Artificial intelligence-based hierarchical chronic disease management and referral system

Abstract

The present invention discloses an artificial intelligence-based hierarchical chronic disease management and referral system, which belongs to the field of smart medical technology and comprises: setting upstream and downstream referral paths composed of diagnosis and treatment institutions, and binding the responsible doctors corresponding to chronic diseases for diagnosis and treatment institutions at each level to form a referral relationship mapping table; predicting the dynamic risk of chronic diseases based on multi-source and multi-modal data of the target population and classifying the target population, generating chronic disease management suggestions for each type of population and pushing them to chronic disease doctors for contract management; generating referral suggestions based on the dynamic risk changes of chronic diseases in each type of population and pushing them to chronic disease doctors, the chronic disease doctors determine the referral suggestions and set referral information based on the upstream and downstream referral paths and the referral relationship mapping table and simultaneously notify the target diagnosis and treatment institutions and patients; and providing performance incentives to chronic disease doctors based on whether they have completed the contract signing of chronic disease patients, implemented hierarchical referrals, and saved medical expenses. This can improve the accuracy of medical services, improve the efficiency of diagnosis and treatment, and promote medical cost saving.

Description

Technical Field

[0001] The present invention belongs to the field of smart medical technology, and specifically relates to a hierarchical and graded chronic disease management and referral system based on artificial intelligence. Background Art

[0002] Stratified diagnosis and treatment, as well as referrals, play a crucial role in a health-centered healthcare model. Currently, however, patient referrals and stratification rely primarily on manual judgment by healthcare professionals based on guidelines and experience, as well as spontaneous referrals by patients. Due to their demand for high-quality medical services, patients, even during the stable phase of their chronic illness, tend to seek treatment at higher-tier hospitals. This places a heavy burden on these hospitals and underutilizes primary healthcare services.

[0003] Artificial intelligence (AI) technology and big data have developed rapidly in recent years. The deep integration of the AI ​​industry across various sectors continues to spawn new business models. AI, with its superior capabilities in big data-assisted diagnosis, intelligent decision support, real-time information sharing, resource allocation prediction, and automated processing, is helping to advance the reform and optimization of tiered medical diagnosis and referral systems. Consequently, an AI-based tiered management and referral system is urgently needed. Summary of the Invention

[0004] In view of the above, the purpose of the present invention is to provide a hierarchical and graded chronic disease management and referral system based on artificial intelligence, establish a hierarchical and graded diagnosis and treatment system through artificial intelligence technology, classify the target management population into layers, build an upstream and downstream referral chain of medical institutions and medical teams, carry out active management and active referral based on artificial intelligence risk prediction, build a reverse incentive mechanism, and optimize medical resource allocation and medical insurance management.

[0005] To achieve the above-mentioned purpose of the invention, the embodiment provides an artificial intelligence-based hierarchical chronic disease management and referral system, including:

[0006] The target population classification and contract management module is used to predict the dynamic risk of chronic diseases and classify the target population based on multi-source and multi-modal data of the target population, generate chronic disease management recommendations for each group of people, and push them to chronic disease doctors for contract management;

[0007] A hierarchical referral path setting module is used to provide a hierarchical referral path setting function. Based on the referral path setting function, an upstream and downstream referral path consisting of medical institutions is set, and the responsible doctors corresponding to chronic diseases are bound to each level of medical institutions to form a referral relationship mapping table;

[0008] Active referral module, which is used to generate referral suggestions based on the dynamic risk changes of chronic diseases in each group of people and push them to chronic disease doctors. Chronic disease doctors determine the referral suggestions and set referral information based on upstream and downstream referral paths and referral relationship mapping tables, and simultaneously notify the target diagnosis and treatment institutions and patients of the referral;

[0009] The reverse incentive module is used to provide performance incentives to chronic disease doctors based on whether they have completed the signing of chronic disease patients, implemented tiered referrals, and saved medical expenses.

[0010] Preferably, the target population classification and contract management module predicts the dynamic risk of chronic diseases based on the multi-source and multi-modal data of the target population and classifies the target population, generates chronic disease management suggestions for each type of population and pushes them to chronic disease doctors for contract management, including:

[0011] Use chronic disease prediction models to predict the dynamic risk of chronic diseases based on multi-source and multi-modal data of the target population;

[0012] The target population whose chronic disease dynamic risk prediction results are lower than the first threshold are classified as healthy people. For the healthy population, follow-up reminders and follow-up projects are regularly sent to chronic disease doctors and patients, and follow-up project data of the healthy population are actively tracked, and the prediction of progression to high-risk groups is observed;

[0013] The target population whose dynamic risk prediction results for chronic diseases exceed the second threshold is classified as a high-risk population. Chronic disease doctors are signed with the high-risk population on a voluntary basis. Follow-up frequency is increased based on the personalized disease risk of high-risk individuals. Follow-up frequency data of high-risk individuals is actively tracked. If follow-up is not completed within a certain number of expected follow-up intervals, a warning value will be triggered and chronic disease doctors will be proactively reminded to pay attention to high-risk individuals.

[0014] The population that has been clinically diagnosed with the disease is classified as the confirmed population, and the population that is actually ill but has not been diagnosed by a doctor is classified as the population to be managed. All the confirmed population and the population to be managed are required to sign contracts with chronic disease doctors, and sign contracts and long-term management are achieved through telephone mobilization, APP reminders, tracking of medical records, and high-risk health education.

[0015] Preferably, a chronic disease prediction model is used to predict the dynamic risk of chronic diseases based on multi-source and multi-modal data of the target population, including:

[0016] First, based on three types of structured multimodal data of the target population: historical diagnostic data, biochemical indicators, and follow-up information, a multimodal health trajectory in the form of a time series is constructed.

[0017] Secondly, the chronic disease prediction model adopts a fusion of dynamic risk encoding mechanism and attention mechanism. Multimodal health trajectory data is input into the chronic disease prediction model according to a certain time window. The encoded features are first encoded and the attention mechanism is used on the encoded features to capture the nonlinear change characteristics of the key variables of interest in the data under different time windows. The nonlinear change characteristics of each key variable are integrated through the convolutional neural network and long short-term memory network adopted by the multi-task learning head. Based on the fusion results, the dynamic risk of chronic diseases and health status trends in the future time period are predicted. The dynamic risk of chronic diseases is expressed in the form of probability, and the larger the probability, the higher the risk.

[0018] An adaptive time decay function is also embedded in the chronic disease prediction model, which introduces a weight reduction mechanism for distant historical data. Different decay weights are set for the input data within the time window according to the distance from the prediction time to adjust the degree of influence on the prediction results. Specifically, the farther the input data is from the prediction time, the larger the decay weight is set.

[0019] Preferably, long-term management for confirmed and awaiting care populations includes:

[0020] Gradual management is implemented based on the progression of chronic diseases. Specifically, based on the consensus and guidelines of a multidisciplinary clinical team of doctors as prior knowledge, cardiovascular disease risk stratification is determined for patients with different chronic disease progression according to the cardiovascular disease risk stratification model, which includes low risk, moderate risk, high risk, and very high risk. Follow-up frequencies and follow-up items with different requirements are carried out according to the stratification results.

[0021] Among them, the risk level of disease progression is determined according to the risk stratification model, including:

[0022] (1) Construct a knowledge map based on clinician consensus to identify chronic disease risk factors and their importance;

[0023] (2) Using the historical patient database, with the patient's physiological indicators as independent variables and each type of chronic disease risk factor judged by complications as the dependent variable, a logistic regression model was constructed to obtain the regression coefficient of each indicator for each type of chronic disease risk factor. The absolute value of each regression coefficient was then normalized to obtain the contribution ratio of each indicator to each type of chronic disease risk factor.

[0024] (3) Based on the importance of each type of chronic disease risk factor and the contribution of each indicator to each type of chronic disease risk factor, the patient's real-time risk score is calculated using the weighted summation method;

[0025] (4) Apply the entropy weight method to dynamically analyze the risk score distribution of all patients and set dynamic thresholds for risk stratification based on the risk score distribution;

[0026] (5) Real-time collection of individual patient physiological indicator data and calculation of real-time risk scores, accurately stratifying patients into corresponding risk levels according to the dynamic thresholds of risk stratification.

[0027] Preferably, the hierarchical referral path setting module is used to set the upstream and downstream referral paths when a super administrator with high authority logs in, and the set upstream and downstream referral paths include: community health service stations at level 1 → community health service centers at level 2 → district hospitals at level 3 → tertiary hospitals at level 4; referral hospitals and target responsible doctor teams are also separately implemented for high-risk groups for chronic diseases;

[0028] When the chronic disease doctor logs in, the hierarchical referral path setting module fills in the list of superior referral doctors, and the nurse fills in the assistance management path.

[0029] Preferably, the active referral module generates referral recommendations based on the dynamic risk changes of chronic diseases in each group of people and pushes them to chronic disease doctors, including:

[0030] When the confirmed population or the population under management is judged to be at continuous high risk based on the dynamic risk changes of chronic diseases, or if follow-up is not completed, they will be judged as high-risk. When the high-risk population is continuously at high risk, they will also be judged as high-risk.

[0031] The referral recommendation generation model is used to generate referral recommendations for people in high-risk situations and push them to chronic disease doctors. The referral recommendations include patient risk analysis reports containing indicator trend charts and risk assessments, recommended referral hospitals and doctors matched according to pre-set hierarchical referral paths, and recommended referral times dynamically recommended based on the source data of the superior hospital number.

[0032] Preferably, the process of the referral recommendation generation model generating referral recommendations for people in high-risk situations is as follows:

[0033] (1) Calculate the medical gap index based on the difference in grade and disease handling capacity between the patient's current medical institution and the target medical institution, where the medical gap index is the product of the grade difference between the two medical institutions and the disease handling capacity;

[0034] (2) Real-time monitoring of patients' key physiological indicators and complication risk scores, and inputting the monitoring data into the gated recurrent neural network in combination with the medical gap index, outputting the emergency score in real time, and identifying high-risk patients based on the emergency score;

[0035] (3) Input the urgency score into the asymmetric priority algorithm. For high-risk patients, the urgency score increases by a given value, and the queue priority weight increases by a given multiple, thereby updating the priority order of the patient queue.

[0036] (4) The medical resource status of the target hospital is collected and a resource heat map is generated. Based on the patient sequence after the updated priority, the patient priority and the hospital resource heat map are input into the multi-objective integer programming model. The multi-objective integer programming model uses the patient priority as the input variable and the medical resource status of the target hospital as the constraint condition to construct an objective function. The objective function aims to maximize the sum of the product of the patient's priority weight and the medical resource satisfaction. Then, integer programming is used to solve the model to obtain the optimal referral path, achieve the best match between patient needs and medical resources, and determine the optimal referral path between patient needs and medical resources.

[0037] (5) Automatically generate referral reports, including the optimal referral path, the patient's real-time urgency score, an overview of the medical resource status of the referral hospital, the expected risk reduction after referral, and the reasons for the referral recommendation.

[0038] Preferably, in the active referral module, the chronic disease doctor determines the referral suggestion and sets the referral information based on the upstream and downstream referral paths and the referral relationship mapping table and simultaneously notifies the target diagnosis and treatment institution and patient of the referral, including:

[0039] After receiving the referral suggestion and reviewing the prediction basis, the chronic disease doctor can choose to accept the referral suggestion or manually modify it to confirm and generate the final referral information. That is, the final referral information is generated by setting information based on the upstream and downstream referral paths and referral relationship mapping table. Manual modification takes precedence over the generated referral suggestion.

[0040] When chronic disease doctors reject referral recommendations, they must manually fill in the reasons and sign electronically.

[0041] Preferably, the active referral module also records in real time when the target diagnosis and treatment institution accepts the referral and generates a treatment plan, and synchronizes the treatment plan to the chronic disease doctor.

[0042] Preferably, in the reverse incentive module, performance incentives are provided to chronic disease doctors based on whether they have completed the signing of chronic disease patients, implemented graded referrals, and saved medical expenses, including:

[0043] The referral acceptance rate and medical cost savings are calculated based on the specific referral circumstances recorded in the system, the model prediction results, whether the referral is made and the record is modified, the follow-up completion status, and the effectiveness of the referral. The performance incentive amount is calculated based on the basic salary, contract signing rate, referral acceptance rate, and medical cost savings.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The artificial intelligence-based hierarchical chronic disease management and referral system provided by the present invention improves the efficiency of chronic disease management, optimizes the allocation of medical resources, and saves medical expenses through various models constructed by artificial intelligence. Specifically, the target population is classified through a hierarchical diagnosis and treatment system, and upstream and downstream referrals are made based on risk predictions and set hierarchical referral paths. The manually driven active screening and active referral mechanism can efficiently identify high-risk patients and automatically generate management and referral recommendations to ensure timely intervention. Through the reverse incentive mechanism, the enthusiasm of medical staff is stimulated. The system can improve the accuracy of medical services, improve the efficiency of diagnosis and treatment, and promote cost savings in medical expenses. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 This is a schematic diagram of the structure of an artificial intelligence-based hierarchical chronic disease management and referral system provided in an embodiment;

[0048] Figure 2 This is a flowchart of hierarchical chronic disease management and referral provided by the embodiment. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not limit the scope of protection of the present invention.

[0050] The inventive concept of the present invention is as follows: The embodiment of the present invention provides a hierarchical chronic disease management and referral system based on artificial intelligence, which is guided by artificial intelligence and uses a chronic disease prediction model to predict the risk of chronic disease onset and disease progression for individuals, and divides individuals into high-risk and low-risk. It takes a two-way dual-path as the core and manual as the final review, and actively recommends and reviews chronic disease doctors based on the risk prediction results and makes the final decision, and performs corresponding active management and active referral. The two-way refers to upward referral and downward referral to form a complete upstream and downstream chain of diagnosis and treatment, and promote the rational flow of medical resources and social medical insurance. The dual paths refer to the path of chronic disease population (diagnosed population + population to be managed) and the path of high-risk population.

[0051] The embodiment provides an artificial intelligence-based hierarchical chronic disease management and referral system. In order to achieve accurate health and chronic disease management, it establishes a hierarchical diagnosis and treatment system based on risk prediction, which is specifically reflected in the hierarchical referral paths and classified management of target populations to ensure that each group of people receives appropriate medical resource allocation; it also has an artificial intelligence-driven active management and active referral mechanism. Artificial intelligence will play a core role in the hierarchical diagnosis and treatment management of chronic diseases, actively screening different categories of people and incorporating them into management, and actively making referral recommendations. It uses artificial intelligence to assist in improving the screening rate, diagnosis rate and treatment rate of chronic diseases, assisting in the reasonable allocation of medical resources and assisting in medical insurance diversion; it also has a reverse incentive mechanism to stimulate the initiative and enthusiasm of the medical team, and continuously optimize the referral path and management.

[0052] like Figure 1 As shown, the artificial intelligence-based hierarchical chronic disease management and referral system 10 provided in the embodiment includes a target population classification and contract management module 11, a hierarchical referral path setting module 12, an active referral module 13, and a reverse incentive module 14.

[0053] In this embodiment, the target population classification and contract management module 11 is used to divide the target management population included in the database into confirmed population, waiting population, high-risk population and healthy population, and perform hierarchical management. Among them, the confirmed population refers to the population that has been diagnosed with a chronic disease by a doctor, the waiting population refers to the population that actually meets the diagnostic criteria for a chronic disease but has not yet been diagnosed by a doctor, the high-risk population refers to the population that does not currently meet the diagnostic criteria for a chronic disease but is identified by the chronic disease prediction model as being at high risk of developing the disease, and the healthy population refers to the population that is identified by the chronic disease prediction model as being at low risk of developing the disease.

[0054] During patient information entry, chronic disease doctors assist patients in logging in for the first time, filling in their personal information and choosing whether to sign a contract. For those diagnosed with the disease and those awaiting care, we ensure they have signed a contract with a chronic disease doctor; high-risk individuals are encouraged to sign up for a guaranteed contract. Using an AI-based chronic disease prediction model, we predict the dynamic risk of chronic diseases and further refine the risk level of disease progression for those diagnosed with the disease and those awaiting care, facilitating further management.

[0055] Specifically, based on the multi-source and multi-modal data of the target population, the dynamic risk of chronic diseases is predicted and the target population is classified. Chronic disease management recommendations are generated for each group of people and pushed to chronic disease doctors for contract management. Different contract management models and chronic disease management approaches are adopted for different groups of people, including:

[0056] Use chronic disease prediction models to predict the dynamic risk of chronic diseases based on multi-source and multi-modal data of the target population. Specifically include:

[0057] First, based on three types of structured multimodal data: historical diagnostic data, biochemical indicators, and follow-up information of the target population, a multimodal health trajectory in the form of a time series is constructed. This multimodal health trajectory reflects the dynamic changes in the patient's health status through a time series approach. Among them, historical diagnostic data includes the diagnosis time and severity of hypertension and diabetes, and biochemical indicator data includes regular test results such as fasting blood sugar, glycosylated hemoglobin, and total cholesterol.

[0058] Secondly, the chronic disease prediction model adopts a fusion of dynamic risk encoding mechanism and attention mechanism. That is, multimodal health trajectory data is input into the chronic disease prediction model according to a certain time window. It is first encoded and the attention mechanism is used on the encoded features to capture the nonlinear change characteristics of the key variables of interest in the data under different time windows. This can more effectively identify the key nodes of risk change (for example, fasting blood sugar suddenly rises from 5.5mmol / L to 7.8mmol / L). The convolutional neural network (CNN) and long short-term memory network (LSTM) used in the multi-task learning head are used to fuse the nonlinear change characteristics of each key variable. Based on the fusion results, the dynamic risk of chronic diseases and health status trends in the future time period are predicted. The dynamic risk of chronic diseases is expressed in the form of probability, and the larger the probability, the higher the risk. The health status trend includes, for example, that glycated hemoglobin may reach 8.5% in the next few months.

[0059] It's also worth noting that the chronic disease prediction model also incorporates an adaptive time decay function, introducing a weight-reduction mechanism for distant historical data. Specifically, different decay weights are assigned to input data within the time window based on their distance from the prediction time, adjusting their impact on the prediction results. Specifically, the further away from the prediction time, the greater the decay weight. For example, input data from the last six months has a weight of 1, input data from 6-12 months has a weight of 0.6, and input data older than 12 months has a weight reduced to 0.3.

[0060] For healthy individuals: Target populations whose chronic disease dynamic risk prediction results are below the first threshold are classified as healthy. Follow-up reminders and follow-up items are regularly sent to chronic disease doctors and patients via PCs and desktops. For example, for hypertension, follow-up frequency is set to once a year. At the same time, follow-up item data for healthy individuals is actively tracked, and predictions of progression to high-risk individuals are monitored.

[0061] For high-risk populations: Target populations whose dynamic chronic disease risk prediction results exceed the second threshold are classified as high-risk. This threshold is greater than the first threshold. The requirement for chronic disease physicians in high-risk populations is to sign up as many as possible, and active mobilization is implemented. This principle is to sign up as many chronic disease physicians as possible. High-risk populations are no longer based on traditional guidelines, but rather on individual patient risk. Follow-up frequency is increased based on the individual risk profile of high-risk individuals. For example, for patients with both diabetes and chronic kidney disease, the actual AI threshold for inclusion in the high-risk hypertension population will be lower than the guideline-based systolic blood pressure of 120-130 mmHg or diastolic blood pressure of 80-89 mmHg. The expected follow-up frequency for these patients will be increased to 2-3 times per year. Follow-up frequency data for high-risk individuals is also actively tracked. Failure to complete follow-up within a certain number of expected follow-up intervals triggers a warning, alerting chronic disease physicians to monitor the high-risk individual.

[0062] For the confirmed population and the population to be managed, the population with clinically confirmed diseases is classified as the confirmed population, and the population who is actually ill but has not been diagnosed by a doctor is classified as the population to be managed. All the confirmed population and the population to be managed are required to sign a contract with a chronic disease doctor, and achieve signing and long-term management through telephone mobilization, APP reminders, tracking of medical records, and high-risk health education.

[0063] Long-term management includes: graded management according to the progression of chronic diseases, specifically based on the diagnosis and treatment consensus and guidelines of the clinical multidisciplinary medical team as prior knowledge, and judging the risk level of disease progression for patients with different chronic disease processes (for example, patients with stage 1 / 2 / 3 hypertension) according to the risk stratification model. The stratification levels include low risk (low risk), medium risk (medium risk), high risk (high risk), and very high risk (very high risk), and carry out follow-up frequencies and follow-up items with different requirements based on the stratification results.

[0064] Among them, the risk level of disease progression is determined according to the risk stratification model, including:

[0065] (1) Construct a knowledge map based on clinician consensus to identify chronic disease risk factors and their importance (e.g., the importance score of hypertension level 1 is 0.3, level 2 is 0.6, level 3 is 0.9, and diabetes is 0.7);

[0066] (2) Using the historical patient database, with the patient's physiological indicators as independent variables and each type of chronic disease risk factor (such as complications) as the dependent variable, a logistic regression model was constructed to obtain the regression coefficient of each indicator for each type of chronic disease risk factor. The absolute value of each regression coefficient was then normalized to obtain the contribution ratio of each indicator to each type of chronic disease risk factor. (For example, if the coefficient of a certain indicator for hypertension is 1.5, for diabetes is 0.9, and for chronic kidney disease is 0.6, then the contribution ratio of hypertension is 1.5 / (1.5+0.9+0.6)=0.5, for diabetes is 0.3, and for chronic kidney disease is 0.2);

[0067] (3) Based on the importance of each type of chronic disease risk factor and the contribution ratio of each indicator to each type of chronic disease risk factor, the patient's real-time risk score is calculated using the weighted summation method. That is, the importance of each risk factor is multiplied by the corresponding contribution ratio and then added together to obtain the risk score (for example, the risk score of patient A = hypertension stage 3 0.9 × 0.5 + diabetes 0.7 × 0.3 + chronic kidney disease 0.8 × 0.2 = 0.82);

[0068] (4) Apply the entropy weight method to dynamically analyze the risk score distribution of all patients (e.g., the median of the current patient group is 0.6), and set the dynamic threshold of risk stratification based on the risk score distribution (low risk <0.4, intermediate risk 0.4-0.6, high risk 0.6-0.8, very high risk ≥0.8);

[0069] (5) Real-time collection of individual patient physiological indicator data (e.g., patient A’s current systolic blood pressure is 190 mmHg, and fasting blood glucose is 8.0 mmol / L) and calculation of real-time risk scores, accurately stratifying patients into corresponding risk levels according to the dynamic thresholds of risk stratification.

[0070] Table 1 shows the detection frequency and follow-up items for different cardiovascular disease risk stratification in hypertensive patients.

[0071] Table 1

[0072]

[0073] The Target Population Classification and Contract Management Module 11 performs real-time classification and screening of target populations based on a multi-source, multimodal database. Using dynamic risk prediction from a chronic disease prediction model, it proactively identifies different population groups and sends relevant management recommendations to chronic disease physicians. This helps patients sign up with chronic disease physicians before or during the early stages of disease onset, enabling them to enter a full-process proactive management phase to better prevent chronic diseases and delay disease progression. Management pathways are adjusted based on the patient's risk level, ensuring that each patient receives a personalized medical plan tailored to their health status.

[0074] In the embodiment, the hierarchical referral path setting module 12 is used to establish upstream and downstream referral paths, establish upstream and downstream corresponding medical communities and specific docking medical teams, and specifically provide a hierarchical referral path setting function. Based on the referral path setting function, the upstream and downstream referral paths composed of medical institutions are set. The set upstream and downstream referral paths include: community health service stations (primary screening) at level 1 → community health service centers (basic diagnosis and treatment) at level 2 → district hospitals (specialized diagnosis and treatment) at level 3 → tertiary hospitals (complex diseases) at level 4. It should be noted that some levels can be skipped for critically ill people. Through the hierarchical referral path setting function, the responsible doctors corresponding to chronic diseases are also bound to the medical institutions at each level to form a referral relationship mapping table.

[0075] Specifically, in the referral path setting, when the super administrator logs in, he needs to set up the referral path, that is, the referral chain of hospitals at all levels, clarify the upstream and downstream referral hospitals and responsible doctors, and separately implement referral hospitals for high-risk groups for chronic diseases; when the hospital manager logs in, he needs to fill in and complete the list of medical and nursing teams for each chronic disease to correspond to the hospital referral chain.

[0076] During the pre-diagnosis stage, the patient will be helped by the nurse or attending doctor, chronic disease doctor, and contracted doctor to confirm the referral chain between the contracted doctor and the patient (a corresponding referral hospital at the village health center, township health center, and county hospital levels).

[0077] In an embodiment, the active referral module 13 is used to generate referral suggestions based on the dynamic risk changes of chronic diseases in each group of people and push them to chronic disease doctors. The chronic disease doctors determine the referral suggestions and set the referral information based on the upstream and downstream referral paths and referral relationship mapping tables and simultaneously notify the target medical institutions and patients of the referral.

[0078] Specifically, in risk prediction, when the confirmed population or the population to be managed is judged to be at persistent high risk based on the dynamic risk changes of chronic diseases, or follow-up is not completed, it is judged to be a high-risk situation. High-risk populations with continuous high risks will also be judged to be in a high-risk situation.

[0079] Persistent high risk is defined as multiple consecutive assessments of a high risk for disease onset or progression. The criteria for disease onset were developed by a multidisciplinary team of physicians, combining guidelines with AI-based logic. Criteria for disease progression include disease grade progression, target organ damage, complications, hospitalization, and death. The specific number of consecutive incomplete follow-up visits is determined by disease severity.

[0080] Upward referral recommendations will be automatically generated for patients identified as high-risk. Specifically, a referral recommendation generation model will be used to generate referral recommendations for people in high-risk situations and push them to chronic disease doctors. The referral recommendations include a patient risk analysis report containing indicator trend charts and risk assessments, recommended referral hospitals and doctors matched according to pre-set hierarchical referral paths, real-time emergency scores for patients, an overview of the medical resource status of referral hospitals, and recommended referral times dynamically recommended based on the number source data of higher-level hospitals.

[0081] The process of generating referral recommendations by the referral recommendation generation model is as follows:

[0082] (1) Calculate the medical gap index based on the level difference and disease handling capacity between the patient's current medical institution and the target medical institution. The medical gap index is the level difference between the two medical institutions multiplied by the disease handling capacity. For example, if the level difference between the current secondary hospital and the target tertiary hospital is 1 level, then the medical gap index is 1×1.5=3, where 1.5 is the disease handling capacity of the given higher-level medical institution.

[0083] (2) Real-time monitoring of patients' key physiological indicators and complication risk scores, and inputting the monitoring data into the gated recurrent neural network (GRU) in combination with the medical gap index, to output the emergency score in real time (ranging from 0 to 100 points, with patients over 80 points defined as high-risk patients);

[0084] (3) Input the urgency score into the asymmetric priority algorithm. For high-risk patients, the queue priority weight increases by a given multiple for every increase in the urgency score of the patient. For example, for high-risk patients with an urgency score ≥80, the queue priority weight increases exponentially by 1.5 times for every increase of 5 points, thereby updating the priority order of the patient queue.

[0085] (4) The medical resource status of the target hospital (such as outpatient appointment rate 85%, registration rate 80%, doctor scheduling, etc.) is collected to generate a resource heat map. Based on the patient sequence after the updated priority, the patient priority and the hospital resource heat map are input into the multi-objective integer programming model. The multi-objective integer programming model uses the patient priority as the input variable and the medical resource status of the target hospital as the constraint condition to construct an objective function. The objective function aims to maximize the sum of the product of the patient's priority weight (representing the patient's urgency or demand level, such as high-risk 0.9, ordinary 0.5) and the medical resource satisfaction (calculated according to the hospital resource heat map, such as a comprehensive score of outpatient appointment status, doctor scheduling, etc., such as 0.9 when resources are abundant and 0.3 when resources are tight). Then, integer programming is used (the variable takes the value 0 or 1, representing whether the referral path is selected) to solve the model to obtain the optimal referral path, achieve the best match between patient needs and medical resources, and determine the optimal referral path between patient needs and medical resources (such as from secondary hospital → tertiary hospital A → tertiary hospital B);

[0086] (5) Automatically generate a complete referral report, including the optimal referral path, the patient's real-time urgency score (e.g., 87 points, high risk), an overview of the medical resource status of the referral hospital (tertiary hospital A outpatient appointment rate 85%, registration rate 80%), the expected risk reduction after referral (e.g., the risk of complications decreased by 25%), and the reasons for the referral recommendation.

[0087] Among them, the chronic disease upward referral standards based on which the referral recommendation generation model generates referral recommendations include: ① The system discovers for the first time that high-risk populations are actually ill but have not been diagnosed by doctors; ② The confirmed population or the population to be managed have a continuous high risk of disease progression; ③ The confirmed population or the population to be managed have a continuous high risk of disease progression; ④ The confirmed population or the population to be managed have failed to meet the disease control standards for consecutive periods; ⑤ The confirmed population or the population to be managed have acute exacerbations or critical illnesses of chronic diseases.

[0088] After receiving the referral suggestion and checking the prediction basis, the chronic disease doctor chooses to accept the referral suggestion or manually modify it to confirm and generate the final referral information. That is, the information is set according to the upstream and downstream referral paths and referral relationship mapping table to generate the final referral information. Among them, manual modification takes precedence over the generated referral suggestion and has the final decision-making power. When the chronic disease doctor rejects the referral suggestion, he must also manually fill in the reason and sign electronically.

[0089] The active referral module 13 automatically sends referral notifications to the patient and the senior physician, providing the attending physician's information. The responsible physician is responsible for guiding the patient through the consultation and recording the referral's completion. After the senior physician's consultation, the system records the treatment plan and provides feedback to the original responsible physician. This system also records the referral and treatment plan generated by the target medical institution in real time, synchronizing the plan with the chronic disease physician.

[0090] In the AI-based hierarchical chronic disease management and referral system of the present invention, the AI-assisted active referral process has the characteristics of prevention, efficiency, and precision, which improves the response speed and accuracy of medical services.

[0091] In this embodiment, the reverse incentive module 14 is used to provide performance incentives to chronic disease doctors based on whether they have completed the contract signing of chronic disease patients, implemented graded referrals, and saved medical expenses. Through the reverse incentive mechanism, the initiative and enthusiasm of the medical team are stimulated, and the referral path and management are continuously optimized.

[0092] The assessment of chronic disease doctors focuses on whether they have completed the signing of chronic disease patients, implemented hierarchical referrals, and saved medical expenses. The system automatically records the specific circumstances of the referral, the model's prediction results, whether the referral and modification records are made, the completion of follow-up, and the effectiveness of the referral. It is evaluated based on the allocated performance based on the system's recorded data. If chronic disease doctors save medical expenses, they will receive positive performance incentives based on the proportion of medical expenses. On the contrary, they will be subject to a certain degree of reverse deduction. Specifically, the referral acceptance rate and the amount of medical expenses saved are counted, and the performance incentive amount is calculated based on the basic salary, contract signing rate, referral acceptance rate, and amount of medical expenses saved.

[0093] Among them, the amount of medical cost savings = the average cost of similar patients in the same period last year - the current actual cost; the referral acceptance rate = the number of recommended referrals / the number of referrals actually performed by doctors.

[0094] Performance incentive amount = basic salary × (1 + signing rate × a) + (referral acceptance rate × b%) + (medical cost savings × c yuan) + d, where a, b, c, and d are all preset coefficients or amounts, and are adjusted regionally based on the regional financial and health service status.

[0095] The AI-based hierarchical chronic disease management and referral system provided in the above embodiment, through AI-assisted hierarchical referral, assists the intelligent flow of patients upstream and downstream of the referral system based on disease progression prediction, optimizes the allocation and utilization of medical resources, builds a close-knit medical community, and realizes the complementary advantages of urban and rural medical care; screens the target population for chronic disease management, implements classified management, reduces the incidence and mortality and disability rates of chronic diseases, and alleviates the heavy medical burden caused by chronic diseases.

[0096] The above-mentioned embodiment provides an artificial intelligence-based hierarchical chronic disease management and referral system to optimize the process and monitoring of stratified diagnosis and treatment of chronic diseases, reduce the disease burden caused by death and disability due to chronic diseases, reduce personal health expenditures and medical cost burdens; convert medical cost savings into performance feedback and stimulate the initiative of the medical team.

[0097] The embodiment also provides a chronic disease management and referral method using the above-mentioned hierarchical chronic disease management and referral system, such as Figure 2 As shown, based on clinical guidelines for chronic diseases, an algorithm is used to identify chronic disease management targets. The number of confirmed chronic disease patients is based on physicians' diagnoses. The number of patients awaiting management is determined from the pool of suspected patients who have yet to be effectively managed. High-risk chronic disease patients are identified through risk assessment based on clinical findings. For those diagnosed with chronic diseases and those awaiting management, contracts with the chronic disease physician management system are ensured. For those at high risk, a strategy focusing on supervision and education is employed to achieve contractual binding whenever possible, forming an AI-powered proactive management model. Building on the traditional manual referral model from the medical insurance system and social service family medicine system to district hospitals, an AI-assisted proactive referral mechanism is introduced. When a patient's condition changes, a rapid assessment process is initiated to make a preliminary decision on whether to refer the patient, and the results are submitted to the physician for review. If the AI ​​and manual judgments disagree, the physician's judgment is implemented and manually signed and confirmed. Ultimately, patients with persistently high risk or those without follow-up are promptly referred to higher-level hospitals.

[0098] Taking the various medical communities in a certain district as an example, the medical data of 1.26 million people in the district (including medical institution data, public health data, physical examination data, and medical insurance data) were included. Through the target population classification and contract management module, the chronic disease model was used for dynamic risk prediction, and 400,000 target people for chronic disease management were screened. These people include: 180,000 confirmed people with chronic diseases who have been diagnosed by doctors, 60,000 people waiting to be managed who actually meet the diagnostic criteria for chronic diseases but have not yet been diagnosed by doctors, and 160,000 high-risk people who do not currently meet the diagnostic criteria for chronic diseases but are identified as high-risk by AI. In the hierarchical referral path setting module, hospitals are allocated in a hierarchical manner, and upstream and downstream referral chains are constructed. A separate referral chain is established for high-risk patients, including information such as responsible doctors, affiliated hospitals, and superior referrals.

[0099] For chronic diseases such as diabetes, hypertension, chronic kidney disease, and cardiovascular and cerebrovascular diseases, the system implements proactive management and referral mechanisms based on the active referral module. New patients are dynamically added to the management pool from medical insurance and family doctor populations. If a patient remains at high risk or has not received follow-up, the system will proactively refer the patient to a higher-level district hospital.

[0100] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An artificial intelligence-based hierarchical chronic disease management and referral system, characterized by: include: The target population classification and contract management module is used to predict the dynamic risk of chronic diseases based on the multi-source and multi-modal data of the target population and classify the target population, generate chronic disease management suggestions for each type of population and push them to chronic disease doctors for contract management, including: using the chronic disease prediction model to predict the dynamic risk of chronic diseases based on the multi-source and multi-modal data of the target population, specifically: first, based on the three types of structured multimodal data of the target population's historical diagnostic data, biochemical indicators and follow-up information, a multimodal health trajectory in the form of a time series is constructed; secondly, the chronic disease prediction model adopts a fusion of dynamic risk coding mechanism and attention mechanism, and inputs the multimodal health trajectory data into the chronic disease prediction model according to a certain time window, firstly after encoding and sampling the encoding features An attention mechanism is used to capture the nonlinear variation characteristics of key variables of interest in data under different time windows. The convolutional neural network and long short-term memory network adopted by the multi-task learning head are used to fuse the nonlinear variation characteristics of each key variable. Based on the fusion results, the dynamic risk of chronic diseases and health status trends in future time periods are predicted. The dynamic risk of chronic diseases is expressed in the form of probability, where a larger probability indicates a higher risk. An adaptive time decay function is also embedded in the chronic disease prediction model, introducing a weight reduction mechanism for long-term historical data. Different decay weights are set for the input data in the time window according to the distance from the prediction time to adjust the degree of influence on the prediction result. Specifically, the farther the input data is from the prediction time, the larger the decay weight is set. A hierarchical referral path setting module is used to provide a hierarchical referral path setting function. Based on the referral path setting function, an upstream and downstream referral path consisting of medical institutions is set, and the responsible doctors corresponding to chronic diseases are bound to each level of medical institutions to form a referral relationship mapping table; Active referral module, which is used to generate referral suggestions based on the dynamic risk changes of chronic diseases in each group of people and push them to chronic disease doctors. Chronic disease doctors determine the referral suggestions and set referral information based on upstream and downstream referral paths and referral relationship mapping tables, and simultaneously notify the target diagnosis and treatment institutions and patients of the referral; The reverse incentive module is used to provide performance incentives to chronic disease doctors based on whether they have completed the signing of chronic disease patients, implemented tiered referrals, and saved medical expenses.

2. The artificial intelligence-based hierarchical chronic disease management and referral system according to claim 1 is characterized in that: The target population classification and contract management module predicts the dynamic risk of chronic diseases based on the multi-source and multi-modal data of the target population and classifies the target population. Chronic disease management suggestions are generated for each group of people and pushed to chronic disease doctors for contract management. It also includes: The target population whose chronic disease dynamic risk prediction results are lower than the first threshold are classified as healthy people. For the healthy population, follow-up reminders and follow-up projects are regularly sent to chronic disease doctors and patients, and follow-up project data of the healthy population are actively tracked, and the prediction of progression to high-risk groups is observed; The target population whose dynamic risk prediction results for chronic diseases exceed the second threshold is classified as a high-risk population. Chronic disease doctors are signed with the high-risk population on a voluntary basis. Follow-up frequency is increased based on the personalized disease risk of high-risk individuals. Follow-up frequency data of high-risk individuals is actively tracked. If follow-up is not completed within a certain number of expected follow-up intervals, a warning value will be triggered and chronic disease doctors will be proactively reminded to pay attention to high-risk individuals. The population that has been clinically diagnosed with the disease is classified as the confirmed population, and the population that is actually ill but has not been diagnosed by a doctor is classified as the population to be managed. All the confirmed population and the population to be managed are required to sign contracts with chronic disease doctors, and sign contracts and long-term management are achieved through telephone mobilization, APP reminders, tracking of medical records, and high-risk health education.

3. The artificial intelligence-based hierarchical chronic disease management and referral system according to claim 2 is characterized in that: Long-term management of confirmed cases and those awaiting care includes: Gradual management is implemented based on the progression of chronic diseases. Specifically, based on the consensus and guidelines of a multidisciplinary clinical team of doctors as prior knowledge, cardiovascular disease risk stratification is determined for patients with different chronic disease progression according to the cardiovascular disease risk stratification model, which includes low risk, moderate risk, high risk, and very high risk. Follow-up frequencies and follow-up items with different requirements are carried out according to the stratification results. Among them, the risk level of disease progression is determined according to the risk stratification model, including: (1) Construct a knowledge map based on clinician consensus to identify chronic disease risk factors and their importance; (2) Using the historical patient database, with the patient's physiological indicators as independent variables and each type of chronic disease risk factor judged by complications as the dependent variable, a logistic regression model was constructed to obtain the regression coefficient of each indicator for each type of chronic disease risk factor. The absolute value of each regression coefficient was then normalized to obtain the contribution ratio of each indicator to each type of chronic disease risk factor. (3) Based on the importance of each type of chronic disease risk factor and the contribution of each indicator to each type of chronic disease risk factor, the patient's real-time risk score is calculated using the weighted summation method; (4) Apply the entropy weight method to dynamically analyze the risk score distribution of all patients and set dynamic thresholds for risk stratification based on the risk score distribution; (5) Real-time collection of individual patient physiological indicator data and calculation of real-time risk scores, accurately stratifying patients into corresponding risk levels according to the dynamic thresholds of risk stratification.

4. The artificial intelligence-based hierarchical chronic disease management and referral system according to claim 1 is characterized in that: The hierarchical referral path setting module is used to set the upstream and downstream referral paths when the super administrator with high authority logs in. The set upstream and downstream referral paths include: community health service stations at level 1 → community health service centers at level 2 → district hospitals at level 3 → tertiary hospitals at level 4; referral hospitals and target responsible doctor teams are also separately implemented for high-risk groups for chronic diseases; When the chronic disease doctor logs in, the hierarchical referral path setting module fills in the list of superior referral doctors, and the nurse fills in the assistance management path.

5. The artificial intelligence-based hierarchical chronic disease management and referral system according to claim 1 is characterized in that: In the active referral module, referral recommendations are generated based on the dynamic risk changes of chronic diseases in each group of people and pushed to chronic disease doctors, including: When the confirmed population or the population under management is judged to be at continuous high risk based on the dynamic risk changes of chronic diseases, or if follow-up is not completed, they will be judged as high-risk. When the high-risk population is continuously at high risk, they will also be judged as high-risk. The referral recommendation generation model is used to generate referral recommendations for people in high-risk situations and push them to chronic disease doctors. The referral recommendations include patient risk analysis reports containing indicator trend charts and risk assessments, recommended referral hospitals and doctors matched according to pre-set hierarchical referral paths, and recommended referral times dynamically recommended based on the source data of the superior hospital number.

6. The artificial intelligence-based hierarchical chronic disease management and referral system according to claim 5 is characterized in that: The process of the referral recommendation generation model to generate referral recommendations for people in high-risk situations is as follows: (1) Calculate the medical gap index based on the difference in grade and disease handling capacity between the patient's current medical institution and the target medical institution, where the medical gap index is the product of the grade difference between the two medical institutions and the disease handling capacity; (2) Real-time monitoring of patients' key physiological indicators and complication risk scores, and inputting the monitoring data into the gated recurrent neural network in combination with the medical gap index, outputting the emergency score in real time, and identifying high-risk patients based on the emergency score; (3) Input the urgency score into the asymmetric priority algorithm. For high-risk patients, the urgency score increases by a given value, and the queue priority weight increases by a given multiple, thereby updating the priority order of the patient queue. (4) The medical resource status of the target hospital is collected and a resource heat map is generated. Based on the patient sequence after the updated priority, the patient priority and the hospital resource heat map are input into the multi-objective integer programming model. The multi-objective integer programming model uses the patient priority as the input variable and the medical resource status of the target hospital as the constraint condition to construct an objective function. The objective function aims to maximize the sum of the product of the patient's priority weight and the medical resource satisfaction. Then, integer programming is used to solve the model to obtain the optimal referral path, achieve the best match between patient needs and medical resources, and determine the optimal referral path between patient needs and medical resources. (5) Automatically generate referral reports, including the optimal referral path, the patient's real-time urgency score, an overview of the medical resource status of the referral hospital, the expected risk reduction after referral, and the reasons for the referral recommendation.

7. The artificial intelligence-based hierarchical chronic disease management and referral system according to claim 1 is characterized in that: In the active referral module, the chronic disease doctor determines the referral recommendation and sets the referral information based on the upstream and downstream referral paths and referral relationship mapping table, and simultaneously notifies the target diagnosis and treatment institution and patient of the referral, including: After receiving the referral suggestion and reviewing the prediction basis, the chronic disease doctor can choose to accept the referral suggestion or manually modify it to confirm and generate the final referral information. That is, the final referral information is generated by setting information based on the upstream and downstream referral paths and referral relationship mapping table. Manual modification takes precedence over the generated referral suggestion. When chronic disease doctors reject referral recommendations, they must manually fill in the reasons and sign electronically.

8. The artificial intelligence-based hierarchical chronic disease management and referral system according to claim 1 is characterized in that: The active referral module also records in real time when the target diagnosis and treatment institution accepts the referral and generates a treatment plan, and synchronizes the treatment plan to the chronic disease doctor.

9. The artificial intelligence-based hierarchical chronic disease management and referral system according to claim 1 is characterized in that: In the reverse incentive module, performance incentives are provided to chronic disease doctors based on whether they have completed the signing of chronic disease patients, implemented graded referrals, and saved medical expenses, including: The referral acceptance rate and medical cost savings are calculated based on the specific referral circumstances recorded in the system, the model prediction results, whether the referral is made and the record is modified, the follow-up completion status, and the effectiveness of the referral. The performance incentive amount is calculated based on the basic salary, contract signing rate, referral acceptance rate, and medical cost savings.

Images