A method for constructing a risk assessment model for patients undergoing intravenous therapy

Abstract

The present invention relates to the technical field of intravenous therapy risk assessment, and in particular to a method for constructing a risk assessment model for intravenous therapy patients. The method comprises the following steps: constructing a multidimensional correlation matrix for treatment risk patient data to obtain an intravenous risk correlation matrix; extracting similarity-related features in the time series window before and after risk marking for normal treatment patient data to obtain pre-risk similarity-related data and risk-related data; performing feature state division based on an initial model of intravenous risk probability to obtain a recovery state data set and a severity state data set; and performing shallow multidimensional assessment model learning on the initial model of intravenous risk probability based on the recovery state data set and the severity state data set to obtain a patient risk assessment model. The present invention achieves accurate and efficient assessment of intravenous therapy risk.

Description

Technical Field

[0001] The present invention relates to the technical field of intravenous therapy risk assessment, and in particular to a method for constructing a risk assessment model for intravenous therapy patients. Background Art

[0002] With the continuous advancement of medical technology and the intensification of the aging population trend, intravenous therapy has been widely used as a common medical method in the treatment and care of various diseases, and patient risk assessment is an important issue faced when formulating treatment plans and making decisions. However, although intravenous therapy has significant advantages in improving patients' conditions and improving their quality of life, in practice, various complications and adverse events may occur during intravenous therapy, such as thromboembolism, venous channel failure, allergic reactions, etc. Traditional risk assessment methods are often based only on individual clinical characteristics and historical data, and lack full utilization of time and sequence information. The patient's status and risk level may change over time. Traditional risk assessment models often cannot handle these time series and correlation characteristics well. Moreover, in the process of intravenous treatment of patients, due to the influence of various factors during treatment and various physical factors of different patients themselves, the final constructed risk assessment model is difficult to accurately predict the various risk probabilities of patients during intravenous treatment. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention proposes a method for constructing a risk assessment model for intravenous therapy patients to solve at least one of the above technical problems.

[0004] To achieve the above objectives, the present invention provides a method for constructing a risk assessment model for intravenous therapy patients, comprising the following steps:

[0005] Step S1: Acquire historical data of intravenous treatment patients; perform risk classification on the historical data of intravenous treatment patients, thereby obtaining normal treatment patient data and treatment risk patient data; construct a multidimensional correlation matrix for the treatment risk patient data, thereby obtaining an intravenous risk correlation matrix;

[0006] Step S2: Extract similarity-related features of the time series window before and after risk marking from the data of patients undergoing normal treatment according to the venous risk association matrix, thereby obtaining similarity-related data before risk and similarity-related data during risk; construct a naive Bayesian probability model for the similarity-related data before risk and similarity-related data during risk, thereby obtaining an initial venous risk probability model;

[0007] Step S3: Based on the initial model of venous risk probability, the characteristic state division of different risk symptom recovery conditions and the characteristic state division of different risk severity are performed to obtain a recovery state data set and a severity state data set; based on the recovery state data set and the severity state data set, the initial model of venous risk probability is shallowly learned into a multidimensional evaluation model to obtain a patient risk assessment model.

[0008] The present invention obtains historical data of patients undergoing intravenous treatment and classifies them into risk categories, thereby dividing patients into two categories: normal treatment patients and treatment risk patients. This classification helps to distinguish between patients with lower and higher risk levels; by constructing a multidimensional correlation matrix for the treatment risk patient data, the correlation between the patient's medical history and various factors in the treatment process can be captured. Furthermore, by generating a venous risk correlation matrix, the interaction between different risk factors can be more comprehensively described; the method of similar correlation feature extraction and naive Bayesian probability model construction can be used to extract similar correlation features of patients at risk from the normal treatment patient data, and to construct an initial venous risk probability model to improve the recognition rate of the probability model; the method of feature state division and multidimensional evaluation model shallow learning can be used to further optimize the initial venous risk probability model to obtain a more accurate patient risk assessment model, so as to more comprehensively consider the recovery of different risk symptoms and the feature states of different risk severities, thereby more accurately assessing the patient's risk. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 Schematic diagram of the steps of the method for constructing the intravenous therapy patient risk assessment model of the present invention;

[0010] Figure 2 Detailed implementation flow chart of step S1;

[0011] Figure 3 4 is a flowchart of the detailed implementation steps of step S14. DETAILED DESCRIPTION

[0012] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] This application provides a method for constructing a risk assessment model for intravenous therapy patients. The execution entities of this method include, but are not limited to, the following: mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc., which can be considered general computing nodes of this application. The data processing platform includes, but is not limited to, at least one of an audio and image management system, an information management system, and a cloud data management system.

[0014] See also Figures 1 to 3 The present invention provides a method for constructing a risk assessment model for intravenous therapy patients, comprising the following steps:

[0015] Step S1: Acquire historical data of intravenous treatment patients; perform risk classification on the historical data of intravenous treatment patients, thereby obtaining normal treatment patient data and treatment risk patient data; construct a multidimensional correlation matrix for the treatment risk patient data, thereby obtaining an intravenous risk correlation matrix;

[0016] Step S2: Extract similarity-related features of the time series window before and after risk marking from the data of patients undergoing normal treatment according to the venous risk association matrix, thereby obtaining similarity-related data before risk and similarity-related data during risk; construct a naive Bayesian probability model for the similarity-related data before risk and similarity-related data during risk, thereby obtaining an initial venous risk probability model;

[0017] Step S3: Based on the initial model of venous risk probability, the characteristic state division of different risk symptom recovery conditions and the characteristic state division of different risk severity are performed to obtain a recovery state data set and a severity state data set; based on the recovery state data set and the severity state data set, the initial model of venous risk probability is shallowly learned into a multidimensional evaluation model to obtain a patient risk assessment model.

[0018] The present invention obtains historical data of patients undergoing intravenous treatment and classifies them into risk categories, thereby dividing patients into two categories: normal treatment patients and treatment risk patients. This classification helps to distinguish between patients with lower and higher risk levels; by constructing a multidimensional association matrix for the treatment risk patient data, the correlation between the patient's medical history and various factors in the treatment process can be captured. Furthermore, by generating a venous risk association matrix, the interaction between different risk factors can be more comprehensively described; the method of similar related feature extraction and naive Bayesian probability model construction can be used to extract similar related features of patients at risk from the normal treatment patient data, and to construct an initial venous risk probability model to improve the recognition rate of the probability model; the method of feature state division and multidimensional evaluation model shallow learning can be used to further optimize the initial venous risk probability model to obtain a more accurate patient risk assessment model, so as to more comprehensively consider the recovery of different risk symptoms and the feature states of different risk severities, thereby more accurately assessing the patient's risk.

[0019] In this embodiment, reference Figure 1 The above is a schematic flow chart of the steps of the method for constructing the intravenous therapy patient risk assessment model of the present invention. The method for constructing the intravenous therapy patient risk assessment model includes the following steps:

[0020] Step S1: Acquire historical data of intravenous treatment patients; perform risk classification on the historical data of intravenous treatment patients, thereby obtaining normal treatment patient data and treatment risk patient data; construct a multidimensional correlation matrix for the treatment risk patient data, thereby obtaining an intravenous risk correlation matrix;

[0021] In this embodiment, relevant data from patients who have undergone intravenous therapy, including various monitoring indicators and symptom records during treatment, is collected and organized. Based on the characteristics of these historical data, appropriate methods are used to categorize patients into two groups: those with normal treatment and those with treatment risk. This classification can be based on known risk indicators, the frequency of abnormalities, and other factors. Using the data from patients with treatment risk, a multidimensional association matrix is constructed, reflecting the relationships between different risk factors. Based on this multidimensional association matrix, a venous risk association matrix is generated, which records the degree of association and correlation between different risk factors.

[0022] Step S2: Extract similarity-related features of the time series window before and after risk marking from the data of patients undergoing normal treatment according to the venous risk association matrix, thereby obtaining similarity-related data before risk and similarity-related data during risk; construct a naive Bayesian probability model for the similarity-related data before risk and similarity-related data during risk, thereby obtaining an initial venous risk probability model;

[0023] In this embodiment, the data of normal treatment patients are analyzed based on the venous risk association matrix to extract risk-related features. These features can be monitoring indicators, symptom changes, etc. related to treatment risks within a time window before and after the risk mark. By extracting similar related features in the time series window before and after the risk mark, similar related data before the risk and similar related data at the time of risk are obtained. Similar related data before the risk reflects the characteristic state before the risk occurs, while similar related data at the time of risk reflects the characteristic state when the risk occurs. Using similar related data before the risk and similar related data at the time of risk, a naive Bayesian probability model is constructed. This model can make a preliminary assessment of venous risk based on the probability of occurrence of the characteristic state.

[0024] Step S3: Based on the initial model of venous risk probability, the characteristic state division of different risk symptom recovery conditions and the characteristic state division of different risk severity are performed to obtain a recovery state data set and a severity state data set; based on the recovery state data set and the severity state data set, the initial model of venous risk probability is shallowly learned into a multidimensional evaluation model to obtain a patient risk assessment model.

[0025] In this embodiment, the recovery status of different risk symptoms is divided based on the initial model of venous risk probability. This can be defined according to the patient's characteristic state and related indicators, such as the degree of symptom relief, recovery speed, etc., and the severity of different risks is also divided based on the initial model of venous risk probability. This can be defined according to the patient's characteristic state and related indicators, such as the severity of symptoms, the abnormality of monitoring indicators, etc., and according to the characteristic state division of the recovery status and severity of different risk symptoms, a recovery state data set and a severity state data set are obtained. These data sets record the characteristic state information of patients under different risks. The recovery state data set and the severity state data set are used to perform shallow learning of the multidimensional evaluation model of the initial model of venous risk probability. This can use a machine learning algorithm to further optimize the risk assessment model according to the patient's characteristic state and risk indicators to improve accuracy and predictive ability.

[0026] In this embodiment, reference Figure 2 The above is a flowchart of the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include:

[0027] Step S11: Acquiring historical data of intravenous treatment patients;

[0028] Step S12: Risk classification is performed on the historical data of intravenous treatment patients, thereby obtaining normal treatment patient data and treatment risk patient data, wherein the treatment risk patient data includes patient basic data, treatment observation data, and patient vital sign data, wherein the patient vital sign data includes blood pressure data, heart rate data, respiratory rate data, body temperature data, body mass index data, skin data, and white blood cell count data;

[0029] Step S13: marking the treatment observation data with risk periods, thereby obtaining risk period marking data, wherein the risk period marking data includes thromboembolism marking data, hematoma marking data, infection marking data, venous access failure marking data, allergy marking data, and phlebitis marking data;

[0030] Step S14: constructing a multidimensional association matrix for the patient's basic data and the patient's vital sign data according to the risk period marking data, thereby obtaining a venous risk association matrix.

[0031] Embodiments of the present invention obtain historical data of intravenous therapy patients from medical record systems, electronic health records, medical records, or other relevant data sources. This data may include basic patient information, treatment indicators, and symptom records. The collected data is then organized and cleaned to ensure accuracy and consistency. This may include removing missing values, processing outliers, and converting data formats. The historical data of intravenous therapy patients is then categorized based on known risk indicators, such as thromboembolism, infection, and intravenous access failure. Expert knowledge or statistical analysis methods can be used to determine the classification criteria, dividing the patient historical data into two categories: normal treatment patient data and treatment risk patient data. Normal treatment patient data represents patients without significant risk, while treatment risk patient data includes patients with certain risk factors. From the treatment risk patient data, basic patient data (such as age and gender), treatment observation data (such as treatment monitoring indicator records), and patient vital signs (such as blood pressure, heart rate, respiratory rate, temperature, body mass index, skin condition, and white blood cell count) are extracted. Based on the different risks associated with intravenous therapy, the type of risk period marker is determined. Common risk period markers include thromboembolism, hematoma, infection, venous access failure, allergies, and phlebitis. Based on treatment observation data, periods associated with specific risks are marked. This can be determined by setting thresholds, rules, or models. For example, if blood pressure rises above a certain threshold or symptoms associated with infection occur, this is marked as an infection risk period. Patient basic data and vital signs are extracted from the risk period marker data. This data includes basic patient information, as well as monitoring indicators and vital signs during the risk period. A multidimensional association matrix is constructed using this extracted data. This matrix can be used to calculate the degree of association between different data sets using statistical analysis methods (such as correlation analysis) or machine learning algorithms (such as association rule mining). The constructed intravenous risk association matrix can be used for further analysis and applications, such as risk prediction and risk screening. By analyzing the associations in the matrix, factors and patterns closely associated with intravenous treatment risks can be identified, thereby supporting patient treatment decisions and risk management.

[0032] The present invention first provides a comprehensive understanding of the patient's condition, treatment process and risk factors by obtaining historical data of intravenous treatment patients. By analyzing the patient's historical data, risk factors, patient characteristics and abnormal conditions during the treatment process related to intravenous treatment can be identified, providing a basis for risk assessment and prediction. Risk classification can divide intravenous treatment patients into normal treatment patients and treatment risk patients, thereby better distinguishing patient groups with different risk levels. Normal treatment patient data can be used as a benchmark for establishing a risk assessment model, while treatment risk patient data provides a more challenging sample, which helps to identify risk factors and predict risk development. Risk period marking can divide treatment observation data into different risk periods and clarify the time range for the occurrence of risk events. By marking risk periods, patient observation data related to specific risks can be more accurately captured and analyzed, providing a more refined time dimension for risk assessment and prediction. Constructing a multidimensional correlation matrix based on risk period marking data can reveal the mutual relationship and correlation strength between basic data and vital sign data of different patients. These correlation matrices can be used to identify the correlation between patient characteristics and specific risks, further understand the importance of risk factors and the comprehensive assessment of patient risks.

[0033] In this embodiment, reference Figure 3 The above is a flowchart of the detailed implementation steps of step S14. In this embodiment, the detailed implementation steps of step S14 include:

[0034] Step S141: performing deep vein thrombosis time-frequency correlation analysis and pulmonary embolism time-frequency correlation analysis on the blood pressure data, heart rate data, and respiratory rate data based on the thromboembolism marker data, thereby obtaining thrombosis sign time-frequency correlation data, wherein the deep vein thrombosis time-frequency correlation analysis and pulmonary embolism time-frequency correlation analysis include blood pressure elevation time-frequency correlation analysis, arrhythmia time-frequency correlation analysis, and tachypnea time-frequency correlation analysis;

[0035] Step S142: performing hematoma feature association clustering on the skin data according to the hematoma marking data, thereby obtaining hematoma association clustering data, wherein the hematoma feature association clustering includes hematoma location association clustering, hematoma complication association clustering, bleeding volume association clustering, and hematoma size association clustering;

[0036] Step S143: performing blood infection change gradient correlation analysis and local infection change gradient correlation analysis on the body temperature data and the white blood cell count data according to the infection marker data, thereby obtaining infection sign change correlation data;

[0037] Step S144: performing association clustering on the body mass index data according to the venous access failure mark data, thereby obtaining venous access failure association clustering data, wherein the association clustering includes catheter blockage association clustering and needle drop association clustering;

[0038] Step S145: performing association clustering of allergy characteristics during intravenous treatment and skin test on the skin data according to the allergy marker data, thereby obtaining allergy association clustering data;

[0039] Step S146: performing a local edema degree time series correlation analysis on the skin data according to the phlebitis marker data, thereby obtaining phlebitis sign correlation data, wherein the local edema degree time series correlation analysis includes a chemical phlebitis time series correlation analysis and a mechanical phlebitis time series correlation analysis;

[0040] Step S147: construct a multidimensional correlation matrix of the patient's basic data, thrombosis sign time-frequency correlation data, hematoma correlation cluster data, infection sign change correlation data, venous channel failure correlation cluster data, allergy correlation cluster data, and phlebitis sign correlation data, thereby obtaining a venous risk correlation matrix.

[0041] In embodiments of the present invention, relevant blood pressure, heart rate, and respiratory rate data are extracted from treatment observation data. These data can be values recorded regularly during treatment. Time-frequency correlation analysis is performed on the thromboembolic marker data and the blood pressure, heart rate, and respiratory rate data. Methods such as wavelet transform and time-frequency atlas can be used to determine the correlation between thrombosis and blood pressure, heart rate, and respiratory rate. Time-frequency correlation analysis is performed on the thromboembolic marker data and the blood pressure, heart rate, and respiratory rate data to determine the correlation between pulmonary embolism and blood pressure, heart rate, and respiratory rate. Similarly, methods such as wavelet transform and time-frequency atlas can be used for analysis. Based on the analysis results, correlation data between thrombotic signs (such as elevated blood pressure, irregular heartbeat, and tachypnea) and time-frequency data are obtained to describe the temporal relationship between thromboembolic risk and blood pressure, heart rate, and respiratory rate. Skin-related data, including skin condition, color, and degree of swelling, are extracted from the treatment observation data. This data can be obtained by observing and recording the patient's skin condition. Correlation cluster analysis is performed on the extracted skin data based on the hematoma marker data. Clustering algorithms (such as K-means clustering, hierarchical clustering, etc.) can be used to group skin data with similar hematoma characteristics into one category. Based on the results of the cluster analysis, hematoma-related clustering data is obtained, including hematoma location-related clustering (grouping data with similar hematoma locations into one category), hematoma complication-related clustering (grouping data with similar complications into one category), bleeding volume-related clustering (grouping data with similar bleeding volume into one category), and hematoma size-related clustering (grouping data with similar hematoma size into one category). Data related to body temperature and white blood cell count are extracted from the treatment observation data. These data can be values recorded regularly during the treatment process. Based on the infection marker data, the gradient correlation analysis of blood infection changes is performed with the body temperature data and the white blood cell count data. Correlation analysis, regression analysis, and other methods can be used to determine the degree of correlation and gradient change between body temperature and white blood cell count and blood infection, and the local infection gradient correlation analysis is performed on the infection marker data, body temperature data, and white blood cell count data. This can include analysis of the degree of correlation and gradient change between infection marker data at specific sites and body temperature and white blood cell count data at the corresponding sites. Based on the analysis results, data on the correlation between infection signs (such as increased body temperature, increased white blood cell count, etc.) and time are obtained to describe the temporal relationship between infection risk and body temperature and white blood cell count. Data related to body mass index are extracted from treatment observation data. These data can be weight and height data recorded regularly during treatment, so as to calculate the body mass index. Based on the venous access failure marker data, the extracted body mass index data is subjected to association cluster analysis.Clustering algorithms (such as K-means clustering and hierarchical clustering) can be used to group body mass index data with similar failure characteristics. Based on the results of the cluster analysis, cluster data associated with intravenous access failure is obtained, including clusters associated with catheter blockage (grouping data with similar catheter blockage characteristics) and clusters associated with styecaulk loss (grouping data with similar styecaulk loss characteristics). Skin-related data, including skin condition, color, and degree of swelling, can be extracted from the treatment observation data. This data can be obtained by observing and recording the patient's skin condition. Based on the allergy marker data, an association cluster analysis is performed on the extracted skin data. Clustering algorithms (such as K-means clustering and hierarchical clustering) can be used to group skin data with similar allergy characteristics. Based on the results of the cluster analysis, allergy-related cluster data is obtained, including clusters associated with allergies during intravenous treatment (grouping data with similar allergy characteristics) and clusters associated with allergies during skin testing (grouping data with similar allergy characteristics). Skin-related data, including skin condition, color, and degree of swelling, can be extracted from the treatment observation data. These data can be obtained by observing and recording the patient's skin condition, and a time-series correlation analysis of chemical phlebitis can be performed based on the phlebitis marker data and the skin data. Correlation analysis, regression analysis and other methods can be used to determine the degree of correlation and time-series changes between chemical phlebitis and the degree of local edema, and a time-series correlation analysis of mechanical phlebitis can be performed based on the phlebitis marker data and the skin data. Similarly, correlation analysis, regression analysis and other methods can be used to determine the degree of correlation and time-series changes between mechanical phlebitis and the degree of local edema. According to the analysis results, phlebitis sign correlation data are obtained, including local edema degree correlation data of chemical phlebitis and local edema degree correlation data of mechanical phlebitis. According to the thrombosis marker data and the time-frequency correlation analysis method, the correlation degree and time-series changes between thrombosis signs and other related factors are analyzed. According to the hematoma marker data and the correlation cluster analysis method, the hematoma data with similar characteristics are classified into one category to obtain hematoma correlation cluster data. According to the infection marker data and the correlation analysis method, the correlation degree and time between the changes in infection signs and other factors are analyzed. According to the sequence changes, the failure data with similar characteristics are classified into one category according to the venous channel failure marker data and the association cluster analysis method to obtain the venous channel failure association cluster data. According to the allergy marker data and the association cluster analysis method, the data with similar allergy characteristics are classified into one category to obtain the allergy association cluster data. According to the phlebitis sign association data, including the local edema degree association data of chemical phlebitis and the local edema degree association data of mechanical phlebitis, the patient's basic data, thrombosis sign time-frequency association data, hematoma association cluster data, infection sign change association data, venous channel failure association cluster data, allergy association cluster data and phlebitis sign association data are integrated together to construct a multidimensional association matrix.Statistical analysis methods (such as correlation analysis, covariance analysis, etc.) can be used to calculate the degree of association between different factors, and the results can be filled into the matrix. Based on the constructed multidimensional association matrix, the venous risk association matrix can be obtained, which can be used to evaluate the risk level of patients in intravenous treatment and the correlation between different factors.

[0042] The present invention first obtains the time-frequency correlation data of thrombosis signs through time-frequency correlation analysis. Specifically, deep vein thrombosis and pulmonary embolism are serious risk complications in intravenous treatment, which are closely related to physiological indicators such as blood pressure, heart rate and respiratory rate. By performing time-frequency correlation analysis, the early signs and change patterns of thrombosis and pulmonary embolism can be identified, and possible risk situations can be discovered; hematoma is one of the common complications in intravenous treatment. By correlating and clustering the hematoma characteristics, the correlation between the location, complications, amount of bleeding and size of the hematoma can be identified, which helps to understand the mechanism of hematoma occurrence, predict risks and take corresponding preventive measures; infection is one of the common complications in intravenous treatment. By performing blood infection change gradient correlation analysis and local infection change gradient correlation analysis, the change pattern of body temperature and white blood cell count and the relationship between infection and infection can be identified. By correlating and clustering the body mass index data, we can identify factors related to venous channel failure, such as catheter blockage and needle detachment, which helps to assess the risk of channel failure; allergy is an issue that requires special attention in intravenous treatment. By correlating and clustering the allergy characteristics of skin data, we can identify features related to allergic reactions during intravenous treatment and skin tests, which helps to predict the patient's allergy risk; phlebitis is one of the common complications of intravenous treatment. By analyzing the temporal correlation of local edema, we can identify early signs and change patterns of phlebitis; integrating the basic data of patients and the correlation data of various risk factors to form a multidimensional correlation matrix helps to comprehensively assess the patient's intravenous treatment risk.

[0043] Preferably, the specific steps of step S147 are:

[0044] Extracting treatment operation features of the time-series marked coincident segments of the venous access failure associated cluster data and the phlebitis sign associated data, thereby obtaining operation mark data, wherein the treatment operation feature extraction includes intubation operation feature extraction and infusion operation feature extraction;

[0045] Clustering the drug treatment time series characteristics of the treatment observation data to obtain drug time series clustering data;

[0046] According to the drug time series clustering data, correlation matching is performed on the thrombosis sign time-frequency correlation data and the allergy correlation clustering data to obtain drug matching data;

[0047] Based on the basic data of the patients, multiple regression analysis was performed on the time-frequency correlation data of thrombosis signs, the clustering data of hematoma correlation, the correlation data of infection signs changes, the clustering data of venous access failure correlation, the clustering data of allergy correlation, and the correlation data of phlebitis signs, thereby obtaining the multiple regression correlation data;

[0048] According to the multivariate regression correlation data, the patient's basic data is enhanced with the long-term and short-term memory features of the patient's related medical history, thereby obtaining a synthetic related medical history dataset;

[0049] Perform random forest risk prediction on the synthetic relevant medical history dataset and multivariate regression association data to obtain multivariate risk prediction data, where the random forest risk prediction includes age risk prediction, gender risk prediction, and patient-related medical history risk prediction;

[0050] A multidimensional association matrix was constructed for the multivariate risk prediction data, drug matching data, and operation marker data to obtain the venous risk association matrix.

[0051] The embodiment of the present invention uses association cluster analysis to classify venous access failure data with similar characteristics into one category, analyzes the degree of association between phlebitis sign data, and performs time series marking of overlapping segments of venous access failure association cluster data and phlebitis sign association data. In the overlapping segments, features related to treatment operations are extracted, including intubation operation features (such as intubation time, intubation position, etc.) and infusion operation features (such as infusion speed, infusion drugs, etc.). Based on the extracted treatment operation features, operation mark data are generated, including intubation operation mark data and infusion operation mark data. Features related to drug treatment are extracted from the treatment observation data, including drug usage time, drug dosage, drug type, etc. Cluster analysis methods (such as K-means clustering, hierarchical clustering, etc.) are used to classify data with similar drug treatment time series features. Based on the results of cluster analysis, drug time series cluster data are obtained, including cluster data of different drug treatment time series features. Based on the drug time series cluster data, thrombosis sign time-frequency association data and allergy association cluster data are correlated with it to determine the degree of correlation between drugs and thrombosis signs and allergies. Based on the results of correlation matching, drug matching data are obtained, including correlation data between drugs and thrombosis signs and correlation data between drugs and allergies. A multivariate regression analysis was performed using patient basic data as independent variables and time-frequency correlation data of thrombosis signs, clustered data associated with hematomas, correlation data of changes in infection signs, clustered data associated with venous access failure, clustered data associated with allergies, and correlation data associated with phlebitis signs as dependent variables. The results of the multivariate regression analysis determined the degree of correlation between the patient basic data and various related data, generating multivariate regression correlation data. Using the multivariate regression correlation data as input, a long short-term memory (LSTM) model or other appropriate sequence model was used to enhance the features of the patient basic data. This process captures both long-term and short-term dependencies related to the patient's medical history. Based on the results of the LSTM feature enhancement, a synthetic related medical history dataset was generated, which included the patient basic data and its correlation with various related data. Risk prediction was then performed using the synthetic related medical history dataset and the multivariate regression correlation data as input using a random forest model. Age risk, gender risk, and patient-related medical history risk can be predicted separately. Based on the results of random forest risk prediction, multivariate risk prediction data can be obtained, including age risk prediction, gender risk prediction, and patient-related medical history risk prediction. The multivariate risk prediction data can be comprehensively considered and the patient's age risk, gender risk, and patient-related medical history risk can be comprehensively evaluated based on pre-set evaluation indicators and weights. Weighted average, logistic regression, or other appropriate methods can be used for evaluation. Based on the results of the comprehensive evaluation, the patient's risk assessment result can be obtained, which can be a categorical result (such as low risk, medium risk, high risk) or a continuous value result (such as a risk score).

[0052] The present invention first extracts the treatment operation features of the time-series marked overlapping segments of the intravenous channel failure associated clustering data and the phlebitis sign associated data, and can obtain the treatment operation features related to intubation and infusion. In the process of intravenous treatment, different operation features will also lead to changes in patient risks. Therefore, these key operation features need to be used as an important influencing factor in risk assessment. Drug treatment plays an important role in intravenous treatment. The timing of drug use has a certain impact on the treatment effect and safety. By clustering the drug treatment timing features of the treatment observation data, the timing patterns and features of different drug uses can be identified. These timing clustering data help to reveal the regularity and potential problems of drug treatment. In the process of intravenous treatment, the drugs used by patients will affect the changes in patient risks. Therefore, these key drug features also need to be used as an important influencing factor in risk assessment; specifically, drug use and thrombosis signs There is a certain correlation between the drug and allergic reactions. By matching the drug time series clustering data with the thrombosis sign time-frequency association data and the allergy association clustering data, the potential association between the drug and these risk factors can be determined; multiple regression analysis can be performed to correlate and couple the patient's basic data with the risk factors related to intravenous treatment, helping to build a model in the future. The patient's basic data serves as an important reference for the model, improving the practicality and applicability of the model. At the same time, by performing multiple regression analysis on the patient's basic data and thrombosis sign time-frequency association data, hematoma association clustering data, infection sign change association data, venous channel failure association clustering data, allergy association clustering data and phlebitis sign association data, the interaction between these variables and their impact on risk can be revealed. The multiple regression association data provides a quantitative assessment of the contribution of different variables to risk, which can help identify important risk factors.Enhanced long- and short-term memory features can help fill in information gaps in relevant medical history data. For example, a patient may not have a certain medical history, or some of the patient's medical history data may be missing. However, based on the changing trend characteristics of relevant data such as physical signs and age, the probability of these risks occurring can be used to simulate the relevant medical history data that may occur, and provide sufficient training data for subsequent medical history-related questions such as allergies and thrombosis. The synthesized relevant medical history dataset can provide more comprehensive and accurate patient medical history information. Random forest risk prediction can comprehensively consider the impact of multiple variables on risk, which can more accurately predict risk. Multivariate risk prediction data includes age risk prediction, gender risk prediction, and patient-related medical history risk prediction. This key prediction information directly related to the patient can provide an important risk assessment dimension for the subsequent construction of an associated risk matrix. Constructing a multidimensional association matrix of multivariate risk prediction data, drug matching data, and operation marker data can comprehensively consider the correlation between multiple risk factors, namely, the patient's physical sign-related data, as well as the patient's age, gender, and medical history-related data, as well as the impact data of drug use and operation conditions during intravenous treatment. The intravenous risk association matrix provides an intuitive way to visualize and analyze the relationship between different risk factors.

[0053] Preferably, the specific steps of enhancing the long-term and short-term memory features of the patient's medical history based on the patient's basic data according to the multivariate regression correlation data are as follows:

[0054] Obtain relevant medical history regression coefficients based on multiple regression correlation data;

[0055] According to the relevant medical history regression coefficient, the patient's basic data is statistically analyzed in terms of the relevant medical history time series, thereby obtaining the medical history time series data;

[0056] Perform sliding window feature extraction on the patient's basic data based on the medical history time series data to obtain average drug use window data, disease type window data, and surgery type window data;

[0057] Perform discrete feature encoding on the average drug use window data, disease type window data, and surgery type window data to obtain time window encoding data;

[0058] The long and short-term memory features of the multivariate regression correlation data are enhanced according to the time window encoding data to obtain a synthetic related medical history dataset.

[0059] In an embodiment of the present invention, regression analysis is performed using patient basic data and relevant medical history data as independent variables and relevant medical history time series data as dependent variables. The regression analysis yields a regression coefficient for the relevant medical history. Time series statistics of each patient's relevant medical history are then generated using the patient basic data and the relevant medical history regression coefficient. The relevant medical history values for each patient at different time points can be calculated, or the changing trend of the relevant medical history can be statistically analyzed. The statistically generated relevant medical history time series data can be organized into a suitable format, such as chronologically arranged, to form a medical history time series data set. A sliding window approach is then used to extract features from the patient basic data. The window size and step size can be defined to extract features within different time windows from the time series. During the sliding window feature extraction process, the average drug usage of each patient within each window is calculated based on medical history features related to drug use, resulting in average drug usage window data. During the sliding window feature extraction process, the type of disease suffered by the patient within each window is statistically analyzed based on medical history features related to disease type, resulting in disease type window data. During the sliding window feature extraction process, the type of surgery performed by the patient within each window is statistically analyzed based on medical history features related to surgery type, resulting in surgery type window data. Discrete feature encoding is performed on the average medication usage window data, disease type window data, and surgery type window data. Methods such as one-hot encoding and label encoding can be used to convert discrete features into numerical representations that can be used by machine learning algorithms. The average medication usage window data, disease type window data, and surgery type window data that have undergone discrete feature encoding are then integrated to form a time window encoded dataset. Using this time window encoded data as input, a long short-term memory (LSTM) neural network or other applicable sequence model is applied to perform feature enhancement on the multivariate regression correlation data. LSTM can capture long-term dependencies in time series data, thereby extracting more informative features. The results of LSTM feature enhancement are used as a synthetic related medical history dataset. This dataset combines multivariate regression correlation data with medical history time series data, providing richer information for subsequent analysis and application.

[0060] The present invention first obtains relevant medical history regression coefficients by analyzing multivariate regression correlation data. These coefficients represent the importance of relevant medical history in predicting risk. The relevant medical history regression coefficients can be used to evaluate the contribution of different medical history factors to risk, so as to determine the medical history information that needs to be focused on. The relevant medical history regression coefficients can be used to perform time series statistics on the relevant medical history in the patient's basic data, so as to capture the changing trends and patterns of the relevant medical history, and provide more comprehensive information for predicting and managing intravenous treatment risks. The sliding window feature extraction of the medical history time series data can convert the time series data into window-level features. The average drug use window data, disease type window data and surgery type window data are window-level statistical features extracted from the medical history time series data, which can more effectively represent the patient's historical treatment situation and disease characteristics. Specifically, the medical history time series data contains a lot of information, but it is not directly applicable to prediction. The model may lead to problems of excessive dimensionality and overfitting, and the feature data within these time windows are the most easily found relevant data in the patient's medical history, which can be directly used as the relevant change features in different time windows of the model predictor; by discrete feature encoding of the average drug usage window data, disease type window data and surgery type window data, these data can be converted into discrete features, and different window types and degrees of influence are converted into discrete features, which is helpful for the subsequent marking and feature representation of event nodes in the memory model; using time window encoded data as input, the LSTM model can be used to enhance the features of multivariate regression correlation data, capture long-term dependencies and short-term patterns in the data, so as to better predict the impact of relevant medical history, and combine the original multivariate regression correlation data with medical history time series information to generate a synthetic relevant medical history data set for further analysis and prediction model construction.

[0061] Preferably, the specific steps of enhancing the long-short term memory features of the multivariate regression correlation data according to the time window encoding data are as follows:

[0062] Perform time series decay trend analysis on the time window coded data to obtain the thrombus time decay weight and allergy time decay weight;

[0063] A bidirectional long-short term memory model was constructed for the multivariate regression correlation data based on the thrombosis time decay weight and the allergy time decay weight, thereby obtaining a bidirectional memory model for medical history.

[0064] Based on the medical history bidirectional memory model, the long-term and short-term memory features of the multivariate regression correlation data are enhanced to obtain a synthetic related medical history dataset.

[0065] The embodiments of the present invention use time window encoding data to perform time series decay analysis on features related to thrombosis and allergies. Methods such as exponentially weighted averaging or sliding average can be used to assign different weights to features within the time window. Based on the results of the time series decay trend analysis, thrombosis time decay weights and allergy time decay weights are obtained. These weights are used in subsequent model construction and feature enhancement to adjust the importance of different features based on changes in time. The thrombosis time decay weights and allergy time decay weights are used as input weights to construct a bidirectional long short-term memory (BLSTM) model. The BLSTM model can simultaneously consider past and future information in the time series, more comprehensively capturing time dependencies. Based on the construction results of the bidirectional long short-term memory model, a medical history bidirectional memory model is obtained. This model can perform time series analysis and prediction on multivariate regression correlation data, taking into account the time series decay weights of thrombosis time and allergy time. The medical history bidirectional memory model is used to enhance the features of multivariate regression correlation data. Taking multivariate regression correlation data as input, more informative features are extracted through the medical history bidirectional memory model. The results after enhancement through long-term and short-term memory features are used as a synthetic related medical history dataset. This dataset combines multivariate regression correlation data with time-decay weights, providing richer information for subsequent analysis and application.

[0066] The present invention first obtains the thrombosis time decay weight and the allergy time decay weight by performing a time series decay trend analysis, which is used to measure the importance of the time factor to the prediction result. Different time points in the time window encoding data have different degrees of influence on the prediction result. The thrombosis time decay weight and the allergy time decay weight can help adjust the prediction model's attention to different time points and improve the modeling ability of the model for time factors; using the thrombosis time decay weight and the allergy time decay weight, the weight distribution of the BiLSTM model can be adjusted according to the time factor, the modeling ability for different time points can be enhanced, and the forward and backward dependencies of the time series data can be captured at the same time; using the medical history bidirectional memory model, the multivariate regression correlation data can be feature enhanced to better capture the time series characteristics of the relevant medical history. The synthesized relevant medical history data set contains richer time series information, which helps to improve the performance and interpretability of the prediction model.

[0067] Preferably, the specific steps of step S2 are:

[0068] Step S21: extracting similarity-related features of the time series windows before and after risk marking from the data of patients undergoing normal treatment according to the venous risk association matrix, thereby obtaining similarity-related data before risk and similarity-related data during risk;

[0069] Step S22: integrating adjacent time series segments of the similarity-related data before the risk and the similarity-related data during the risk, thereby obtaining similarity-related data during the risk;

[0070] Step S23: performing risk trend feature analysis on risk similarity related data to obtain risk trend data;

[0071] Step S24: performing correlation weight determination on the venous risk association matrix based on the risk trend data, thereby obtaining a risk-weighted association matrix;

[0072] Step S25: constructing a naive Bayesian probability model for the risk-weighted association matrix, thereby obtaining an initial venous risk probability model.

[0073] In this embodiment of the present invention, data from patients undergoing normal treatment are divided into pre- and post-temporal windows based on a venous risk association matrix. These windows can be fixed time periods of equal length or divided according to specific events or markers. Within each time window, risk-related features are extracted. These features can include physiological indicators, medical history data, laboratory results, and so on. Correlation analysis and other methods are used to identify risk-related features. Pre-risk similarity-related data and risk-related similarity-related data are obtained based on the extracted similarity-related features. The former represents risk-related feature data before the risk occurs, while the latter represents risk-related feature data at the time of the risk. The pre-risk similarity-related data and risk-related similarity-related data are then integrated into adjacent time series segments. A sliding window approach can be used to merge adjacent time series segments into a longer time series sequence. The integrated time series sequence is the risk similarity-related data. This dataset captures the temporal continuity of risk-related features. Trend analysis is performed on the risk similarity-related data to explore risk trends. Statistical methods, time series analysis, or machine learning methods can be used to extract features related to risk trends. Risk trend data is then obtained based on the results of the risk trend feature analysis. This data reflects the risk change trend in risk-related data with similar risks. Risk trend data is used to perform weighted discrimination on the correlation in the venous risk association matrix. The weights of each correlation item can be calculated using correlation coefficients, information entropy, machine learning methods, etc. Based on the results of correlation weight discrimination, the weights are applied to the venous risk association matrix to obtain a risk-weighted correlation matrix. This matrix reflects the importance of different correlation items in risk prediction. Based on the risk-weighted correlation matrix, a venous risk probability model is constructed using the naive Bayes algorithm. The naive Bayes model can use the conditional independence assumption between correlation items to calculate the probability of each correlation item under given risk conditions. Based on the results of the naive Bayes probability model construction, an initial venous risk probability model is obtained. This model can be used to preliminarily assess a patient's venous risk.

[0074] The present invention first captures the change pattern of patient data before and after the risk occurs by extracting similar related features of the time series windows before and after the risk mark, so as to better understand the state difference of patient data before and after the risk occurs; by integrating adjacent time series segments, the dimension of the data can be reduced, and the features before and after the risk in the complete time series features of a certain risk are compared. If they are similar to the change features of the normal treatment patient data, they are integrated as adjacent features, thereby forming a variety of feature combinations, but all are similar, which improves the completeness and standardization of the features in the time series; by performing trend feature analysis on risk-similar related data, the risk change trend of patient data can be revealed, reflecting the state change of patient data before and after the risk occurs; by performing correlation weight judgment on risk trend data, the importance weight of different patient data in risk prediction can be determined. After obtaining the risk-weighted correlation matrix, its contribution to the risk prediction model can be adjusted according to the correlation weight of the patient data, thereby improving the accuracy of the model; by constructing a naive Bayesian probability model based on the risk-weighted correlation matrix, the correlation relationship and weight between patient data can be considered, thereby further improving the accuracy of risk prediction.

[0075] Preferably, the specific steps of step S3 are:

[0076] Step S31: performing a risk prediction factor probability analysis based on the initial venous risk probability model to obtain risk prediction factor data, wherein the risk prediction factor data includes thromboembolism prediction factor data, hematoma prediction factor data, infection prediction factor data, and venous access failure prediction factor data;

[0077] Step S32: performing risk probability change analysis on the risk prediction factor data, thereby obtaining cannula thrombosis risk probability data, coagulation thrombosis risk probability data, platelet hematoma risk probability data, cannula infection risk probability data, and venous failure location risk probability data;

[0078] Step S33: Based on the cannula thrombosis risk probability data, the coagulation thrombosis risk probability data, the platelet hematoma risk probability data, the cannula infection risk probability data, and the venous failure location risk probability data, the venous risk probability initial model is divided into characteristic states of different risk symptom recovery conditions and characteristic states of different risk severity levels, thereby obtaining a recovery state data set and a severity state data set;

[0079] Step S34: performing shallow multi-dimensional assessment model learning on the initial model of venous risk probability according to the recovery status data set and the severity status data set, thereby obtaining a patient risk assessment model.

[0080] The embodiments of the present invention analyze patient data based on an initial venous risk probability model to calculate the probability of each risk prediction factor. These risk prediction factors can include characteristic data related to thromboembolism, hematoma, infection, and venous access failure. Based on the results of the probability analysis, a risk prediction factor dataset is obtained. This dataset includes thromboembolism prediction factor data, hematoma prediction factor data, infection prediction factor data, and venous access failure prediction factor data, which are used for subsequent risk probability change analysis. The risk prediction factor data is analyzed to calculate the risk probability changes for cannula thrombosis, coagulation thrombosis, platelet hematoma, cannula infection, and venous access failure location. Statistical methods, machine learning, and other techniques can be used to calculate risk probabilities based on the combination and change trends of the predictors. Based on the results of the risk probability change analysis, cannula thrombosis risk probability data, coagulation thrombosis risk probability data, platelet hematoma risk probability data, cannula infection risk probability data, and venous access failure location risk probability data are obtained. These data reflect the probability information of different risks. Based on data on the probability of cannula thrombosis, coagulation thrombosis, platelet hematoma, cannula infection, and venous malfunction location, recovery status is classified into characteristic states for different risk symptoms. This can be determined using thresholds or other statistical methods. Based on the results of the characteristic state classification, a recovery status dataset and a severity status dataset are generated. The recovery status dataset includes sample data on the recovery status of different risk symptoms and is used to assess patient recovery. The severity status dataset includes sample data at different risk severity levels and is used to assess the severity of the patient's condition. Using the recovery status and severity status datasets as training data, shallow machine learning is performed on the initial venous risk probability model. Various machine learning algorithms, such as decision trees, logistic regression, and support vector machines, can be used to construct a patient risk assessment model based on the input features and label data. The patient risk assessment model is then generated based on the shallow learning results of the multidimensional assessment model. This model can be used to predict patients' venous-related risks based on their risk predictor data and assess their recovery status and severity.

[0081] The present invention first analyzes the probabilities of predictors such as thromboembolism, hematoma, infection, and venous access failure to assess a patient's potential risk level across different risk categories. Based on an initial venous risk probability model, a probability analysis of different risk predictors can be performed to understand their contribution to venous risk. Probabilistic change analysis of risk predictor data can be performed to understand the probability change trends of different risks, allowing for a more accurate assessment of a patient's risk level across different risk categories. Based on risk probability data for cannula thrombosis, coagulation thrombosis, platelet hematoma, cannula infection, and venous access failure location, the recovery status of different risk symptoms can be classified into characteristic states. Similarly, the severity of different risks can be classified into characteristic states based on the risk probability data. By generating recovery status and severity status datasets, the characteristic states of patients across different risk categories can be better described, providing a more comprehensive data foundation for the risk assessment model. Using these recovery status and severity status datasets, a multidimensional assessment model can be constructed that comprehensively considers the characteristic states of different risk factors and their recovery status. Using a shallow learning algorithm, the laws and patterns of the risk assessment model can be learned from the data, extracting valuable features for risk assessment.

[0082] Preferably, the specific steps of step S32 are:

[0083] Step S321: performing a risk probability change analysis of the central venous catheter and peripheral venous catheter intubation time on the thromboembolism prediction factor data, thereby obtaining intubation thrombosis risk probability data;

[0084] Step S322: performing quantitative risk probability change analysis of coagulation factors and antithrombin on the thromboembolism prediction factor data, thereby obtaining coagulation and thrombosis risk probability data, wherein the quantitative risk probability change analysis includes tumor probability change analysis and cancer probability change analysis;

[0085] Step S323: performing a platelet count risk probability change analysis on the hematoma prediction factor data to obtain platelet hematoma risk probability data, wherein the platelet count risk probability change analysis includes a risk probability change analysis when anticoagulants are used and a risk probability change analysis when non-anticoagulants are used;

[0086] Step S324: performing an analysis on the infection prediction factor data regarding the probability of intubation time risk for different immune functions, thereby obtaining intubation infection risk probability data;

[0087] Step S325: Perform risk probability change analysis on the venous access failure prediction factor data relative to the central venous catheter position, thereby obtaining venous failure position risk probability data, wherein the risk probability change analysis relative to the central venous catheter position includes blockage risk probability change analysis, leakage risk probability change analysis, and displacement risk probability change analysis.

[0088] In an embodiment of the present invention, a risk probability change analysis is performed based on the intubation time of central venous catheters and peripheral venous catheters in the thromboembolism prediction factor data. Statistical methods or machine learning techniques can be used to calculate the risk probability change of intubation thrombosis based on the intubation time and related feature data. Based on the results of the risk probability change analysis, intubation thrombosis risk probability data is obtained. This data reflects the impact of intubation time on thrombosis risk and can be used to assess a patient's thrombosis risk during intubation. A risk probability change analysis is performed based on the amount of coagulation factors and antithrombin in the thromboembolism prediction factor data. The relationship between the amount of coagulation factors and antithrombin and thrombosis risk is analyzed, including the risk probability change associated with changes in tumor probability and cancer probability. Based on the results of the risk probability change analysis, coagulation thrombosis risk probability data is obtained. This data reflects the impact of the amount of coagulation factors and antithrombin on thrombosis risk and can be used to assess a patient's coagulation thrombosis risk.

[0089] The present invention first analyzes the risk probability changes of the intubation time of central venous catheters and peripheral venous catheters to evaluate the risk of thromboembolism in patients during the intubation process. The risk probability of thromboembolism of central venous catheters and peripheral venous catheters is proportional to the increase of intubation time. The relevant risk probabilities of different venous catheters are distinguished to obtain the intubation thrombosis risk probability data, so as to consider more and more critical risk change factors in the evaluation of patients. By analyzing the risk probability changes of coagulation factors and antithrombin amounts, the risk of coagulation thrombosis in patients can be evaluated. The probability change analysis of tumors and cancer can be used to Identify the patient's additional risk factors for thrombosis. By performing a risk probability change analysis on the platelet count, the patient's risk of hematoma formation can be assessed to improve the risk prediction effect. By performing a risk probability change analysis on the intubation time of patients with different immune functions, the patient's risk of infection during the intubation process can be assessed. By performing a risk probability change analysis on the venous channel failure prediction factor data, the differences in failure risks of catheters in different locations can be assessed. The blockage risk probability change analysis, leakage risk probability change analysis, and displacement risk probability change analysis are all key features that can be directly and easily obtained.

[0090] Preferably, the specific steps of step S34 are:

[0091] Step S341: constructing a multi-state feature matrix for the recovery state data set and the severity state data set to obtain a risk state matrix, wherein the multi-state feature matrix construction includes constructing a probability state of risk occurrence, constructing a probability state of severity when the risk occurs, and constructing a recovery state when the risk occurs;

[0092] Step S342: assigning weights to the risk prediction factor data according to the risk status matrix to obtain weighted risk factor data;

[0093] Step S343: Performing shallow multi-dimensional assessment model learning on the initial model of venous risk probability according to the weighted risk factor data, thereby obtaining a patient risk assessment model.

[0094] The embodiments of the present invention analyze risk-related features, such as the patient's medical history and physiological indicators, based on a recovery status dataset. Based on these features, a probability state for risk occurrence is constructed, reflecting the likelihood of the patient experiencing the risk at different time points. Based on a severity status dataset, features related to the severity of the risk at the time of occurrence, such as symptoms and laboratory test results, are analyzed. Based on these features, a probability state for severity at the time of risk occurrence is constructed, reflecting the severity faced by the patient at the time of risk occurrence. Based on a recovery status dataset, features related to the degree of recovery at the time of risk occurrence, such as treatment measures and recovery status, are analyzed. Based on these features, a probability state for degree of recovery at the time of risk occurrence is constructed, reflecting the patient's recovery after the risk occurs. The probability state for risk occurrence, the probability state for severity at the time of risk occurrence, and the degree of recovery at the time of risk occurrence are integrated to form a risk state matrix, and weights are assigned to each feature in the risk prediction factor data. Weights can be determined based on the importance of each state in the risk state matrix. For example, weights can be calculated using expert knowledge, statistical analysis, or machine learning algorithms. The risk prediction factor data is multiplied by the corresponding weights to generate weighted risk factor data. These data take into account the degree of influence of different risk states and can be used for subsequent risk assessment model construction. Based on the weighted risk factor data and the existing initial model of venous risk probability, shallow machine learning algorithms (such as logistic regression, decision trees, etc.) are used for model training and learning. Using weighted risk factor data as input features, the model is trained to learn a multidimensional assessment model related to risk and venous thrombosis. After shallow learning, a patient risk assessment model is obtained. The model can predict the patient's risk level for venous thrombosis based on the input patient information and weighted risk factor data. The final output can include risk scores, probabilities, and risk classification results for different risk indicator directions and specific risk assessment predictors that change over time.

[0095] The present invention first constructs a multi-state feature matrix, which can combine the patient's recovery state and severity state with the probability state of risk occurrence to form a comprehensive risk state matrix. The three state construction dimensions provided by the probability state construction of risk occurrence, the probability state construction of severity when risk occurs, and the recovery state construction when risk occurs can provide a more direct and complete risk assessment direction. By weighting the risk prediction factor data according to the risk state matrix, the contribution of different factors to the risk can be taken into account, thereby improving the accuracy of risk assessment. The weight allocation can be determined based on the importance of different states in the risk state matrix, making it more objective and accurate when calculating weighted risk factor data. By using weighted risk factor data to perform shallow learning of the multidimensional evaluation model on the initial model of venous risk probability, the predictive ability and accuracy of the model can be improved. The multidimensional evaluation model can comprehensively consider the influence of multiple risk factors, learn and optimize based on the weighted risk factor data, and generate a risk assessment model for individual patients.

[0096] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0097] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0098] The foregoing description is intended only to provide specific embodiments of the present invention, intended to enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing a risk assessment model for intravenous therapy patients, characterized in that: The following steps are involved: Step S1: Obtaining historical data of intravenous therapy patients; Risk classification is performed on historical data of intravenous treatment patients to obtain normal treatment patient data and treatment risk patient data; a multidimensional association matrix is constructed for the treatment risk patient data to obtain an intravenous risk association matrix, wherein step S1 is specifically as follows: Step S11: Acquiring historical data of intravenous treatment patients; Step S12: Risk classification is performed on the historical data of intravenous treatment patients, thereby obtaining normal treatment patient data and treatment risk patient data, wherein the treatment risk patient data includes patient basic data, treatment observation data, and patient vital sign data, wherein the patient vital sign data includes blood pressure data, heart rate data, respiratory rate data, body temperature data, body mass index data, skin data, and white blood cell count data; Step S13: marking the treatment observation data with risk periods, thereby obtaining risk period marking data, wherein the risk period marking data includes thromboembolism marking data, hematoma marking data, infection marking data, venous access failure marking data, allergy marking data, and phlebitis marking data; Step S14: construct a multidimensional correlation matrix for the patient's basic data and the patient's vital signs data according to the risk period mark data, thereby obtaining a venous risk correlation matrix, wherein step S14 is specifically as follows: Step S141: performing deep vein thrombosis time-frequency correlation analysis and pulmonary embolism time-frequency correlation analysis on the blood pressure data, heart rate data, and respiratory rate data based on the thromboembolism marker data, thereby obtaining thrombosis sign time-frequency correlation data, wherein the deep vein thrombosis time-frequency correlation analysis and pulmonary embolism time-frequency correlation analysis include blood pressure elevation time-frequency correlation analysis, arrhythmia time-frequency correlation analysis, and tachypnea time-frequency correlation analysis; Step S142: performing hematoma feature association clustering on the skin data according to the hematoma marking data, thereby obtaining hematoma association clustering data, wherein the hematoma feature association clustering includes hematoma location association clustering, hematoma complication association clustering, bleeding volume association clustering, and hematoma size association clustering; Step S143: performing blood infection change gradient correlation analysis and local infection change gradient correlation analysis on the body temperature data and the white blood cell count data according to the infection marker data, thereby obtaining infection sign change correlation data; Step S144: performing association clustering on the body mass index data according to the venous access failure mark data, thereby obtaining venous access failure association clustering data, wherein the association clustering includes catheter blockage association clustering and needle drop association clustering; Step S145: performing association clustering of allergy characteristics during intravenous treatment and skin test on the skin data according to the allergy marker data, thereby obtaining allergy association clustering data; Step S146: performing a local edema degree time series correlation analysis on the skin data according to the phlebitis marker data, thereby obtaining phlebitis sign correlation data, wherein the local edema degree time series correlation analysis includes a chemical phlebitis time series correlation analysis and a mechanical phlebitis time series correlation analysis; Step S147: construct a multidimensional correlation matrix based on the patient's basic data, thrombosis sign time-frequency correlation data, hematoma correlation cluster data, infection sign change correlation data, venous access failure correlation cluster data, allergy correlation cluster data, and phlebitis sign correlation data, thereby obtaining a venous risk correlation matrix. Step S147 is specifically as follows: Extracting treatment operation features of the time-series marked coincident segments of the venous access failure associated cluster data and the phlebitis sign associated data, thereby obtaining operation mark data, wherein the treatment operation feature extraction includes intubation operation feature extraction and infusion operation feature extraction; Clustering the drug treatment time series characteristics of the treatment observation data to obtain drug time series clustering data; According to the drug time series clustering data, correlation matching is performed on the thrombosis sign time-frequency correlation data and the allergy correlation clustering data to obtain drug matching data; Based on the basic data of the patients, multiple regression analysis was performed on the time-frequency correlation data of thrombosis signs, the clustering data of hematoma correlation, the correlation data of infection signs changes, the clustering data of venous access failure correlation, the clustering data of allergy correlation, and the correlation data of phlebitis signs, thereby obtaining the multiple regression correlation data; According to the multivariate regression correlation data, the long-term and short-term memory features of the patient's related medical history are enhanced on the patient's basic data, thereby obtaining a synthetic related medical history data set. The long-term and short-term memory features of the patient's related medical history are enhanced on the patient's basic data according to the multivariate regression correlation data as follows: Obtain relevant medical history regression coefficients based on multiple regression correlation data; According to the relevant medical history regression coefficient, the patient's basic data is statistically analyzed in terms of the relevant medical history time series, thereby obtaining the medical history time series data; Perform sliding window feature extraction on the patient's basic data based on the medical history time series data to obtain average drug use window data, disease type window data, and surgery type window data; Perform discrete feature encoding on the average drug use window data, disease type window data, and surgery type window data to obtain time window encoding data; The long-term and short-term memory features of the multivariate regression correlation data are enhanced according to the time window encoding data, thereby obtaining a synthetic related medical history data set. The long-term and short-term memory features of the multivariate regression correlation data are enhanced according to the time window encoding data as follows: Perform time series decay trend analysis on the time window coded data to obtain the thrombus time decay weight and allergy time decay weight; A bidirectional long-short term memory model was constructed for the multivariate regression correlation data based on the thrombosis time decay weight and the allergy time decay weight, thereby obtaining a bidirectional memory model for medical history. Based on the medical history bidirectional memory model, the long-term and short-term memory features of the multivariate regression correlation data are enhanced to obtain a synthetic related medical history dataset; Perform random forest risk prediction on the synthetic relevant medical history dataset and multivariate regression association data to obtain multivariate risk prediction data, where the random forest risk prediction includes age risk prediction, gender risk prediction, and patient-related medical history risk prediction; A multidimensional correlation matrix is constructed for the multivariate risk prediction data, drug matching data, and operation marker data to obtain a venous risk correlation matrix; Step S2: extract similarity-related features of the time series window before and after risk marking from the data of patients with normal treatment according to the venous risk association matrix, thereby obtaining similarity-related data before risk and similarity-related data during risk; A naive Bayesian probability model was constructed for the similarity-related data before risk and the shape-related data during risk, thereby obtaining an initial model of venous risk probability; Step S3: Based on the initial venous risk probability model, characteristic state division of different risk symptom recovery conditions and characteristic state division of different risk severity levels are performed to obtain a recovery state data set and a severity state data set; The multidimensional evaluation model of the initial venous risk probability model is shallowly learned based on the recovery status dataset and the severity status dataset to obtain a patient risk assessment model.

2. The method for constructing a risk assessment model for intravenous therapy patients according to claim 1, characterized in that: The specific steps of step S2 are: Step S21: extracting similarity-related features of the time series windows before and after risk marking from the data of patients undergoing normal treatment according to the venous risk association matrix, thereby obtaining similarity-related data before risk and similarity-related data during risk; Step S22: integrating adjacent time series segments of the similarity-related data before the risk and the similarity-related data during the risk, thereby obtaining similarity-related data during the risk; Step S23: performing risk trend feature analysis on risk similarity related data to obtain risk trend data; Step S24: performing correlation weight determination on the venous risk association matrix based on the risk trend data, thereby obtaining a risk-weighted association matrix; Step S25: constructing a naive Bayesian probability model for the risk-weighted association matrix, thereby obtaining an initial venous risk probability model.

3. The method for constructing a risk assessment model for intravenous therapy patients according to claim 1, characterized in that: The specific steps of step S3 are: Step S31: performing a risk prediction factor probability analysis based on the initial venous risk probability model to obtain risk prediction factor data, wherein the risk prediction factor data includes thromboembolism prediction factor data, hematoma prediction factor data, infection prediction factor data, and venous access failure prediction factor data; Step S32: performing risk probability change analysis on the risk prediction factor data, thereby obtaining cannula thrombosis risk probability data, coagulation thrombosis risk probability data, platelet hematoma risk probability data, cannula infection risk probability data, and venous failure location risk probability data; Step S33: Based on the cannula thrombosis risk probability data, the coagulation thrombosis risk probability data, the platelet hematoma risk probability data, the cannula infection risk probability data, and the venous failure location risk probability data, the venous risk probability initial model is divided into characteristic states of different risk symptom recovery conditions and characteristic states of different risk severity levels, thereby obtaining a recovery state data set and a severity state data set; Step S34: performing shallow multi-dimensional assessment model learning on the initial model of venous risk probability according to the recovery status data set and the severity status data set, thereby obtaining a patient risk assessment model.

4. The method for constructing a risk assessment model for intravenous therapy patients according to claim 3, characterized in that: The specific steps of step S32 are: Step S321: performing a risk probability change analysis of the central venous catheter and peripheral venous catheter intubation time on the thromboembolism prediction factor data, thereby obtaining intubation thrombosis risk probability data; Step S322: performing quantitative risk probability change analysis of coagulation factors and antithrombin on the thromboembolism prediction factor data, thereby obtaining coagulation and thrombosis risk probability data, wherein the quantitative risk probability change analysis includes tumor probability change analysis and cancer probability change analysis; Step S323: performing a platelet count risk probability change analysis on the hematoma prediction factor data to obtain platelet hematoma risk probability data, wherein the platelet count risk probability change analysis includes a risk probability change analysis when anticoagulants are used and a risk probability change analysis when non-anticoagulants are used; Step S324: performing an analysis on the infection prediction factor data regarding the probability of intubation time risk for different immune functions, thereby obtaining intubation infection risk probability data; Step S325: Perform risk probability change analysis on the venous access failure prediction factor data relative to the central venous catheter position, thereby obtaining venous failure position risk probability data, wherein the risk probability change analysis relative to the central venous catheter position includes blockage risk probability change analysis, leakage risk probability change analysis, and displacement risk probability change analysis.

5. The method for constructing a risk assessment model for intravenous therapy patients according to claim 3, characterized in that: The specific steps of step S34 are: Step S341: constructing a multi-state feature matrix for the recovery state data set and the severity state data set to obtain a risk state matrix, wherein the multi-state feature matrix construction includes constructing a probability state of risk occurrence, constructing a probability state of severity when the risk occurs, and constructing a recovery state when the risk occurs; Step S342: assigning weights to the risk prediction factor data according to the risk status matrix to obtain weighted risk factor data; Step S343: Performing shallow multi-dimensional assessment model learning on the initial model of venous risk probability according to the weighted risk factor data, thereby obtaining a patient risk assessment model.

Images