An artificial intelligence-based all-medical order rational drug use auditing method and platform

Abstract

The application discloses an artificial intelligence-based all-medical order rational drug use auditing method and platform, which comprises the following steps: by integrating patient structured data and unstructured text, determining physiological safety range in time sequence, and constructing panoramic contraindication medical image; after deep information extraction and multi-source data fusion, forming a patient panoramic image, and comparing it with medical order drugs for contraindication, triggering hard blocking warning to avoid drug conflict; further establishing a digital twin model, setting the medical order compared by contraindication as a baseline scheme, simulating the dynamic change of physiological indicators within 72 hours, and assisting doctors in adjusting drug dosage combined with clinical experience; finally, according to the simulation curve, the individualized drug safety baseline is established, precise simulation, dynamic early warning and scientific decision are realized; the scheme significantly improves the treatment safety and precision, provides all-process intelligent drug use auditing support for complex patients, and has important clinical application value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical information processing, and in particular to a full medical order rational drug use auditing method and platform based on artificial intelligence. BACKGROUND

[0002] In the traditional medical scene, the rationality of medical orders highly depends on the personal experience and static knowledge base of doctors, and there are problems of insufficient information integration and lack of dynamic evaluation. Patient data is scattered in structured forms (such as electronic medical records, test reports) and unstructured texts (such as medical records, consultation opinions), and doctors need to manually cross-check contraindication information, which is time-consuming and easy to miss. For example, the adjustment of drug dosage for patients with abnormal liver function needs to be combined with test results and drug instructions, but existing systems cannot correlate multiple data sources in real time and construct individualized contraindication images. In addition, the traditional audit does not consider the dynamic changes of drug interactions and patient physiological states, which increases the risk of drug conflicts, and intelligent technology is needed to realize full data fusion and real-time risk warning.

[0003] Existing drug use auditing systems mostly focus on the static safety verification of single medical orders, and lack the ability to optimize complex patient multi-specialty treatment. When a patient receives multiple department treatments such as cardiovascular and nephrology, the metabolic pathways of different drugs and physiological goals may conflict, but traditional methods cannot simulate the changes of physiological indicators under the combined action of multiple medical orders. In addition, the influence of individual differences on drug metabolism is not quantified, resulting in dose adjustment relying on experience and trial and error. Although digital twin technology is widely used in the industrial field, it has not been realized in the medical field based on physiological simulation of dynamic drug use deduction, and an intelligent auditing platform needs to be developed to simulate the drug metabolism process and predict long-term efficacy. SUMMARY

[0004] The present application provides a full medical order rational drug use auditing method and platform based on artificial intelligence, which provides doctors with scientific drug use reference, significantly improves the accuracy and safety of drug treatment, and has high clinical application value.

[0005] The present application provides a full medical order rational drug use auditing method based on artificial intelligence, which comprises:

[0006] S1, obtaining structured data and unstructured text of a patient, real-time preprocessing according to time sequence to determine the physiological safety range of the patient, and constructing a full medical panorama image of the patient;

[0007] S2, deep information extraction is performed on the unstructured text, the structured data and the unstructured extraction results are fused, and a full medical panorama image of the patient is constructed;

[0008] S3, each drug in the medical order is compared with each item in the patient's panoramic image for contraindication, and when an absolute contraindication is found, a hard blocking warning is triggered to prevent the medical order from being issued and the conflict evidence is clearly presented;

[0009] S4, an initialized digital twin model is established for the patient, the current medical order obtained through contraindication comparison is set as the baseline scheme of the digital twin simulation, and the physiological safety range of the patient is defined as the hard constraint boundary of the model; wherein the initialized digital twin model specifically comprises: S41, adding pharmacokinetic and pharmacodynamic models in the digital twin model, collecting treatment targets set by doctors according to the patient's disease, and translating them into specific constraint conditions for physiological index characteristic parameters in the digital twin model; S42, when multiple specialist medical orders are injected into the digital twin model, the multiple specialist medical order scheme is executed on the patient's digital twin model to simulate the change process of the patient's physiological index after the implementation of the multiple specialist medical order; S43, observing the change trajectory of each physiological index characteristic parameter of the model after the simulation is executed, and judging whether the simulated parameter trajectory completely satisfies the hard constraint boundary of the model; if not, it is confirmed that there is a conflict; S44, after the conflict is confirmed, a multi-objective optimization algorithm is started to find a dose combination that maximizes the global benefit function of the patient; S45, according to the results of the collaborative optimization, each specialist determines the final medication scheme based on the simulated change trajectory and dynamically adjusts the patient's individual medication safety baseline; S46, comparing the physiological index characteristic parameters monitored during the patient's medication with the simulated change trajectory, and using the generated difference data to calibrate the patient's digital twin model in reverse;

[0010] Preferably, the collaborative optimization finds a dose combination that maximizes the global benefit function of the patient, which is specifically calculated according to the following formula, wherein represents the achievement degree index of the i th treatment target, is the weight coefficient of the target, reflecting the importance of different treatment targets in the overall treatment scheme; once the conflict is confirmed, a multi-objective optimization algorithm is started to consider the balance between multiple treatment targets and find the optimal solution; the algorithm searches and adjusts in the parameter space of the digital twin model, trying to fine-tune the drug doses in each specialist medical order; after each dose adjustment, the model simulates the change process of the physiological index and calculates the corresponding target achievement degree index;

[0011] S5, within the set safety boundary, fine-tune the drug doses in the baseline scheme, and after each fine-tuning, the digital twin model simulates the change of the patient's physiological index in the next 72 hours under the scheme;

[0012] S6, according to the physiological index dynamic curve simulated by the digital twin model, the doctor confirms the patient's final medical order in combination with clinical experience, and takes the final confirmed physiological index change as the patient's individual medication safety baseline.

[0013] Preferably, the construction of the patient panoramic profile includes: basic patient information, a dynamic clinical panoramic profile, and a panoramic contraindication profile for medical visits. The panoramic contraindication profile for medical visits involves standardized text sequence processing of structured data and unstructured text obtained from the hospital system. For physiological indicators determined by normal reference ranges in the test results of the standardized text sequence, the normal reference range is directly used as the physiological safety range. For patients with special disease states, the physiological safety range of relevant indicators is adjusted according to the test results. Contraindications for drugs are determined by combining drug instructions and clinical guidelines. The patient's physiological indicators are compared with the contraindications, and the contraindication relationship between the drug and the patient is recorded. Keyword searches are performed on the standardized text sequence to find contraindication-related statements and record them in the panoramic profile for medical visits. The results of structured data contraindication analysis and unstructured text contraindication mining are integrated to construct a panoramic contraindication profile for medical visits centered on the patient and including basic patient information, contraindicated drugs, and contraindicated disease state information.

[0014] Preferably, defining the patient's physiological safety range as the hard constraint boundary of the model includes: constructing a multi-scale physiological simulation model framework based on the characteristics of the patient's physiological indicators and medical knowledge, using a system of differential equations to describe the dynamic physiological process; using static attributes in the patient's panoramic profile as initial parameters of the model, and the characteristics of the patient's physiological indicators as time series inputs; converting medical orders that have passed contraindication comparisons into baseline inputs for the digital twin model, and aligning the execution time of the medical orders with the model's time axis; mapping each drug dosage in the medical orders to dosage parameters in the model, with the mapping relationship as follows: ,in, This refers to the dosage of each drug. For standard dose, It is a patient-specific regulatory factor.

[0015] Preferably, in step S42, simulating the changes in the patient's physiological indicators after the implementation of multi-specialty medical orders further includes:

[0016] For various medications in multidisciplinary prescriptions, the system automatically matches their pharmacokinetic parameters and, combined with patient physiological indicators, uses a physiological pharmacokinetic model algorithm to generate a patient-specific drug metabolism simulation prediction model. Counterfactual clinical scenario simulations are then conducted within this model to simulate potential changes in the patient's condition. For each counterfactual clinical scenario, a new drug metabolism simulation prediction is performed to identify vulnerabilities in the multidisciplinary prescription medication regimen. Based on these vulnerabilities, it is determined whether drug dosage adjustments are necessary. Finally, based on the range of dosage adjustments, the permissible deviation thresholds for the patient's medication safety baseline are determined.

[0017] Preferably, the vulnerability point includes: analyzing the vulnerability of the medication regimen based on the simulated drug concentration change curve over time; and setting the effective drug concentration range as... ,in This indicates the minimum effective concentration of the drug. This indicates the maximum range of drug concentrations; when the drug concentration is below... The time indicates poor treatment effect, higher than The system alerts users to potential toxicity risks, thereby identifying counterfactual clinical scenarios that could lead to drug concentrations exceeding the effective range. These scenarios represent the vulnerabilities in the drug administration process.

[0018] Preferably, conducting counterfactual clinical scenario simulations in the generated patient-specific drug metabolism simulation prediction model specifically includes: dividing the patient's treatment process into stages according to treatment indicators and detecting the physiological characteristics of different treatment stages; determining stage goals and metabolic constraints in different treatment stages; designing personalized dose adjustment ranges and frequencies for each treatment stage based on the prediction results of the drug metabolism model; continuously monitoring indicators during treatment stage transitions, and designing a gradual transition plan when a stage transition signal is detected, gradually adjusting to the target dosing regimen of the new treatment stage within 1-2 dosing cycles; setting different effect evaluation indicators for each treatment stage, and readjusting the stage division when monitoring data indicates that the current treatment stage division is no longer appropriate.

[0019] Preferably, the progressive transition scheme includes: selecting key indicators for monitoring treatment phase transitions based on the characteristics of different diseases and treatment stages; setting normal ranges or change thresholds for each monitoring indicator in different treatment stages; determining a phase transition signal when the monitoring indicator exceeds the normal range of the current treatment stage or reaches the preset change threshold; determining 1-2 dosing cycles as a transition period based on the pharmacokinetic characteristics of the drug and the treatment stage, wherein the dosing cycle is determined according to the therapeutic characteristics of the drug; gradually adjusting the drug dosage to the target dosage of the new treatment stage; and gradually adjusting the dosing frequency during the transition period according to the requirements of the new treatment stage.

[0020] An AI-based platform for reviewing rational drug use across all medical orders, applied to an AI-based method for reviewing rational drug use across all medical orders, the platform comprising:

[0021] The data acquisition and preprocessing module is used to acquire patients' structured data and unstructured text, perform real-time preprocessing in chronological order to determine the patient's physiological safety range, and construct a panoramic profile of the patient's medical visits that includes all contraindications.

[0022] The information extraction and fusion module is used to perform deep information extraction from unstructured text, and to fuse structured data with unstructured extraction results to build a comprehensive patient profile.

[0023] The contraindication comparison and warning module is used to compare each drug in the medical order with each item in the patient's panoramic profile for contraindications. When an absolute contraindication is found, a hard block warning is triggered to prevent the medical order from being issued and to clearly present the conflict evidence.

[0024] The digital twin model building module is used to establish an initial digital twin model for patients, set the current medical orders that have been compared with contraindications as the baseline scheme for digital twin simulation, and define the patient's physiological safety range as the hard constraint boundary of the model.

[0025] The dosage adjustment and simulation module is used to fine-tune the drug dosage in the baseline protocol within the set safety boundaries. After each fine-tuning, the changes in the patient's physiological indicators in the next 72 hours under the protocol are simulated through a digital twin model.

[0026] The prescription confirmation and baseline setting module is used to confirm the patient's final prescription based on the dynamic curve of physiological indicators simulated by the digital twin model, combined with the doctor's clinical experience, and to use the final confirmed changes in physiological indicators as the patient's personal medication safety baseline.

[0027] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0028] By comprehensively integrating structured and unstructured patient data, a panoramic profile of the patient's medical journey is constructed. Utilizing deep information extraction and digital twin models, precise simulation and prediction of patients' physiological indicators are achieved. This approach effectively identifies drug contraindications, avoids medication conflicts, and ensures treatment safety. Simultaneously, by establishing individualized safety baselines, it provides physicians with scientific medication guidelines, significantly improving the accuracy and safety of treatment, and possesses extremely high clinical application value.

[0029] By integrating pharmacokinetic and pharmacodynamic models into a digital twin model, the changes in patients' physiological indicators under multiple specialist medical orders are accurately simulated. This effectively detects efficacy issues caused by physiological conflicts among multiple medical orders. Furthermore, a multi-objective optimization algorithm is used to collaboratively optimize drug dosages and find the globally optimal medication regimen. Based on this, specialists determine medication and dynamically adjust their individual medication safety baselines. Finally, the model is calibrated using real data, significantly improving the accuracy, safety, and overall efficacy of treatment for complex patients with multiple medical orders.

[0030] By automatically matching drug pharmacokinetic parameters and combining them with patient physiological indicators, a personalized drug metabolism simulation and prediction model is generated using a physiological pharmacokinetic model. Counterfactual clinical scenario simulations are then conducted to identify vulnerabilities in the medication regimen. This allows for adjustments to drug dosage and determination of the threshold that can be deviated from the medication safety baseline. This approach can effectively improve the safety and effectiveness of medication regimens, provide precise evidence for personalized medication, and reduce adverse drug reactions and poor treatment outcomes.

[0031] By dividing treatment into stages based on therapeutic indicators and detecting the physiological characteristics of each stage, and combining this with factors such as the patient's liver and kidney function and age to determine the metabolic constraints at different stages, a personalized dosage adjustment plan is designed based on a drug metabolism model. Stage transition indicators are monitored, and a gradual transition plan is designed. Simultaneously, efficacy evaluation indicators for different stages are set. This plan can more accurately match the patient's treatment progress, dynamically adjust medication, improve treatment effectiveness and safety, reduce adverse drug reactions, and provide strong support for personalized medicine. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating an artificial intelligence-based method for reviewing rational drug use across all medical orders, according to an embodiment of the present invention.

[0033] Figure 2 This is a structural block diagram of an AI-based platform for reviewing rational drug use across all medical orders, according to an embodiment of the present invention. Detailed Implementation

[0034] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0035] It should be noted that the terms "vertical," "horizontal," "up," "down," "left," "right," and similar expressions used in this article are for illustrative purposes only and do not represent the only possible implementation.

[0036] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0037] Example 1: Figure 1 This is a flowchart illustrating an artificial intelligence-based method for reviewing rational drug use across all medical orders, according to an embodiment of the present invention.

[0038] For example Figure 1 As shown, an artificial intelligence-based method for auditing the rational use of all medical orders includes the following steps:

[0039] S1. Obtain the structured data and unstructured text of the patient, perform real-time preprocessing in chronological order to determine the physiological safety range of the patient, and simultaneously construct a panoramic taboo portrait of the patient's medical treatment.

[0040] Specifically, obtain the patient's basic information (age, name, allergy history, past medical history, etc.), diagnosis codes, medical order records, and test results from the hospital system as structured data; regard free texts such as admission records, progress logs, consultation opinions, and descriptive texts of imaging reports as unstructured text.

[0041] Perform word segmentation on the unstructured text, segment the continuous text into single words and perform word-tagging to obtain a sequence of words S with word tags. Define a stop word set (such as "de", "le"), remove the words belonging to the stop word set and irrelevant symbols such as punctuation and spaces from the sequence S to obtain the processed text sequence S'.

[0042] Establish a medical term mapping table Map (such as Map = {"heart attack" → "acute myocardial infarction", "high blood sugar" → "hyperglycemia"}), traverse the words in S', if the word is in the key set of Map, then replace it with the corresponding value to obtain the standardized text sequence S''.

[0043] For the physiological indicators determined by the normal reference range in the test results (such as blood pressure, heart rate, body temperature, blood sugar, respiratory rate, metabolic conditions, etc.), directly use the normal reference range as the physiological safety range; for patients with special disease states, adjust the physiological safety range of relevant indicators according to the test results. Starting from the drug information in the medical order record, combine the drug instructions and clinical guidelines to determine the taboo conditions of the drugs. Compare the patient's physiological indicators with the taboo conditions, record the taboos of the drugs and the patient, perform keyword search on S'', find the expressions related to the taboos and record them into the panoramic portrait of the medical treatment. Integrate the results of structured data taboo analysis and unstructured text taboo mining, and construct a panoramic taboo portrait of the patient-centered, including the patient's basic information, taboo drugs, taboo disease states, etc.

[0044] S2. Perform deep information extraction on the unstructured text, fuse the structured data with the unstructured extraction results, and construct a panoramic portrait of the patient.

[0045] The extraction process involves combining word segmentation, part-of-speech tagging, and named entity recognition (NER) of unstructured text such as electronic medical records and examination reports to extract medical entities such as diseases, symptoms, drugs, and examinations. Continuous text is split into independent lexical units and tagged with parts of speech. A pre-trained model (BERT + CRF) is used to identify medical entities in the text and label their categories. The entity recognition confidence threshold is set at 0.7-0.9, retaining entities whose predicted probabilities are higher than the threshold. Relationships between entities (e.g., "disease causes symptoms") and entity attributes (e.g., "symptom severity") are extracted from the preprocessed text. Entity pair relationships are identified using dependency parsing or graph neural networks (GNNs). The relationship prediction confidence threshold is set at 0.6-0.8, and the attribute value confidence threshold is set at 0.7-0.9, filtering out low-confidence results.

[0046] This approach maps unstructured extraction results to structured data within a unified semantic framework, resolving conflicts between multi-source data. It defines mapping rules between entities and values ​​(e.g., mapping the text "hemoglobin 120 g / L" to the structured field "HEMOGLOBIN = 120"). If the same indicator has multiple source values, a weighted average is used based on the reliability of the data source. The fused data is organized into a graph structure, forming a dynamic clinical panorama. Longitudinal data is sorted by time, and missing values ​​are filled using linear interpolation or LSTM models to analyze indicator trends. Entities such as patients, diseases, and drugs are constructed as heterogeneous graphs, with nodes representing entities and edges representing relationships. Entity association features are learned using the GraphSAGE algorithm.

[0047] The patient panoramic profile includes basic patient information, dynamic clinical panoramic profile, and panoramic profile of contraindications for seeking medical treatment.

[0048] S3 compares each medication in the prescription with each item in the patient's panoramic profile for contraindications. When an absolute contraindication is found, a hard-block warning is triggered to prevent the prescription from being issued and to clearly present conflict evidence.

[0049] Specifically, authoritative medical resources (such as drug instructions, clinical guidelines, and the DrugBank database) are integrated to construct a rule base for the association between drugs and contraindications.

[0050] Contraindication-related items are extracted from the patient's comprehensive profile to form a patient contraindication set.

[0051] For example, the system filters contraindication-related entities (such as "gastric ulcer" or "penicillin allergy") from entries in the patient profile, including "medical history," "allergy history," and "surgical history." Contraindication expressions are standardized to standard medical terminology (e.g., mapping "gastric bleeding" to "upper gastrointestinal bleeding") and matched against knowledge base rules. If an entry in the patient profile matches a drug contraindication (e.g., the patient has a history of "gastric ulcer" and the prescription includes aspirin), it is marked as a potential conflict.

[0052] The system iterates through each medication in the prescription, checking for absolute contraindications (by iterating through all medications in the prescription, querying the knowledge base for their associated absolute contraindication sets, and checking if the patient's contraindication set intersects with the absolute contraindication set). If an intersection exists, it is considered an "absolute contraindication conflict," triggering a hard-block warning. When an absolute contraindication is detected, the prescription is prevented from being issued, and conflict evidence is clearly displayed (the conflicting drug name is clearly marked, the matching contraindication item in the patient's profile is listed, a prominent warning box pops up on the prescription issuance page, requiring confirmation or modification of the prescription, and providing a link to the contraindication details). Doctors are allowed to override warnings in special circumstances (requiring secondary confirmation), and all conflict events are recorded for quality review. Doctors must enter a reason for overriding; the system records the operation time, doctor ID, and reason. Conflict event reports are generated regularly, analyzing frequently conflicting drugs and contraindication types to optimize the association rule base.

[0053] S4 establishes an initial digital twin model for the patient, sets the current medical orders that have passed the contraindication comparison as the baseline scheme for the digital twin simulation, and defines the patient's physiological safety range as the hard constraint boundary of the model.

[0054] Based on the characteristics of patient physiological indicators (such as blood pressure trends, heart rate changes, drug metabolism capacity, and temperature variation) and medical knowledge, a multi-scale physiological simulation model framework is constructed using a system of differential equations to describe dynamic physiological processes. Static attributes from the patient's comprehensive profile (such as age, gender, and underlying diseases) are used as initial parameters of the model, while the characteristics of the patient's physiological indicators are used as time-series inputs.

[0055] The medical orders verified through contraindication comparison are transformed into baseline inputs for the digital twin model, aligning the order execution time with the model's timeline. Each drug dosage in the medical order is mapped to a dosage parameter in the model, with the following mapping relationship:

[0056]

[0057] in, This refers to the dosage of each drug. For standard dose, The patient-specific adjustment factor is the difference between the adjusted drug dose and the normal drug dose determined based on the patient's overall profile. For example, if the dose is reduced by 5 mg (0.005 g) in patients with renal insufficiency, then the patient-specific adjustment factor = 0.005.

[0058] S5, within the set safety boundaries, fine-tunes the drug dosage in the baseline protocol. After each fine-tuning, the digital twin model simulates the changes in the patient's physiological indicators over the next 72 hours under this protocol.

[0059] Specifically, the safe range for drug dosage adjustment is determined based on the drug's pharmacological properties, clinical trial data, and past patient treatment experience. For example, for a certain antihypertensive drug, the lower limit of the safe dose is 20 mg per day, and the upper limit is 100 mg per day, i.e., the safety boundary is [20 mg, 100 mg].

[0060] Based on the patient's comprehensive profile, confirm the initial drug dosage of the current treatment plan and determine the range of drug dosage adjustments.

[0061] The fine-tuned drug dosage is input into the digital twin model to simulate the changes in the patient's physiological indicators over the next 72 hours, including blood pressure (BP(t)) and heart rate (HR(t)), where t represents time and t∈[0,72h].

[0062] Set threshold ranges for physiological indicators (e.g., normal blood pressure range is 90-139 mmHg systolic and 60-89 mmHg diastolic; normal heart rate range is 60-100 beats / minute), and check whether the physiological indicators at each time point within the simulated 72-hour period are all within the threshold range. If all indicators are within the range, the fine-tuning plan can be considered for further application; if any indicator exceeds the range, the drug dosage needs to be readjusted.

[0063] S6. Based on the dynamic curves of physiological indicators simulated by the digital twin model, doctors combine clinical experience to confirm the patient's final medical orders and use the final confirmed changes in physiological indicators as the patient's personal medication safety baseline.

[0064] Specifically, doctors identify key indicator fluctuation patterns based on the 72-hour dynamic curves of physiological indicators (such as blood pressure-time curves and heart rate-time curves) output by the digital twin model, combined with the patient's past medical history, real-time symptoms, and clinical experience.

[0065] For example, if the simulated blood pressure value shows a peak fluctuation 6 hours after medication, it is necessary to assess whether the fluctuation is within the patient's individual tolerance threshold.

[0066] Based on the data range of the simulated physiological indicator dynamic curve, the fluctuation range of the patient's physiological indicators is determined. Changes in physiological indicators within this fluctuation range constitute an individualized safety range. The physician selects the drug dosage within the safety boundaries to ensure that the simulated physiological indicators remain within the individualized safety range. The final confirmed 72-hour dynamic dataset within the individualized safety range of physiological indicators is defined as the patient-specific safety baseline. The data in the baseline must include the indicator values ​​and fluctuation ranges at key time points.

[0067] It should be noted that the functional indicator parameters in this application are only illustrative examples, and the specific functional indicator parameters need to be limited according to the application scenario; the formulas involving multi-parameter calculations, such as mapping relationships and parameter calculations in this application, are all calculated after normalized data processing.

[0068] The technical solutions described in the embodiments of this application above have at least the following technical effects or advantages:

[0069] By comprehensively integrating structured and unstructured patient data, a panoramic profile of the patient's medical journey is constructed. Utilizing deep information extraction and digital twin models, precise simulation and prediction of patients' physiological indicators are achieved. This approach effectively identifies drug contraindications, avoids medication conflicts, and ensures treatment safety. Simultaneously, by establishing individualized safety baselines, it provides physicians with scientific medication guidelines, significantly improving the accuracy and safety of treatment, and possesses extremely high clinical application value.

[0070] Example 2: In Example 1, although the system can construct a digital twin model based on a single patient's prescription and determine an individualized safety baseline, ensuring the safety of a single prescription at a static level, this solution falls short when faced with the reality of multiple doctors prescribing multiple prescriptions for the same complex patient. Because the treatment goals of different departments may inherently conflict at the physiological level, the safety assurance of a single prescription cannot collaboratively address the overall efficacy issues among multiple prescriptions. To comprehensively consider the treatment goals of multiple specialties, accurately coordinate the relationships between various prescriptions, achieve the globally optimal medication plan, and dynamically adjust the individual medication safety baseline, this solution is needed.

[0071] Therefore, the embodiments of this application are optimized based on the above embodiments.

[0072] In some embodiments, step S4, establishing an initial digital twin model for the patient, further includes:

[0073] S41. Add pharmacokinetic and pharmacodynamic models to the digital twin model, collect treatment goals set by various specialists according to the patient's symptoms, and translate them into specific constraints on the physiological characteristic parameters of the patient in the digital twin model.

[0074] S42, When multi-specialty medical orders are injected into the digital twin model, the multi-specialty medical order plan is executed on the patient's digital twin model to simulate the changes in the patient's physiological indicators after the implementation of multi-specialty medical orders.

[0075] The model calculates the values ​​of physiological indicators such as blood pressure, heart rate, and blood sugar at different time points based on the drug information in the doctor's order, the patient's own physiological indicators, and the relationship between pharmacokinetics and pharmacodynamics.

[0076] S43. Observe the change trajectory of various physiological indicator characteristic parameters of the model after the simulation is performed, and determine whether the simulated parameter trajectory fully meets the hard constraint boundary of the model; if it cannot be fully met, it is confirmed that there is a conflict.

[0077] S44. After conflict confirmation, a multi-objective optimization algorithm is initiated to collaboratively optimize and find the dose combination that maximizes the patient's global benefit function.

[0078] Specifically, once a conflict is confirmed, a multi-objective optimization algorithm is immediately initiated. This algorithm aims to comprehensively consider the balance between multiple treatment objectives and find the optimal solution.

[0079] The algorithm searches and adjusts within the parameter space of the digital twin model, attempting to fine-tune the drug dosages in various specialty prescriptions. After each dosage adjustment, the model re-simulates the changes in physiological indicators and calculates the corresponding target achievement indicators.

[0080] Define a global benefit function that comprehensively considers the achievement of various treatment objectives. For example, the global benefit function can be expressed as: ,in The indicator representing the achievement rate of the i-th treatment goal. These are the weighting coefficients for the target, reflecting the importance of different treatment targets in the overall treatment plan. The algorithm aims to find the drug dosage combination that maximizes the global benefit function F.

[0081] S45. Based on the results of collaborative optimization, each specialist doctor determines the final medication plan based on the simulated change trajectory and dynamically adjusts the patient's personal medication safety baseline.

[0082] For example, if optimization results show that adjusting the dosage of a certain drug can reduce the risk of drug side effects while meeting the main treatment goals, doctors can determine the final drug dosage and administration method accordingly.

[0083] S46 compares the physiological indicators and characteristic parameters monitored during the patient's medication period with the simulated and predicted change trajectory, and uses the resulting difference data to calibrate the patient's digital twin model in reverse.

[0084] The technical solutions described in the embodiments of this application above have at least the following technical effects or advantages:

[0085] By integrating pharmacokinetic and pharmacodynamic models into a digital twin model, the changes in patients' physiological indicators under multiple specialist medical orders are accurately simulated. This effectively detects efficacy issues caused by physiological conflicts among multiple medical orders. Furthermore, a multi-objective optimization algorithm is used to collaboratively optimize drug dosages and find the globally optimal medication regimen. Based on this, specialists determine medication and dynamically adjust their individual medication safety baselines. Finally, the model is calibrated using real data, significantly improving the accuracy, safety, and overall efficacy of treatment for complex patients with multiple medical orders.

[0086] Example 3: While Example 2 analyzed the medication regimen based on certain rules, it did not fully consider individual patient physiological differences and the complex drug metabolism changes under various counterfactual clinical scenarios. Different patients have different physiological indicators, and the impact of organ function changes, pathophysiological changes, and drug interactions on drug metabolism varies. To more accurately simulate the drug metabolism process in patients, identify the vulnerabilities of the medication regimen under different conditions, and thus formulate a more patient-specific medication strategy, Example 3 underwent the following optimizations and improvements.

[0087] Therefore, the embodiments of this application are optimized based on the above embodiments.

[0088] In some embodiments, step S42, simulating the changes in the patient's physiological indicators after the implementation of multi-specialty medical orders, further includes:

[0089] S421 automatically matches the pharmacokinetic parameters of various drugs in multi-specialty prescriptions, combines them with the patient's physiological indicators, and uses physiological pharmacokinetic model algorithms to generate a patient-specific drug metabolism simulation prediction model.

[0090] The pharmacokinetic parameters include absorption rate constant, volume of distribution, elimination rate constant, plasma protein binding rate, metabolic pathway (e.g., which cytochrome P450 enzyme metabolizes it), and bioavailability. Using the physiological pharmacokinetic model (PBPK) algorithm, combined with drug pharmacokinetic parameters and patient physiological characteristics, a patient-specific drug metabolism simulation and prediction model is generated.

[0091] The following is a simplified core formula for the PBPK model: Drug concentration calculation in tissues: Assuming the drug distribution in the tissue follows a one-compartment model, the drug concentration in the tissue... The change with time (t) can be expressed as:

[0092]

[0093] Where Q represents tissue blood flow (unit: L / h). This represents the drug concentration in plasma (unit: mg / L). Tissue distribution volume (unit: L). Tissue blood flow can be calculated based on the patient's cardiac output (CO, unit: L / min) and the proportion of cardiac output in that tissue; tissue distribution volume can be estimated based on the patient's weight and tissue characteristics.

[0094] Considering the absorption, distribution, and elimination processes of drugs, the drug concentration in plasma The change over time (t) can be represented by the following differential equation (representing the rate of change of drug concentration in plasma):

[0095]

[0096] Where D is the dosage (unit: mg). Indicates the drug absorption rate. It is the absorption rate constant, which reflects the rate at which a drug enters the systemic circulation from the site of administration (such as the gastrointestinal tract). A higher value indicates faster drug absorption; for example, some oral medications, if they have a higher value... This allows the drug to reach an effective concentration in the blood more quickly. D is the dosage, referring to the total amount of drug administered to the patient, usually measured in milligrams (mg) or grams (g). The dosage directly affects the amount of drug absorbed into the body. t is time, representing the time elapsed since drug administration, and can be measured in hours (h), minutes (min), etc. Volume of distribution is a hypothetical volume, representing the volume of body fluid required for a drug to be uniformly distributed in the body at the same concentration as in plasma. It reflects the distribution characteristics of a drug in the body, and different drugs... The values ​​differ greatly. This indicates the rate of drug elimination. The elimination rate constant represents the rate at which a drug is eliminated from the body. The higher the value, the faster the drug is eliminated from the body. Drug elimination is mainly achieved through two mechanisms: metabolism (such as biotransformation in the liver) and excretion (such as excretion by the kidneys). Plasma drug concentration refers to the concentration of a drug in plasma, usually measured in milligrams per liter (mg / L) or micrograms per milliliter (μg / mL). Plasma drug concentration reflects the amount of drug circulating in the blood and is an important indicator of the presence of a drug in the body.

[0097] S422, conduct counterfactual clinical scenario simulations in the generated patient-specific drug metabolism simulation prediction model to simulate possible changes in the patient's condition.

[0098] The counterfactual clinical scenario simulates changes in patient organ function (considering abnormal changes in the function of vital organs such as the liver and kidneys. For example, impaired liver function may lead to reduced activity of drug-metabolizing enzymes, and reduced kidney function may affect drug excretion), changes in pathophysiological indicators (simulating changes in patient pathophysiological indicators, such as infection, fever, shock, etc. These changes in state can affect the absorption, distribution, metabolism, and excretion of drugs), and drug interactions (considering drug interactions caused by adding or discontinuing drugs in multidisciplinary medication settings. Drug interactions may alter drug pharmacokinetic parameters, thereby affecting drug efficacy and safety).

[0099] Specifically, parameters set in scenarios involving changes in organ function, pathophysiological indicators, and drug interactions are input into a patient-specific drug metabolism simulation and prediction model to obtain time-varying drug concentration curves in tissues and plasma under different counterfactual clinical scenarios. Based on these simulated drug concentration-time curves, potential changes in the patient's condition under different counterfactual clinical scenarios are analyzed. For example, excessively high drug concentrations may lead to drug toxicity; excessively low drug concentrations may fail to achieve the desired therapeutic effect.

[0100] S423 involves re-conducting drug metabolism simulation predictions for each counterfactual clinical scenario to identify vulnerabilities in multidisciplinary medication regimens.

[0101] Specifically, the vulnerability of the medication regimen is analyzed based on the simulated drug concentration change curve over time. The effective drug concentration range is set as follows: ( and The specific value is determined based on the drug characteristics and clinical treatment requirements. This indicates the minimum effective concentration of the drug. This indicates the maximum range of drug concentrations; when the drug concentration is below... The time indicates poor treatment effect, higher than The system alerts users to potential toxicity risks, thereby identifying counterfactual clinical scenarios that could lead to drug concentrations exceeding the effective range. These scenarios represent the vulnerabilities in the drug administration process.

[0102] S424 determines whether drug dosage needs to be adjusted based on the vulnerabilities of the medication regimen.

[0103] Specifically, the analysis examines the impact of identified vulnerabilities on drug efficacy and safety. If a vulnerability leads to drug concentrations remaining below a certain level for an extended period... This could seriously affect the treatment outcome; if it leads to a drug concentration that is consistently higher than [a certain level] for an extended period... This could potentially trigger severe toxic reactions.

[0104] The need to adjust the drug dosage is determined based on the degree of impact. If the impact is significant, meaning the drug concentration frequently or for extended periods exceeds the effective concentration range, then the drug dosage needs to be adjusted. If the impact is minor, and the drug concentration only slightly exceeds the effective range for a short period, observation can be conducted first, and the dosage can be left unchanged.

[0105] S425, based on the range of drug dosage adjustments, determines the acceptable deviation thresholds from the patient's baseline medication safety for certain medications.

[0106] Specifically, the range of drug dosage adjustments is determined by considering factors such as drug characteristics (e.g., therapeutic window width, dose-response relationship), individual patient differences (e.g., age, weight, liver and kidney function), and clinical treatment goals. For example, the dosage adjustment range may be narrower for drugs with a narrower therapeutic window and wider for drugs with a wider therapeutic window.

[0107] Based on the dose adjustment range, and considering the dose-effect relationship and toxicity-dose relationship of drugs, we determine the thresholds that some drugs may deviate from the baseline of medication safety. Let the lower limit of the dose adjustment range be... The upper limit is The safe baseline dose is The allowable deviation threshold can be set as a certain percentage of the safe baseline dose, such as ±θ (the value of θ can be determined according to the drug characteristics, generally 10% - 30%). That is, the lower limit of the allowable deviation threshold is... The upper limit can deviate from the threshold by: And it is necessary to ensure that the adjusted dose is within Within the range.

[0108] The technical solutions described in the embodiments of this application above have at least the following technical effects or advantages:

[0109] By automatically matching drug pharmacokinetic parameters and combining them with patient physiological indicators, a personalized drug metabolism simulation and prediction model is generated using a physiological pharmacokinetic model. Counterfactual clinical scenario simulations are then conducted to identify vulnerabilities in the medication regimen. This allows for adjustments to drug dosage and determination of the threshold that can be deviated from the medication safety baseline. This approach can effectively improve the safety and effectiveness of medication regimens, provide precise evidence for personalized medication, and reduce adverse drug reactions and poor treatment outcomes.

[0110] Example 4: While Example 3 could generate patient-specific drug metabolism simulation prediction models for multi-specialty prescriptions, conduct counterfactual clinical scenario simulations, and identify vulnerabilities in medication regimens, thereby adjusting drug dosages and determining the acceptable deviation thresholds from the medication safety baseline, this approach focused on theoretical simulation and adjustments based on simulation results. It lacked in-depth consideration of the dynamic changes during the patient's actual treatment process and did not fully integrate the physiological characteristics and metabolic changes of patients at different treatment stages. To achieve more precise personalized medication and improve treatment efficacy and safety, Example 4 proposes a dynamic medication adjustment scheme based on treatment stage divisions.

[0111] Therefore, the embodiments of this application are optimized based on the above embodiments.

[0112] In some embodiments, step S422, conducting counterfactual clinical scenario simulations in the generated patient-specific drug metabolism simulation prediction model, further includes:

[0113] 2A, the patient's treatment process is divided into stages according to treatment indicators, and the physiological characteristics of different treatment stages are detected.

[0114] Specifically, key treatment indicators are determined based on the patient's disease type and the established treatment plan. For example, treatment indicators for cardiovascular disease patients may include heart rate, blood pressure, and blood lipid levels; important treatment indicators for diabetic patients include blood glucose and glycated hemoglobin. These indicators can directly reflect the treatment effect and changes in the patient's condition.

[0115] Based on clinical experience and relevant disease treatment guidelines, the entire treatment process for patients is initially divided into different stages, commonly including the acute phase, subacute phase, and recovery phase. For example, the acute phase for fracture patients may be the first 1-2 weeks after the fracture, with the main goal of reducing pain and controlling inflammation; the subacute phase may be 2-6 weeks, focusing on promoting fracture healing; and the recovery phase is after 6 weeks, dedicated to restoring joint function and muscle strength.

[0116] During treatment, the patient's physiological indicators are continuously monitored. The monitoring frequency varies for different indicators. Vital signs such as heart rate and blood pressure may need to be monitored in real time or multiple times a day; while indicators such as blood lipids and glycated hemoglobin are monitored relatively less frequently, possibly once every few weeks or months.

[0117] Analyze the physiological characteristics at different stages. In the acute phase, physiological indicators typically change drastically. Taking infectious diseases as an example, patients in the acute phase may experience a significant increase in body temperature and white blood cell count, accompanied by changes in vital signs such as increased heart rate and rapid breathing. This is a stress response of the body to pathogens. In the subacute phase, changes in physiological indicators tend to moderate. As the infection is brought under control, body temperature gradually returns to normal, and the white blood cell count begins to decrease, although it may still be above the normal range. At this time, the body's repair mechanisms begin to function. In the recovery phase, physiological indicators gradually approach the normal range, the patient's vital signs stabilize, and various laboratory test indicators also basically return to normal, indicating that bodily functions are gradually recovering.

[0118] 2B, Determine stage goals and metabolic constraints at different treatment phases.

[0119] In terms of metabolic constraints, drug metabolism is considered in relation to the patient's liver and kidney function. For patients with hepatic impairment, strict dosage control is crucial during the acute phase when using hepatically metabolized drugs to prevent drug accumulation and toxicity. The degree of liver dysfunction (mild, moderate, or severe) can be determined based on the specific range of liver function indicators such as alanine aminotransferase (ALT) and aspartate aminotransferase (AST), referring to clinical standards. A liver function adjustment coefficient (0.7-0.8 for mild abnormalities, 0.5-0.6 for moderate abnormalities, and assessment for severe abnormalities based on specific circumstances) is then established. The adjusted dose is calculated using the formula: Adjusted dose = Normal dose × Liver function adjustment coefficient. For patients with renal impairment, the dosage of drugs excreted through the kidneys is adjusted based on the glomerular filtration rate (GFR). A normal GFR is defined as 120 ml / min. The adjusted dose is calculated using the formula: Adjusted dose = Normal dose × (Patient GFR / 120). When GFR decreases, the drug dosage is reduced accordingly.

[0120] In terms of physiological metabolism, factors such as the patient's age, gender, and basal metabolic rate should be considered. Elderly patients have a lower basal metabolic rate, resulting in slower drug metabolism and excretion. Therefore, more caution should be exercised when administering medication at each stage of treatment to avoid drug accumulation in the body. For example, the dosage of sedative-hypnotic drugs in elderly patients during the stable and recovery phases should be appropriately reduced compared to younger patients. Simultaneously, attention should be paid to the patient's nutritional and metabolic status to ensure adequate nutritional support during treatment and to maintain normal metabolic function.

[0121] 2C, based on the prediction results of the drug metabolism model, designs personalized dose adjustment range and frequency for each treatment stage.

[0122] Specifically, using a pre-constructed patient drug metabolism model, and inputting the patient's physiological parameters and current treatment stage, the model simulates the drug's metabolic process in the patient's body, predicting key indicators such as blood drug concentration and metabolic rate at different time points. Because a rapid achievement of effective therapeutic concentrations is crucial during the acute phase, dosage adjustments may be significant. Based on the drug metabolism model's predictions, if the prediction shows a slow increase in blood drug concentration at the current dose, failing to reach an effective concentration within the expected timeframe, a substantial increase in dose can be considered. During the stable phase, the focus is on maintaining stable blood drug concentrations, with relatively smaller dosage adjustments. If the model predicts a blood drug concentration within the effective range but close to the lower limit, a small increase in dose can be made. During the recovery phase, considering the need to gradually reduce drug dependence, dosage adjustments primarily involve gradual reduction.

[0123] 2D continuously monitors indicators during treatment phase transitions, but once a phase transition signal is detected, a gradual transition plan is designed to gradually adjust to the target dosing regimen for the new treatment phase within 1-2 dosing cycles.

[0124] Specifically, key indicators are selected for monitoring the transition between different stages of treatment, based on the characteristics of different diseases and treatment phases. During treatment, the selected indicators are continuously measured at regular time intervals. The monitoring frequency can be adjusted according to the severity of the condition and the treatment phase; more frequent monitoring may be needed during the acute phase, such as multiple times per hour or per day; the monitoring frequency can be appropriately reduced during the stable phase, such as once per day or every few days.

[0125] Set normal ranges or threshold values ​​for each monitoring indicator at different treatment stages. When a monitoring indicator exceeds the normal range for the current treatment stage or reaches the preset threshold value and persists for a certain period of time, a stage transition signal is identified. For example, in the acute phase of an infectious disease, if the body temperature returns to normal for three consecutive days, and the white blood cell count and C-reactive protein also continue to decrease to near the normal range, this can be considered a signal of transition to the stable phase.

[0126] Based on the pharmacokinetic characteristics of the drug and the stage of treatment, determine 1-2 dosing cycles (the number of dosing cycles is determined according to the drug's therapeutic characteristics) as a transition period. During the transition period, gradually adjust the drug dosage to the target dosage for the new treatment stage. A linear adjustment method can be used, that is, adjusting the dosage proportionally for each dosing cycle. According to the requirements of the new treatment stage, gradually adjust the dosing frequency during the transition period. For example, if changing from multiple daily doses in the acute phase to once daily doses in the stable phase, the dosing frequency can be reduced first during the transition period, such as gradually reducing from 4 times a day to 2 times a day, and then transitioning to once a day.

[0127] 2E sets different efficacy evaluation indicators for each treatment stage, and readjusts the stage division when monitoring data indicates that the current treatment stage division is no longer appropriate.

[0128] The technical solutions described in the embodiments of this application above have at least the following technical effects or advantages:

[0129] By dividing treatment into stages based on therapeutic indicators and detecting the physiological characteristics of each stage, and combining this with factors such as the patient's liver and kidney function and age to determine the metabolic constraints at different stages, a personalized dosage adjustment plan is designed based on a drug metabolism model. Stage transition indicators are monitored, and a gradual transition plan is designed. Simultaneously, efficacy evaluation indicators for different stages are set. This plan can more accurately match the patient's treatment progress, dynamically adjust medication, improve treatment effectiveness and safety, reduce adverse drug reactions, and provide strong support for personalized medicine.

[0130] Furthermore, this embodiment of the invention also provides an artificial intelligence-based platform for reviewing rational drug use across all medical orders.

[0131] Figure 2 This is a schematic diagram of the structure of an artificial intelligence-based rational drug use review platform for all medical orders according to an embodiment of the present invention.

[0132] For example Figure 2 As shown, an AI-based platform for reviewing rational drug use across all medical orders includes: a data acquisition and preprocessing module, an information extraction and fusion module, a contraindication comparison and warning module, a digital twin model construction module, a dosage adjustment and simulation module, and a medical order confirmation and baseline setting module.

[0133] The data acquisition and preprocessing module is used to acquire patients' structured data and unstructured text, perform real-time preprocessing in chronological order to determine the patient's physiological safety range, and construct a panoramic profile of the patient's medical visits that includes all contraindications.

[0134] The information extraction and fusion module is used to perform deep information extraction from unstructured text, and to fuse structured data with unstructured extraction results to build a comprehensive patient profile.

[0135] The contraindication comparison and warning module is used to compare each drug in the medical order with each item in the patient's panoramic profile for contraindications. When an absolute contraindication is found, a hard block warning is triggered to prevent the medical order from being issued and to clearly present the conflict evidence.

[0136] The digital twin model building module is used to establish an initial digital twin model for patients, set the current medical orders that have been compared with contraindications as the baseline scheme for digital twin simulation, and define the patient's physiological safety range as the hard constraint boundary of the model.

[0137] The dosage adjustment and simulation module is used to fine-tune the drug dosage in the baseline protocol within the set safety boundaries. After each fine-tuning, the changes in the patient's physiological indicators in the next 72 hours under the protocol are simulated through a digital twin model.

[0138] The prescription confirmation and baseline setting module is used to confirm the patient's final prescription based on the dynamic curve of physiological indicators simulated by the digital twin model, combined with the doctor's clinical experience, and to use the final confirmed changes in physiological indicators as the patient's personal medication safety baseline.

[0139] It should be noted that other specific implementation details of the AI-based rational drug use review platform for all medical orders in this embodiment of the invention can refer to the above-described AI-based rational drug use review method for all medical orders.

[0140] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for reviewing rational drug use across all medical orders based on artificial intelligence, characterized in that, The method includes: S1: Obtain the patient's structured data and unstructured text, perform real-time preprocessing in chronological order to determine the patient's physiological safety range, and construct a panoramic profile of the patient's medical visits that includes all contraindications. S2 performs deep information extraction on unstructured text, integrates structured data with the unstructured extraction results, and constructs a panoramic profile of the patient. S3 compares each drug in the prescription with each item in the patient's panoramic profile for contraindications. When an absolute contraindication is found, a hard-block warning is triggered to prevent the prescription from being issued and to clearly present conflict evidence. S4. Establish an initial digital twin model for the patient, set the current medical orders that have been compared with contraindications as the baseline scheme for the digital twin simulation, and define the patient's physiological safety range as the hard constraint boundary of the model. The initialization of the digital twin model specifically includes: S41, adding pharmacokinetic and pharmacodynamic models to the digital twin model, collecting treatment goals set by various specialists according to the patient's symptoms, and translating them into specific constraints on the patient's physiological indicator parameters in the digital twin model; S42, when multi-specialty medical orders are injected into the digital twin model, executing the multi-specialty medical order plan on the patient's digital twin model, simulating the changes in the patient's physiological indicators after the implementation of the multi-specialty medical orders; S43, observing the change trajectory of various physiological indicator parameters in the model after the simulation, and determining whether the simulated parameter trajectory fully meets the hard constraint boundary of the model; if it cannot be fully met, a conflict is confirmed; S44, after the conflict is confirmed, a multi-objective optimization algorithm is started to collaboratively optimize and find the dosage combination that maximizes the patient's global benefit function; S45, based on the collaborative optimization results, each specialist determines the final medication plan based on the simulated change trajectory, and dynamically adjusts the patient's personal medication safety baseline; S46, comparing the physiological indicator parameters monitored during the patient's medication period with the simulated predicted change trajectory, and using the generated difference data to back-calibrate the patient's digital twin model; The collaborative optimization finds the dose combination that maximizes the patient's global benefit function, specifically calculated according to the following formula. ,in The indicator representing the achievement rate of the i-th treatment goal. This is the weighting coefficient of the objective, reflecting the importance of different treatment objectives in the overall treatment plan; once a conflict is confirmed, a multi-objective optimization algorithm is immediately activated to consider the balance between multiple treatment objectives and find the optimal solution; the algorithm searches and adjusts within the parameter space of the digital twin model, attempting to fine-tune the drug dosage in each specialist prescription; after each dosage adjustment, the model re-simulates the physiological indicator change process and calculates the corresponding objective achievement index. S5, within the set safety boundaries, fine-tunes the drug dosage in the baseline protocol. After each fine-tuning, the digital twin model simulates the changes in the patient's physiological indicators over the next 72 hours under this protocol. S6. Based on the dynamic curves of physiological indicators simulated by the digital twin model, doctors combine clinical experience to confirm the patient's final medical orders and use the final confirmed changes in physiological indicators as the patient's personal medication safety baseline.

2. The method for reviewing rational drug use across all medical orders based on artificial intelligence as described in claim 1, characterized in that, The construction of the patient panoramic profile includes: basic patient information, dynamic clinical panoramic profile, and panoramic contraindication profile. The panoramic contraindication profile is obtained from the hospital system through standardized text sequence processing of structured data and unstructured text. For physiological indicators determined by normal reference ranges in the test results of the standardized text sequence, the normal reference range is directly used as the physiological safety range. For patients with special disease states, the physiological safety range of relevant indicators is adjusted according to the test results. Contraindications of drugs are determined by combining drug instructions and clinical guidelines. The patient's physiological indicators are compared with the contraindications to record the contraindication relationship between the drug and the patient. Keyword search is performed on the standardized text sequence to find contraindication-related statements and record them in the panoramic contraindication profile. The results of structured data contraindication analysis and unstructured text contraindication mining are integrated to construct a patient-centered panoramic contraindication profile that includes basic patient information, contraindicated drugs, and contraindicated disease states.

3. The method for reviewing rational drug use across all medical orders based on artificial intelligence as described in claim 1, characterized in that, The definition of the patient's physiological safety range as the hard constraint boundary of the model includes: constructing a multi-scale physiological simulation model framework based on the characteristics of the patient's physiological indicators and medical knowledge, using a system of differential equations to describe the dynamic physiological process; using the static attributes in the patient's panoramic portrait as the initial parameters of the model, and the characteristics of the patient's physiological indicators as the time series input; converting the medical orders that have passed the contraindication comparison into the baseline input of the digital twin model, and aligning the execution time of the medical orders with the model's time axis; mapping each drug dosage in the medical orders to the dosage parameters in the model, with the mapping relationship as follows: ,in, This refers to the dosage of each drug. For standard dose, It is a patient-specific regulatory factor.

4. The method for reviewing rational drug use across all medical orders based on artificial intelligence as described in claim 1, characterized in that, The process of changes in patients' physiological indicators after the implementation of the simulated multi-specialty medical orders also includes: For various medications in multidisciplinary prescriptions, the system automatically matches their pharmacokinetic parameters and, combined with patient physiological indicators, uses a physiological pharmacokinetic model algorithm to generate a patient-specific drug metabolism simulation prediction model. Counterfactual clinical scenario simulations are then conducted within this model to simulate potential changes in the patient's condition. For each counterfactual clinical scenario, a new drug metabolism simulation prediction is performed to identify vulnerabilities in the multidisciplinary prescription medication regimen. Based on these vulnerabilities, it is determined whether drug dosage adjustments are necessary. Finally, based on the range of dosage adjustments, the permissible deviation thresholds for the patient's medication safety baseline are determined.

5. The method for reviewing rational drug use across all medical orders based on artificial intelligence as described in claim 4, characterized in that, The vulnerabilities include: analyzing the vulnerabilities of the medication regimen based on the simulated drug concentration-time curve; and setting the effective drug concentration range as... ,in This indicates the minimum effective concentration of the drug. This indicates the maximum range of drug concentrations; when the drug concentration is below... The time indicates poor treatment effect, higher than The system alerts users to potential toxicity risks, thereby identifying counterfactual clinical scenarios that could lead to drug concentrations exceeding the effective range. These scenarios represent the vulnerabilities in the drug administration process.

6. The method for reviewing rational drug use across all medical orders based on artificial intelligence as described in claim 4, characterized in that, Conducting counterfactual clinical scenario simulations in the generated patient-specific drug metabolism simulation prediction model specifically includes: dividing the patient's treatment process into stages according to treatment indicators and detecting the physiological characteristics of different treatment stages; determining stage goals and metabolic constraints for different treatment stages; designing personalized dose adjustment ranges and frequencies for each treatment stage based on the prediction results of the drug metabolism model; continuously monitoring indicators during treatment stage transitions, but designing a gradual transition plan after detecting a stage transition signal, gradually adjusting to the target dosing regimen of the new treatment stage within 1-2 dosing cycles; setting different effect evaluation indicators for each treatment stage, and readjusting the stage division when monitoring data indicates that the current treatment stage division is no longer appropriate.

7. The method for reviewing rational drug use across all medical orders based on artificial intelligence as described in claim 6, characterized in that, The proposed progressive transition scheme includes: selecting key indicators for monitoring treatment phase transitions based on the characteristics of different diseases and treatment stages; setting normal ranges or change thresholds for each monitoring indicator at different treatment stages; determining a phase transition signal when a monitoring indicator exceeds the normal range of the current treatment stage or reaches a preset change threshold; determining 1-2 dosing cycles as a transition period based on the pharmacokinetic characteristics of the drug and the treatment stage, wherein the dosing cycle is determined according to the therapeutic characteristics of the drug; gradually adjusting the drug dosage to the target dosage of the new treatment stage; and gradually adjusting the dosing frequency within the transition period according to the requirements of the new treatment stage.

8. An AI-based platform for reviewing rational drug use across all medical orders, applied to the AI-based method for reviewing rational drug use across all medical orders as described in any one of claims 1 to 7, characterized in that, The platform includes: The data acquisition and preprocessing module is used to acquire patients' structured data and unstructured text, perform real-time preprocessing in chronological order to determine the patient's physiological safety range, and construct a panoramic profile of the patient's medical visits that includes all contraindications. The information extraction and fusion module is used to perform deep information extraction from unstructured text, and to fuse structured data with unstructured extraction results to build a comprehensive patient profile. The contraindication comparison and warning module is used to compare each drug in the medical order with each item in the patient's panoramic profile for contraindications. When an absolute contraindication is found, a hard block warning is triggered to prevent the medical order from being issued and to clearly present the conflict evidence. The digital twin model building module is used to establish an initial digital twin model for patients, set the current medical orders that have been compared with contraindications as the baseline scheme for digital twin simulation, and define the patient's physiological safety range as the hard constraint boundary of the model. The dosage adjustment and simulation module is used to fine-tune the drug dosage in the baseline protocol within the set safety boundaries. After each fine-tuning, the changes in the patient's physiological indicators in the next 72 hours under the protocol are simulated through a digital twin model. The prescription confirmation and baseline setting module is used to confirm the patient's final prescription based on the dynamic curve of physiological indicators simulated by the digital twin model, combined with the doctor's clinical experience, and to use the final confirmed changes in physiological indicators as the patient's personal medication safety baseline.

Images