Drug interaction screening method based on drug use records

Abstract

The invention belongs to the technical field of medical information, and relates to a drug interaction screening method based on drug use records. The core structure comprises: a data integration module, which obtains a multi-source medication record of a patient in real time through an HL7 / FHIR interface and carries out structured processing; the metabolism engine calculation module is used for identifying a metabolism action mode based on the drug-metabolic enzyme influence relation database and quantifying a net effect value by adopting a dynamic time weighting algorithm; the risk decision module is used for integrating liver and kidney functions and gene polymorphism data and generating a three-dimensional risk matrix through a multi-factor prediction model; and the clinical output module is used for executing risk visualization early warning, alternative medicine recommendation and individualized monitoring scheme generation. Dynamic medication time sequence analysis, individualized risk quantification and clinical decision closed-loop management are realized, and medication safety is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical information technology, more particularly, the present application relates to a drug interaction screening method based on medication records. BACKGROUND

[0002] Drug-Drug Interaction (DDI) screening is the core of clinical rational drug use. With the popularity of polypharmacy in the treatment of chronic diseases, more than 40% of elderly patients use ≥5 drugs at the same time, which significantly increases the risk of complex DDIs. The current mainstream screening relies on the rule engine of static databases (such as Micromedex, Lexicomp) to achieve basic warning through drug pairing contraindication tables, but it cannot adapt to the dynamic medication scene in the real world. According to the statistics of the World Health Organization, 28% of preventable adverse drug events are caused by DDIs, which causes about 26 billion US dollars of economic loss worldwide each year, and a new generation of intelligent screening technology is urgently needed to break through the existing limitations. The existing technology still has some shortcomings:

[0003] 1. Dynamic medication timing analysis is missing

[0004] Existing systems only detect the current prescription combination, ignoring the influence of historical medication residues and the overlapping effect of drug administration time windows. For example, the CYP3A4 enzyme activity still continues to rise for 2 weeks after the withdrawal of rifampicin (a strong inducer), and traditional methods cannot capture this time-dependent interaction, resulting in missed reports of bleeding risks for narrow therapeutic window drugs such as warfarin after the withdrawal of inducers.

[0005] 2. Insufficient individual risk quantification capability

[0006] The mainstream tools use binary risk classification (presence / absence of DDI), without integrating key factors such as patient liver and kidney function, genetic polymorphism. CYP2C19 slow metabolizer patients using omeprazole + clopidogrel, the anti-platelet effect is reduced by more than 45%, but existing systems cannot dynamically adjust the risk level based on patient genotypes.

[0007] 3. Clinical decision-making loop is broken

[0008] Existing warnings only indicate the presence of risks, and lack of clinical action guidelines such as alternative drug recommendations and individualized monitoring plans. Doctors need to manually consult multiple sources of information to develop intervention strategies, resulting in more than 60% of high-risk DDIs not being timely processed, and the electronic prescription system and risk disposal process are in a fragmented state.

[0009] Therefore, the drug interaction screening method based on medication records is proposed to solve the above problems. SUMMARY

[0010] To overcome the above-mentioned defects of the prior art, embodiments of the present application provide a drug interaction screening method based on medication records to solve the problems raised in the above background.

[0011] To achieve the above object, the present application provides the following technical solution: a drug interaction screening method based on medication records, comprising the following consecutive steps: acquiring a full-dimension medication record data set of a target patient in real time through a hospital information system interface, the data set containing current prescription drugs, historical medication records and non-prescription drug use information, and the structured fields including at least drug generic name, trade name, dosage specification, medication frequency, medication start and end time stamp, administration route and treatment cycle; based on a pre-constructed drug action mechanism knowledge graph, identifying the induction or inhibition mode of each drug in the medication record on each subtype of the cytochrome P450 enzyme system (CYP1A2, CYP2C9, CYP2C19, CYP2D6, CYP3A4), the UGT enzyme system and the transporter protein (P-gp, OATP1B1, BCRP), and quantifying the intensity level thereof; according to the time overlap degree of drug combination, the metabolic path competition relationship and the clearance half-life parameters, using a dynamic time weighting algorithm to calculate the net metabolic influence effect value in the multi-drug combination state; inputting the net effect value and the patient's real-time liver and kidney function indicators (including creatinine clearance rate, Child-Pugh classification), genetic polymorphism data (CYP2C9*3, VKORC1 allele type) and combined disease information into a multi-factor risk prediction model to generate individualized drug interaction risk quantification score and graded clinical intervention strategy.

[0012] Preferably, the construction of the drug action mechanism knowledge graph comprises: enzyme induction intensity adopts a three-level classification system, wherein strong induction is defined as increasing enzyme activity by ≥3 times (such as rifampicin on CYP3A4), moderate induction is 1.5-3 times (such as carbamazepine on CYP2C19), and weak induction is <1.5 times (such as St. John's wort extract on CYP2E1); enzyme inhibition types are distinguished as reversible inhibition (such as fluconazole on CYP2C9), time-dependent irreversible inhibition (such as erythromycin on CYP3A4) and mechanism-based inhibition (such as grapefruit juice on CYP3A4); the transporter interaction is marked with specific action sites and inhibition constant Ki value, and the data sources are integrated from the FDA Adverse Event Reporting System, the PharmGKB clinical pharmacology database and more than 2000 pharmacokinetic literature evidence-based evidences.

[0013] Preferably, the dynamic quantitative evaluation is specifically implemented as: for a drug combination with overlapping medication time, calculating the net effect value according to the metabolic pathway correlation, the calculation formula is Σ (drug influence intensity coefficient × medication duration days × metabolic enzyme clinical weight factor), wherein the induced agent takes a positive value, the inhibitor takes a negative value, and the metabolic enzyme clinical weight factor is set according to evidence-based medicine evidence.

[0014] Preferably, the multi-factor risk prediction model comprises: a pharmacokinetics parameter module for calculating AUCR (AUC change rate) and C_maxR (peak concentration change rate) prediction values; a clinical consequence evaluation module for correlating QT interval prolongation, bleeding risk, and nephrotoxicity serious adverse reaction threshold values; and an output comprising a three-dimensional risk matrix of metabolic influence degree, clinical severity, and evidence level, wherein the evidence level is determined according to the warning level of the drug instruction and the literature evidence strength.

[0015] Preferably, the system further comprises a real-time warning mechanism: when the electronic prescription system adds a new drug, an interactive screening with existing medication records is automatically triggered, and if a high-risk interaction is detected, a visual warning icon is generated, and a substitute medication regimen recommendation and blood drug concentration monitoring suggestion are outputted, and the warning information is embedded into the hospital information system through the HL7 protocol.

[0016] Preferably, the system comprises: a data integration module configured to extract structured medication records from a hospital HIS system and a pharmacy management system; a metabolic engine calculation module internally storing the defined drug-enzyme relationship graph and executing the dynamic quantitative evaluation algorithm; a risk decision module integrating the multi-factor risk prediction model; and a clinical output module generating an interactive risk report conforming to the FHIR standard and executing the real-time warning of claim 5.

[0017] The technical effects and advantages of the present application are as follows:

[0018] Compared with the prior art, the core advantage of the present application is that a dynamic risk management system throughout the whole medication cycle is constructed, real-time structured analysis of patient historical and current medication records is realized through a multi-source heterogeneous data fusion engine, and the time sequence limitation of traditional static screening is broken through; a drug action mechanism knowledge graph constructed based on evidence-based medicine is combined with a metabolic pathway specific analysis module to accurately quantify the net influence effect on key metabolic enzymes and transporters under multi-drug combination; an innovative dynamic time weighting algorithm solves the problem of risk accumulation evaluation in long-term medication scenarios through sliding analysis of the medication time window, dose standardization correction, and metabolic enzyme clinical weight adaptation; a multi-factor risk prediction model integrates genetic polymorphism analysis and pharmacodynamic threshold to generate a three-dimensional risk quantization matrix; finally, a clinical decision support engine realizes risk visualization warning, intelligent recommendation of substitute medication, and generation of individualized monitoring scheme, and is deeply integrated with a hospital information system to form a "monitoring-evaluation-intervention" closed-loop management, thereby comprehensively improving the efficiency and safety of clinical medication decision-making. Attached Figure Description

[0019] Fig. 1 This is a system framework diagram of the present invention.

[0020] Fig. 2 This is a flowchart of the process of the present invention.

[0021] Fig. 3 This is an execution diagram of the dynamic time-weighted algorithm of the present invention. Detailed Implementation

[0022] 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.

[0023] Example 1:

[0024] As attached Figs. 1-3As shown, (1) a drug interaction screening method based on medication records, comprising the following consecutive steps: obtaining a full-dimensional medication record dataset of a target patient in real time through a hospital information system interface, the dataset containing current prescription drugs, historical medication records and non-prescription drug use information, and the structured fields including at least drug generic name, trade name, dosage specification, medication frequency, medication start and end timestamp, administration route and treatment cycle; based on a pre-constructed drug action mechanism knowledge graph, identifying the induction or inhibition mode of each drug in the medication record on each subtype of cytochrome P450 enzyme system (CYP1A2, CYP2C9, CYP2C19, CYP2D6, CYP3A4), UGT enzyme system and transporter protein (P-gp, OATP1B1, BCRP), and quantifying its intensity level; according to the time overlap degree of drug combination, the competition relationship of metabolic path and the clearance half-life parameter, the net metabolic effect value under the condition of multi-drug combination is calculated by using dynamic time weighting algorithm; the net effect value and the real-time liver and kidney function indicators of the patient (including creatinine clearance rate, Child-Pugh classification), genetic polymorphism data (CYP2C9*3, VKORC1 allele genotype) and combined disease information are input into a multi-factor risk prediction model to generate individualized drug interaction risk quantification score and graded clinical intervention strategy. In clinical implementation, first, the patient's medication data is collected in real time through the FHIR API interface of the hospital information system, including: the current electronic prescription in HIS (such as "Atorvastatin Calcium Tablets 20mg qd"), the dispensing record of the pharmacy management system (such as "Rivaroxaban Tablets 15mg bid-28 days have been dispensed"), and the patient's self-reported non-prescription drug use history (such as "continuous use of St. John's wort extract 300mg / day"). These data are parsed into structured datasets by a natural language processing engine, with key fields including drug generic name (atorvastatin), dosage specification (20mg), medication frequency (qd), and precise medication timestamp (2025-06-01 to 2025-06-28), etc. Then the pre-constructed drug action mechanism knowledge graph (stored in a Neo4j graph database) is called to identify atorvastatin as a CYP3A4 substrate and OATP1B1 transporter substrate, rivaroxaban as a CYP3A4 / P-gp substrate, and St. John's wort extract as a strong inducer of CYP3A4. According to the overlapping period of the three drugs (such as 2025-06-15 to 2025-06-28, a total of 14 days overlap) and the half-life of rivaroxaban (5-9 hours), the net metabolic effect value is calculated by using dynamic time weighting algorithm. Finally, the calculation results and the patient's real-time test data (such as creatinine clearance rate 45mL / min, CYP2C9*3 mutant genotype) are input into the risk model to generate a quantitative risk report (such as "Rivaroxaban AUCR is reduced to 0.4, bleeding risk level IV").

[0025] (2) The drug mechanism knowledge graph is constructed by: enzyme induction intensity adopts a three-level classification system, wherein strong induction is defined as enzyme activity being increased by more than 3 times (such as rifampicin on CYP3A4), moderate induction is 1.5-3 times (such as carbamazepine on CYP2C19), and weak induction is less than 1.5 times (such as St. John's wort extract on CYP2E1); enzyme inhibition types are distinguished as reversible inhibition (such as fluconazole on CYP2C9), time-dependent irreversible inhibition (such as erythromycin on CYP3A4), and mechanism-based inhibition (such as grapefruit juice on CYP3A4); the specific action site and inhibition constant Ki value of the transporter interaction are marked, and the data sources are integrated from the FDA Adverse Event Reporting System, the PharmGKB clinical pharmacology database, and more than 2000 pharmacokinetic literature evidence, wherein the construction process of the drug mechanism knowledge graph is: 1.16 million adverse reaction reports in the past 10 years are extracted from the FDA FAERS database, the CYP enzyme metabolism path data (such as the Km value of CYP2C19 metabolizing omeprazole) in PharmGKB are integrated, and the pharmacokinetic parameters (such as rifampicin increasing CYP3A4 activity by 5.8 times) in more than 2000 PubMed literature are extracted. After being reviewed by an expert committee, the enzyme induction effect is classified: strong inducers (such as rifampicin CYP3A4↑5.8 times), moderate inducers (such as carbamazepine CYP3A4↑2.2 times), and weak inducers (such as modafinil CYP3A4↑1.3 times). The specific mechanism of enzyme inhibition type is marked, such as clarithromycin inhibiting CYP3A4 by time-dependent irreversible inhibition (Ki=0.1 μM), and fluconazole is a reversible competitive inhibitor of CYP2C9 (Ki=2.8 μM). The key parameters of transporter interaction are marked, such as amiodarone inhibiting P-gp with IC50=4.6 μM, and rosuvastatin as an OATP1B1 substrate with Vmax=8.3 pmol / mg / min. All data are stored in the form of RDF triples to form a knowledge graph containing 380,000 relationships.

[0026] (3) The dynamic quantitative evaluation is specifically implemented as follows: for a drug combination with overlapping medication time, the net effect value is calculated according to the grouping of metabolic pathway correlation, and the calculation formula is Σ (drug influence intensity coefficient × medication duration days × metabolic enzyme clinical weight factor), wherein the induced agent takes a positive value, the inhibitor takes a negative value, and the metabolic enzyme clinical weight factor is set according to evidence-based medicine evidence, wherein the specific execution of the dynamic time weighting algorithm is as follows: taking June 1, 2025 00:00 to June 30, 2025 23:59 as the analysis period, the time window is divided by 24 hours step by step. For the drugs used in each time window, group them according to the shared metabolic pathway (such as CYP3A4 group: St. John's wort extract + atorvastatin + rivaroxaban). When calculating the net effect value, first standardize the dose parameter (such as St. John's wort extract 300 mg / day, the standard dose is 300 mg, D_i=1.0), determine the influence coefficient (strong inducer S_i=+2), and combine the clinical weight factor (CYP3A4 W_e=1.0). For drugs with a half-life of ≤12 hours (such as rivaroxaban t1 / 2=9 hours), introduce a time decay factor F_t=1-e^(-0.693*t / t1 / 2), t is the time from the last dose. The final calculation formula is: Net_Effect=[2(St. John's wort)×1.0(dose)×14(days)×1.0(W_e)×1.0(F_t)]+[-1(atorvastatin weak inhibitor)×1.0×14×0.8(W_e)]=22.4-11.2=11.2.

[0027] (4) The multi-factor risk prediction model comprises: a pharmacokinetic parameter module for calculating AUCR (AUC change rate) and C_maxR (peak concentration change rate) prediction values; a clinical consequence evaluation module for associating QT interval prolongation, bleeding risk, and renal toxicity severe adverse reaction threshold; and an output comprising a three-dimensional risk matrix of metabolic influence degree, clinical severity, and evidence level, wherein the evidence level is determined according to the drug instruction warning level and the literature evidence strength, wherein the three-layer architecture of the multi-factor risk prediction model operates as follows: the pharmacokinetic module: based on the net effect value 11.2, the output of the AUCR of rivaroxaban is 0.42 (warning threshold <0.5) through the simulation of the physiological pharmacokinetic model, the pharmacodynamic module: associated with the international normalized ratio (INR) risk threshold (>4.5), combined with the patient genotype (CYP2C9*3 / *3), the bleeding risk probability is predicted to be 32%, and the risk integration module: generates a three-dimensional matrix, the metabolic influence index is 8.5 / 10 (based on the AUCR deviation degree), the clinical severity is IV level (possibly fatal bleeding), and the evidence level is B level (Micromedex 1A evidence + FDA black box warning)

[0028] Wherein the evidence level determination rule: Level A = randomized controlled trial, Level B = observational study + package insert warning, Level C = case report, Level D = theoretical speculation.

[0029] (5) Further comprising a real-time warning mechanism: when the electronic prescription system adds a new drug, it automatically triggers an interaction screening with existing medication records, and if a high-risk interaction is detected, a visual warning icon is generated, and a substitute medication regimen recommendation and blood drug concentration monitoring suggestion are output, the warning information is embedded in the hospital information system through the HL7 protocol, and when a high-risk level IV is detected, the system performs the following clinical decision support: a red warning matrix pops up on the doctor's workstation, labeled "St. John's wort strongly induces CYP3A4, causing a 58% decrease in rivaroxaban blood drug concentration", and recommends the following alternative: first choice: stop St. John's wort and use sertraline (interaction risk level I), second choice: replace rivaroxaban with apixaban (CYP3A4 metabolic rate < 25%), generate monitoring plan: INR monitoring: detect on day 1 / 3 / 7, kidney function monitoring: serum creatinine every week, symptom tracking: daily stool occult blood self-test, packaged through FHIR resources, wherein the DetectedIssue resource records the risk mechanism, and the MedicationRequest resource pushes the alternative to the EMR system.

[0030] (6) The system comprises: a data integration module configured to interface with the hospital HIS system and the pharmacy management system to extract structured medication records; a metabolic engine calculation module that internally stores the drug-enzyme relationship graph defined in claim 2 and executes the dynamic quantitative evaluation algorithm of claim 3; a risk decision module that integrates the multi-factor risk prediction model of claim 4; a clinical output module that generates an interaction risk report in accordance with the FHIR standard and performs real-time warnings according to claim 5, wherein the system hardware implementation scheme: data integration engine: deployed in a Docker container, EMR data (Epic Systems C-CDA document) is called through HL7 FHIR API, BERT model is used to parse prescription text (accuracy rate 98.2%), metabolic analysis engine: equipped with NVIDIA T4 GPU, running knowledge graph query (Cypher statement) and dynamic algorithm (CUDA parallel computing, processing speed up to 1500 / second), risk decision engine: running XGBoost model in Kubernetes cluster, real-time receiving liver and kidney function data (AST / ALT value from LIS system), clinical interaction engine: warning information is pushed to the doctor's workstation (C# client) through WebSocket, and is also pushed to the mobile APP (iOS / Android) in JSON format.

[0031] Example Two:

[0032] Step 1: Multi-source medication data collection and structured processing

[0033] Operation details: Real-time retrieval of three aspects of data through the FHIR API interface (version R4) of the hospital information system:

[0034] ① Current prescription in the electronic medical record (such as "Warfarin Sodium Tablets 3mg qd" in the MedicationRequest resource)

[0035] ② Dispensing record of the pharmacy management system (such as "Amiodarone Tablets 200mg bid - Dispensed for 30 days" in the MedicationDispense resource)

[0036] ③ Non-prescription drug use reported by the patient mobile APP (such as "Continuous use of Coenzyme Q10 Soft Capsules 100mg / day")

[0037] Natural language processing: Use the BERT-Base pre-training model to analyze free-text prescriptions, for example, convert "Lipitor 20mg every night"

[0038] to structured data:

[0039] Generic name = Atorvastatin Calcium, Dose = 20mg, Frequency = QD, Time of administration = daily 21:00. Step 2: Analysis of drug metabolism mechanism

[0040] Knowledge graph query: Perform Cypher query statements based on the pre-built Neo4j graph database: MATCH (d: Drug {name: "Amiodarone"} ) - [: INHIBITS ->] (e: Enzyme {name: "CYP2C9"} ) RETURN e. inhibition_strength.

[0041] Intensity classification:

[0042] Strong inhibitor: Reduce enzyme activity by ≥80% (such as amiodarone with 92% inhibition of CYP2C9);

[0043] Moderate inhibitor: 50%-80% (such as fluoxetine with 65% inhibition of CYP2D6);

[0044] Weak inhibitor: <50% (such as cimetidine with 40% inhibition of CYP3A4);

[0045] Transporter annotation: Label warfarin as a P-gp transporter substrate and amiodarone as a P-gp inhibitor. Step 3: Dynamic time weighting algorithm execution

[0046] Time window division: Divide the patient's medication period into 31 time windows according to 24 hours (for example, from 2025-07-01 to 2025-07-31).

[0047] Parameter assignment rules:

[0048] Drug impact intensity coefficient (S_i): +2 for strong inducer, -2 for strong inhibitor, -1 for moderate inhibitor, 0 for substrate.

[0049] Dose normalized value (D_i): actual daily dose ÷ standard therapeutic dose (e.g. warfarin 3 mg / day, standard dose 5 mg, D_i = 0.6).

[0050] Effective days (T_i): actual days of drug overlap within the time window (e.g. warfarin and amiodarone overlap for 25 days in July).

[0051] Metabolic enzyme weight factor (W_e): CYP3A4 = 1.0, CYP2D6 = 0.8, CYP2C9 = 0.9 (assigned based on clinical importance).

[0052] Time decay factor (F_t): 1.0 for drugs with half-life > 12 hours; for drugs with half-life ≤ 12 hours, calculated by formula: 1 - e^(-0.693 x time to last dose ÷ half-life).

[0053] Net effect value calculation:

[0054] For each combination of drugs sharing the same metabolic pathway within a time window, perform the following calculation procedure:

[0055] ① Identify drugs in the same group: e.g. warfarin (substrate) and amiodarone (inhibitor) in CYP2C9 metabolic group

[0056] ② Calculate single drug contribution value:

[0057] Amiodarone contribution value = inhibition intensity coefficient (-2) x dose normalized value (200 mg ÷ 400 mg = 0.5) x effective days (25) x enzyme weight factor (0.9) x time decay factor (1.0) = (-2) x 0.5 x 25 x 0.9 x 1.0 = -22.5.

[0058] ③ Accumulate net effect value: net effect value for this group = -22.5 (warfarin as substrate does not participate in calculation).

[0059] Step 4: Multi-factor risk prediction model operation

[0060] Pharmacokinetic simulation: input net effect (-22.5) into physiologically-based pharmacokinetic model, calculate the change of pharmacokinetic parameters of warfarin: AUC change rate (AUCR) = 1 / [1+(absolute value of net effect / 50)], in this case AUCR = 1 / [1+(22.5 / 50)] = 0.69; peak concentration change rate (C_maxR) = 1 / [1+(absolute value of net effect / 100)] = 0.82.

[0061] Pharmacodynamic assessment:

[0062] Alert triggered when AUCR > 2.0 or < 0.5, in this case AUCR = 0.69 (alert threshold not triggered) but combined with patient genotype (CYP2C9*3 / *3), predicted INR value fluctuation risk increased 3-fold.

[0063] Three-dimensional risk matrix generation:

[0064]

[0065] Step 5: Clinical decision support execution

[0066] High-risk handling: when clinical severity is ≥ grade III (as in this case grade III):

[0067] Yellow alert box pops up on physician workstation: "Warning: Amiodarone inhibits CYP2C9, resulting in 31% decrease in warfarin metabolism (CYP2C9*3 / *3 gene exacerbates risk)".

[0068] Recommended alternative:

[0069] Preferred: switch to rivaroxaban (not metabolized by CYP2C9);

[0070] Alternative: adjust amiodarone to propafenone;

[0071] Generate monitoring plan:

[0072] INR monitoring: detect on day 1 / 3 / 7 / 14;

[0073] Sign observation: daily self-check of gum bleeding;

[0074] System integration: generate DetectedIssue resource through FHIR standard:

[0075]

[0076] Step 6: Dynamic risk tracking and audit

[0077] Timeline construction: X-axis is date sequence (2025-07-01 to 07-31), Y-axis records daily metabolic impact index.

[0078] Automatic re-evaluation: when acetaminophen was added on July 15th: detected weak inhibition on CYP2C9 (30% inhibition rate), recalculate net effect value: original -22.5 + (-0.5 x 0.3 x 17 x 0.9) = -24.8, update risk level to level IV (high risk).

[0079] Blockchain audit: an audit record is generated each time the evaluation is performed, containing:

[0080] Timestamp: 2025-07-15T14:22:36Z;

[0081] Operator ID: Physician_0238;

[0082] Data hash: SHA256("Warfarin 3mg | Amiodarone 200mg | Acetaminophen 500mg") is stored on-chain by Ethereum smart contract, with a Gas Limit of 50,000.

[0083] Finally, it should be noted that in the description of the present application, it should be noted that unless otherwise specified and limited, the terms "mounting", "connecting", "connecting" should be broadly understood, which can be mechanical connection or electrical connection, or the internal connection of two elements, or direct connection, "up", "down", "left", "right" and the like are only used to indicate relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may change;

[0084] Secondly: the drawings of the disclosed embodiments only involve the structures involved in the disclosed embodiments, other structures can refer to the usual design, and in the case of no conflict, the same embodiment and different embodiments of the present application can be combined with each other;

[0085] Finally: the above only describes the preferred embodiments of the present application and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for drug interaction screening based on medication records, characterized by It comprises the following consecutive steps: obtaining a full-dimension medication record dataset of a target patient in real time through a hospital information system interface, the dataset containing current prescription drugs, historical medication records and non-prescription drug use information, and the structured fields including at least drug generic name, trade name, dosage specification, medication frequency, medication start and end time stamp, administration route and treatment cycle; based on a pre-constructed drug action mechanism knowledge graph, identifying the induction or inhibition mode of each drug in the medication record on each subtype of cytochrome P450 enzyme system (CYP1A2, CYP2C9, CYP2C19, CYP2D6, CYP3A4), UGT enzyme system and transporter protein (P-gp, OATP1B1, BCRP), and quantifying the intensity level; according to the time overlap degree of drug combination, metabolic pathway competition relationship and elimination half-life parameters, using a dynamic time weighting algorithm to calculate the net metabolic effect value under the condition of multiple drug combination; inputting the net effect value and real-time liver and kidney function indicators of the patient (including creatinine clearance rate, Child-Pugh classification), genetic polymorphism data (CYP2C9*3, VKORC1 allele type) and combined disease information into a multifactor risk prediction model to generate individualized drug interaction risk quantification score and graded clinical intervention strategy.

2. The method of claim 1, wherein the method is based on a medication record. The construction of the drug action mechanism knowledge graph includes: enzyme induction intensity adopts a three-level classification system, where strong induction is defined as increasing enzyme activity by ≥3 times (such as rifampicin on CYP3A4), moderate induction is 1.5-3 times (such as carbamazepine on CYP2C19), and weak induction is <1.5 times (such as St. John's wort extract on CYP2E1); enzyme inhibition types are distinguished as reversible inhibition (such as fluconazole on CYP2C9), time-dependent irreversible inhibition (such as erythromycin on CYP3A4) and mechanism-based inhibition (such as grapefruit juice on CYP3A4); transporter interaction is annotated with specific action sites and inhibition constant Ki value, and the data sources integrate evidence-based evidence from FDA Adverse Event Reporting System, PharmGKB clinical pharmacology database and more than 2000 pharmacokinetic literatures. 3.The medication record-based drug interaction screening method of claim 1, wherein The dynamic quantification evaluation is specifically implemented as follows: for drug combinations with overlapping medication times, the net effect value is calculated according to the metabolic pathway correlation, and the calculation formula is Σ (drug influence intensity coefficient × medication duration × metabolic enzyme clinical weight factor), where the induction agent takes a positive value and the inhibitor takes a negative value, and the metabolic enzyme clinical weight factor is set according to evidence-based medical evidence. 4.The method of claim 1, wherein The multifactor risk prediction model includes: a pharmacokinetic parameter module for calculating AUCR (AUC change rate) and C_maxR (peak concentration change rate) prediction values; a clinical consequence evaluation module for correlating QT interval prolongation, bleeding risk and nephrotoxicity serious adverse reaction threshold values; and an output containing a three-dimensional risk matrix of metabolic influence degree, clinical severity and evidence level, wherein the evidence level is determined according to the warning level of the drug instruction and the strength of the literature evidence. 5.The medication record-based drug interaction screening method of claim 1, wherein Also included is a real-time alert mechanism: when an electronic prescription system adds a new drug, an automatic interaction screening with existing medication records is triggered, and if a high-risk interaction is detected, a visual warning icon is generated, along with an alternative medication regimen recommendation and blood drug concentration monitoring suggestion, and the warning information is embedded in the hospital information system through the HL7 protocol.

6. A drug interaction screening system characterized by Comprise: a data integration module configured to interface with hospital HIS systems and pharmacy management systems to extract structured medication records; a metabolism engine calculation module that incorporates the drug-enzyme relationship map defined in claim 2 and executes the dynamic quantitative evaluation algorithm of claim 3; a risk decision module that integrates the multi-factor risk prediction model of claim 4; a clinical output module that generates an interaction risk report in accordance with the FHIR standard and executes the real-time alert of claim 5.

Images