Diagnostic model for sarcopenia of cardiovascular disease patient and construction method

Abstract

The invention belongs to the technical field of diagnostic model construction, and particularly relates to a diagnostic model for sarcopenia of cardiovascular disease patients and a construction method. Comprising the steps that health related data of a tester are collected, the health related data serve as candidate variables, key factors related to SP are obtained through screening with a multi-dimensional feature selection method, and the key factors comprise continuous variables and classification variables; performing standardization or normalization on the continuous variables to obtain processed continuous variables, and performing One-Hot on the classification variables to obtain processed classification variables; constructing an extreme gradient enhancement model according to the processed continuous variable and the processed classification variable; and performing parameter tuning on the extreme gradient enhancement model by adopting a grid search method in combination with a five-time cross validation method to obtain a diagnosis model. The selected sample has representativeness, and the constructed model is high in diagnosis accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of diagnostic model construction, and specifically relates to a diagnostic model and a construction method for sarcopenia in patients with cardiovascular diseases. Background Art

[0002] Sarcopenia (SP), also known as sarcopenia, is a problem of reduction in the body's entire muscle mass, including its strength, which is usually associated with the elderly and can lead to decreased physical function and quality of life in patients, increased risk of frailty and adverse events, and is considered a complication of cardiovascular disease (CVD).

[0003] At present, the detection methods of SP include: Dual-Energy X-ray Absorptiometry (DXA) is the most commonly used method for measuring human body composition in clinical practice. Its principle is to distinguish different human body components (bone, muscle, fat) based on the difference in the attenuation rate of X-rays passing through different tissues, but it will expose the body of the person being measured to radiation and is expensive. The principle of bioelectrical impedance analysis (BIA) is that when a microcurrent passes through the human body, the resistance of muscle, fat, and bone is different, so the proportion of different components of the human body can be calculated by resistance, thereby evaluating muscle mass. However, although BIA is portable and reasonably priced, the diagnostic critical point is limited to methods for specific populations and specific equipment, which limits its accuracy.

[0004] It is worth noting that with the rapid development of artificial intelligence-assisted diagnosis technology, relevant prediction models for the disease have been developed with high accuracy and sensitivity. In previous studies, a multivariate LR model was used to analyze and screen independent risk factors, and a nomogram model was developed to predict the risk of SP in community elderly people. The results showed that the areas under the curve of the training set and the validation set were 0.827 (95% confidence interval [CI]: 0.792-0.860) and 0.755 (95% CI: 0.680-0.837), respectively. In the West China Health and Aging Trends (WCHAT), serum biomarkers, anthropometric measurements, and physical activity were collected from people aged ≥50 years, and SVM, RF, XGBoost, and W&D machine learning techniques were used to predict SP results with high accuracy and AUC values. Although machine learning models have performed well in previous SP diagnostic studies, the samples selected by the existing SP diagnostic model construction methods are not representative, resulting in low diagnostic accuracy of the constructed models. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the defects in the prior art and provide a diagnostic model and a construction method for sarcopenia in patients with cardiovascular diseases.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions.

[0007] The first object of the present invention is to provide a method for constructing a diagnostic model for sarcopenia in patients with cardiovascular diseases, comprising:

[0008] Screening variables and processing: Collect the health-related data of the test persons, and use the health-related data as candidate variables. Use the multidimensional feature selection method to screen out key factors related to SP, which include continuous variables and categorical variables; standardize or normalize the continuous variables to obtain processed continuous variables; perform One-Hot on the categorical variables to obtain processed categorical variables.

[0009] Model construction and parameter tuning: The extreme gradient boosting model was constructed based on the processed continuous variables and the processed categorical variables. The grid search method combined with the 5-fold cross-validation method was used to tune the parameters of the extreme gradient boosting model to obtain the diagnostic model.

[0010] Preferably, when the proportion of patients with sarcopenia among the test persons is less than or equal to 30%, it is also necessary to use the synthetic minority oversampling technology (SMOTE) to process the proportion of patients with sarcopenia among the test persons so that the proportion of patients with sarcopenia among the test persons is 45% to 55%, and then construct an extreme gradient enhancement model based on the results of the synthetic minority oversampling technology, the processed continuous variables and the processed categorical variables.

[0011] Preferably, the health-related data include main judgment indicators: BMI, height, age, waist circumference, diastolic blood pressure, uric acid, blood sugar and high-density lipoprotein.

[0012] Preferably, the health-related data also include auxiliary judgment indicators: gender, weekly contact with children, taking any antihypertensive drugs, CESD scale, vision or hearing problems, complications, average nightly sleep time and blood urea nitrogen in the past month.

[0013] The continuous variables corresponded to blood urea nitrogen, high-density lipoprotein, blood glucose, uric acid, age, diastolic blood pressure, average sleep time per night in the past month, height, BMI and waist circumference; the categorical variables corresponded to weekly contact with the child, taking any antihypertensive medication, gender, CESD scale, vision or hearing problems, and complications.

[0014] Preferably, the continuous variable is standardized or normalized to obtain a processed continuous variable, and the method is as follows:

[0015] Check whether the continuous variables of all testers conform to the normal distribution, standardize the continuous variables that conform to the normal distribution, and normalize the continuous variables that do not conform to the normal distribution. After standardization and normalization, summarize the processed continuous variables.

[0016] The second object of the present invention is to provide a diagnostic model for sarcopenia in patients with cardiovascular diseases constructed by the construction method.

[0017] The third object of the present invention is to provide a diagnostic system for sarcopenia in patients with cardiovascular diseases, comprising:

[0018] Data entry module, used to collect and enter health-related data of the subjects to be tested;

[0019] A diagnostic model processing module, using the diagnostic model to process the health-related data of the test subjects and calculate the highest Youden index;

[0020] A judgment module is used to judge the risk of sarcopenia in patients with cardiovascular diseases among the test subjects according to the highest Youden index.

[0021] Preferably, the calculation formula of the Youden Index is:

[0022] Youden index = sensitivity + specificity – 1.

[0023] Preferably, when the Youden Index is higher than 0.457, the risk of sarcopenia in patients with cardiovascular diseases among the test subjects is higher than when the Youden Index is lower than 0.457. That is, when the Youden Index is higher than 0.457, the risk of sarcopenia in patients with cardiovascular diseases among the test subjects is high risk; when the Youden Index is lower than 0.457, the risk of sarcopenia in patients with cardiovascular diseases among the test subjects is low risk.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention provides a method for constructing sarcopenia in patients with cardiovascular diseases, including the steps of screening variables and processing (i.e., collecting health-related data of test persons, and using the health-related data as candidate variables, and then screening and processing them through a multidimensional feature selection method to obtain processed continuous variables and processed categorical variables), model construction and parameter tuning, which is used to overcome the problem that the samples selected by the existing SP diagnostic model construction method are not representative, resulting in low diagnostic accuracy of the constructed model.

[0026] We also compared it with three models: logistic regression (LR), support vector machine (SVM), and random forest (RF). The results showed that the discrimination and calibration of the diagnostic model constructed by the present invention were better than those of other models. The diagnostic model constructed by the present invention has high accuracy when used in scenarios such as risk stratification. It can conveniently predict the SP of CVD patients and provide a basis for formulating personalized treatment measures.

[0027] The general inventive concept of the present invention is as follows:

[0028] First, given the increased likelihood of sampling error in small, unrepresentative samples, we used a large Chinese cohort to increase the accuracy and breadth of our results.

[0029] Secondly, because SP is considered a complication after CVD and its incidence is relatively high among the elderly, we constructed a model for middle-aged and elderly people with CVD in the Chinese community.

[0030] Furthermore, the present invention predicts the prevalence of SP in CVD patients by using a parameter-tuned XGBoost model, that is, a diagnostic model. The diagnostic model is an implementation of the gradient boosting technique, which is based on a collection of weak learners and has the characteristics of high bias and low variance. The diagnostic model uses a second-order Taylor series to approximate the value of the loss function, and further reduces the possibility of overfitting through regularization, thereby improving the flexibility and scalability of the diagnostic model.

[0031] When the proportion of patients with sarcopenia among the test subjects is less than or equal to 30%, the diagnostic model may tend to predict the majority class. Therefore, in such a situation, it is also necessary to use synthetic minority oversampling technology to process the proportion of patients with sarcopenia among the test subjects, so that the proportion of patients with sarcopenia among the test subjects is 45%~55%, thereby overcoming the class imbalance problem of whether the test subjects have SP or not. Then, based on the results of the synthetic minority oversampling technology, the processed continuous variables and the processed categorical variables, an extreme gradient enhancement model is constructed to further improve the prediction accuracy of the diagnostic model.

[0032] Finally, the present invention uses SHAP to analyze the importance and direction of action of each post-screening variable in the model, which helps to identify factors with high impact and improve clinical outcomes through early intervention. In addition, it has been verified that the diagnostic model of the present invention has higher accuracy, calibration, and better interpretability than other models, which provides a non-invasive, economical, and harmless alternative imaging method for the early detection of CVD patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a result diagram of the accuracy of the diagnostic model of the present invention in risk stratification; Figure 1A in the figure represents the ROC curve of the prediction model in the test set; Figure 1 B in Figure 2 shows the risk probability scatter plot of SP in CVD patients.

[0034] Figure 2 The results of multivariate LR analysis are shown in Figure 2.

[0035] Figure 3 A result diagram showing the contribution of each screened variable to risk prediction in the diagnostic model provided by the present invention; Figure 3 A in the figure represents the average importance of the variables after screening (|SHAP value|); Figure 3 B in the figure indicates the importance result diagram of all screened variables calculated using the diagnostic model.

[0036] Figure 4 Explanation plot for SHAP prediction results in terms of individual local post-screening variable explanations.

[0037] Figure 5 A result diagram showing the accuracy of the diagnostic model provided by the present invention in stratifying high-risk and low-risk groups of different genders; Figure 5 A in the figure represents the average value of the importance of all screened variables in males (|SHAP value|); Figure 5 B in the figure represents the average importance of all filtered variables in women (|SHAP value|).

[0038] Figure 6 The following is a graph showing the prediction results for risk stratification in different gender groups; Figure 6 The A in the figure indicates the risk stratification prediction results in men; Figure 6 B in the figure indicates the risk stratification prediction results in women. DETAILED DESCRIPTION

[0039] The present invention is described in detail below in conjunction with specific examples, but it should not be construed as limiting the present invention. Unless otherwise specified, the technical means used in the following examples are conventional means well known to those skilled in the art, and the materials, reagents, etc. used in the following examples, unless otherwise specified, can be obtained from commercial sources.

[0040] 1. Testers

[0041] The China Health and Retirement Longitudinal Study (CHARLS) is a national longitudinal survey of Chinese people aged 45 years and over and their spouses. 17,708 participants were recruited from June 2011 to March 2012, and multiple follow-ups were conducted thereafter. The database uses face-to-face computer-assisted personal interviews (CAPI) to survey nearly 10,000 households in 150 counties and 450 villages in 28 provinces in China over an average of 2-3 years, providing relevant social, economic, and health data of community residents. CHARLS was approved by the Biomedical Ethics Review Committee of Peking University, and all participants provided written informed consent.

[0042] Blood tests and body measurements were performed in 2011 and 2015, respectively, so the diagnostic model for SP was developed using data from 2011 and 2015. We excluded testers with a diagnosis of SP at baseline and included testers aged ≥60 years with a history of CVD (n = 1,080) who had blood tests and other health-related data at baseline.

[0043] The test persons who meet the above screening process are used as test persons for constructing the diagnostic model of SP in CVD patients of the present invention.

[0044] 2. Determination of SP

[0045] In this invention, the diagnostic criteria for SP were defined according to the latest version of the Asian Working Group on Sarcopenia 2019 (AWGS2019), which included three components: low muscle mass (<5.61 kg / m 2 , male <7.35kg / m 2 ), muscle strength (<28.0 kg for men and <18.0 kg for women), and low physical performance (gait speed <1 m / s, chair stand test ≥12 s).

[0046] First, muscle mass was assessed by appendicular skeletal muscle mass (ASM) and height. The equation model for ASM was validated in Chinese residents and had strong consistency with DXA: ASM = 0.193 × weight + 0.107 × height - 4.157 × sex - 0.037 × age - 2.631. Height, weight, and age are in centimeters, kilograms, and years, respectively. For sex, a value of 1 represents males and a value of 2 represents females. Subsequently, the researchers adjusted muscle mass according to height in the study population (ASM / height 2 )The lowest 20% for both sexes determined the threshold for low muscle mass.

[0047] In addition, the muscle strength of the test subjects' dominant and non-dominant hands was assessed using a YUEJIANTM WL-1000 dynamometer.

[0048] Finally, in terms of physical performance, the assessment of gait speed and the chair stand test were chosen. To measure gait speed, the researchers asked the test subjects to walk a 2.5-meter distance twice at a normal speed and then took the average. To measure the chair stand test, the test subjects were asked to stand on a 47-centimeter-high chair five times in a row with their arms crossed over their chest, and the time taken (in seconds) was recorded.

[0049] The above method is used to diagnose whether the test person suffers from SP based on clinical symptoms.

[0050] 3. Identification of CVD patients

[0051] Testers identified CVD patients by assessing three questions: “Have you ever been told by your doctor that you have had a heart attack, angina, coronary artery disease, heart failure, or other heart problems?” or “Have you ever been told by your doctor that you have had a stroke?” or “Have you ever been told by your doctor that you have HBP?”

[0052] Specifically, the present invention includes the following embodiments.

[0053] Example 1

[0054] The method for constructing a diagnostic model for SP in CVD patients includes the following steps.

[0055] 1. Screening variables and processing

[0056] In order to efficiently and comprehensively extract key factors related to SP and construct a diagnostic model for SP in CVD patients with accurate diagnostic effect, the present invention collects data from multiple aspects including sociodemographic information, physical, exercise, nutritional factors and clinical factors. The health-related data measured at baseline for all test subjects are used as candidate variables, and the detailed information of the candidate variables is shown in Table 1.

[0057] Table 1 Detailed information of health-related data (i.e., candidate variables)

[0058]

[0059]

[0060] The above candidate variables were screened by multidimensional feature selection method to obtain key factors related to SP based on the following criteria: (1) the missing rate of candidate variables was less than 30%, and the missing data were filled by multiple interpolation, (2) there was a potential causal relationship between the candidate variables and SP, or the LR model analysis showed a significant correlation between the two, and (3) the Mann-Whitney U test and Pearson correlation coefficient were used to exclude candidate variables with low variance and strong correlation with correlation coefficient >0.90.

[0061] The key factors related to SP were screened out through the multidimensional feature selection method. These key factors are the screened variables, including continuous variables and categorical variables.

[0062] Since the variables after screening may have different value ranges, and some machine learning (ML) algorithms are very sensitive to the scale of the variables after screening, the present invention standardizes or normalizes the continuous variables; and performs One-Hot processing on the categorical variables.

[0063] The continuous variables included blood urea nitrogen, high-density lipoprotein, blood glucose, uric acid, age, diastolic blood pressure, average sleep time per night in the past month, height, BMI, and waist circumference; the categorical variables included weekly contact with the child, taking any antihypertensive medication, gender, CESD scale, vision or hearing problems, and complications.

[0064] Among them, the continuous variables of all test persons are checked to see whether they conform to the normal distribution, the continuous variables that conform to the normal distribution are standardized, and the continuous variables that do not conform to the normal distribution are normalized. After standardization and normalization, the processed continuous variables are summarized.

[0065] 2. Derivation and evaluation of prediction models

[0066] The synthetic minority oversampling technology was used to make the proportion of patients with sarcopenia in the test personnel reach 50%. Then, four machine learning models were established according to the results of the synthetic minority oversampling technology, the processed continuous variables and the processed categorical variables. The four machine learning models were LR model, support vector machine (SVM) model, random forest (RF) model and extreme gradient boosting (XGBoost) model.

[0067] After screening the variables, the testers randomly divided the 1,080 testers into training and test sets at a ratio of 7:3. The training set was then preprocessed, parameter tuned, and model trained. Considering the imbalance between positive and negative classes, SMOTE was used to perform class balancing on the XGBoost model.

[0068] The technical means for parameter tuning is the grid search method combined with the 5-fold cross-validation method, that is, tuning the parameter combination of the model. The specific parameter tuning results are shown in Table 2.

[0069] XGBoost after parameter tuning is an innovative diagnostic model for sarcopenia in patients with cardiovascular diseases of the present invention.

[0070] Table 2 Grid search method combined with 5-fold cross validation to determine the optimal parameters of the model

[0071]

[0072]

[0073] The area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity and F1 value were used to evaluate the discrimination and accuracy of the model, and the Brier score was selected to evaluate the calibration of the model. In order to ensure the robustness of the results and limit the overfitting of the model, the present invention set 100 different random seeds to repeat the above process, and calculated the average performance of these 100 repetitions.

[0074] 3. Model Interpretation and Feature Importance

[0075] The "black box" problem is considered to be one of the main obstacles to the further development of ML, so it is necessary to increase the interpretability of the results by visualizing the model results globally and locally. SHapley Additive Interpretation (SHAP) assigns the contribution of feature values ​​to different post-screening variables, and can globally explain and visualize the contribution of each feature in the diagnostic model (corresponding to the post-screening variables in the embodiments of the present invention) to risk prediction. SHAP generates graphics and quantitative explanations to help interpret the model and make more accurate clinical decisions. Its performance has been verified in previous studies. Subsequently, the present invention used SHAP to calculate the importance of features in the diagnostic model (corresponding to the importance of the post-screening variables) to determine the main predictive variables for the occurrence of SP in the elderly population with cardiovascular disease.

[0076] 4. Statistical analysis methods

[0077] To better understand the data distribution and post-screening variables, continuous variables were expressed as mean ± SD and tested using independent sample t-test or Mann-Whitneyu test. Categorical variables were expressed as frequency (percentage) and tested using chi-square test. The highest Youden index was used to define the optimal threshold and distinguish low-risk CVD patients from high-risk CVD patients. Scatter plots were then drawn to describe the distribution of high ML risk and low ML risk. Finally, the relationship between ML risk and 4-year SP was evaluated using univariate and multivariate LR (controlling for the top 5 significant predictors). Two-tailed P value ≤ 0.05 was considered statistically significant. All analyses and calculations were performed using R version 4.2.1 (caret, XGBoost, kernlab, randomForest, DMwR, and shapviz packages). Among them, ML risk refers to the various uncertainties and potential losses that may be encountered when using ML techniques for data processing, model training, and prediction. These risks may come from multiple aspects such as data quality, model performance, algorithm selection, and computing resource limitations.

[0078] 5. Results

[0079] (1) Screening variables of testers in the training and test sets

[0080] The baseline characteristics of the candidate variables of all testers of the present invention are shown in Table 3. The baseline characteristics of the variables after screening are shown in Table 4. The data of the variables after screening are divided into 70% training set and 30% test set. Among them, the prevalence of SP in CVD patients accounted for 18.61% of the total number of testers, and 141 cases of CVD patients with SP in the training set accounted for 18.63% of the testers in the training set; 60 cases in the validation group accounted for 18.58%. Except for the significant differences in high-density lipoprotein HDL (training set 48.53 vs test set 45.32, P = 0.001) and uric acid (training set 4.65 vs test set 4.86, P = 0.018), there were no statistically significant differences in the training set and the test set in the screening variables such as gender, weekly contact with children, whether taking any antihypertensive drugs, CESD scale (10-item scale for depression level assessment of the Center for Epidemiological Survey), vision or hearing problems, complications, blood urea nitrogen (BUN), blood sugar, age, average sleep time per night in the past month, diastolic blood pressure (DBP), height and waist circumference.

[0081] Table 3 Baseline characteristics of candidate variables for all test subjects

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] Note: In Table 3, “-” means that there is no such item, the data in “()” is the percentage of the total number of people in the column, and the rightmost column is the P value of the candidate variable in the corresponding first column.

[0088] Table 4 Baseline characteristics of screening variables

[0089]

[0090] Note: In Table 4, “-” means that there is no such item. The data in “()” is the percentage of the total number of people in the column. The rightmost column is the P value of the selected variable in the corresponding first column.

[0091] (2) Machine learning predicts outcomes

[0092] Table 5 shows the prediction performance of the four machine learning models in the test set. Overall, the performance of the SVM model, RF model, and diagnostic model is better than that of the traditional LR model. More specifically, firstly, the AUC is compared. From Table 1, it can be seen that the RF model has the highest AUC, and the following values ​​are all average values: 0.8769, 95% CI: 0.8367-0.9171; the LR model has the lowest AUC, AUC = 0.8459, 95% CI: 0.7922-0.8996. Secondly, the SVM model has the highest accuracy (0.9681, 95% CI: 0.9429-0.9932) and specificity (0.9000, 95% CI: 0.8241-0.9759). The diagnostic model scored the highest accuracy (0.8050, 95% CI: 0.7575-0.8467), F1 score (0.8706, 95% CI: 0.8408-0.9004), sensitivity (0.8061, 95% CI: 0.7583-0.8539) and Brier score (0.1950, 95% CI: 0.1486-0.2414) performed best. In addition, the AUC of the diagnostic model (0.8670, 95% CI: 0.8184-0.9156) was second only to the RF model. Therefore, the present invention selected the diagnostic model for further prediction.

[0093] Table 5 Results of machine learning models for CVD patients diagnosed with SP

[0094]

[0095] Note: The contents in brackets in Table 5 represent 95% confidence intervals, i.e. 95% CI; the contents before the brackets are specific values.

[0096] (3) Prediction scoring and risk stratification

[0097] See also Figure 1 In Figure A, we use the maximum Youden index as the optimal threshold of 0.457, sensitivity = 0.8500, specificity = 0.7985. It should be noted that Figure 1 After making the ROC curve in Figure A, select the threshold to convert the predicted probability into a classification result to determine whether the patient is sick. Threshold: refers to the specific point selected on the ROC curve, which determines the probability used by the model to distinguish between positive and negative samples. Different thresholds will affect the values ​​of TPR and FPR, thereby affecting the classification effect of the model.

[0098] Youden Index: It is a statistic used to evaluate the selection of thresholds. The higher the value of Youden Index, the better the classification effect of the model. Usually, the point with the largest Youden Index is selected as the optimal threshold.

[0099] After the diagnostic model was established in the test set, the CVD patients were divided into a high-risk group with a Youden index higher than 0.457 and a low-risk group with a Youden index lower than 0.457 according to the calculated maximum Youden index of 0.457. The risk probability scatter plot determined by the diagnostic model ( Figure 1 (B) The results showed that some CVD patients diagnosed with SP had obvious clustering, further demonstrating the accuracy of this diagnostic model in risk stratification.

[0100] (4) Logistic regression analysis

[0101] In the one-way LR analysis, after patients were divided into low-risk and high-risk groups according to the maximum Youden index, the high-risk group was significantly associated with SP after CVD, with an unadjusted odds ratio [OR]: 12.45, 95% CI: 1.87-23.02, P = 4.74 × vacuum.

[0102] After controlling for the five most important predictive factors, BMI, height, age, waist circumference, and diastolic blood pressure, a multivariate LR analysis was performed, and the correlation still existed, with an adjusted OR of 6.98, 95% CI: 1.33-12.63, P = 3.96 × vacuum. The results of the multivariate LR analysis are as follows Figure 2 shown.

[0103] (5) Visualization of feature importance and interpretation of individual variable predictions after screening

[0104] Therefore, the present invention uses SHAP to visualize all the post-screening variables in Table 4. In each post-screening variable importance row, all testers' attribution of the results are represented by dots of different colors. Purple dots represent negative correlations between post-screening variables and results, and yellow dots represent positive correlations. Figure 3 Figure A. Figure 3 Panel B of shows the importance of all post-screening variables screened using the diagnostic model. The X-axis represents a unified index that responds to the impact of a post-screening variable in the diagnostic model and the importance of the post-screening variable, respectively. Age, uric acid, blood glucose, HDL, CESD scale, blood urea nitrogen, comorbidities, taking any medication for antihypertensive treatment, and vision or hearing problems were positively associated with SP. In contrast, BMI, height, waist circumference, diastolic blood pressure, average sleep hours per night in the past month, gender, and weekly contact with children were negatively associated with SP, see Figure 3 Figure A.

[0105] We also give an example to illustrate the power of SHAP in locally explaining individual filtered variables, see Figure 4 Although the specific contribution of each screened variable may be different, the overall trend is consistent with previous research experience or trends documented in the literature.

[0106] (6) Gender subgroup analysis

[0107] In the gender subgroup analysis, we compared Figure 3 The top 10 post-screening variables listed in , which are the most important predictors. Average nightly sleep duration in the past month and HDL, which were significant predictors only in men; in women, they were not among the top 10 predictors. Similarly, some factors, such as CESD scale and comorbidities, were significant top predictors only in women; in men, they were not among the top 10 predictors, see Figure 5 Figure A and Figure B of the diagram.

[0108] The risk probability scatter plot stratified by gender also showed that CVD patients diagnosed with SP were clustered, suggesting that the diagnostic model was also highly accurate in stratifying high-risk and low-risk groups in different genders. Figure 5 Figures C and D of the study.

[0109] 6. Conclusion

[0110] In the Chinese elderly CVD population, this paper demonstrates the feasibility of using a diagnostic model and easily accessible post-screening variables to effectively predict the risk of SP. Combined with the use of SHAP, it can provide a clear explanation for individualized risk prediction and is expected to serve as a decision-making tool for clinicians to screen for SP.

[0111] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0112] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for constructing a diagnostic model for sarcopenia in patients with cardiovascular disease, characterized in that: include: Collecting health-related data of the test persons, and using the health-related data as candidate variables, screening key factors related to SP through a multidimensional feature selection method, wherein the key factors include continuous variables and categorical variables; Standardize or normalize continuous variables to obtain processed continuous variables; perform One-Hot on categorical variables to obtain processed categorical variables; An extreme gradient boosting model was constructed based on the processed continuous variables and the processed categorical variables. The grid search method combined with the 5-fold cross-validation method was used to tune the parameters of the extreme gradient boosting model to obtain a diagnostic model.

2. The method for constructing a diagnostic model for sarcopenia in patients with cardiovascular disease according to claim 1, characterized in that: When the proportion of patients with sarcopenia among the test subjects is less than or equal to 30%, it is also necessary to use synthetic minority oversampling technology to make the proportion of patients with sarcopenia among the test subjects reach 45% to 55%. Then, an extreme gradient enhancement model is constructed based on the results of synthetic minority oversampling technology, the processed continuous variables and the processed categorical variables.

3. The method for constructing a diagnostic model for sarcopenia in patients with cardiovascular disease according to claim 1, characterized in that: The health-related data include the main judgment indicators: BMI, height, age, waist circumference, diastolic blood pressure, uric acid, blood sugar and high-density lipoprotein.

4. The method for constructing a diagnostic model for sarcopenia in patients with cardiovascular disease according to claim 3, characterized in that: The health-related data also include auxiliary judgment indicators: gender, weekly contact with children, taking any antihypertensive medications, CESD scale, vision or hearing problems, complications, average nightly sleep time in the past month and blood urea nitrogen.

5. The method for constructing a diagnostic model for sarcopenia in patients with cardiovascular disease according to claim 4, characterized in that: The continuous variables corresponded to blood urea nitrogen, high-density lipoprotein, blood glucose, uric acid, age, diastolic blood pressure, average sleep time per night in the past month, height, BMI and waist circumference; the categorical variables corresponded to weekly contact with the child, taking any antihypertensive medication, gender, CESD scale, vision or hearing problems, and complications.

6. The method for constructing a diagnostic model for sarcopenia in patients with cardiovascular disease according to claim 1, characterized in that: Standardize or normalize continuous variables to obtain processed continuous variables as follows: Check whether the continuous variables of all testers conform to the normal distribution, standardize the continuous variables that conform to the normal distribution, and normalize the continuous variables that do not conform to the normal distribution. After standardization and normalization, summarize the processed continuous variables.

7. A diagnostic model for sarcopenia in patients with cardiovascular diseases constructed by the construction method according to any one of claims 1 to 6.

8. A diagnostic system for sarcopenia in patients with cardiovascular disease, characterized in that: include: Data entry module, used to collect and enter health-related data of the subjects to be tested; A diagnostic model processing module, using the diagnostic model of claim 7 to process the health-related data of the test subject and calculate the highest Youden index; A judgment module is used to judge the risk of sarcopenia in patients with cardiovascular diseases among the test subjects according to the highest Youden index.

9. The diagnostic system for sarcopenia in patients with cardiovascular diseases according to claim 8, characterized in that: The calculation formula of the Youden Index is: Youden index = sensitivity + specificity – 1.

10. The diagnostic system for sarcopenia in patients with cardiovascular diseases according to claim 9, characterized in that: When the Youden Index is higher than 0.457, the risk of sarcopenia in patients with cardiovascular disease among the trial participants is higher than the risk of sarcopenia in patients with cardiovascular disease among the trial participants when the Youden Index is lower than 0.457.

Images