A system for determining venous thromboembolism prevention plans based on diagnosis and treatment information

Abstract

The present invention discloses a system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information, and relates to the field of data processing technology. The system for determining a venous thromboembolism prevention plan includes: a feature matching module, which is used to establish a symptom matching model based on the symptom characteristics of venous thromboembolism, and collect the real-time diagnosis and treatment characteristics of the target object and input them into the symptom matching model for feature matching; a risk assessment and grading module, which is used to predict the risk probability of venous thromboembolism in the target object according to the feature matching results; and a personalized plan formulation module, which is used to formulate personalized venous thromboembolism prevention plans for target objects of different risk levels in combination with the preference characteristics of the target object. The present invention can accurately capture multiple factors such as the target object's treatment preference, drug selection preference, and risk tolerance preference through the preference analysis module, and then further optimize the prevention plan based on these personalized data.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information. Background Art

[0002] Venous thromboembolism (VTE) is a serious disease that includes deep vein thrombosis and pulmonary embolism. VTE is common in special populations such as hospitalized patients, long-term bedridden people, and patients recovering from surgery, but it may also occur in other high-risk environments. Due to its potential harm, the prevention of VTE is crucial in modern medical management.

[0003] By implementing effective preventive measures, the occurrence of VTE complications can be significantly reduced, the length of hospital stay can be shortened, the overall quality of life of patients can be improved, and the waste of medical resources can be reduced, which not only helps reduce patients' health risks but also has significant economic benefits.

[0004] However, if the individual preferences of the target subjects are ignored when formulating VTE prevention programs, it may have a negative impact on treatment compliance. Different target subjects have their own preferences for treatment methods, such as the method of drug administration (injection or oral) or the degree of acceptance of a specific treatment frequency. If the prevention program fails to fully consider these preferences, the target subjects may find it difficult to receive treatment on time and in the right amount, ultimately affecting the prevention effect.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In view of the problems in the related art, the present invention proposes a system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information to overcome the above-mentioned technical problems existing in the existing related art.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] A system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information, the system comprising:

[0009] A symptom feature recognition module, used to obtain historical diagnosis and treatment information from a predefined database, and extract symptom features of venous thromboembolism from the historical diagnosis and treatment information;

[0010] A feature matching module is used to establish a symptom matching model based on the symptom characteristics of venous thromboembolism, and collect the real-time diagnosis and treatment characteristics of the target object and input them into the symptom matching model for feature matching;

[0011] The risk assessment and grading module is used to predict the risk probability of venous thromboembolism in the target object based on the feature matching results, and to classify the target object into risk levels;

[0012] A personalized program formulation module is used to formulate personalized venous thromboembolism prevention programs for target subjects of different risk levels based on their preference characteristics;

[0013] The feature matching module includes a feature preprocessing module, a feature clustering module and an association matching module;

[0014] A feature preprocessing module is used to fuse the symptom features of venous thromboembolism to form a feature data set, and use a wavelet algorithm to remove noise in the feature data set;

[0015] A feature clustering module is used to classify similar symptom features in the preprocessed feature data set into the same cluster to obtain several symptom feature clusters;

[0016] The association matching module is used to input the real-time diagnosis and treatment characteristics of the target object into the symptom matching model to match the symptom characteristics and output the feature matching results.

[0017] Preferably, when the association matching module inputs the real-time diagnosis and treatment characteristics of the target object into the symptom matching model to match the symptom characteristics and outputs the characteristic matching results, it includes:

[0018] A symptom matching model is established based on the feature data set, and real-time diagnosis and treatment features are extracted from the pre-collected real-time diagnosis and treatment information of the target object and used as the input of the symptom matching model;

[0019] The symptom matching model is integrated with the tree matching algorithm to analyze the mapping relationship between real-time diagnosis and treatment features and symptom features, and the associated node information between the two sets of features is obtained based on the mapping relationship;

[0020] The log-likelihood function of the symptom matching model was calculated based on the maximum likelihood estimation technique, and the estimated prior parameters were obtained by combining the Lagrangian operator.

[0021] The correlation between the real-time diagnosis and treatment characteristics and the symptom characteristics is calculated by combining the associated node information and the estimated prior parameters, and the correlation is ranked;

[0022] The real-time diagnosis and treatment features with the highest correlation with the symptom features are selected and divided into corresponding symptom feature clusters to obtain feature matching results.

[0023] Preferably, the risk assessment and grading module includes:

[0024] A primary risk prediction model is constructed based on the feature matching results, and the primary risk prediction model is used to predict the risk probability of venous thromboembolism in the target subject;

[0025] The risk level of the target object is divided into low risk, medium risk and high risk based on the predicted risk probability;

[0026] The personal vital sign data of the low-risk target object and the medium-risk target object are respectively obtained and fused with the feature matching results to obtain the fused data of the target object;

[0027] A secondary risk prediction model is constructed based on the fusion data of the target object, and the probability of risk evolution of low-risk target objects and medium-risk target objects into high-risk target objects is predicted;

[0028] The risk levels are revised based on the probability of risk evolution, and the revised risk levels include low risk, medium risk, high risk and critical risk.

[0029] Preferably, constructing a secondary risk prediction model based on the fusion data of the target object and predicting the risk evolution probability of a low-risk target object and a medium-risk target object evolving into a high-risk target object includes:

[0030] Extract the empirical mode function from the fused data and establish an autoregressive model for the empirical mode function;

[0031] The empirical mode function is tested for stationarity and white noise respectively, and the non-white noise sequence characteristics are obtained based on the test results;

[0032] Determine the autoregressive model based on the characteristics of the non-white noise sequence and perform a residual sequence test, and modify the autoregressive model that fails the residual sequence test;

[0033] The autoregressive model and the least squares regression model are integrated to obtain a secondary risk prediction model, and the probability of low-risk and medium-risk target objects evolving into high-risk target objects is predicted.

[0034] Preferably, the personalized solution formulation module includes a preference analysis module, a preference weight allocation module, a personalized solution configuration module and a personalized solution optimization module;

[0035] The preference analysis module is used to collect the personal preference data of the target object and analyze the preference factors of the target object based on the personal preference data using the collaborative filtering algorithm;

[0036] The weight allocation module is used to assign weights to the preference factors of the target object;

[0037] The personalized program configuration module is used to preliminarily formulate a VTE prevention program for target subjects of different risk levels based on the weighted preference factors;

[0038] The personalized program optimization module is used to use the venous thromboembolism prevention program as a tree branch and perform iterative search to obtain the optimal branch, and at the same time form the optimal venous thromboembolism prevention program for the target object.

[0039] Preferably, the preference factors include at least treatment method preference, drug selection preference and risk tolerance preference.

[0040] Preferably, the weight allocation module includes the following steps when performing weighting processing for the preference factors of the target object:

[0041] The principal hierarchy analysis method was used to compare treatment preference, drug selection preference and risk tolerance preference in pairs, and several weight vectors of each preference factor were obtained;

[0042] Calculate the density distribution of the weight vector to determine the confidence interval of each preference factor, and assign weights to each preference factor based on the confidence interval;

[0043] Conduct consistency test on the weighted preference factors to determine whether the comparison matrix of each preference factor has logical consistency;

[0044] Based on the preference factors that fail the consistency check, the minimum weight vector is selected for adjustment.

[0045] Preferably, the personalized program optimization module takes the venous thromboembolism prevention program as a tree branch and performs iterative search to obtain the optimal branch, and simultaneously forms the optimal venous thromboembolism prevention program for the target object, including:

[0046] The root node in the tree structure represents the risk level of the target object, and the root node is gradually expanded according to the preference factors of the target object to form a tree branch corresponding to the venous thromboembolism prevention program;

[0047] Use the upper confidence bound formula to randomly select a reward node on the tree branch and determine whether the reward node is a terminal node. If the reward node is not a terminal node, expand the current reward node to form a new node.

[0048] Based on the new node, it is passed along the tree branch from the return node back to the root node, and the number of collisions and visits of each node passed through is obtained during the transmission process;

[0049] The nodes with the most visits are selected in turn and the corresponding tree branches are taken as the optimal branches. Meanwhile, the venous thromboembolism prevention plan corresponding to the optimal branch is taken as the optimal venous thromboembolism prevention plan.

[0050] Preferably, the expression of the autoregressive model order is:

[0051] ;

[0052] In the formula, The order of the autoregressive model is p Akaike Information Criterion value when ; Q Indicates the length of the sequence; p represents the order of the autoregressive model; Indicates defective a The second-order origin moment of ln Represents self-identified logarithm.

[0053] Preferably, the expression of the upper confidence bound is:

[0054] ;

[0055] In the formula, UCB represents the upper confidence bound; W i Indicates i The number of collisions of nodes; N i Indicates i The number of visits to a node; N parent Indicates the number of visits to the parent node; c represents a constant; log represents a logarithmic function.

[0056] The beneficial effects of the present invention are:

[0057] 1. The present invention obtains historical diagnosis and treatment information from a predefined database through a symptom feature recognition module, which can accurately extract symptom features related to venous thromboembolism, and help improve the accuracy of diagnosis and prevention; through a feature matching module, the real-time diagnosis and treatment features of the target object are collected and input into the model to achieve individualized symptom feature matching; through a risk assessment and grading module, the VTE risk of the target object is predicted according to the feature matching results and the risk level is divided, which helps to identify high-risk patients and take more active preventive measures; through a personalized program formulation module, prevention plans are tailored for different target objects in combination with the personal preferences of the target object, thereby improving the effectiveness and satisfaction of the plan.

[0058] 2. The present invention utilizes a multi-level risk assessment model and an integrated algorithm to accurately assess and grade the risks of the target object, ensuring that the risks are identified in a timely manner and effectively managed. The feature clustering and matching modules can quickly locate the correlation between the diagnostic and treatment characteristics and symptom characteristics of the target object, thereby improving the efficiency of diagnosis and plan recommendations. At the same time, the preference analysis module can accurately capture multiple factors such as the target object's treatment preference, drug selection preference, and risk tolerance preference, thereby further optimizing the prevention plan based on these personalized data, thereby improving the personalization of the plan and the target matching degree.

[0059] 3. The present invention can accurately predict the risk level of VTE in the early stage through personalized risk prediction, providing a basis for early intervention, and through the secondary risk prediction model, it can not only predict the current risk but also predict the evolution trend of the risk, thereby not only predicting the probability of venous thromboembolism, but also warning of future risk evolution, providing doctors with sufficient time and information for intervention, and significantly improving the effectiveness and safety of VTE prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0061] Figure 1 The present invention is a block diagram of a system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information according to an embodiment of the present invention.

[0062] In the figure:

[0063] 1. Symptom feature identification module; 2. Feature matching module; 3. Risk assessment and grading module; 4. Personalized plan formulation module. DETAILED DESCRIPTION

[0064] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention and are mainly used to illustrate the embodiments. They can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention.

[0065] According to an embodiment of the present invention, a system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information is provided.

[0066] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1As shown, according to a system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information according to an embodiment of the present invention, the system for determining a venous thromboembolism prevention plan includes:

[0067] The symptom feature recognition module 1 is used to obtain historical diagnosis and treatment information from a predefined database, and extract the symptom features of venous thromboembolism from the historical diagnosis and treatment information.

[0068] It should be noted that the historical diagnosis and treatment information is obtained from the predefined database, and the symptom characteristics of venous thromboembolism are extracted from the historical diagnosis and treatment information, including:

[0069] Historical medical data were extracted from a structured and predefined database containing clinical information of a large number of patients diagnosed with VTE (venous thromboembolism), such as symptoms, treatment methods, disease progression, etc.

[0070] Diagnostic and treatment information related to venous thromboembolism was extracted from historical records, including medical records, examination results, patient symptom descriptions, laboratory test data, imaging results, etc., and the risk factors for VTE included:

[0071] Long-term bed rest or inactivity: Sitting for a long time, lying in bed for a long time after surgery, etc. can cause blood flow to slow down and increase the risk of thrombosis.

[0072] Surgery: Particularly orthopedic surgery (such as hip or knee replacement) significantly increases the risk of blood clots.

[0073] Trauma or serious illness: Severe trauma, infection, or tumors can cause increased blood coagulability.

[0074] Age: Older people are at higher risk for VTE.

[0075] Obesity: Obese people have slower blood flow and are more prone to blood clots.

[0076] Family history: People with a family history of thrombosis are at a higher risk.

[0077] The extracted historical data were analyzed to identify typical symptom characteristics of VTE and convert them into model parameters that can be used for subsequent matching and prediction.

[0078] The feature matching module 2 is used to establish a symptom matching model based on the symptom characteristics of venous thromboembolism, and collect the real-time diagnosis and treatment characteristics of the target object and input them into the symptom matching model for feature matching.

[0079] The feature matching module 2 includes a feature preprocessing module, a feature clustering module and an association matching module;

[0080] The feature preprocessing module is used to fuse the symptom features of venous thromboembolism to form a feature data set, and use a wavelet algorithm to eliminate noise in the feature data set.

[0081] It should be noted that the symptom features of venous thromboembolism are integrated to form a feature data set, and the wavelet algorithm is used to remove the noise in the feature data set, including:

[0082] Symptom features are constructed into feature data sets, and features with diagnostic value are selected for encoding processing, such as converting categorical variables into numerical variables to ensure that the data can be input into the algorithm model.

[0083] If the data dimensions of different symptoms or characteristics are different, the data needs to be standardized or normalized to ensure that all data features are processed on a uniform scale so that the subsequent noise removal process is more stable.

[0084] Select wavelet basis functions, such as Haar wavelet, Daubechies wavelet, Coiflet wavelet, etc. Different wavelet functions are suitable for different types of data and noise characteristics.

[0085] Perform wavelet decomposition on the feature data set to decompose the original feature signal into multiple detail components (high-frequency parts, usually containing noise) and approximation components (low-frequency parts, containing main signal information) of different scales.

[0086] The high-frequency components are processed using soft threshold or hard threshold methods, the part exceeding the threshold is retained and the part below the threshold is considered as noise and eliminated; the soft threshold method will smooth the signal, while the hard threshold method will directly remove the noise.

[0087] The processed signal is reconstructed by wavelet to reconstruct a feature data set after noise removal. The reconstructed data set retains the main features of the original signal while removing the noise.

[0088] The feature clustering module is used to classify similar symptom features in the preprocessed feature data set into the same cluster to obtain several symptom feature clusters.

[0089] It should be noted that in order to ensure that all feature data work on the same scale, the symptom features should be standardized or normalized. For example, z-score normalization is used to convert the data into a standardized form with a mean of 0 and a standard deviation of 1, or minimum-maximum normalization is used to map the data to the range of [0, 1].

[0090] The similar symptom features in the preprocessed feature data set are divided into the same cluster, and several symptom feature clusters are obtained, including:

[0091] Initialize clusters based on the selected clustering algorithm (such as the initial cluster centers in K-Means and the initial tree structure in hierarchical clustering), and assign the preprocessed symptom features to each cluster based on similarity or distance metrics.

[0092] The center or boundary of the cluster is updated through algorithm iteration until the cluster converges (for example, the center of the cluster no longer changes in the K-Means algorithm, or the hierarchical clustering algorithm reaches a specific level), and the silhouette coefficient is used to evaluate the clustering effect. The closer the value is to 1, the better the separation of the clusters.

[0093] Measures the separation and closeness between clusters. The intra-class distance should be as small as possible, while the inter-class distance should be as large as possible. The hyperparameters in the algorithm (such as the number of clusters in K-Means) are adjusted based on the clustering effect evaluation.

[0094] Assign labels to each symptom feature cluster and analyze the features of each cluster. For example, one cluster may mainly include symptoms related to lower extremity venous thrombosis, while another cluster may contain high-risk features for pulmonary embolism.

[0095] The association matching module is used to input the real-time diagnosis and treatment characteristics of the target object into the symptom matching model to match the symptom characteristics and output the feature matching results.

[0096] Among them, when the association matching module inputs the real-time diagnosis and treatment characteristics of the target object into the symptom matching model to match the symptom characteristics and outputs the characteristic matching results, it includes:

[0097] A symptom matching model is established based on the feature data set, and real-time diagnosis and treatment features are extracted from the pre-collected real-time diagnosis and treatment information of the target object and used as the input of the symptom matching model;

[0098] The symptom matching model is integrated with the tree matching algorithm to analyze the mapping relationship between real-time diagnosis and treatment features and symptom features, and the associated node information between the two sets of features is obtained based on the mapping relationship.

[0099] It should be noted that the symptom matching model is integrated with the tree matching algorithm to analyze the mapping relationship between the real-time diagnosis and treatment characteristics and the symptom characteristics, and the associated node information between the two sets of characteristics is obtained based on the mapping relationship, including:

[0100] A symptom matching model is established to map the real-time features of the target object with the features in the historical VTE feature library. The matching degree between the features is quantified by feature similarity measurement methods (such as cosine similarity or Euclidean distance).

[0101] A tree structure is used to represent the hierarchical structure of symptom characteristics. Each node represents a symptom characteristic or diagnostic and treatment characteristic. The root node represents the overall diagnostic goal or risk assessment; the leaf nodes represent specific symptoms or diagnostic and treatment characteristics (such as leg swelling, dyspnea, blood D-dimer level, etc.); the intermediate nodes represent the combination or summary of different characteristics (such as the overall risk assessment of VTE or the risk of deep vein thrombosis).

[0102] According to the real-time diagnosis and treatment characteristics of the target object, the nodes in the tree are traversed, and the matching tree nodes are found by calculating the similarity between the real-time characteristics and the symptom characteristics.

[0103] After the initial matching, if the real-time feature has a high similarity with a specific symptom feature, the corresponding child node is further expanded to analyze other related symptom features or risk factors.

[0104] The heuristic search in the tree matching algorithm is used to find the path with the highest matching degree from the root node to the leaf node, which represents the best symptom matching solution.

[0105] Among them, the mapping relationship between real-time diagnosis and treatment characteristics and symptom characteristics includes:

[0106] The real-time features of the target object are input into the matching model to calculate their similarity with the historical VTE symptom features.

[0107] The symptom node with the highest similarity to the real-time feature is located in the tree structure, and the symptom feature node corresponding to each real-time diagnosis and treatment feature is obtained through the mapping model.

[0108] The mapping results are analyzed, and the mapping results include:

[0109] Matching degree between diagnosis and treatment characteristics and symptom characteristics: Measures the matching degree between each real-time characteristic and symptom characteristic.

[0110] Node correlation information: The correlation nodes between two sets of features are obtained through the tree structure to determine the direct or indirect connection between diagnosis and treatment features and symptom features.

[0111] Based on the mapping relationship between real-time features and symptom features, their associated nodes are extracted in the tree structure. The associated nodes can be reflected as:

[0112] Multiple real-time diagnosis and treatment features of the target object may correspond to the same symptom feature node; some nodes in the tree structure act as a bridge in feature mapping, connecting multiple related features and representing key diagnostic basis; through the information transmission of associated nodes, the matching results of real-time diagnosis and treatment features and symptom features are traced back and optimized to ensure that the selected path (plan) tends to be optimal after multiple iterations.

[0113] The log-likelihood function of the symptom matching model was calculated based on the maximum likelihood estimation technique, and the estimated prior parameters were obtained by combining the Lagrangian operator.

[0114] The correlation between the real-time diagnosis and treatment characteristics and the symptom characteristics is calculated by combining the associated node information and the estimated prior parameters, and the correlation is ranked;

[0115] The real-time diagnosis and treatment features with the highest correlation with the symptom features are selected and divided into corresponding symptom feature clusters to obtain feature matching results.

[0116] The risk assessment and grading module 3 is used to predict the risk probability of venous thromboembolism in the target object based on the feature matching results, and to classify the target object into risk levels.

[0117] The risk assessment and grading module 3 predicts the risk probability of venous thromboembolism in the target object according to the feature matching results, and divides the risk level of the target object into the following steps:

[0118] A primary risk prediction model is constructed based on the feature matching results, and the primary risk prediction model is used to predict the risk probability of venous thromboembolism in the target subject.

[0119] It should be noted that the personal vital sign data of the target object has been matched with the VTE symptom characteristics through the feature matching module to generate the feature matching results of the target object.

[0120] On this basis, a primary risk prediction model is constructed based on the feature matching results of the target object. This model is usually based on a statistical learning algorithm or a machine learning model, such as a logistic regression model or a decision tree model, which can associate the feature input of the target object with the known VTE risk data.

[0121] The real-time diagnostic and treatment characteristics of the target subject (such as physical signs, laboratory data, medical history, etc.) are used as the input of the primary risk prediction model, and the model calculates the risk probability of VTE in the target subject based on these inputs.

[0122] The risk level of the target object is divided based on the predicted risk probability, and the risk levels after division include low risk, medium risk and high risk.

[0123] It should be noted that the risk level classification is based on the risk probability of the target object. First, different probability thresholds need to be set. These thresholds determine the low, medium and high risk intervals. The threshold selection can use the empirical data in medical practice to set reasonable risk level thresholds, or through the risk incidence rate in historical data, combined with the data distribution to reasonably set the probability range of each risk level.

[0124] For example: low risk: P(VTE) < 0.1; medium risk: 0.1 ≤ P(VTE) < 0.3; high risk: P(VTE) ≥ 0.3; P(VTE) represents the predicted risk probability of the target object developing venous thromboembolism; for each target object, the risk level is divided into the following according to the risk probability generated by the prediction model:

[0125] When the risk probability P of the target subject is less than 0.1, it is considered to be in a low-risk state. Such subjects are less likely to develop VTE in the short term (such as in the next week or month) and usually do not require special medical intervention, and routine monitoring is sufficient.

[0126] When the risk probability of the target subject is between 0.1≤P<0.3, it is considered to be in a medium-risk state. Such subjects have a certain risk of VTE. It is recommended to closely observe their clinical symptoms. Regular follow-up may be required, and preventive treatment (such as physical therapy or anticoagulants) may be carried out according to the situation.

[0127] When the risk probability of the target subject is P ≥ 0.3, it is considered that the subject is in a high-risk state. The probability of VTE occurrence in such subjects is high, and timely medical intervention measures are usually required, including anticoagulant drug treatment, bed rest monitoring, reduction of surgery and other triggering factors, and close attention should be paid to changes in the subject's symptoms.

[0128] It should be noted that the health status and physical signs of the target subjects may change over time, so it is necessary to reassess their risk probability regularly. Specifically, the risk probability can be dynamically adjusted and the risk level can be reclassified based on new diagnosis and treatment data and laboratory test results.

[0129] For example, in one specific embodiment:

[0130] Patient A: A 50-year-old male who recently underwent knee replacement surgery and has a history of long-term bed rest. The model predicts that Patient A has a VTE risk probability of 0.12.

[0131] Patient B: A 60-year-old female with a recent surgery with a significantly elevated D-dimer level and a family history of thrombosis. The prediction model gives a risk probability of 0.35.

[0132] Patient A: Risk probability P=0.12, which is medium risk. Close monitoring of the condition is recommended and preventive anticoagulant therapy may be required.

[0133] Patient B: Risk probability P=0.35, which is high risk. Immediate anticoagulant treatment and intensive monitoring are required to prevent the occurrence of VTE.

[0134] The personal vital sign data of the low-risk target object and the medium-risk target object are respectively obtained and fused with the feature matching results to obtain the fused data of the target object;

[0135] A secondary risk prediction model is constructed based on the fused data of the target object, and the risk evolution probability of low-risk target objects and medium-risk target objects evolving into high-risk target objects is predicted.

[0136] Preferably, constructing a secondary risk prediction model based on the fusion data of the target object and predicting the risk evolution probability of a low-risk target object and a medium-risk target object evolving into a high-risk target object includes:

[0137] The empirical mode function in the fused data is extracted, and an autoregressive model is established for the empirical mode function.

[0138] It should be noted that extracting the empirical mode function from the fused data and establishing an autoregressive model for the empirical mode function includes:

[0139] The fused data of the target object is subjected to empirical mode decomposition, and the complex nonlinear and non-stationary data is decomposed into several empirical mode functions (IMFs), each of which reflects the fluctuations and trends of the data at different time scales. These IMFs become the key inputs of the secondary risk prediction model.

[0140] An autoregressive model is established for each extracted IMF, assuming that the current value of each IMF can be predicted by a linear combination of its past values, where the expression of the autoregressive model is:

[0141] ;

[0142] In the formula, X t Represents the time vector t The empirical mode function of represents the autoregressive coefficient; p represents the autoregressive order; represents the error term.

[0143] The empirical mode function is tested for stationarity and white noise respectively, and the non-white noise sequence characteristics are obtained based on the test results.

[0144] It should be noted that the stationarity test uses a unit root test (such as an ADF test) to check whether each IMF series is a stationary series. A stationary series is the basis of the autoregressive model. If the series is not stationary, difference processing may be required to achieve stationarity.

[0145] The white noise test uses the autocorrelation function (ACF) test to detect whether the IMF sequence is a white noise sequence. If the IMF is a white noise sequence, the sequence does not contain useful information and should be removed from the model.

[0146] The autoregressive model is determined based on the characteristics of the non-white noise sequence and a residual sequence test is performed. The autoregressive model that fails the residual sequence test is corrected.

[0147] Among them, the expression of the autoregressive model order is:

[0148] ;

[0149] In the formula, The order of the autoregressive model is p Akaike Information Criterion value when ; Q Indicates the length of the sequence; p represents the order of the autoregressive model; Indicates defective a The second-order origin moment of ln Represents self-identified logarithm.

[0150] It should be noted that the Akaike information criterion is used to perform a residual test on the fitted autoregressive model to check whether the residual is white noise; if the residual is non-white noise, the model may be underfit or misfit and needs to be corrected; if the residual test fails, the autoregressive model is corrected (such as increasing the order or transforming the input data) until the residual meets the white noise assumption.

[0151] Akaike Information Criterion (AIC) is a criterion for model selection. Its core idea is to strike a balance between goodness of fit and model complexity. AIC not only considers the fitting effect of the model, but also the complexity of the model to prevent overfitting.

[0152] The autoregressive model and the least squares regression model are integrated to obtain a secondary risk prediction model, and the probability of low-risk and medium-risk target objects evolving into high-risk target objects is predicted.

[0153] The risk levels are revised based on the probability of risk evolution, and the revised risk levels include low risk, medium risk, high risk and critical risk.

[0154] It should be noted that the integrated secondary risk prediction model is used to input the fusion data of low-risk and medium-risk objects to predict the probability of their evolution from the current state to the high-risk state. This probability reflects the risk evolution trend of the object.

[0155] The risk evolution probability output by the model is usually a value between 0 and 1, indicating the possibility that the target object will evolve from the current risk state (low / medium) to a high risk state.

[0156] The risk level of the target object is dynamically adjusted according to the predicted risk evolution probability. According to the evolution probability, the target object can be divided into:

[0157] Low risk: The probability of risk evolution is lower than a certain threshold (such as 0.1).

[0158] Medium risk: The probability of risk evolution is in the middle (such as between 0.1 and 0.3).

[0159] High risk: The probability of risk evolution exceeds a high threshold (e.g., above 0.3).

[0160] Critical risk: The probability of risk evolution approaches the limit value (such as exceeding 0.6 or higher), indicating that the target object may quickly enter a high-risk state.

[0161] The personalized program formulation module 4 is used to formulate personalized venous thromboembolism prevention programs for target subjects of different risk levels in combination with the preference characteristics of the target subjects.

[0162] Among them, the personalized plan formulation module includes a preference analysis module, a preference weight allocation module, a personalized plan configuration module and a personalized plan optimization module;

[0163] The preference analysis module is used to collect the personal preference data of the target object and analyze the preference factors of the target object based on the personal preference data using a collaborative filtering algorithm.

[0164] Among them, preference factors include at least treatment method preference, drug selection preference and risk tolerance preference;

[0165] Treatment preference: Whether the target subject has clearly expressed a preference for certain treatment methods (such as surgery, medication, physical therapy, etc.).

[0166] Drug selection preference: The target subject's rating or selection of specific drugs or drug types (such as traditional Chinese medicine, chemical drugs, over-the-counter drugs, brand drugs, etc.).

[0167] Risk tolerance preference: The target subject's acceptance of different treatment risks. For example, when choosing a treatment plan, the target subject prefers a safe but slow-acting method, or a high-risk but fast-acting method.

[0168] It should be noted that the personal preference data of the target object is collected, and the preference factors of the target object are analyzed based on the personal preference data using the collaborative filtering algorithm, including:

[0169] The weight assignment module is used to assign weights to the preference factors of the target object.

[0170] Among them, the weight allocation module includes:

[0171] The principal hierarchy process was used to compare treatment preference, drug selection preference and risk tolerance preference in pairs, and several weight vectors of each preference factor were obtained.

[0172] It should be noted that the analytic hierarchy process (AHP) is a decision-making method that decomposes complex decision-making problems into simple comparison problems in a hierarchical manner, and finally obtains the weights of each factor based on these comparison results. For the three major preference factors of treatment preference, drug selection preference, and risk tolerance preference, AHP can determine the relative weight of each preference factor through pairwise comparison. Using AHP to compare these factors pairwise and obtain the weight vector includes:

[0173] First, we need to construct a pairwise comparison matrix to represent the relative importance of each preference factor. The basic idea of ​​​​the principal hierarchy analysis method is to determine the relative importance of each preference factor through the judgment of experts or users.

[0174] In AHP, a scale of 1-9 (Saaty scale) is often used to express the strength of pairwise comparisons:

[0175] 1: Both factors are equally important.

[0176] 3: One of the factors is slightly more important.

[0177] 5: One of these factors is clearly more important.

[0178] 7: One of these factors is very important.

[0179] 9: One of these factors is extremely important.

[0180] The other numbers (2, 4, 6, 8) represent intermediate values ​​and are used for further subdivision. If the preference for treatment method is judged to be slightly more important than the preference for drug choice, a rating of 3 can be given.

[0181] In the comparison matrix, each element represents the relative importance of one preference factor with respect to another preference factor. For three preference factors, the comparison matrix will have a dimension of 3×3 and the following structure:

[0182] The rows and columns represent treatment preference, drug selection preference, and risk tolerance preference, respectively; each element in the matrix corresponds to the relative importance of two preference factors, for example:

[0183] Suppose the target subject believes that the weight of treatment preference relative to drug selection preference is 3, indicating that treatment preference is slightly important; at the same time, the target subject believes that drug selection preference and risk tolerance preference are equally important, so the value is assigned to 1; if the target subject believes that treatment preference is far more important than risk tolerance preference, the value may be assigned to 7.

[0184] After filling the comparison matrix, the next step is to calculate the weight vector of each preference factor from the matrix. The weight vector represents the relative importance of each preference factor in the entire system. This step can be obtained by calculating the normalized value of each row element in the matrix, that is, normalizing each element in the matrix relative to other elements, and finally obtaining the weight of each factor. The weight vector results include:

[0185] The weight of treatment method preference: indicates the user's attention to the treatment method in the entire preference system.

[0186] Weight of drug selection preference: indicates the importance of drug selection in user preference.

[0187] Weight of risk tolerance preference: Indicates the importance the user places on risk tolerance.

[0188] Calculate the density distribution of the weight vector to determine the confidence interval of each preference factor, and assign weights to each preference factor based on the confidence interval;

[0189] Conduct consistency test on the weighted preference factors to determine whether the comparison matrix of each preference factor has logical consistency;

[0190] Based on the preference factors that fail the consistency check, the minimum weight vector is selected for adjustment.

[0191] The personalized program configuration module is used to preliminarily formulate a VTE prevention program for target subjects of different risk levels based on the weighted preference factors, for example:

[0192] 1. Prevention plan for low-risk targets:

[0193] Considerations for empowerment factors:

[0194] Treatment preference: The target group prefers non-invasive or conservative treatment.

[0195] Drug selection preference: Prefer over-the-counter drugs or drugs with lower prices.

[0196] Risk tolerance: The desire to avoid drug side effects or potential risks during treatment.

[0197] Prevention options:

[0198] Non-drug intervention: It is recommended to wear medical elastic stockings and maintain appropriate activities (such as getting up and moving around regularly every day, leg stretching exercises) to improve venous blood flow and reduce the risk of thrombosis.

[0199] Drug prevention: Consider using low-dose, low-risk medications to ensure that the medication has minimal side effects on the intended recipient.

[0200] Lifestyle recommendations: Maintain good living habits, such as drinking plenty of water, avoiding sitting for long periods of time, and doing moderate exercise to prevent blood clots.

[0201] 2. Prevention plan for medium-risk targets:

[0202] Considerations for empowerment factors:

[0203] Treatment preference: May accept a certain degree of drug intervention, but still hope to avoid invasive treatment as much as possible.

[0204] Drug selection preference: There may be a preference for certain types of drugs, and it is recommended to choose an anticoagulant drug that matches your preference.

[0205] Risk tolerance: willingness to take slight treatment risks in exchange for more effective prevention effects.

[0206] Prevention options:

[0207] Drug prevention: It is recommended to use anticoagulant drugs that have been proven to be effective, and the dose should be adjusted according to the target subject's weight and clinical needs.

[0208] Non-pharmacological interventions: Continue to emphasize the use of medical elastic stockings and advise patients to regularly move their lower limbs to keep the veins open.

[0209] Monitoring and Adjustment: It is recommended that the target subjects undergo regular blood coagulation function tests to ensure appropriate drug dosage and avoid the risk of excessive anticoagulation.

[0210] 3. Prevention plan for high-risk targets:

[0211] Considerations for empowerment factors:

[0212] Treatment preference: Ability to accept more powerful drug interventions and even a certain degree of invasive treatment when necessary.

[0213] Medication selection preference: You may want to use a proven and effective medication.

[0214] Risk tolerance preference: willingness to assume a certain degree of treatment risk to ensure effective prevention.

[0215] Prevention options:

[0216] Drug intervention: Standard doses of drugs are recommended for more effective prevention, and the choice of drugs should be consistent with the target group's drug brand or type preference.

[0217] Enhanced monitoring: It is recommended to conduct regular coagulation function monitoring and pay attention to the liver and kidney function of the target subjects to ensure that the drug does not have a negative impact on the body.

[0218] 4. Prevention plan for critical risk target objects:

[0219] Considerations for empowerment factors:

[0220] Treatment preference: Can accept invasive treatment and hope to achieve preventive effect as soon as possible.

[0221] Drug selection preference: They have extremely high requirements for drug effectiveness and may prefer imported drugs or more effective anticoagulation regimens.

[0222] Risk tolerance preference: having a higher tolerance for treatment risks and being willing to choose a combination of strong or multiple interventions.

[0223] Prevention options:

[0224] Intensified drug intervention: Prioritize the use of more potent anticoagulants, such as higher doses or the combination of two anticoagulants.

[0225] Intensive monitoring and adjustment: Perform coagulation function tests every 48 hours, closely monitor the patient's vital signs, and adjust drug dosage and treatment strategies at any time.

[0226] It should be noted that the present invention only uses treatment method preference, drug selection preference and risk tolerance preference as examples for explanation. In specific implementation, a more targeted VTE prevention program can be formulated in combination with other preference characteristics of the target subject.

[0227] The personalized program optimization module is used to use the venous thromboembolism prevention program as a tree branch and perform iterative search to obtain the optimal branch, and at the same time form the optimal venous thromboembolism prevention program for the target object.

[0228] The personalized program optimization module includes the following steps when taking the venous thromboembolism prevention program as a tree branch and performing iterative search to obtain the optimal branch and forming the optimal venous thromboembolism prevention program for the target object:

[0229] The root node in the tree structure represents the risk level of the target object, and the root node is gradually expanded according to the preference factors of the target object to form a tree branch corresponding to the venous thromboembolism prevention program;

[0230] The upper confidence bound formula is used to randomly select a reward node on the tree branch and determine whether the reward node is a terminal node. If the reward node is not a terminal node, the current reward node is expanded to form a new node.

[0231] It should be noted that during the search process, reward nodes need to be selected on different branches for further expansion. The upper confidence bound formula (UCB) is used to balance the trade-off between exploration and exploitation during search. The expression of the upper confidence bound is:

[0232] ;

[0233] In the formula, UCB represents the upper confidence bound; W i Indicates i The number of collisions of nodes; N i Indicates i The number of visits to a node; N parent Indicates the number of visits to the parent node; c represents a constant; log represents a logarithmic function.

[0234] Based on the new node, it is passed along the tree branch from the return node back to the root node, and the number of collisions and visits of each node passed through is obtained during the transmission process;

[0235] The nodes with the most visits are selected in turn and the corresponding tree branches are taken as the optimal branches. Meanwhile, the venous thromboembolism prevention plan corresponding to the optimal branch is taken as the optimal venous thromboembolism prevention plan.

[0236] It should be noted that the score of each node is calculated according to the upper confidence bound formula, and a node with a higher UCB value is randomly selected as the reward node in each search; if the selected node is not a terminal node (that is, the final solution has not been reached), the node will continue to be expanded to form a new child node, and further search will be performed on the child node.

[0237] Determine whether the return node is a termination node; if the return node is a termination node, it means that the plan corresponding to the node is a complete VTE prevention plan and will not be expanded; if the return node is not a termination node, continue to expand the node to form a new child node, indicating further treatment plan selection (such as drug dosage adjustment or treatment method change).

[0238] The newly generated node will be marked and added to the tree branch, and information will be passed from the node up along the tree branch to update the number of collisions and visits for each passed node; the status of each node will be recorded during the transmission process so that the optimal path can be selected in future iterations.

[0239] Once the search is completed or the preset number of iterations is reached, the node with the most visits is selected as the best choice. Usually, a node with a large number of visits means that the path has been repeatedly verified and may be a better solution.

[0240] The path corresponding to the optimal branch is the optimal venous thromboembolism prevention plan, which not only takes into account the preference factors of the target object, but also finds the prevention path with the highest return and the best effect through iterative search and upper confidence bound optimization.

[0241] For example, the VTE prevention program optimization process includes:

[0242] Assume that the target object is a medium-risk patient, whose preference factors are:

[0243] Treatment preference: Prefer non-invasive treatments.

[0244] Drug selection preference: Low molecular weight heparin is preferred.

[0245] Risk tolerance: Moderate treatment risk is acceptable.

[0246] The root node represents medium risk, and then branches are expanded according to treatment preference, drug selection preference, etc., forming different treatment paths.

[0247] Through multiple iterations and selecting reward nodes according to the UCB formula, new branches are expanded and generated.

[0248] Finally, a path is selected based on the nodes with the most visits, which may be a regimen based on non-invasive treatment and low molecular weight heparin use.

[0249] The optimal VTE prevention plan for the target subject is the plan formed along this path, which is consistent with his preferences for treatment methods, drugs and risks, while ensuring the maximization of prevention effects.

[0250] It should be noted that the design of the tree structure in the present invention can decompose each prevention plan into multiple branch nodes, optimize and expand them in turn, and randomly select reward nodes in the tree branches and expand them, which can achieve a balance between exploration and utilization. Through iterative search of reward nodes, the path with the most visits is finally selected as the optimal plan, which greatly improves the reliability and effectiveness of the prevention plan.

[0251] In summary, with the help of the above-mentioned technical scheme of the present invention, the present invention obtains historical diagnosis and treatment information from a predefined database through a symptom feature recognition module, can accurately extract symptom features related to venous thromboembolism, and help improve the accuracy of diagnosis and prevention; through the feature matching module, the real-time diagnosis and treatment features of the target object are collected and input into the model to achieve individualized symptom feature matching; through the risk assessment and grading module, the VTE risk of the target object is predicted according to the feature matching results and the risk level is divided, which helps to identify high-risk patients and take more active preventive measures; through the personalized program formulation module, prevention plans are tailored for different target objects in combination with the personal preferences of the target object, thereby improving the effectiveness and satisfaction of the plan; the present invention uses a multi-level risk assessment model and an integrated algorithm to accurately assess and grade the risks of the target object, ensuring that the risks are obtained and The invention can identify and effectively manage VTE in real time, and through the feature clustering and matching module, it can quickly locate the correlation between the diagnosis and treatment characteristics and the symptom characteristics of the target object, thereby improving the efficiency of diagnosis and scheme recommendation. At the same time, the preference analysis module can accurately capture multiple factors such as the treatment method preference, drug selection preference and risk tolerance preference of the target object, so as to further optimize the prevention scheme according to these personalized data, thereby improving the personalization and target matching of the scheme. The present invention can accurately predict the risk level of VTE in the early stage through personalized risk prediction, providing a basis for early intervention, and through the secondary risk prediction model, it can not only predict the current risk but also predict the evolution trend of the risk, so as to not only predict the probability of venous thromboembolism, but also warn of future risk evolution, provide doctors with sufficient time and information for intervention, and significantly improve the effectiveness and safety of VTE prevention.

[0252] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information, characterized in that: The system includes: A symptom feature recognition module, used to obtain historical diagnosis and treatment information from a predefined database, and extract symptom features of venous thromboembolism from the historical diagnosis and treatment information; A feature matching module is used to establish a symptom matching model based on the symptom characteristics of venous thromboembolism, and collect the real-time diagnosis and treatment characteristics of the target object and input them into the symptom matching model for feature matching; The risk assessment and grading module is used to predict the risk probability of venous thromboembolism in the target object based on the feature matching results, and to classify the target object into risk levels; A personalized program formulation module is used to formulate personalized venous thromboembolism prevention programs for target subjects of different risk levels based on their preference characteristics; The feature matching module includes a feature preprocessing module, a feature clustering module and an association matching module; The personalized solution formulation module includes a preference analysis module, a personalized solution configuration module, a personalized solution optimization module, and a weight allocation module for weighting the preference factors of the target object; The weight allocation module includes the following steps when performing weighting processing on the preference factors of the target object: The principal hierarchy analysis method was used to compare treatment preference, drug selection preference and risk tolerance preference in pairs, and several weight vectors of each preference factor were obtained; Calculate the density distribution of the weight vector to determine the confidence interval of each preference factor, and assign weights to each preference factor based on the confidence interval; Conduct consistency test on the weighted preference factors to determine whether the comparison matrix of each preference factor has logical consistency; Based on the preference factors that fail the consistency check, the minimum weight vector is selected for adjustment.

2. A system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information according to claim 1, characterized in that: The feature matching module also includes: The feature preprocessing module is used to fuse the symptom features of venous thromboembolism to form a feature data set, and use a wavelet algorithm to remove noise in the feature data set; The feature clustering module is used to classify similar symptom features in the preprocessed feature data set into the same cluster to obtain a number of symptom feature clusters; The association matching module is used to input the real-time diagnosis and treatment characteristics of the target object into the symptom matching model to match the symptom characteristics and output the characteristic matching results.

3. A system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information according to claim 2, characterized in that: The personalized program formulation module also includes: The preference analysis module is used to collect personal preference data of the target object, and analyze the preference factors of the target object based on the personal preference data using a collaborative filtering algorithm, wherein the preference factors at least include treatment method preference, drug selection preference and risk tolerance preference; The personalized program configuration module is used to preliminarily formulate a venous thromboembolism prevention program for target subjects of different risk levels in combination with the weighted preference factors; The personalized program optimization module is used to use the venous thromboembolism prevention program as a tree branch and perform iterative search to obtain the optimal branch, thereby forming an optimal venous thromboembolism prevention program for the target object.

4. A system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information according to claim 3, characterized in that: The association matching module includes: when inputting the real-time diagnosis and treatment characteristics of the target object into the symptom matching model to match the symptom characteristics and outputting the characteristic matching results: A symptom matching model is established based on the feature data set, and real-time diagnosis and treatment features are extracted from the pre-collected real-time diagnosis and treatment information of the target object and used as the input of the symptom matching model; The symptom matching model is integrated with the tree matching algorithm to analyze the mapping relationship between real-time diagnosis and treatment features and symptom features, and the associated node information between the two sets of features is obtained based on the mapping relationship; The log-likelihood function of the symptom matching model was calculated based on the maximum likelihood estimation technique, and the estimated prior parameters were obtained by combining the Lagrangian operator. The correlation between the real-time diagnosis and treatment characteristics and the symptom characteristics is calculated by combining the associated node information and the estimated prior parameters, and the correlation is ranked; The real-time diagnosis and treatment features with the highest correlation with the symptom features are selected and divided into corresponding symptom feature clusters to obtain feature matching results.

5. A system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information according to claim 4, characterized in that: The risk assessment and grading module predicts the risk probability of venous thromboembolism in the target object according to the feature matching result and divides the risk level of the target object into the following steps: A primary risk prediction model is constructed based on the feature matching results, and the primary risk prediction model is used to predict the risk probability of venous thromboembolism in the target subject; The risk level of the target object is divided into low risk, medium risk and high risk based on the predicted risk probability; The personal vital sign data of the low-risk target object and the medium-risk target object are respectively obtained and fused with the feature matching results to obtain the fused data of the target object; A secondary risk prediction model is constructed based on the fusion data of the target object, and the probability of risk evolution of low-risk target objects and medium-risk target objects into high-risk target objects is predicted; The risk levels are revised based on the probability of risk evolution, and the revised risk levels include low risk, medium risk, high risk and critical risk.

6. A system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information according to claim 5, characterized in that: The method of constructing a secondary risk prediction model based on the fusion data of the target object and predicting the risk evolution probability of a low-risk target object and a medium-risk target object evolving into a high-risk target object includes: Extract the empirical mode function from the fused data and establish an autoregressive model for the empirical mode function; The empirical mode function is tested for stationarity and white noise respectively, and the non-white noise sequence characteristics are obtained based on the test results; Determine the autoregressive model based on the characteristics of the non-white noise sequence and perform a residual sequence test, and modify the autoregressive model that fails the residual sequence test; The autoregressive model and the least squares regression model are integrated to obtain a secondary risk prediction model, and the probability of low-risk target objects and medium-risk target objects evolving into high-risk target objects is predicted.

7. A system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information according to claim 6, characterized in that: The personalized program optimization module takes the venous thromboembolism prevention program as a tree branch and performs iterative search to obtain the optimal branch, and simultaneously forms the optimal venous thromboembolism prevention program for the target object, including: The root node in the tree structure represents the risk level of the target object, and the root node is gradually expanded according to the preference factors of the target object to form a tree branch corresponding to the venous thromboembolism prevention program; Use the upper confidence bound formula to randomly select a reward node on the tree branch and determine whether the reward node is a terminal node. If the reward node is not a terminal node, expand the current reward node to form a new node. Based on the new node, it is passed along the tree branch from the return node back to the root node, and the number of collisions and visits of each node passed through is obtained during the transmission process; The nodes with the most visits are selected in turn and the corresponding tree branches are taken as the optimal branches. Meanwhile, the venous thromboembolism prevention plan corresponding to the optimal branch is taken as the optimal venous thromboembolism prevention plan.

8. The system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information according to claim 7, characterized in that: The expression of the autoregressive model order is: ； In the formula, The order of the autoregressive model is p Akaike Information Criterion value when ; Q Indicates the length of the sequence; p represents the order of the autoregressive model; Indicates defective a The second-order origin moment of ln Represents self-identified logarithm.

9. A system for determining a venous thromboembolism prevention plan based on diagnosis and treatment information according to claim 8, characterized in that: The expression of the upper confidence bound is: ； In the formula, UCB represents the upper confidence bound; W i Indicates i The number of collisions of nodes; N i Indicates i The number of visits to a node; N parent Indicates the number of visits to the parent node; c represents a constant; log represents a logarithmic function.

Images