Medication data supervision and evaluation system based on artificial intelligence

Abstract

The invention discloses a medication data supervision and evaluation system based on artificial intelligence, and belongs to the technical field of medication data supervision. The method and the device are used for solving the technical problems of poor active supervision reliability and poor active optimization iteration effect of drug administration data supervision evaluation in an existing scheme. Credible label values in the standardized matrix can provide clear input for subsequent risk assessment; through weighted integration of multi-dimensional factors of dosage, solvent and pathology, quantitative processing of medication risks is realized, a matching degree algorithm based on historical optimal prescriptions and real-time guide update enables a system to dynamically adapt to clinical changes, and through construction of a three-dimensional evaluation tensor, complex data can be converted into visual scores, so that the accuracy of evaluation is improved. Unnecessary intervention of low-risk prescriptions is reduced; the credible label value, the medication risk index and the rule adaptation degree are subjected to conjoint analysis and judgment, and a hierarchical optimization mechanism is dynamically triggered according to a judgment result, so that active optimization management on implementation of an existing supervision and evaluation scheme is realized.

Description

Technical Field

[0001] This invention relates to the field of medication data monitoring technology, and more specifically to an artificial intelligence-based medication data monitoring and evaluation system. Background Technology

[0002] The scope of medication data is very broad, ranging from information on drug instructions to medication records in hospital information systems.

[0003] The existing drug use data monitoring and evaluation system lacks sufficient intelligence. For example, it cannot verify the rationality of excessive dosage or special solvents, and has a high false alarm rate, such as directly judging drugs used with caution as unreasonable. Furthermore, it cannot proactively optimize and iterate intelligent tools in multiple dimensions, resulting in poor reliability of proactive monitoring and poor proactive optimization and iteration effects in drug prescription data monitoring and evaluation. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence-based drug use data monitoring and evaluation system to solve the technical problems of poor reliability of proactive monitoring and poor proactive optimization and iteration effects in existing solutions for drug prescription data monitoring and evaluation.

[0005] The objective of this invention can be achieved through the following technical solutions: An AI-based drug use data monitoring and evaluation system includes: Multi-source data acquisition and processing module: Acquires and preprocesses multi-source data, constructs a dual-source verification data sequence, and uses the dual-source verification data sequence to construct a standardized data matrix with trusted label values. The dual-source verification data sequence includes the original medication data from the medical institution information system and the third-party verification data stored on the blockchain. Three-dimensional evaluation tensor processing module: Calculates dynamic evaluation indicators based on standardized data matrices. Dynamic evaluation indicators include medication risk index and rule fit. Use dynamic evaluation indicators to construct a three-dimensional evaluation tensor. Joint Judgment Dynamic Grading Optimization Module: Based on the three-dimensional evaluation tensor, the module performs joint analysis and judgment on the credible label value, medication risk index, and rule fit. If the result is a judgment failure, the grading optimization mechanism is triggered, and the implementation of the existing regulatory assessment scheme is optimized and managed. Multi-dimensional implementation effect evaluation and management module: Conducts multi-dimensional regulatory analysis on the implementation effect of existing regulatory evaluation schemes, proactively evaluates the implementation effect of existing regulatory evaluation schemes in corresponding local and overall dimensions, and implements targeted closed-loop iterative management; closed-loop iterative management includes local closed-loop iterative management and overall closed-loop iterative management.

[0006] Preferably, when collecting original medication data from the medical institution's information system and third-party verified data stored on the blockchain, the following information is collected: drug name, single dose, solvent type, route of administration, and prescription timestamp; patient pathological characteristics are collected through the electronic medical record system; and full lifecycle data of the drug, including drug batch number, solvent compliance certificate, and manufacturer qualifications, are obtained through the alliance link interface.

[0007] Preferably, when constructing a standardized data matrix with trusted label values ​​using dual-source verification data sequences, the row index of the data matrix is ​​designed to be a single prescription, and the data is arranged in ascending order by prescription timestamp; The column index is a feature variable, including basic features, verification features, and trusted label values.

[0008] Preferably, when calculating the medication risk index based on a standardized data matrix, the formula is used. The calculation results are as follows: RI is the medication risk index; a, b, and c are all weighting coefficients, a+b+c=1; D is the dose deviation; S is the solvent risk level value; and P is the patient pathological fit.

[0009] Preferably, several target vectors are obtained from a standardized data matrix, and the feature vector distance is calculated using weighted cosine similarity, and then calculated using the formula... Calculate the rule fit FA; where DX is the feature vector distance.

[0010] Preferably, when integrating the credible label value, the drug risk index RI, and the rule fit FA to construct a three-dimensional evaluation tensor, the X-axis corresponds to the data credibility, the Y-axis corresponds to the drug risk index, and the Z-axis corresponds to the rule fit. Furthermore, the clinical intervention necessity score P is obtained through calculation, and the clinical intervention necessity score is set as the element value. The relevant expression is as follows: Where XB is the trusted label value.

[0011] Preferably, the trusted label value, medication risk index, and rule fit are analyzed using a joint model, and the joint analysis value LF is output; wherein, the expression of the joint model is: ;in, These are the trusted label threshold, the medication risk threshold, and the rule adaptation threshold, respectively. If the joint analysis value is 0, the result is considered passed; If the joint analysis value is 1, the result is deemed unsuccessful, and a hierarchical optimization mechanism is triggered.

[0012] Preferably, when conducting multi-dimensional regulatory analysis, the total number of first-level regulatory actions N1 and the total number of second-level regulatory actions N2 that trigger the hierarchical optimization mechanism are statistically analyzed using a first-level iterative identification model, and the first-level identification value DS1 is output. Similarly, the total number of second-level regulatory actions is analyzed using a second-level iterative identification model, and the second-level identification value DS2 is output.

[0013] Preferably, the first and second iteration identification values ​​are jointly analyzed. If both values ​​are 0, the existing regulatory assessment scheme is maintained. If a value 'a' exists, then implement targeted overall closed-loop iterative management. If a value b exists, then targeted local closed-loop iterative management is implemented.

[0014] Preferably, the expression of the first iterative recognition model is: Where K is the optimization trigger threshold; a is a natural number greater than 0; The expression for the second-iteration recognition model is: Where N0 is the optimization implementation threshold; b is a natural number greater than 0.

[0015] Compared to existing solutions, the beneficial effects achieved by this invention are: This invention solves the problem of easy tampering with single-source data in traditional systems by cross-verifying medical institution systems with blockchain data, effectively improving data reliability. The trusted label values ​​in the standardized matrix can provide clear input for subsequent risk assessment, avoiding interference from low-quality data in the assessment results, and effectively improving the system's autonomous optimization efficiency.

[0016] This invention quantifies medication risk by weighted integration of multiple dimensions including dosage, solvent, and pathology. This effectively improves the accuracy of identifying unconventional dosages and contraindicated solvents, and significantly reduces the false positive rate compared to traditional single-threshold judgments. A matching algorithm based on historical best prescriptions and real-time guideline updates allows the system to dynamically adapt to clinical changes, effectively improving the speed of adaptation to new diseases. By constructing a three-dimensional assessment tensor, complex data can be transformed into intuitive scores, reducing unnecessary interventions in low-risk prescriptions and achieving risk visualization and tiered intervention.

[0017] This invention achieves proactive optimization and management of the implementation of existing regulatory assessment schemes by jointly analyzing and judging the trust label value, medication risk index, and rule adaptability, and dynamically triggering a hierarchical optimization mechanism based on the judgment results, thereby improving the proactive optimization effect of the implementation of existing regulatory assessment schemes.

[0018] This invention conducts multi-dimensional regulatory analysis on the implementation effects of existing regulatory assessment schemes, proactively evaluates the implementation effects of existing regulatory assessment schemes in corresponding local and overall dimensions, and implements targeted closed-loop iterative management. This enables the expansion and extension of existing regulatory assessment schemes and regulatory optimization, ensuring the reliability of the implementation of existing regulatory assessment schemes and the diversity of regulation from different dimensions. Attached Figure Description

[0019] The invention will now be further described with reference to the accompanying drawings.

[0020] Figure 1 This is a flowchart illustrating the operation of the AI-based drug data monitoring and evaluation system of this invention. Detailed Implementation

[0021] 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.

[0022] like Figure 1 As shown, this invention is an artificial intelligence-based drug use data monitoring and evaluation system, comprising: Multi-source data acquisition and processing module: Acquires and preprocesses multi-source data, constructs a dual-source verification data sequence, and uses the dual-source verification data sequence to construct a standardized data matrix with trusted label values. The dual-source verification data sequence includes original medication data from the medical institution's information system and third-party verification data stored on the blockchain. Specific steps include: When collecting raw medication data from medical institution information systems and third-party verified data stored on the blockchain, data is collected in real time through standardized interfaces of Hospital Information System (HIS), Laboratory Information System (LIS), and Electronic Medical Record System (EMR). Specifically, this includes: Collect drug name, single dose, solvent type, route of administration, and prescription timestamp through the hospital information system; The drug name includes the generic name and brand name, such as sodium penicillin for injection (North China Pharmaceutical); the unit for a single dose is mg / vial, accurate to one decimal place; the solvent type is such as 0.9% sodium chloride injection; the route of administration is such as intravenous infusion, oral administration, or intramuscular injection; the prescription timestamp format is YYYY-MM-DD-HH-MM-SS, specifically corresponding to year-month-day-hour-minute-second; The patient's pathological characteristics are collected through the electronic medical record system (EMR), including but not limited to patient identification, age, liver and kidney function indicators, and allergy history; liver and kidney function indicators include but are not limited to alanine aminotransferase (ALT), aspartate aminotransferase (AST), serum creatinine (Cr), and blood urea nitrogen (BUN). Among them, the reference ranges for liver and kidney function indicators are marked according to the "Normal Reference Values ​​for Clinical Laboratory Results". For example, the normal range for alanine aminotransferase (ALT) is 0-40 U / L. In addition, the entire lifecycle data of a drug can be obtained through the alliance link interface, including the drug batch number, solvent compliance certificate, and manufacturer qualifications. The entire lifecycle data of a drug is automatically synchronized to the local system through smart contracts. Among them, the batch number corresponds to a 20-digit code; the solvent compliance certificate corresponds to the drug regulatory authority's filing number; and the manufacturer's qualification corresponds to the GMP certification status. In addition, blockchain data comes with its own timestamps and hash values, and only the hash digest is retained when stored locally to avoid leakage of the original data; When preprocessing the collected multi-source data, including but not limited to timestamp alignment, hash value comparison and tamper detection, and outlier identification and processing, outlier identification and processing can be implemented based on the improved Z-score algorithm. These are all existing conventional technical solutions, and the specific implementation steps are not elaborated here. The preprocessed multi-source data is sorted and combined to obtain a dual-source verification data sequence. When constructing a standardized data matrix with trusted label values ​​using dual-source verified data sequences, the row index of the data matrix is ​​designed to be a single prescription, and the data is sorted in ascending order by prescription timestamp; The column index is a feature variable, including basic features, verification features, and trusted label values; Among them, the basic features include drug name code, standardized dose value, and solvent type code; Verification features include timestamp alignment results, hash comparison results, and outlier flags. The trust label value, ranging from 0 to 100, is calculated by comprehensively considering data integrity, tamper-proof nature, absence of outliers, and matching with blockchain data. The higher the trust label value, the higher the trustworthiness. When calculating the trust label value, for data integrity, 30 points are awarded for no missing fields, and 10 points are deducted for each missing key field, such as dosage and solvent. For tamper-proof hash comparisons, a match results in 20 points, while a mismatch results in 0 points. For cases without outliers, 30 points are awarded when flag=0 and 0 points are awarded when flag=1. For blockchain data matching, 20 points are awarded for timestamp alignment and batch compliance, 10 points are awarded for timestamp alignment but batch non-compliance, and 0 points are awarded for timestamp misalignment and batch non-compliance. Example: For a prescription, data completeness is 30 points, tamper-proof is 20 points, no outliers are 30 points, and blockchain matching is 20 points. The trustworthy label value = 30 + 20 + 30 + 20 = 100 points; data with a score of 80 or above is considered highly reliable. When standardizing the constructed data matrix, continuous variables are standardized using the Min-Max method, such as dose and liver and kidney function indicators. And perform one-heat encoding on categorical variables to obtain a standardized data matrix; where categorical variables include, for example, solvent type; Example: 0.9% sodium chloride is coded as [1,0,0], and 5% glucose is coded as [0,1,0]. The standardization process is carried out using existing conventional technical means, and the specific implementation steps will not be elaborated here.

[0023] In this embodiment of the invention, cross-validation between the medical institution system and blockchain data can solve the problem of easy tampering of single-source data in traditional systems, effectively improving data reliability; the trusted label values ​​in the standardized matrix can provide clear input for subsequent risk assessment, avoid low-quality data interfering with the assessment results, and effectively improve the system's autonomous optimization efficiency.

[0024] The 3D evaluation tensor processing module calculates dynamic evaluation indicators based on a standardized data matrix. These indicators include a medication risk index and rule fit. A 3D evaluation tensor is then constructed using these dynamic indicators. Specific steps include: When calculating the medication risk index based on a standardized data matrix, the formula is used. The calculation results are as follows: RI is the medication risk index; a, b, and c are all weighting coefficients, a+b+c=1, with default values ​​of a=0.4, b=0.3, and c=0.3, which can be customized according to actual clinical needs; D is the dose deviation; S is the solvent risk level value; and P is the patient pathological fit. Among them, the dose deviation degree D is the proportion of the deviation between the actual dose and the disease-specific routine dosage threshold table in the dynamic rule base, which reflects the rationality of the dose; If the actual dose is less than or equal to the lower limit of the threshold, the dose deviation D = 0.2; If the lower limit of the threshold is less than the actual dose and less than or equal to the upper limit of the threshold, the dose deviation D = 0.5; If the actual dose is greater than the upper threshold, the dose deviation D = actual dose / upper threshold. Example of parameters: In the dynamic rule base, the upper limit of the conventional threshold for penicillin in adult community-acquired pneumonia is 1200mg. If the actual standardized dose for a certain patient is 1500mg, then D=1500 / 1200=1.25; If the actual dose is 800mg, which is between the lower threshold of 500mg and the upper threshold of 1200mg, then D=0.5; The solvent risk level value S ranges from 0 to 5, where 0 indicates no risk and 5 indicates incompatible substances. Specifically, the basic risk level of the drug-solvent combination can be directly read from the compatibility matrix. For example, the basic risk level of penicillin plus 5% glucose is 2. When adjusting for liver and kidney function indicators, if the patient's creatinine is >133μmol / L, corresponding to renal insufficiency, the risk level is increased by 1 when the solvent is a "potassium-containing solution". For example, the risk level of the solvent is 3 after the base level of 2 is adjusted. Example parameters: If the patient has normal renal function and the prescription is penicillin plus 0.9% sodium chloride, the base level of the compatibility matrix is ​​0, then S=0; if the patient has renal insufficiency and the solvent is 10% potassium chloride solution, the base level is 3, and the corrected solvent risk level value S=4. The patient pathological fit P is the degree of matching between the patient's pathological characteristics and the drug warning level, which is calculated by summing the allergy history matching degree, age stratification fit degree, and complication correlation degree. Specifically, regarding the allergy history matching score, if the patient's allergy history list overlaps with the drug's contraindicated ingredients, the corresponding value is 1; if there is no overlap, the value is 0. Regarding age-stratified suitability, the value is 1 for children and elderly patients using adult doses, and 0 for age-stratified doses. Regarding the correlation with complications, if the drug instructions clearly state that a complication should be used with caution, the corresponding value is 1; if there are no complications, the value is 0. When calculating rule fit based on a standardized data matrix, a set of rules related to the current disease and drug is extracted from the rule base, including clinical guideline rules and historical best prescription rules. Among them, clinical guidelines, such as the "maximum daily dose of penicillin ≤4000mg" in the "National Guidelines for Antimicrobial Therapy"; Historical best prescription rules, prescription characteristics of the same disease that have been "passed without intervention" in the past 3 months, such as solvent selection preference and dose fluctuation range; Obtain several target vectors from a standardized data matrix. These target vectors specifically represent dosage, solvent, route of administration, and pathological suitability. Calculate the feature vector distance using weighted cosine similarity, and then apply the formula... Calculate the rule fit FA; where DX is the feature vector distance. , where i is the feature dimension index, corresponding to 4 target vectors; Let be the weighting coefficient for the i-th feature. The values ​​corresponding to dosage, solvent, route of administration, and pathological suitability are 0.3, 0.2, 0.2, and 0.3, respectively, satisfying the following condition: ; This is the i-th feature vector component of the current prescription, with specific values ​​ranging from: dose [0,2]; solvent [0,5]; route of administration [0,1]; and pathological suitability [0,1]. For the i-th feature vector component of the optimal prescription in the dynamic rule base, based on historical risk-free prescriptions and predefined values ​​from clinical guidelines, the value range is [missing information]. Consistent; When integrating the confidence label value, medication risk index RI, and rule fit FA to construct a three-dimensional evaluation tensor, the confidence level of the data corresponding to the X-axis is set to 0-100 points, with an interval of 10 points, and divided into 10 intervals; for example, [80,90) and [90,100]. The Y-axis represents the drug application risk index: 0-5 points, with intervals of 0.5 points, divided into 10 intervals; for example, [0,0.5), [0.5,1.0]. Z-axis rule fit: 0-100 points, with 10-point intervals, divided into 10 intervals; for example, [70, 80), [80, 90); Furthermore, the clinical intervention necessity score P is obtained through calculation, and the clinical intervention necessity score is set as the element value. The relevant expression is as follows: Where XB is the confidence label value; the higher the clinical intervention necessity score, the lower the intervention necessity; when the clinical intervention necessity score is <5, manual review is automatically triggered. When performing tensor visualization and storage, it is displayed in the form of a 3D heatmap, with red areas corresponding to low scores, indicating high-risk prescriptions; and green areas corresponding to high scores, indicating low-risk prescriptions. Tensor data is stored in CSV format, containing coordinates (X,Y,Z) and corresponding clinical intervention necessity scores, supporting clinical retrospective analysis.

[0025] In this embodiment of the invention, by weighted integration of multiple dimensions of factors such as dosage, solvent, and pathology, the risk of medication can be quantified, which can effectively improve the accuracy of identifying unconventional dosages and contraindicated solvents. Compared with traditional single-threshold judgment, it can effectively reduce the false judgment rate. The matching degree algorithm based on historical best prescriptions and real-time guideline updates enables the system to dynamically adapt to clinical changes, which can effectively improve the adaptation speed to new diseases. By constructing a three-dimensional assessment tensor, complex data can be transformed into intuitive scores, reducing unnecessary intervention for low-risk prescriptions and realizing risk visualization and graded intervention.

[0026] The joint judgment dynamic grading optimization module: Based on the three-dimensional evaluation tensor, it performs joint analysis and judgment on the credible label value, medication risk index, and rule fit; if the result is a judgment failure, it triggers the grading optimization mechanism and optimizes the implementation of the existing regulatory assessment scheme; the specific steps include: The trusted label value, medication risk index, and rule fit are analyzed using a joint model, and the joint analysis value LF is output. The expression of the joint model is as follows: ;in, These are the trusted label threshold, the medication risk threshold, and the rule adaptation threshold, which can be determined based on data such as the "Medical Institution Data Quality Specification" or by combining the application requirements of the actual application scenario with expert consensus. The specific values ​​are not limited. If the joint analysis value is 0, the result is considered passed; If the joint analysis value is 1, the test is deemed unsuccessful and a hierarchical optimization mechanism is triggered. When implementing a tiered optimization mechanism, if If so, data credibility optimization is performed, and the reasons for low credibility are located through the sub-dimensions of the credibility label value. These sub-dimensions include, for example, patient information completeness and drug code matching accuracy. Reasons for low credibility include, for example, the drug specification not being filled in, resulting in a code matching accuracy of 60%. When the backup data source is enabled, a pop-up interactive interface prompts pharmacists to manually enter missing information, such as drug specifications and solvent volume, and then the trusted label value is recalculated after the missing information is added. If the missing data is a functional test indicator, the latest test results will be automatically obtained by calling the hospital's LIS system interface; like Then, the weight parameters of the rule base are updated through reinforcement learning algorithms to optimize the risk assessment model; Reinforcement learning is a type of machine learning method that uses an agent to interact with the environment and continuously try and optimize decision-making strategies based on reward signals. Its core objective is to enable the agent to learn to take a series of actions in a specific environment in order to maximize long-term cumulative rewards. The weight parameters of the rule base updated by the reinforcement learning algorithm are existing conventional techniques, and the specific implementation steps will not be elaborated here.

[0027] In this embodiment of the invention, by jointly analyzing and judging the trust label value, medication risk index, and rule adaptability, and dynamically triggering a hierarchical optimization mechanism based on the judgment results, the implementation of the existing regulatory assessment scheme is proactively optimized and managed, thereby improving the proactive optimization effect of the implementation of the existing regulatory assessment scheme.

[0028] Multi-dimensional implementation effect evaluation and management module: Conducts multi-dimensional regulatory analysis on the implementation effect of existing regulatory evaluation schemes, proactively evaluates the implementation effect of existing regulatory evaluation schemes in corresponding local and overall dimensions, and implements targeted closed-loop iterative management; closed-loop iterative management includes local closed-loop iterative management and overall closed-loop iterative management.

[0029] When conducting multi-dimensional regulatory analysis, the total number of first-level regulatory actions N1 and the total number of second-level regulatory actions N2 that trigger the hierarchical optimization mechanism are statistically analyzed using the first iterative identification model, and the first iterative identification value DS1 is output. Similarly, the total number of second-level regulatory actions is analyzed using the second iterative identification model, and the second iterative identification value DS2 is output. The expression for the first iterative recognition model is: Where K is the optimization trigger threshold; a is a natural number greater than 0; The expression for the second-iteration recognition model is: Where N0 is the optimization implementation threshold; b is a natural number greater than 0; the optimization trigger threshold and the optimization implementation threshold can be determined based on the test simulation data in the early stage of the application of the regulatory assessment scheme, or they can be customized in combination with the application requirements and specifications of the actual application scenario; A joint analysis is performed on the first and second iteration identification values. If both values ​​are 0, the existing regulatory assessment scheme will continue to be implemented. If a value 'a' exists, then implement targeted overall closed-loop iterative management. If a value b exists, targeted local closed-loop iterative management is implemented. Specifically, local closed-loop iterative management involves iteratively upgrading the hierarchical optimization mechanism; overall closed-loop iterative management involves iteratively upgrading both the existing regulatory assessment scheme and the hierarchical optimization mechanism. The specific content and rules of different iterative upgrades can be determined according to the application requirements and specifications of the actual application scenario, and this embodiment of the invention does not impose specific limitations. By analyzing data and implementing targeted local closed-loop iterative management and overall closed-loop iterative management, the reliability of the existing drug use data regulatory assessment scheme and the reliability of the self-optimization scheme are further improved.

[0030] Unlike existing technical solutions that only focus on single-dimensional regulatory optimization without analyzing and iteratively managing the implementation and optimization effects from different dimensions, this invention analyzes the implementation effects of existing regulatory assessment schemes from multiple dimensions. It proactively evaluates the implementation effects of existing regulatory assessment schemes in both local and overall dimensions, and implements targeted closed-loop iterative management. This extends and expands existing regulatory assessment schemes and optimizations, ensuring the reliability of existing regulatory assessment schemes and the diversity of regulation from different dimensions.

[0031] In the several embodiments provided by this invention, it should be understood that the disclosed methods can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0032] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0033] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0034] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0035] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A drug use data monitoring and evaluation system based on artificial intelligence, characterized in that: include: Multi-source data acquisition and processing module: Acquires and preprocesses multi-source data, constructs a dual-source verification data sequence, and uses the dual-source verification data sequence to construct a standardized data matrix with trusted label values. The dual-source verification data sequence includes the original medication data from the medical institution information system and the third-party verification data stored on the blockchain. Three-dimensional evaluation tensor processing module: Calculates dynamic evaluation indicators based on standardized data matrices. Dynamic evaluation indicators include medication risk index and rule fit. Use dynamic evaluation indicators to construct a three-dimensional evaluation tensor. Joint Judgment Dynamic Grading Optimization Module: Based on the three-dimensional evaluation tensor, the module performs joint analysis and judgment on the credible label value, medication risk index, and rule fit. If the result is a judgment failure, the grading optimization mechanism is triggered, and the implementation of the existing regulatory assessment scheme is optimized and managed. Multi-dimensional implementation effect evaluation and management module: Conducts multi-dimensional regulatory analysis on the implementation effect of existing regulatory evaluation schemes, proactively evaluates the implementation effect of existing regulatory evaluation schemes in corresponding local and overall dimensions, and implements targeted closed-loop iterative management; closed-loop iterative management includes local closed-loop iterative management and overall closed-loop iterative management.

2. The drug use data monitoring and evaluation system based on artificial intelligence according to claim 1, characterized in that, When collecting original medication data from medical institution information systems and third-party verified data stored on the blockchain, the hospital information system is used to collect drug name, single dose, solvent type, route of administration, and prescription timestamp; the electronic medical record system is used to collect patient pathological characteristics; and the alliance link is used to obtain full life-cycle data of drugs, including drug batch number, solvent compliance certificate, and manufacturer qualifications.

3. The drug use data monitoring and evaluation system based on artificial intelligence according to claim 2, characterized in that, When constructing a standardized data matrix with trusted label values ​​using dual-source verified data sequences, the row index of the data matrix is ​​designed to be a single prescription, and the data is sorted in ascending order by prescription timestamp; The column index is a feature variable, including basic features, verification features, and trusted label values.

4. The drug use data monitoring and evaluation system based on artificial intelligence according to claim 3, characterized in that, When calculating the medication risk index based on a standardized data matrix, the formula is used. The calculation results are as follows: RI is the medication risk index; a, b, and c are all weighting coefficients, a+b+c=1; D is the dose deviation; S is the solvent risk level value; and P is the patient pathological fit.

5. The drug use data monitoring and evaluation system based on artificial intelligence according to claim 4, characterized in that, Obtain several target vectors from a standardized data matrix, calculate the feature vector distance using weighted cosine similarity, and then use the formula... Calculate the rule fit FA; where DX is the feature vector distance.

6. The drug use data monitoring and evaluation system based on artificial intelligence according to claim 5, characterized in that, When constructing a three-dimensional evaluation tensor by integrating trusted label values, drug risk index RI, and rule fit FA, the X-axis corresponds to data trustworthiness, the Y-axis corresponds to drug risk index, and the Z-axis corresponds to rule fit. Furthermore, the clinical intervention necessity score P is obtained through calculation, and the clinical intervention necessity score is set as the element value. The relevant expression is as follows: Where XB is the trusted label value.

7. The drug use data monitoring and evaluation system based on artificial intelligence according to claim 6, characterized in that, The trusted label value, medication risk index, and rule fit are analyzed using a joint model, and the joint analysis value LF is output. The expression of the joint model is as follows: ;in, These are the trusted label threshold, the medication risk threshold, and the rule adaptation threshold, respectively. If the joint analysis value is 0, the result is considered passed; If the joint analysis value is 1, the result is deemed unsuccessful, and a hierarchical optimization mechanism is triggered.

8. The drug use data monitoring and evaluation system based on artificial intelligence according to claim 7, characterized in that, When conducting multi-dimensional regulatory analysis, the total number of first-level regulatory actions (N1) and the total number of second-level regulatory actions (N2) that trigger the hierarchical optimization mechanism are statistically analyzed using the first-level regulatory assessment scheme. The first and second total number of regulatory actions are analyzed using the first-level iterative identification model, and the first-level identification value (DS1) is output. The second total number of regulatory actions is analyzed using the second-level iterative identification model, and the second-level identification value (DS2) is output.

9. The drug use data monitoring and evaluation system based on artificial intelligence according to claim 8, characterized in that, A joint analysis is performed on the first and second iteration identification values. If both values ​​are 0, the existing regulatory assessment scheme will continue to be implemented. If a value 'a' exists, then implement targeted overall closed-loop iterative management. If a value b exists, then targeted local closed-loop iterative management is implemented.

10. The drug use data monitoring and evaluation system based on artificial intelligence according to claim 8, characterized in that, The expression for the first iteration recognition model is: Where K is the optimization trigger threshold; a is a natural number greater than 0; The expression for the second-iteration recognition model is: Where N0 is the optimization implementation threshold; b is a natural number greater than 0.

Images