Orthopedic trauma patient VTE risk prediction model establishment method and device

Abstract

The invention relates to a method and a device for establishing a VTE risk prediction model of an orthopedic trauma patient. The method comprises the following steps: acquiring orthopedic patient data; carrying out modeling based on a plurality of preset machine learning methods, carrying out training, and carrying out performance evaluation; determining a machine learning method corresponding to the initial model with the best selection performance as a preliminary target learning method; based on a plurality of feature screening algorithms, screening features related to the VTE risk in the orthopedic patient data to obtain a plurality of feature sets; summarizing the multiple groups of feature sets to obtain a total feature set; using a feature incremental modeling test to select the number of incorporated model features; determining a target feature set; respectively adopting different machine learning methods to construct a machine learning model, and carrying out training, tuning and performance evaluation; and selecting the orthopaedic trauma patient VTE risk prediction model with the best performance as a target orthopaedic trauma patient VTE risk prediction model.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for establishing a VTE risk prediction model for orthopedic trauma patients. Background Art

[0002] Venous thromboembolism (VTE) is a serious complication in orthopedic trauma patients, with an incidence rate of 40% to 60%. The occurrence of VTE not only increases the medical risk of patients, but may also lead to long-term health problems and economic burdens. Therefore, it is an important clinical need to conduct effective VTE risk assessment for orthopedic trauma patients so that preventive measures can be taken in a timely manner.

[0003] Based on this, a method for establishing a VTE risk prediction model for orthopedic trauma patients is needed to establish a VTE risk prediction model for orthopedic trauma patients and predict the VTE risk in orthopedic trauma patients. Summary of the invention

[0004] In view of this, a method and device for establishing a VTE risk prediction model for orthopedic trauma patients are provided.

[0005] The present invention adopts the following technical solution: An embodiment of the present invention provides a method for establishing a VTE risk prediction model for orthopedic trauma patients, the method comprising: Obtain orthopedic patient data; Modeling is performed based on each of the preset multiple machine learning methods to obtain multiple initial prediction models; For each initial model, the initial prediction model is trained based on the orthopedic patient data, and a performance evaluation is performed; Determine the machine learning method corresponding to the initial model with the best performance as the preliminary target learning method; Based on multiple feature screening algorithms, features related to VTE risk in orthopedic patient data were screened to obtain multiple sets of feature sets; Summarizing the multiple feature sets to obtain a total feature set, and sorting each feature in the total feature set from most to least according to the number of times it appears in the multiple feature sets; Determine a target feature set; wherein determining the target feature set includes: step A1: set n=n+1; wherein the initial value of n is 0; step A2: construct a risk prediction model based on a preliminary target learning method of the top n features, and train and evaluate the performance of the model; repeatedly iteratively execute steps A1 and A2 until the error of the index obtained by the performance evaluation in the most recent two iterations is lower than a preset value; determine the top n-1 features as the target feature set; Select the machine learning methods corresponding to the top m initial models with the best performance to form a target machine learning method set; Based on each machine learning method in the target machine learning method set and the features corresponding to the target feature set, a machine learning model is constructed to obtain multiple VTE risk prediction models for orthopedic trauma patients; Train, tune, and evaluate the performance of multiple VTE risk prediction models for orthopedic trauma patients; The VTE risk prediction model for orthopedic trauma patients with the best performance was selected as the target VTE risk prediction model for orthopedic trauma patients.

[0006] In some embodiments, the Shapley additive interpretation method is used to perform individual interpretation and variable-dependent interpretation of the target orthopedic trauma patient VTE risk prediction model, and the SHAP value of each feature is calculated.

[0007] In some embodiments, the method further includes: drawing a bar chart, a honeycomb chart and / or a waterfall chart based on the SHAP value to reflect the relative importance of each feature and the quantitative relationship between the feature and the probability of VTE occurrence.

[0008] In some embodiments, the orthopedic patient data include: age, gender, VTE history, comorbidities, time from injury to hospitalization, upper limb fractures, lower limb fractures, high-energy injuries, multiple injuries, open fractures, comminuted fractures, combined vascular and nerve injuries, operation duration, operation type, whether general anesthesia was used, amount of red blood cell suspension infusion, number of days in bed after surgery, whether drug prevention was used, and ICU admission information.

[0009] In some embodiments, performing performance evaluation includes: performing performance evaluation based on accuracy, area under the curve, recall, precision, F1 score, Kappa value and / or MCC.

[0010] In some embodiments, the multiple feature screening algorithms include: Pearson Correlation Coefficient algorithm (PearsonCorr, Pearson Correlation Coefficient), Spearman's Rank Correlation Coefficient algorithm (SpearmanCorr, Spearman's Rank Correlation Coefficient), double centering algorithm (DC, Double Centering), mutual information algorithm (MI, Mutual Information), maximum information coefficient algorithm (MIC, Maximum Information Coefficient), minimum absolute value shrinkage and selection operator algorithm (lasso, Least Absolute Shrinkage and Selection Operator), ridge regression algorithm (ridge, RidgeRegression), multiple discriminant analysis algorithm (MDA, Multiple Discriminant Analysis).

[0011] In some embodiments, the preset multiple machine learning methods include: Gradient Boosting Classifier (GBC), Ridge Regression (Ridge), Light Gradient Boosting Machine (LightGBM), Random Forest (RF), Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), Logistic Regression (LR), Extreme Random Trees (ET), Naive Bayes (NB), Quadratic Discriminant Analysis (QDA), Adaptive Boosting (AdaBoost), Decision Tree (DT), Dummy Classifier (Dummy).

[0012] The present application also provides a device for establishing a VTE risk prediction model for orthopedic trauma patients, comprising: The acquisition module is used to obtain orthopedic patient data; A feature determination module is used to respectively perform modeling based on each of a plurality of preset machine learning methods to obtain a plurality of initial prediction models; for each initial model, the initial prediction model is trained based on the orthopedic patient data and a performance evaluation is performed; the machine learning method corresponding to the initial model with the best performance is determined as the preliminary target learning method; based on a plurality of feature screening algorithms, features related to VTE risk in the orthopedic patient data are screened to obtain a plurality of feature sets; the plurality of feature sets are summarized to obtain a total feature set, and each feature in the total feature set is sorted from most to least according to the number of occurrences in the plurality of feature sets; a target feature set is determined; wherein determining the target feature set comprises: step A1: setting n=n+1; wherein the initial value of n is 0; step A2: constructing a risk prediction model based on the preliminary target learning method of the top n features, and training and evaluating the performance of the model; iteratively executing steps A1 and A2 until the error of the index obtained by the performance evaluation in the two most recent iterations is lower than the preset value; determining the top n-1 features as the target feature set; The model determination module is used to select the machine learning methods corresponding to the top m initial models with the best performance to form a target machine learning method set; construct machine learning models based on each machine learning method in the target machine learning method set and the features corresponding to the target feature set to obtain multiple VTE risk prediction models for orthopedic trauma patients; train, tune and evaluate the performance of multiple VTE risk prediction models for orthopedic trauma patients; and select the VTE risk prediction model for orthopedic trauma patients with the best performance as the target VTE risk prediction model for orthopedic trauma patients.

[0013] The present application also provides an electronic device, comprising: A processor, and a memory connected to the processor; The memory is used to store a computer program, and the computer program is at least used to execute the above-mentioned method for establishing a VTE risk prediction model for orthopedic trauma patients; The processor is used to call and execute the computer program in the memory.

[0014] The present application also provides a storage medium, which stores a computer program. When the computer program is executed by a processor, each step of the above-mentioned method for establishing a VTE risk prediction model for orthopedic trauma patients is implemented.

[0015] The present invention adopts the above technical scheme to obtain orthopedic patient data; modeling is performed based on each of the preset multiple machine learning methods to obtain multiple initial prediction models; for each initial model, the initial prediction model is trained based on the orthopedic patient data, and performance evaluation is performed; the machine learning method corresponding to the initial model with the best performance is determined as the preliminary target learning method; based on multiple feature screening algorithms, features related to VTE risk in orthopedic patient data are screened to obtain multiple groups of feature sets; the multiple groups of feature sets are summarized to obtain a total feature set, and each feature in the total feature set is sorted from most to least according to the number of occurrences in the multiple groups of feature sets; a target feature set is determined; wherein determining the target feature set includes: step A1: let n=n+1; wherein the initial value of n is 0; step A 2: Based on the initial target learning method of the top n features, a risk prediction model is constructed, and the model is trained and performance evaluated; Step A1 and Step A2 are repeatedly executed until the error of the index obtained by the performance evaluation in the last two iterations is lower than the preset value; The top n-1 features are determined as the target feature set; The machine learning methods corresponding to the top m initial models with the best performance are selected to form the target machine learning method set; Based on each machine learning method in the target machine learning method set and the features corresponding to the target feature set, a machine learning model is constructed to obtain multiple VTE risk prediction models for orthopedic trauma patients; Multiple VTE risk prediction models for orthopedic trauma patients are trained, tuned and performance evaluated; The VTE risk prediction model for orthopedic trauma patients with the best performance is selected as the target VTE risk prediction model for orthopedic trauma patients. In this way, by collecting comprehensive orthopedic patient data, it is ensured that the model can analyze VTE risk factors from multiple angles and improve the accuracy of prediction. Using actual clinical data, the model is closer to the actual medical situation and enhances the practicality and reliability of the model. Using multiple machine learning methods can explore the fitting ability of different models to data and increase the possibility of finding the best model. By comparing the performance of different models, the most suitable model for predicting VTE risk in orthopedic patients can be selected. Through training and evaluation, model parameters can be adjusted and optimized to improve the predictive performance of the model. Based on objective performance indicators such as accuracy, AUC, etc., the best model can be scientifically selected to avoid subjective judgment bias. Quickly locate the model with the best performance to reduce unnecessary calculations and resource waste. Provide a clear direction for subsequent feature screening and model building. Use multiple feature screening algorithms to screen relevant features. Reduce model complexity, avoid overfitting, and improve the generalization ability of the model. By sorting the number of occurrences, the features that are most likely to affect the risk of VTE can be identified. Dynamically determine the optimal number of features through iteration to balance the complexity and predictive performance of the model. By monitoring the errors of performance evaluation indicators, ensure that the performance of the model is substantially improved when adding features.Avoid the risk of relying on a single model and disperse the prediction risk by combining a diverse set of model methods. By building multiple models, you can compare the performance of different models on the target feature set and select the best model. Ensure that the model has high prediction accuracy and reliability in practical applications. Select the optimal model after rigorous evaluation and tuning to ensure the performance of the model in practical applications. Provide a validated and reliable tool for VTE risk prediction in target orthopedic trauma patients, which will help clinical decision-making and implementation of preventive measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 It is a flow chart of a method for establishing a VTE risk prediction model for orthopedic trauma patients provided by an embodiment of the present invention; Figure 2 It is a flow chart of a method for establishing a VTE risk prediction model for orthopedic trauma patients provided by an embodiment of the present invention; Figure 3 It is a structural schematic diagram of a device for establishing a VTE risk prediction model for orthopedic trauma patients provided by an embodiment of the present invention; Figure 4 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0019] First, the application scenario of the embodiment of the present invention is explained. Orthopedic trauma patients are a high-risk group for venous thromboembolism (VTE). Previous studies have shown that their thrombosis incidence rate can reach 40% to 60%. Efficient and accurate thrombosis risk assessment is an important prerequisite for the effective prevention of VTE in orthopedic trauma patients. VTE risk assessment is conducive to subsequent decision-making on whether to use VTE prevention measures and what kind of prevention measures to use. It is the focus of clinical medical staff, but there are still many problems in the accurate assessment of VTE in orthopedic trauma patients.

[0020] Risk assessment is an important research issue in the medical field. The core is to evaluate the risk factors related to the occurrence of a certain outcome event. Through scoring, fitting formula calculations and other means, the probability of a patient being "ill" or "not ill" is obtained, or a judgment of "high risk" or "non-high risk" is made, thereby guiding the implementation of clinical intervention measures. One of the key technologies is how to achieve accurate assessment by selecting, scoring and fitting a certain algorithm for risk factors.

[0021] Historically, VTE risk assessment for orthopedic trauma patients has gone through a process from general to specific, from group to individual, and from extensive to precise. At first, medical staff used a group risk assessment method to assess the VTE risk of orthopedic trauma patients. Once the patient was confirmed to have a certain fracture type, or to have certain risk factors such as hip fracture, joint replacement, and surgery time > 90 minutes, the patient was automatically classified as a high-risk group for VTE and given intervention; since then, a scale-based individual risk assessment method has been further formed. According to the commonly used VTE risk assessment scales such as Caprini risk assessment and Rogers risk assessment, the number and weight of the patient's risk factors are accumulated and scored, and then the scores are stratified, usually divided into low risk, medium risk and high risk, and then intervention is carried out according to different risk levels. The above method can give a simple and intuitive risk level, but there may be problems such as insufficient consideration of risk factors, systematic overestimation or underestimation of individual risks, and inability to update dynamically.

[0022] In view of this, the present application provides a comprehensive risk prediction model establishment and interpretation method, which can better solve the problems brought by the aforementioned methods, and use accurate risk factors to accurately evaluate and interpret the VTE risk of orthopedic trauma patients. First, an orthopedic trauma patient database is established to collect basic patient information and VTE risk factors; in different machine learning models, the model construction method and evaluation indicators for risk factor screening (feature screening) are selected. The feature incremental modeling test is used to select the number of model features to be included. After obtaining the corresponding risk factors, the hyperparameter optimization and bootstrap validation of the modeling model are further performed. The indicators for verifying the performance of the model mainly include AUC, F1, Recall and Precision. The Shapley additive interpretation method is used to calculate the SHAP value of each risk factor. According to the SHAP value, a bar chart, a honeycomb chart and a waterfall chart are drawn to reflect the relative importance of the risk factors and the quantitative relationship between them and the probability of VTE.

[0023] Example Figure 1 The present invention provides a flowchart of a method for establishing a VTE risk prediction model for orthopedic trauma patients, which can be executed by the electronic device provided by the present invention. Figure 1, the method may specifically include the following steps: S101. Obtain orthopedic patient data; Specifically, the medical data of orthopedic patients are extracted from the hospital's electronic health record (EHR) system. The collected data includes the patient's basic information, medical history, treatment records, laboratory test results, etc. Ensure the integrity and accuracy of the data, and handle missing or abnormal data.

[0024] S102, performing modeling based on each of the preset multiple machine learning methods to obtain multiple initial prediction models; In this step, we use a variety of preset machine learning methods to build multiple initial prediction models. These methods may include but are not limited to decision trees, random forests, support vector machines, gradient boosting machines, neural networks, etc. Each method aims to learn patterns from the data in order to predict VTE risk.

[0025] S103, for each initial model, training the initial prediction model based on the orthopedic patient data, and performing performance evaluation; For each initial model, we trained the model using orthopedic patient data. After training, the performance of each model was evaluated, and commonly used performance indicators included accuracy, recall, precision, F1 score, AUC value, etc. The purpose of this step is to determine which models perform best in predicting VTE risk.

[0026] S104, determining to select the machine learning method corresponding to the initial model with the best performance as the preliminary target learning method; After comparing the performance of all initial models, the machine learning method corresponding to the best performing model is selected as the preliminary target learning method, which means that the method provides the best prediction results on the current dataset.

[0027] S105. Based on multiple feature screening algorithms, screen features related to VTE risk in orthopedic patient data to obtain multiple feature sets; Use a variety of feature screening algorithms to identify features associated with VTE risk. These algorithms may include correlation analysis, mutual information, LASSO regression, Ridge regression, etc. Through these algorithms, we can obtain multiple sets of feature sets, each set containing different feature combinations.

[0028] S106, summarizing the multiple feature sets to obtain a total feature set, and sorting each feature in the total feature set from most to least according to the number of times it appears in the multiple feature sets; All feature sets are aggregated into a total feature set and sorted according to the number of times each feature appears in multiple feature sets. Features with more occurrences are considered to be more important features.

[0029] S107, determining a target feature set; wherein determining the target feature set comprises: step A1: setting n=n+1; wherein the initial value of n is 0; step A2: constructing a risk prediction model based on a preliminary target learning method of the top n features, and training and evaluating the performance of the model; repeatedly iterating step A1 and step A2 until the error of the index obtained by the performance evaluation in the most recent two iterations is lower than a preset value; determining the top n-1 features as the target feature set; This step is an iterative process to determine the final set of features to be used for modeling: feature incremental modeling testing can be used to select the number of features to include in the model.

[0030] S108, selecting the machine learning methods corresponding to the first m initial models with the best performance to form a target machine learning method set; In this step, the machine learning methods corresponding to the top m initial models with the best performance are selected to form the target machine learning method set. These methods will be used for the next step of model construction.

[0031] S109, constructing a machine learning model based on each machine learning method in the target machine learning method set and the features corresponding to the target feature set, to obtain a plurality of VTE risk prediction models for orthopedic trauma patients; Based on each method in the target machine learning method set and the features corresponding to the target feature set, multiple VTE risk prediction models for orthopedic trauma patients are constructed.

[0032] S110. Train, tune and evaluate the performance of multiple VTE risk prediction models for orthopedic trauma patients; The multiple VTE risk prediction models constructed were trained, tuned, and evaluated. The purpose of this step is to optimize the model parameters and improve the prediction accuracy of the model.

[0033] S111. Select the VTE risk prediction model for orthopedic trauma patients with the best performance as the target VTE risk prediction model for orthopedic trauma patients.

[0034] After all training and evaluation are completed, the model with the best performance is selected as the target orthopedic trauma patient VTE risk prediction model. This model will be used in practical applications to predict the VTE risk of orthopedic trauma patients.

[0035] Together, these steps constitute a systematic approach to develop and validate an accurate VTE risk prediction model aimed at improving the efficiency and effectiveness of clinical decision making.

[0036] Specifically, in some embodiments, the method further includes: using the Shapley additive interpretation method to perform individual interpretation and variable dependency interpretation on the target orthopedic trauma patient VTE risk prediction model, and calculating the SHAP value of each feature. According to the SHAP value, a bar chart, a honeycomb chart and / or a waterfall chart are drawn to reflect the relative importance of each feature and the quantitative relationship between the feature and the probability of VTE occurrence.

[0037] Specifically, the SHAP value is a method for interpreting model predictions that is based on the Shapley value in cooperative game theory. The SHAP value of each feature represents the average contribution of the feature to the model prediction. A positive value indicates that the feature has a positive impact on the prediction result, and a negative value indicates a negative impact. For each prediction of the target VTE risk prediction model, the SHAP value of all features is calculated. This involves simulating the contribution of each feature to the model prediction value and averaging these contributions to obtain the SHAP value of each feature.

[0038] For the prediction results of individual patients, a waterfall chart is used to show the impact of each feature on the prediction results. The waterfall chart displays the SHAP value of each feature through a bar chart, so that you can intuitively see which features have the greatest positive or negative impact on the VTE risk prediction of a specific patient.

[0039] The waterfall plot can explain why the model predicts a specific VTE risk level for a specific patient. This helps doctors understand the basis for the model's prediction and may influence clinical decision making.

[0040] Draw a bar graph to show the average absolute SHAP value of each feature in all samples, thus reflecting the relative importance of the feature. The height of the bar graph represents the average contribution of the feature to the model prediction.

[0041] The honeycomb diagram is a heat map in which each cell represents a sample and the SHAP value of a feature. The color depth indicates the size of the SHAP value, so that the relationship between different feature values ​​and VTE risk can be intuitively seen.

[0042] Specifically, create a bar chart where the x-axis represents the features and the y-axis represents the average absolute SHAP value of the features. This helps to quickly identify which features have a greater impact on the model prediction. Create a honeycomb chart where the x-axis and y-axis represent the feature value and sample, respectively, and the color depth represents the size of the SHAP value. This helps to understand how changes in feature values ​​affect the prediction of VTE risk. For the selected individuals, create a waterfall chart to show the SHAP value of each feature to explain the model's prediction for a specific individual.

[0043] By analyzing SHAP values ​​and graphs, we explain how each feature affects the prediction of VTE risk. These explanations can be used in clinical decision support to help physicians understand the basis of the model's predictions and potentially guide treatment decisions.

[0044] Through these steps, the SHAP method not only provides transparency of model predictions but also enhances the application value of the model in clinical practice.

[0045] Specifically, the orthopedic patient data include: age, gender, VTE history, comorbidities, time from injury to hospitalization, upper limb fractures, lower limb fractures, high-energy injuries, multiple injuries, open fractures, comminuted fractures, combined vascular and nerve injuries, operation duration, operation type, whether general anesthesia was used, amount of red blood cell suspension infusion, number of days in bed after surgery, whether drug prevention was used, and ICU admission information.

[0046] In some embodiments, performing performance evaluation includes: performing performance evaluation based on accuracy, area under the curve, recall, precision, F1 score, Kappa value and / or MCC.

[0047] The multiple feature screening algorithms include: PearsonCorr (Pearson correlation coefficient) algorithm, SpearmanCorr (Spearman rank correlation coefficient) algorithm, DC (distance correlation analysis) algorithm, MI (mutual information) algorithm, MIC (maximum information coefficient) algorithm, lasso (minimum absolute shrinkage and selection operator) algorithm, ridge (ridge regression) algorithm and MDA (multivariate adaptive regression spline) algorithm.

[0048] In some embodiments, the preset multiple machine learning methods include: GBC machine learning method, Ridge machine learning method, LightGBM machine learning method, RF machine learning method, LDA machine learning method, SVM machine learning method, LR machine learning method, ET machine learning method, NB machine learning method, QDA machine learning method, AdaBoost machine learning method, DT machine learning method and / or Dummy machine learning method.

[0049] Based on the above preferred embodiments, Figure 2 , the solutions provided in this application include: Based on the given inclusion and exclusion criteria, hospitalized orthopedic trauma patients were screened, and medical data were extracted with the informed consent of the patients to establish a database of venous thromboembolism in orthopedic trauma patients. The medical data included age, gender, history of VTE, comorbidities, time from injury to hospitalization, upper limb fractures, lower limb fractures, high-energy injuries, multiple injuries, open fractures, comminuted fractures, combined vascular and nerve injuries, duration of surgery, type of surgery, whether general anesthesia was used, amount of red blood cell suspension transfusion, number of days in bed after surgery, whether drug prevention was used, and ICU admission.

[0050] The assignment method of the above medical information is shown in the following table: Table 1 Assignment method of medical data

[0051] Use a variety of machine learning methods (GBC, Ridge, LightGBM, RF, LDA, SVM, LR, ET, NB, QDA, AdaBoost, DT, Dummy), incorporate all risk factors for modeling, and select the machine learning model with the best performance parameters. The performance parameters evaluated include: Accuracy, Area Under the Curve (AUC), Recall, Precision, F1 score, Kappa value, MCC. For specific examples, the performance indicators of different machine learning models are shown in the following table: Table 2 Performance indicators of different machine learning models

[0052] It can be seen that the random forest model has the best indicators in all aspects. Therefore, the random forest model is further used to optimize hyperparameters through grid search. The specific parameter settings are shown in the table below: Table 3 Specific parameter settings

[0053] For the aforementioned risk factors, PearsonCorr, SpearmanCorr, DC, MI, MIC, lasso, ridge and MDA methods were used for feature screening. A voting table was established based on the feature selection method and features, and the features were sorted by the number of times they were selected by different methods. See the table below: Table 4 Voting table

[0054] Based on the obtained optimal model, the set of predictive variables is added in the order of each feature in the voting table. As the number of predictive features increases, all indicators show a trend of first rising and then leveling off. When the predictive index reaches a certain number, all indicators of the model no longer increase further, indicating that the addition of other additional features can no longer bring additional information to the target classification prediction, and the subsequent added features are redundant or irrelevant to the model. Specifically in this case, the number of optimal predictive indicators is 9, that is, there are 9 VTE risk factors for orthopedic trauma patients screened. The changes in the prediction performance indicators of the random forest model before and after dimensionality reduction are shown in the following table.

[0055] Table 5 Comparison table before and after dimensionality reduction

[0056] Based on the selected features, four machine learning models, including random forest, gradient boosting tree, multilayer perceptron neural network and support vector machine (SVM), were established. The hyperparameters were optimized based on 5-fold cross validation and grid search to obtain the optimal hyperparameters of the models, as shown in the following table: Table 6 Hyperparameter settings

[0057] The model was built based on the optimized hyperparameters, and the model performance was tested using 1000 bootstraps to compare the model's AUC, F1, Precision, and Recall. The results in this example are shown in the following table, and the best model can be identified as the multilayer perceptron neural network (MLP) model, which has the highest AUC, F1 score, and recall.

[0058] Table 7 Model performance test table

[0059] According to the optimal model screened out, the Shapley additive interpretation method was used to interpret various classification and regression models in machine learning to quantify the linear and nonlinear effects of each feature on the model prediction results. This method calculates the SHAP value based on each feature, and the SHAP value shapi of each feature Xi reflects the direction and intensity of the impact on the target. Shapi>0 or <0 indicates that Xi has a positive or negative impact on the target result, thereby obtaining a higher or lower prediction result than the baseline value (i.e., the average prediction value on all analyzed samples). Subsequently, bar charts and honeycomb charts were drawn based on the SHAP value to illustrate the impact of each feature on the occurrence of VTE. Furthermore, with the feature value as the horizontal axis and the risk probability prediction value given by the model as the vertical axis, a partial dependence plot (PDP) was drawn to illustrate the changing relationship between the changing trend of each feature and the outcome of VTE. In terms of individual interpretation, a waterfall chart was used to explain the impact of individual characteristics on the risk probability of VTE.

[0060] Specifically in the example, this study can draw SHAP value bar charts and honeycomb charts. The rows in the SHAP value bar chart correspond to features, which are sorted from high to low according to importance. The number after the feature name represents the feature importance score obtained by the SHAP value, that is, the mean of the absolute value of the SHAP value of the feature on all samples. For the patient population as a whole, AGE has the greatest impact on the risk probability of VTE, with a score of 0.13; ICU has the lowest impact, with a score of only 0.002. In the honeycomb chart, each point represents a sample. The redder or bluer the point, the larger or lower the value of the corresponding row feature, and the horizontal axis is the SHAP score value obtained by the feature on the sample. It can be seen from the honeycomb chart that, except for CHEMOPROPHYLAXIS and ICU, all other feature values ​​are positively correlated with the risk probability of VTE, and the boundary between the feature value and the SHAP score in each sample is clear and less mixed. As the feature importance score gradually decreases, the SHAP value of the feature on each corresponding sample gradually concentrates toward the baseline represented by the vertical solid line.

[0061] In each PDP diagram corresponding to the variable Xi and the target Y, the horizontal axis represents the variable value xi; the vertical axis represents the model output, i.e., the predicted value of VTE risk probability f (xi); the vertical dotted line represents the expected E[Xi] of Xi; the horizontal dotted line represents the expected E[f (xi)] of the model output (i.e., the predicted value of VTE risk probability predicted on all samples). Compared with the honeycomb diagram, the PDP further clearly shows the linear and nonlinear effects of each feature value on the probability of VTE occurrence and the corresponding critical values. For example, when the patient is under 55 years old, the VTE risk probability is lower than the baseline value; when the patient is over 55 years old, the VTE risk probability will be higher than the baseline value, and from the changes in the ICE and PDP curves in the figure, it can be seen that the VTE risk probability has a nonlinear change trend with the increase of AGE, which is first gentle and then accelerated.

[0062] A waterfall chart is used to explain the VTE prediction results of an individual. The horizontal axis in the figure represents the VTE risk probability predicted by the model, where the baseline level can be 0.354; the vertical axis is the characteristics and corresponding values ​​that are arranged from strong to weak for the individual. The VTE risk probability predicted by the model for this individual is 0.809, which is higher than the baseline level. The factors that increase the risk of VTE and the corresponding degree, i.e., the SHAP value, can be shown in red in the figure, and the factors that reduce the risk and the SHAP value are shown in blue. In some embodiments, for some individuals, the operation duration of 10 is the main reason for increasing their VTE prediction risk. In addition, MULTIPLE_TRUMA, LE, and POSTOPRATIVE_BEDREST also have a positive impact on the risk of VTE.

[0063] Figure 3 : is a schematic diagram of a structure of a device for establishing a VTE risk prediction model for orthopedic trauma patients provided by another embodiment of the present invention, and the device is suitable for executing a method for establishing a VTE risk prediction model for orthopedic trauma patients provided by an embodiment of the present invention. Figure 3 As shown, the device may specifically include: An acquisition module 31 is used to acquire orthopedic patient data; The feature determination module 32 is used to respectively perform modeling based on each of the preset multiple machine learning methods to obtain multiple initial prediction models; for each initial model, the initial prediction model is trained based on the orthopedic patient data and the performance is evaluated; the machine learning method corresponding to the initial model with the best performance is determined as the preliminary target learning method; based on multiple feature screening algorithms, the features related to VTE risk in the orthopedic patient data are screened to obtain multiple groups of feature sets; the multiple groups of feature sets are summarized to obtain a total feature set, and each feature in the total feature set is sorted from most to least according to the number of occurrences in the multiple groups of feature sets; the target feature set is determined; wherein, determining the target feature set includes: step A1: let n=n+1; wherein the initial value of n is 0; step A2: based on the preliminary target learning method of the top n features, a risk prediction model is constructed, and the model is trained and the performance is evaluated; iteratively execute steps A1 and A2 until the error of the index obtained by the performance evaluation in the last two iterations is lower than the preset value; the top n-1 features are determined as the target feature set; The model determination module 33 is used to select the machine learning methods corresponding to the top m initial models with the best performance to form a target machine learning method set; construct machine learning models based on each machine learning method in the target machine learning method set and the features corresponding to the target feature set to obtain multiple VTE risk prediction models for orthopedic trauma patients; train, tune and evaluate the performance of multiple VTE risk prediction models for orthopedic trauma patients; and select the VTE risk prediction model for orthopedic trauma patients with the best performance as the target VTE risk prediction model for orthopedic trauma patients.

[0064] The device for establishing a VTE risk prediction model for orthopedic trauma patients provided in an embodiment of the present invention can execute the method for establishing a VTE risk prediction model for orthopedic trauma patients provided in any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method. Since they have been elaborated in detail in the method embodiment, they are not further described in the device embodiment. For details, please refer to the description in the method embodiment.

[0065] The present invention also provides an electronic device. Figure 4 , Figure 4 is a schematic diagram of the structure of an electronic device, such as Figure 4 As shown, the electronic device includes: a processor 41, and a memory 42 connected to the processor 41; the memory 42 is used to store a computer program, and the computer program is at least used to execute the method for establishing a VTE risk prediction model for orthopedic trauma patients in an embodiment of the present invention; the processor 41 is used to call and execute the computer program in the memory.

[0066] An embodiment of the present invention further provides a storage medium storing a computer program. When the computer program is executed by a processor, each step of the method for establishing a VTE risk prediction model for orthopedic trauma patients in the embodiment of the present invention is implemented.

[0067] It can be understood that the same or similar parts of the above embodiments can be referenced to each other, and the contents not described in detail in some embodiments can refer to the same or similar contents in other embodiments.

[0068] It should be noted that, in the description of the present invention, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of the present invention, unless otherwise specified, the meaning of "plurality" refers to at least two.

[0069] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0070] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0071] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0072] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0073] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0074] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0075] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for establishing a VTE risk prediction model for orthopedic trauma patients, characterized in that: include: Obtain orthopedic patient data; Modeling is performed based on each of the preset multiple machine learning methods to obtain multiple initial prediction models; For each initial model, the initial prediction model is trained based on the orthopedic patient data, and a performance evaluation is performed; Determine the machine learning method corresponding to the initial model with the best performance as the preliminary target learning method; Based on multiple feature screening algorithms, features related to VTE risk in orthopedic patient data were screened to obtain multiple sets of feature sets; Summarizing the multiple feature sets to obtain a total feature set, and sorting each feature in the total feature set from most to least according to the number of times it appears in the multiple feature sets; Determine a target feature set; wherein determining the target feature set includes: step A1: set n=n+1; wherein the initial value of n is 0; step A2: construct a risk prediction model based on a preliminary target learning method of the top n features, and train and evaluate the performance of the model; repeatedly iteratively execute steps A1 and A2 until the error of the index obtained by the performance evaluation in the most recent two iterations is lower than a preset value; determine the top n-1 features as the target feature set; Select the machine learning methods corresponding to the top m initial models with the best performance to form a target machine learning method set; Based on each machine learning method in the target machine learning method set and the features corresponding to the target feature set, a machine learning model is constructed to obtain multiple VTE risk prediction models for orthopedic trauma patients; Train, tune, and evaluate the performance of multiple VTE risk prediction models for orthopedic trauma patients; The VTE risk prediction model for orthopedic trauma patients with the best performance was selected as the target VTE risk prediction model for orthopedic trauma patients.

2. The method according to claim 1, characterized in that Also includes: The Shapley additive interpretation method was used to perform individual interpretation and variable-dependent interpretation of the VTE risk prediction model for target orthopedic trauma patients, and the SHAP value of each feature was calculated.

3. The method according to claim 1, characterized in that Also includes: According to the SHAP value, a bar chart, honeycomb chart and / or waterfall chart is drawn to reflect the relative importance of each feature and its quantitative relationship with the probability of VTE occurrence.

4. The method according to claim 1, characterized in that: The orthopedic patient data include: age, gender, VTE history, comorbidities, time from injury to hospital admission, upper limb fractures, lower limb fractures, high-energy injuries, multiple injuries, open fractures, comminuted fractures, combined vascular and nerve injuries, operation duration, operation type, whether general anesthesia was used, amount of red blood cell suspension transfusion, number of days in bed after surgery, whether drug prevention was used, and ICU admission information.

5. The method according to claim 1, characterized in that The performance evaluation includes: performing performance evaluation based on accuracy, area under the curve, recall rate, precision, F1 score, Kappa value and / or MCC.

6. The method according to claim 1, characterized in that The multiple feature screening algorithms include: PearsonCorr algorithm, SpearmanCorr algorithm, DC algorithm, MI algorithm, MIC algorithm, lasso algorithm, ridge algorithm and MDA algorithm.

7. The method according to claim 1, characterized in that The preset multiple machine learning methods include: GBC, Ridge, LightGBM, RF, LDA, SVM, LR, ET, NB, QDA, AdaBoost, DT and / or Dummy.

8. A device for establishing a VTE risk prediction model for orthopedic trauma patients, characterized in that: include: The acquisition module is used to obtain orthopedic patient data; A feature determination module, used for performing modeling based on each of a plurality of preset machine learning methods to obtain a plurality of initial prediction models; For each initial model, the initial prediction model is trained based on the orthopedic patient data, and a performance evaluation is performed; Determine the machine learning method corresponding to the initial model with the best performance as the preliminary target learning method; Based on multiple feature screening algorithms, features related to VTE risk in orthopedic patient data are screened to obtain multiple feature sets; the multiple feature sets are summarized to obtain a total feature set, and each feature in the total feature set is sorted from most to least according to the number of occurrences in the multiple feature sets; Determine a target feature set; wherein determining the target feature set includes: step A1: set n=n+1; wherein the initial value of n is 0; step A2: construct a risk prediction model based on a preliminary target learning method of the top n features, and train and evaluate the performance of the model; repeatedly iteratively execute steps A1 and A2 until the error of the index obtained by the performance evaluation in the two most recent iterations is lower than a preset value; determine the top n-1 features as the target feature set; The model determination module is used to select the machine learning methods corresponding to the top m initial models with the best performance to form a target machine learning method set; construct machine learning models based on each machine learning method in the target machine learning method set and the features corresponding to the target feature set to obtain multiple VTE risk prediction models for orthopedic trauma patients; train, tune and evaluate the performance of multiple VTE risk prediction models for orthopedic trauma patients; and select the VTE risk prediction model for orthopedic trauma patients with the best performance as the target VTE risk prediction model for orthopedic trauma patients.

9. An electronic device, characterized in that: include: A processor, and a memory connected to the processor; The memory is used to store a computer program, and the computer program is used at least to execute the method for establishing a VTE risk prediction model for orthopedic trauma patients according to any one of claims 1 to 7; The processor is used to call and execute the computer program in the memory.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, each step of the method for establishing a VTE risk prediction model for orthopedic trauma patients as described in any one of claims 1 to 7 is implemented.

Images