Method and system for constructing PICC (Peripherally Inserted Central Catheter) related thrombus risk prediction model

Abstract

The invention relates to the technical field of medical risk models, in particular to a method and a system for constructing a PICC (peripherally inserted central catheter) related thrombus risk prediction model. Relevant indexes of the PICC catheter patient are obtained; on the basis of the thrombus risk distinguishing ability of the related indexes, determining thrombus risk distinguishing ability of different feature values under the related indexes; based on the difference of different characteristic values under the same related index, correcting the thrombus risk distinguishing ability to obtain the dividing ability of the related index; determining the division cost of the related indexes by using the similarity between the label distribution characteristics before and after the division of the related indexes; and constructing a corresponding PICC correlation thrombus prediction model according to the division cost of different correlation indexes. According to the method, the optimal attribute of the decision tree at the current node position is selected according to the division cost, so that the decision tree is constructed, and the PICC thrombus risk judgment is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of medical risk models, and particularly to a method and system for constructing a risk prediction model for PICC catheter-related thrombosis. Background Art

[0002] A peripherally inserted central catheter (PICC) refers to a catheter that is inserted through the basilic vein, median cubital vein, cephalic vein, brachial vein, external jugular vein (in neonates, it can also be inserted through the great saphenous vein in the lower limb, temporal vein in the head, posterior auricular vein, etc.) in the upper limb, and the tip is located in the superior vena cava or inferior vena cava. The infused drugs can enter the central vein with a large blood flow and fast flow rate through this pipeline for rapid in vivo circulation. It is a way to provide medium- and long-term intravenous treatment, reducing the damage to peripheral blood vessels caused by long-term infusion or infusion of hypertonic and irritating drugs, thus avoiding the pain of patients, ensuring the smooth progress of important treatments such as parenteral nutrition support and chemotherapy, and improving patient satisfaction. PICC catheter-related thrombosis is the process of forming blood clots on the inner wall of the blood vessel where the PICC is located and on the catheter wall after catheterization. Most patients with PICC catheter-related thrombosis do not have obvious clinical symptoms, and may present as occasional swelling or numbness in the limb on the side of catheterization, which is easily overlooked, leading to the further development of thrombosis, thus inducing pulmonary embolism and endangering life. The subsequent post-thrombotic syndrome hinders the function of venous valves, resulting in limb pain, swelling and dysfunction, affecting the quality of life, and delaying the hospital stay. Therefore, it is necessary to predict the risk of PICC catheter-related thrombosis, judge the thrombosis risk according to the prediction results, and take corresponding preventive measures to ensure the safety of the patients themselves.

[0003] Currently, traditional machine learning-related algorithms such as random forest are usually used to construct a PICC catheter-related thrombosis prediction model. However, when using the traditional random forest algorithm to construct a PICC catheter-related thrombosis prediction model, the following problems exist: Since the incidence of PICC catheter-related thrombosis is relatively low, fewer cases are included in the training set when constructing the prediction model, resulting in an inaccurate prediction model and poor applicability; the time of thrombosis occurrence is not analyzed as a variable, and nurses cannot pay attention to the high-incidence period of PICC catheter-related thrombosis, resulting in poor management efficiency. Summary of the Invention

[0004] In order to solve the technical problem that the construction of a decision tree using the traditional random forest algorithm has inaccurate prediction of PICC catheter-related thrombosis and cannot pay attention to the high-incidence period, thus affecting the thrombosis prediction ability of the random forest, the purpose of the present invention is to provide a method and system for constructing a risk prediction model for PICC catheter-related thrombosis, and the specific technical solutions adopted are as follows:

[0005] In a first aspect, an embodiment of the present invention provides a method for constructing a risk prediction model for PICC catheter-related thrombosis, the method comprising:

[0006] Relevant indicators of PICC catheter patients;

[0007] Based on the ability of the relevant indicators to distinguish thrombosis risks, determine the thrombosis risk discrimination ability of different eigenvalue under the relevant indicators;

[0008] Based on the differences between different eigenvalues under the same relevant indicator, correct the thrombosis risk discrimination ability to obtain the partitioning ability of the relevant indicator;

[0009] Combining the thrombosis risk discrimination ability and the partitioning ability, and using the similarity between the label distribution characteristics before and after the partitioning of the relevant indicators, determine the partitioning cost of the relevant indicators;

[0010] According to the partitioning costs of different types of relevant indicators, construct a corresponding PICC-related thrombosis prediction model.

[0011] Further, the determining the thrombosis risk discrimination ability of different eigenvalues under the relevant indicators based on the ability of the relevant indicators to distinguish thrombosis risks includes:

[0012] Assign corresponding PICC thrombosis risk labels to the eigenvalues of the relevant indicators;

[0013] Based on the types of PICC thrombosis risk labels corresponding to each eigenvalue under the same type of relevant indicator and the proportion of different types of PICC thrombosis risk labels, determine the category characteristics of each eigenvalue under the same type of relevant indicator in the corresponding different types of PICC thrombosis risk labels;

[0014] Construct a data feature vector for each eigenvalue from the category characteristics of each eigenvalue under the same type of relevant indicator corresponding to all types of PICC thrombosis risk labels;

[0015] Based on the data feature vector, determine the thrombosis risk discrimination ability of different eigenvalues under the relevant indicators.

[0016] Further, the determining the category characteristics of each eigenvalue under the same type of relevant indicator in the corresponding different types of PICC thrombosis risk labels based on the types of PICC thrombosis risk labels corresponding to each eigenvalue under the same type of relevant indicator and the proportion of different types of PICC thrombosis risk labels includes:

[0017] Take any eigenvalue under any relevant indicator as a reference value; obtain the number of types of PICC thrombosis risk labels corresponding to the same reference value as the feature dispersion;

[0018] Take any PICC thrombus risk label as the reference thrombus risk label; calculate the proportion of the total number of other PICC thrombus risk labels corresponding to the same reference value except the reference thrombus risk label as the feature similarity proportion.

[0019] Based on the feature dispersion degree and the feature similarity proportion, determine the category feature when the reference value corresponds to the reference thrombus risk label.

[0020] Further, the determining the thrombus risk discrimination ability of different eigenvalue under the relevant index based on the data feature vector includes:

[0021] Take any eigenvalue under any relevant index as the reference value;

[0022] Based on the data feature vector and the number of the reference value corresponding to the relevant index, determine the thrombus risk discrimination ability of the reference value.

[0023] Further, the determining the thrombus risk discrimination ability of the reference value based on the data feature vector and the number of the reference value corresponding to the relevant index includes:

[0024] For the element values in the data feature vector corresponding to the reference value, calculate the ratio of the element value to the number of parameter values respectively to form an adjusted feature vector;

[0025] Based on the number of types of PICC thrombus risk labels, correct the adjusted feature vector to obtain a corrected feature vector;

[0026] Take the modulus of the corrected feature vector as the thrombus risk discrimination ability of the reference value.

[0027] Further, the correcting the thrombus risk discrimination ability based on the difference of different eigenvalues under the same relevant index to obtain the partitioning ability of the relevant index includes:

[0028] Take any relevant index as the reference relevant index, and determine the importance degree of the partitioning ability based on the distribution of the thrombus risk discrimination ability corresponding to different eigenvalues under the same reference relevant index;

[0029] Select the representative value of the thrombus risk discrimination ability from the thrombus risk discrimination abilities corresponding to different eigenvalues of the reference relevant index;

[0030] Based on the importance degree of the reference relevant index, correct the representative value of the thrombus risk discrimination ability corresponding to the reference relevant index to obtain the partitioning ability of the reference relevant index.

[0031] Further, the selecting the representative value of the thrombus risk discrimination ability from the thrombus risk discrimination abilities corresponding to different eigenvalues of the reference relevant index includes:

[0032] Select the maximum value among the thrombus risk discrimination capabilities corresponding to different characteristic values of the reference relevant indicators as the representative value of the thrombus risk discrimination ability of the reference relevant indicators.

[0033] Furthermore, determining the partitioning cost of the relevant indicators based on the similarity between the label distribution characteristics before and after partitioning the relevant indicators includes:

[0034] Compare the similarity between the thrombus risk discrimination ability of the relevant indicators before partitioning and the thrombus risk discrimination abilities of different characteristic values under the relevant indicators after partitioning to determine the partitioning improvement value;

[0035] Combine the partitioning improvement value and the partitioning ability to determine the partitioning cost of the relevant indicators.

[0036] Furthermore, constructing a corresponding PICC-related thrombus prediction model according to the partitioning costs of different types of relevant indicators includes:

[0037] Based on the partitioning costs of different types of relevant indicators of PICC catheter patients, construct multiple decision trees to obtain a random forest model as the corresponding PICC-related thrombus prediction model.

[0038] In a second aspect, a system for constructing a PICC catheter-related thrombus risk prediction model is provided. The system includes:

[0039] An acquisition module that acquires relevant indicators of PICC catheter patients;

[0040] An ability determination module that determines the thrombus risk discrimination abilities of different characteristic values under the relevant indicators based on the thrombus risk discrimination ability of the relevant indicators;

[0041] A correction module that corrects the thrombus risk discrimination ability based on the differences between different characteristic values under the same relevant indicator to obtain the partitioning ability of the relevant indicator;

[0042] A cost determination module that combines the thrombus risk discrimination ability and the partitioning ability and determines the partitioning cost of the relevant indicators by using the similarity between the label distribution characteristics before and after partitioning the relevant indicators;

[0043] A model construction module that constructs a corresponding PICC-related thrombus prediction model according to the partitioning costs of different types of relevant indicators.

[0044] In a third aspect, a server is provided, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory so that the device executes the method in the first aspect or any possible implementation manner of the first aspect.

[0045] In a fourth aspect, a computer program product is provided, which includes computer program code that, when running on a computer, causes the computer to execute the method in the first aspect or any possible implementation of the first aspect as described above.

[0046] In a fifth aspect, a computer-readable storage medium is provided, which stores computer program code that, when running on a computer, causes the computer to execute the method in the first aspect or any possible implementation of the first aspect as described above.

[0047] The embodiments of the present invention have at least the following beneficial effects:

[0048] By obtaining relevant indicators of PICC catheter patients, since the importance degrees of different relevant indicators are different and their influence degrees on the thrombus risk are different, it is necessary to judge the partitioning ability of relevant indicators according to the discrimination ability of thrombus risk. Therefore, further, based on the discrimination ability of the relevant indicators on thrombus risk, determine the thrombus risk discrimination ability of different eigenvalue under the relevant indicators; because there is a problem of low accuracy in determining the discrimination ability of PICC thrombus risk only relying on the distribution characteristics of a single eigenvalue under the same relevant indicator, the discrimination ability of thrombus risk is corrected through the differences of different eigenvalues under the same relevant indicator to obtain the partitioning ability of relevant indicators, thus avoiding the problem that eigenvalues belonging to the same relevant indicator all show a certain difference between the eigenvalue distribution and the uniform distribution to a certain extent, but in fact, the concentrated distribution under one eigenvalue of the same relevant indicator is a normal phenomenon and is an attribute with little correlation with PICC thrombus risk and cannot be used as the main basis for judging thrombus risk; utilize the similarity between the label distribution characteristics before and after the partitioning of relevant indicators to determine the partitioning cost of relevant indicators; construct a corresponding PICC-related thrombus prediction model according to the partitioning costs of different types of relevant indicators. In the present invention, the best attribute at the current node position of the decision tree is selected according to the partitioning cost, thereby completing the construction of the decision tree, making the judgment of PICC thrombus risk more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0050] Figure 1The method flow chart of a method for constructing a PICC catheter-related thrombosis risk prediction model provided by an embodiment of the present invention;

[0051] Figure 2 Another method flow chart of a method for constructing a PICC catheter-related thrombosis risk prediction model provided by an embodiment of the present invention;

[0052] Figure 3 Another method flow chart of a method for constructing a PICC catheter-related thrombosis risk prediction model provided by an embodiment of the present invention;

[0053] Figure 4 Another method flow chart of a method for constructing a PICC catheter-related thrombosis risk prediction model provided by an embodiment of the present invention;

[0054] Figure 5 It is the composition structure diagram of a system for constructing a PICC catheter-related thrombosis risk prediction model provided by an embodiment of the present application;

[0055] Figure 6 It is the structure diagram of a computer device provided by an embodiment of the present application. Detailed implementation manners

[0056] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following combines the drawings and preferred embodiments to elaborate in detail on a method and system for constructing a PICC catheter-related thrombosis risk prediction model proposed according to the present invention, its specific implementation manners, structures, features and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0058] The embodiment of the present invention provides a specific implementation method of a method and system for constructing a PICC catheter-related thrombosis risk prediction model. This method is applicable to the scenario of a PICC-related thrombosis prediction model. This method can determine the selection of the division criterion of the decision tree suitable for thrombus risk judgment by analyzing the importance of relevant indicators in PICC thrombus characteristics and the differentiation of thrombus risk labels before and after division, thereby improving the accuracy of thrombus risk model judgment.

[0059] The following specifically describes the specific solutions of a method and system for constructing a PICC catheter-related thrombosis risk prediction model provided by the present invention with reference to the drawings.

[0060] Please refer to Figure 1 , which shows the flowchart of the steps of a method for constructing a risk prediction model for PICC catheter-related thrombosis provided by an embodiment of the present invention. The method includes the following steps:

[0061] Step S100, obtain relevant indicators of PICC catheter patients.

[0062] Collect relevant information of patients with indwelling PICC catheters, including demographics, disease information, medication information, catheterization, clinical laboratory information, maintenance, and PICC catheter-related thrombosis information.

[0063] Demographic information includes the patient's gender, age, BMI, and educational level.

[0064] Disease information includes the main diagnosis of the patient at admission, tumor stage, whether there is hypertension, diabetes, coronary heart disease, deep vein thrombosis and thrombophlebitis, history of deep vein catheterization, and the reason for catheter indwelling.

[0065] Medication information includes the chemotherapy cycle and the types of chemotherapy drugs.

[0066] Clinical laboratory information includes the values of white blood cells, hemoglobin, platelets, D-dimer, prothrombin time, fibrinogen, activated partial thromboplastin time, albumin, and C-reactive protein.

[0067] Catheterization information includes the model, lumen number, opening type of the PICC catheter, catheterization method, catheter / vessel diameter ratio, puncture site, puncture vessel, number of punctures, catheter tip positioning method, and catheter tip position.

[0068] Maintenance information includes the catheter fixation method, whether the sealing solution is pre-filled, the type of sealing solution, and the type of connector.

[0069] PICC catheter-related thrombosis information includes whether PICC catheter-related thrombosis has occurred and the occurrence date.

[0070] Based on algorithms such as univariate analysis, Cox survival analysis, DeepSurv, and DeepHit, construct a PICC-related thrombosis prediction model and obtain the weight of each factor.

[0071] The relevant indicators of PICC catheter patients mentioned here include: indicators related to body posture such as the age, gender, body mass index, D-dimer level, hemoglobin, demographic information, disease information, medication information, clinical laboratory information, catheterization information, and maintenance information of PICC catheter patients. The relevant indicators of PICC catheter patients can be collected through the hospital's medical record system. The relevant indicators of PICC catheter patients can also be understood as the PICC thrombosis characteristics of PICC catheter patients. It is equivalent to calling the age, gender, body mass index, D-dimer level, hemoglobin, demographic information, disease information, medication information, clinical laboratory information, catheterization information, and maintenance information of PICC catheter patients relevant indicators, but they are divided into different types of relevant indicators. For example, age is a kind of relevant indicator, and gender is a kind of relevant indicator, etc. It should be noted that, for example, if gender is a characteristic expressed in words, it can be replaced with a pre-set label value, or in other embodiments, only relevant indicators represented by numerical values may be considered, without considering relevant indicators represented by words.

[0072] Each relevant indicator has its corresponding characteristic value, and the data is diverse, that is, each relevant indicator corresponds to multiple characteristic values of different sizes.

[0073] As an embodiment of the present invention, the relevant indicators of PICC catheter patients can be input into the deepsurv model to obtain the PICC thrombosis risk label; this deepsurv model is the PICC-related thrombosis prediction model.

[0074] Specifically: significant relevant variables are selected through statistical analysis, including but not limited to: gender, age, height, education level, preliminary diagnosis at admission, activated partial thromboplastin time (APTT), albumin level, catheter fixation situation, type of catheter flushing solution, connector type, catheter opening type, dressing, certain chemotherapy drugs (such as platinum drugs, anthracycline drugs, alkylating agents, plant alkaloids);

[0075] The relevant indicators are input into the deepsurv model, and the risk assessment result of the patient having PICC, that is, the PICC thrombosis risk label, is input. The obtained PICC thrombosis risk label is used as the category label of the corresponding relevant data. By using the PICC thrombosis risk label obtained from the deepsurv model as the label of the decision tree data, the construction of the PICC-related thrombosis prediction model is completed through the supervised learning model of the decision tree.

[0076] Reasons for using the evaluation results of the deepsurv model for the decision tree model: 1. In the first stage, the deepsurv model is used for risk assessment to extract preliminary risk features. In the second stage, the decision tree model is used to further learn and optimize these risk features to improve the accuracy of risk assessment. 2. In this way, secondary optimization of risk assessment can be achieved, that is, on the basis of the deepsurv model, the decision tree model is used to further refine and improve the accuracy of risk prediction. 3. The decision tree is a supervised learning model, and the accuracy and robustness of the judgment of PICC thrombosis risk can be improved by learning the classification labels.

[0077] Furthermore, the PICC thrombosis risk label of each PICC catheter patient is determined, and thus the PICC thrombosis risk label L corresponding to each eigenvalue V of each relevant indicator is determined. Therefore, the eigenvalue of the relevant indicator and the corresponding PICC thrombosis risk label can be expressed as X = {V, L}; where V is the eigenvalue of the relevant indicator; L is the PICC thrombosis risk label corresponding to V.

[0078] Step S200, based on the ability of the relevant indicator to distinguish thrombosis risk, determine the thrombosis risk discrimination ability of different eigenvalues under the relevant indicator.

[0079] The thrombosis risk discrimination ability of the relevant indicator is obtained through the ability of each relevant indicator showing thrombosis characteristics to divide thrombosis risk and the accuracy of the division. The division cost of the relevant indicator is determined by using whether the accuracy of the decision tree for judging thrombosis risk is improved before and after the division, and the selection of the decision tree attribute is realized according to the division cost.

[0080] Since the data obtained from the hospital, especially patient data, is actually biased in reality, a certain proportion of the population has a risk of PICC thrombosis, while the population without thrombosis risk is small, resulting in relatively unbalanced sample sizes under the relevant indicators collected. And for different relevant indicators related to PICC thrombosis, their importance for judging thrombosis risk is different, so the established decision tree model is inaccurate, affecting the risk assessment.

[0081] Therefore, the thrombosis risk discrimination ability of the current relevant indicator is determined by using the accuracy of the division of thrombosis risk by different relevant indicators and the change situation between normal data and data with thrombosis risk after attribute division.

[0082] When constructing the corresponding PICC-related thrombosis prediction model, the traditional decision tree is used to analyze and select relevant indicators in the following way: the selection of relevant indicators is determined according to the purity represented by entropy. However, the samples obtained are clinical data from hospitals, resulting in most users having a certain risk of PICC thrombosis. Therefore, there are only a small number of people without thrombosis risk. Moreover, the relevant indicators of PICC catheter patients are complex and diverse, and not every relevant indicator is the main basis for judging thrombosis risk. Therefore, the attribute selection method of the traditional decision tree algorithm may have a large error in the construction of the PICC catheter-related thrombosis risk prediction model. As a result, in the study of relevant indicators of thrombosis, relevant indicators with high purity are common phenomena among the population and cannot be used as the main basis for distinguishing and judging thrombosis risk, thus affecting the construction of the decision tree.

[0083] On the other hand, different relevant indicators representing thrombosis characteristics have different degrees of importance. For example, although factors such as age have a certain impact on the risk of PICC thrombosis, the older the age, the weaker the immunity and the greater the thrombosis risk. However, age cannot be used as the main basis for thrombosis risk. It is still necessary to look at direct relevant indicators related to PICC thrombosis such as hemoglobin. Therefore, it is necessary to judge the partitioning ability of relevant indicators according to their ability to distinguish thrombosis risk.

[0084] Therefore, in the embodiments of the present invention, based on the ability of different relevant indicators to distinguish thrombosis risk, the ability of different characteristic values of relevant indicators to distinguish thrombosis risk is determined.

[0085] In some embodiments, based on the ability of the relevant indicators to distinguish thrombosis risk, the ability of different characteristic values of the relevant indicators to distinguish thrombosis risk is determined. That is, the above step 200 can be implemented through Figure 2 the steps shown as follows:

[0086] Step S210, label the characteristic values of the relevant indicators with the corresponding PICC thrombosis risk labels.

[0087] It should be noted that the PICC thrombosis risk label corresponding to the characteristic value of the relevant indicator is the PICC thrombosis risk label of each PICC catheter patient predicted by the deepsurv model. That is, when the PICC thrombosis risk label corresponding to patient c1 is c2, then all the characteristic values of the relevant indicators corresponding to patient c1 are the PICC thrombosis risk label c2 corresponding to this patient c1.

[0088] Step S220, based on the types of PICC thrombosis risk labels corresponding to each characteristic value under the same type of relevant indicators and the proportion of different types of PICC thrombosis risk labels, determine the category characteristics of each characteristic value under the same type of relevant indicators corresponding to different types of PICC thrombosis risk labels.

[0089] In some possible implementation manners, by analyzing the number of types of PICC thrombosis risk labels corresponding to each eigenvalue and the proportion of PICC thrombosis risk labels under the relevant indicators of the same type, the category characteristics of each eigenvalue corresponding to different types of PICC thrombosis risk labels under the relevant indicators of the same type can be quickly calculated. That is, the above step S220 can be implemented through the following steps S221, S222, and S223 (not shown in the figure):

[0090] Step S221: Take any eigenvalue under any relevant indicator as a reference value; obtain the number of types of PICC thrombosis risk labels corresponding to the same reference value as the feature dispersion degree.

[0091] For example, for the relevant indicator: age, age can correspond to multiple eigenvalues. Take any eigenvalue as a reference value; and each eigenvalue has its own corresponding PICC thrombosis risk label, and count the number of types of PICC thrombosis risk labels corresponding to the reference value.

[0092] Step S222: Take any PICC thrombosis risk label as a reference thrombosis risk label; calculate the total number proportion of other PICC thrombosis risk labels except the reference thrombosis risk label corresponding to the same reference value as the feature similarity proportion.

[0093] As an embodiment of the present invention, the proportion of the number of reference thrombosis risk labels corresponding to the same reference value can be calculated first and denoted as η j , and take the difference between the preset complete value and η j as the feature similarity proportion, that is, 1 - η j , and this feature similarity proportion is used to make up for the sample imbalance problem existing in the initially collected data. Since the proportion of data is calculated here, the total number proportion of all corresponding PICC thrombosis risk labels is 1, so the value of the corresponding preset complete value is 1. As another embodiment of the present invention, the total number proportion of other PICC thrombosis risk labels except the reference thrombosis risk label corresponding to the same reference value can also be directly calculated as the feature similarity proportion.

[0094] Step S223: Determine the category characteristics of the reference value when corresponding to the reference thrombosis risk label based on the feature dispersion degree and the feature similarity proportion.

[0095] As an embodiment of the present invention, the product value of the feature dispersion degree and the feature similarity proportion can be used as the category feature corresponding to the reference thrombus risk label. It can also be understood that the feature similarity proportion is used as the weight of the feature dispersion degree, and the feature dispersion degree is weighted to obtain the category feature corresponding to the reference value for the reference thrombus risk label. For example, when the k-th feature value of the i-th related index is used as the reference value and the j-th PICC thrombus risk label is used as the reference thrombus risk label, the category feature can be expressed as: α(v ik ,L j ) = N(v ik ,L j ) × (1 - η j ). Wherein, α(v ik ,L j ) is the category feature corresponding to the j-th PICC thrombus risk label L ik under the k-th feature value v j of the i-th type; η j is the proportion of the number of the j-th PICC thrombus risk labels under the k-th feature value v ik of the i-th related index.

[0096] Step S230: Construct a data feature vector for each feature value under the same type of related index corresponding to all types of PICC thrombus risk labels.

[0097] This data feature vector has a data feature vector corresponding to each feature value under the same type of related index. The number of elements in this data feature vector is n, where n is the total number of feature values under the same type of related index. It should be noted that the same feature values under the same type of related index are classified into the same category. Here, the total number of feature values refers to the total number of feature values with different numerical values. For example, if all the feature values corresponding to the same type of related index are: 5, 2, 2, 5, then the total number of feature values under this type of related index is 2, and the corresponding feature values are 2 and 5 respectively.

[0098] Step S240: Based on the data feature vector, determine the thrombus risk discrimination ability of different feature values under the related index.

[0099] Still on the premise that any feature value under any related index is used as the reference value, based on the data feature vector of the related index corresponding to the reference value and the number of reference values, determine the thrombus risk discrimination ability of the reference value.

[0100] In some possible implementations, by analyzing the data feature vector of the relevant indicators corresponding to the reference value and the number of reference values, the thrombus risk discrimination ability of the reference value can be quickly calculated. That is, the above step S240 can be implemented through the following steps S241, S242, and S243 (not shown in the figure):

[0101] Step S241: For the element values in the data feature vector of the relevant indicators corresponding to the reference value, calculate the ratio of the element value to the number of parameter values respectively to form an adjusted feature vector.

[0102] It can also be understood as calculating the ratio cn of all element values in the data feature vector of the relevant indicators corresponding to the reference value to the number of parameter values respectively, and taking the vector composed of all ratios cn as the adjusted feature vector.

[0103] Step S242: Based on the number of types of PICC thrombus risk labels, correct the adjusted feature vector to obtain a corrected feature vector.

[0104] As an embodiment of the present invention, the reciprocal of the number of types of PICC thrombus risk labels is subtracted from the elements in the adjusted feature vector respectively, and the vector composed of all the obtained differences as elements is used as the corrected feature vector.

[0105] Step S243: Use the norm of the corrected feature vector as the thrombus risk discrimination ability of the reference value.

[0106] When the k-th eigenvalue of the i-th relevant indicator is used as the reference value, the calculation formula for the thrombus risk discrimination ability of this reference value is: where, ‖γ(v ik )‖ is the norm of the corrected feature vector, γ(v ik ) is the corrected feature vector; β(v ik ) is the data feature vector corresponding to the k-th eigenvalue of the i-th relevant indicator; num(v ik ) is the number of the k-th eigenvalue of the i-th relevant indicator; [num(v ik )] [n,1] is a quantity vector with n elements composed of the number of the k-th eigenvalue of the i-th relevant indicator, where the element values in the quantity vector are all num(v ik ); is the adjusted feature vector of the k-th eigenvalue of the i-th relevant indicator; n is the total number of eigenvalues under the i-th type of relevant indicator; is a reverse vector with n elements composed of the reciprocals of the total number of eigenvalues under the i-th type of relevant indicator, where the element values in the reverse vector are all 1 / n.

[0107] In the calculation formula for the thrombus risk discrimination ability, the data feature vector β(v ik ) reflects the data feature corresponding to the k-th eigenvalue of the i-th relevant index. This data feature is determined by the features under each PICC thrombus risk label at the k-th eigenvalue of the i-th relevant index. It reflects the distribution feature of the data at the k-th eigenvalue of the i-th relevant index; Indicates that when the eigenvalues are in a balanced state, the eigenvalues corresponding to each PICC thrombus risk label show a uniform distribution. When the element values in the feature vector are adjusted closer to , it means that the relevant index corresponding to this eigenvalue has a weaker ability to distinguish PICC thrombus risks because it fails to significantly distinguish PICC thrombus risks based on this feature. Therefore, the smaller the difference between the data distribution feature of this relevant index and the uniform distribution, the more similar the data distribution under this node is to the uniform distribution, without showing an obvious identification of a certain PICC thrombus risk, and the weaker the ability to distinguish PICC thrombus risks; ‖γ(v ik )‖ represents the ability of the eigenvalue of this relevant index to distinguish thrombus risks, reflecting the concentration of the PICC thrombus risk labels of the data under the eigenvalue of this relevant index. The larger this value, the greater the difference between the distribution of the PICC thrombus risk labels and the uniform distribution, and the data under the eigenvalue of this relevant index has a significant ability to distinguish certain categories in thrombus risks.

[0108] Step S300, based on the differences between different eigenvalues under the same relevant index, correct the thrombus risk discrimination ability to obtain the division ability of the relevant index.

[0109] Since different relevant indices corresponding to thrombus characteristics have different degrees of importance for thrombus, relying solely on the distribution feature of a single eigenvalue under the same relevant index to determine the ability to distinguish PICC thrombus risks has a problem of low accuracy because it does not compare with the data of other eigenvalues under this relevant index. The eigenvalues under the same relevant index all show a certain difference between the eigenvalue distribution and the average distribution to some extent. However, in fact, the occurrence of a concentrated distribution under a certain eigenvalue under the same relevant index is a normal phenomenon and is an attribute with a relatively small correlation with PICC thrombus risks and cannot be used as the main basis for judging thrombus risks.

[0110] Each eigenvalue of the same relevant index has a label distribution feature, that is, the corrected feature vector γ(v ik) The corrected eigenvector is an n-dimensional vector, and the importance of relevant indicators is determined according to the similarity between vectors. If the label distributions among different eigenvalues under the same relevant indicator are similar, it is considered a normal distribution feature unrelated to the PICC thrombus risk, and thus this relevant indicator cannot be used as the main basis for judging thrombus risk. Therefore, the ability to distinguish thrombus risk is corrected by the differences in data points under different eigenvalues of the same relevant indicator.

[0111] In some embodiments, the ability to distinguish thrombus risk can be corrected according to the differences between different eigenvalues of the relevant indicator to obtain the partitioning ability of the relevant indicator. Among them, the different eigenvalues of the relevant indicator can also reflect the similarity between different eigenvalues, that is, the above step S300 can be achieved through Figure 3 the steps shown as follows:

[0112] Step S310: Take any relevant indicator as the reference relevant indicator, and determine the importance of the partitioning ability based on the distribution of the thrombus risk discrimination ability corresponding to different eigenvalues under the same reference relevant indicator.

[0113] In the field of mathematics, variance is usually used to reflect the degree of dispersion or fluctuation of a set of data. The larger the variance, the greater the fluctuation of this set of data and the more unstable the data; conversely, the smaller the variance, the smaller the fluctuation of this set of data and the more stable the data. Specifically, variance can help understand how data points are distributed around their mean value, that is, the degree of concentration and dispersion, fluctuation and stability of the data. A smaller variance means that the data points are more concentrated, that is, most data are close to the mean value, while a larger variance means that the data points are more dispersed, that is, the data fluctuates greatly above and below the mean value. Therefore, as an embodiment of the present invention, the distribution of the thrombus risk discrimination ability corresponding to different eigenvalues under the same reference relevant indicator is reflected by variance, that is, the variance of the thrombus risk discrimination ability corresponding to different eigenvalues under the same reference relevant indicator is used as the importance of the partitioning ability.

[0114] Step S320: Screen out the representative value of the thrombus risk discrimination ability from the thrombus risk discrimination abilities corresponding to different eigenvalues of the reference relevant indicator.

[0115] As an embodiment of the present invention, the maximum value among the thrombus risk discrimination capabilities corresponding to different characteristic values of the reference related index is selected as the representative value of the thrombus risk discrimination capability of the reference related index. As another embodiment of the present invention, the mean value of the thrombus risk discrimination capabilities corresponding to different characteristic values of the reference related index can also be calculated as the representative value of the thrombus risk discrimination capability of the reference related index. As another embodiment of the present invention, the median of the thrombus risk discrimination capabilities corresponding to different characteristic values of the reference related index can also be used as the representative value of the thrombus risk discrimination capability of the reference related index.

[0116] Step S330: Based on the importance degree of the reference related index, correct the representative value of the thrombus risk discrimination capability corresponding to the reference related index to obtain the discrimination capability of the reference related index.

[0117] In the embodiment of the present invention, the product of the importance degree of calculating the discrimination capability of the reference related index and the representative value of the thrombus risk discrimination capability is used as the discrimination capability of the reference related index, so as to achieve the purpose of correcting the representative value of the thrombus risk discrimination capability corresponding to the reference related index through the importance degree of the reference related index and obtaining the discrimination capability of the reference related index.

[0118] In the above steps S310 to S330, based on the differences of different characteristic values under the same related index, correct the thrombus risk discrimination capability to obtain the discrimination capability of the related index. The acquisition method of the discrimination capability of the related index can also be reflected by the calculation formula of the discrimination capability of the related index.

[0119] The magnitude of the characteristic value reflects the similarity of the characteristic distribution. The greater the variance of the thrombus risk discrimination capabilities corresponding to these characteristic values, the smaller the similarity of the characteristic distribution, and thus the greater the importance degree corresponding to this attribute. The greater the importance degree of the discrimination capability, the smaller the similarity, and the greater the importance of the related index. Taking the i-th related index as the reference related index as an example, the calculation formula for the discrimination capability of the related index is constructed as: W(V i ) = p(V i ) × mt k ‖γ(v ik )‖; where W(V i ) is the discrimination capability of the i-th related index; p(V i ) is the importance degree of the discrimination capability of the i-th related index; mt k ‖γ(v ik )‖ is the representative value of the thrombus risk discrimination capability of the i-th related index. The greater the importance degree of the discrimination capability of the related index, the greater the importance degree of this related index in distinguishing the PICC thrombus risk; mt k ‖γ(v ik) ‖As a representative value, as the partitioning ability of the relevant index, the importance level p(V i ) As a correction factor for the partitioning ability, W(V i ) Represents the corrected partitioning ability of the i-th relevant index V i .

[0120] So far, the partitioning ability of this attribute has been obtained through the difference in thrombus risk under the same attribute representing thrombus characteristics.

[0121] Step S400, combining the thrombus risk discrimination ability and the partitioning ability, and using the similarity between the label distribution characteristics before and after partitioning by the relevant index, determine the partitioning cost of the relevant index.

[0122] However, the following problems will occur in the construction of the decision tree for thrombus risk: The obtained relevant index has a strong partitioning ability because the thrombus risk has been well partitioned before the partitioning by this relevant index, resulting in a good partitioning effect in subsequent partitioning. But in fact, it has little correlation with the partitioning ability of this relevant index. Therefore, the obtained partitioning ability is inaccurate. It is necessary to use the similarity in data distribution before and after partitioning to judge the partitioning cost of the relevant index.

[0123] If there is no obvious improvement in the judgment of thrombus risk before and after partitioning, it is considered that this relevant index has not played a good role in the judgment of thrombus risk, and the relevant index has no obvious impact on the decision tree, resulting in the need for a deeper decision tree to achieve the discrimination of PICC thrombus risk. For the decision tree, the importance level of this relevant index is relatively small at this position, and the partitioning cost of the relevant index is also smaller. In the embodiments of the present invention, as much as possible, the selected relevant index is made to have an obvious improvement effect on the judgment of thrombus risk.

[0124] In some embodiments, by using the similarity between the label distribution characteristics before and after partitioning by the relevant index, the partitioning cost of the relevant index can be analyzed more accurately. That is, the above step S400 can be implemented through Figure 4 the steps shown:

[0125] Step S410, compare the similarity between the thrombus risk discrimination ability of the relevant index before partitioning and the thrombus risk discrimination ability of different eigenvalue under the relevant index after partitioning, and determine the partitioning improvement value.

[0126] First, calculate the thrombus risk discrimination ability of relevant indicators as the discrimination ability to be partitioned. For each eigenvalue, calculate the difference between the thrombus risk discrimination ability of each eigenvalue under the relevant indicator and the discrimination ability to be partitioned as the single partition improvement value of each eigenvalue under the relevant indicator; select the representative value from the single partition improvement values of each eigenvalue under the relevant indicator as the partition improvement value. Compare the change in the discrimination ability before and after the partition of the relevant indicator, that is, compare the change in the discrimination ability to be partitioned and the discrimination ability. The difference between the two reflects the magnitude of the improvement in the partition ability. The greater the improvement, the greater the partition cost of the relevant indicator. Select the maximum difference between the thrombus risk discrimination ability of the relevant indicator and the thrombus risk discrimination ability of the eigenvalue as the improvement in the final partition ability of the relevant indicator, denoted as the partition improvement value.

[0127] Therefore, in an embodiment of the present invention, select the maximum value in the single partition improvement values as the partition improvement value corresponding to the relevant indicator.

[0128] It should be noted that the method for obtaining the thrombus risk discrimination ability of the relevant indicator is the same as the method for obtaining the thrombus risk discrimination ability of each eigenvalue under the relevant indicator. More specifically: label the corresponding PICC thrombus risk label for the relevant indicator; based on the type of the PICC thrombus risk label corresponding to the same type of relevant indicator and the proportion of different types of PICC thrombus risk labels, determine the to-be-partitioned category features of the same type of relevant indicator for different types of PICC thrombus risk labels; construct the to-be-partitioned data feature vector of each eigenvalue from the to-be-partitioned category features of the same type of relevant indicator for different types of PICC thrombus risk labels; based on the to-be-partitioned data feature vector, determine the thrombus risk discrimination ability of the relevant indicator.

[0129] Among them, based on the type of the PICC thrombus risk label corresponding to the same type of relevant indicator and the proportion of different types of PICC thrombus risk labels, determine the to-be-partitioned category features of the same type of relevant indicator for different types of PICC thrombus risk labels. Specifically: obtain the number of types of the PICC thrombus risk label corresponding to the same type of relevant indicator as the to-be-partitioned feature dispersion; use any PICC thrombus risk label as the reference thrombus risk label; calculate the total number proportion of other PICC thrombus risk labels except the reference thrombus risk label corresponding to the same type of relevant indicator as the to-be-partitioned feature similarity proportion; based on the to-be-partitioned feature dispersion and the to-be-partitioned feature similarity proportion, determine the to-be-partitioned category features of the same type of relevant indicator when corresponding to the reference thrombus risk label.

[0130] Among them, based on the feature vectors of the data to be partitioned, the thrombus risk discrimination ability of relevant indicators is determined. Specifically: for the element values within the feature vectors of the data to be partitioned corresponding to the same type of relevant indicators, the ratios of the element values to the quantity of the same type of relevant indicators are calculated respectively to form the adjusted feature vectors to be partitioned; based on the number of types of PICC thrombus risk labels, the adjusted feature vectors to be partitioned are corrected to obtain the corrected feature vectors to be partitioned; the norm of the corrected feature vectors to be partitioned is used as the thrombus risk discrimination ability of the reference value. It should be noted that the method for obtaining the thrombus risk discrimination ability of relevant indicators is the same as that of the thrombus risk discrimination ability of the eigenvalue of relevant indicators. The only difference is that when calculating the thrombus risk discrimination ability of the eigenvalue of relevant indicators, the PICC thrombus risk labels corresponding to the eigenvalues are analyzed respectively, while when calculating the thrombus risk discrimination ability of relevant indicators, the PICC thrombus risk labels corresponding to the relevant indicators are analyzed and calculated. Therefore, it will not be elaborated here.

[0131] Step S420: Combine the partition improvement value and the partition ability to determine the partition cost of the relevant indicator.

[0132] Among them, both the partition improvement value and the partition ability are positively correlated with the partition cost of the relevant indicator. As an embodiment of the present invention, the square root of the sum of the squares of the partition improvement value and the partition ability is used as the partition cost of the relevant indicator.

[0133] In the above steps S410 to S420, whether the partition of the relevant indicator has an obvious effect on the judgment of thrombus risk is judged by the similarity between the PICC thrombus risk labels before and after the partition of the relevant indicator, so as to determine the partition cost t(V i ). Taking the i-th type of relevant indicator as the reference relevant indicator, the calculation formula for constructing the partition cost cost(V i ) of the reference relevant indicator is as follows: Among them, t(V i ) is the partition improvement value of the i-th type of relevant indicator; W(V i ) is the partition ability of the i-th type of relevant indicator.

[0134] The partition cost is obtained by analyzing the changes before and after the partition of the relevant indicator. So far, the partition cost is obtained through the analysis of the relevant indicator, and further the selection of the decision tree attribute is realized according to the partition cost.

[0135] Step S500: Construct a corresponding PICC-related thrombus prediction model according to the partition costs of different types of relevant indicators.

[0136] After obtaining the division cost through the changes before and after the body posture division according to steps S100 to S400, multiple decision trees are constructed based on the division costs of different types of relevant indicators of PICC catheter patients. More specifically: The relevant indicator with the largest division cost is selected for the construction of the decision tree, which can also be understood as using the relevant indicator with the largest division cost as the division criterion of the decision tree.

[0137] Based on the results of different random samplings, the selected N samples are used as the samples of the root node of the decision tree to train a decision tree. Through different randomly sampled training sets, multiple decision trees can be constructed, and then a random forest is obtained as the corresponding PICC-related thrombosis prediction model, completing the construction of the PICC catheter-related thrombosis risk prediction model. In the embodiment of the present invention, each relevant indicator is equivalent to a parent node of the decision tree, and the eigenvalue under the relevant indicator is equivalent to the child node further divided by the parent node of the decision tree.

[0138] By inputting the relevant indicators of the thrombus risk, the PICC thrombus risk level of the relevant indicator can be obtained, and medical staff can give different intervention measures to the people coming for examination according to the obtained PICC thrombus risk level. For example, relevant preventive suggestions can be given to normal healthy people; for low-risk patients, only routine care and monitoring may be required; for medium-risk patients, more frequent monitoring and some preventive measures may be needed; and for high-risk patients, more aggressive preventive and treatment measures may be required.

[0139] As an embodiment of the present invention, after obtaining the PICC-related thrombosis prediction model, the C-index and Brier score can also be used to evaluate the models constructed by different algorithms to obtain the best prediction model.

[0140] The embodiment of the present application provides a system for constructing a PICC catheter-related thrombosis risk prediction model, as Figure 5 shown. The system 600 includes:

[0141] A collection module 610 that obtains relevant indicators of PICC catheter patients;

[0142] An ability determination module 620 that determines the thrombus risk discrimination ability of different eigenvalues under the relevant indicator based on the discrimination ability of the relevant indicator for the thrombus risk;

[0143] A correction module 630 that corrects the thrombus risk discrimination ability based on the differences of different eigenvalues under the same relevant indicator to obtain the division ability of the relevant indicator;

[0144] A cost determination module 640, in combination with the thrombus risk differentiation ability and the partitioning ability, determines the partitioning cost of relevant indicators by using the similarity between the label distribution characteristics before and after partitioning with relevant indicators;

[0145] A model construction module 650 constructs a corresponding PICC-related thrombus prediction model according to the partitioning costs of different types of relevant indicators.

[0146] As an embodiment of the present invention, the probability of thrombus occurrence in each PICC catheter patient can be calculated according to the weight values of various factors through the implanted PICC catheter-related thrombus prediction model.

[0147] As an embodiment of the present invention, a result output module 660 can also be added. For the case of each PICC catheter patient, the risk level is determined according to the constructed PICC catheter-related thrombus prediction model for result output; for the high-incidence period of PICC catheter-related thrombus, key reminders are made.

[0148] It should be noted that: for the system provided in the above embodiment, only the above-mentioned division of each functional module is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the construction system of a PICC catheter-related thrombus risk prediction model provided in the above embodiment and the construction method embodiment of a PICC catheter-related thrombus risk prediction model belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0149] Figure 6 It is a schematic structural diagram of a computer device provided by an embodiment of the present application. Exemplarily, as Figure 6 shown, the computer device 700 includes: a memory 701, a processor 702, and a computer program 703 stored in the memory 701 and running on the processor 702. When the processor 702 executes the computer program 703, the computer device can execute any one of the above-described PICC catheter-related thrombus risk prediction model construction methods.

[0150] In addition, an embodiment of the present application also protects a device, which may include a memory and a processor. The memory stores executable program code, and the processor is used to call and execute the executable program code to execute a PICC catheter-related thrombus risk prediction model construction method provided by an embodiment of the present application.

[0151] In this embodiment, the device can be divided into functional modules according to the above method examples. For example, it can correspond to each functional module, or two or more functions can be integrated into one processing module. The above integrated module can be implemented in the form of hardware. It should be noted that the division of modules in this embodiment is illustrative, only a logical function division, and there can be other division methods in actual implementation.

[0152] In the case of dividing each module according to each function, the device may further include a signal uploading module, a determination module, an adjustment module, etc. It should be noted that all relevant contents of each step involved in the above method embodiment can be cited in the function description of the corresponding functional module, and will not be repeated here.

[0153] It should be understood that the device provided in this embodiment is used to execute the above method for constructing a PICC catheter-related thrombus risk prediction model, so the same effect as the above implementation method can be achieved.

[0154] In the case of adopting an integrated unit, the device may include a processing module and a storage module. Among them, when the device is applied to a device, the processing module can be used to control and manage the actions of the device. The storage module can be used to support the device to execute mutual program codes, etc.

[0155] Among them, the processing module can be a processor or a controller, which can implement or execute various exemplary logical blocks, modules, and circuits described in combination with the disclosure of this application. The processor can also be a combination that realizes computing functions, such as including a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc. The storage module can be a memory.

[0156] In addition, the device provided in the embodiment of the present application can specifically be a chip, a component, or a module. The chip may include a connected processor and a memory; among them, the memory is used to store instructions, and when the processor calls and executes the instructions, the chip can execute the method for constructing a PICC catheter-related thrombus risk prediction model provided in the above embodiment.

[0157] This embodiment also provides a computer-readable storage medium, in which computer program code is stored. When the computer program code runs on a computer, it causes the computer to execute the above relevant method steps to implement the method for constructing a PICC catheter-related thrombus risk prediction model provided in the above embodiment.

[0158] This embodiment also provides a computer program product. When the computer program product runs on a computer, it causes the computer to execute the above-related steps to implement the method for constructing a risk prediction model for PICC catheter-related thrombosis provided in the above embodiment.

Claims

1. A method for constructing a risk prediction model for PICC catheter-related thrombosis, characterized in that, The method includes the following steps: Obtain relevant indicators of patients with PICC catheters; Based on the ability of the relevant indicators to distinguish thrombus risks, determine the thrombus risk discrimination ability of different eigenvalue under the relevant indicators; Based on the differences of different eigenvalues under the same relevant indicator, correct the thrombus risk discrimination ability to obtain the classification ability of the relevant indicator; Combining the thrombus risk discrimination ability and the classification ability, and using the similarity between the label distribution characteristics before and after the division of the relevant indicators, determine the division cost of the relevant indicators; According to the division costs of different types of relevant indicators, construct a corresponding PICC-related thrombus prediction model.

2. The method for constructing a risk prediction model for PICC catheter-related thrombosis according to claim 1, wherein The step of determining the thrombus risk discrimination ability of different eigenvalues under the relevant indicators based on the ability of the relevant indicators to distinguish thrombus risks includes: Mark the eigenvalues of the relevant indicators with corresponding PICC thrombus risk labels; Based on the types of PICC thrombus risk labels corresponding to each eigenvalue under the same type of relevant indicators and the proportion of different types of PICC thrombus risk labels, determine the category characteristics of each eigenvalue under the same type of relevant indicators in the corresponding different types of PICC thrombus risk labels; Construct a data feature vector for each eigenvalue from the category characteristics of each eigenvalue under the same type of relevant indicators corresponding to all types of PICC thrombus risk labels; Based on the data feature vector, determine the thrombus risk discrimination ability of different eigenvalues under the relevant indicators.

3. The method for constructing a PICC catheter-related thrombus risk prediction model according to claim 2, wherein The step of determining the category characteristics of each eigenvalue under the same type of relevant indicators in the corresponding different types of PICC thrombus risk labels based on the types of PICC thrombus risk labels corresponding to each eigenvalue under the same type of relevant indicators and the proportion of different types of PICC thrombus risk labels includes: Take any eigenvalue under any relevant indicator as a reference value; obtain the number of types of PICC thrombus risk labels corresponding to the same reference value as the feature dispersion; Take any PICC thrombus risk label as a reference thrombus risk label; calculate the total number proportion of other PICC thrombus risk labels except the reference thrombus risk label corresponding to the same reference value as the feature similarity proportion; Based on the feature dispersion and the feature similarity proportion, determine the category characteristics of the reference value when corresponding to the reference thrombus risk label.

4. The method for constructing a risk prediction model for PICC catheter-related thrombosis according to claim 2, wherein The step of determining the thrombus risk discrimination ability of different eigenvalues under the relevant indicators based on the data feature vector includes: Take any eigenvalue under any relevant indicator as a reference value; Based on the data feature vector of the relevant indicator corresponding to the reference value and the number of the reference value, determine the thrombus risk discrimination ability of the reference value.

5. The method for constructing a PICC catheter-related thrombosis risk prediction model according to claim 4, wherein The step of determining the thrombus risk discrimination ability of the reference value based on the data feature vector of the relevant indicator corresponding to the reference value and the number of the reference value includes: For the element values in the data feature vector of the relevant indicator corresponding to the reference value, calculate the ratio of the element value to the number of parameter values respectively to form an adjusted feature vector; Based on the number of types of PICC thrombus risk labels, correct the adjusted feature vector to obtain a corrected feature vector; Take the norm of the corrected feature vector as the thrombus risk discrimination ability of the reference value.

6. The method for constructing a PICC catheter-related thrombosis risk prediction model according to claim 1, wherein Based on the differences in different eigenvalue under the same relevant index, correcting the ability to distinguish thrombosis risk to obtain the partitioning ability of the relevant index, including: Taking any relevant index as the reference relevant index, and determining the importance degree of the partitioning ability based on the distribution of the thrombosis risk discrimination ability corresponding to different eigenvalues under the same reference relevant index; Selecting the representative value of the thrombosis risk discrimination ability from the thrombosis risk discrimination abilities corresponding to different eigenvalues of the reference relevant index; Based on the importance degree of the reference relevant index, correcting the representative value of the thrombosis risk discrimination ability corresponding to the reference relevant index to obtain the partitioning ability of the reference relevant index.

7. The method for constructing a PICC catheter-related thrombus risk prediction model according to claim 6, characterized in that, The selecting the representative value of the thrombosis risk discrimination ability from the thrombosis risk discrimination abilities corresponding to different eigenvalues of the reference relevant index includes: Selecting the maximum value among the thrombosis risk discrimination abilities corresponding to different eigenvalues of the reference relevant index as the representative value of the thrombosis risk discrimination ability of the reference relevant index.

8. The method for constructing a PICC catheter-related thrombosis risk prediction model according to claim 1, wherein Using the similarity between the label distribution characteristics before and after the partitioning of the relevant index to determine the partitioning cost of the relevant index, including: Comparing the similarity between the thrombosis risk discrimination ability of the relevant index before partitioning and the thrombosis risk discrimination abilities of different eigenvalues under the relevant index after partitioning to determine the partitioning improvement value; Combining the partitioning improvement value and the partitioning ability to determine the partitioning cost of the relevant index.

9. The method for constructing a PICC catheter-related thrombosis risk prediction model according to claim 1, wherein, According to the partitioning costs of different types of relevant indexes, constructing the corresponding PICC-related thrombosis prediction model, including: Based on the partitioning costs of different types of relevant indexes of PICC catheter patients, constructing multiple decision trees to obtain a random forest model as the corresponding PICC-related thrombosis prediction model.

10. A construction system for a risk prediction model of PICC catheter-related thrombosis, characterized in that, The system includes: An acquisition module, which acquires the relevant indexes of PICC catheter patients; An ability determination module, which determines the thrombosis risk discrimination abilities of different eigenvalues under the relevant index based on the ability of the relevant index to distinguish thrombosis risk; A correction module, which corrects the thrombosis risk discrimination ability based on the differences in different eigenvalues under the same relevant index to obtain the partitioning ability of the relevant index; A cost determination module, which combines the thrombosis risk discrimination ability and the partitioning ability, and uses the similarity between the label distribution characteristics before and after the partitioning of the relevant index to determine the partitioning cost of the relevant index; A model construction module, which constructs the corresponding PICC-related thrombosis prediction model according to the partitioning costs of different types of relevant indexes.

Images