Diagnosis and treatment guide self-adaptive medical advice compliance verification method

Abstract

The invention relates to the technical field of medical informatics, and discloses a diagnosis and treatment guide self-adaptive medical advice compliance verification method. The doctor's advice terms are extracted and standardized, so that the problems of mixed multi-source expression and inconsistent semantics are solved; according to the method, terms are mapped into nodes, a traceable undirected graph is constructed according to guide co-occurrence and interdisciplinary reference relations, and the shortest hop count and the unique path are automatically obtained through matrix operation. And setting a hop count threshold, binarizing to judge term compliance, and combining with path density marking credibility to realize quantitative standard and structural risk prompt. And integrating judgment and credibility of all terms, outputting medical advice compliance results in a grading manner, and dynamically adjusting a threshold value according to batch data to adapt to an actual environment. All non-compliant paths are archived as an abnormal graph, and follow-up knowledge optimization and auditing are assisted.

Description

Technical Field

[0001] This invention relates to the field of medical informatics technology, specifically to a method for verifying the compliance of medical orders that is adaptive to treatment guidelines. Background Technology

[0002] With the continuous improvement of hospital informatization, electronic medical records (EMR) and hospital information systems (HIS) have become the core carriers of clinical workflow. Medical orders, as the execution instructions for diagnosis and treatment, are directly related to medication safety, the rationality of examinations and tests, and the feasibility of interdisciplinary collaboration. At the same time, clinical practice guidelines are constantly being updated, covering multiple specialties, multiple stages, and multiple conditions and constraints (indications, contraindications, dosage ranges, combination relationships, sequential timing, etc.). In real-world scenarios, physicians need to make treatment decisions within a limited timeframe under complex guideline constraints, which can easily lead to inconsistencies with the guidelines, fragmented interdisciplinary terminology, or incomplete chains of evidence.

[0003] Existing common approaches to compliance verification include "condition-action" judgments primarily based on rule engines, and correlation alerts based on statistical or learning methods. The former relies on manually maintaining a large number of discrete rules, making it difficult to cover cross-disciplinary scenarios; when guideline entries or pharmacopoeias are updated, rule synchronization costs are high, and conflicts or omissions are prone to occur between rules. The latter largely relies on empirical weights or model parameters, resulting in insufficient transparency of conclusions and difficulty in providing traceable, structured evidence for clinical review; when sample distributions drift, data is sparse, or novel combinations appear, the uncertainty of the learning method's output increases, easily leading to over-warnings or missed detections. Furthermore, medical order terminology comes from diverse sources (synonyms, aliases, abbreviations, and hospital jargon coexist). Without unified terminology standardization and equivalence class processing, it will lead to synonyms and failed cross-disciplinary mappings, further weakening the coverage and accuracy of verification. Multidisciplinary team (MDT) approaches present new technical challenges. Direct constraints within a single discipline are often insufficient to explain the rationale for cross-disciplinary collaboration. Clinical rationale is often manifested as an "indirect chain of connections," such as: underlying condition—pre-medication examination—anesthesia assessment—perioperative medication—postoperative review—follow-up examination. If these "multi-hop" relationships are exhaustively enumerated using traditional rules, the number of rules increases exponentially, leading to soaring maintenance costs and consistency risks. If a black-box model is used instead, it becomes difficult to provide an interpretable path between each medical order and the guidelines, affecting clinical adoption.

[0004] To address this, this study proposes an adaptive compliance verification method for medical orders based on treatment guidelines. It extracts all clinically relevant terms from the medical order text using a structured approach and establishes a mapping relationship between synonymous, homographed, and coded standardized terms based on a standardized terminology set from the treatment guidelines. This resolves the ambiguity and unification issues caused by different descriptions and annotations. Furthermore, the solution uses graph theory as its core, modeling the multidimensional relationships between standardized terms as an undirected graph. Dynamic construction of the graph structure is achieved through node co-occurrence and explicit cross-disciplinary referencing. Utilizing the path computation capabilities of adjacency matrices and matrix powers, the shortest hop count between medical order terms and guideline terms is efficiently calculated, and this is used to determine the compliance level of the medical order and identify potential abnormal paths or knowledge breakpoints. Simultaneously, the solution introduces a data-driven adaptive threshold adjustment mechanism, enabling intelligent evolution and dynamic optimization of the compliance judgment threshold, effectively adapting to changes in treatment scenarios across different departments and time batches in actual business operations. Through the archiving and graph evolution of abnormal paths, traceable evidence is further provided for guideline maintenance and cross-disciplinary knowledge integration, significantly improving the efficiency of compliance auditing and quality improvement. Summary of the Invention

[0005] This invention provides a method for verifying the compliance of medical orders that is adaptive to treatment guidelines, thereby helping to solve the problems mentioned in the background section.

[0006] This invention provides the following technical solution: a method for verifying the compliance of medical orders that is adaptive to treatment guidelines, comprising:

[0007] Extract all clinically relevant terms from the medical order text to be verified, and compare them with the standardized terminology set in the corresponding diagnosis and treatment guidelines for synonyms, homographs, and homocodes, and output the standardized representative terms;

[0008] All standard representative terms are mapped to graph nodes, and an undirected term association graph is constructed based on the co-occurrence of terms in the guide chapters and the rules of explicit cross-disciplinary reference.

[0009] Based on the adjacency matrix of the graph, the shortest hop count between medical order term nodes and guideline term nodes is calculated by matrix exponentiation, and the shortest path is uniquely determined under total order of nodes;

[0010] Set a maximum acceptable hop count threshold, construct a binary indicator function, and determine the compliance of the shortest hop count for each term;

[0011] The path density value is calculated based on the edge density of the subgraph induced by the shortest path, and low or high confidence labels are generated accordingly.

[0012] By combining compliance indicators and credibility markers for various terms, a compliance score for medical orders is calculated, and the verification results are classified as compliant, partially compliant, or non-compliant.

[0013] Based on the actual compliance rate and terminology distribution of a predefined batch of medical orders, the jump threshold is dynamically adjusted in conjunction with the effective diameter of the graph.

[0014] For all non-compliant or anomalous terms, record their unique witness path or unreachable state, and construct and output the evolutionary version sequence of the anomalous term graph.

[0015] Optionally, the step of extracting all clinically relevant terms from the medical order text to be verified and comparing them with the standardized terminology set in the corresponding treatment guidelines for synonymy, homonymy, and homocoding, and outputting standardized representative terms, specifically includes:

[0016] Extract all clinically relevant terms from the medical order text to be verified, including disease, drug, procedure, dosage form and test item, by syntactic word segmentation;

[0017] Let the extracted term set be . ;in, For set The Middle A term; For terms in a set The serial number in; The total number of terms extracted from the medical order;

[0018] Extract all standardized terms from the corresponding department's treatment guidelines and transcribe them as follows:

[0019] ;in, For set The first in A term; For terms in a set The serial number in; Total number of terms in the guide; This refers to a standardized set of terms corresponding to the clinical guidelines.

[0020] For any term pair If the two are equivalent in meaning, form, and encoding, then denote them as... ;in, Indicates the equivalence relation between terms; For any two terms;

[0021] Define encoding function This maps terms to unique canonical codes; and provides total order relations in the code space. ;in, It is a total order symbol, defined as the lexicographical order over the encoded value domain, used to compare the size of two codes;

[0022] Constructing a canonical representative function ; where, function Input terms Output and The term with the smallest encoding in the equivalence class;

[0023] right Implement one by one Combine equivalent terms to construct a complete set of terms. :

[0024] ;in, For the complete set of terms The first in item; For the complete set of terms The serial number in; For the complete set of terms The total number of elements in the middle satisfies ;

[0025] For each term Assign a unique number Let the mapping function be... ;

[0026] like ,but and It refers to the same term and the same number.

[0027] Optionally, the step of mapping all standard representative terms into graph nodes, and constructing an undirected term association graph based on the co-occurrence of terms in the guide chapters and cross-disciplinary explicit reference rules, specifically includes:

[0028] Construct a graph representing the relationships between terms. , , ;in, It is a set of nodes in a graph, where each node corresponds to a canonical term. For a set of nodes The Middle Each node, and the terminology One-to-one correspondence; Let be the set of edges of the graph;

[0029] Let the guide structure unit set Each This refers to a discrete unit within a guideline chapter, section, or treatment pathway window; where, For the first One structural unit; For structural unit indexes;

[0030] Set up the assignment mapping from terms to structural units. ;in, For set-value mapping functions, input terms Output the subset of structural units that have appeared in it; For set The power set, i.e. The set of all subsets;

[0031] If terminology Appearing in unit ,but ;

[0032] Suppose an interdisciplinary citation set If guidelines from different disciplines show cross-references, then ; Candidate pairs of undirected edges representing explicit cross-references between guidelines from different disciplines; For ordered pairs, it is only used as a notation when constructing candidate sets, and is subsequently transformed from symmetry into undirected edges; explicit cross-references refer to cross-disciplinary references or citations that appear explicitly in the text and can be identified and traced in a rule-based manner.

[0033] Set the adjacency matrix elements:

[0034] ;in, , , ; Matrix elements, adjacency matrix The Yuan, value ;

[0035] get Adjacency matrix .

[0036] Optionally, based on the adjacency matrix of the graph, the shortest hop count between medical order terminology nodes and guideline terminology nodes is calculated using matrix exponentiation, and the shortest path is uniquely determined under total order of nodes, specifically including:

[0037] Index of medical order terminology nodes:

[0038] , ;in, Standardized medical order terminology exist Node index in; Standardized Guide Terminology exist Node index in;

[0039] Construct a matrix exponentiation sequence: , ;in, For matrix of Power of 1 Time is defined as the identity matrix ; for A 1-dimensional identity matrix with 1 dimension on the diagonal and 0 on the rest.

[0040] Construct the shortest hop count function:

[0041] ;in, For the node To the node through Count the number of paths connected by edges; Input is term subscript The output is from Corresponding node arrive Corresponding node If the shortest number of hops is unattainable, then it is considered as... ;

[0042] like ,but ;

[0043] Obtaining medical order terminology To any set of guide terms Shortest number of hops :

[0044] ;

[0045] like This indicates that it is not connected to any guide terminology;

[0046] when hour:

[0047] Make it possible The length is ;

[0048] Set the guide endpoint index set ;

[0049] Construct the set of shortest paths for witnesses :

[0050] ;in, For the product step index, the set of values ; For sequence elements, represent the ordered node indices of a candidate path; It is a product expression, indicating that from arrive Does all adjacent edges exist? A value of 1 indicates that each segment exists.

[0051] when hour, ;

[0052] In the natural total order of the node index Below, set the unique path as follows:

[0053] ;in, Indicates in The unique shortest path selected lexicographically;

[0054] remember The node sequence is , ;in, For sequence elements, Rewritten as from the 1st to the 1st Index of each node; This represents the number of nodes in the path.

[0055] like The path contains only nodes. .

[0056] Optionally, the step of setting a maximum acceptable hop count threshold and constructing a binary indicator function to determine the compliance of the shortest hop count for each term specifically includes:

[0057] Set the maximum acceptable number of hops threshold to ;

[0058] Construct compliance indicator functions: ;in, To indicate the function value, for the term The compliance is judged using a binary method;

[0059] like Then the terminology is compliant;

[0060] like Then the terminology is non-compliant.

[0061] Optionally, the step of calculating the path density value based on the edge density of the subgraph induced by the shortest path and generating low-confidence or high-confidence labels accordingly specifically includes:

[0062] Constructing the density of induced subgraphs: ;in, The density of the induced subgraph composed of nodes with the shortest paths; Let be an element of the adjacency matrix, representing the th element in the path. The node and the first Does an edge exist between the nodes? For path location index, indicating the path node sequence Two different positions in the middle; For the node index, respectively, the first node in the path... With the The global node number of each node;

[0063] For each unit Set its node index set: ;in, Includes appearing in structural units The node index corresponding to all terms in;

[0064] when At that time, the unit density ;in, For set The number of elements; Representation unit Density of the internal inducible factor plot;

[0065] Set threshold ;

[0066] Constructing low-confidence tags .

[0067] Optionally, the process of integrating compliance indicators and credibility markers for various terms to calculate a compliance score for medical orders, and classifying the verification results as compliant, partially compliant, or non-compliant, specifically includes:

[0068] Calculate the compliance score of medical orders ;

[0069] Constructing a set of non-compliant terms: ;in, To eliminate all non-compliant terms in the current medical orders The set that constitutes;

[0070] when Output compliance and retain each of and ;

[0071] when At that time, compliance of the output portion needs to be reviewed and listed in One by one:

[0072] Minimum number of hops ;like Give the unique shortest path Node sequence index and , ;like Mark as unreachable, no path, no output. and ;

[0073] when The output is non-compliant, and all terms are entered into the exception log.

[0074] Optionally, the step of dynamically adjusting the jump threshold based on the actual compliance rate and terminology distribution of a predefined batch of medical orders, combined with the effective diameter of the graph, specifically includes:

[0075] Construct the first Collection of medical orders ;in, For batch indexing; For the first batch A doctor's order; The serial number of the medical order within this batch; For batch Total number of internal medical orders;

[0076] For the A medical order, assuming the number of its terms is . The terminology compliance indicator quantity is Its medical order compliance score ; For the first The first doctor's order Compliance of the terminology; For the first The percentage of medical orders that comply with terminology;

[0077] Set a fully compliant indicator function : , ;

[0078] Calculate the first Batch of actual compliance rate ;

[0079] Set the effective diameter of the diagram. :

[0080] ;

[0081] If the graph is completely discrete, let ;

[0082] The term "completely discrete" means that no two pairs exist. Connected nodes;

[0083] Initial run initialization: ;in, For the first run Setting value;

[0084] Batch Update: ;in, For the first The threshold for batch execution time; It is a symbolic function;

[0085] like ,but Increase by 1, not exceeding ;

[0086] like ,but Subtract 1, but not less than 1;

[0087] like , remains unchanged.

[0088] Optionally, the step of recording the unique witness path or unreachable state of all non-compliant or anomalous terms, and constructing and outputting the evolutionary version sequence of the anomalous term graph, specifically includes:

[0089] Set the abnormal terminology set ;

[0090] For each ,Record ;

[0091] like Record the unique witness path ;

[0092] like The recorded state is unreachable;

[0093] Constructing anomaly graphs ;in, This is the set of node indices corresponding to the exception terms; The set of edges in the anomaly graph;

[0094] In the Output version at the end of the batch , forming a sequence ;in, In the first The abnormal map version that is frozen after batching is completed.

[0095] The present invention has the following beneficial effects:

[0096] 1. By comparing medical order texts with guideline standard terminology through a triple comparison of synonyms, homographs, and identical codes, and employing unique coding and total sequence mapping, this solution completely resolves the industry challenge of inconsistently classifying the same medical entity under different descriptions. Compared to existing methods relying primarily on simple keyword searches or manual comparisons, this solution enables automatic semantic consistency determination, effectively preventing misjudgments and omissions caused by "same thing, different name" or "same name, different thing," thus improving the accuracy and reusability of subsequent determinations. Furthermore, the standardized numbering system and equivalence class merging operation lay a solid data foundation for large-scale automated processing, giving the system excellent scalability and maintainability. In the real-world environment of multi-source, heterogeneous clinical data with varying descriptive styles, this standardization mechanism greatly enhances data availability and cross-system compatibility, creating conditions for downstream algorithms to eliminate redundancy, unify semantics, and enhance the reliability of discrimination.

[0097] 2. Abstract terminology entities are modeled using graph nodes, integrating co-occurrence relationships within guideline chapters and explicit cross-disciplinary references to form a multi-dimensional, multi-level undirected terminology relationship graph. Unlike traditional linear rule chains or planar relationship tables, this approach can comprehensively express complex interdisciplinary and knowledge reference networks, enhancing semantic coverage depth. Especially in today's increasingly interdisciplinary collaborative clinical activities, the introduction of graph structures not only provides structural support for complex path tracing and dynamic relationship evolution but also reserves space for the future introduction of advanced reasoning algorithms such as graph neural networks. Furthermore, the rule-based and traceable edge definitions eliminate subjective weighting and "black box" risks, achieving objectivity and interpretability of judgment logic, and providing high reference value for structural optimization and knowledge maintenance of guideline entries. Compared to existing rule-based systems that primarily rely on planar comparisons, the graph mechanism in this step improves the model's flexibility, accuracy, and scalability.

[0098] 3. Based on the exponentiation of adjacency matrices, this innovative approach transforms the complex inter-termine path problem into a programmable matrix operation process. It achieves efficient and automated calculation of the shortest hop count between medical order terms and guideline terms, and introduces a total order path selection mechanism to ensure uniqueness and reproducibility. Compared to manual judgment or heuristic traversal methods, this method boasts extremely high computational efficiency and consistency, stably outputting the optimal witness path even in complex graph structures, eliminating loops, ambiguities, and high implementation complexity issues common in traditional algorithms. Furthermore, the matrix exponentiation concept is compatible with the characteristics of sparse graph data, facilitating batch processing in large-scale terminology networks. This step lays a solid foundation for compliance quantification and batch statistical analysis, solving the previous problems of incomparable distances between multiple targets and the inability to automatically output auditable paths. It represents a key technological breakthrough in improving the objectivity and scalability of automated verification.

[0099] 4. Compliance judgment is quantified into a precise delimitation of the shortest hop count and a threshold. A binary indicator function clearly defines compliance at the terminology level, and explicit rejection is applied to "unreachable" cases. Unlike previous subjective standards or multiple fuzzy judgments, this mechanism uses the inherent properties of graph structures to objectify and proceduralize compliance judgments, achieving the effects of rule-based pre-implementation, high consistency, and strong reusability. Especially in setting the hop count threshold, the solution combines actual business data with structural statistics such as the effective diameter of the graph, balancing clinical needs with mathematical rationality, providing support for differentiated settings across disciplines and subsequent adaptive optimization. This method reduces the risk of false positives and false negatives that are easily caused by traditional keyword-based or human experience-based approaches, achieving an adjustable and controllable compliance standard system.

[0100] 5. A density assessment mechanism based on shortest path induced subgraphs is introduced, using the connectivity between nodes along the path as a supplementary indicator of credibility in compliance judgment. Compared to traditional methods that only focus on path length, this approach can promptly mark short but structurally loose paths as low-credibility, assisting clinical staff in prioritizing review and avoiding illusions caused by accidental connectivity. This density and risk warning mechanism does not require subjective weighting and can dynamically adapt to the inherent structure of the actual knowledge network, effectively improving the system's risk identification capability and decision support level. Furthermore, the comparison mechanism between path density and unit mean provides quantitative early warning for potential problems such as structural breaks and knowledge silos, which is an important supplement to improving the reliability of the compliance verification system.

[0101] 6. By organically integrating the binary results of terminology-level compliance with credibility markers, a comprehensive scoring mechanism at the prescription level and a three-tiered classification (compliant, partially compliant, and non-compliant) are proposed, achieving holistic risk identification from fine-grained judgment to the business level. Unlike previous simple statistical or all-or-nothing judgment methods, this solution can intelligently output specific problem terms and detailed evidence for "partially compliant" scenarios, providing strong technical support for subsequent manual review and continuous improvement. This method not only improves the precision of statistical analysis and quality management but also provides a fully automated and traceable solution for batch verification, regulatory reports, and subject quality control, improving work efficiency and the scientific nature of decision-making.

[0102] 7. For the first time in the industry, an adaptive optimization mechanism is applied to compliance threshold adjustment. Through batch-level feedback on actual compliance rates and graph structure characteristics, dynamic and intelligent optimization of the hop count threshold is achieved. Compared to fixed parameters or static configurations, the solution can automatically converge to the optimal range based on actual business conditions and historical performance. This avoids unintended negative impacts on legitimate paths and effectively suppresses false positive path penetration, continuously optimizing the judgment strategy. This mechanism, based on batch-level small-step adjustments and smooth transitions, ensures the stability and scientific nature of the system's long-term evolution, enhances its adaptability and generalization capabilities across different disciplines and business stages, and represents a significant innovation in promoting self-learning and refined management of compliance algorithms.

[0103] 8. A full-scale archiving mechanism for the unique path and unreachable state of anomalous terms is proposed, and an evolutionary sequence of the anomalous terminology graph is constructed in a versioned manner, providing a solid structured chain of evidence for the maintenance of the clinical knowledge base, guideline optimization, and historical tracing. Unlike the previous passive system that only judges right or wrong without tracing the root cause, this solution achieves dynamic identification and visualization of structural problems such as high-incidence fault lines and interdisciplinary barriers through full-process anomaly tracking, enhancing the interpretability and proactive governance capabilities of the knowledge network. The batch and versioned output mechanism of the anomalous graph not only facilitates historical comparison and evolutionary analysis but also provides decision-making references for regulatory authorities and guideline maintenance teams, helping to form a continuously optimized, auditable, and scientifically closed-loop compliance assurance system. Attached Figure Description

[0104] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0105] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0106] Example, refer to Figure 1 An adaptive method for verifying the compliance of medical orders based on treatment guidelines, comprising:

[0107] Extract all clinically relevant terms from the medical order text to be verified, and compare them with the standardized terminology set in the corresponding diagnosis and treatment guidelines for synonyms, homographs, and homocodes, and output the standardized representative terms;

[0108] All standard representative terms are mapped to graph nodes, and an undirected term association graph is constructed based on the co-occurrence of terms in the guide chapters and the rules of explicit cross-disciplinary reference.

[0109] Based on the adjacency matrix of the graph, the shortest hop count between medical order term nodes and guideline term nodes is calculated by matrix exponentiation, and the shortest path is uniquely determined under total order of nodes;

[0110] Set a maximum acceptable hop count threshold, construct a binary indicator function, and determine the compliance of the shortest hop count for each term;

[0111] The path density value is calculated based on the edge density of the subgraph induced by the shortest path, and low or high confidence labels are generated accordingly.

[0112] By combining compliance indicators and credibility markers for various terms, a compliance score for medical orders is calculated, and the verification results are classified as compliant, partially compliant, or non-compliant.

[0113] Based on the actual compliance rate and terminology distribution of a predefined batch of medical orders, the jump threshold is dynamically adjusted in conjunction with the effective diameter of the graph.

[0114] For all non-compliant or anomalous terms, record their unique witness path or unreachable state, and construct and output the evolutionary version sequence of the anomalous term graph.

[0115] This paper proposes an adaptive compliance verification method for medical orders based on clinical practice guidelines, encompassing a complete process from medical order terminology extraction, terminology standardization, knowledge graph modeling, path reasoning, compliance determination, to anomaly tracing and dynamic threshold adaptation. First, by comprehensively extracting and standardizing all clinical terms involved in medical orders, the semantic barriers of mixed expressions and polysemy in multi-source data are effectively addressed. Second, standardized terms are mapped to graph nodes, and a traceable undirected terminology association graph is constructed using co-occurrence relationships and cross-disciplinary references within the clinical practice guidelines. This overcomes the limitations of previous rule-based systems that only perform static comparisons and cannot represent complex disciplinary networks. Through matrix operations on the adjacency relationships of the graph structure, the shortest hop count between medical order terms and guideline terms is automatically calculated, ensuring unique path determination and avoiding subjective and non-repeatable judgment risks. Next, an adjustable hop count threshold is set, and terminology compliance is determined using a binary function, quantifying compliance standards and eliminating the arbitrariness of traditional manual interpretation. Finally, credibility markers are generated based on indicators such as path density to highlight potential structural risks. Ultimately, by considering both terminology compliance and credibility, an overall compliance rating for medical orders is provided, and parameters are dynamically adjusted to adapt to changes in the actual clinical environment. All abnormal paths are archived and their evolutionary graphs are output, facilitating subsequent knowledge optimization and audit backtracking. Overall, this method effectively improves the automation, interpretability, and adaptability of verification, solving common industry challenges such as the strong subjectivity, difficulty in batch traceability, and difficulty in knowledge maintenance inherent in existing manual verification methods. It strongly supports the standardization of clinical medical orders and the improvement of interdisciplinary collaboration.

[0116] The process involves extracting all clinically relevant terms from the medical order text to be verified, comparing them with the standardized terminology set in the corresponding treatment guidelines for synonymy, homonymy, and homocoding, and outputting standardized representative terms, specifically including:

[0117] Extract all clinically relevant terms from the medical order text to be verified, including disease, drug, procedure, dosage form and test item, by syntactic word segmentation;

[0118] The method of extracting all clinically relevant terms by syntactic word segmentation is an existing technology.

[0119] Let the extracted term set be . ;in, For set The Middle A term; For terms in a set The serial number in; The total number of terms extracted from the medical order;

[0120] Ensure that all units related to guideline judgment in the medical order text are fully enumerated and structured to form a unique set of input sources for subsequent graph construction and jump count calculation, avoiding false non-compliance caused by omissions or statistical bias caused by duplication;

[0121] Extract all standardized terms from the corresponding department's treatment guidelines and transcribe them as follows:

[0122] ;in, For set The first in A term; For terms in a set The serial number in; Total number of terms in the guide; This refers to a standardized set of terms corresponding to the clinical guidelines.

[0123] Establish a targeted reference terminology domain as the "target set" for compliance assessment. All subsequent accessibility, shortest path, and density calculations will be based on this domain. For reference, ensure that the judgment strictly aligns with the guidelines;

[0124] For any term pair If the two are equivalent in meaning, form, and encoding, then denote them as... ;in, Indicates the equivalence relation between terms; For any two terms;

[0125] Define encoding function This maps terms to unique, standardized codes, such as the Unified Subject Code; and provides total order relations in the coding space. ;in, It is a total order symbol, defined as the lexicographical order over the encoded value domain, used to compare the size of two codes;

[0126] Constructing a canonical representative function ; where, function Input terms Output and The term with the smallest encoding in the equivalence class;

[0127] right Implement one by one Combine equivalent terms to construct a complete set of terms. :

[0128] ;in, For the complete set of terms The first in item; For the complete set of terms The serial number in; For the complete set of terms The total number of elements in the middle satisfies ;

[0129] For each term Assign a unique number Let the mapping function be... ;

[0130] like ,but and Use the same terminology and the same number;

[0131] Redundancy and standardization of terms from different sources and with different annotation styles are achieved, forming a globally unique numbering system. This standardization is the foundation for one-to-one correspondence between graph nodes, avoiding the splitting of multiple points with "same name for different things" or the erroneous merging of "different things with the same name" in subsequent path calculations.

[0132] Excerpt from the original medical order:

[0133] "Azithromycin was administered intravenously. Preoperative liver function tests were performed. The patient has a history of penicillin allergy." (Rule extraction yielded:) , , , ;

[0134] Appears in the guide section:

[0135] "Liver function assessment required before medication," "Alternatives for penicillin allergy," and "Precautions for azithromycin use." These can be summarized as follows: , , , .

[0136] Equivalent folding: "Archie" "Unify to "Azithromycin use"; "Liver function" is now unified under "liver function test";

[0137] Generate the complete set , , , , , ;

[0138] ;

[0139] Consistent alignment:

[0140] , , ;

[0141] , , .

[0142] This approach focuses on the extraction and standardization of medical order terminology. Specifically, it utilizes natural language processing technologies such as syntactic word segmentation to meticulously extract various relevant terms from medical orders, including those related to diseases, medications, procedures, dosage forms, and tests. These terms are then compared with standard terminology in treatment guidelines for synonymy, homonymy, and identical coding. The innovation lies in unifying terms with different expressions, sources, and styles under a global numbering system through equivalent merging and standardized representative coding. This completely avoids the normalization failures, path calculation confusion, and statistical errors caused by "different names for the same thing" or "different things with the same name" issues in traditional systems. Furthermore, this process ensures that all judgment-related units are structured, serving as the sole input source for subsequent graph construction and compliance judgment, significantly improving the completeness and accuracy of the input data. Compared to solutions relying solely on manual rules or simple word list comparisons, this method supports larger-scale, cross-domain, and multi-dimensional data fusion processing. This meticulous terminology standardization step resolves the risks associated with data heterogeneity and inconsistent expression, providing a solid and reliable foundation for downstream knowledge network analysis and automated judgment, effectively improving the traceability and automation level of the entire process.

[0143] The process involves mapping all representative terms of the specifications to graph nodes, constructing an undirected term association graph based on the co-occurrence of terms in the guide sections and the rules for explicit cross-disciplinary citations. Specifically, this includes:

[0144] Construct a graph representing the relationships between terms. , , ;in, It is a set of nodes in a graph, where each node corresponds to a canonical term. For a set of nodes The Middle Each node, and the terminology One-to-one correspondence; It is the set of edges of a graph; it transforms abstract terminology entities into graph nodes, establishes a computable structural carrier, and provides a strict object basis for the subsequent definition of graph theory quantities such as "adjacency", "path", and "hop count";

[0145] Let the guide structure unit set Each This refers to a discrete unit within a guideline chapter, section, or treatment pathway window; where, For the first One structural unit; For structural unit indexes;

[0146] Set up the assignment mapping from terms to structural units. ;in, For set-value mapping functions, input terms Output the subset of structural units that have appeared in it; For set The power set, i.e. The set of all subsets;

[0147] If terminology Appearing in unit ,but ;

[0148] Suppose an interdisciplinary citation set If guidelines from different disciplines show cross-references, then ; Candidate pairs of undirected edges representing explicit cross-references between guidelines from different disciplines; For ordered pairs, this notation is used only when constructing candidate sets, and is subsequently transformed from symmetry to undirected edges; explicit cross-references refer to cross-disciplinary references or citations that are explicitly present in the text and can be identified and traced in a regular manner; for example: direct wording: such as "see Section 3.2 of the Anesthesiology Guidelines", "anticoagulation management requirements according to the Cardiovascular Disease Guidelines", "according to the Obstetric Imaging Safety Specifications", etc.; coded or numbered citations: such as "see the liver function assessment process in Appendix A" and "see the dose adjustment principles in Table B-4"; linking or bibliographical citations: such as giving the chapter number, title, or standard code of another discipline in the text;

[0149] Set the adjacency matrix elements:

[0150] ;in, , , ; Matrix elements, adjacency matrix The Yuan, value ;

[0151] The two types of provable connections, "co-occurrence within the same unit" and "explicit cross-disciplinary references," are transformed into undirected edges in a graph, ensuring that the origin of each edge is traceable and its semantics are clear. This rule avoids subjective weighting and uninterpretable similarity, ensuring an objective and compliant basis for subsequent hop count and density calculations.

[0152] get Adjacency matrix It completes the transformation from text corpus to computable graph matrix, providing a unified and sparse-friendly data structure for subsequent matrix exponentiation, reachability and shortest path solutions.

[0153] Establish a node set ;

[0154] Guide structure unit: , , , ;

[0155] Assignment mapping: , , , , ;

[0156] Explicit cross-reference (text mentions "Azithromycin usage terms and conditions, see pre-treatment assessment requirements"): ,express and There is a clear cross-unit indication;

[0157] Edge positioning rules:

[0158] Co-occurrence in the same unit: in the same The terms appearing in the text are linked together, such as make and Connected, make and Connected;

[0159] Interdisciplinary citation: Add The edge;

[0160] Obtain the undirected edge set: , and have , ;

[0161] hour, The non-zero symmetric elements are: , , Explanation: Medications are linked to assessments, examinations are linked to assessments, allergies are linked to alternatives; all others are 0.

[0162] The key focus is on constructing a knowledge graph, systematically mapping standardized terminology to graph nodes, and establishing edge relationships based on co-occurrence relationships within the clinical guidelines and interdisciplinary references. This organically combines abstract terminology with a structured knowledge network in clinical practice, ensuring that each node and edge has authentic and interpretable source evidence, unlike existing approaches that rely on simple co-occurrence of the same term or flat databases. Furthermore, by introducing structural units (such as chapters, sections, and clinical pathway windows) and explicit referencing mechanisms, the actual flow and interaction of knowledge across different disciplines and items can be accurately tracked and recorded. This approach addresses the static nature and lack of dynamic adaptability of previous knowledge systems, laying a computable and visual foundation for path reasoning, anomaly tracing, and interdisciplinary collaboration. Since the graph's adjacency relationships are entirely derived from the logical structure and authentic references of clinical guidelines, without relying on subjective experience or manual weighting, the subsequent algorithm output possesses extremely high interpretability and scientific validity. This is highly beneficial in multidisciplinary collaboration and variable clinical pathway applications, effectively improving the system's flexibility and the reliability of its judgments.

[0163] Based on the adjacency matrix of the graph, the shortest hop count between medical order terminology nodes and guideline terminology nodes is calculated using matrix exponentiation, and the shortest path is uniquely determined under total order of nodes. Specifically, this includes:

[0164] Index of medical order terminology nodes:

[0165] , ;in, Standardized medical order terminology exist Node index in; Standardized Guide Terminology exist Node index in;

[0166] Construct a matrix exponentiation sequence: , ;in, For matrix of Power of 1 Time is defined as the identity matrix ; for A 1-dimensional identity matrix with 1 dimension on the diagonal and 0 on the rest.

[0167] Construct the shortest hop count function:

[0168] ;in, For the node To the node through Count the number of paths connected by edges; Input is term subscript The output is from Corresponding node arrive Corresponding node If the shortest number of hops is unattainable, then it is considered as... ;

[0169] like ,but ;

[0170] Using matrix exponentiation, we can determine the "step size". The reachability of "reachability" is transformed into algebraic discrimination, thereby obtaining the shortest hop count between any pair of terms using deterministic rules, laying the foundation for subsequent compliance threshold comparison and path extraction;

[0171] Obtaining medical order terminology To any set of guide terms Shortest number of hops :

[0172] The "shortest distance from terminology to guidelines" is condensed into a single numerical indicator, which is used for length constraints when determining compliance at the terminology level and extracting paths, thus avoiding ambiguity and implementation complexity caused by multiple objectives.

[0173] like This indicates that it is not connected to any guide terminology;

[0174] when hour:

[0175] Make it possible The length is ;

[0176] Set the guide endpoint index set ;

[0177] Construct the set of shortest paths for witnesses :

[0178] ;in, For the product step index, the set of values ; For sequence elements, represent the ordered node indices of a candidate path; It is a product expression, indicating that from arrive Does all adjacent edges exist? A value of 1 indicates that each segment exists.

[0179] when hour, ;

[0180] In the natural total order of the node index Below, set the unique path as follows:

[0181] ;in, Indicates in The unique shortest path selected lexicographically;

[0182] remember The node sequence is , ;in, For sequence elements, Rewritten as from the 1st to the 1st Index of each node; This represents the number of nodes in the path.

[0183] like The path contains only nodes. ;

[0184] The unique "witness" is identified among all shortest feasible paths, which facilitates stable reproduction and audit trail; at the same time, the "empty product convention" and "simple path property" eliminate the ambiguity of zero jumps and loops, ensuring the readability and comparability of the path output.

[0185] Example:

[0186] right (Mapped to) To each guide term: ( () is a zero jump; to ( )because , is 1 jump; to ( The shortest route must be taken Or, not connected to 5, not reachable in the current sample;

[0187] get , ,

[0188] ;

[0189] ( )arrive :because , ;

[0190] right The set of candidate shortest paths is The only witness ;

[0191] If there are two paths of the same length (such as...) and (If the length is the same), the unique one is selected according to the lexicographical order of the node index; if there are duplicate nodes in the path, they are removed because they contradict the "shortest" property.

[0192] The algorithm revolves around reasoning algorithms for paths and hop counts between terms. Through matrix operations, the adjacency relationships in the terminology graph are transformed into a programmable process for calculating the shortest hop count and unique witness path, enabling automated, batch-based path reasoning. This innovation addresses the shortcomings of traditional manual interpretation or heuristic traversal methods, such as low efficiency, difficulty in standardization, and susceptibility to errors. A total-order optimal path mechanism ensures that each group of nodes outputs a unique, traceable, and reproducible shortest path, facilitating subsequent verification and automated tracing. Complex cases such as unreachable paths and loops are handled uniformly through explicit conventions, eliminating ambiguity and incomparability in the output. This technical detail improves the accuracy and transparency of compliance judgments, while providing stable structured data support for subsequent statistical analysis and knowledge optimization. Compared to existing approaches using single-rule judgments or fuzzy matching, path reasoning capabilities enable the system to solve complex logical chains across multiple disciplines, enhancing scalability and adaptability.

[0193] The process of setting a maximum acceptable hop count threshold and constructing a binary indicator function to determine the compliance of the shortest hop count for each term specifically includes:

[0194] Set the maximum acceptable number of hops threshold to Quantify the "permissible distance for compliance" in advance and solidify the clinically acceptable depth of interdisciplinary connections into a clear discrete threshold, so that the compliance judgment results are consistent and interpretable.

[0195] The maximum acceptable number of hops threshold is Its function: Indicates the maximum allowed number of reasoning steps between medical order terminology and guideline terminology. If the shortest number of steps does not exceed this threshold, the terminology is considered compliant; if it exceeds or is unattainable, it is considered non-compliant. It represents the upper limit on the number of allowed interdisciplinary "intermediary nodes." A smaller value (stricter) reduces the allowance of "accidental connections," lowering false positives, but may reject some reasonable but slightly longer scenarios, increasing false negatives. A larger value (more lenient) covers more interdisciplinary collaborations, lowering false negatives, but may introduce structurally fragile paths, requiring reliance on "path density and low-confidence markers" for verification. Therefore, Controlling the "radius" of compliance requires a balance between the institution's risk appetite, interdisciplinary collaboration habits, and the granularity of guidelines. Value selection criteria: Consistency with the effective diameter of the graph: It should not exceed the upper bound of the maximum shortest distance on "reachable pairs"; otherwise, the threshold loses its practical meaning. Linkage with historical batch performance: In step seven, this solution uses "full compliance rate" for single-step discrete adaptive fine-tuning, causing the threshold to gradually approach the target level, while setting upper and lower limits (not less than 1, not greater than the effective diameter) to prevent loss of control. Combining structural statistics and departmental consensus: If the guideline chapters are finely divided and have high edge density, a smaller threshold is usually sufficient to cover most reasonable paths; if the interdisciplinary chain is naturally long (e.g., perioperative period—anesthesia—anticoagulation), the initial value can be slightly higher. Value recommendations: Initial value: It can be "half the effective diameter of the graph," then clamped between 1 and the effective diameter, as a general starting point. Scenario-based fine-tuning: For scenarios involving direct intradisciplinary connections between medication and examinations, a value of 0 or 1 can be used; for common scenarios requiring interdisciplinary mediation (e.g., "pre-medication—a certain assessment—medication"), a value of 1 or 2 can be used; for pathways with known long chains but clinical rationality (e.g., multi-stage perioperative linkage), an initial value of 2 can be used, followed by fine-tuning based on batch data using an adaptive mechanism. Governance collaboration: When low-confidence markers frequently appear at a certain hop count level, consideration can be given to not relaxing the threshold. Under the premise of improving the connection of the relevant sections or refining the guide items, we should prioritize improving the structural density and naturally reduce the proportion of low credibility.

[0196] Construct compliance indicator functions: ;in, To indicate the function value, for the term The compliance is judged using a binary method;

[0197] like Then the terminology is compliant;

[0198] like Then the terminology is non-compliant;

[0199] The "shortest distance" is directly mapped to an actionable compliance conclusion, and "unreachable" situations are explicitly rejected, ensuring that each term has a definite decision exit and can be used for statistical summarization.

[0200] Example:

[0201] set up (Direct hits or hits via one intermediary node are allowed). Calibration: smaller values ​​are more stringent; larger values ​​are more lenient; subsequent steps will be adaptively fine-tuned based on batch data.

[0202] For the example above: , ;

[0203] If there is a certain term or In the current The judgment was made as .

[0204] By setting a maximum acceptable hop count threshold and designing a binary compliance indicator function, compliance judgment is transformed into a strictly programmable standard process. Compared to traditional solutions relying solely on experience or manual judgment, this method can flexibly address the varying requirements of different clinical departments regarding pathway rationality, quantifying and pre-positioning compliance judgment standards. Especially when interdisciplinary pathways are long and knowledge links are complex, setting appropriate thresholds can reduce compliance judgment bias caused by false positives and false negatives. The upper and lower limits of the hop count threshold and its dynamic adjustment mechanism effectively avoid potential risks such as misjudgment and loss of control in judgment. The solution also clarifies the handling of unreachable situations, ensuring the uniqueness and generalizability of the judgment results. In summary, this process not only makes the verification results more consistent and scientific but also greatly reduces the burden of manual quality control, facilitating its promotion and large-scale application in actual medical management.

[0205] The calculation of path density values ​​based on the edge density of the subgraph induced by the shortest path, and the generation of low-confidence or high-confidence labels accordingly, specifically includes:

[0206] Constructing the density of induced subgraphs: ;in, The density of the induced subgraph composed of nodes with the shortest paths; Let be an element of the adjacency matrix, representing the th element in the path. The node and the first Does an edge exist between the nodes? For path location index, indicating the path node sequence Two different positions in the middle; For the node index, respectively, the first node in the path... With the The global node number of each node; measures the overall connectivity density between nodes traversed by the shortest path. When the path is short but the connections between nodes are sparse, low density will indicate potential "accidental connectivity," providing a basis for prioritizing manual review without changing the compliance binary conclusion;

[0207] For each unit Set its node index set: ;in, Includes appearing in structural units The node index corresponding to all terms in;

[0208] when At that time, the unit density ;in, For set The number of elements; Representation unit Density of the internal inducible factor plot;

[0209] Set threshold ;

[0210] Constructing low-confidence tags ;

[0211] Using the average experience density within the same guideline unit as a "credible lower bound" to mark the shortest path with sparse structure, a feasible review priority mechanism and risk warning channel can be formed without relying on external weights.

[0212] Example:

[0213] right The two nodes are completely connected. ;

[0214] If the shortest path of a certain term is Then an inspection is required. , , Three pairs of connections; if only and Existence, and The density is then "number of connected edges / maximum possible number of edges", that is... ;

[0215] Statistical Guidelines for Unit Density:

[0216] In Chinese terminology, for fully connected elements, the element density is denoted as 1. In Chinese terminology, for fully connected elements, the element density is denoted as 1; taking the average yields... ;

[0217] If a shortest path density If it is not changed, a low-confidence flag is set for priority manual review; otherwise, it is left unchanged. Whether it is compliant or not.

[0218] A path density-based credibility assessment is introduced. Unlike previous methods that relied solely on path length, this approach further examines the overall connectivity compactness between nodes on the shortest path, serving as a basis for risk alerts and review priority. When a path is short but structurally sparse, low-credibility markers effectively warn of "accidental connectivity" or potential structural vulnerabilities. This mechanism requires no additional subjective weighting and is entirely based on the dynamic adaptation of the knowledge network's inherent structure, providing clear guidance for subsequent manual review and continuous optimization. Within the guidance unit, the average empirical density serves as a "credibility lower bound," automatically marking high-risk paths to ensure that the judgment is both rigorous and flexible enough to handle real-world complexities. This approach overcomes the shortcomings of traditional pure path-based methods that "only consider distance and not quality," providing strong guarantees for system security, reliability, and risk controllability.

[0219] The system integrates compliance indicators and credibility markers for various terms to calculate a compliance score for medical orders, and categorizes the verification results as compliant, partially compliant, or non-compliant. Specifically, this includes:

[0220] Calculate the compliance score of medical orders The terminology-level binary results are aggregated into overall compliance at the order level for use in decision-making classification (compliant / partial / non-compliant) and batch statistics, facilitating regulatory reports and departmental quality control.

[0221] Constructing a set of non-compliant terms: ;in, To eliminate all non-compliant terms in the current medical orders The set that constitutes;

[0222] when Output compliance and retain each of and ;

[0223] when At that time, compliance of the output portion needs to be reviewed and listed in One by one:

[0224] Minimum number of hops ;like Give the unique shortest path Node sequence index and , ;like Mark as unreachable, no path, no output. and ;

[0225] when The output is non-compliant, and all terms are entered into the exception log;

[0226] The overall status of medical orders is clearly classified and the problem points are listed. For each non-compliant term, traceable quantitative evidence (shortest number of hops, path sequence, density and low confidence label) is provided, which provides direct clues for clinical review and revision.

[0227] Example:

[0228] A certain medical order contains three terms: ,but This system organically integrates terminology-level compliance with credibility markers, outputting a comprehensive compliance score and grading (compliant, partially compliant, non-compliant) at the order level. This multi-level grading not only quickly identifies the overall risk level but also outputs specific non-compliant terms, minimum number of hops, and path details, providing detailed operational clues for clinical review and subsequent revisions. Systematically aggregating terminology compliance to the overall order level improves management convenience and statistical comprehensiveness. Compared to traditional systems that only focus on a single dimension, this solution's granular quantifiable output better supports batch quality control, tiered supervision, and continuous optimization, improving implementation efficiency and operability in practice.

[0229] The method of dynamically adjusting the jump threshold based on the actual compliance rate and terminology distribution of a predefined batch of medical orders, combined with the effective diameter of the graph, specifically includes:

[0230] Construct the first Collection of medical orders ;in, For batch indexing; For the first batch A doctor's order; The serial number of the medical order within this batch; For batch Total number of internal medical orders;

[0231] For the A medical order, assuming the number of its terms is . The terminology compliance indicator quantity is Its medical order compliance score ; For the first The first doctor's order Compliance of the terminology; For the first The percentage of medical orders that comply with terminology;

[0232] Set a fully compliant indicator function : , ;

[0233] Calculate the first Batch of actual compliance rate ;

[0234] Set the effective diameter of the diagram. :

[0235] ;

[0236] If the graph is completely discrete, let ;

[0237] The term "completely discrete" means that no two pairs exist. Connected nodes;

[0238] Based on the proportion of true and complete compliance at the batch level We measure the external achievement of the current threshold strategy and give an upper limit constraint through the effective diameter of the graph to ensure that the threshold update does not go out of bounds and is interpretable.

[0239] Initial run initialization: ;in, For the first run Setting value;

[0240] Batch Update: ;in, For the first The threshold for batch execution time; It is a symbolic function; Target compliance rate;

[0241] like ,but Increase by 1, not exceeding ;

[0242] like ,but Subtract 1, but not less than 1;

[0243] like It remains unchanged;

[0244] Adjust using single-step integers to make Focusing on the target compliance rate Adaptive convergence, and always subject to The double constraint with the lower bound of 1 avoids policy oscillations and ineffective widening of the unreachable interval.

[0245] Example:

[0246] A total of [number] batches were reviewed Of the 15 medical orders, 15 were "fully compliant" (i.e., each was...). );

[0247] but ;

[0248] Effective diameter estimation of the diagram: In the current Enumerating the shortest distance between connected pairs yields a maximum value of 3. ;

[0249] Set target compliance rate ,current Adjusted according to the rules:

[0250] If the current ,but and not exceeding ;

[0251] Explanation: The next batch will be more stringent, encouraging clinicians to transform "two-hop rationale" into "direct connection or one-hop rationale" (e.g., by supplementing guideline sections or optimizing pathways to make the structure more compact).

[0252] An adaptive optimization mechanism for the hop count threshold based on batch-level medical order data is proposed. By statistically analyzing the actual compliance rate and terminology distribution of each batch of medical orders and combining this with the effective diameter of the knowledge graph, the hop count threshold is dynamically adjusted, enabling the judgment criteria to automatically optimize according to actual business performance and subject characteristics. This mechanism solves the problem that static parameters are easily fixed and difficult to adapt to changing clinical scenarios, allowing the system to ensure the stability of the judgment criteria while evolving flexibly based on feedback. Through limited single-step fine-tuning, the system can smoothly and gradually converge to the optimal strategy, preventing parameter oscillations and strategy runaway. Compared with previous experience-based settings or single-batch calibration methods, this mechanism improves the scientific rigor, continuous improvement capability, and applicability of the judgment process, providing important technical support for large-scale automated quality control.

[0253] The process of recording the unique witness path or unreachable state of all non-compliant or anomalous terms, and constructing and outputting the evolutionary version sequence of the anomalous term graph, specifically includes:

[0254] Set the abnormal terminology set ;

[0255] For each ,Record ;

[0256] like Record the unique witness path ;

[0257] like The recorded state is unreachable;

[0258] The most crucial structural evidence for post-event review and knowledge maintenance is accumulated: non-compliance points, corresponding shortest paths, or unreachable facts, thereby supporting targeted revisions and structured reviews across disciplines;

[0259] Constructing anomaly graphs ;in, This is the set of node indices corresponding to the exception terms; It is the set of edges in the anomaly graph; it connects scattered anomaly terms with each other to form an anomaly subgraph, which is used to identify high-incidence patterns (such as a cross-disciplinary node cluster) and structural fracture zones, providing a visual tool for guide maintenance and path optimization;

[0260] In the Output version at the end of the batch , forming a sequence ;in, In the first Anomaly graph versions are frozen after approval; the temporal changes of anomaly graphs are recorded in version sequence, supporting longitudinal evaluation of the effectiveness of rule adjustments and the evolution of subject boundaries, and allowing for retrospective analysis of the basis at any point in time, meeting the evidence chain requirements for compliance audits and continuous quality improvement.

[0261] A structured archiving and graph evolution mechanism for anomalous terms and paths is proposed. For all non-compliant terms, their unique witness path or unreachable state is recorded, and a sequence of anomalous term graph versions is output in batches. This approach breaks through the passive mode of the traditional system where "judgment is the endpoint," and achieves proactive governance and continuous optimization of the knowledge network through dynamic monitoring and summarization of high-incidence anomalies and fault lines. The temporal evolution of the anomaly graph not only facilitates historical data tracing but also provides an intuitive structural basis for guideline maintenance, path optimization, and interdisciplinary collaboration. The system can backtrack the anomaly structure at any point in time, establishing a complete chain of evidence for compliance auditing and quality improvement. Compared with existing simple log archiving or manual summarization methods, this mechanism greatly improves the intelligence and efficiency of knowledge management, promoting the long-term healthy development of the medical order compliance system.

[0262] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0263] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for verifying the compliance of medical orders that is adaptive to treatment guidelines, characterized in that, include: Extract all clinically relevant terms from the medical order text to be verified, and compare them with the standardized terminology set in the corresponding diagnosis and treatment guidelines for synonyms, homographs, and homocodes, and output the standardized representative terms; All standard representative terms are mapped to graph nodes, and an undirected term association graph is constructed based on the co-occurrence of terms in the guide chapters and the rules of explicit cross-disciplinary reference. Based on the adjacency matrix of the graph, the shortest hop count between medical order term nodes and guideline term nodes is calculated by matrix exponentiation, and the shortest path is uniquely determined under total order of nodes; Set a maximum acceptable hop count threshold, construct a binary indicator function, and determine the compliance of the shortest hop count for each term; The path density value is calculated based on the edge density of the subgraph induced by the shortest path, and low or high confidence labels are generated accordingly. By combining compliance indicators and credibility markers for various terms, a compliance score for medical orders is calculated, and the verification results are classified as compliant, partially compliant, or non-compliant. Based on the actual compliance rate and terminology distribution of a predefined batch of medical orders, the jump threshold is dynamically adjusted in conjunction with the effective diameter of the graph. For all non-compliant or anomalous terms, record their unique witness path or unreachable state, and construct and output the evolutionary version sequence of the anomalous term graph.

2. The method for verifying the compliance of medical orders based on adaptive treatment guidelines according to claim 1, characterized in that, The process involves extracting all clinically relevant terms from the medical order text to be verified, comparing them with the standardized terminology set in the corresponding treatment guidelines for synonymy, homonymy, and homocoding, and outputting standardized representative terms, specifically including: Extract all clinically relevant terms from the medical order text to be verified, including disease, drug, procedure, dosage form and test item, by syntactic word segmentation; Let the extracted term set be . ;in, For set The Middle A term; For terms in a set The serial number in; The total number of terms extracted from the medical order; Extract all standardized terms from the corresponding department's treatment guidelines and transcribe them as follows: ;in, For set The first in A term; For terms in a set The serial number in; Total number of terms in the guide; This refers to a standardized set of terms corresponding to the clinical guidelines. For any term pair If the two are equivalent in meaning, form, and encoding, then denote them as... ;in, Indicates the equivalence relation between terms; For any two terms; Define encoding function This maps terms to unique canonical codes; and provides total order relations in the code space. ;in, It is a total order symbol, defined as the lexicographical order over the encoded value domain, used to compare the size of two codes; Constructing a canonical representative function ; where, function Input terms Output and The term with the smallest encoding in the equivalence class; right Implement one by one And combine equivalent terms to construct a complete set of terms. : ;in, For the complete set of terms The first in item; For the complete set of terms The serial number in; For the complete set of terms The total number of elements in the middle satisfies ; For each term Assign a unique number Let the mapping function be... ; like ,but and It refers to the same term and the same number.

3. The method for verifying the compliance of medical orders based on adaptive treatment guidelines according to claim 2, characterized in that, The process involves mapping all representative terms of the specifications to graph nodes, constructing an undirected term association graph based on the co-occurrence of terms in the guideline chapters and the rules for explicit cross-disciplinary citations. Specifically, this includes: Construct a graph representing the relationships between terms. , , ;in, It is a set of nodes in a graph, where each node corresponds to a canonical term. For a set of nodes The Middle Each node, and the terminology One-to-one correspondence; Let be the set of edges of the graph; Let the guide structure unit set each This refers to a discrete unit within a guideline chapter, section, or treatment pathway window; where, For the first One structural unit; For structural unit indexes; Set up the assignment mapping from terminology to structural units. ;in, For set-value mapping functions, input terms Output the subset of structural units that have appeared in it; For set The power set, i.e. The set of all subsets; If terminology Appearing in unit ,but ; Suppose an interdisciplinary citation set If guidelines from different disciplines show cross-references, then ; Candidate pairs of undirected edges representing explicit cross-references between guidelines from different disciplines; For ordered pairs, it is only used as a notation when constructing candidate sets, and is subsequently transformed from symmetry into undirected edges; explicit cross-references refer to cross-disciplinary references or citations that appear explicitly in the text and can be identified and traced in a rule-based manner. Set the adjacency matrix elements: ;in, , , ; Matrix elements, adjacency matrix The Yuan, value ; get Adjacency matrix .

4. The method for verifying the compliance of medical orders based on adaptive treatment guidelines according to claim 3, characterized in that, Based on the adjacency matrix of the graph, the shortest hop count between medical order terminology nodes and guideline terminology nodes is calculated using matrix exponentiation, and the shortest path is uniquely determined under total order of nodes. Specifically, this includes: Index of medical order terminology nodes: , ;in, Standardized medical order terminology exist Node index in; Standardized Guide Terminology exist Node index in; Construct a matrix exponentiation sequence: , ;in, For matrix of Power of 1 Time is defined as the identity matrix ; for A 1-dimensional identity matrix with 1 dimension on the diagonal and 0 on the rest. Construct the shortest hop count function: ;in, For the node To the node through Count the number of paths connected by edges; Input is term subscript The output is from Corresponding node arrive Corresponding node If the shortest number of hops is unattainable, then it is considered as... ; like ,but ; Obtaining medical order terminology To any set of guide terms Shortest number of hops : ； like This indicates that it is not connected to any guide terminology; when hour: Make it possible The length is ; Set the guide endpoint index set ; Construct the set of shortest paths for witnesses : ;in, For the product step index, the set of values ; For sequence elements, represent the ordered node indices of a candidate path; It is a product expression, indicating that from arrive Does all adjacent edges exist? A value of 1 indicates that each segment exists. when hour, ; In the natural total order of the node index Below, set the unique path selection as follows: ;in, Indicates in The unique shortest path selected lexicographically; remember The node sequence is , ;in, For sequence elements, Rewritten as from the 1st to the 1st Index of each node; This represents the number of nodes in the path. like The path contains only nodes. .

5. The method for verifying the compliance of medical orders based on adaptive treatment guidelines according to claim 4, characterized in that, The process of setting a maximum acceptable hop count threshold and constructing a binary indicator function to determine the compliance of the shortest hop count for each term specifically includes: Set the maximum acceptable number of hops threshold to ; Construct compliance indicator functions: ;in, To indicate the function value, for the term The compliance is judged using a binary method; like Then the terminology is compliant; like Then the terminology is non-compliant.

6. The method for verifying the compliance of medical orders based on adaptive treatment guidelines according to claim 5, characterized in that, The calculation of path density values ​​based on the edge density of the subgraph induced by the shortest path, and the generation of low-confidence or high-confidence labels accordingly, specifically includes: Constructing the density of induced subgraphs: ;in, The density of the induced subgraph composed of nodes with the shortest paths; Let be an element of the adjacency matrix, representing the th element in the path. The node and the first Does an edge exist between the nodes? For path location index, indicating the path node sequence Two different positions in the middle; For the node index, respectively, the first node in the path... With the The global node number of each node; For each unit Set its node index set: ;in, Includes appearing in structural units The node index corresponding to all terms in; when At that time, the unit density ;in, For set The number of elements; Representation unit Density of the internal inducible factor plot; Set threshold ; Constructing low-confidence tags .

7. The method for verifying the compliance of medical orders based on adaptive treatment guidelines according to claim 6, characterized in that, The system integrates compliance indicators and credibility markers for various terms to calculate a compliance score for medical orders, and categorizes the verification results as compliant, partially compliant, or non-compliant. Specifically, this includes: Calculate the compliance score of medical orders ; Constructing a set of non-compliant terms: ;in, To eliminate all non-compliant terminology in the current medical orders The set that constitutes; when Output compliance and retain each of and ; when At that time, compliance of the output portion needs to be reviewed and listed in One by one: Minimum number of hops ;like Give the unique shortest path Node sequence index and , ;like Mark as unreachable, no path, no output. and ; when The output is non-compliant, and all terms are entered into the exception log.

8. The method for verifying the compliance of medical orders based on adaptive treatment guidelines according to claim 7, characterized in that, The method of dynamically adjusting the jump threshold based on the actual compliance rate and terminology distribution of a predefined batch of medical orders, combined with the effective diameter of the graph, specifically includes: Construct the first Collection of medical orders ;in, For batch indexing; For the first batch A doctor's order; The serial number of the medical order within this batch; For batch Total number of internal medical orders; For the first A medical order, assuming the number of its terms is . Terminology compliance indicators are Its medical order compliance score ; For the first The first medical order Compliance of the terminology; For the first The percentage of medical orders with compliant terminology; Set a fully compliant indicator function : , ; Calculate the first Batch of actual compliance rate ; Set the effective diameter of the diagram. : ； If the graph is completely discrete, let ; The term "completely discrete" means that no two pairs exist. Connected nodes; Initial run initialization: ;in, For the first run Setting value; Batch Update: ;in, For the first The threshold for batch execution time; It is a symbolic function; like ,but Increase by 1, not exceeding ; like ,but Subtract 1, but not less than 1; like , remains unchanged.

9. The method for verifying the compliance of medical orders based on adaptive treatment guidelines according to claim 8, characterized in that, The process of recording the unique witness path or unreachable state of all non-compliant or anomalous terms, and constructing and outputting the evolutionary version sequence of the anomalous term graph, specifically includes: Set the abnormal terminology set ; For each ,Record ; like Record the unique witness path ; like The recorded state is unreachable; Constructing anomaly graphs ;in, This is the set of node indices corresponding to the exception terms; The set of edges in the anomaly graph; In the Output version at the end of the batch , forming a sequence ;in, In the first The abnormal map version that is frozen after batching is completed.

Images