding based on its data type to obtain a uniformly expressed field value.

[0019] For text fields containing unstructured target information, information extraction is performed using a combination of template parsing, pattern matching, and terminology dictionary matching to form standardized field values, while preserving the original text position of the extracted content in the original record.

[0020] Based on the unified field category and the field values ​​obtained after the format conversion, standard encoding mapping or information extraction processing, a field-level record unit is generated; the field-level record unit includes a field content part consisting of a field name and a field value, and an additional attribute part consisting of an object identifier, an event identifier, a source record type, an event occurrence time, a record generation time, a source system type, a creator identifier, a number of modifications, and the original text position.

[0021] Preferably, in step S2, the dependency relationship types include temporal dependency, business dependency, and reference inheritance.

[0022] The temporal sequence refers to the temporal order between the content of the preceding node and the content of the following node in two nodes with a dependency relationship.

[0023] The business dependency relationship refers to the fact that the content of the previous node constitutes the basis for generating the content of the next node, or constitutes the basis for judging the content of the next node.

[0024] The reference inheritance relationship refers to the fact that the content of the subsequent node references, uses, or copies at least part of the content of the previous node.

[0025] Preferably, in step S2, the dependency strength is defined for each edge according to a preset inter-node dependency level, and the expression for the dependency strength is:

[0026] ;

[0027] in, For nodes To the node The edge between The strength of dependence; The baseline coefficient corresponding to the dependency relationship type; The coefficient representing the degree of dependency between nodes; For nodes With nodes Terminology overlap; The time interval between nodes. This represents the time decay parameter corresponding to the dependency type; and This is the adjustment coefficient.

[0028] Preferably, the data quality verification method in step S3 includes at least one of integrity verification, rationality verification, consistency verification, timeliness verification, and source cross-verification.

[0029] Preferably, in step S3, a conflict intensity is defined for each abnormal node, and the expression for the conflict intensity is:

[0030] ;

[0031] in, For nodes The intensity of the conflict; This is the node criticality coefficient, used to represent the degree of criticality of the node's content; To determine the degree of inconsistency in content; The degree of timing violation; To the degree of disagreement regarding the source; , and All are adjustment coefficients.

[0032] Preferably, the preset calculation rule in step S4 is as follows:

[0033] The expression for calculating the propagation strength between nodes is:

[0034] ;

[0035] in, For nodes For nodes The intensity of transmission; For the node To the node The set of propagation paths, where p is the index of a single propagation path; For a single propagation path, the edge The strength of dependence; These are path modulation coefficients, used to suppress the propagation of errors along the propagation path;

[0036] The expression for the path modulation coefficient is:

[0037] ;

[0038] in, This is the path length suppression coefficient; For the propagation path The number of sides; The total time span of the propagation path from the starting point to the end point. This is the decay parameter over time.

[0039] Preferably, in step S5, in the current conflict propagation subgraph consisting of conflicting nodes and the set of affected nodes corresponding to the conflicting nodes, the expression for the error source score is:

[0040] ;

[0041] in, For nodes Error source score, For nodes The corresponding set of affected nodes For nodes For nodes The intensity of transmission; For nodes The intensity of the conflict; Represents a node The time of the event, This indicates the earliest occurrence time of the event among the conflicting nodes and affected nodes in the current conflict propagation subgraph. For time adjustment parameters; For nodes The upstream support is used to represent the node's upstream support. The degree to which it is referenced by upstream nodes. This is a source reliability suppression term, used to indicate the degree of reliability of the field's source.

[0042] Preferably, in step S6, before generating the quality control result containing the error source node, the following is further performed: a review of the error source node and the downstream affected nodes corresponding to the error source node is conducted; the specific method for the review is as follows:

[0043] The error source node is identified as a priority for repair.

[0044] From the set of affected nodes corresponding to the conflict node, extract the downstream affected nodes associated with the source node of the error to form the review queue corresponding to the source node of the error;

[0045] Based on the dependency relationship type between the downstream affected nodes and the error source node, a preset review rule is established, and the downstream affected nodes in the review queue are reviewed according to the preset review rule.

[0046] An electronic medical record quality control system based on temporal causal conflict propagation includes:

[0047] The data acquisition and preprocessing unit is used to collect multi-source medical record data generated by patients during the diagnosis and treatment process, and to preprocess the collected multi-source medical record data to generate field-level record units.

[0048] The graph construction unit is used to construct a medical record time-series dependency graph with the field-level record units as nodes and the dependencies between nodes as edges; the degree of dependency between nodes is preset according to the dependency type, and the dependency strength is defined for each edge according to the preset degree of dependency between nodes.

[0049] The conflict node identification unit is used to verify the nodes in the medical record time-series dependency graph based on a preset data quality verification method, identify abnormal nodes, define the conflict intensity for each abnormal node, and identify nodes whose conflict intensity reaches a first preset threshold as conflict nodes.

[0050] The propagation analysis unit is used for the path constraint-based conflict propagation analysis method. Taking the conflict node as the propagation starting point, it searches downstream nodes along its dependency relationship. According to the dependency strength, it calculates the inter-node propagation strength of the conflict node on the propagation path using a preset calculation rule. Nodes whose inter-node propagation strength reaches a second preset threshold are identified as affected nodes, and the set of affected nodes corresponding to the conflict node is obtained.

[0051] An error source localization unit is used to assign an error source score to each conflicting node and the affected node corresponding to the conflicting node according to the propagation strength and conflict strength between the nodes, and to determine the error source node according to the error source score;

[0052] The result generation unit is used to generate quality control results that include the error source nodes.

[0053] The beneficial effects of this invention are as follows:

[0054] This invention addresses multi-source medical record data in electronic medical records, including diagnoses, laboratory tests, examinations, surgeries, medical orders, nursing care, medication administration, and document modification records, by performing field mapping and standardization. A medical record time-series dependency graph is constructed based on chronological order, business dependencies, and reference / inheritance relationships. Combining data quality verification methods and defining conflict intensity, conflict nodes are identified. The propagation scope and impact of conflicts are analyzed along the medical record time-series dependency graph, identifying the affected nodes corresponding to the conflict nodes. This extends traditional single-point verification to associated quality control, improving the completeness and accuracy of medical record anomaly identification. Furthermore, the invention pinpoints the error source nodes that trigger multiple anomalies, transforming existing single-point error reporting into source localization. This prioritizes identifying the source nodes most likely to cause subsequent related anomalies, reducing manual item-by-item checks and improving medical record quality control efficiency. Finally, the error source nodes and their corresponding downstream affected nodes are reviewed, generating repair suggestions and quality control results, thereby improving the efficiency and traceability of electronic medical record quality control processing. Attached Figure Description

[0055] The present invention will now be described in further detail with reference to the accompanying drawings:

[0056] Figure 1 This is a method block diagram of the present invention;

[0057] Figure 2 This is a schematic diagram of the medical record time-series dependency graph of the present invention. Detailed Implementation

[0058] like Figure 1 As shown, the present invention provides an electronic medical record quality control method based on temporal causal conflict propagation, which includes the following steps.

[0059] S1: Collect multi-source medical record data generated during the patient's diagnosis and treatment process, and preprocess the collected multi-source medical record data to generate field-level record units. Multi-source medical record data includes admission records, progress notes, discharge records, diagnostic information, laboratory results, examination reports, surgical records, nursing records, medical order records, medication execution records, and document modification logs, and synchronously collects the record generation time, signature time, operator identifier, number of modifications, and source system identifier.

[0060] To address the issues of inconsistent field names and formats across different systems and documents, specific methods for preprocessing collected multi-source medical record data include:

[0061] Field mapping is performed on the collected multi-source medical record data, mapping fields with different names but the same meaning to a unified field category.

[0062] After the field mapping is performed, the format is converted or standard encoding is mapped according to the data type of the field value to obtain a uniformly expressed field value. For example, the field value named "Time" is converted to standard encoding, such as 2000-00-00; the field values ​​named "Gender," "Department," "Test Item," "Drug Name," and "Surgery Name" are converted to a uniform format.

[0063] For text fields containing unstructured target information, information extraction is performed using a combination of template parsing, pattern matching, and terminology dictionary matching to form standardized field values, while preserving the original textual position of the extracted content in the original record. For example, key fields from long text fields such as diagnosis names, examination conclusions, surgery names, medication information, and time expressions in medical record data are extracted to form standardized field values.

[0064] Based on a unified field category and field values ​​obtained through format conversion, standard encoding mapping, or information extraction, field-level record units are generated. Each field-level record unit includes a field content portion consisting of a field name and field value, such as the field name being gender, age, diagnosis name, surgery name, test result, or discharge date; and an additional attribute portion consisting of one or more of the following: object identifier, event identifier, source record type, event occurrence time, record generation time, source system type, creator identifier, number of modifications, and original text location. For example, "Gender = Male in admission record" can be considered a field-level record unit, where:

[0065] Field name: Gender;

[0066] Field value: Male;

[0067] Object identifier: A patient's ID;

[0068] Event identifier: Hospitalization ID;

[0069] Source record type: Admission record;

[0070] Time of the incident (patient's hospitalization period): 2000-00-00;

[0071] Record generation time (the time this record was written): 2000-00-00.

[0072] To address the issues of inconsistent field names and formats across different systems and documents, field mapping and standardization are employed to unify and organize time, encoding, text content, and source identifiers, thereby generating field-level record units and forming computable data units.

[0073] S2: Construct a medical record time-series dependency graph using field-level record units as nodes and the dependencies between nodes as edges; preset the degree of dependency between nodes according to the dependency type, and define the dependency strength for each edge according to the preset degree of dependency between nodes.

[0074] like Figure 2 As shown, dependency types include temporal dependencies, business dependencies, and reference inheritance. Specifically, a temporal dependency means that, among two nodes with a dependency relationship, there is a temporal order between the content of the preceding node and the content of the following node. A business dependency means that the content of the preceding node forms the basis for generating the content of the following node, or forms the basis for judging the content of the following node. For example, there is a business dependency between node A1 (anomaly detection result) and node A2 (diagnostic conclusion), where the anomaly detection result can serve as the basis for the diagnostic conclusion; there is a business dependency between node B1 (preoperative diagnosis) and node B2 (surgery name), where the preoperative diagnosis determines the name of the subsequent surgery. A reference inheritance means that the content of the following node references, uses, or copies at least part of the content of the preceding node. For example, node C1 (admission record) and node C2 (daily medical record) have a reference inheritance relationship, where the daily medical record uses the diagnostic description from the admission record; node C2 (daily medical record) and node C3 (discharge record) have a reference inheritance relationship, where the discharge record uses the treatment process from the daily medical record.

[0075] In this embodiment, the method for constructing a medical record temporal dependency graph is as follows: Field-level record units are grouped according to object identifiers and event identifiers, and then sorted within each group according to the event occurrence time or record generation time. Using the sorted field-level record units of each group as nodes, edges are established based on preset temporal relationships, business dependencies, and reference inheritance relationships. Specifically, a time window is set between nodes to determine whether two nodes belong to a reasonable candidate association range, excluding nodes that are significantly too far apart in time, and retaining only node pairs that may have a dependency relationship within a reasonable time range. For example, the time window from test results to diagnosis is set to 7 days, the time window from preoperative diagnosis to surgical record is 14 days, and the time window from prescription to execution record is 72 hours.

[0076] In the medical record time sequence dependency graph, the dependency strength is defined for each edge according to the preset dependency degree between nodes. The expression for the dependency strength is:

[0077] ;

[0078] in, For nodes To the node The edge between The strength of dependence; The base coefficients are the base coefficients corresponding to the dependency relationship types. The base coefficients for time-sequence relationships range from 0.40 to 0.65, the base coefficients for business dependencies range from 0.80 to 0.95, and the base coefficients for reference-inheritance relationships range from 0.70 to 0.90. This is the dependency coefficient between nodes, used to indicate whether there is a clear correspondence or reference relationship between nodes. Its value ranges from 0 to 1. For example, if a later node directly references a previous node, then... Take 1; if there is no correspondence at all, then... Take 0. For nodes With nodes The terminology overlap is set to a value between 0 and 1. For example, if the medical terms in the content of two nodes completely overlap, then... Take 1; if they do not overlap at all, then Take 0. The time interval between nodes. This represents the time decay parameter corresponding to the dependency type, with a value range of 6 hours to 72 hours. and The adjustment coefficient ranges from 1.2 to 2.0. In this embodiment, both the time decay parameter and the adjustment coefficient are set manually.

[0079] This invention obtains the reference inheritance relationship between field-level record units based on the medical record generation process, and obtains the business dependency relationship between field-level record units based on medical business logic, so as to establish the sequential association between field-level record units, thereby obtaining the dependency relationship between nodes and constructing a medical record time-series dependency graph, thus organizing the originally scattered multi-source medical record data into a propagable and traceable association structure.

[0080] S3: Based on the preset data quality verification method, the nodes in the medical record time-series dependency graph are verified, abnormal nodes are identified, and a conflict intensity is defined for each abnormal node. Nodes whose conflict intensity reaches the first preset threshold are identified as conflict nodes.

[0081] In this embodiment, the data quality verification methods include integrity verification, rationality verification, consistency verification, timeliness verification, and source cross-verification. Integrity verification identifies missing key fields; consistency verification identifies inconsistencies in field values ​​across different stages of the same medical visit; rationality verification identifies records that do not conform to basic medical logic or business processes; timeliness verification identifies anomalies such as overdue writing and delayed supplementary entries; and source cross-verification identifies contradictions between the same field from different sources. For example, consistency verification is performed on the "admission diagnosis—surgical record—discharge diagnosis" sequence; time sequence rationality verification is performed on the "medical order issuance time—execution time" sequence; and source cross-verification is performed on the "laboratory report conclusion—disease description" sequence. If a node is identified as an abnormal node by two or more verification methods simultaneously, it is marked as a high-concern conflict field, and its priority in propagation analysis, error source location, and review is increased.

[0082] Define the conflict strength for each anomalous node. The expression for the conflict strength is:

[0083] ;

[0084] in, For nodes The conflict intensity, with a value ranging from 0 to 1; The node criticality coefficient is used to represent the criticality of the node content. In this embodiment, the node criticality coefficient of key fields such as main diagnosis on the homepage, surgery name, discharge date, and length of hospital stay ranges from 1.20 to 1.60, while the value range of general description fields ranges from 0.80 to 1.00. The value represents the degree of inconsistency in the content and ranges from 0 to 1. The value represents the degree of timing violation and ranges from 0 to 1. The value ranges from 0 to 1, representing the degree of divergence in the source. , and All are adjustment coefficients. The value range is 0.35 to 0.50. The value range is 0.20 to 0.35. The value range is 0.20 to 0.35. , and The sum of the three values ​​ranges from 0.85 to 1.00.

[0085] In this embodiment, the first preset threshold is 0.4. When a node is not considered a conflict node, it is not considered a conflict node. To further differentiate the conflict strength of conflict nodes, we define: when... When, it is determined to be a node with significant conflict; when When a node is identified as a suspected conflict node, both significant conflict nodes and suspected conflict nodes are input as conflict nodes into the conflict propagation analysis in step S4.

[0086] S4: The conflict propagation analysis method based on path constraints takes the conflict node as the propagation starting point, searches for downstream nodes along its dependency relationship, calculates the inter-node propagation strength of the conflict node on the propagation path according to the dependency strength using a preset calculation rule, identifies the nodes whose inter-node propagation strength reaches a second preset threshold as affected nodes, and obtains the set of affected nodes corresponding to the conflict node.

[0087] The preset calculation rules are as follows: The expression for calculating the propagation strength between nodes is:

[0088] ;

[0089] in, For nodes For nodes The propagation intensity, with a value ranging from 0 to 1; For the node To the node The set of propagation paths, where p is the index of a single propagation path; For a single propagation path, the edge The strength of dependence; These are path modulation coefficients used to suppress the false propagation of excessively long paths and weakly correlated propagation paths.

[0090] The expression for the path modulation coefficient is:

[0091] ;

[0092] in, This is the path length suppression coefficient, with a value ranging from 0.25 to 0.6. For the propagation path The number of sides, ranging from 2 to 5; The total time span of the propagation path from the starting point to the end point. This is the time span decay parameter, with a value range of 12h to 120h.

[0093] By setting path modulation coefficients, path constraints are applied to propagation analysis to avoid excessive accumulation of long paths due to simple addition, and to suppress erroneous propagation of excessively long paths and weakly correlated propagation paths.

[0094] In this embodiment, the second preset threshold is 0.3. When [a node is not affected], it is not considered an affected node. To further differentiate the strength of influence relationships, we define: when [a node is affected]... At that time, the node Determined as a node The propagation node of substantial impact (i.e., the node of substantial impact), when 0.3 When the node is identified as having a weak impact, both the node with a substantial impact and the node with a weak impact are included as affected nodes in the error source localization analysis in step S5.

[0095] For example, when there are significant conflicts at the admission diagnosis node, the system can calculate the propagation intensity along the path of "admission diagnosis → surgical record → postoperative course of illness → discharge diagnosis," thereby identifying which subsequent fields are associated anomalies rather than independent errors. Through the above steps, the originally scattered single-point quality control results can be organized into a propagation structure of "conflict origin - propagation path - affected fields," providing a basis for subsequent root cause field location.

[0096] By analyzing the propagation of identified conflicting fields along the time-series dependency graph of medical records to their downstream related nodes, it is determined whether the conflict will have a cascading impact on subsequent nodes such as diagnosis, surgery, medication, discharge records, and homepage reporting. By employing a graph-based propagation analysis method, combined with dependency type and time interval, the propagation intensity of the conflict is calculated, thereby expanding from a single conflicting node to a set of affected nodes. This extends traditional single-point verification to associated quality control, thereby improving the completeness and accuracy of medical record anomaly identification.

[0097] S5: Assign error source scores to each conflicting node and the affected nodes corresponding to the conflicting node based on the propagation strength and conflict strength between nodes, and determine the error source node based on the error source scores.

[0098] In the current conflict propagation subgraph, which consists of conflicting nodes and the set of affected nodes corresponding to the conflicting nodes, the expression for the error source score is:

[0099] ;

[0100] in, For nodes Error source score, For nodes The corresponding set of affected nodes For nodes For nodes The intensity of transmission; For nodes The intensity of the conflict; Represents a node The time of the event, This indicates the earliest occurrence time of the event among the conflicting nodes and affected nodes in the current conflict propagation subgraph. This is a time adjustment parameter, with a value range of 6 hours to 48 hours; For nodes The upstream support is used to represent the node's upstream support. The degree to which it is referenced by upstream nodes, with a value ranging from 0 to 1; This is a source reliability suppression term, used to indicate the reliability level of the field source. When the field source is feedback from the inspection system, feedback from the testing system, or equipment testing, the value range is 0.30 to 0.60; when the field source is manual entry, supplementary entry, or multiple rewrites, the value range is 0.05 to 0.30.

[0101] By combining factors such as the timing of the event, upstream support, and source reliability, the most likely source node to trigger multiple anomalies is identified. If a node experiences an event earlier, can be linked to more subsequently affected nodes, and lacks effective support from earlier nodes, then that node is identified as a more likely source of error. This transforms existing single-point error reporting into source localization, prioritizing the identification of the most likely source node to cause subsequent chain anomalies, thereby reducing the workload of manual item-by-item investigation and improving the efficiency of medical record quality control.

[0102] In this embodiment, the node with the largest error source score in the current conflict propagation subgraph is determined as the error source node; or when the difference between the largest error source score and the second largest error source score is less than a third preset threshold (such as 0.05), the two nodes corresponding to it are listed together as error source nodes.

[0103] S6: Review the error source node and the downstream affected nodes corresponding to the error source node, and generate a quality control result containing the error source node.

[0104] The specific review method is as follows:

[0105] Identify the error source node as the priority target for repair;

[0106] From the set of affected nodes corresponding to the conflict node, extract the downstream affected nodes associated with the source node of the error to form the review queue corresponding to the source node of the error.

[0107] For example, in a review queue: if the system determines that the node "Admission Diagnosis = Acute Appendicitis" is the source of the error, and subsequent fields such as the surgery name, postoperative progress notes, discharge diagnosis, and the main diagnosis on the medical record cover all have business dependencies or inheritance relationships with this source node, then the system extracts these subsequent fields as downstream affected nodes. Therefore, the resulting review queue is:

[0108] Admission diagnosis;

[0109] Surgical name;

[0110] Diagnostic descriptions in postoperative medical records;

[0111] Discharge diagnosis;

[0112] The main diagnosis on the first page of the medical record.

[0113] Based on the dependency relationship type between the downstream affected nodes and the error source node, preset review rules are established, and the downstream affected nodes in the review queue are reviewed according to the preset review rules.

[0114] Specifically, if the downstream affected node and the error source node have a reference inheritance relationship, then after the error source node is corrected, the downstream affected node will undergo a consistency review; if the downstream affected node and the error source node have a business dependency relationship, then after the error source node is corrected, the downstream affected node will undergo a business rationality review; if the downstream affected node only has a temporal relationship with the error source node, then the downstream affected node will be prompted with an association, and the review will not be performed directly.

[0115] The quality control results include at least: the source node of the error, the affected nodes, the original text location of each node, the conflict type, the repair priority, and the repair suggestions. The repair suggestions include manual verification and correction, linked review and confirmation, and automatic prompt write-back. Manual verification and correction are applicable to core nodes such as diagnosis names, surgery names, and discharge diagnoses; linked review and confirmation are applicable to nodes referenced, inherited, or copied from upstream nodes; automatic prompt write-back is applicable to format-related, time-related, and coding-related nodes. After the review and repair are completed, the conflict identification in step S3 and the propagation analysis in step S4 are performed again on the corrected nodes and their downstream affected nodes. If the corresponding conflict disappears, the node's status is updated to "repaired," and the quality control status of the entire medical record is updated synchronously. If conflicts still exist, they are retained in the review queue for further review. Thus, through review and correction, the electronic medical record quality control process is transformed from the traditional single-point error reporting method into a closed-loop processing flow of "error source location—linked review of affected fields—status write-back update."

[0116] By generating a review queue based on the relationship between the error source node and the affected nodes, and performing status write-back and re-verification analysis after repair, closed-loop processing of medical record quality control can be achieved, thereby improving the executability, consistency and traceability of quality control results.

[0117] This invention also provides an electronic medical record quality control system based on temporal causal conflict propagation, comprising:

[0118] The data acquisition and preprocessing unit is used to collect multi-source medical record data generated by patients during the diagnosis and treatment process, and to preprocess the collected multi-source medical record data to generate field-level record units.

[0119] The graph construction unit is used to construct a time-series dependency graph of medical records, with field-level record units as nodes and the dependencies between nodes as edges. The degree of dependency between nodes is preset according to the type of dependency relationship, and the dependency strength is defined for each edge according to the preset degree of dependency between nodes.

[0120] The conflict node identification unit is used to verify the nodes in the medical record time-series dependency graph based on a preset data quality verification method, identify abnormal nodes, define the conflict intensity for each abnormal node, and identify nodes whose conflict intensity reaches a first preset threshold as conflict nodes.

[0121] The propagation analysis unit is used for the path constraint-based conflict propagation analysis method. It takes the conflict node as the propagation starting point, searches for downstream nodes along its dependency relationship, calculates the inter-node propagation strength of the conflict node on the propagation path according to the dependency strength using a preset calculation rule, identifies the nodes whose inter-node propagation strength reaches a second preset threshold as affected nodes, and obtains the set of affected nodes corresponding to the conflict node.

[0122] The error source localization unit is used to assign an error source score to each conflicting node and the affected node corresponding to the conflicting node based on the propagation strength and conflict strength between nodes, and to determine the error source node based on the error source score;

[0123] The result generation unit is used to generate quality control results that include the error source nodes.

[0124] The system of the present invention is implemented based on the above method, and will not be described in detail here.

[0125] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A method for quality control of electronic medical records based on temporal causal conflict propagation, characterized in that, include: S1: Collect multi-source medical record data generated by patients during the diagnosis and treatment process, and preprocess the collected multi-source medical record data to generate field-level record units; S2: Construct a medical record time-series dependency graph using the field-level record units as nodes and the dependencies between nodes as edges; The degree of dependency between nodes is preset according to the type of dependency relationship, and the dependency strength is defined for each edge according to the preset degree of dependency between nodes; S3: Based on a preset data quality verification method, the nodes in the medical record time-series dependency graph are verified, abnormal nodes are identified, and a conflict intensity is defined for each abnormal node. Nodes whose conflict intensity reaches a first preset threshold are identified as conflict nodes. S4: A path constraint-based conflict propagation analysis method, taking the conflict node as the propagation starting point, searches downstream nodes along its dependency relationship, calculates the inter-node propagation strength of the conflict node on the propagation path according to the dependency strength using a preset calculation rule, identifies nodes whose inter-node propagation strength reaches a second preset threshold as affected nodes, and obtains the set of affected nodes corresponding to the conflict node. S5: Assign an error source score to each conflicting node and the affected node corresponding to the conflicting node based on the propagation strength and conflict strength between the nodes, and determine the error source node based on the error source score; S6: Generate quality control results containing the error source nodes.

2. The method according to claim 1, characterized in that, The method for preprocessing the collected multi-source medical record data in step S1 includes: Perform field mapping on the collected multi-source medical record data, mapping fields with different names but the same meaning to a unified field category; After the field mapping is performed, the field value is formatted or mapped to a standard encoding based on its data type to obtain a uniformly expressed field value. For text fields containing unstructured target information, information extraction is performed using a combination of template parsing, pattern matching, and terminology dictionary matching to form standardized field values, while preserving the original text position of the extracted content in the original record. Based on the unified field category and the field values ​​obtained after the format conversion, standard encoding mapping or information extraction processing, a field-level record unit is generated; the field-level record unit includes a field content part consisting of a field name and a field value, and an additional attribute part consisting of an object identifier, an event identifier, a source record type, an event occurrence time, a record generation time, a source system type, a creator identifier, a number of modifications, and the original text position.

3. The method according to claim 1, characterized in that, In step S2, the dependency relationship types include time sequence relationship, business dependency relationship and reference inheritance relationship; The temporal sequence refers to the temporal order between the content of the preceding node and the content of the following node in two nodes with a dependency relationship. The business dependency relationship refers to the fact that the content of the previous node constitutes the basis for generating the content of the next node, or constitutes the basis for judging the content of the next node. The reference inheritance relationship refers to the fact that the content of the subsequent node references, uses, or copies at least part of the content of the previous node.

4. The method according to claim 1, characterized in that, In step S2, the dependency strength of each edge is defined according to the preset inter-node dependency degree. The expression for the dependency strength is: ； in, For nodes To the node The edge between Dependence strength; The baseline coefficient corresponding to the dependency relationship type; The coefficient representing the degree of dependency between nodes; For nodes With nodes Terminology overlap; The time interval between nodes. This represents the time decay parameter corresponding to the dependency type; and This is the adjustment coefficient.

5. The method according to claim 1, characterized in that, The data quality verification method in step S3 includes at least one of integrity verification, rationality verification, consistency verification, timeliness verification, and source cross-verification.

6. The method according to claim 1, characterized in that, In step S3, a conflict intensity is defined for each abnormal node, and the expression for the conflict intensity is: ； in, For nodes The intensity of the conflict; This is the node criticality coefficient, used to represent the degree of criticality of the node's content; To determine the degree of inconsistency in content; The degree of timing violation; To the degree of disagreement regarding the source; , and All are adjustment coefficients.

7. The method according to claim 1, characterized in that, The preset calculation rule in step S4 is as follows: The expression for calculating the propagation strength between nodes is: ； in, For nodes For nodes The intensity of transmission; For the node To the node The set of propagation paths, where p is the index of a single propagation path; For a single propagation path, the edge Dependence strength; These are path modulation coefficients, used to suppress the propagation of errors along the propagation path; The expression for the path modulation coefficient is: ； in, This is the path length suppression coefficient; For the propagation path The number of sides; The total time span of the propagation path from the starting point to the end point. This is the decay parameter over time.

8. The method according to claim 1, characterized in that, In step S5, in the current conflict propagation subgraph consisting of conflicting nodes and the set of affected nodes corresponding to the conflicting nodes, the expression for the error source score is: ； in, For nodes Error source score, For nodes The corresponding set of affected nodes For nodes For nodes The intensity of transmission; For nodes The intensity of the conflict; Represents a node The time of the event, This indicates the earliest occurrence time of the event among the conflicting nodes and affected nodes in the current conflict propagation subgraph. For time adjustment parameters; For nodes The upstream support is used to represent the node's upstream support. The degree to which it is referenced by upstream nodes. This is a source reliability suppression term, used to indicate the degree of reliability of the field's source.

9. The method according to claim 1, characterized in that, In step S6, before generating the quality control result containing the error source node, the following is also performed: a review of the error source node and the downstream affected nodes corresponding to the error source node is conducted; the specific method for the review is as follows: The error source node is identified as a priority for repair. From the set of affected nodes corresponding to the conflict node, extract the downstream affected nodes associated with the source node of the error to form the review queue corresponding to the source node of the error; Based on the dependency relationship type between the downstream affected nodes and the error source node, a preset review rule is established, and the downstream affected nodes in the review queue are reviewed according to the preset review rule.

10. An electronic medical record quality control system based on temporal causal conflict propagation, characterized in that, include: The data acquisition and preprocessing unit is used to collect multi-source medical record data generated by patients during the diagnosis and treatment process, and to preprocess the collected multi-source medical record data to generate field-level record units. The graph construction unit is used to construct a medical record time-series dependency graph with the field-level record units as nodes and the dependencies between nodes as edges. The degree of dependency between nodes is preset according to the type of dependency relationship, and the dependency strength is defined for each edge according to the preset degree of dependency between nodes; The conflict node identification unit is used to verify the nodes in the medical record time-series dependency graph based on a preset data quality verification method, identify abnormal nodes, define the conflict intensity for each abnormal node, and identify nodes whose conflict intensity reaches a first preset threshold as conflict nodes. The propagation analysis unit is used for the path constraint-based conflict propagation analysis method. Taking the conflict node as the propagation starting point, it searches downstream nodes along its dependency relationship. According to the dependency strength, it calculates the inter-node propagation strength of the conflict node on the propagation path using a preset calculation rule. Nodes whose inter-node propagation strength reaches a second preset threshold are identified as affected nodes, and the set of affected nodes corresponding to the conflict node is obtained. An error source localization unit is used to assign an error source score to each conflicting node and the affected node corresponding to the conflicting node according to the propagation strength and conflict strength between the nodes, and to determine the error source node according to the error source score; The result generation unit is used to generate quality control results that include the error source nodes.

Images