Intelligent Conversation System for Patient Medication Management Driven by Large Language Models

Abstract

The present invention discloses an intelligent conversation system for patient medication management driven by a large model, which relates to the technical field of intelligent medication management. It collects real-time health feedback data of high-risk patients and issues a warning instruction when the data is abnormal; uses the medical large model to identify the relevant data of high-risk patients and gives medical reply information, and aggregates to generate a set of medical reply information; restricts the follow-up frequency of high-risk patients, optimizes the medical reply information received by high-risk patients, and obtains the optimized medical reply information; generates a rehabilitation coefficient from the health risk values of high-risk patients within the observation period. If the rehabilitation coefficient does not exceed the expectation, the medical large model is optimized, and the maintenance frequency is restricted according to the rehabilitation coefficients of different high-risk patients. It actively intervenes when the health status of high-risk patients is abnormal and gives medication reference suggestions to achieve health protection for high-risk patients.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent medication management, and particularly to an intelligent conversation system for patient medication management driven by a large model. Background Art

[0002] Patient medication management is a crucial link in medical care work, aiming to ensure that patients use medications safely, effectively, and rationally. It involves providing clear medication guidance, regularly reminding and supervising patients to take medications, preparing and checking medications, observing patients' reactions after taking medications, as well as storing and managing medications.

[0003] Intelligent patient medication management based on a medical large model is an innovative medical care model that uses advanced medical large model technology to intelligently and individually manage the patient medication process. By integrating and analyzing patients' medical data, this model can provide accurate medication guidance, real-time reminder and supervision of patients taking medications, predict and monitor drug interactions and adverse reactions, thereby ensuring that patients use medications safely, effectively, and rationally, and improving the treatment effect and the quality of life of patients.

[0004] In the Chinese invention patent with the application publication number CN118197533A, a medical medication information management system and method, the system includes a medication information library module, a patient information acquisition module, a patient information template generation module, a patient information library module, a keyword retrieval module, a matching module, a screening module, and a medication guidance book generation module. The medication information library module is used to store medication information, the patient information acquisition module is used to acquire patient information, the patient information template generation module is used to generate patient information templates, the patient information library module is used to store and modify patient information templates, the keyword retrieval module is used to perform keyword retrieval, the matching module is used to match information based on the keyword retrieval results and medication information, the screening module is used to screen patients' medications based on the information matching results, and the medication guidance book generation module is used to generate patients' medication guidance books.

[0005] Combined with the above application and the content in the prior art:

[0006] During the outbreak of influenza or similar infectious diseases, the number of patients visiting the hospital will increase significantly. Considering the reception pressure and workload of the medical staff team, the hospital may have difficulty completing the reception task for patients, resulting in patient detention in the hospital. In such a situation, in order to reduce the hospital's reception load, the hospital or the medical staff team usually connects to medical assistance systems, such as patient information management systems, intelligent auxiliary diagnosis systems, and medical diagnosis large models, etc. Through these intelligent systems or intelligent devices, the efficiency of reception and diagnosis can be improved.

[0007] In the existing intelligent dialogue system for patient medication management, patients or doctors usually input symptom information into the medical big model, which then outputs corresponding medication plans or reference suggestions. However, during influenza or similar epidemic outbreaks, some high-risk patients, such as the elderly, pregnant women, and patients recovering from surgery, need to be provided with high-precision medication plans, otherwise there may be certain medication safety risks. When the concentration of high-risk patients is high, the existing medical big model lacks timely health monitoring and data collection for high-risk patients, resulting in the inability to timely detect and deal with the medication risks of high-risk patients, resulting in insufficient health protection for high-risk patients.

[0008] To this end, the present invention provides an intelligent dialogue system for patient medication management driven by a large model. Summary of the invention

[0009] 1. Technical issues to be resolved

[0010] In view of the deficiencies of the prior art, the present invention provides an intelligent dialogue system for patient medication management driven by a large model. After identification using a medical large model, medical response information is given, and a medical response information set is generated by aggregation; the frequency of return visits for high-risk patients is constrained, and the medical response information received by high-risk patients is optimized to obtain the optimized medical response information; the rehabilitation coefficient is generated from the health risk value of high-risk patients within the observation period. If the rehabilitation coefficient does not exceed the expectation, the medical large model is optimized, and the maintenance frequency is constrained according to the rehabilitation coefficients of different high-risk patients. Active intervention is performed when the health status of high-risk patients is abnormal, and medication reference suggestions are given to achieve health protection for high-risk patients; thereby solving the technical problems raised in the background technology.

[0011] (II) Technical solution

[0012] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0013] Intelligent dialogue system for patient medication management driven by large models, including:

[0014] The aggregation analysis unit uses the trained patient risk model to identify high-risk patients among the received patients, and generates the aggregation degree Fto according to the reception time node distribution of high-risk patients. If the aggregation degree Fto exceeds the aggregation threshold, the high-risk patient information is collected and a monitoring instruction is issued to the outside. The method of analyzing and obtaining the aggregation degree Fto of high-risk patients is as follows:

[0015]

[0016] Weight coefficient, 0≤p i ≤1, p iis the weighted value of the ith high-risk patient, and its value falls within [0, 1]; k is the number of high-risk patients, T ij is the time interval from the i-th appearance of a high-risk patient to the j-th appearance of a high-risk patient, T a is the time interval average;

[0017] The abnormal alarm unit collects real-time health feedback data of high-risk patients and issues an external warning instruction when there is an abnormality in the data; when there is no response to the warning instruction, the urgency Hto is constructed according to the issuance status of the warning instruction. If the urgency Hto exceeds the emergency threshold, an alarm instruction is issued to the outside;

[0018] The medical information generating unit takes the relevant data of the high-risk patients as input, uses the medical big model to identify and then gives the medical response information, and summarizes and generates a medical response information set; generates an information value Xso according to the medical response information, and selects an encryption method according to the information value Xso to encrypt the medical response information;

[0019] The return visit unit constrains the return visit frequency of high-risk patients, and after collecting feedback data from high-risk patients, optimizes the medical response information received by the high-risk patients to obtain optimized medical response information;

[0020] The rehabilitation assessment optimization unit generates a rehabilitation coefficient Ktu based on the health risk value of high-risk patients during the observation period. If the rehabilitation coefficient Ktu does not exceed expectations, the medical large model is optimized and the maintenance frequency is constrained based on the rehabilitation coefficient Ktu of different high-risk patients.

[0021] Furthermore, information is collected from the patient, and the patient information is used as input, and the trained patient risk model is used to identify whether the patient is a high-risk patient, and after risk assessment of the patient, the health risk value of the high-risk patient is obtained;

[0022] Record the time nodes when high-risk patients appear during the reception period, analyze and obtain the concentration Fto of high-risk patients. If the concentration Fto exceeds the concentration threshold, establish a health information file for the high-risk patients based on the high-risk patient file information.

[0023] Furthermore, after receiving the monitoring instruction, health indicator data of high-risk patients are collected and monitored, daily health status and medication data provided by high-risk patients are collected, and the collected data are summarized as a real-time feedback data set;

[0024] The real-time feedback data set is used as input and the trained abnormal data recognition model is used for identification. When there are abnormal values ​​in the real-time feedback data set, an early warning instruction is issued to the outside.

[0025] Further, taking the time node of the received early warning instruction as the early warning node, construct the urgency Hto of the early warning instruction under dimensionless conditions in the following manner:

[0026]

[0027] where, Jy ij is the time interval from the i-th early warning instruction to the j-th early warning instruction, Cy ij is the difference in the abnormal degree of the abnormal health indicators of high-risk patients from the i-th early warning instruction to the j-th early warning instruction; ρ is the weight coefficient, 0 ≤ ρ ≤ 1; m is the number of times of issuing early warning instructions.

[0028] Further, after receiving the alarm instruction, collect the health information files and real-time feedback data sets of high-risk patients, use them as inputs, and use a medical large model to identify them, identify the current health and medication risks of high-risk patients, formulate personalized medication plans for high-risk patients, generate corresponding medication and related suggestions, and after review and confirmation, summarize them to generate a medical reply information set.

[0029] Further, when sending the medical reply information in the medical reply information set to high-risk patients, generate the information density Xt for the information, perform linear normalization on the information density Xt, map the corresponding data values to the interval [0, 1], and generate the information value Xso according to the following method:

[0030]

[0031] where, i = 1, 2,... p, p is the number of sub-stages; Xt i is the information density within the i-th sub-stage, Xt a is the mean value of the information density, the weight coefficient, 0 ≤ F 1 ≤ 1, 0 ≤ F 2 ≤ 1, and F 1 + F 2 = 1;

[0032] According to the positive correlation between the information value Xso and the encryption difficulty, select the corresponding encryption method from the pre-constructed encryption scheme library.

[0033] Further, if the medical reply information received by high-risk patients is followed up on high-risk patients after the observation stage, collect the feedback data of high-risk patients, and summarize it as the dialogue feedback data set;

[0034] Obtain the feedback data and real-time health index data of high-risk patients, take reducing the health risk value of high-risk patients as the optimization goal, and use the pre-trained genetic algorithm to optimize the medical reply information.

[0035] Further, after obtaining the health risk value of a high-risk patient, the follow-up frequency of the high-risk patient is restricted, and the high-risk patient is followed up at the follow-up nodes that meet the restricted conditions, as follows:

[0036]

[0037] Weight coefficients, 0 ≤ α ≤ 1, 0 ≤ β ≤ 1; Fot is the health risk value of the high-risk patient, n is the number of follow-ups within the follow-up period, C ij is the time interval from the i-th follow-up node to the j-th follow-up node, C a is the average value of the time intervals.

[0038] Further, after collecting the health indicators and related data of high-risk patients in each sub-period, a corresponding health risk value is regenerated; the health risk values Fot of high-risk patients in a number of consecutive sub-periods are sorted according to the time axis, and a recovery coefficient Ktu is generated based on the change of the health risk value Fot, as follows:

[0039]

[0040] where i = 1, 2,..., t, t is the number of sub-periods, weight coefficients: 0 ≤ F 1 ≤ 1, 0 ≤ F 2 ≤ 1 and F 2 + F 1 = 1; Fot i is the health risk value in the i-th sub-period, Fot y is the initial value of the health risk value, Fot p is the acceptable value of the health risk value; Kt i is the i-th recovery intermediate value, Kt a is its average value.

[0041] Further, after receiving an optimization instruction, various feedback data of different high-risk patients are collected, the feedback data is identified to obtain the points to be optimized, and the medical large model is maintained regularly based on the points to be optimized. The maintenance frequency is determined as follows:

[0042]

[0043] Weight coefficients, 0 ≤ F 1 ≤ 1, 0 ≤ F 2 ≤ 1, and F 1 + F 2 = 1, Ktu i is the recovery coefficient of the i-th high-risk patient, Ktu a is the average value of the recovery coefficients, i = 1, 2,..., k, k is the number of high-risk patients, and Hp is the maintenance frequency.

[0044] (3) Beneficial effects

[0045] The present invention provides an intelligent dialogue system for patient medication management driven by a large model, which has the following beneficial effects:

[0046] 1. High-risk patients are screened out from a number of patients based on the health risk value. Targeted treatment can be carried out for high-risk patients to improve the pertinence of patient management; the density of high-risk patients is judged according to the aggregation degree Fto to evaluate the reception difficulty of high-risk patients. When the reception difficulty is relatively high, a health information file is established for high-risk patients to achieve targeted information management and improve the reception efficiency.

[0047] 2. The trained abnormal data recognition model is used for recognition. When the physical health indicators of high-risk patients are abnormal, timely treatment can be carried out to achieve the health guarantee of high-risk patients; by constructing the urgency Hto based on the distribution state of the warning nodes, when the health indicators of high-risk patients are continuously abnormal, an alarm instruction is issued after the warning to achieve forced reminder to ensure that high-risk patients can receive attention and treatment.

[0048] 3. After collecting various information of high-risk patients, a medication plan and other similar reference suggestions are given to high-risk patients. When the health status of high-risk patients is abnormal, active intervention is carried out to achieve the health guarantee of high-risk patients; according to the content and importance of the medical reply information, a corresponding encryption scheme is selected for it to achieve targeted encryption, which can avoid the loss of important data.

[0049] 4. By collecting the feedback data of high-risk patients, when the obtained feedback does not match the actual expectation, timely treatment can be carried out; by restricting the return visit frequency according to the health risk value of high-risk patients, the return visit frequency is adapted to the health status of high-risk patients, avoiding the inconvenience brought to high-risk patients by too high or too low return visit frequency.

[0050] 5. By optimizing and adjusting the return visit information, the adaptability between the medical reply information and high-risk patients is higher, and when high-risk patients are in the recovery state, the efficiency of recovery or treatment is improved.

[0051] 6. The recovery coefficient Ktu is constructed based on the change of the health risk value of high-risk patients. The change of the health status of high-risk patients is evaluated according to the recovery coefficient Ktu. If the recovery degree of high-risk patients does not reach the expectation, the medical large model is optimized.

[0052] 7. Conduct targeted optimization and supplementation of the medical large model. When using the medical large model for targeted management of high-risk patients, better results can be achieved, improving the rehabilitation efficiency of high-risk patients. Based on obtaining the rehabilitation coefficients of several high-risk patients, optimize the maintenance frequency of the medical large model as a whole, so that the technical support ability of the medical large model can keep up in real time. When used for managing high-risk users, better results can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic flowchart of the intelligent dialogue method for patient medication management driven by a large model according to the present invention;

[0054] Figure 2 It is a schematic structural diagram of the intelligent dialogue system for patient medication management driven by a large model according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0056] Please refer to Figure 1 , the present invention provides an intelligent dialogue method for patient medication management driven by a large model, including,

[0057] Step 1. Use the trained patient risk model to identify high-risk patients among the received patients. Generate the aggregation degree Fto based on the distribution of the reception time nodes of the high-risk patients. If the aggregation degree Fto exceeds the aggregation threshold, collect the information of the high-risk patients and send a monitoring instruction to the outside;

[0058] The above Step 1 includes the following contents:

[0059] Step 101. Collect patient information, including basic information: age, gender, height, weight, contact information, etc.; medical history information: including whether there are chronic diseases (such as diabetes, hypertension, etc.), past surgical history, allergy history (drug allergy, food allergy, etc.); medication records: currently taking medications (drug name, dosage, usage frequency), past medication history;

[0060] Train a convolutional neural network with the labeled sample data to obtain the trained patient risk identification model; use the patient information as the input, use the trained patient risk model to identify whether the patient is a high-risk patient, and after risk assessment of the patient, obtain the health risk value of the high-risk patient;

[0061] When in use, after collecting various information of a patient, the trained patient risk model can be used to score the current health risk of the patient, and high-risk patients can be screened out from a number of patients based on the health risk value. For high-risk patients, targeted treatment can be carried out to improve the pertinence of patient management.

[0062] Step 102: After presetting the reception cycle of patients, record the time nodes when high-risk patients occur within the reception cycle, and analyze and obtain the aggregation degree Fto of high-risk patients according to the distribution state of the time nodes. The method is as follows:

[0063]

[0064] Weight coefficient, 0 ≤ p i ≤ 1, p i is the weighted value of the i-th high-risk patient, and the value falls within [0, 1]; k is the number of high-risk patients, T ij is the time interval from the i-th occurrence of a high-risk patient to the j-th occurrence of a high-risk patient, T a is the average value of the time intervals;

[0065] According to historical data and the acceptance degree of high-risk patients, preset the aggregation threshold.

[0066] If the aggregation degree Fto exceeds the aggregation threshold, it indicates that the current number of high-risk patients is large and the distribution is relatively dense. At this time, collect the information of high-risk patients, record a detailed health file, sort out and digitize the information such as the patient's medical history, allergy history, and current health status, and record the patient's sleep situation (using a sleep monitoring device), exercise habits (daily exercise volume, exercise type, etc.); record the patient's diet situation through a diet recording software or diary; obtain the patient's various health indicators (such as blood pressure, blood sugar, cholesterol level, etc.) through regular physical examination reports; summarize and generate high-risk patient plan information.

[0067] Establish a health information file for high-risk patients according to the high-risk patient plan information and send a monitoring instruction to the outside.

[0068] When in use, combine the content in Steps 101 and 102:

[0069] After continuously screening out a number of high-risk patients, generate the aggregation degree Fto according to the distribution state of the time nodes for receiving high-risk patients, judge the density of high-risk patients based on the aggregation degree Fto, realize the evaluation of the reception difficulty of high-risk patients, and when the reception difficulty is relatively high, establish a health information file for high-risk patients to achieve targeted information management and improve the reception efficiency.

[0070] In the existing intelligent dialogue system for patient medication management, usually after a patient or a doctor inputs symptom information into a medical large model, the medical large model outputs the corresponding medication plan or reference suggestions. However, during the outbreak of influenza or similar epidemics, some high-risk patients, such as the elderly, pregnant women, and patients in the postoperative recovery period, etc., these high-risk patients need to be provided with highly accurate medication plans, otherwise there may be certain medication safety hazards. And when the concentration of high-risk patients in the existing medical large model is relatively high, due to the lack of timely health monitoring and data collection for high-risk patients, it is impossible to timely detect and handle the medication risks of high-risk patients, resulting in insufficient health protection for high-risk patients.

[0071] Step 2: Collect the real-time health feedback data of high-risk patients, and when the data is abnormal, send a warning instruction to the outside; when there is no response to the warning instruction, construct the urgency Hto according to the sending status of the warning instruction, and if the urgency Hto exceeds the emergency threshold, send an alarm instruction to the outside;

[0072] The above Step 2 includes the following content:

[0073] Step 201: After receiving the monitoring instruction, use wearable devices, such as smart bracelets, sphygmomanometers, blood glucose meters, etc., to collect and monitor the health index data of high-risk patients, and real-time monitor the health indexes of high-risk patients, including heart rate, blood pressure, blood glucose, sleep quality, respiratory rate, body temperature, etc.; collect the daily health status and medication data provided by high-risk patients, and summarize the collected data as the real-time feedback data set;

[0074] Step 202: Train a convolutional neural network with the labeled sample data to obtain a trained abnormal data recognition model; use the trained abnormal data recognition model for recognition with the real-time feedback data set as the input. When there are abnormal values in the real-time feedback data set, such as too high blood pressure, too low blood glucose, etc., send a warning instruction to the patient and the medical team;

[0075] During use, after real-time monitoring and collection of the health indexes of high-risk patients, use the trained abnormal data recognition model for recognition, and send a warning instruction to the outside when the health indexes of high-risk patients are abnormal. After the outside receives the warning instruction, it can be processed in a timely manner to achieve the health protection of high-risk patients;

[0076] Step 203: When there is no response to several consecutive warning instructions sent, use the time node of the received warning instruction as the warning node, and construct the urgency Hto of the warning instruction sent in the current stage under dimensionless conditions to judge the urgency of the current warning, and the method is as follows:

[0077]

[0078] Among them, Jy ij is the time interval from the i-th warning instruction to the j-th warning instruction, Cy ij The difference in the degree of abnormality of abnormal health indicators of high-risk patients from the i-th warning instruction to the j-th warning instruction; ρ is the weight coefficient, 0≤ρ≤1; m is the number of warning instructions issued;

[0079] Pre-set emergency thresholds based on historical data and the degree of management of abnormal health indicators of high-risk patients;

[0080] If the urgency Hto exceeds the emergency threshold, it means that the high-risk patient is currently experiencing an abnormal situation that is of high urgency and needs to be processed in a timely manner. At this time, an alarm instruction is issued to the outside;

[0081] When using, combine the contents in steps 201 to 203:

[0082] Taking into account that when the physical health indicators of high-risk patients continuously produce abnormalities but are not treated, the health of high-risk patients may have greater hidden dangers. At this time, by constructing the urgency Hto according to the distribution status of the early warning nodes, when the health indicators of high-risk patients continuously produce abnormalities, an alarm command is issued after the early warning to implement mandatory reminders and ensure that high-risk patients can receive attention and treatment.

[0083] Step 3: Taking the relevant data of high-risk patients as input, using the medical big model to identify and give medical response information, and summarizing and generating a medical response information set; generating an information value Xso based on the medical response information, and selecting an encryption method to encrypt the medical response information based on the information value Xso;

[0084] The step three includes the following contents:

[0085] Step 301: After receiving the alarm instruction, collect the health information files and real-time feedback data sets of high-risk patients, use them as input, and use the medical big model to identify them, wherein the medical big model includes the following, which can be one or a combination of several of them; including IBM Watson Health, IBM Watson for Oncology, Dxplain, Google Health, Health Catalyst, Babylon Health, Livongo, BlueDot and ClinicalBERT; identify the current health and medication risks of high-risk patients, and obtain corresponding identification results;

[0086] Based on the identification results obtained, develop personalized medication plans for high-risk patients, including drug types, dosages, and time of administration; and generate corresponding medication and related recommendations, such as considering drug interactions to avoid adverse reactions; pay special attention to the safety and indications of drugs for special populations (such as pregnant women and postoperative patients); explain medication plans in detail - explain to patients the mechanism of action, correct method of administration, dosage, time of administration, possible side effects and precautions of each drug through illustrated instructions; provide precautions and life advice - diet control: provide dietary advice that matches the drug to avoid some foods that affect drug absorption; recommend a healthy lifestyle, such as quitting smoking, limiting alcohol consumption, and exercising appropriately; remind patients to avoid other drugs or supplements that conflict with medication;

[0087] After review and confirmation, the medical response information is summarized and generated;

[0088] When in use, when the current health status of high-risk patients may be in an emergency state and has not received attention, with the help of existing medical big models or pre-built medical big models, after collecting various information of high-risk patients, medication plans and other similar reference suggestions are given to high-risk patients, and active intervention is taken when the health status of high-risk patients is abnormal, so as to achieve health protection for high-risk patients;

[0089] Step 302: When sending the medical response information in the medical response information set to the high-risk patient, the medical response information set needs to be encrypted, and an information value Xso is generated based on the medical response information, where:

[0090] The medical information response time is divided into several sub-stages at equal distances, and the ratio of the amount of information to the amount of data in each sub-stage is obtained and used as the information density Xt. The information density Xt is linearly normalized and the corresponding data values ​​are mapped to the interval [0,1]. The information value Xso is generated according to the following method:

[0091]

[0092] Where i = 1, 2, ... p, p is the number of sub-stages; Xt i is the information density in the ith sub-stage, Xt a is the mean value of information density, weight coefficient, 0≤F 1 ≤1,0≤F 2 ≤1, and F 1 +F 2 =1, the weight coefficient can be obtained by referring to the hierarchical analysis method;

[0093] Pre-constructing an encryption scheme library containing several information encryption methods with different encryption difficulties, and selecting a corresponding encryption method from the pre-constructed encryption scheme library according to the positive correlation between information value and encryption difficulty to encrypt the medical response information;

[0094] When using, combine the contents in steps 301 to 302:

[0095] After the medical big model completes the generation and output of various medical response information, it selects the corresponding encryption scheme according to the content and importance of the medical response information to achieve targeted encryption, which can avoid the loss of important data;

[0096] Step 4: restrict the frequency of return visits for high-risk patients, and after collecting feedback data from high-risk patients, optimize the decrypted medical response information received by the high-risk patients to obtain optimized medical response information;

[0097] The step 4 includes the following contents:

[0098] Step 401: If the high-risk patient receives the decrypted medical response information, a return visit is conducted on the high-risk patient after the observation period. For example, through a questionnaire survey or telephone return visit, feedback data on medication regimens and life suggestions from the high-risk patient is collected, and discomfort or side effects experienced by the patient during medication are recorded. The obtained return visit feedback data is aggregated as a dialogue feedback data set;

[0099] When in use, after a high-risk patient receives medical response information, by collecting feedback data from high-risk patients, timely processing can be carried out when the feedback obtained does not meet actual expectations;

[0100] Step 402: After obtaining the health risk value of the high-risk patient, constrain the revisit frequency of each high-risk patient, and revisit the high-risk patient at the revisit node that meets the constraint conditions, in the following manner:

[0101]

[0102] Weight coefficient, 0≤α≤1, 0≤β≤1; the weight coefficient can be obtained by referring to the hierarchical analysis method; Fot is the health risk value of high-risk patients, n is the number of follow-up visits within the follow-up period, C ij is the time interval from the i-th revisited node to the j-th revisited node, C a is the time interval average;

[0103] When in use, the frequency of return visits is constrained according to the health risk value of high-risk patients, so that the frequency of return visits is adapted to the health status of high-risk patients, avoiding inconvenience caused to high-risk patients by return visits that are too high or too low in frequency;

[0104] Step 403: After the return visit, feedback data and real-time health index data of high-risk patients are obtained, and the decrypted medical response information received by the high-risk patients is optimized using a pre-trained genetic algorithm to obtain optimized medical response information, with the reduction of the health risk value of the high-risk patients as the optimization goal;

[0105] When using, combine the contents in steps 401 to 403:

[0106] On the basis of completing the follow-up visit, by optimizing and adjusting the follow-up information, the medical response information can be made more adaptable to high-risk patients, thereby improving the efficiency of rehabilitation or treatment when the high-risk patients are in a state of recovery.

[0107] Step 5: Generate a recovery coefficient Ktu based on the health risk value of high-risk patients during the observation period. If the recovery coefficient Ktu does not exceed expectations, optimize the medical model and constrain the maintenance frequency based on the recovery coefficients Ktu of different high-risk patients.

[0108] The step five includes the following contents:

[0109] Step 501: After the high-risk patient receives the optimized medical response information, an observation period including several sub-periods is pre-set, and after collecting the health indicators and related data of the high-risk patient in each sub-period, the corresponding health risk value is regenerated;

[0110] The sub-periods are marked with the obtained health risk values, and the health risk values ​​Fot of high-risk patients in several consecutive sub-periods are sorted according to the time axis. The rehabilitation coefficient Ktu is generated according to the change of the health risk value Fot, as follows:

[0111]

[0112] Among them, Fot i is the health risk value in the ith sub-period, Fot y is the initial value of the health risk value, Fot p is the acceptable value of health risk; Kt i is the i-th recovery median value, Kt a is its mean;

[0113] i=1,2,…,t,t is the number of sub-periods, weight coefficient: 0≤F 1 ≤1,0≤F 2 ≤1 and F 2 +F 1 =1; the weight coefficient is consistent with the previous value;

[0114] Pre-set recovery thresholds based on historical data and expectations for recovery management of high-risk patients;

[0115] Based on the recovery coefficient Ktu, it is possible to judge whether the high-risk patients have recovered and the corresponding degree of recovery. If the high-risk patients have not improved and if the recovery coefficient Ktu does not exceed the recovery threshold, after optimizing the medical response information, the medical big model needs to be optimized. At this time, an optimization instruction is issued to the outside.

[0116] When in use, after high-risk patients have obtained medical response information several times, the rehabilitation coefficient Ktu is constructed according to the changes in the health risk value of the high-risk patients. The rehabilitation coefficient Ktu can be used to evaluate and judge the changes in the health status of high-risk patients. If the degree of recovery of high-risk patients does not meet expectations, it means that the current medical big model has not achieved the expected effect in the medical response information given after obtaining various information data of high-risk patients, and the medical big model needs to be optimized;

[0117] Step 502: After receiving the optimization instruction, various feedback data of different high-risk patients are collected, and after combining their change states and reference values, the feedback data are identified to obtain corresponding points to be optimized, for example, the accuracy of the health and work schedule recommendations is insufficient, the error rate of personalized medication plans is high, or the identification of the patient's symptoms is inaccurate, etc.;

[0118] Regularly maintain the medical model based on the points to be optimized. The maintenance frequency is determined as follows:

[0119]

[0120] Weight coefficient, 0≤F 1 ≤1,0≤F 2 ≤1, and F 1 +F 2 =1, the value of the weight coefficient is consistent with the previous value; Ktu i is the recovery coefficient of the i-th high-risk patient, Ktu a is the mean value of the recovery coefficient, k is the number of high-risk patients, Hp is the maintenance frequency;

[0121] When using, combine the contents in steps 501 and 502:

[0122] After identifying the points to be optimized in the medical big model, targeted optimization, adjustment and supplementation are carried out on the medical big model. When the medical big model is used for targeted management of high-risk patients, better results can be achieved and the rehabilitation efficiency of high-risk patients can be improved. At the same time, based on the recovery coefficients of several high-risk patients, the maintenance frequency of the medical big model is optimized as a whole, so that the technical support capabilities of the medical big model can be followed up in real time, and better results can be achieved when used to manage high-risk users.

[0123] The Analytic Hierarchy Process (AHP) is a decision-making method that decomposes decision-related elements into levels such as goals, criteria, and plans, and conducts qualitative and quantitative analysis on this basis. It is particularly suitable for dealing with target systems with hierarchical and staggered evaluation indicators, and when the target value is difficult to describe quantitatively, the AHP is an effective decision-making tool.

[0124] The core of the hierarchical analysis method is to decompose the decision problem into multiple levels to form a hierarchical structure, which usually includes the target level, the criterion level, the sub-criterion level and the solution level. By solving the eigenvector of the judgment matrix, the priority weight of each element of each level to an element of the previous level is obtained, and the final weight of each alternative solution to the total goal is recursively merged by the weighted sum method, so as to find the optimal solution.

[0125] See also Figure 2 The present invention provides a patient medication management intelligent dialogue system based on a large model driven system, including:

[0126] The aggregation analysis unit uses the trained patient risk model to identify high-risk patients among the received patients, and generates the aggregation degree Fto according to the reception time node distribution of high-risk patients. If the aggregation degree Fto exceeds the aggregation threshold, the high-risk patient information is collected and a monitoring instruction is issued to the outside. The method of analyzing and obtaining the aggregation degree Fto of high-risk patients is as follows:

[0127]

[0128] Weight coefficient, 0≤p i ≤1, p i is the weighted value of the ith high-risk patient, and its value falls within [0, 1]; k is the number of high-risk patients, T ij is the time interval from the i-th appearance of a high-risk patient to the j-th appearance of a high-risk patient, T a is the time interval average;

[0129] The abnormal alarm unit collects real-time health feedback data of high-risk patients and issues an external warning instruction when there is an abnormality in the data; when there is no response to the warning instruction, the urgency Hto is constructed according to the issuance status of the warning instruction. If the urgency Hto exceeds the emergency threshold, an alarm instruction is issued to the outside;

[0130] The medical information generating unit takes the relevant data of the high-risk patients as input, uses the medical big model to identify and then gives the medical response information, and summarizes and generates a medical response information set; generates an information value Xso according to the medical response information, and selects an encryption method according to the information value Xso to encrypt the medical response information;

[0131] The return visit unit constrains the return visit frequency of high-risk patients, and after collecting feedback data from high-risk patients, optimizes the medical response information received by the high-risk patients to obtain optimized medical response information;

[0132] The rehabilitation assessment optimization unit generates a rehabilitation coefficient Ktu based on the health risk value of high-risk patients during the observation period. If the rehabilitation coefficient Ktu does not exceed expectations, the medical large model is optimized and the maintenance frequency is constrained based on the rehabilitation coefficient Ktu of different high-risk patients.

[0133] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0134] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0135] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0136] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only some logical function divisions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0137] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0138] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0139] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks or optical disks.

[0140] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An intelligent dialogue system for patient medication management driven by a large model, characterized by: include, The aggregation analysis unit uses the trained patient risk model to identify high-risk patients among the received patients, and generates the aggregation degree Fto according to the reception time node distribution of high-risk patients. If the aggregation degree Fto exceeds the aggregation threshold, the high-risk patient information is collected and a monitoring instruction is issued to the outside. The method of analyzing and obtaining the aggregation degree Fto of high-risk patients is as follows: Weight coefficient, 0≤p i ≤1, p i is the weighted value of the i-th high-risk patient; k is the number of high-risk patients, T ij is the time interval from the i-th appearance of a high-risk patient to the j-th appearance of a high-risk patient, T a is the time interval average; The abnormal alarm unit collects real-time health feedback data of high-risk patients and issues an external warning instruction when there is an abnormality in the data; when there is no response to the warning instruction, the urgency Hto is constructed according to the issuance status of the warning instruction. If the urgency Hto exceeds the emergency threshold, an alarm instruction is issued to the outside; The medical information generating unit takes the relevant data of the high-risk patients as input, uses the medical big model to identify and then gives the medical response information, and summarizes and generates a medical response information set; generates an information value Xso according to the medical response information, and selects an encryption method according to the information value Xso to encrypt the medical response information; The return visit unit constrains the return visit frequency of high-risk patients, and after collecting feedback data from high-risk patients, optimizes the medical response information received by the high-risk patients to obtain optimized medical response information; The rehabilitation assessment optimization unit generates a rehabilitation coefficient Ktu based on the health risk value of high-risk patients during the observation period. If the rehabilitation coefficient Ktu does not exceed expectations, the medical large model is optimized and the maintenance frequency is constrained based on the rehabilitation coefficient Ktu of different high-risk patients.

2. The patient medication management intelligent dialogue system according to claim 1, characterized in that: Using the collected patient information as input, the trained patient risk model is used to identify whether the patient is a high-risk patient. After performing a risk assessment on the patient, the health risk value of the high-risk patient is obtained; Record the time nodes when high-risk patients appear during the reception period, analyze and obtain the concentration Fto of high-risk patients. If the concentration Fto exceeds the concentration threshold, establish a health information file for the high-risk patients based on the high-risk patient file information.

3. The patient medication management intelligent dialogue system according to claim 2, characterized in that: After receiving the monitoring instruction, the health indicator data of high-risk patients is monitored, and the daily health status and medication data provided by high-risk patients are collected, and the collected data are summarized as a real-time feedback data set; The real-time feedback data set is used as input and the trained abnormal data recognition model is used for identification. When there are abnormal values ​​in the real-time feedback data set, an early warning instruction is issued to the outside.

4. The patient medication management intelligent dialogue system according to claim 3 is characterized by: Taking the time node of the received warning instruction as the warning node, the urgency Hto of the warning instruction is constructed under dimensionless conditions as follows: Among them, Jy ij is the time interval from the i-th warning instruction to the j-th warning instruction, Cy ij The difference in the degree of abnormality of abnormal health indicators of high-risk patients from the i-th warning instruction to the j-th warning instruction; ρ is the weight coefficient, 0≤ρ≤1; m is the number of times the warning instruction is issued.

5. The patient medication management intelligent dialogue system according to claim 4, characterized in that: After receiving the alarm command, the health information files and real-time feedback data sets of high-risk patients are collected and used as input. The medical big model is used to identify them, identify the current health and medication risks of high-risk patients, formulate personalized medication plans for high-risk patients, and generate corresponding medication and related suggestions. After review and confirmation, the medical response information set is summarized and generated.

6. The patient medication management intelligent dialogue system according to claim 5, characterized in that: When sending the medical response information in the medical response information set to the high-risk patient, the information density Xt is generated based on the medical response information, and the information density Xt is linearly normalized, and the corresponding data value is mapped to the interval [0,1], and the information value Xso is generated according to the following method: Where i = 1, 2, ... p, p is the number of sub-stages; is the information density in the ith sub-stage, Xt a is the mean value of information density, weight coefficient, 0≤F1≤1, 0≤F2≤1, and F1+F2=1; According to the correlation between the information value Xso and the encryption difficulty, a corresponding encryption method is selected from the encryption scheme library.

7. The patient medication management intelligent dialogue system according to claim 6, characterized in that: If a high-risk patient receives medical response information, a return visit will be conducted to the high-risk patient after the observation phase to collect feedback data from the high-risk patient, which will be summarized as a dialogue feedback data set; Obtain feedback data and real-time health indicator data from high-risk patients, take reducing the health risk value of high-risk patients as the optimization goal, and use pre-trained genetic algorithms to optimize medical response information.

8. The patient medication management intelligent dialogue system according to claim 7, characterized in that: After obtaining the health risk value of high-risk patients, the frequency of revisiting high-risk patients is constrained, and revisiting high-risk patients at revisit nodes that meet the constraints is performed as follows: Weight coefficient, 0≤α≤1, 0≤β≤1; Fot is the health risk value of high-risk patients, n is the number of follow-up visits within the follow-up period, C ij is the time interval from the i-th revisited node to the j-th revisited node, C a is the average value of the time interval.

9. The patient medication management intelligent dialogue system according to claim 8, characterized in that: After collecting the health indicators and related data of high-risk patients in the sub-period, the corresponding health risk values ​​are regenerated; The health risk values ​​Fot of high-risk patients in several consecutive sub-periods are sorted according to the time axis, and the recovery coefficient Ktu is generated according to the change of the health risk value Fot, as follows: Where i = 1, 2, ..., t, t is the number of sub-periods, weight coefficient: 0 ≤ F1 ≤ 1, 0 ≤ F2 ≤ 1 and F2 + F1 = 1; Fot i is the health risk value in the ith sub-period, Fot y is the initial value of the health risk value, Fot p is the acceptable value of health risk; Kt i is the i-th recovery median value, Kt a is its mean value.

10. The patient medication management intelligent dialogue system according to claim 9, characterized in that: After receiving the optimization instruction, various feedback data of different high-risk patients are collected, and the points to be optimized are obtained after the feedback data are identified. The medical model is regularly maintained based on the points to be optimized. The maintenance frequency is determined as follows: Weight coefficient, 0≤F1≤1, 0≤F2≤1, and F1+F2=1, Ktu i is the recovery coefficient of the i-th high-risk patient, Ktu a is the mean of the recovery coefficient, i = 1, 2, … k, k is the number of high-risk patients, and Hp is the maintenance frequency.

Images