A hospital drug catalog matching system based on multi-strategy hierarchical progression

Abstract

The application provides a hospital drug directory matching system based on multi-strategy hierarchical progression, comprising: a drug knowledge base module for constructing a drug knowledge base; a to-be-matched module for constructing a hospital drug dictionary table, performing standardization processing on the drug knowledge base, and obtaining to-be-matched data; an intelligent decision-making module for performing model training and evaluation by using historical drug pairing data and a machine learning algorithm, and obtaining an intelligent decision-making model; a candidate pairing module for performing hierarchical progression matching on the to-be-matched data by using the intelligent decision-making model, and obtaining a candidate pairing result set; and a review and feedback module for manual review and model iteration optimization. The application can provide an intelligent matching tool with high accuracy, high coverage and high efficiency, so as to solve the semantic alignment problem of the drug knowledge base and the hospital drug directory in the reasonable drug use review system, and realize accurate and efficient association between the hospital drug directory and the drug knowledge base.

Description

Technical Field

[0001] This invention relates to the field of drug information management technology, and in particular to a hospital drug catalog matching system based on a multi-strategy hierarchical progression. Background Technology

[0002] To ensure the normal and stable operation of the rational drug use review system within the clinical decision support system, it is essential to accurately correlate and match the system's built-in drug knowledge base with the hospital's HIS drug catalog (which typically includes information such as drug name, dosage form, specifications, and manufacturer). This matching relationship is a core prerequisite for the rational drug use review system to correctly identify doctors' orders and trigger the corresponding review rules.

[0003] Currently, the commonly used matching methods in the field of drug matching mainly include: manual comparison matching, rule-based matching, semantic matching based on large language models, and drug similarity weight scoring matching. Each method has its limitations. While manual matching can guarantee accuracy and handle complex situations such as aliases, abbreviations, misspellings, and non-standard spellings, it is inefficient, lacks scalability, and struggles to cope with the continuous updating of large drug catalogs. Rule-based matching is suitable for structured information matching, but it lacks flexibility and cannot cover scenarios involving abbreviations and misspellings of hospital drugs. Semantic matching based on large language models, while possessing strong text understanding capabilities, has significant limitations in standardized drug matching scenarios. It lacks sensitivity to conditions such as drug dosage form, specifications, and manufacturer, and is prone to misjudgments. For example, it may fail to recognize key differences in dosage form and specifications between "Amoxicillin Capsules 0.25g" and "Amoxicillin Granules 0.125g," classifying them as highly similar. While drug similarity weighting and matching can avoid some misjudgments, it relies too heavily on fixed weight settings and has poor adaptability to dirty data such as drug aliases, abbreviations, and misspellings. It is difficult to cover complex drug description variations and cannot achieve dynamic optimization and adjustment. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide a hospital drug catalog matching system based on a multi-strategy hierarchical progression. This system provides a highly accurate, comprehensive, and efficient intelligent matching tool to solve the semantic alignment problem between the drug knowledge base and the hospital drug catalog in the rational drug use review system, thereby achieving a precise and efficient association between the hospital drug catalog and the drug knowledge base.

[0005] To achieve the above objectives, the present invention provides the following solution: a hospital drug catalog matching system based on a multi-strategy hierarchical progression, comprising: The drug knowledge base module is used to construct basic drug information, a drug naming system, a dosage form terminology standard system, a manufacturer name association table, and a unified specification representation to obtain the drug knowledge base. The matching module is used to construct a hospital drug dictionary table and perform standardization processing on the drug knowledge base to obtain matching data that retains the core semantic information of the drugs; The intelligent decision-making module is used to train and evaluate models using historical drug pairing data and machine learning algorithms to obtain intelligent decision-making models. The candidate matching module is used to perform hierarchical progressive matching on the data to be matched using the intelligent decision-making model to obtain a candidate matching result set. The review and feedback module is used to manually review the candidate pairing results, and then perform model iterative optimization based on the candidate pairing results and the manual review results. The drug knowledge base module, the matching module, the intelligent decision-making module, the candidate matching module, and the review and feedback module are interconnected.

[0006] Optionally, the drug knowledge base module includes: The standard drug unit is used to collect the standard name, dosage form, specification description, registration and manufacturing enterprise, drug approval number, medical insurance drug code, drug code base code and drug identification code of marketed drugs to obtain the original drug information. The basic drug information is then cleaned, normalized and aligned with entities and identifiers are constructed to obtain the basic drug information. The naming system unit is used to construct a standard drug name, trade name, generic name, chemical name, active ingredient name, and ontology name based on the basic drug information, thereby obtaining a multi-level drug naming ontology. For the multi-level drug naming ontology, a synonym evolution relationship network, a salt evolution relationship network, and a lipid evolution relationship network are constructed to obtain the drug naming system. The dosage form classification unit is used to construct a standard dosage form name, common aliases and variants, and dosage form semantic classification based on the basic information of the drug and the drug name system, thereby obtaining a standard system of dosage form terminology. The manufacturer system unit is used to standardize manufacturer names and related attributes, integrate business registration information, drug approval information and industry abbreviations, and obtain a table of manufacturer name association relationships between the standard name and aliases of the manufacturing enterprise. The specification parsing unit is used to construct a content unit conversion table and international units based on the basic information of the drug, and to automatically parse and structure the drug specification string to obtain a unified specification representation; The knowledge base unit is used to integrate the basic drug information, the drug name system, the dosage form terminology standard system, the manufacturer name association table, and the unified specification representation to obtain the drug knowledge base.

[0007] Optionally, the module to be matched includes: The hospital drug collection unit is used to periodically collect data from the hospital drug dictionary table, store the data in a temporary table, and then perform a full comparison between the temporary table and the hospital drug dictionary table to identify newly added, invalid, and changed drug information, obtain the comparison results, and update and maintain the hospital drug dictionary table based on the comparison results. The standardization processing unit is used to perform drug information processing, dosage form information processing, structured decomposition of drug specifications, cleaning of manufacturer information, and other information standardization processing based on the drug knowledge base, to obtain matching data that retains the core semantic information of the drug.

[0008] Optionally, the standardization processing unit includes: The drug information processing subunit is used to remove dosage form information, specification description and non-drug semantic tags before and after the drug name from the hospital drug names in the drug knowledge base using regular expressions and the drug name system. The dosage form information processing subunit is used to extract dosage form information contained in the names of hospital drugs based on the drug knowledge base, and store it as an independent field. It compares whether the extracted dosage form information is the same as the dosage form information of the corresponding drug in the dosage form terminology standard system. If not, the extracted dosage form information is used as subsequent matching information. The drug specification decomposition subunit is used to decompose unstructured specification text into content value, content unit, packaging quantity and packaging unit based on the drug knowledge base, and to perform standardized conversion according to the preset content unit conversion table and international units. The production information cleaning subunit is used to remove redundant characters, special symbols, and meaningless articles from the original manufacturer's name based on the drug knowledge base, thus completing the text cleaning. Other information processing subunits are used to standardize the approval number format, correct character errors and extra spaces in the approval number, and split the approval number into letter categories and numeric codes for storage.

[0009] Optionally, the intelligent decision-making module includes: The training data unit is used to collect historical drug pairing data, and to perform rule-weighted scoring matching and manual annotation dual verification on the historical drug pairing data to obtain a benchmark sample set. The similarity calculation unit is used to calculate the similarity of the core fields of the drug for the benchmark sample set, obtain the field matching score, and perform normalization processing on abnormal data according to the drug name knowledge system to obtain standardized training data. The model training unit is used to select a machine learning algorithm for binary classification tasks based on the standardized training data, using field matching scores and manual annotations as feature vectors, and to train and evaluate the model through cross-validation and hyperparameter tuning to obtain an intelligent decision model.

[0010] Optionally, the candidate pairing module includes: The rule matching unit is used to preset structured rules and match the data to be matched according to the structured rules to obtain a first pairing result; The candidate discrimination unit is used to determine whether the specification and manufacturer fields are consistent based on the first pairing result. If not, it calculates the semantic similarity between the specification feature value and the manufacturer. Based on the calculated semantic similarity, it uses the intelligent decision model to filter out the drug pairing results and obtain the second pairing result. The high similarity discrimination unit is used to select data with semantic similarity greater than a preset threshold of 70% based on the second matching result to obtain a candidate dataset, and then use the intelligent decision model to score and rank the candidate dataset to obtain a high confidence matching result. The pairing result unit is used to integrate the first pairing result, the second pairing result, and the high-confidence matching result to obtain a candidate pairing result set.

[0011] Optionally, the structured rules include a first rule and a second rule. The first rule is used to perform a complete match based on the unique code, and the second rule is used to perform a multi-field combination match if the first rule has no result. The second rule must satisfy the following conditions: the hospital drug and the standard library drug have the same standard name or ontology name semantics, the same dosage form semantics, the same specifications, and the same manufacturer.

[0012] Optionally, the review and feedback module includes: The review and return unit is used to manually review the matching results with medium confidence in the candidate matching result set, and then integrate the candidate matching results and the manually reviewed results into an optimal matching result set. The optimal matching result set is returned to the system front-end and back-end interfaces in a unified standard format to perform automatic matching and mapping between the hospital drug dictionary table and the standard drug dictionary table. The closed-loop update unit is used to periodically update the drug knowledge base, perform compliance verification on the candidate pairing results and manual review results, and perform model iterative optimization based on the candidate pairing results and manual review results.

[0013] This invention discloses the following technical effects by providing a hospital drug catalog matching system based on a multi-strategy hierarchical progression: 1. High accuracy: Through a triple guarantee mechanism of "precise rule-based backup, machine learning intelligent judgment, and manual review and error correction", the reliability of the final matching relationship is ensured, providing a solid data foundation for the rational drug use review system.

[0014] 2. High coverage: Through the enhanced knowledge system (alias and dosage form mapping) and the generalization ability of machine learning models, it can effectively handle non-standard expressions such as aliases, abbreviations, and misspellings in hospital drug catalogs, breaking through the coverage bottleneck of pure rule matching.

[0015] 3. High efficiency: The hierarchical and progressive mechanism optimizes the allocation of computing resources. Most simple matching is completed by fast rules, and only a small number of complex cases enter the more computationally intensive model inference and manual processing stages, thus achieving scalable processing capabilities.

[0016] 4. High controllability and interpretability: The methodology is transparent and does not rely entirely on a large, black-box end-to-end model. Each matching decision is based on explicit structured rules or interpretable features, and the manual review mechanism ensures final control over key results.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the system architecture provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the data collection process for a hospital drug dictionary table, as provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of data processing provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the automatic matching of hospital drugs and standard drugs in an embodiment of the present invention. Detailed Implementation

[0020] 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.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] like Figure 1 As shown, this invention provides a hospital drug catalog matching system based on a multi-strategy hierarchical progression, comprising: 1. A drug knowledge base module, used to construct basic drug information, a drug naming system, a dosage form terminology standard system, a manufacturer name association table, and a unified specification representation, resulting in a drug knowledge base; the drug knowledge base module includes: 1.1 Standard Drug Unit, used to collect the standard drug name, dosage form, specification description, registration and manufacturing enterprise, drug approval number, medical insurance drug code, drug code base code, and drug identification code of marketed drugs to obtain the original drug information. This basic drug information is then cleaned, normalized, and aligned with entities, and identifiers are constructed to obtain the basic drug information. Specifically: The data sources for the Standard Drug Product Basic Information Database include multiple authoritative data sources such as the National Medical Products Administration, the National Healthcare Security Administration, and the National Drug Use Monitoring Data. It integrates drug product information, including but not limited to drug standard names, dosage forms, specifications, registration and manufacturing enterprises, drug approval numbers, national medical insurance drug codes, national drug code base codes, and drug identification codes. The multi-source data is cleaned, normalized, and entity aligned to form a structured and versioned standard drug product information database. In conjunction with the drug name system sub-module, dosage form classification system sub-module, manufacturer system sub-module, and drug specification structured parsing sub-module, each record and each field (drug name, dosage form, specification, manufacturer) is assigned a unique identifier and structured data.

[0023] 1.2 The nomenclature system unit is used to construct a standard drug name, trade name, generic name, chemical name, active ingredient name, and ontological name based on the basic drug information, thereby obtaining a multi-level drug nomenclature ontology. Furthermore, based on the multi-level drug nomenclature ontology, a synonym evolution relationship network, a salt evolution relationship network, and a lipid evolution relationship network are constructed to obtain the drug nomenclature system. Specifically: Based on a comprehensive analysis of generic names, brand names, and alternative names of marketed drugs, this system supports precise matching at the name level. By integrating drug instructions, pharmacopoeias, clinical practice guidelines, hospital prescription collections, and publicly available company information, a multi-level drug naming ontology is constructed. This system includes, but is not limited to: Standard drug name: such as "Levofloxacin Hydrochloride Tablets"; Product names: such as "Zuofuke" and "Weilitai"; Generic name: such as "levofloxacin hydrochloride"; Chemical name and active ingredient name: such as "(S)-(-)-9-fluoro-2,3-dihydro-3-methyl-10-(4-methyl-1-piperazinyl)-7-oxo-7H-pyrido[1,2,3-de]-1,4-benzoxazine-6-carboxylic acid"; The name of the substance is such as "levofloxacin".

[0024] Further establish a network of synonym, salt, and lipid evolution relationships between names to support ontology-based name standardization and semantic mapping.

[0025] 1.3 Dosage Form Classification Unit, used to construct a standard dosage form name, common aliases and variants, and dosage form semantic classification based on the basic drug information and the drug naming system, thereby obtaining a standard system of dosage form terminology; specifically: Based on drug instructions and marketed drug information, a standardized dosage form classification system will be constructed to strengthen the matching constraints of dosage form dimensions; dosage form descriptions used within the system, industry, and in actual clinical practice will be collected to construct a standard system of dosage form terminology, including: Standard dosage form name: such as "tablet"; Common aliases and variations: such as "tablets", "plain tablets", "film-coated tablets", "enteric-coated tablets"; Dosage form semantic classification: such as classifying it into the hierarchical structure of "conventional oral tablets → oral solid dosage forms".

[0026] 1.4 Manufacturer System Unit: This unit standardizes manufacturer names and related attributes, integrates business registration information, drug approval information, and industry abbreviations, and generates a table showing the association between the standard name and aliases of manufacturing enterprises; specifically: This system integrates drug instructions and information on manufacturers of marketed drugs, standardizing manufacturer names and related attributes to ensure accurate matching at the manufacturer level. It also integrates business registration information, drug approval information, and commonly used industry abbreviations to establish a table relating standard manufacturer names to their aliases. For example: Standard manufacturer name: "Jiangsu Hengrui Medicine Co., Ltd." Common aliases: "Hengrui Medicine", "Jiangsu Hengrui", "Hengrui Pharmaceutical"; This mapping relationship supports fuzzy matching and normalized recognition of company names.

[0027] 1.5 Specification parsing unit, used to construct a content unit conversion table and international units based on the basic drug information, and to automatically parse and structure the drug specification string to obtain a unified specification representation; specifically: The system automates the parsing and structured storage of drug specification strings, establishing a unified specification representation model. For example, "10mg×10 tablets / box" is parsed as: single dose content: numerical value "10" + unit "mg"; packaging quantity: numerical value "10" + unit "tablet"; packaging unit: "box". Furthermore, a content unit conversion table (e.g., g↔mg↔µg, IU↔mg, etc.) is constructed to achieve equivalent numerical matching across unit specifications.

[0028] 1.6 Knowledge base unit, used to integrate the basic drug information, the drug name system, the dosage form terminology standard system, the manufacturer name association table, and the unified specification representation to obtain a drug knowledge base.

[0029] 2. Modules to be matched, such as Figure 2 , Figure 3 As shown, a hospital drug dictionary table is constructed, and the drug knowledge base is standardized to obtain matching data that retains the core semantic information of the drugs; the matching module includes: 2.1 The hospital drug acquisition unit is used to periodically collect data from the hospital drug dictionary table, store the data in a temporary table, and then perform a full comparison between the temporary table and the hospital drug dictionary table to identify newly added, expired, and changed drug information. Based on the comparison results, the hospital drug dictionary table is updated and maintained. Specifically: A timed mechanism is adopted to automatically acquire hospital drug dictionary data every night. The collected drug dictionary data is first stored in a temporary table, and then the data in the temporary table is compared with the hospital drug dictionary already stored in the tool. Through intelligent analysis, newly added, invalid, and changed drug information is identified, and the target dictionary is automatically updated and maintained based on the comparison results. This ensures that the hospital drug dictionary data is always consistent with the source, providing accurate and real-time basic data support for subsequent drug inventory management, prescription issuance, and rational drug use monitoring.

[0030] 2.2 A standardization processing unit, used to perform drug information processing, dosage form information processing, structured decomposition of drug specifications, cleaning of manufacturer information, and other information standardization processes based on the drug knowledge base, to obtain matching data that retains the core semantic information of the drug. The standardization processing unit includes: 2.2.1 The drug information processing subunit is used to remove dosage form information, specification descriptions, and non-drug semantic tags from the hospital drug names in the drug knowledge base using regularization expressions and the drug name system; specifically: Using hospital drug names as the core processing content, regular expressions and the drug name system are used to accurately strip and remove dosage form information (such as tablets, capsules, injections, ointments) and specification descriptions (such as 0.25g, 10mg, 5ml) attached to the name. Common non-drug semantic tags are filtered out, including but not limited to: source tags (such as self-provided), temporary tags, and drug property tags (such as essential drugs, Class A medical insurance drugs, Class I drugs, etc., management attribute fields).

[0031] 2.2.2 Dosage Form Information Processing Subunit: Based on the drug knowledge base, this subunit extracts dosage form information contained in the names of hospital drugs and stores it as an independent field. It then compares the extracted dosage form information with the dosage form information of the corresponding drug in the dosage form terminology standard system. If they are different, the extracted dosage form information is used as subsequent matching information. Specifically: In addition to the dosage form information already present in the drug dictionary, the dosage form information contained in the hospital drug names is extracted and retained separately and stored as an independent field. Based on the dosage form classification subsystem, it is determined whether two dosage forms have the same semantics (e.g., both belong to conventional oral tablets). If they are essentially the same, both can be used as subsequent matching fields; if the semantics are completely different, only the dosage form in the name is considered for subsequent matching.

[0032] 2.2.3 The drug specification decomposition subunit is used to decompose unstructured specification text into content values, content units, packaging quantities, and packaging units based on the drug knowledge base, and to perform standardized conversion according to a preset content unit conversion table and international units; specifically: Unstructured specification text is broken down into independent fields such as content value, content unit, packaging quantity and packaging unit, and then standardized and converted according to a pre-built unit conversion rule library.

[0033] 2.2.4 The production information cleaning subunit is used to remove redundant characters, special symbols, and meaningless articles from the original manufacturer names based on the drug knowledge base, thus completing text cleaning; specifically: Text cleaning is performed on the original manufacturer's name. The cleaning includes, but is not limited to, the following: removing redundant characters, special symbols, and meaningless articles (such as various variations of "Limited Liability Company", parenthetical notes, etc.) from the name.

[0034] 2.2.5 Other information processing subunits are used to standardize the approval document number format, correct character errors and extra spaces in the document number, and split the approval document number into letter categories and numeric codes for storage. Specifically: Standardized processing of approval document numbers and other key information: For example, by unifying the format of approval document numbers, the system will automatically identify and correct errors in full-width / half-width characters and extra spaces in the document number, and store it in three parts according to national standards: letter category (H, Z, S, etc.) and numeric code.

[0035] After the above processing, a clean semantic information is obtained, retaining only the core semantic information of the drug (e.g., standardizing amoxicillin capsules 0.5g*20 capsules / box (essential drug) to amoxicillin, specification 0.5g, 20 capsules, dosage form capsule, etc.). These core words will serve as the sole semantic anchor for all subsequent structured rules, text similarity calculations, and matching.

[0036] 3. An intelligent decision-making module, used to train and evaluate models using historical drug pairing data and machine learning algorithms to obtain an intelligent decision-making model; the intelligent decision-making module includes: 3.1 Training data unit, used to collect historical drug pairing data, and to perform rule-weighted scoring matching and manual annotation dual verification on the historical drug pairing data to obtain a benchmark sample set; specifically: The training data source is constructed based on the historical drug pairing results accumulated over a long period of time by the rational drug use system. This type of historical pairing data has undergone rule-weighted scoring matching and dual verification by humans, possessing high annotation quality and can be directly used as the benchmark sample set for model training.

[0037] 3.2 A similarity calculation unit is used to calculate the similarity of core drug fields for the benchmark sample set, obtain field matching scores, and perform normalization processing on abnormal data according to the drug name knowledge system to obtain standardized training data; specifically: After obtaining correct and incorrect data from historical data, abnormal samples are first standardized based on the drug name knowledge system to achieve accurate alignment between the hospital drug field and the standard matching field.

[0038] The specific scoring rules are as follows: A score of 1 is recorded when the drug name, the drug's original name, the dosage form after removing the drug name, the original dosage form field, and the manufacturer are semantically consistent; if the names are inconsistent, their semantic similarity is calculated and included as a continuous score. A score of 1 is recorded when the converted structured fields such as single-dose specification, concentration, packaging unit, approval number, medical insurance code, and YPID are consistent; otherwise, a score of 0 is recorded and the data is not included in subsequent weight calculations.

[0039] After summarizing the matching results of each field, the weight thresholds are iteratively optimized by combining the manually labeled correct and incorrect sample labels, and finally a standardized dataset that can be directly used for model training is formed.

[0040] 3.3 The model training unit is used to select a machine learning algorithm for binary classification tasks based on the standardized training data, using field matching scores and manual annotations as feature vectors. The model is then trained and evaluated through cross-validation and hyperparameter tuning to obtain an intelligent decision-making model. Specifically: Based on the standardized training dataset of the output, the matching scores of each field and the manually labeled correct and incorrect tags are used to construct a feature vector, which serves as the basic input for model training. During feature construction, for semantically similar fields, their calculated similarity scores are incorporated as continuous features; for structured fields, the matching results are transformed into binary features for modeling. Building upon this, the XGBoost binary classification algorithm is selected, and the model is trained and evaluated through cross-validation and hyperparameter tuning, enabling the model to fully learn the weight contributions and discrimination rules of different field combinations in drug matching.

[0041] The trained model has the ability to autonomously judge the rationality of drug pairings. It can automatically score and classify newly added hospital drugs and standard drug candidate pairings, accurately output high-confidence rational pairing results, and provide efficient support for subsequent candidate set confirmation and business implementation.

[0042] 4. Candidate pairing module, such as Figure 4 As shown, the system utilizes the intelligent decision-making model to perform hierarchical progressive matching on the data to be matched, obtaining a candidate pairing result set; the candidate pairing module includes: 4.1 Rule matching unit, used to preset structured rules, and match the data to be matched according to the structured rules to obtain a first pairing result; the structured rules include a first rule and a second rule, the first rule is used for complete consistency matching based on unique codes, and the second rule is used for multi-field combination matching when the first rule has no result; wherein, the second rule must satisfy the following requirements: the hospital drug and the standard library drug have the same standard name or ontology name semantics, the same dosage form semantics, the same specifications, and the same manufacturer. Specifically: After data processing is completed, structured rules are used to achieve completely accurate drug matching, ensuring the accuracy of the matching results.

[0043] Based on the priority of the preset structured rules, multi-level exact matching rules are executed sequentially: a) Prioritize matching based on unique codes such as drug standard codes, approval numbers, and medical insurance codes. Then, perform multi-field combination matching, with specific requirements including: the hospital drug and the standard library drug must have identical semantics in their "standard name or ontology name," identical semantics in their dosage form (or the dosage form portion of the name), identical specifications / key specifications, and identical manufacturers / aliases.

[0044] b) If no match is found for the unique code, a multi-field combination matching rule will be used. The combination match must meet the following conditions simultaneously: the "standard name or ontology name" of the hospital drug and the standard library drug are semantically identical; the dosage form (or the dosage form part in the name) is semantically identical; the specifications or key specifications are the same; and the manufacturer or alias is the same.

[0045] Once the above rules are successfully matched, it is considered that a completely accurate drug matching result has been obtained, and the execution of subsequent rules will be terminated.

[0046] 4.2 Candidate discrimination unit, used to determine whether the specification and manufacturer fields are consistent based on the first pairing result; if not, it calculates the semantic similarity between the specification feature value and the manufacturer, and uses the intelligent decision-making model to filter out the drug pairing results based on the calculated semantic similarity, thus obtaining the second pairing result; specifically: By using machine learning models, candidate datasets for precise drug / dosage form matching are identified, and accurate matching results are precisely selected.

[0047] When a hospital drug and a drug in the standard library have the same semantic "standard name or ontology name" and the same semantic "dosage form" (or the dosage form part in the name), but the specifications and manufacturer fields are not completely consistent, the system first calculates the semantic similarity between the feature value of the specifications and the manufacturer, and then uses a machine learning model to accurately select the correct drug pairing results.

[0048] 4.3 A high similarity discrimination unit is used to select data with a semantic similarity greater than a preset threshold of 70% based on the second matching result, to obtain a candidate dataset, and then use the intelligent decision-making model to score and rank the candidate dataset to obtain a high-confidence matching result; specifically: Machine learning models are used to identify candidate datasets where drug similarity reaches a set threshold, thereby expanding the matching coverage.

[0049] When hospital drugs and standard library drugs fail to meet the exact matching module's requirement of complete consistency in the key field of "standard name or ontology name", but the semantic similarity of the drug name is higher than the preset threshold of 70%, a candidate dataset is obtained. Then, based on multi-field feature extraction and semantic similarity calculation, the candidate set is scored and ranked by a machine learning model, and matching results with sufficient confidence are selected as an effective supplement to exact matching.

[0050] The medium-confidence matching results output by the model are manually reviewed and confirmed to ensure the accuracy and completeness of the matching results.

[0051] 4.4 Pairing result unit, used to integrate the first pairing result, the second pairing result and the high confidence matching result to obtain a candidate pairing result set.

[0052] For example, the drug name is binding, and all the above judgments will take priority based on the drug name: 1) Hospital drugs: Standardization of drug A.

[0053] 2) After standardization, drug A is directly matched to the standard drug A through the relevant fields of the structured rules, and at this point, it does not proceed to the subsequent matching stage. Result-based rules represent rigid rules that are 100% correct.

[0054] 3) If drug A does not match any record in the standard drug database after the rule-based process, then for drug A, if there are multiple records in the standard database using identical drug name and dosage form, semantic analysis of the manufacturer and specifications will be performed, and the model will be used to determine the most similar record. The result selected by the model will be directly used as the matching result and will not participate in subsequent processes.

[0055] 4) If drug A does not find a matching result through the above two steps, then find drug records with identical or extremely high similarity in name according to the standard library, and then use the model to determine the optimal matching result based on the detailed feature values ​​of the drug. This optimal matching result is directly used as the matching result and does not participate in the subsequent process.

[0056] 5) And so on... 5. A review and feedback module, used to manually review the candidate pairing results, and then perform iterative model optimization based on the candidate pairing results and the manual review results; the review and feedback module includes: 5.1 The review and return unit is used to manually review the matching results with medium confidence in the candidate pairing result set, then integrate the candidate pairing results and the manually reviewed results into an optimal pairing result set, and return the optimal pairing result set to the system front-end and back-end interfaces in a unified standard format for automated matching and mapping between the hospital drug dictionary table and the standard drug dictionary table; specifically: The medium-confidence matching results output by the model are manually reviewed to ensure the accuracy and completeness of the matching results.

[0057] The review scope mainly includes two types of data: first, the matching results whose confidence level does not meet the standard; and second, the candidate matching results obtained by the machine learning discrimination model with the semantic similarity of drug names in the range of 50% to 70%. Valid matches are ultimately confirmed through manual review, erroneous associations are eliminated, and the overall matching quality is improved.

[0058] Results returned: The final confirmation results after integrating precise matching, high similarity matching, and manual review are formed into the optimal matching result set, and returned to the front-end and back-end interfaces of the drug matching tool in a unified standard format to complete the automatic matching and mapping between the hospital drug dictionary table and the standard drug dictionary table.

[0059] 5.2 A closed-loop update unit is used to periodically update the drug knowledge base, perform compliance checks on the candidate pairing results and manual review results, and perform iterative optimization of the model based on the candidate pairing results and manual review results. Specifically: To address the need for long-term stable system operation and improved matching accuracy, we will refine the closed-loop management of data and models: Data updates are conducted regularly to address situations such as new drug additions, withdrawals from the market, changes in specifications, and manufacturer name changes. This ensures dynamic data updates, guarantees the timeliness of matching benchmarks, and provides reliable data support for subsequent matching and model training. Matching result verification and feedback: The system performs compliance verification on the automatic matching and manual review results, and feeds back the matching error cases and manual review results to the machine learning model to continuously support model iteration and optimization, and gradually improve the accuracy of the system's automatic matching.

[0060] Therefore, this invention provides a hospital drug catalog matching system based on a multi-strategy hierarchical progression, which can provide a highly accurate, comprehensive, and efficient intelligent matching tool to solve the semantic alignment problem between the drug knowledge base and the hospital drug catalog in the rational drug use review system, and achieve accurate and efficient association between the hospital drug catalog and the drug knowledge base.

[0061] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0062] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A hospital drug catalog matching system based on a multi-strategy hierarchical progression, characterized in that, include: The drug knowledge base module is used to construct basic drug information, a drug naming system, a dosage form terminology standard system, a manufacturer name association table, and a unified specification representation to obtain the drug knowledge base. The matching module is used to construct a hospital drug dictionary table and perform standardization processing on the drug knowledge base to obtain matching data that retains the core semantic information of the drugs; The intelligent decision-making module is used to train and evaluate models using historical drug pairing data and machine learning algorithms to obtain intelligent decision-making models. The candidate matching module is used to perform hierarchical progressive matching on the data to be matched using the intelligent decision-making model to obtain a candidate matching result set. The review and feedback module is used to manually review the candidate pairing results, and then perform model iterative optimization based on the candidate pairing results and the manual review results. The drug knowledge base module, the matching module, the intelligent decision-making module, the candidate matching module, and the review and feedback module are interconnected.

2. The hospital drug catalog matching system based on a multi-strategy hierarchical progression according to claim 1, characterized in that, The drug knowledge base module includes: The standard drug unit is used to collect the standard name, dosage form, specification description, registration and manufacturing enterprise, drug approval number, medical insurance drug code, drug code base code and drug identification code of marketed drugs to obtain the original drug information. The basic drug information is then cleaned, normalized and aligned with entities and identifiers are constructed to obtain the basic drug information. The naming system unit is used to construct a standard drug name, trade name, generic name, chemical name, active ingredient name, and ontology name based on the basic drug information, thereby obtaining a multi-level drug naming ontology. For the multi-level drug naming ontology, a synonym evolution relationship network, a salt evolution relationship network, and a lipid evolution relationship network are constructed to obtain the drug naming system. The dosage form classification unit is used to construct a standard dosage form name, common aliases and variants, and dosage form semantic classification based on the basic information of the drug and the drug name system, thereby obtaining a standard system of dosage form terminology. The manufacturer system unit is used to standardize manufacturer names and related attributes, integrate business registration information, drug approval information and industry abbreviations, and obtain a table of manufacturer name association relationships between the standard name and aliases of the manufacturing enterprise. The specification parsing unit is used to construct a content unit conversion table and international units based on the basic information of the drug, and to automatically parse and structure the drug specification string to obtain a unified specification representation; The knowledge base unit is used to integrate the basic drug information, the drug name system, the dosage form terminology standard system, the manufacturer name association table, and the unified specification representation to obtain the drug knowledge base.

3. A hospital drug catalog matching system based on a multi-strategy hierarchical progression according to claim 2, characterized in that, The modules to be matched include: The hospital drug collection unit is used to periodically collect data from the hospital drug dictionary table, store the data in a temporary table, and then perform a full comparison between the temporary table and the hospital drug dictionary table to identify newly added, invalid, and changed drug information, obtain the comparison results, and update and maintain the hospital drug dictionary table based on the comparison results. The standardization processing unit is used to perform drug information processing, dosage form information processing, structured decomposition of drug specifications, cleaning of manufacturer information, and other information standardization processing based on the drug knowledge base, to obtain matching data that retains the core semantic information of the drug.

4. A hospital drug catalog matching system based on a multi-strategy hierarchical progression according to claim 3, characterized in that, The standardization processing unit includes: The drug information processing subunit is used to remove dosage form information, specification description and non-drug semantic tags before and after the drug name from the hospital drug names in the drug knowledge base using regular expressions and the drug name system. The dosage form information processing subunit is used to extract dosage form information contained in the names of hospital drugs based on the drug knowledge base, and store it as an independent field. It compares whether the extracted dosage form information is the same as the dosage form information of the corresponding drug in the dosage form terminology standard system. If not, the extracted dosage form information is used as subsequent matching information. The drug specification decomposition subunit is used to decompose unstructured specification text into content value, content unit, packaging quantity and packaging unit based on the drug knowledge base, and to perform standardized conversion according to the preset content unit conversion table and international units. The production information cleaning subunit is used to remove redundant characters, special symbols, and meaningless articles from the original manufacturer's name based on the drug knowledge base, thus completing the text cleaning. Other information processing subunits are used to standardize the approval number format, correct character errors and extra spaces in the approval number, and split the approval number into letter categories and numeric codes for storage.

5. A hospital drug catalog matching system based on a multi-strategy hierarchical progression according to claim 4, characterized in that, The intelligent decision-making module includes: The training data unit is used to collect historical drug pairing data, and to perform rule-weighted scoring matching and manual annotation dual verification on the historical drug pairing data to obtain a benchmark sample set. The similarity calculation unit is used to calculate the similarity of the core fields of the drug for the benchmark sample set, obtain the field matching score, and perform normalization processing on abnormal data according to the drug name knowledge system to obtain standardized training data. The model training unit is used to select a machine learning algorithm for binary classification tasks based on the standardized training data, using field matching scores and manual annotations as feature vectors, and to train and evaluate the model through cross-validation and hyperparameter tuning to obtain an intelligent decision model.

6. A hospital drug catalog matching system based on a multi-strategy hierarchical progression according to claim 5, characterized in that, The candidate pairing module includes: The rule matching unit is used to preset structured rules and match the data to be matched according to the structured rules to obtain a first pairing result; The candidate discrimination unit is used to determine whether the specification and manufacturer fields are consistent based on the first pairing result. If not, it calculates the semantic similarity between the specification feature value and the manufacturer. Based on the calculated semantic similarity, it uses the intelligent decision model to filter out the drug pairing results and obtain the second pairing result. The high similarity discrimination unit is used to select data with semantic similarity greater than a preset threshold of 70% based on the second matching result to obtain a candidate dataset, and then use the intelligent decision model to score and rank the candidate dataset to obtain a high confidence matching result. The pairing result unit is used to integrate the first pairing result, the second pairing result, and the high-confidence matching result to obtain a candidate pairing result set.

7. A hospital drug catalog matching system based on a multi-strategy hierarchical progression according to claim 6, characterized in that, The structured rules include a first rule and a second rule. The first rule is used to perform a complete match based on the unique code, and the second rule is used to perform a multi-field combination match if the first rule has no result. The second rule must satisfy the following conditions: the hospital drug and the standard library drug have the same standard name or ontology name semantics, the same dosage form semantics, the same specifications, and the same manufacturer.

8. A hospital drug catalog matching system based on a multi-strategy hierarchical progression according to claim 7, characterized in that, The review and feedback module includes: The review and return unit is used to manually review the matching results with medium confidence in the candidate matching result set, and then integrate the candidate matching results and the manually reviewed results into an optimal matching result set. The optimal matching result set is returned to the system front-end and back-end interfaces in a unified standard format to perform automatic matching and mapping between the hospital drug dictionary table and the standard drug dictionary table. The closed-loop update unit is used to periodically update the drug knowledge base, perform compliance verification on the candidate pairing results and manual review results, and perform model iterative optimization based on the candidate pairing results and manual review results.

Images