Clinical prediction model construction method for severe diabetes mellitus based on machine learning

By constructing a clinical prediction model of diabetes based on machine learning, the problem of low prediction accuracy in the existing technology is solved, and early prediction of high sensitivity and accuracy is achieved, the model structure is simplified, suitable for clinical applications, and the medical burden is reduced .

CN118969245APending Publication Date: 2024-11-15QILU INST OF TECH
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411025753.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

In the prior art, the prediction model of diabetes patients is not very accurate and cannot effectively assist medical staff in making decisions, resulting in low efficiency in diagnosis and treatment of patients, and relying on doctors' professional knowledge and clinical experience, which has limitations.

Method used

Using a clinical prediction model for severe diabetes based on machine learning, a variety of machine learning models (such as ridge regression, support vector machine, decision tree, etc.) are constructed, and the features are explained using SHAP Importance, screen out the most important features and build performance-optimized prediction models for early disease prediction and intervention.

Benefits of technology

It improves the accuracy and sensitivity of early predictions of diabetic patients, reduces medical burden, can take timely intervention measures, improves patient survival and treatment effects, simplifies the model structure, and is easy to promote clinically.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118969245A_ABST
    Figure CN118969245A_ABST
Patent Text Reader

Abstract

The invention discloses a method for constructing a clinical prediction model of severe diabetes based on machine learning, and belongs to the technical field of data processing, and the method comprises the steps: obtaining initial physiological and biochemical data from a database when a patient checks in, carrying out the preprocessing of the initial physiological and biochemical data, and dividing the data into a training set and a test set; inputting the training set into a plurality of machine learning models for training; by constructing a confusion matrix of a plurality of models, evaluating the performance of the plurality of models by using a test set, and screening out a model with optimal performance; the importance of each feature is explained by using SHAP, and a plurality of most important features are screened out; reconstructing a prediction model according to the plurality of screened most important features and a performance optimal model; and evaluating the prediction model by using the test set, and performing targeted nursing on the patient according to an output result of the model. Through the prediction model, doctors can be assisted to predict the condition of a patient in an early stage, and the treatment effect and the survival rate of the patient are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for constructing a clinical prediction model for severe diabetes based on machine learning. Background Art

[0002] With the rapid development of urbanization and commercialization, people's lifestyles have changed, leading to an increase in diabetes and non-communicable diseases. Diabetes has changed from a rare disease to an epidemic, and even from a single disease to severe diabetes, requiring admission to the intensive care unit. Diabetes is a chronic disease characterized by high blood sugar. Its etiology is relatively complex and is usually not caused by a single disease. Therefore, if the patient's condition can be predicted at an early stage and intervention measures can be taken in a timely manner, it is of great significance to improve the patient's survival rate and treatment effect.

[0003] However, the traditional prediction method for diabetic patients often requires doctors to conduct on-site assessment of the risk of diabetes based on abnormal data of physiological indicators. This not only requires doctors to have high professional knowledge and rich treatment experience, which has certain limitations, but also has problems such as heavy medical burden and high medical pressure on patients. The use of prediction models can play a role in early intervention for high-risk patients, thereby reducing the economic and medical burden on society. At the same time, it can break the limitations of traditional diagnosis and treatment based on personal professional knowledge and clinical experience of doctors, and provide auxiliary support for disease diagnosis and treatment for medical personnel. However, the prediction model in the existing technology has low prediction accuracy and cannot effectively assist medical staff in making reasonable decisions, resulting in low efficiency in patient disease diagnosis and treatment. Summary of the invention

[0004] In order to solve the above technical problems, this application proposes the following technical solutions:

[0005] In a first aspect, an embodiment of the present application provides a method for constructing a clinical prediction model for severe diabetes based on machine learning, comprising: obtaining initial physiological and biochemical data of a patient upon admission from a database and preprocessing the initial physiological and biochemical data and dividing it into a training set and a test set;

[0006] Inputting the training set into a plurality of machine learning models for training;

[0007] By constructing confusion matrices of multiple machine learning models, the performance of the multiple machine learning models is evaluated using the test set to select the model with the best performance;

[0008] Use SHAP to explain the importance of each feature and select the most important features;

[0009] Rebuilding the prediction model based on the screened most important features and the optimal performance model;

[0010] The prediction model is evaluated using a test set, and targeted care is provided to the patient based on the output of the model.

[0011] In a possible implementation, preprocessing the initial physiological and biochemical data includes:

[0012] Lasso regression is used to analyze and screen the initial physiological and biochemical data;

[0013] Identify features that are highly correlated with severe diabetes, while ignoring features that are either irrelevant or have low correlation;

[0014] After deleting the null value data and the front-end and back-end data of the initial physiological and biochemical data, SMOTE was used to oversample the samples.

[0015] In a possible implementation, the multiple machine learning models include: ridge regression RR, support vector machine SVM, decision tree Decision Tree, multilayer perceptron Multilayer Perceptron, adaptive enhancement Ada Boost, gradient boosting decision tree GBDT and voting classifier Vot.

[0016] In a possible implementation, the training set is input into a plurality of machine learning models for training, including: constructing RR, SVM, Decision Tree, Multilayer Perceptron, Ada Boost, GBDT and Vot respectively, wherein:

[0017] Constructing RR includes: standardizing features and eliminating dimensions between different features; fitting the model using training data, adjusting regularization parameters, and using the test set to evaluate the model’s predictive performance;

[0018] Constructing SVM, including: standardizing features, eliminating dimensions between different features; adjusting parameters after selecting kernel function, and training and evaluating the model;

[0019] Constructing a decision tree includes: selecting the best features for splitting, and dividing the data set into smaller subsets based on the selected features until the stopping condition is met; preventing overfitting through pruning;

[0020] Construct a Multilayer Perceptron, including: determining the number of hidden layers, the number of neurons in each layer, and the activation function; randomly initializing weights and biases; selecting a backpropagation training algorithm to minimize the loss function by iteratively optimizing weights and biases;

[0021] Constructing Ada Boost, including: selecting a base classifier, training the base classifier with weighted training data, updating the weight of the sample according to the performance of the base classifier, and combining multiple base classifiers to form a strong classifier;

[0022] Constructing GBDT, including: initializing the predicted value by setting the mean of the target variable, for each round of iteration, constructing a decision tree to fit the current residual and update the predicted value, adjusting the learning rate, number of data and depth of data, and evaluating and selecting the model;

[0023] Constructing Vot includes: selecting multiple classifiers of different types or classifiers of the same type with different parameters, training each base classifier separately with training data, voting on the prediction results of multiple base classifiers, and selecting the category with the most votes as the final prediction result.

[0024] In a possible implementation, the confusion matrices of multiple machine learning models are constructed, and the performance of the multiple machine learning models is evaluated using the test set to select the model with the best performance, including:

[0025] Construct a confusion matrix using sklearn's confusion-matrix function and matplotlib;

[0026] Fill each sample in the test set into the corresponding position of the confusion matrix according to the labels of the true category and the predicted category;

[0027] The accuracy, recall, specificity, false positive rate (FPR), false negative rate (FNR), positive predictive value (PPV), negative predictive value (NPV), F1 and area under the ROC curve (AUC) of each model were calculated respectively;

[0028] The model corresponding to the optimal value of the Specificity parameter is taken as the model with the best performance.

[0029] In a possible implementation, SHAP is used to explain the importance of each feature and select the most important features, including:

[0030] Calculate the marginal contribution of each feature to the model output and sort them by the contribution;

[0031] Select the preset top-ranked features as important features.

[0032] In a possible implementation, selecting the preset top-ranked features as important features includes:

[0033] Coagulation normalized ratio, prothrombin index, blood sodium ion content, chloride ion content, hematocrit, Glasgow Coma Scale eye reaction, hemoglobin content, blood carbon dioxide partial pressure and anion gap.

[0034] In a possible implementation, reconstructing the prediction model based on the screened most important features and the optimal performance model includes:

[0035] The initial physiological and biochemical data are reprocessed based on the screened features with higher contribution;

[0036] Divide the data into a training set and a test set in a ratio of 8:2;

[0037] The RR algorithm is used to perform model training using the training set.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] This application uses lasso regression for data analysis and screening, which helps to identify variables that are highly correlated with severe diabetes, can effectively reduce unnecessary features in the model, and improve the simplicity and accuracy of the model.

[0040] The prediction model of this application has high sensitivity and accuracy, and can assist doctors in predicting patients' conditions in the early stages, helping to take timely intervention measures, improve treatment outcomes and patient survival rates. At the same time, the prediction model is constructed using easily accessible vital signs and routine blood test indicators, without the need for a large amount of clinical data, making this technology easier to promote and use in clinical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A method for constructing a clinical prediction model for severe diabetes based on machine learning is provided in an embodiment of the present application;

[0042] Figure 2 A confusion matrix architecture diagram provided for an embodiment of the present application;

[0043] Figure 3 A feature importance ranking diagram provided for an embodiment of the present application;

[0044] Figure 4 A case analysis diagram provided for the embodiments of the present application;

[0045] Figure 5 This is a graph of the accuracy of the prediction model provided in the embodiments of the present application. DETAILED DESCRIPTION

[0046] The present solution is described below in conjunction with the accompanying drawings and specific implementation methods.

[0047] Figure 1 A schematic diagram of a method for constructing a clinical prediction model for severe diabetes based on machine learning provided in an embodiment of the present application, see Figure 1 In this embodiment, a method for constructing a clinical prediction model for severe diabetes based on machine learning includes:

[0048] S101, obtaining the initial physiological and biochemical data of the patient when he / she is admitted to the hospital from a database and pre-processing the initial physiological and biochemical data to divide the data into a training set and a test set.

[0049] In order to improve the importance of nursing care for patients at risk of death and improve the survival rate of patients with severe diabetes. This embodiment is a clinical prediction model for patients with severe diabetes. The physiological and biochemical data measured on the first day after the patient entered the ICU are used to predict the patient's possible clinical outcome, wherein the physiological and biochemical data measured on the first day after the patient entered the ICU are obtained from the MIMIC-IV database, and then the initial physiological and biochemical data are analyzed and screened using lasso regression to screen out features that are highly correlated with in-hospital death from diabetes, while ignoring those features that are irrelevant or less correlated. Patient data containing null values ​​and patient data under 20 years old and over 80 years old are deleted, and 4753 available data are obtained, of which 467 people died. The SMOTE method is used to oversample the samples to solve the problem of imbalance between the number of survivors and deaths.

[0050] S102, inputting the training set into a plurality of machine learning models for training.

[0051] In this embodiment, multiple machine learning models are used to predict the inpatient mortality rate of diabetic patients in the intensive care unit. Among them, multiple machine learning models include: ridge regression RR, support vector machine SVM, decision tree Decision Tree, multilayer perceptron Multilayer Perceptron, adaptive enhancement Ada Boost, gradient boosting decision tree GBDT and voting classifier Vot.

[0052] Among them, RR is a linear regression model that prevents overfitting by introducing an L2 regularization term (i.e., the sum of squares of weights). In classification tasks, RR can be used to predict probabilities and converted to categories through thresholds. Constructing RR includes: standardizing features to eliminate the impact of dimensions between different features. Fitting the model using training data, adjusting regularization parameters, and evaluating the model's predictive performance using a test set.

[0053] SVM is a classification method based on statistical learning theory. It finds a hyperplane to divide the training data and maximize the interval between two categories. Constructing SVM includes: standardizing features, eliminating the dimensions between different features, selecting kernel functions such as linear kernel, polynomial kernel, radial basis function kernel, adjusting the regularization parameters and radial basis kernel parameter values, and training and evaluating the model.

[0054] MLP is a feed-forward artificial neural network that contains multiple hidden layers, each layer consisting of multiple neurons. Constructing a Multilayer Perceptron involves: determining the number of hidden layers, the number of neurons in each layer, and the activation function, randomly initializing weights and biases, selecting a back-propagation training algorithm, iteratively optimizing weights and biases to minimize the loss function, and evaluating the model.

[0055] Decision tree is a classification method based on tree structure, which classifies samples through a series of conditional judgments. Building a decision tree includes: selecting the best features for splitting, recursively dividing the data set into smaller subsets based on the selected features until the stopping conditions (such as the size and purity of the subsets) are met, and preventing overfitting through pruning.

[0056] Ada Boost is an ensemble learning method that improves prediction performance by combining multiple weak classifiers into a strong classifier. Each weak classifier focuses on samples that were misclassified by the previous classifier. Constructing Ada Boost includes: selecting a single-layer decision tree base classifier such as a decision stump, training the base classifier with weighted training data, updating the weights of the samples based on the performance of the base classifier, combining multiple base classifiers into a strong classifier, and evaluating the model.

[0057] GDTB is another ensemble learning method that fits the residuals by building a series of decision trees and adds the predictions of these trees to get the final prediction. Building GBDT includes: initializing the prediction value by setting the mean of the target variable, building a decision tree for each iteration to fit the current residuals, and updating the prediction value, adjusting the learning rate, the number of trees, and the depth of the tree, and evaluating and selecting the model.

[0058] Vot is an integrated strategy that votes on the prediction results of multiple classifiers and selects the category with the most votes as the final prediction result. Constructing Vot includes: selecting multiple base classifiers, which can be different types of classifiers or classifiers of the same type but with different parameters, training each base classifier separately with training data, voting on the prediction results of multiple base classifiers, selecting the category with the most votes as the final prediction result, and evaluating the model.

[0059] S103, constructing confusion matrices of multiple machine learning models, and using the test set to evaluate the performance of the multiple machine learning models to select the model with the best performance.

[0060] The model constructed in this embodiment belongs to a two-classification model, and the confusion matrix architecture is as follows: Figure 2 As shown, the confusion matrix is ​​constructed by using sklearn's confusion-matrix function and matplotlib. Each sample in the test set is filled into the corresponding position of the confusion matrix according to the labels of the true category and the predicted category. If the true category of a sample is positive and the predicted category is also positive, the element of the positive category row and column of the confusion matrix is ​​increased by 1. This is repeated for all samples in the test set.

[0061] True class: The true class of the sample is the positive class, and the result of model recognition is also the positive class.

[0062] False negative: The true category of the sample is positive, but the model identifies it as negative.

[0063] False positive: The true category of the sample is negative, but the model identifies it as positive.

[0064] True Negative: The true class of the sample is negative, and the model identifies it as negative.

[0065] According to the above classification, the accuracy, recall, specificity, false positive rate FPR, false negative rate FNR, positive predictive value PPV, negative predictive value NPV, F1 and area under the ROC curve AUC of each model can be calculated respectively. The calculation formula of Accuracy is Accuracy = (TP + TN) / (TP + FN + FP + TN), where TP represents true positive (that is, the number of samples that are actually positive and correctly predicted as positive samples), TN represents true negative (that is, the number of samples that are actually negative and correctly predicted as negative samples), FP represents false positive (that is, the number of samples that are actually negative but predicted as positive samples), and FN represents false negative (that is, the number of samples that are actually positive but predicted as negative samples).

[0066] The calculation formula of Recall is Recall=TP / (TP+FN), where TP and FN have the same meanings as described above.

[0067] The calculation formula of Specificity is Specificity=TN / (TN+FP), wherein TN and FP have the same meanings as described above.

[0068] The calculation formula of FPR is FPR=FP / (FP+TN), where FP represents false positives (i.e., the number of samples that are actually negative but predicted as positive samples), and TN represents true negatives (i.e., the number of samples that are actually negative but correctly predicted as negative samples).

[0069] The calculation formula of FNR is FNR=FN / (TP+FN), where FN represents false negatives (i.e., the number of samples that are actually positive but predicted as negative samples), and TP represents true positives (i.e., the number of samples that are actually positive and correctly predicted as positive samples).

[0070] The calculation formula of PPV is PPV=TP / (TP+FP), where TP represents true positive and FP represents false positive.

[0071] The calculation formula of NPV is NPV = TN / (TN + FN), where TN represents true negative and FN represents false negative. It represents the proportion of samples that are actually negative among all samples predicted to be negative samples.

[0072] F1 is the harmonic mean of precision and recall, and is used to measure the accuracy of a classification model. It is an indicator that takes precision and recall into account, and can more comprehensively evaluate the performance of a model. The calculation formula for F1 is F1 = 2*(Accuracy*Recall) / (Accuracy+Recall). Precision refers to the proportion of samples predicted as positive that are actually positive, and recall refers to the proportion of samples correctly predicted as positive by the classifier to all true positives.

[0073] AUC refers to the area under the ROC curve, which is used to evaluate the performance of a classification model. The larger the AUC value, the better the classification performance of the model. The calculation of AUC is usually based on the ROC curve, which is plotted with the false positive rate (FPR) as the horizontal axis and the true positive rate (TPR) as the vertical axis. The AUC value can be obtained by calculating the area under the ROC curve. Common calculation methods include the direct calculation method and the trapezoidal method. The value range of AUC is between 0 and 1. When AUC is 1, it indicates a perfect classifier, and when AUC is 0.5, it indicates a random classifier.

[0074] The model corresponding to the optimal value of the Specificity parameter is taken as the model with the best performance.

[0075] See Table 1 for relevant parameter indicators of each model. Since this embodiment defines in-hospital death patients as negative samples during model construction, the specificity parameter is focused on. The RR model can reach 91%, indicating that 91% of the death patients are correctly identified, which is the best performance among all models. At the same time, its AUC and NPV performance indicators are at an upper-middle level compared with other models. Therefore, it is expected to be used for the identification of patients with a high clinical risk of death. Therefore, this embodiment selects the RR model for subsequent analysis and optimization.

[0076] Table 1 Confusion matrix related parameter indicators

[0077]

[0078] S104, use SHAP to explain the importance of each feature and select the most important features.

[0079] The SHAP algorithm is used to quantify the contribution of each feature in a machine learning model to the prediction results. It provides a "fair" way to assign the contribution value of each feature to the model prediction results, namely the SHAP value. Figure 3 In this embodiment, the marginal contribution of each feature to the model output is calculated, and the features are sorted according to the contribution, and important features such as coagulation normalization ratio, prothrombin index, sodium ion content in blood, chloride ion content, hematocrit, Glasgow Coma Scale eye reaction, hemoglobin content, blood carbon dioxide partial pressure and anion gap are selected.

[0080] By visualizing the SHAP values, we can further understand the positive and negative relationships between each feature and the predicted results. For example, a high pt_min value (which may refer to the minimum value of the prothrombin time) has a positive effect on the predicted result, meaning that the higher the value, the more the predicted result tends to a certain direction (such as death). On the contrary, a high inr_min value (which may refer to the minimum value of the international normalized ratio) has a negative effect on the predicted result, meaning that the higher the value, the more the predicted result tends to another direction (such as survival), see Figure 4 In the embodiment of the present application, the patient is a deceased patient, and the model predicts that the mortality rate of this patient is 90%, among which the excessively high prothrombin time is the main cause of death.

[0081] S105: Reconstructing a prediction model based on the screened most important features and the optimal performance model.

[0082] In this embodiment, the initial physiological and biochemical data are reprocessed according to the selected features with higher contribution, such as coagulation normalization ratio, prothrombin index, sodium ion content in blood, chloride ion content, hematocrit, Glasgow Coma Scale eye reaction, hemoglobin content, blood carbon dioxide partial pressure and anion gap, and the physiological and biochemical data corresponding to other features are deleted. The data are divided into a training set and a test set at a ratio of 8:2, and the RR algorithm is used, the regularization strength is set to 0.5, the convergence threshold is set to 0.0001, and the training set data is used for model training.

[0083] S106, using the test set to evaluate the prediction model, and providing targeted care to the patient based on the output results of the model.

[0084] This example uses a test set to evaluate the performance of the model and calculate the accuracy of the model. Figure 5 ,After calculation, the true positive rate of the test set reached 84%, which is comparable to the ,prediction accuracy of the unsimplified model and can be used for ,auxiliary prediction in clinical settings.

[0085] In the embodiments of the present application, "at least one" means one or more, and "more" means two or more. "And / or" describes the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B can mean that A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0086] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0087] The above is only a specific implementation of the present application. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for constructing a clinical prediction model for severe diabetes based on machine learning, characterized in that: include: Acquire the initial physiological and biochemical data of the patient when he / she is admitted to the hospital from the database and pre-process the initial physiological and biochemical data and divide them into a training set and a test set; Inputting the training set into a plurality of machine learning models for training; By constructing confusion matrices of multiple machine learning models, the performance of the multiple machine learning models is evaluated using the test set to select the model with the best performance; Use SHAP to explain the importance of each feature and select the most important features; Rebuilding the prediction model based on the screened most important features and the optimal performance model; The prediction model is evaluated using a test set, and targeted care is provided to the patient based on the output of the model.

2. The method for constructing a clinical prediction model for severe diabetes based on machine learning according to claim 1, characterized in that: Preprocessing the initial physiological and biochemical data includes: Lasso regression is used to analyze and screen the initial physiological and biochemical data; Identify features that are highly correlated with severe diabetes, while ignoring features that are irrelevant or have low correlation; After deleting the null value data and the front-end and back-end data of the initial physiological and biochemical data, SMOTE was used to oversample the samples.

3. The method for constructing a clinical prediction model for severe diabetes based on machine learning according to claim 1, characterized in that: The multiple machine learning models include: ridge regression RR, support vector machine SVM, decision tree Decision Tree, multilayer perceptron Multilayer Perceptron, adaptive enhancement Ada Boost, gradient boosting decision tree GBDT and voting classifier Vot.

4. The method for constructing a clinical prediction model for severe diabetes based on machine learning according to claim 3, characterized in that: The training set is input into a variety of machine learning models for training, including: constructing RR, SVM, Decision Tree, Multilayer Perceptron, Ada Boost, GBDT and Vot respectively, where: Constructing RR includes: standardizing features and eliminating dimensions between different features; fitting the model using training data, adjusting regularization parameters, and using the test set to evaluate the model’s predictive performance; Constructing SVM, including: standardizing features, eliminating dimensions between different features; adjusting parameters after selecting kernel function, and training and evaluating the model; Constructing a decision tree includes: selecting the best features for splitting, and dividing the data set into smaller subsets based on the selected features until the stopping condition is met; preventing overfitting through pruning; Construct a Multilayer Perceptron, including: determining the number of hidden layers, the number of neurons in each layer, and the activation function; randomly initializing weights and biases; selecting a backpropagation training algorithm to minimize the loss function by iteratively optimizing weights and biases; Constructing Ada Boost, including: selecting a base classifier, training the base classifier with weighted training data, updating the weight of the sample according to the performance of the base classifier, and combining multiple base classifiers to form a strong classifier; Constructing GBDT, including: initializing the predicted value by setting the mean of the target variable, for each round of iteration, constructing a decision tree to fit the current residual and update the predicted value, adjusting the learning rate, number of data and depth of data, and evaluating and selecting the model; Constructing Vot includes: selecting multiple classifiers of different types or classifiers of the same type with different parameters, training each base classifier separately with training data, voting on the prediction results of multiple base classifiers, and selecting the category with the most votes as the final prediction result.

5. The method for constructing a clinical prediction model for severe diabetes based on machine learning according to claim 1, characterized in that: The method constructs confusion matrices of multiple machine learning models and uses the test set to evaluate the performance of multiple machine learning models to select the model with the best performance, including: Construct a confusion matrix using sklearn's confusion-matrix function and matplotlib; Fill each sample in the test set into the corresponding position of the confusion matrix according to the labels of the true category and the predicted category; The accuracy, recall, specificity, false positive rate (FPR), false negative rate (FNR), positive predictive value (PPV), negative predictive value (NPV), F1 and area under the ROC curve (AUC) of each model were calculated respectively; The model corresponding to the optimal value of the Specificity parameter is taken as the model with the best performance.

6. The method for constructing a clinical prediction model for severe diabetes based on machine learning according to claim 5, characterized in that: The SHAP method is used to explain the importance of each feature and select the most important features, including: Calculate the marginal contribution of each feature to the model output and sort them by the contribution; Select the preset top-ranked features as important features.

7. The method for constructing a clinical prediction model for severe diabetes based on machine learning according to claim 6, characterized in that: The selecting of the preset top-ranked features as important features includes: Coagulation normalized ratio, prothrombin index, blood sodium ion content, chloride ion content, hematocrit, Glasgow Coma Scale eye reaction, hemoglobin content, blood carbon dioxide partial pressure and anion gap.

8. The method for constructing a clinical prediction model for severe diabetes based on machine learning according to claim 1, characterized in that: Rebuilding the prediction model based on the most important features selected and the optimal performance model, including: The initial physiological and biochemical data are reprocessed based on the screened features with higher contribution; Divide the data into a training set and a test set in a ratio of 8:2; The RR algorithm is used to perform model training using the training set.

Citation Information

Cited By

  • Orthopedic trauma patient VTE risk prediction model establishment method and device

    CN119296808A

  • Feature selection method for multi-label prediction of diabetic complications

    CN120724115A

  • Wine production area identification method and system based on volatile compound analysis and machine learning

    CN120927844A