Intelligent detection system and method for drug compatibility risk
By combining graph databases and graph neural network algorithms with pharmacokinetic models, the problems of data silos and system integration in drug compatibility review systems have been solved. This enables dynamic and accurate detection and graded early warning of multi-drug combinations, provides alternative solutions and dosing interval guidance, and improves drug compatibility safety.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE THIRD PEOPLES HOSPITAL OF CHENGDU
- Filing Date
- 2026-03-30
- Publication Date
- 2026-06-26
AI Technical Summary
Existing drug compatibility review systems suffer from data silos, are limited to single departments, cannot identify complex cascade reactions of multi-drug combinations, lack timely assessment and graded early warning, have poor system integration, and are difficult to integrate efficiently with hospital information systems.
By employing graph databases and graph neural network algorithms, combined with pharmacokinetic models, and integrating hospital-wide medication data, dynamic, real-time detection and tiered early warning of multi-drug combinations are achieved. Through the FHIR standard interface, it is integrated with the hospital information system, and multi-level compatibility risk detection algorithms and graph neural network models are used to identify potential complex interactions.
It enables dynamic and accurate detection of drug compatibility risks, can identify complex cascade reactions, provide graded early warnings and alternative recommendations, improves the depth and breadth of detection, and ensures the safety and effectiveness of clinical operations.
Smart Images
Figure CN122290862A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical information technology, specifically relating to an intelligent detection system and method for drug compatibility risks. Background Technology
[0002] In modern medical practice, multidrug combination therapy is a common approach, especially in complex settings such as intensive care units (ICUs) and emergency departments, where patients often receive multiple medications simultaneously. The safety of drug compatibility is crucial to treatment efficacy and patient safety. Drug incompatibilities refer to undesirable physical or chemical incompatibilities that occur when two or more drugs are mixed in vitro or in vivo, potentially leading to reduced efficacy, increased toxicity, precipitation, or gas formation, among other serious adverse consequences.
[0003] In the process of developing this invention, the inventors discovered that existing drug compatibility review systems or methods have at least the following problems: 1. Data silos and limitations of single departments: Existing systems are mostly limited to medication records of a single department, and cannot obtain the complete medication history of patients before referral or in other departments, resulting in the neglect of cross-departmental compatibility risks.
[0004] 2. Limitations of two-drug detection: Most systems can only detect incompatibilities between any two drugs and cannot identify complex cascade reactions that occur when a third or more drugs are added. For example, drug A and drug B can be used alone without any problems, but when coexisting with drug C, precipitation will occur.
[0005] 3. Lack of timeliness assessment: The system generally lacks consideration of drug metabolism time. A drug used 12 hours ago may still have a residual concentration in the body that can react with newly used drugs, but existing systems miss this risk because they cannot perform dynamic assessment.
[0006] 4. Limited early warning mechanism: Most systems only provide simple text prompts and lack an early warning mechanism that grades risks according to their severity, making it difficult to effectively guide clinical operations.
[0007] 5. Poor system integration: Many systems use proprietary API interfaces, making it difficult to achieve efficient and standardized data integration with existing hospital HIS (Hospital Information System), EMR (Electronic Medical Record System), or PDA medication verification terminals. Summary of the Invention
[0008] The present invention aims to at least partially solve the aforementioned technical problems. Therefore, the present invention aims to provide an intelligent detection system and method for drug compatibility risks, capable of integrating hospital-wide medication data and combining pharmacokinetics and advanced graph neural network algorithms to achieve dynamic, real-time, accurate detection and graded early warning of complex multi-drug combination compatibility risks.
[0009] To achieve the above objectives, the present invention provides an intelligent detection method for drug compatibility risks, comprising the following steps: a. Establish and maintain a multi-dimensional drug compatibility knowledge base based on a graph database, wherein the knowledge base includes drug nodes, solvent nodes, and edges representing the compatibility relationships between drugs; b. Automatically obtain the drug information to be executed through a standardized medical information exchange interface, and retrieve the target patient's medication history throughout the hospital; c. Based on the pharmacokinetic compartment model, calculate the estimated blood concentration C(t) of each drug in the medication history at the current time, screen out drugs with estimated blood concentrations greater than a preset threshold, and construct an effective drug list; d. Execute a multi-level compatibility risk detection algorithm to combine and analyze the drug information to be executed with the drugs in the effective drug list; e. Based on the output of the detection algorithm, generate and output graded early warning information containing risk level and specific description.
[0010] In a preferred embodiment, in step c, the estimated blood drug concentration C(t) is calculated using a single-compartment model, with the formula: C(t) = C0 × e^(-k × t); where the initial blood drug concentration C0 is calculated based on the drug's dosage, bioavailability corresponding to the route of administration, and apparent volume of distribution; the elimination rate constant k is calculated based on the drug's elimination half-life and is individually corrected based on the target patient's liver and kidney function indicators (such as creatinine clearance).
[0011] In a preferred embodiment, the multi-level compatibility risk detection algorithm in step d includes two levels: First, based on the traversal query of the graph database, known and explicitly recorded compatibility contraindications are detected for any combination of two drugs in the knowledge base; second, when the number of drug combinations reaches or exceeds three, a cascade reaction risk prediction based on a graph neural network model (GNN model) is initiated to discover those complex interactions that are not explicitly recorded but may exist.
[0012] In a preferred embodiment, the risk prediction based on the GNN model includes: constructing a high-dimensional feature vector for each drug in the combination, containing molecular structural features and pharmacological properties; aggregating information from each drug node and its neighboring nodes within the combination through the multi-layer information transfer mechanism of the GNN to learn the contextual embedding representation of each drug in the current combination environment; and finally predicting the risk probability of the entire drug combination based on this representation. The GNN model can also incorporate preset dangerous three-drug combination patterns to identify specific high-risk combinations such as pH oscillations and competitive metabolism.
[0013] In a preferred embodiment, the graded warning information generated in step e includes a clear risk level, such as "absolute contraindication", "relative contraindication" or "safe", and provides a specific textual description of the detected contraindication, such as "may produce precipitation when mixed with drug X".
[0014] In a preferred embodiment, the method further includes a step of providing decision support. This step is activated after a high-risk warning is generated and may include: querying a knowledge base and recommending safe alternative drugs based on the drug's ATC classification code; or, calculating a safe dosing interval for the user using a pharmacokinetic model. This time interval t is derived by reverse engineering the basic pharmacokinetic formula C(t) = C0 × e^(-k × t), by solving for the concentration C when C(t) decreases to a preset safe threshold concentration C. safe The time t required is calculated using the formula: t = ln(C0 / C safe ) / k.
[0015] In a preferred embodiment, the standardized medical information interaction interface is the FHIR standard interface, which ensures efficient interoperability with mainstream medical information systems.
[0016] In a preferred embodiment, the maintenance of the knowledge base is automated by deploying an NLP pipeline to continuously extract new knowledge from medical literature and add it to the database after review by pharmacists.
[0017] The present invention also provides an intelligent detection system for drug compatibility risks, which is equipped with a processor and a memory and is capable of executing the above-described intelligent detection method.
[0018] The beneficial effects of this invention are as follows: This invention breaks down information barriers between departments by adopting the FHIR standard interface, enabling real-time acquisition and analysis of a patient's entire hospital medication history, effectively preventing cross-departmental medication risks caused by incomplete information. This invention uses graph databases and GNN models, upgrading from pairwise detection to multi-drug combination prediction, which can efficiently identify complex, nonlinear cascade reactions, significantly improving the depth and breadth of detection.
[0019] By introducing a pharmacokinetic compartmental model (PK model), this invention can dynamically assess the residual effects of historical drug use in the body, transforming the detection from a static "whether medication was used" to a dynamic "current drug concentration in the body," making risk assessment more accurate and closer to clinical practice. This invention can not only issue early warnings, but also provide alternative recommendations based on ATC classification and dosing interval calculations based on the PK model, providing clinicians with an actionable solution. Attached Figure Description
[0020] Figure 1This is a flowchart illustrating an intelligent detection method for drug compatibility risks according to the present invention.
[0021] Figure 2 This is a functional module architecture diagram of a drug compatibility risk intelligent detection system according to the present invention.
[0022] Figure 3 This is a schematic diagram of the internal workflow of the intelligent analysis engine in this invention. Detailed Implementation
[0023] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0024] It should be understood that, and also noted, in the embodiments, the functions / actions may appear in a different order than those shown in the figures. For example, depending on the functions / actions involved, they may actually be performed substantially concurrently, or sometimes the two figures shown consecutively may be performed in reverse order.
[0025] like Figure 1 As shown in this embodiment, the intelligent drug compatibility risk detection method disclosed in this embodiment has the following process: Step a: Establish and maintain a multi-dimensional drug compatibility knowledge base.
[0026] The first step of this method is to build a robust knowledge base. This step is accomplished by the system's knowledge base module and automated knowledge extraction module.
[0027] Knowledge Base Construction: The system employs graph database technology (such as Neo4j), whose network structure is naturally suitable for expressing complex relationships between drugs. The data model is meticulously designed, including nodes such as Drug, Solvent, and Patient, as well as edges such as INCOMPATIBLE_WITH and TRIGGERS_CASCADE. The Drug node records detailed information on key attributes such as ATC code, half-life, and molecular structure (SMILES string).
[0028] Knowledge Base Maintenance: To ensure the freshness of knowledge, the system continuously learns from authoritative sources such as PubMed and FDA drug instructions through an NLP pipeline. This pipeline automatically extracts drug compatibility relationships and pushes them to a review interface designed specifically for pharmacists. After pharmacist review and confirmation, the new knowledge is seamlessly integrated into the graph database, ensuring the authority and timeliness of the knowledge base.
[0029] Step b: Obtain hospital-wide medication data through a standardized interface.
[0030] This method is triggered when clinical staff (such as nurses) scan a patient's wristband and medication barcode using a handheld smart device (PDA) during medication administration. This step is performed by the system's FHIR standard interface module.
[0031] This module adheres to the international FHIR R4 / R5 standard, ensuring seamless integration with various hospital information systems (HIS / EMR). It initiates a series of standardized data requests to efficiently and accurately obtain two key types of information: 1) information on currently pending medications; and 2) the patient's complete medication history across all departments in the hospital, including drug name, dosage, administration time, and route. Simultaneously, it retrieves personalized data such as the patient's liver and kidney function indicators.
[0032] Step c: Construct a time-sensitive medication list based on the PK model.
[0033] After acquiring the data, the method enters the core intelligent analysis engine. The first step is to perform pharmacokinetic (PK) timeliness filtering.
[0034] This step aims to address the question of whether yesterday's medication still has an effect today. The analysis engine calculates the effect for each drug in the historical medication records using the single-compartment model formula C(t)=C0×e^(-k×t).
[0035] The parameters C0 (initial concentration) and k (elimination rate constant) in the model are calculated individually: C0 takes into account the route of administration (intravenous or oral) and dosage; k is based on the drug half-life and dynamically adjusted according to the latest liver and kidney function data of patients obtained from the FHIR interface. For example, the k value is lowered for patients with renal insufficiency to simulate the real situation of slowed drug metabolism.
[0036] After the calculation is completed, only those drugs whose current estimated blood drug concentration C(t) is higher than a very small clinically significant threshold are considered to still be active and are screened out to form an effective drug list.
[0037] Step d: Perform multi-level compatibility risk detection.
[0038] Next, the multi-level risk detector within the analysis engine performs in-depth analysis of the combination of "pending drugs" and "effective drug list".
[0039] First layer: Two-drug detection. The system utilizes the efficient query capabilities of graph databases to quickly traverse any two drugs in the combination and check whether there is a known INCOMPATIBLE_WITH relationship edge between them. This can quickly eliminate known and simple incompatibilities.
[0040] The second layer: Multi-drug combination prediction. When the number of drugs in a combination reaches three or more, the method activates a GNN model for prediction. By simulating the information transfer between drug molecules, the GNN can learn how drug properties interact in a specific multi-drug environment. For example, through five layers of information transfer, the model can capture the complex effect of drug A indirectly affecting drug D through B and C. The model is also trained to identify specific dangerous three-way combination patterns, such as pH oscillation patterns, thereby discovering hidden cascade reaction risks.
[0041] Step e: Generate tiered early warning information and provide decision support.
[0042] After the test is completed, the analysis results are sent to the system's early warning and decision support module.
[0043] Tiered Warning: This module generates and pushes intuitive tiered warning information to the user's PDA based on the severity of the risk. For example: Red Alert (Absolutely Prohibited): "Danger! Mixing ceftriaxone sodium with calcium-containing solutions will produce a deadly precipitate!" Yellow warning (relative contraindication): "Caution: This combination may increase the risk of bleeding. Please monitor coagulation parameters closely." Green label (safe): "No incompatibilities found." Decision Support: Upon issuing a red or yellow alert, this module also activates decision support functionality, providing two feasible solutions: Alternative Solution Recommendation: Based on the ATC classification code of the drug to be used, the system searches for similar alternative drugs in the knowledge base, and uses a GNN model to quickly simulate the compatibility of the alternative drugs with the current combination, recommending one or more safe options.
[0044] Intelligent dosing interval calculation: If alternatives are not feasible, the system will calculate a safe dosing waiting time for the user. This function is based on the core pharmacokinetic model. Given the formula for drug concentration decay over time as C(t) = C0 × e^(-k × t), the system sets a target concentration as a pre-existing safe threshold C in the knowledge base. safe The time t is calculated in reverse. The derivation process is as follows: Let C(t) = C safe Then C safe =C0×e^(-k×t), taking the natural logarithm of both sides gives ln(C safe / C0)=-kt, and finally we get t=ln(C0 / C safe The system clearly displays the calculated time t (e.g., "Please wait 2 hours and 15 minutes") to the user, providing precise data guidance for clinical operations.
[0045] The method of this invention breaks down a complex clinical safety issue into four stages—data acquisition, timely filtering, in-depth detection, and intelligent decision-making—through a series of interconnected intelligent processing steps. These stages are implemented by a well-designed set of system functional modules, thereby providing a comprehensive, dynamic, and accurate solution for drug compatibility risks.
[0046] This invention is not limited to the above-described optional embodiments. Anyone can derive other various forms of products under the guidance of this invention. However, regardless of any changes made in their shape or structure, any technical solution that falls within the scope of the claims of this invention shall be protected by this invention.
Claims
1. A smart detection method for drug compatibility risk, characterized in that, Includes the following steps: a. Establish and maintain a multi-dimensional drug compatibility knowledge base based on a graph database, wherein the knowledge base includes drug nodes, solvent nodes, and edges representing the compatibility relationships between drugs; b. Automatically obtain the drug information to be executed through a standardized medical information exchange interface, and retrieve the target patient's medication history throughout the hospital; c. Based on the pharmacokinetic compartment model, calculate the estimated blood concentration C(t) of each drug in the medication history at the current time, screen out drugs with estimated blood concentrations greater than a preset threshold, and construct an effective drug list; d. Execute a multi-level compatibility risk detection algorithm to combine and analyze the drug information to be executed with the drugs in the effective drug list; e. Based on the output of the detection algorithm, generate and output graded early warning information containing risk level and specific description.
2. The intelligent detection method according to claim 1, characterized in that, In step c, the formula for calculating the estimated blood drug concentration C(t) is: C(t) = C0 × e^(-k × t); The initial blood drug concentration C0 is calculated based on the drug's dosage, route of administration, corresponding bioavailability, and apparent volume of distribution; the elimination rate constant k is calculated based on the drug's elimination half-life and is individually adjusted according to the liver and kidney function indicators of the target patient.
3. The intelligent detection method according to claim 1, characterized in that, The multi-level compatibility risk detection algorithm in step d includes: Based on graph database traversal query, known incompatibilities are detected for any combination of two drugs; Additionally, when the number of drug combinations is greater than or equal to three, a cascade reaction risk prediction based on a graph neural network model is initiated.
4. The intelligent detection method according to claim 3, characterized in that, The risk prediction based on the graph neural network model includes: For each drug in the combination, construct a high-dimensional feature vector containing molecular structural features and pharmacological properties; By leveraging the multi-layered information transfer mechanism of GNN, information from each drug node and its neighboring nodes within the combination is aggregated to learn the contextual embedding representation of each drug in the current combination environment. Based on the contextual embedding representation, the compatibility risk probability and response type of the entire drug combination are predicted.
5. The intelligent detection method according to claim 4, characterized in that, The graph neural network model has a built-in preset dangerous three-drug combination mode, which is used to identify specific high-risk combinations, including pH oscillation mode, competitive metabolism mode or ion complexation cascade mode.
6. The intelligent detection method according to claim 1, characterized in that, The graded early warning information generated in step e includes at least one of the following risk levels: absolute taboo, relative taboo, or safe; wherein, the early warning information for absolute taboo and relative taboo levels also includes a textual description of the potential risk.
7. The intelligent detection method according to claim 1, characterized in that, It also includes a step of providing decision support after the early warning information is generated, which includes at least one of the following: When a high compatibility risk is detected, the system queries the knowledge base for alternative drugs of the same category based on the ATC classification code, simulates the check of the compatibility of the alternative drugs with the list of effective drugs, and recommends safe alternative options. The calculation reduces the estimated blood drug concentration of a certain drug in the body to a safety threshold C safe The time interval t required to wait, where t = ln(C0 / C safe ) / k.
8. The intelligent detection method according to claim 1, characterized in that, The standardized medical information interaction interface is the FHIR standard interface, which obtains medication records, patient basic information and liver and kidney function indicators, and laboratory test results by calling the MedicationRequest, Patient, and Observation resources, respectively.
9. The intelligent detection method according to claim 1, characterized in that, The maintenance steps for the knowledge base include: Deploy a natural language processing pipeline to automatically extract drug compatibility relationship entities from external medical literature data sources; Provide a pharmacist review interface for manual review and confirmation of automatically extracted compatibility relationships; The approved matching relationships will be automatically updated to the graph database.
10. An intelligent detection system for drug compatibility risks, characterized in that, include: One or more processors; A memory, wherein the memory stores instructions that can be executed by the one or more processors; When the instruction is executed, it causes the one or more processors to implement the intelligent detection method as described in any one of claims 1 to 9.