Cross-institutional referral intelligent routing method and system based on medical agent middleware

Abstract

The application provides a cross-institutional referral intelligent routing method and system based on a medical intelligent agent middleware, relates to the technical field of medical informatization, and comprises the following steps: establishing a sign data monitoring window, mapping sign index changes and medication record time sequences, constructing a drug reaction sensitivity curve to calculate a treatment effective coefficient; adopting a dynamic time warping algorithm to match a drug-sign correlation graph, and generating a treatment decision tree; and predicting disease development, identifying the best referral opportunity and target institution based on the treatment effective coefficient and specialty characteristics. The application improves cross-institutional referral efficiency, optimizes medical resource allocation, and enhances patient treatment effect.

Description

TECHNICAL FIELD

[0001] The present application relates to medical informatization technology, and in particular to a cross-institutional referral intelligent routing method and system based on a medical intelligent agent middleware. BACKGROUND

[0002] With the promotion of hierarchical diagnosis and treatment policy, cross-institutional referral has become an important means to optimize medical resource allocation. However, the current referral process between medical institutions still has problems such as information asymmetry and lack of scientific basis for referral decision-making, making it difficult to ensure the timeliness and accuracy of referral.

[0003] The existing referral system mainly matches institutions based on static rules, and does not fully consider the individualized treatment response characteristics of patients, making it difficult to predict the effectiveness of subsequent treatment after referral. At the same time, data barriers between medical institutions hinder the effective flow of patient clinical information, affecting referral quality and medical continuity.

[0004] There is a lack of intelligent referral decision support system between medical institutions, which cannot realize dynamic assessment of patient condition evolution and grasp of the best referral opportunity. Medical staff are difficult to accurately assess the matching degree of treatment capacity of different specialist medical institutions and patient needs, so there is an urgent need for a cross-institutional referral intelligent routing method based on a medical intelligent agent middleware. SUMMARY

[0005] The embodiments of the present application provide a cross-institutional referral intelligent routing method and system based on a medical intelligent agent middleware, which can solve the problems in the prior art.

[0006] In a first aspect, the embodiments of the present application provide a cross-institutional referral intelligent routing method based on a medical intelligent agent middleware, comprising:

[0007] Receiving referral application information through a medical intelligent agent middleware, the referral application information including patient medical record information and referral department information, wherein the patient medical record information includes physical sign data and medication records;

[0008] Establishing a monitoring window for physical sign data, time-series mapping the change rate of physical sign indicators with the administration time points and doses in the medication records, identifying the response delay time and peak level of physical sign indicators to different drug doses, constructing a drug response sensitivity curve, and calculating a treatment effectiveness coefficient based on the drug response sensitivity curve;

[0009] The diagnosis and treatment data of the candidate medical institutions are obtained from the medical institution collaborative database, the drug reaction sensitivity curve is segmented according to the fluctuation period, the time sequence feature vector is obtained by extracting the feature of each segment curve, the drug-sign correlation graph is constructed, the similarity between the drug-sign correlation graph and the historical correlation graph in the diagnosis and treatment data is calculated by using the dynamic time warping algorithm, the sign index set is determined according to the similarity matching result, and the treatment decision tree containing the disease evolution path is generated;

[0010] According to the treatment decision tree, the patient's disease development trajectory is predicted, the best referral opportunity is identified based on the treatment effective coefficient and the specialty of the candidate medical institution, the target medical institution is determined, and the data transmission channel is established by the medical intelligent agent middleware to complete data exchange.

[0011] In an optional embodiment,

[0012] A monitoring window of the sign data is established, the sign index change rate is time-mapped with the drug administration time point and the dose in the medication record, the response delay time and the peak level of the sign index to different drug doses are identified, and the drug reaction sensitivity curve is constructed, including:

[0013] The wavelet denoising is performed on the sign data to obtain smoothed sign data, the initial monitoring window is divided according to the interval of adjacent drug administration time points in the medication record, and the window size is dynamically adjusted based on the autocorrelation characteristics of the sign data to obtain the windowed sign sequence;

[0014] The time difference of the windowed sign sequence is calculated to obtain the sign change rate sequence, the drug administration dose in the medication record is normalized to obtain the dose sequence, the probability distribution difference between the sign change rate sequence and the dose sequence is calculated as the transmission loss, and the time sequence mapping relationship is obtained by minimizing the transmission loss;

[0015] In the time-frequency decomposition of the time sequence mapping relationship, the continuous trajectory with the maximum energy density is extracted as the feature ridge line, the starting point of the feature ridge line is marked as the response delay time, and the maximum point of the ridge line is marked as the peak response level;

[0016] According to the response delay time and the peak response level, a feature point set is constructed, the bandwidth of the kernel function is dynamically adjusted based on the local distribution rule of the feature point set, the feature point set is interpolated and fitted by using the kernel function, and the drug reaction sensitivity curve is constructed.

[0017] In an optional embodiment,

[0018] The treatment effective coefficient is calculated based on the drug reaction sensitivity curve, including:

[0019] The drug response sensitivity curve is segmented according to the curvature change points to obtain characteristic segments reflecting the drug action process, the mapping relationship between the sign change trend in each characteristic segment and the drug dose is calculated, and a dose response characteristic is generated;

[0020] A compensation coefficient is constructed, which changes with the response sensitivity of the sign index to the drug dose, the compensation coefficient is multiplied by the dose response characteristic to generate a compensated response value;

[0021] A pulse response sequence is constructed based on the compensated response value, which reflects the response mode of the sign index to single drug administration, the superposition effect of the pulse response sequence is analyzed to obtain a drug accumulation feature, and the drug accumulation feature is combined with the peak value of the pulse response sequence to obtain a treatment effective coefficient.

[0022] In an alternative embodiment,

[0023] The drug response sensitivity curve is segmented according to the fluctuation period, the time sequence feature vector is obtained by extracting the characteristics of each segment curve, and the drug-sign correlation graph is constructed, including:

[0024] The drug response sensitivity curve is subjected to wavelet transform to obtain periodic characteristics, and the fluctuation time window is identified based on the periodic characteristics, and the drug response sensitivity curve is divided into multiple subsequences according to the fluctuation time window;

[0025] The time sequence feature vector of the subsequence is extracted, and principal component analysis is used to reduce the dimension of the time sequence feature vector to obtain a reduced dimension feature vector;

[0026] An initial heterogeneous graph network is constructed based on the reduced dimension feature vector, which includes drug nodes and sign nodes, and the Pearson correlation coefficient is used to calculate the correlation value between the drug nodes and the sign nodes, and the correlation value is used as the initial weight between the drug nodes and the sign nodes;

[0027] The initial weight is dynamically adjusted using a graph attention mechanism to obtain a dynamic weight, the dynamic weight is combined with time encoding to calculate the corresponding delay response value, and a dynamic heterogeneous graph network is constructed according to the dynamic weight and the delay response value, the node interaction feature is extracted from the dynamic heterogeneous graph network using a multi-head attention mechanism, and a drug-sign correlation graph is constructed according to the node interaction feature.

[0028] In an alternative embodiment,

[0029] The similarity between the drug-sign correlation graph and the historical correlation graph in the diagnosis and treatment data is calculated using a dynamic time warping algorithm, the sign index set is determined according to the similarity matching result, and a treatment decision tree containing a disease evolution path is generated, including:

[0030] The drug-sign correlation graph is constructed as a time sequence matrix, a time sequence matching window is set, a dynamic time warping algorithm is executed in the time sequence matching window, and an initial distance value of the time sequence matrix and a historical correlation graph in diagnosis and treatment data is calculated;

[0031] A strip-shaped penalty function is constructed, the initial distance value is weighted and corrected according to the strip-shaped penalty function to obtain a corrected distance value, the corrected distance values under different time scales are calculated and weighted combined to obtain a graph similarity value;

[0032] According to the graph similarity value, historical cases are clustered and grouped, a sign change mode is extracted from the clustered and grouped cases to obtain a sign feature sequence, an information correlation degree between features in the sign feature sequence is calculated, the sign feature sequence is quantitatively scored based on the importance of the information correlation degree, and a sign index set is constructed by selecting features with a quantitative score higher than a preset score;

[0033] Based on the sign index set, a decision tree node is constructed, an index determination interval, a drug response probability and a time window parameter are set in the decision tree node, an initial decision tree is generated, the initial decision tree is pruned and optimized based on the minimum information principle, a structure constraint term is introduced to limit the branch size of the initial decision tree, and a treatment decision tree of a disease evolution path is obtained.

[0034] In an optional embodiment,

[0035] The initial decision tree is pruned and optimized based on the minimum information principle, a structure constraint term is introduced to limit the branch size of the initial decision tree, and a treatment decision tree of a disease evolution path is obtained, including:

[0036] The classification result, node transition time and subtree information of the non-leaf node of the initial decision tree are extracted, a node information vector containing a time sequence consistency constraint is constructed, the node information vector is weighted and combined based on the time sequence consistency constraint, and node information quantity is generated;

[0037] The information quantity change value of the non-leaf node under the minimum information principle is calculated, and an adaptive threshold curve is constructed according to the depth distribution and sample distribution of the initial decision tree, the information quantity change value is projected onto the adaptive threshold curve for comparison, pruning is performed when the projection result is lower than the adaptive threshold curve, and a pruned decision tree is obtained;

[0038] The number of leaf nodes, tree depth and path complexity of the pruned decision tree are extracted and combined to construct a structure constraint term, the pruned decision tree is dynamically planned and searched based on the structure constraint term, and a plurality of candidate decision trees are generated;

[0039] The feature statistics of the nodes of the candidate decision tree are performed, the information quantity gain upper limit is calculated combining the disease evolution law, the candidate decision tree is screened according to the information quantity gain upper limit, the decision tree with the minimum information quantity is selected as the optimal decision tree, the boundary optimization is performed on the branch of the optimal decision tree, and the treatment decision tree containing the disease evolution path is generated.

[0040] In an alternative embodiment,

[0041] According to the treatment decision tree, the patient's disease development trajectory is predicted, the best referral opportunity is identified based on the treatment effective coefficient and the specialty of the candidate medical institution, and the target medical institution is determined, including:

[0042] The pre-acquired sign index set is subjected to time sequence decomposition and feature extraction, the state transition probability and transition time of each node are calculated by inputting the treatment decision tree, and the multi-dimensional state space is constructed based on the state transition probability and transition time;

[0043] In the multi-dimensional state space, the state transition sequence is subjected to Monte Carlo sampling in combination with the treatment effective coefficient, and a set of disease development trajectories with weights is generated;

[0044] The historical diagnosis and treatment cases of the candidate medical institutions are extracted from the medical institution cooperation database, the historical diagnosis and treatment cases are clustered according to the disease types and treatment stages, the treatment success rate and disease duration of each category are calculated, and a specialty strength matrix is generated;

[0045] The set of disease development trajectories is mapped to the specialty strength matrix, the trajectory-specialty matching degree is calculated, the candidate medical institutions are dynamically sorted based on the matching degree, and the referral trigger threshold is set according to the treatment effective coefficient, when the treatment effective coefficient is lower than the referral trigger threshold and there is a candidate medical institution with a matching degree higher than a preset matching threshold, the current time is determined as the best referral opportunity, and the medical institution with the highest matching degree is selected as the target medical institution.

[0046] The second aspect of the embodiment of the application provides a cross-institution referral intelligent routing system based on a medical intelligent agent middleware, including:

[0047] The first unit is configured to receive referral application information through the medical intelligent agent middleware, the referral application information including patient medical record information and referral department information, wherein the patient medical record information includes sign data and medication records;

[0048] The second unit is configured to establish a monitoring window of the sign data, map the sign index change rate to the administration time points and doses in the medication records in time sequence, identify the response delay time and peak level of the sign index to different drug doses, construct a drug reaction sensitivity curve, and calculate a treatment effective coefficient based on the drug reaction sensitivity curve;

[0049] The third unit is configured to obtain diagnosis and treatment data of the candidate medical institution from the medical institution collaborative database, segment the drug reaction sensitivity curve according to a fluctuation period, extract features of each segment of the curve to obtain a time sequence feature vector, construct a drug-sign correlation graph, calculate a similarity between the drug-sign correlation graph and a historical correlation graph in the diagnosis and treatment data by using a dynamic time warping algorithm, determine a sign index set according to a similarity matching result, and generate a treatment decision tree containing a disease evolution path;

[0050] The fourth unit is configured to predict a patient condition development trajectory according to the treatment decision tree, identify an optimal referral time based on a treatment effective coefficient and a specialty of the candidate medical institution, determine a target medical institution, and establish a data transmission channel by the medical intelligent agent middleware to complete data exchange.

[0051] In a third aspect, an electronic device is provided, including:

[0052] a processor;

[0053] a memory for storing processor-executable instructions;

[0054] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0055] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0056] In the embodiment, the medical intelligent agent middleware is used to realize intelligent routing of cross-institution referral, the drug reaction sensitivity curve and the drug-sign correlation graph can be constructed, the response characteristics of the patient to the treatment can be accurately evaluated, the accuracy of the referral decision can be improved, and diagnosis and treatment delay and resource waste caused by blind referral can be avoided. By dynamically monitoring the time sequence relationship between the sign data change and the medication record, and combining the dynamic time warping algorithm, the accurate identification of the individualized treatment response mode of the patient is realized, the disease development trajectory can be scientifically predicted, data-driven decision support for patient distribution among medical institutions is provided, and the rationality and timeliness of the referral process are significantly improved. The medical intelligent agent middleware establishes a safe and efficient data transmission channel among different medical institutions, realizes seamless exchange of medical data on the premise of protecting the privacy of the patient, promotes the optimized allocation of medical resources, provides technical support for building a hierarchical diagnosis and treatment system and improving the overall quality of medical services, has significant social benefits and clinical application value. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1A flowchart of a cross-institutional referral intelligent routing method based on a medical intelligent agent middleware according to an embodiment of the present application is shown in

[0058] Figure 2 A schematic diagram of a drug-sign related graph construction system architecture is shown in

[0059] Figure 3 A simulation schematic diagram of drug-sign related graph similarity and sign index importance is shown in DETAILED DESCRIPTION

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0061] The technical solutions of the present application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and some embodiments may not be described again for the same or similar concepts or processes.

[0062] Figure 1 A flowchart of a cross-institutional referral intelligent routing method based on a medical intelligent agent middleware according to an embodiment of the present application is shown in Figure 1 The method comprises the following steps:

[0063] The referral application information is received through the medical intelligent agent middleware, the referral application information comprises patient medical record information and referral department information, wherein the patient medical record information comprises sign data and medication records;

[0064] A monitoring window of the sign data is established, the sign index change rate is time-mapped with the administration time point and the dose in the medication records, the response delay time and the peak level of the sign index to different drug doses are identified, a drug reaction sensitivity curve is constructed, and a treatment effective coefficient is calculated based on the drug reaction sensitivity curve;

[0065] The diagnosis and treatment data of the candidate medical institutions are obtained from a medical institution collaborative database, the drug reaction sensitivity curve is segmented according to the fluctuation period, the feature extraction is performed on each segment of the curve to obtain a time sequence feature vector, a drug-sign related graph is constructed, the similarity between the drug-sign related graph and the historical related graph in the diagnosis and treatment data is calculated by using a dynamic time warping algorithm, a sign index set is determined according to the similarity matching result, and a treatment decision tree containing a disease evolution path is generated;

[0066] According to the treatment decision tree, the patient's disease development trajectory is predicted, the best referral opportunity is identified based on the treatment effective coefficient and the specialty of the candidate medical institution, the target medical institution is determined, and the data transmission channel is established by the medical intelligent agent middleware to complete data exchange.

[0067] The medical intelligent agent middleware receives the referral application information from the medical institution through the standardized interface protocol. The application information is packaged in JSON format and includes the patient's unique identifier, basic information field, complete medical record data, and target referral department identifier. The patient's medical record information includes time series records of physiological parameters such as blood pressure, heart rate, body temperature, and blood oxygen saturation. Each record contains a measurement timestamp, value, measurement device identifier, and data quality identifier. The medication record details the drug name, administration time, dosage, administration method, and patient response record, forming a complete medication timeline.

[0068] In an alternative embodiment, a monitoring window for vital sign data is established, the rate of change of vital sign indicators is time-mapped with the administration time points and doses in the medication record, the response delay time and peak level of vital sign indicators to different drug doses are identified, and a drug response sensitivity curve is constructed, including:

[0069] Wavelet denoising is performed on the vital sign data to obtain smoothed vital sign data. The initial monitoring window is divided based on the interval between adjacent administration time points in the medication record. The window size is dynamically adjusted based on the autocorrelation characteristics of the vital sign data to obtain the windowed vital sign sequence.

[0070] The windowed vital sign sequence is calculated to obtain the vital sign change rate sequence. The administration dose in the medication record is normalized to obtain the dose sequence. The probability distribution difference between the vital sign change rate sequence and the dose sequence is calculated as the transmission loss. The time-mapping relationship is obtained by minimizing the transmission loss.

[0071] In the time-frequency decomposition of the time-mapping relationship, the continuous trajectory with the maximum energy density is extracted as the feature ridge line. The starting point of the feature ridge line is marked as the response delay time, and the maximum point of the ridge line is marked as the peak response level.

[0072] According to the response delay time and peak response level, a feature point set is constructed. The kernel function bandwidth is dynamically adjusted based on the local distribution rule of the feature point set. The feature point set is interpolated and fitted using the kernel function to construct a drug response sensitivity curve.

[0073] In an embodiment, the wavelet denoising processing is performed on the vital sign data to obtain smoothed vital sign data. Specifically, a db4 wavelet basis function is adopted, the original vital sign data is decomposed for 5 layers, a threshold of 0.05 is set for high-frequency coefficients, soft threshold shrinkage is performed on the coefficients exceeding the threshold, and then wavelet reconstruction is performed to obtain the smoothed vital sign data. For example, for the blood pressure data of a patient, the original value is [135, 142, 138, 146, 140, 143, 141, 138] mmHg, and after wavelet denoising, [136.5, 140.2, 139.3, 143.8, 141.2, 142.1, 140.5, 138.8] mmHg is obtained, effectively eliminating the measurement noise.

[0074] The initial monitoring window is divided according to the interval of adjacent administration time points in the medication record. For example, the patient is administered 100 mg of drug A at 08:00 and 50 mg of drug A at 14:00, and the initial monitoring window is set to 6 hours. Considering that there may be individual differences in drug metabolism rate, the window size is dynamically adjusted based on the autocorrelation characteristics of the vital sign data. Specifically, the autocorrelation function of the vital sign data is calculated, and the time point at which the autocorrelation coefficient first falls below 0.3 is found as the actual window boundary. In actual application, if the autocorrelation coefficient of blood pressure data falls to 0.28 after 5 hours, the monitoring window is adjusted to 5 hours. In this way, multiple windowed vital sign sequences are generated for different administration events, each sequence corresponding to a complete drug response process.

[0075] The windowed vital sign sequence is calculated to obtain a vital sign change rate sequence. For blood pressure data with a sampling interval of 10 minutes, the difference between adjacent two measurement values is calculated and divided by the time interval to obtain a blood pressure change rate sequence with a unit of mmHg / hour. For example, the original blood pressure values [140, 142, 145, 143, 140] mmHg at adjacent 10-minute sampling points are converted to a change rate of [12, 18, -12, -18] mmHg / hour. At the same time, the administration dose in the medication record is normalized, for example, the dose of drug A is divided by the maximum recommended dose of 200 mg to obtain a normalized dose sequence [0.5, 0.25].

[0076] The probability distribution difference between the vital sign change rate sequence and the dose sequence is calculated as the transport loss. In a specific implementation, the kernel density estimation method is used to estimate the probability distribution of the vital sign change rate and the normalized dose respectively, and the Wasserstein distance between the two distributions is calculated as the transport loss. By iteratively adjusting the time offset, the time offset value that minimizes the transport loss is found, establishing a time sequence mapping relationship between the vital sign change rate and the drug dose. In an example, for drug A, it is found that when the time offset is 45 minutes, the transport loss is minimized, that is, the effect of the drug on blood pressure begins to appear about 45 minutes after administration.

[0077] Feature ridge is extracted in time-frequency decomposition of the time series. The transformed time series is analyzed in time-frequency domain using continuous wavelet transform. Morlet wavelet is used as mother wavelet and scale parameter is set as 1-64. Feature ridge is extracted by identifying the continuous region with maximum energy density in the transformed result. In the above patient case, the starting point of feature ridge on time axis is located at 45 minutes after drug administration, which is marked as response delay time; the maximum point of ridge appears at 2.5 hours after drug administration, at which the blood pressure rate reaches 20 mmHg / hour, which is marked as peak response level.

[0078] According to the analysis results of multiple drug administration records, multiple groups of response delay time and peak response level are collected to construct feature point set. For example, for three different dose (50mg, 100mg, 150mg) drug administration records, the extracted feature point set can be {(50, 50, 15), (100, 45, 20), (150, 40, 28)}, which represents dose, response delay time (minute) and peak response level (mmHg / hour). The local distribution rule of feature point set is used to dynamically adjust the bandwidth of kernel function. Specifically, the average distance of adjacent points in feature point set is calculated, and the bandwidth is increased in sparse area and decreased in dense area. The feature point set is fitted by interpolation using adaptive bandwidth Gaussian kernel function to construct drug response sensitivity curve.

[0079] Drug response sensitivity curve is represented as two functions: dose-response delay function and dose-peak response function. These two functions can predict the drug response characteristics under any dose. For example, for drug A, the predicted response delay under 75mg dose is about 47 minutes and the peak response level is about 17.5 mmHg / hour. Medical staff can customize individualized drug administration plan for patients according to these curves, such as for patients who need to quickly reduce blood pressure, they can choose a dose that can quickly reach peak response; for patients who need long-term stable control, they can choose a dose with moderate response delay and moderate peak response.

[0080] In this embodiment, the response relationship between patient sign data and drug dosage can be modeled with high precision, accurately depicting the dynamic influence of drugs on individual signs. The stability of sign data is improved through wavelet denoising processing, and the pertinence and sensitivity of time series analysis are improved through adaptive window division. The transmission loss minimization method between sign change rate and normalized dosage is used to automatically associate the time lag and intensity response of different drugs, thereby avoiding the subjectivity of traditional empirical judgment. By extracting high-energy feature ridges in time-frequency decomposition, the key time points and response amplitudes of sign response are accurately located, which helps to analyze the rhythmicity of drug efficacy. Further introduction of kernel function interpolation method based on local distribution rule adaptive bandwidth makes the constructed drug response sensitivity curve more continuous and interpretable, effectively supporting subsequent individualized treatment effect evaluation and dynamic referral decision, improving the intelligent level and response efficiency of cross-institutional medical collaboration.

[0081] In an alternative embodiment, calculating a treatment effective coefficient based on the drug response sensitivity curve comprises:

[0082] Segmenting the drug response sensitivity curve according to the curvature change points to obtain characteristic segments reflecting the drug action process, calculating the mapping relationship between the sign change trend and the drug dosage in each characteristic segment, and generating a dose response feature;

[0083] Constructing a compensation coefficient that varies with the response sensitivity of the sign index to the drug dosage, multiplying the compensation coefficient by the dose response feature to generate a compensated response value;

[0084] Constructing a pulse response sequence based on the compensated response value, the pulse response sequence reflecting the response mode of the sign index to single drug administration, analyzing the superposition effect of the pulse response sequence to obtain a drug accumulation feature, combining the drug accumulation feature with the peak value of the pulse response sequence to obtain a treatment effective coefficient.

[0085] In this embodiment, the method of calculating a treatment effective coefficient based on a drug response sensitivity curve can more accurately evaluate the effect of drug treatment. The method includes steps such as segmented analysis of the drug response sensitivity curve, calculation of the dose response feature, construction of the compensation coefficient, generation of the pulse response sequence, and analysis of the drug accumulation feature.

[0086] The segmented processing of drug response sensitivity curve is the basis for calculating the treatment effective coefficient. First, the patient's physical data under the action of different doses of drugs is collected, and the drug response sensitivity curve is drawn. By calculating the second derivative of the curve, the point where the curvature changes significantly is determined as the segmentation point. For example, for antihypertensive drugs, when the dose increases from 5 mg to 10 mg, the blood pressure reduction rate may change from 2 mmHg per hour to 3.5 mmHg per hour, and the turning point of this change rate can be used as the segmentation point. In practical application, the analysis of a group of hypertensive patient data shows that the typical antihypertensive drug response curve can be divided into three characteristic segments: initial slow response segment (0-5 mg), rapid response segment (5-15 mg), and stable response segment (more than 15 mg).

[0087] When calculating the dose response characteristics for each characteristic segment, the segmented linear fitting method is used. Within each characteristic segment, the ratio of the rate of change of the physical sign to the dose is calculated. Taking antihypertensive drugs as an example, in the rapid response segment (5-15 mg), the blood pressure reduction rate may be 1.2 mmHg per mg of dose, while in the stable response segment (more than 15 mg), this ratio may drop to 0.5 mmHg per mg. In this way, a dose response characteristic value is generated for each characteristic segment, reflecting the pharmacodynamic characteristics of different dose intervals.

[0088] The construction of the compensation coefficient takes into account the influence of individual differences on drug response. According to the patient's physiological parameters (such as age, weight, liver and kidney function), a compensation model is constructed. The compensation coefficient adjusts with the change of the response sensitivity of the physical sign to the drug dose. For example, for patients with liver dysfunction, the drug metabolism capacity is reduced, and the compensation coefficient may be between 1.2-1.5; while for patients with renal dysfunction, the drug excretion is blocked, and the compensation coefficient may reach 1.8-2.0. Practical application data shows that the average compensation coefficient of patients over 60 years old is 1.35, while the compensation coefficient of healthy adults aged 30-40 is close to 1.0. Multiplying these compensation coefficients with the previously calculated dose response characteristics, the compensated response values are obtained, making the evaluation results more consistent with the actual situation of the individual.

[0089] Based on the compensated response values, a pulse response sequence is constructed to reflect the dynamic response process of the physical sign to a single dose. Through time series analysis, the pattern of the physical sign changing over time is recorded. Taking oral hypoglycemic drugs as an example, the pulse response sequence of blood glucose change after a single dose may show a pattern of "first drop - reach the lowest point - slow rise", and the complete response cycle is 8 hours, among which the time to reach the lowest point is 2.5 hours after administration, and the blood glucose drop is 2.3 mmol / L. The system records the blood glucose values at time points such as 0.5h, 1h, 2h, 4h, 6h, 8h, etc. to construct the complete pulse response sequence.

[0090] The analysis of drug accumulation characteristics is the key to evaluate the long-term drug effect. The residual effect of the previous administration is superimposed on the effect of the new administration under the condition of multiple administrations. By convolution operation, the time shift and superposition of the single administration pulse response sequence are carried out to obtain the cumulative effect curve of multiple administrations. The actual data shows that for a drug with a half-life of 12 hours, under a twice-daily administration scheme, the drug plasma concentration reaches steady state by the fifth day, and the accumulation coefficient is about 1.8, i.e. the steady-state blood drug concentration is 1.8 times the peak concentration of single administration.

[0091] The calculation of the treatment effective coefficient integrates all the above factors. The drug accumulation characteristics and the peak value of the pulse response sequence are combined by weighting to generate the final treatment effective coefficient. This coefficient comprehensively reflects the intensity of the immediate effect and the long-term cumulative effect of the drug. For example, for a lipid-lowering drug, if the peak value of the pulse response of single administration is 0.8 mmol / L for total cholesterol reduction, and the drug accumulation characteristic value is 1.5, the calculated treatment effective coefficient is 1.2 (obtained by a specific weighting algorithm), indicating that the drug has good therapeutic effect under the recommended dose.

[0092] Based on the above technical solutions, the individual level drug treatment effect of the patient can be quantitatively evaluated, thereby assisting intelligent treatment path planning and referral decision-making. By segmenting the curvature change points of the drug response sensitivity curve, the key dynamic characteristics in the drug action process are extracted, and the expression ability of the non-linear pharmacodynamic response law is enhanced. Further combining the dose response characteristics and the compensation coefficient of the response sensitivity adjustment, the adaptability of the model to individual differences is effectively improved, so that the calculation result is closer to the real clinical response. By constructing the pulse response sequence, the reaction mode of the sign to single administration is systematically described, and the accumulation effect of the drug in the continuous administration process is superimposed and analyzed, and then the correlation between the sign peak value and the pharmacodynamic accumulation is comprehensively judged to form a quantifiable treatment effective coefficient. The coefficient reflects the treatment intensity and considers the sign response delay and accumulation change, which can be used to support more accurate efficacy prediction and institution selection, and improve the timeliness and scientificity of cross-institution referral.

[0093] In an optional implementation, the drug response sensitivity curve is segmented according to the fluctuation period, the time sequence feature vector is obtained by extracting the feature of each segment, and the drug-sign association map is constructed, including:

[0094] The periodicity feature of the drug response sensitivity curve is obtained by wavelet transform, the fluctuation time window is identified based on the periodicity feature, and the drug response sensitivity curve is divided into a plurality of subsequences according to the fluctuation time window;

[0095] The time sequence feature vector of the subsequence is extracted, the time sequence feature vector is reduced by principal component analysis to obtain a reduced feature vector;

[0096] An initial heterogeneous graph network is constructed based on the reduced dimension feature vectors, including drug nodes and symptom nodes, and a Pearson correlation coefficient is used to calculate the correlation value between the drug nodes and the symptom nodes, which is used as the initial weight between the drug nodes and the symptom nodes.

[0097] The initial weight is dynamically adjusted using a graph attention mechanism to obtain a dynamic weight, a delay response value corresponding to the dynamic weight is calculated in combination with time coding, and a dynamic heterogeneous graph network is constructed according to the dynamic weight and the delay response value. The multi-head attention mechanism is used to extract node interaction features from the dynamic heterogeneous graph network, and a drug-symptom association graph is constructed according to the node interaction features.

[0098] Figure 2 A system architecture diagram is constructed for the drug-symptom association graph. For example, wavelet transform is performed on the drug response sensitivity curve to obtain periodic features. Specifically, the original sensitivity curve is decomposed using discrete wavelet transform, and a suitable wavelet basis function such as Daubechies wavelet is selected to decompose the signal into coefficients of different frequency bands. By setting a threshold to filter the wavelet coefficients, significant fluctuation information is retained. For example, for a drug response curve containing 1000 time points, four-level wavelet decomposition is used to obtain a wavelet coefficient matrix. By analyzing the energy distribution of the wavelet coefficients, the main periodic components can be identified. Based on the local extreme point distribution of the wavelet coefficients, the fluctuation period is identified, such as for an antibiotic drug, two main fluctuation periods of 24 hours and 72 hours can be identified. According to the identified fluctuation characteristics, the size of the time window is determined, such as setting 12 hours as a window unit, and the entire curve is divided into several sub-sequences.

[0099] For each sub-sequence, a time series feature vector is extracted. The extracted features include two categories of statistical features and morphological features. Statistical features include mean, standard deviation, kurtosis, skewness, quartile, etc.; morphological features include peak number, fluctuation amplitude, rising rate, falling rate, curve curvature, etc. For example, for a 12-hour response window of a certain antihypertensive drug, the extracted feature vector may include an average blood pressure reduction value of 2.3 mmHg, a standard deviation of 1.1 mmHg, a maximum reduction of 4.7 mmHg, and an increase rate of 0.15 mmHg / hour, etc. More than 20 features. Due to the high dimensionality of the original features, principal component analysis technology is used for dimensionality reduction. First, all features are standardized, the feature covariance matrix is calculated, the eigenvalues and eigenvectors of the matrix are solved, and the appropriate number of principal components is selected according to the cumulative contribution rate (usually the principal components with a cumulative contribution rate of more than 85% are selected), and the original features are mapped to the principal component space. For example, the original 20-dimensional features are reduced to 8-dimensional representations, retaining 88.7% of the information content of the original data.

[0100] Based on the reduced feature vectors, an initial heterogeneous graph network is constructed. The graph network contains two types of nodes: drug nodes and symptom nodes. Each drug node represents a drug, and its attributes include drug molecular fingerprint, mechanism of action classification, half-life, etc.; each symptom node represents a physiological indicator, such as blood pressure, heart rate, blood glucose, etc., and its attributes include measurement unit, normal range value, etc. To establish the connection between nodes, the Pearson correlation coefficient between each pair of drug-symptom is calculated. The specific operation is as follows: collect a large amount of patient's symptom change data after taking medicine, for drug A and symptom B, extract the symptom B change sequence of all patients treated with drug A, and calculate the correlation coefficient between the reduced feature vector and the original value of the symptom. For example, the correlation coefficient of drug metronidazole and body temperature index may be 0.73, and the correlation coefficient with heart rate index is 0.32. Set a correlation threshold (such as 0.3), and only keep the connections with correlation higher than the threshold, and the correlation coefficient as the initial weight of the edge.

[0101] The initial weight is dynamically adjusted by using graph attention mechanism. A multi-layer graph attention network is constructed, each layer containing multiple attention heads. Each attention head independently learns the importance weight between nodes. For drug nodes and symptom nodes, first generate query vectors, key vectors and value vectors through linear transformation, calculate the dot product of query vectors and key vectors to get the original attention score, and apply softmax function to normalize the score to get the attention weight. This attention weight represents the importance of different nodes, combined with the initial network weight, to get the dynamically adjusted edge weight. For example, the weight of drug A and symptom C is originally 0.45, and after learning by graph attention mechanism, it is adjusted to 0.58, indicating that this association is more important than the initial evaluation.

[0102] The delay response value is calculated combined with time encoding. To capture the time dynamic characteristics of drug action, the time encoding mechanism is introduced. The absolute time position is encoded into a high-dimensional vector, and different frequencies of sine and cosine functions are used to capture changes at different time scales. Calculate the onset time and peak time of drug action, and construct a time difference matrix. For example, the onset time of drug B on symptom D is 1.5 hours, the peak time is 4 hours, and the duration is 10 hours. Combine these time characteristics with dynamic weights to get the time series association strength containing delay response information.

[0103] A dynamic heterogeneous graph network is constructed according to the dynamic weight and delay response value. In this network, nodes represent drugs or symptoms, and edges represent the association between them, and the weight of the edge contains both association strength and time response characteristics. Apply multi-head attention mechanism to this network, and each attention head is responsible for capturing different types of node interaction patterns. Aggregate the outputs of multiple attention heads to get the integrated node representation vector. For example, through 8-head attention mechanism, each attention head outputs 64-dimensional features, and after aggregation, 512-dimensional node interaction features are obtained.

[0104] Based on the node interaction features, a final drug-sign correlation graph is constructed. The node interaction features are subjected to cluster analysis to identify drug groups and associated sign groups with similar action modes. A correlation strength threshold is set to retain significant correlations. The final graph contains 80 drug nodes, 35 sign nodes, and 320 effective edges. Each edge is labeled with correlation strength, onset time delay, peak time, and duration. The graph structure is presented through visualization techniques. Different strengths of correlation are represented by edges of different thickness and color, facilitating doctors to quickly identify the impact of drugs on specific signs and time characteristics.

[0105] In this embodiment, by introducing wave cycle division and multi-level feature extraction mechanism, the accuracy and expression ability of modeling the relationship between drug action process and sign response are improved. Existing technologies mostly use static statistical features or fixed time windows to model the correlation between drugs and signs, which is difficult to fully capture the dynamic change rule of sign response under different cycles, and generally ignores the influence of response delay and weight dynamics when constructing drug-sign graphs, resulting in rigid graph structure and insufficient expression of individual differences. This application obtains the periodic variation features in the drug response sensitivity curve through wavelet transform, and constructs a wave fluctuation time window accordingly to realize adaptive identification of pharmacodynamic response rhythm. Further extraction of the time sequence feature vector of each subsequence and dimensionality reduction processing enhances the compactness and generalization ability of feature representation. In the graph construction process, Pearson correlation is first used as a static weight to establish the basic connection between drugs and signs, then the weight is dynamically adjusted through the graph attention mechanism, and the delay response information is extracted combined with time coding, so that the finally constructed dynamic heterogeneous graph network can express both structural correlation and response timing. Through the multi-head attention mechanism, the node interaction features are further extracted, and the drug-sign correlation graph constructed in this way not only retains the key correlation relationship, but also strengthens the expression ability of individual treatment response mode, significantly improving the application value in referral path optimization and treatment strategy selection.

[0106] In an alternative embodiment, a dynamic time warping algorithm is used to calculate the similarity between the drug-sign correlation graph and the historical correlation graph in the diagnosis and treatment data, and the sign index set is determined according to the similarity matching result to generate a treatment decision tree containing a disease evolution path.

[0107] The drug-sign correlation graph is constructed as a time series matrix, a time series matching window is set, and the dynamic time warping algorithm is executed within the time series matching window to calculate the initial distance value between the time series matrix and the historical correlation graph in the diagnosis and treatment data.

[0108] A strip-shaped penalty function is constructed, the initial distance value is weighted and corrected according to the strip-shaped penalty function to obtain a corrected distance value, the corrected distance values under different time scales are calculated and weighted combined to obtain a graph similarity value;

[0109] According to the graph similarity value, historical cases are clustered and grouped, a sign change pattern is extracted from the clustered and grouped cases to obtain a sign feature sequence, information correlation between features in the sign feature sequence is calculated, the sign feature sequence is quantitatively scored in importance based on the information correlation, and a sign index set is constructed by selecting features with a quantitative score in importance higher than a preset score value;

[0110] Based on the sign index set, a decision tree node is constructed, an index determination interval, a drug response probability and a time window parameter are set in the decision tree node, an initial decision tree is generated, the initial decision tree is pruned and optimized based on the principle of minimum information, a structure constraint term is introduced to limit the branch size of the initial decision tree, and a treatment decision tree of a disease evolution path is obtained.

[0111] In specific implementation, a dynamic time warping algorithm is used to calculate the similarity of a drug-sign association graph and a historical association graph in diagnosis and treatment data, and a sign index set is determined according to the similarity matching result, and finally a treatment decision tree containing a disease evolution path is generated.

[0112] The drug use records and sign data in the diagnosis and treatment process of a patient are obtained, and a drug-sign association graph is constructed, which reflects the influence relationship of a specific drug on sign indexes. The association graph is converted into a time series matrix, each row in the matrix represents a drug, each column represents a time point, and the matrix element value represents the influence degree of the drug on each sign index at the time point. For example, the influence value of drug A on sign X at time point t1 is 0.75, which means that there is a 75% possibility to cause the change of sign X. The system sets the time sequence matching window to 14 days, and the dynamic time warping algorithm is executed in the window to calculate the initial distance value between the current time series matrix and the association graph in the historical diagnosis and treatment data. The distance value is obtained by accumulating the point-to-point numerical difference, for example, for two sequences with a length of n, the square root of the sum of the square of the difference value of each corresponding point is taken as the initial distance value.

[0113] A strip-shaped penalty function is constructed to modify the initial distance value. The penalty function is defined as a weight function that increases with the time difference of matching points. When the time difference of matching points is less than 3 days, the penalty coefficient is 1.0; when the time difference is between 3-7 days, the penalty coefficient is 1.5; and when the time difference exceeds 7 days, the penalty coefficient is 2.0. By multiplying the initial distance value by the corresponding penalty coefficient, the modified distance value is obtained. The system calculates the modified distance value at different time scales, including day-level, week-level and month-level scales, and assigns weights of 0.5, 0.3 and 0.2 respectively for weighted combination to obtain the atlas similarity value. For example, for patient case P, the modified distance value with historical case H at day-level scale is 0.25, at week-level scale is 0.32, and at month-level scale is 0.40, then the weighted combined atlas similarity value is 1-(0.25x0.5+0.32x0.3+0.40x0.2)=0.695.

[0114] Based on the calculated atlas similarity value, the historical cases are clustered and grouped. The hierarchical clustering algorithm is adopted, and the similarity threshold is set to 0.7. The cases with similarity higher than the threshold are classified into the same group. From the clustering results, the system extracts the sign change pattern within each group to form a sign feature sequence. For example, in a certain clustering group, it is found that blood pressure, heart rate and oxygen saturation have significant time sequence change patterns. The system calculates the information correlation between these features, and uses mutual information as the evaluation index. The higher the mutual information, the closer the correlation between two features. For each sign feature, based on its mutual information with other features and its prediction ability in treatment results, the importance is quantified and scored. The score uses a standardized score of 0-100, for example, the importance score of blood pressure is 85, heart rate is 78, and oxygen saturation is 92. The system selects features with importance scores higher than the preset score of 80 to construct a sign index set, which in this example includes blood pressure and oxygen saturation.

[0115] Based on the selected set of physical indicators, a decision tree node is constructed. Each node is set to contain three key parameters: indicator judgment interval, drug response probability and time window parameter. For example, for the blood pressure indicator, the judgment interval is set to "systolic pressure > 140 mmHg and diastolic pressure > 90 mmHg", the drug response probability is "antihypertensive drug A: 78%, antihypertensive drug B: 65%", and the time window parameter is "expected response time: 24-48 hours". After the system generates the initial decision tree, pruning optimization is performed based on the minimum information principle. This principle calculates the information gain of each node, and when the information gain of a node is less than 0.05, pruning is performed. At the same time, the system introduces structural constraint terms to limit the branch size of the decision tree, such as limiting the maximum depth of the decision tree to 5 and each node to have a maximum of 3 child nodes. Through these optimization steps, a treatment decision tree containing the disease evolution path is finally generated, which can provide personalized treatment recommendations based on real-time physical data of patients.

[0116] In this embodiment, by converting the drug-physical sign association graph into a time series matrix and introducing a dynamic time warping algorithm, sequence alignment is performed within a limited matching window, effectively identifying the dynamic change trend between different historical graphs; through the strip penalty function and multi-scale weighting mechanism, the robustness of similarity calculation is enhanced, making the graph similarity more discriminative. On this basis, the physical sign change pattern of historical cases is mined through clustering analysis, and the importance is quantified combined with information correlation, effectively screening physical indicators with high contribution to disease evolution, enhancing the interpretability of the subsequent decision model and the reliability of the decision basis. The treatment decision tree finally constructed not only reflects the key nodes and response rules of the disease evolution path, but also controls the complexity of the model through pruning and structural constraints, improving the stability and generalization ability of the decision path, providing an efficient and reliable support means for disease trajectory prediction and precise treatment in cross-institutional referral.

[0117] In an alternative embodiment, the initial decision tree is pruned and optimized based on the minimum information principle, and structural constraint terms are introduced to limit the branch size of the initial decision tree, to obtain a treatment decision tree of the disease evolution path, including:

[0118] The classification result, node transition time and sub-tree information are extracted from the non-leaf nodes of the initial decision tree, a node information vector containing time sequence consistency constraints is constructed, the node information vector is weighted and combined based on the time sequence consistency constraints, and a node information amount is generated;

[0119] The information amount change value of the non-leaf node under the minimum information principle is calculated, and an adaptive threshold curve is constructed based on the depth distribution and sample distribution of the initial decision tree, the information amount change value is projected onto the adaptive threshold curve for comparison, pruning is performed when the projection result is lower than the adaptive threshold curve, and a pruned decision tree is obtained;

[0120] extracting a leaf node number, a tree depth and a path complexity of the pruned decision tree, combining to construct a structure constraint term, performing dynamic programming search on the pruned decision tree based on the structure constraint term, and generating a plurality of candidate decision trees;

[0121] performing feature statistics on nodes of the candidate decision tree, calculating an information gain upper bound combining a disease evolution rule, performing screening on the candidate decision tree according to the information gain upper bound, selecting an optimal decision tree with the minimum information, and performing boundary optimization on branches of the optimal decision tree to generate a treatment decision tree containing a disease evolution path.

[0122] Exemplarily, after the initial decision tree is constructed, key information is extracted from all non-leaf nodes for subsequent pruning optimization processing. Each non-leaf node contains classification result data, specifically recording the disease classification label, classification confidence and sample distribution corresponding to the node. The node transition time records the cumulative time information from the root node to the current node, reflecting the time dimension characteristics of disease evolution. The sub-tree information contains the number, depth distribution and sample category statistical information of all child nodes under the node.

[0123] The construction of the time sequence consistency constraint is based on the disease evolution rule in the medical field, which requires that the time transition between adjacent nodes in the decision tree must conform to the natural development process of the disease. For the decision tree of acute myocardial infarction patients, the transition time from the symptom appearance node to the electrocardiogram abnormality node is usually between 30 minutes to 2 hours, and the node transition beyond this time range is considered to violate the time sequence consistency constraint. By checking whether the transition time of each node is within the medical standard range, the node is assigned a time sequence consistency weight value. The weight of the node that meets the constraint is 1.0, the weight of the partially compliant node is between 0.5 and 0.9, and the weight of the non-compliant node is less than 0.5.

[0124] The construction of the node information vector encodes the classification result, node transition time and sub-tree information into a multi-dimensional vector form. The classification result part uses one-hot encoding to represent the disease category, the node transition time is converted to a value between 0 and 1 through standardization processing, and the sub-tree information is represented by a statistical feature vector including the number of child nodes, average depth and sample distribution entropy value. Based on the time sequence consistency constraint, these vector components are combined with weights. The weight allocation follows medical professional knowledge, and the key diagnostic nodes obtain higher weights, and the auxiliary examination nodes have relatively low weights.

[0125] The generation of the node information quantity is achieved by calculating the comprehensive complexity of the weighted node information vector, which reflects the importance and complexity of the node in the decision-making process. For nodes that contain multiple disease classification possibilities, the information quantity is higher, while for nodes that correspond to a single clear diagnosis, the information quantity is lower. The system calculates the variance contribution of each vector component and obtains the overall information quantity value according to the weighting coefficient.

[0126] The calculation of the information quantity change value is achieved by comparing the difference in node information quantity before and after pruning. The system simulates the removal of a specific node and calculates the change in the overall information quantity of the decision tree. If the overall information quantity of the decision tree decreases significantly after removing a node, it indicates that the node plays an important role in the decision-making process and should not be pruned. Conversely, nodes with little change in information quantity after removal can be considered for pruning to simplify the structure of the decision tree.

[0127] The construction of the adaptive threshold curve is based on the depth distribution and sample distribution characteristics of the initial decision tree. Nodes at different depth levels have different pruning threshold standards. Shallow nodes usually contain more important diagnostic information, so a higher pruning threshold is set. Deep nodes may contain over-fitted detailed information, so a relatively lower pruning threshold is set. Sample distribution also affects threshold setting. Nodes with uniform sample distribution obtain a higher threshold, and nodes with extremely uneven sample distribution have a lower threshold. The curve is constructed using piecewise linear interpolation, providing a smooth transition of the threshold between different depth levels of the tree.

[0128] The comparison process of projecting the information quantity change value onto the adaptive threshold curve is achieved by coordinate mapping, taking the node depth as the horizontal coordinate and the information quantity change value as the vertical coordinate to determine the position of the node in two-dimensional space. When the node position is below the threshold curve corresponding to the depth, it is determined that the node should be pruned. For a node with a depth of 5, if its information quantity change value is 0.3 and the threshold value corresponding to this depth is 0.5, the node is marked as a pruning candidate.

[0129] The generation of the pruned decision tree is achieved by traversing the initial decision tree layer by layer and applying the pruning decision rule. The pruning operation is performed from the leaf nodes to the root node, ensuring that the pruning process does not destroy the basic structure of the tree. The pruning operation replaces the selected subtree with a leaf node, and the label of the leaf node adopts the most frequently occurring class label in the subtree. The pruning process maintains the integrity of the decision path, ensuring that each path from the root node to the leaf node represents a complete disease evolution trajectory.

[0130] The construction of structural constraints comprehensively considers multiple structural feature parameters of the pruned decision tree. The number of leaf nodes reflects the final classification ability of the decision tree; too many leaf nodes may lead to overfitting, while too few leaf nodes may lead to underfitting. Tree depth affects the complexity of the decision-making process; excessively deep trees increase decision-making time and computational complexity. Path complexity is measured by calculating the average length and variance of all paths from the root node to the leaf nodes; a decision tree with a uniform distribution of path lengths has better balance. The structural constraints combine these parameters into a comprehensive evaluation index to guide subsequent decision tree optimization.

[0131] Dynamic programming search further optimizes the pruned decision tree based on structural constraints, seeking the structurally optimal tree shape while maintaining decision accuracy. The search algorithm maintains a state transition table, recording the optimal substructure solution under different tree configurations. For each node, the algorithm calculates the overall objective function value under the two options of keeping or removing the node, selecting the scheme with the better objective function value. The search process generates multiple candidate decision trees satisfying different structural constraints, providing alternative solutions for subsequent selection.

[0132] The statistical features of candidate decision trees include node type distribution, branch factor statistics, path length distribution, and leaf node purity analysis. Node type distribution reflects the proportion of different types of diagnostic nodes in the tree; branch factor statistics calculate the distribution of the number of child nodes for each internal node; and path length distribution analyzes the distance changes from the root node to each leaf node. Leaf node purity is measured by calculating the concentration of sample classes in each leaf node; high purity leaf nodes indicate good classification performance.

[0133] The integration of disease evolution patterns is achieved by introducing constraints from medical expertise, maintaining a standard pattern library of disease evolution, including typical developmental stages, key time points, and patterns of complication occurrence for various diseases. The calculation of the upper bound of information gain is based on the expected information gain under the theoretically optimal condition, considering the class distribution of the dataset and the complexity of the feature space. For a dataset containing five disease categories, the theoretically optimal upper bound of information gain is determined by calculating the maximum information gain under a uniform distribution.

[0134] The candidate decision tree selection process evaluates and ranks trees based on the upper bound of their information gain. The actual information gain of each candidate tree on the validation dataset is calculated and compared to the theoretical upper bound. Decision trees with information gain close to the upper bound indicate good learning ability, while those with information gain far below the upper bound may suffer from underfitting. The selection process prioritizes decision trees with high information gain and simple structure, seeking the optimal balance between accuracy and complexity.

[0135] The optimal decision tree is selected based on a comprehensive evaluation index that combines the information minimization objective and the structural constraint requirements. The comprehensive score is calculated for all candidate decision trees, including multiple dimensions such as classification accuracy, structural complexity, temporal consistency, and medical rationality. The decision tree with the highest score is selected as the optimal decision tree, which has the simplest structure and the most consistent branching logic with the disease evolution rule while ensuring diagnostic accuracy.

[0136] Boundary optimization is performed on the branches of the optimal decision tree, aiming to refine the decision boundaries to improve classification accuracy. The optimization process adjusts the split threshold of each internal node by fine-tuning the threshold to reduce the classification error rate. For split nodes of continuous features, the system searches for the optimal split point near the original threshold to maximize the purity of the split child nodes. Boundary optimization uses a local search algorithm to optimize the node split conditions while maintaining the tree structure, generating the final treatment decision tree that includes the disease evolution path.

[0137] In this embodiment, the temporal consistency constraint is introduced to weight the expression of non-leaf node information, making the node retention or pruning process more consistent with the actual trajectory of symptom development; the adaptive threshold curve is constructed based on the initial tree depth and sample distribution, achieving dynamic setting of pruning standards for different nodes, avoiding information loss while significantly reducing redundant nodes; further through the structure constraint term, the leaf node number, path complexity and tree depth are fused for dynamic programming search, selecting the tree model with the smallest information quantity and the optimal structure from multiple candidate schemes, strengthening the discriminability and interpretability of the model; finally, boundary optimization is performed based on the retention of key nodes of disease evolution, making the generated treatment decision tree not only have a simple structure, but also accurately reflect the disease evolution path, showing higher practical value and promotion potential in assisting clinical treatment plan formulation and individualized diagnosis and treatment process.

[0138] Figure 3 The figure is a simulation diagram of the similarity of drug-sign correlation graph and the importance of sign indicators. Eight sign indicators (blood pressure, heart rate, body temperature, blood oxygen, blood sugar, respiratory rate, liver function, and kidney function) are clearly distinguished by different shapes and colors of scatter points. The cluster center point of each indicator is larger, showing the typical distribution position of the indicator. From the scatter distribution, it can be observed that blood oxygen (0.78, 92), liver function (0.75, 88), and blood pressure (0.73, 85) have high similarity and high importance, indicating that they have significant value in predicting the disease evolution path.

[0139] The simulation analysis corrects the initial distance value by constructing a strip-shaped penalty function, and calculates the corrected distance value under multiple time scales, so as to accurately quantify the importance of the physical indicators. The chart clearly shows the physical indicator screening mechanism based on information correlation, and the differentiated contribution of different regional indicators to the construction of the treatment decision tree.

[0140] In an optional implementation, the patient's disease development trajectory is predicted according to the treatment decision tree, the best referral opportunity is identified based on the treatment effective coefficient and the specialty of the candidate medical institution, and the target medical institution is determined, including:

[0141] The pre-acquired physical indicator set is time-decomposed and feature-extracted, and input into the treatment decision tree to calculate the state transition probability and transition time of each node, and a multi-dimensional state space is constructed based on the state transition probability and transition time;

[0142] In the multi-dimensional state space, the state transition sequence is Monte Carlo sampled in combination with the treatment effective coefficient, generating a set of disease development trajectories with weights;

[0143] Historical diagnosis and treatment cases of candidate medical institutions are extracted from a medical institution collaboration database, and the historical diagnosis and treatment cases are clustered according to disease types and treatment stages, the treatment success rate and disease duration of each category are calculated, and a specialty strength matrix is generated;

[0144] The disease development trajectory set is mapped to the specialty strength matrix, the trajectory-strength matching degree is calculated, the candidate medical institutions are dynamically sorted based on the matching degree, and the referral trigger threshold is set according to the treatment effective coefficient. When the treatment effective coefficient is lower than the referral trigger threshold and there is a candidate medical institution with a matching degree higher than a preset matching threshold, the current time is determined as the best referral opportunity, and the medical institution with the highest matching degree is selected as the target medical institution.

[0145] In a specific embodiment, the medical system first acquires a set of physical indicators of a patient, including but not limited to blood pressure, heart rate, blood glucose, body temperature, blood oxygen saturation, and other key physiological parameters. The system time-decomposes these indicators, divides the continuously monitored data according to different time scales (hours, days, weeks), and extracts statistical features (mean, standard deviation, fluctuation trend) and waveform features (peak, valley, periodic change). For example, for a type 2 diabetes patient, the system can extract the blood glucose fluctuation pattern in the past 90 days, including a morning blood glucose mean of 8.2 mmol / L, a standard deviation of 1.3 mmol / L, an average afternoon peak of 10.5 mmol / L, and a minimum night value of 6.1 mmol / L.

[0146] The extracted features are input into a pre-established treatment decision tree. The decision tree consists of multiple nodes, each representing a disease state. Using the diabetic patient as an example, the decision tree includes nodes such as "good glycemic control," "significant glycemic fluctuations," "early diabetic nephropathy," and others. The system calculates the state transition probabilities between nodes, such as a 0.23 transition probability from "significant glycemic fluctuations" to "early diabetic nephropathy" with an estimated transition time of 18 months. Based on these calculations, the system constructs a multi-dimensional state space, where each dimension represents a direction of disease progression.

[0147] In the constructed multi-dimensional state space, a treatment effectiveness coefficient is introduced to evaluate the effectiveness of the current treatment plan. The treatment effectiveness coefficient is determined based on the patient's response to treatment and ranges from 0 to 1, where 0 indicates complete ineffectiveness and 1 indicates complete effectiveness. For example, the diabetic patient's oral hypoglycemic drug treatment effectiveness coefficient is 0.65, indicating that the treatment plan has some effect on controlling blood sugar but has not reached the ideal state. The system uses the Monte Carlo method to generate 10,000 possible disease progression trajectories based on state transition probabilities and treatment effectiveness coefficients. Each trajectory includes a sequence of states and their occurrence time points, and is weighted according to the treatment effectiveness coefficient. Trajectories with higher weights represent more likely disease progression paths under the current treatment effectiveness.

[0148] To evaluate the expertise of candidate medical institutions, the medical institution collaboration database is accessed to extract historical diagnosis and treatment cases from each institution. These cases are clustered by disease type (such as cardiovascular disease, endocrine disease, and nervous system disease) and treatment stage (early intervention, disease control, and complication management). For each category, the treatment success rate and average disease duration are calculated. The treatment success rate is defined as the proportion of cases that achieve the clinical expected goal, and the disease duration is the average number of days from admission to treatment goal achievement. For example, A medical institution has a success rate of 87% and an average disease duration of 45 days in the early intervention of diabetic nephropathy, while B medical institution has a success rate of 92% and an average disease duration of 38 days in this field. These data are organized into a specialty expertise matrix, with rows representing disease types and treatment stages, columns representing different medical institutions, and cell values representing comprehensive scores.

[0149] The generated set of disease progression trajectories is mapped to the specialty expertise matrix, and the trajectory-specialty matching degree is calculated. The matching degree calculation considers three factors: the treatment success rate of the medical institution in predicting the disease state, the matching degree of the average disease duration of the medical institution with the patient's expected disease duration, and the similarity between the medical institution's historical cases and the current patient. For each predicted trajectory, the matching degree score with each candidate medical institution is calculated, ranging from 0 to 100. For example, for the possible development state of "early diabetic nephropathy," the matching degree of A medical institution is 78 points, and that of B medical institution is 85 points.

[0150] The referral trigger threshold is set according to the current treatment effective coefficient. The lower the treatment effective coefficient, the higher the referral trigger threshold. For example, when the treatment effective coefficient is 0.65, the referral trigger threshold is set to 80 minutes. The system continuously monitors the patient's condition and updates the treatment effective coefficient, and when the coefficient drops to 0.52 (lower than the safety threshold 0.6) and there is a matching B medical institution with a matching degree of 85 minutes (higher than the preset matching threshold 80 minutes), the system determines that the current is the best referral opportunity, and identifies B medical institution as the target referral institution. The system generates a referral recommendation report, including patient current condition analysis, disease development prediction, recommended medical institution and its expertise explanation, and referral opportunity rationality analysis.

[0151] In this embodiment, the evolution trend of the patient's condition can be predicted with high accuracy, and the treatment effectiveness and medical institution capacity are intelligently evaluated to dynamically identify the best timing and optimal destination for referral. By fusing time series sign analysis, state transition modeling and Monte Carlo sampling, the individualized disease development path of the patient can be fully restored, and the reliability of the prediction is improved; the specialty matrix is introduced and matched with the disease trajectory to realize the precise connection between medical resources and disease evolution; the referral threshold is set by the treatment effectiveness coefficient to realize the real-time perception of the decline in clinical efficacy, and ensure that the referral decision is timely and based. The overall scheme strengthens the intelligent collaboration of the cross-institution referral process, significantly improves the resource utilization efficiency and patient prognosis level.

[0152] In a second aspect of the embodiment of the present application, a cross-institution referral intelligent routing system based on a medical intelligent agent middleware is provided, which comprises:

[0153] The first unit is configured to receive referral application information through the medical intelligent agent middleware, the referral application information comprising patient medical record information and referral department information, wherein the patient medical record information comprises sign data and medication records;

[0154] The second unit is configured to establish a monitoring window for the sign data, time series map the sign index change rate with the drug administration time points and doses in the medication records, identify the response delay time and peak level of the sign index to different drug doses, construct a drug reaction sensitivity curve, and calculate a treatment effective coefficient based on the drug reaction sensitivity curve;

[0155] The third unit is configured to obtain diagnosis and treatment data of candidate medical institutions from a medical institution collaborative database, segment the drug reaction sensitivity curve according to fluctuation periods, extract features of each segment of the curve to obtain a time series feature vector, construct a drug-sign association graph, calculate the similarity between the drug-sign association graph and historical association graphs in the diagnosis and treatment data using a dynamic time warping algorithm, determine a set of sign indexes according to the similarity matching result, and generate a treatment decision tree containing a disease evolution path;

[0156] A fourth unit is configured to predict a patient's disease development trajectory according to a treatment decision tree, identify an optimal referral opportunity based on a treatment effective coefficient and a specialty of a candidate medical institution, determine a target medical institution, and establish a data transmission channel by a medical intelligent agent middleware to complete data exchange.

[0157] In a third aspect, the present application provides an electronic device, comprising:

[0158] a processor;

[0159] a memory for storing processor-executable instructions;

[0160] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0161] In a fourth aspect, the present application provides a computer-readable storage medium having stored thereon computer program instructions, which, when executed by a processor, implement the method described above.

[0162] The present application can be a method, apparatus, system and / or computer program product. The computer program product can include a computer-readable storage medium having stored thereon computer-readable program instructions that, when executed by a computer, cause the computer to carry out various aspects of the present application.

[0163] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A cross-institutional referral intelligent routing method based on medical intelligent agent middleware, characterized in that, include: The referral application information is received through the medical intelligent agent middleware. The referral application information includes patient medical record information and referral department information, wherein the patient medical record information includes vital sign data and medication records. Establish a monitoring window for vital signs data, map the rate of change of vital signs indicators to the time points and dosages in the medication records, identify the response delay time and peak level of vital signs indicators to different drug doses, construct a drug response sensitivity curve, and calculate the treatment effectiveness coefficient based on the drug response sensitivity curve; The system obtains diagnosis and treatment data of candidate medical institutions from the collaborative database of medical institutions, segments the drug response sensitivity curve according to the fluctuation cycle, extracts features from each segment of the curve to obtain time-series feature vectors, constructs a drug-sign association map, uses the dynamic time warping algorithm to calculate the similarity between the drug-sign association map and the historical association map in the diagnosis and treatment data, determines the set of sign indicators based on the similarity matching results, and generates a treatment decision tree containing the disease evolution path. Based on the treatment decision tree, the patient's disease progression trajectory is predicted. Based on the treatment effectiveness coefficient and the specialty of candidate medical institutions, the best referral time is identified, the target medical institution is determined, and the data exchange is completed by establishing a data transmission channel through the medical intelligent agent middleware. The calculation of the treatment effectiveness coefficient based on the drug response sensitivity curve includes: The drug response sensitivity curve is segmented according to the curvature change point to obtain the characteristic segment reflecting the drug action process. The mapping relationship between the change trend of vital signs and the drug dosage in each characteristic segment is calculated to generate dose response characteristics. A compensation coefficient is constructed, which varies with the sensitivity of vital signs to drug dosage. The compensation coefficient is multiplied by the dose response characteristics to generate a compensated response value. Based on the compensated response values, an impulse response sequence is constructed, which reflects the response pattern of vital signs to a single dose. The superposition effect of the impulse response sequence is analyzed to obtain drug accumulation characteristics. The drug accumulation characteristics are combined with the peak values ​​of the impulse response sequence to obtain the therapeutic efficacy coefficient.

2. The method according to claim 1, characterized in that, A monitoring window for vital signs data is established, and the rate of change of vital signs indicators is time-series mapped with the dosing time points and dosages in the medication records. The response delay time and peak level of vital signs indicators to different drug doses are identified, and drug response sensitivity curves are constructed, including: Wavelet denoising was performed on the vital signs data to obtain smoothed vital signs data. The initial monitoring window was divided according to the interval between adjacent dosing time points in the medication record. The window size was dynamically adjusted based on the autocorrelation characteristics of the vital signs data to obtain the windowed vital signs sequence. The windowed vital sign sequence is calculated by time difference to obtain the vital sign change rate sequence. The dosage in the medication record is normalized to obtain the dosage sequence. The difference in probability distribution between the vital sign change rate sequence and the dosage sequence is calculated as the transmission loss. The time-series mapping relationship is obtained by minimizing the transmission loss. In the time-frequency decomposition of the time-series mapping relationship, the continuous trajectory with the highest energy density is extracted as the feature ridge. The starting point of the feature ridge is marked as the response delay time, and the maximum value point of the ridge is marked as the peak response level. A feature point set is constructed based on the response delay time and peak response level. The kernel function bandwidth is dynamically adjusted based on the local distribution pattern of the feature point set. The kernel function is then used to interpolate and fit the feature point set to construct a drug response sensitivity curve.

3. The method according to claim 1, characterized in that, The drug response sensitivity curve is segmented according to its fluctuation period to extract features from each segment, resulting in a time-series feature vector. A drug-sign association map is then constructed, including: Wavelet transform is performed on the drug response sensitivity curve to obtain periodic features. Based on the periodic features, fluctuation time windows are identified, and the drug response sensitivity curve is divided into multiple subsequences according to the fluctuation time windows. Extract the temporal feature vector of the subsequence, and use principal component analysis to reduce the dimensionality of the temporal feature vector to obtain the dimensionality-reduced feature vector; An initial heterogeneous graph network is constructed based on the dimensionality-reduced feature vector, which includes drug nodes and vital sign nodes. The correlation value between the drug nodes and vital sign nodes is calculated using the Pearson correlation coefficient, and the correlation value is used as the initial weight between the drug nodes and vital sign nodes. The initial weights are dynamically adjusted using a graph attention mechanism to obtain dynamic weights. The delay response value corresponding to the dynamic weights is calculated by combining time coding. A dynamic heterogeneous graph network is constructed based on the dynamic weights and delay response value. A multi-head attention mechanism is used to extract node interaction features from the dynamic heterogeneous graph network. A drug-sign association map is constructed based on the node interaction features.

4. The method according to claim 1, characterized in that, The dynamic time warping algorithm is used to calculate the similarity between the drug-sign association map and the historical association map in the diagnosis and treatment data. Based on the similarity matching results, the set of sign indicators is determined, and a treatment decision tree containing the disease evolution path is generated, including: The drug-sign association map is constructed as a time series matrix. A time series matching window is set, and a dynamic time warping algorithm is executed within the time series matching window to calculate the initial distance value between the time series matrix and the historical association map in the diagnosis and treatment data. Construct a strip penalty function, and perform weighted correction on the initial distance value according to the strip penalty function to obtain the corrected distance value. Calculate the corrected distance values ​​at different time scales and perform weighted combination to obtain the map similarity value. Historical cases are clustered and grouped according to the map similarity value. The pattern of vital signs change is extracted from the clusters to obtain the vital sign feature sequence. The information correlation between features in the vital sign feature sequence is calculated. The importance of the vital sign feature sequence is quantitatively scored based on the information correlation. Features with importance quantitative scores higher than the preset score are selected to construct a set of vital sign indicators. Decision tree nodes are constructed based on the set of vital signs indicators. In the decision tree nodes, indicator judgment intervals, drug response probabilities, and time window parameters are set to generate an initial decision tree. The initial decision tree is pruned and optimized based on the principle of minimum information content. Structural constraint terms are introduced to limit the branch size of the initial decision tree to obtain the treatment decision tree of the disease evolution path.

5. The method according to claim 3, characterized in that, Based on the principle of minimum information content, the initial decision tree is pruned and optimized. Structural constraints are introduced to limit the branch size of the initial decision tree, resulting in a treatment decision tree for the disease evolution path, including: The classification results, node transition time, and subtree information are extracted from the non-leaf nodes of the initial decision tree to construct a node information vector containing temporal consistency constraints. The node information vector is then weighted and combined based on the temporal consistency constraints to generate the node information quantity. The information change value of the non-leaf node under the principle of minimum information is calculated, and an adaptive threshold curve is constructed based on the depth distribution and sample distribution of the initial decision tree. The information change value is projected onto the adaptive threshold curve for comparison. When the projection result is lower than the adaptive threshold curve, pruning is performed to obtain the pruned decision tree. Extract the number of leaf nodes, tree depth, and path complexity of the pruning decision tree, and combine them to construct structural constraints. Based on the structural constraints, perform dynamic programming search on the pruning decision tree to generate multiple candidate decision trees. Feature statistics are performed on the nodes of the candidate decision trees, and the upper bound of information gain is calculated in combination with the disease evolution pattern. The candidate decision trees are then screened according to the upper bound of information gain, and the decision tree with the smallest information gain is selected as the optimal decision tree. Boundary optimization is performed on the branches of the optimal decision tree to generate a treatment decision tree containing the disease evolution path.

6. The method according to claim 1, characterized in that, Based on the prediction of patient disease progression using a treatment decision tree, and considering the treatment effectiveness coefficient and the specialty strengths of candidate medical institutions, the optimal referral timing is identified, and target medical institutions are determined, including: The pre-acquired set of vital signs indicators is decomposed temporally and features are extracted. The results are then input into the treatment decision tree to calculate the state transition probability and transition time of each node. A multidimensional state space is constructed based on the state transition probability and transition time. In the multidimensional state space, Monte Carlo sampling is performed on the state transition sequence in combination with the treatment effectiveness coefficient to generate a weighted set of disease progression trajectories; Historical treatment cases of candidate medical institutions are extracted from the collaborative database of medical institutions. These historical treatment cases are clustered according to disease type and treatment stage. The treatment success rate and disease duration of each category are calculated to generate a specialty matrix. The disease progression trajectory set is mapped to a specialty matrix, the trajectory-specialty matching degree is calculated, and the candidate medical institutions are dynamically sorted based on the matching degree. At the same time, a referral trigger threshold is set according to the treatment effectiveness coefficient. When the treatment effectiveness coefficient is lower than the referral trigger threshold and there are candidate medical institutions with a matching degree higher than the preset matching threshold, the current moment is determined as the best time for referral, and the medical institution with the highest matching degree is selected as the target medical institution.

7. A cross-institutional referral intelligent routing system based on medical intelligent agent middleware, used to implement the method of any one of claims 1-6, characterized in that, include: The first unit is used to receive referral application information through the medical intelligent agent middleware. The referral application information includes patient medical record information and referral department information, wherein the patient medical record information includes vital sign data and medication records. The second unit is used to establish a monitoring window for vital signs data, to time-series map the rate of change of vital signs indicators with the time points and doses in the medication records, to identify the response delay time and peak level of vital signs indicators to different drug doses, to construct a drug response sensitivity curve, and to calculate the treatment effectiveness coefficient based on the drug response sensitivity curve. The third unit is used to obtain the diagnosis and treatment data of candidate medical institutions from the collaborative database of medical institutions, segment the drug response sensitivity curve according to the fluctuation cycle, extract the features of each segment of the curve to obtain the time-series feature vector, construct the drug-sign association map, use the dynamic time warping algorithm to calculate the similarity between the drug-sign association map and the historical association map in the diagnosis and treatment data, determine the set of sign indicators based on the similarity matching results, and generate a treatment decision tree containing the disease evolution path; The fourth unit is used to predict the patient's disease progression trajectory based on the treatment decision tree, identify the best referral time based on the treatment effectiveness coefficient and the specialty of candidate medical institutions, determine the target medical institution, and complete the data exchange by establishing a data transmission channel through the medical intelligent agent middleware.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images