A method, system, device, and medium for constructing a panvascular health index model.
By selecting the optimal model through the Boruta model and multiple machine learning models, correcting the predicted age, and constructing a panvascular health index, the problem of overestimation of low values or underestimation of high values in existing models is solved, and more accurate risk prediction and risk level classification are achieved.
Patent Information
- Application Number
- CN202310744645.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-21
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2043-06-21
AI Technical Summary
Existing panvascular disease risk prediction models suffer from overestimation of low values or underestimation of high values, resulting in unsatisfactory prediction performance.
The Boruta model was used for variable selection, and multiple machine learning models were used to build a preliminary model. The optimal model was selected by the criterion of the highest area under the receiver operating characteristic curve. The difference between the patient's predicted age and actual age was calculated to correct the predicted age. Risk levels were classified based on the assimilated predicted age and the 95% prediction interval, and a pan-vascular health index was constructed.
It improves the accuracy of panvascular disease risk prediction, reduces systematic bias, enables reasonable risk classification, and provides more comprehensive treatment options.
Smart Images

Figure CN116759090B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the fields of clinical medicine and statistics, specifically to a method, system, device, and medium for constructing a panvascular health index model. Background Technology
[0002] Panvascular diseases are a category of vascular system diseases characterized by atherosclerosis, affecting vital organs such as the heart, brain, kidneys, and limbs. In a broader sense, panvascularity refers to small blood vessels, capillaries, veins, as well as tumors, diabetes, and immune disorders. Panvascular diseases are characterized by the systemic nature of vascular diseases, common vascular case features, and shared vascular risk factors. Systemic prevention and treatment begin with understanding the occurrence, development, and characteristics of vascular diseases, shifting from screening for single organs to assessing and stratifying risks based on comprehensive risk factors and multivascular lesions. This holistic approach to systemic factors emphasizes a new disease management model that combines multidisciplinary collaboration and interdisciplinary integration, providing patients with more comprehensive and integrated treatment.
[0003] Existing research includes risk prediction models for vascular diseases. These typically involve multi-dimensional and multi-level data preprocessing and statistical analysis of patient data to select statistically significant data points. The dataset is then used to train and evaluate the model to achieve predictive performance. Algorithms used for training include Logistic Regression, Cox Regression, and Support Vector Machines. However, these models generally exhibit overestimation of low values or underestimation of high values, resulting in unsatisfactory predictive performance. Therefore, a model construction method is needed that can accurately predict the risk of pan-vascular diseases and has a reasonable risk classification. Summary of the Invention
[0004] This disclosure provides a method, system, device, and medium for constructing a pan-vascular health index model, which can solve the problems mentioned in the background art, such as the phenomenon of underestimation or underestimation of low values in the prediction of pan-vascular disease risk, and the unsatisfactory prediction effect. To solve the above technical problems, this disclosure provides the following technical solution:
[0005] As one aspect of this disclosure, a method for constructing a pan-vascular health index model is provided, comprising the following steps:
[0006] Acquire detection data from both vascular disease patients and non-vascular disease patients, and divide the detection data into a modeling dataset and an internal validation dataset;
[0007] The modeling dataset is input into the Boruta model, with age in the detection data as the dependent variable and other detection data in the detection data as independent variables to form a labeled dataset;
[0008] Based on the labeled dataset, a preliminary model was constructed using multiple machine learning models. The optimal parameters, mainly based on the number of iterations, were adjusted. The preliminary model was validated using the highest area under the receiver operating characteristic curve (ROC) as the standard. The preliminary model with the highest area under the ROC curve in the internal validation was selected as the optimal model.
[0009] The difference between the patient's predicted age and actual age is calculated based on the optimal model. The predicted age is then corrected based on the fitting relationship between the difference and the actual age to obtain the assimilated predicted age.
[0010] Based on the assimilation prediction age of the pan-vascular non-disease population, the sample size of each age is supplemented, and a linear function of the predicted age and the 95% prediction interval are fitted.
[0011] Risk levels are classified based on a linear function of predicted age and a 95% prediction interval, thereby constructing a panvascular health index.
[0012] Optionally, the detection data includes at least three of the following parameters:
[0013] Age, sex, systolic blood pressure, diastolic blood pressure, height, weight, waist circumference, total cholesterol, triglycerides, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, lipoprotein (a), non-high-density lipoprotein cholesterol, apolipoprotein B, glycated hemoglobin, fasting glucose, creatinine, blood urea nitrogen, albumin, hemoglobin, D-dimer, cystatin C, homocysteine, C-reactive protein, troponin, brain natriuretic peptide (BNP / NT-proBNP), carotid intima-media thickness, left pulse wave velocity, right pulse wave velocity, left ankle-brachial index, right ankle-brachial index, ST-T changes on ECG, cardiac enlargement on chest X-ray, left ventricular ejection fraction, left ventricular posterior wall thickness at end-diastole, interventricular septum thickness at end-diastole, left ventricular diameter at end-diastole, left atrial volume, E / e'.
[0014] Optionally, after obtaining detection data from pan-vascular disease patients and non-disease patients, before dividing the detection data into a modeling dataset and an internal validation dataset, the following steps are also included:
[0015] Null values and / or outliers in the detection data are removed, and the detection data is divided into a disease population dataset and a non-disease population dataset.
[0016] Optionally, the detection data is divided into a modeling dataset and an internal validation dataset, including the following steps:
[0017] The datasets of the sick population and the non-sick population are divided into internal datasets and external datasets based on time, and the external datasets are used for external validation of the model.
[0018] The internal dataset is further divided into a modeling dataset for building the model and an internal validation dataset for internal validation using a random 8 / 2 partitioning method.
[0019] Optionally, building a preliminary model using multiple machine learning models includes the following steps: building a preliminary model based on the modeling dataset using at least two of the following models: LightGBM, Random Forest, XGBoost, and CatBoost.
[0020] And / or, adjusting the optimal parameters, primarily based on the number of iterations, includes the following steps: using grid tuning to adjust the optimal parameters of the model, primarily based on the number of iterations;
[0021] And / or, the process of selecting the validation model with the highest area under the internally validated receiver operating characteristic (ROC) curve includes the following steps: the ROC curve is plotted with the true positive rate on the ordinate and the false positive rate on the abscissa, which serves as the performance index of the preliminary model, and the preliminary model with the highest area under the internally validated ROC curve is selected as the optimal model.
[0022] Optionally, the difference between the patient's predicted age and actual age is calculated based on the optimal model, and the predicted age is then corrected according to the fitting relationship between the difference and the actual age to obtain the assimilated predicted age, including the following steps:
[0023] Calculate the difference between the predicted age and the actual age for each patient in the optimal model. Then, plot a scatter plot with the difference as the ordinate and the actual age as the abscissa. Fit the scatter points in the scatter plot to obtain a fitted line. Based on the slope and intercept of the fitted line, obtain a fitting equation with the actual age as the independent variable and the predicted age as the dependent variable. Based on the fitted equation, correct the predicted age.
[0024] Optionally, risk levels are classified based on the distribution of predicted age intervals from assimilation prediction, including the following steps:
[0025] All population groups were divided into three categories: low risk, medium risk, and high risk. Low risk meant that the predicted age was lower than the median of the sample fitted line, medium risk meant that the predicted age was slightly higher than the median of the sample fitted line but did not exceed the upper limit of the prediction interval, and high risk meant that the predicted age exceeded the upper limit of the prediction interval.
[0026] As another aspect of this disclosure, a pan-vascular health index model construction system is provided, comprising:
[0027] The detection dataset acquisition module acquires detection data from panvascular disease patients and non-disease patients, and divides the detection data into a modeling dataset and an internal validation dataset.
[0028] The dataset labeling module inputs the modeling dataset into the Boruta model, using age in the detection data as the dependent variable and other detection data in the detection data as independent variables to filter variables and form a labeled dataset.
[0029] The optimal model selection module, based on the labeled dataset, constructs a preliminary model using multiple machine learning models, adjusts the optimal parameters mainly based on the number of iterations, validates the preliminary model using the highest area under the receiver operating characteristic curve as the standard, and selects the preliminary model with the highest area under the internally validated receiver operating characteristic curve as the optimal model.
[0030] The predicted age correction module calculates the difference between the patient's predicted age and actual age based on the optimal model, and then corrects the predicted age according to the fitting relationship between the difference and the actual age to obtain the assimilated predicted age.
[0031] The data resampling and fitting prediction interval module is based on the assimilation prediction age of the pan-vascular non-disease population, supplements the sample size for each age, fits a linear function for the predicted age and the 95% prediction interval.
[0032] The pan-vascular health index construction module classifies risk levels based on a linear function of predicted age and a 95% prediction interval, thereby constructing the pan-vascular health index.
[0033] As another aspect of the present disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the above-described method for constructing a panvascular health index model.
[0034] As another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that the program, when executed by a processor, implements the above-described method for constructing a panvascular health index model.
[0035] Compared to existing technologies, the embodiments of this disclosure construct a pan-vascular health index based on physical examination big data and machine learning algorithms, which can evaluate the overall health status of the pan-vascular system; it can reduce system bias by assimilating prediction results based on actual age, and classify risks based on the distribution of prediction intervals for assimilated prediction age; it can compare the non-compliance rates of various physical examination indicators between diseased and healthy populations, and compare the non-compliance rates under different pan-vascular index risk levels; it can assign scores based on the exceedance of physical examination indicators, and compare the score distribution between diseased and healthy populations under different pan-vascular index risk levels. Attached Figure Description
[0036] Figure 1 This is a flowchart of the method for constructing the panvascular health index model in Embodiment 1 of this disclosure;
[0037] Figure 2 The optimal model obtained from the comparison in Embodiment 1 of this disclosure is the CatBoost model;
[0038] Figure 3 This is a graph showing the fitting results in Embodiment 1 of this disclosure;
[0039] Figure 4 R after assimilation of the predicted age in Embodiment 1 of this disclosure 2 Verification diagram;
[0040] Figure 5 This refers to the difference between the predicted age and the actual age of the diseased population and the healthy population in different age groups in Embodiment 1 of this disclosure;
[0041] Figure 6 , Figure 7 and Figure 8 This is a comparison chart showing the difference in non-compliance rates between the diseased population and the healthy population under different pan-vascular index risk levels in Embodiment 1 of this disclosure.
[0042] Figure 9 A schematic block diagram of the system for constructing the panvascular health index model in Embodiment 2 of this disclosure. Detailed Implementation
[0043] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.
[0044] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.
[0045] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.
[0046] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.
[0047] It is understood that the various method embodiments mentioned above in this disclosure can be combined with each other to form combined embodiments without violating the principle and logic. Due to space limitations, this disclosure will not elaborate further.
[0048] In addition, this disclosure also provides a panvascular health index model construction system, electronic device, computer-readable storage medium, and program, all of which can be used to implement any of the panvascular health index model construction methods provided in this disclosure. The corresponding technical solutions and descriptions are described in the corresponding records in the method section and will not be repeated here.
[0049] The execution entity for the method of constructing the pan-vascular health index model can be a computer or other device capable of constructing the pan-vascular health index model. For example, the method can be executed by a terminal device, server, or other processing device. The terminal device can be user equipment (UE), mobile device, user terminal, cordless phone, personal digital assistant (PDA), handheld device, computing device, in-vehicle device, wearable device, etc. In some possible implementations, this pan-vascular health index model construction method can be implemented by a processor calling computer-readable instructions stored in memory.
[0050] Example 1
[0051] As one aspect of this disclosure, a method for constructing a pan-vascular health index model is provided, such as... Figure 1 As shown, it includes the following steps:
[0052] S10. Obtain detection data from panvascular disease patients and non-disease patients, and divide the detection data into a modeling dataset and an internal validation dataset;
[0053] S20. Input the modeling dataset into the Boruta model, using age in the detection data as the dependent variable and other detection data in the detection data as independent variables to filter variables and form a labeled dataset.
[0054] S30. Based on the labeled dataset, a preliminary model is constructed using multiple machine learning models. The optimal parameters, mainly based on the number of iterations, are adjusted. The preliminary model is validated using the highest area under the receiver operating characteristic curve as the standard. The preliminary model with the highest area under the receiver operating characteristic curve in internal validation is selected as the optimal model.
[0055] S40. Calculate the difference between the patient's predicted age and actual age based on the optimal model, and then correct the predicted age according to the fitting relationship between the difference and the actual age to obtain the assimilated predicted age.
[0056] S50. Based on the assimilation prediction age of the pan-vascular non-disease population, supplement the sample size for each age, fit a linear function for the predicted age and the 95% prediction interval.
[0057] S60. Risk levels are classified based on a linear function of predicted age and a 95% prediction interval, thereby constructing a pan-vascular health index.
[0058] Based on the above steps, the method for constructing a pan-vascular health index model is disclosed in this embodiment. It constructs a pan-vascular health index based on physical examination big data and machine learning algorithms, which can evaluate the rationality of the pan-vascular health index. It can reduce system bias by assimilating prediction results based on actual age, and classify risks based on the distribution of the difference between predicted age and actual age.
[0059] The steps of each embodiment of this disclosure will be described in detail below.
[0060] S10. Obtain detection data from both the pan-vascular diseased population and the non-vascular diseased population. Among them, pan-vascular disease refers to the presence of the following conditions: (1) significant cardiovascular diseases: coronary heart disease, stroke, peripheral artery disease, history of vascular reconstruction, heart failure, atrial fibrillation, dementia, venous thrombosis, etc.; (2) high-risk factors or comorbidities: hypertension, diabetes, dyslipidemia, family history of early-onset CVD, inhaled hormones, malignant tumors, long-term use of nonsteroidal anti-inflammatory drugs or hormones; The detection data are divided into a modeling dataset and an internal validation dataset.
[0061] The original dataset was obtained from a medical institution and included both patients and non-patients. The test data included at least three of the following parameters of the patients: age, gender, systolic blood pressure, diastolic blood pressure, height, weight, waist circumference, total cholesterol, triglycerides, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, lipoprotein(a), non-high-density lipoprotein cholesterol, apolipoprotein B, glycated hemoglobin, fasting glucose, creatinine, blood urea nitrogen, albumin, hemoglobin, D-dimer, cystatin C, homocysteine, C-reactive protein, troponin, brain natriuretic peptide (BNP / NT-proBNP), carotid intima-media thickness, left pulse wave velocity, right pulse wave velocity, left ankle-brachial index, right ankle-brachial index, ST-T changes on electrocardiogram, cardiac enlargement on chest X-ray, left ventricular ejection fraction, left ventricular posterior wall thickness at end-diastole, interventricular septum thickness at end-diastole, left ventricular diameter at end-diastole, left atrial volume, E / e', etc. For example, age, height, and systolic blood pressure can be selected as test data. Preferably, in actual use, all of the above parameters can be included in the original dataset. The more test data selected, the more the test data can reflect the human body's condition.
[0062] Preferably, after obtaining detection data from pan-vascular disease patients and non-disease patients, and before dividing the detection data into a modeling dataset and an internal validation dataset, the following steps are further included:
[0063] Null values and / or outliers in the detection data were removed, and the data was divided into a dataset of patients and a dataset of non-patients. Table 1 shows the number of records before and after the original data was cleaned.
[0064] Table 1. Number of records in the original dataset before and after cleaning
[0065]
[0066]
[0067] Preferably, the detection data is divided into a modeling dataset and an internal validation dataset, including the following steps:
[0068] The datasets of the affected population and the non-affected population are divided into internal and external datasets based on time. The external dataset is used for external validation of the model. The internal and external datasets are divided by time, with January 1, 2019 as the dividing line. The data from 2019 to 2022 is used as the internal dataset, which is further divided into the modeling dataset and the internal validation dataset by random 8 / 2 partitioning, and used for model building and internal validation, respectively. The data from 2018 to 2019 is used as the external dataset for external validation of the model.
[0069] The internal dataset is further divided into a modeling dataset for building the model and an internal validation dataset for internal validation using a random 8 / 2 partitioning method.
[0070] S20. Input the modeling dataset into the Boruta model, using age in the detection data as the dependent variable and other detection data in the detection data as independent variables to filter variables and form a labeled dataset.
[0071] Preferably, the Boruta method in R is used for variable selection, with age as the dependent variable and all other variables such as height, weight, systolic blood pressure, and diastolic blood pressure as independent variables. The modeling dataset is input into the Boruta model. The model will mark variables that will have an effect in the model with age as the dependent variable as "Confirm," meaning they are meaningful in the model and will be included in the output of variables marked "Confirm" during subsequent modeling. In this dataset, all variables are marked "Confirm" and included in subsequent modeling.
[0072] S30. Based on the labeled dataset, a preliminary model is constructed using multiple machine learning models. The optimal parameters, mainly based on the number of iterations, are adjusted. The preliminary model is validated using the highest area under the receiver operating characteristic curve as the standard. The preliminary model with the highest area under the receiver operating characteristic curve in internal validation is selected as the optimal model.
[0073] Preferably, building a preliminary model using multiple machine learning models includes the following steps: building a preliminary model based on the modeling dataset using at least two of the machine learning models such as LightGBM, Random Forest, XGBoost, and CatBoost.
[0074] Optionally, adjusting the optimal parameters, primarily based on the number of iterations, includes the following steps: using GridSearchCV to adjust the optimal parameters of the model, primarily based on the number of iterations;
[0075] Optionally, selecting the validation model with the highest area under the internally validated receiver operating characteristic (AUROC) curve includes the following steps: The receiver operating characteristic curve is a curve plotted using a series of different binary classification methods (such as cutoff values or decision thresholds) with the true positive rate (sensitivity) as the ordinate and the false positive rate (1-specificity) as the abscissa. This curve serves as a performance indicator for the preliminary model, and the preliminary model with the highest area under the internally validated AUROC curve is selected as the optimal model. Figure 2As shown, the optimal model obtained by comparison in this embodiment is the CatBoost model. In reality, due to different data, other models may also yield the optimal model, and it is not limited to the CatBoost model.
[0076] S40. Calculate the difference between the patient's predicted age and actual age based on the optimal model, and then correct the predicted age according to the fitting relationship between the difference and the actual age to obtain the assimilated predicted age.
[0077] Among them, R based on model validation 2 The inventors verified that the model generally exhibited a tendency to overestimate low values and underestimate high values. To address this issue, they first calculated the difference between the predicted age and the actual age for each patient in the optimal model. Then, they plotted a scatter plot with the difference on the ordinate and the actual age on the abscissa. Observing the scatter plot revealed a linear relationship between the difference and the actual age. They then fitted the scatter plot and calculated the slope and intercept of the fitted line. Figure 3 As shown, the fitting result here is y = -0.3703x + 17.087. That is, the actual age is used as x in the model, and a linear equation y = -0.3703x + 17.087 is added to the initial prediction result of the model. Figure 4 As shown, assimilation of the predicted age corrects the common phenomenon of underestimation and underestimation of predicted ages. Internal validation yields R0. 2 The value increased from 0.57 to 0.73. External validation was performed on this assimilated model, and the external validation R value was [value missing]. 2 It reached 0.75.
[0078] S50. Based on the assimilation prediction age of the pan-vascular non-disease population, supplement the sample size for each age, fit a linear function for the predicted age and the 95% prediction interval.
[0079] Due to insufficient sample size for each actual age group in the dataset, the Bootstrap method is used to resample the predicted ages for each actual age group, supplementing the sample size to a level that sufficiently reflects the actual population situation. For example, the 2.5, 50, and 97.5 percentiles of 1000 samples at each actual age are calculated, and the differences between the median and the 2.5 and 97.5 percentiles are calculated. The median for each actual age group is fitted to a straight line y1 = 1.0031x - 1.0038. The mean of the differences between the median and the 2.5 and 97.5 percentiles for each age group is fitted to a quadratic curve y2 = 0.002(x - 20). 2 -0.0462(x-20)+0.8361. With the actual age as x, the line y1 is the median of the prediction interval, y1+y2 is the upper limit of the prediction interval, and y1-y2 is the lower limit of the prediction interval.
[0080] S60. Risk levels are determined based on a linear function of predicted age and a 95% prediction interval. All populations are divided into three categories: low risk, medium risk, and high risk. Low risk means the predicted age is lower than the median of the fitted line of 1000 samples. Medium risk means the predicted age is slightly higher than the median of the fitted line of 1000 samples but does not exceed the upper limit of the prediction interval. High risk means the predicted age exceeds the upper limit of the prediction interval. The prediction interval can be [55, 65].
[0081] In another implementation method, all people are divided into three categories: low-risk, medium-risk, and high-risk. Low-risk means that the predicted age is lower than the actual age, medium-risk means that the predicted age is slightly higher than the actual age (e.g., less than 3.5 years higher than the actual age), and high-risk means that the predicted age is significantly higher than the actual age (e.g., more than 3.5 years higher than the actual age). When 3.5 years is used as the boundary, the number of medium- and high-risk people is roughly the same.
[0082] Based on this, a pan-vascular health index is constructed, such as... Figure 5 As shown. The difference between predicted and actual ages was statistically analyzed for patients and healthy individuals in different age groups, and the model was evaluated through descriptive analysis. This section statistically analyzes the population distribution, mean actual age, mean predicted age, and the difference between mean predicted and actual age for both healthy and diseased individuals in three age groups: under 40 years old, 40-59 years old, and 60 years and older. Overall, the predicted age for healthy individuals is roughly equal to or slightly lower than their actual age, while the predicted age for diseased individuals is slightly higher than their actual age. For example... Figure 6 , Figure 7 and Figure 8 As shown in Table 2, which is a descriptive analysis table of internal data for healthy individuals, and Table 3, which is a descriptive analysis table of internal data for patients, are presented. Table 4 is a descriptive analysis table of external data for healthy individuals, and Table 5 is a descriptive analysis table of external data for patients.
[0083] Table 2. Internal Data: Descriptive Analysis of Healthy Individuals
[0084]
[0085] Table 3. Descriptive Analysis of Patient Population Based on Internal Data
[0086]
[0087] Table 4. Descriptive Analysis of External Data on Healthy Individuals
[0088]
[0089] Table 5. Descriptive Analysis of the Patient Population Based on External Data
[0090]
[0091]
[0092] It is evident that the rate of non-compliance is higher among the sick population than among the healthy population, and the panvascular index shows high risk > medium risk > low risk, which is in line with expectations.
[0093] Preferably, the failure rate of each individual health indicator is calculated for patients and healthy individuals under different risk levels to evaluate the risk classification of the index. For example, systolic blood pressure is judged to be above 130; patients with a blood pressure above 130 are judged to be unqualified for this indicator, while other patients are judged to be qualified. There is no qualified standard for height and weight; BMI is used as the standard. Statistical analysis of the failure rates of each indicator in low-risk, medium-risk, and high-risk groups within both the patient and healthy populations clearly shows that the failure rate in the patient population is higher than that in the healthy population, and the failure rate in the high-risk population is higher than that in the medium- and low-risk populations. This demonstrates that the risk classification of the index is reasonable.
[0094] Example 2
[0095] As another aspect of this disclosure, a panvascular health index model construction system 100 is provided, such as... Figure 9 As shown, it includes:
[0096] The detection dataset acquisition module 1 acquires detection data from panvascular disease patients and non-disease patients, and divides the detection data into a modeling dataset and an internal validation dataset;
[0097] Dataset labeling module 2 inputs the modeling dataset into the Boruta model, using age in the detection data as the dependent variable and other detection data in the detection data as independent variables to filter variables and form a labeled dataset.
[0098] The optimal model selection module 3, based on the labeled dataset, constructs a preliminary model using multiple machine learning models, adjusts the optimal parameters mainly based on the number of iterations, verifies the preliminary model using the highest area under the receiver operating characteristic curve as the standard, and selects the preliminary model with the highest area under the internally validated receiver operating characteristic curve as the optimal model.
[0099] Predicted age correction module 4 calculates the difference between the patient's predicted age and actual age based on the optimal model, and then corrects the predicted age based on the fitting relationship between the difference and the actual age to obtain the assimilated predicted age.
[0100] Module 5, which resamples data to fit the prediction interval, is based on the assimilation prediction age of the pan-vascular non-disease population. It supplements the sample size for each age, fits a linear function for the predicted age, and the 95% prediction interval.
[0101] The Pan-vascular Health Index Construction Module 6 classifies risk levels based on a linear function of predicted age and a 95% prediction interval, thereby constructing the Pan-vascular Health Index.
[0102] In some embodiments, the present disclosure constructs a pan-vascular health index based on physical examination big data and machine learning algorithms, which can evaluate the rationality of the pan-vascular health index; it can reduce system bias by assimilating prediction results based on actual age, and classify risks based on the distribution of the difference between predicted age and actual age.
[0103] The modules of the embodiments of this disclosure will be described in detail below.
[0104] The detection dataset acquisition module 1 is used to acquire detection data from the panvascular disease population and non-disease population, and divide the detection data into a modeling dataset and an internal validation dataset;
[0105] In the dataset acquisition module 1, data obtained from a hospital is used as the original dataset. The original dataset includes two categories: patients and non-patients. The test data includes at least three of the following parameters of the patients: age, gender, systolic blood pressure, diastolic blood pressure, height, weight, waist circumference, total cholesterol, triglycerides, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, lipoprotein(a), non-high-density lipoprotein cholesterol, apolipoprotein B, glycated hemoglobin, fasting glucose, creatinine, blood urea nitrogen, and white blood cell count. The test parameters include: protein, hemoglobin, D-dimer, cystatin C, homocysteine, C-reactive protein, troponin, brain natriuretic peptide (BNP / NT-proBNP), carotid intima-media thickness, left pulse wave velocity, right pulse wave velocity, left ankle-brachial index, right ankle-brachial index, ST-T changes on electrocardiogram, cardiac enlargement on chest X-ray, left ventricular ejection fraction, left ventricular posterior wall thickness at end-diastole, interventricular septum thickness at end-diastole, left ventricular diameter at end-diastole, left atrial volume, and E / e'. For example, age, height, and systolic blood pressure can be selected as test data. Preferably, the more test data selected, the more accurately the test results reflect the body's condition.
[0106] Preferably, after obtaining detection data from pan-vascular disease patients and non-disease patients, and before dividing the detection data into a modeling dataset and an internal validation dataset, the following steps are further included:
[0107] Null values and / or outliers in the detection data were removed, and the data was divided into a dataset of patients and a dataset of non-patients. Table 1 shows the number of records before and after the original data was cleaned.
[0108] Table 1. Number of records in the original dataset before and after cleaning
[0109]
[0110] Preferably, the detection data is divided into a modeling dataset and an internal validation dataset, including the following steps:
[0111] The datasets of the affected population and the non-affected population are divided into internal and external datasets based on time. The external dataset is used for external validation of the model. The internal and external datasets are divided by time, with January 1, 2019 as the dividing line. The data from 2019 to 2022 is used as the internal dataset, which is further divided into the modeling dataset and the internal validation dataset by random 8 / 2 partitioning, and used for model building and internal validation, respectively. The data from 2018 to 2019 is used as the external dataset for external validation of the model.
[0112] The internal dataset is further divided into a modeling dataset for building the model and an internal validation dataset for internal validation using a random 8 / 2 partitioning method.
[0113] Dataset labeling module 2 is used to input the modeling dataset into the Boruta model, using age in the detection data as the dependent variable and other detection data in the detection data as independent variables to filter variables and form a labeled dataset.
[0114] Preferably, the Boruta method in R is used for variable selection, with age as the dependent variable and all other variables such as height, weight, systolic blood pressure, and diastolic blood pressure as independent variables. The modeling dataset is input into the Boruta model. The model will mark variables that will have an effect in the model with age as the dependent variable as "Confirm," meaning they are meaningful in the model and will be included in the output of variables marked "Confirm" during subsequent modeling. In this dataset, all variables are marked "Confirm" and included in subsequent modeling.
[0115] The optimal model selection module 3 is used to construct a preliminary model based on the labeled dataset using multiple machine learning models, adjust the optimal parameters mainly based on the number of iterations, verify the preliminary model with the highest area under the receiver operating characteristic curve as the standard, and select the preliminary model with the highest area under the internally validated receiver operating characteristic curve as the optimal model.
[0116] Preferably, building a preliminary model using multiple machine learning models includes the following steps: building a preliminary model based on the modeling dataset using at least two of the machine learning models such as LightGBM, Random Forest, XGBoost, and CatBoost.
[0117] Optionally, adjusting the optimal parameters, primarily based on the number of iterations, includes the following steps: using GridSearchCV to adjust the optimal parameters of the model, primarily based on the number of iterations;
[0118] Optionally, selecting the validation model with the highest area under the internally validated receiver operating characteristic (AUROC) curve includes the following steps: The receiver operating characteristic curve is a curve plotted using a series of different binary classification methods (such as cutoff values or decision thresholds) with the true positive rate (sensitivity) as the ordinate and the false positive rate (1-specificity) as the abscissa. This curve serves as a performance indicator for the preliminary model, and the preliminary model with the highest area under the internally validated AUROC curve is selected as the optimal model. Figure 2 As shown, the optimal model obtained from comparison in this embodiment is the CatBoost model.
[0119] The optimal model screening module 3 is used to calculate the difference between the patient's predicted age and actual age based on the optimal model, and then correct the predicted age according to the fitting relationship between the difference and the actual age to obtain the assimilated predicted age.
[0120] Among them, R based on model validation 2 The inventors verified that the model generally exhibited a tendency to overestimate low values and underestimate high values. To address this issue, they first calculated the difference between the predicted age and the actual age for each patient in the optimal model. Then, they plotted a scatter plot with the difference on the ordinate and the actual age on the abscissa. Observing the scatter plot revealed a linear relationship between the difference and the actual age. They then fitted the scatter plot and calculated the slope and intercept of the fitted line. Figure 3 As shown, the fitting result here is y = -0.3703x + 17.087. That is, the actual age is used as x in the model, and a linear equation y = -0.3703x + 17.087 is added to the initial prediction result of the model. Figure 4 As shown, assimilation of the predicted age corrects the common phenomenon of underestimation and underestimation of predicted ages. Internal validation yields R0. 2 The value increased from 0.57 to 0.73. External validation was performed on this assimilated model, and the external validation R value was [value missing]. 2 It reached 0.75.
[0121] Predicted age correction module 4 calculates the difference between the patient's predicted age and actual age based on the optimal model, and then corrects the predicted age based on the fitting relationship between the difference and the actual age to obtain the assimilated predicted age.
[0122] Among them, R based on model validation 2 The inventors verified that the model generally exhibited a tendency to overestimate low values and underestimate high values. To address this issue, they first calculated the difference between the predicted age and the actual age for each patient in the optimal model. Then, they plotted a scatter plot with the difference on the ordinate and the actual age on the abscissa. Observing the scatter plot revealed a linear relationship between the difference and the actual age. They then fitted the scatter plot and calculated the slope and intercept of the fitted line. Figure 3 As shown, the fitting result here is y = -0.3703x + 17.087. That is, the actual age is used as x in the model, and a linear equation y = -0.3703x + 17.087 is added to the initial prediction result of the model. Figure 4 As shown, assimilation of the predicted age corrects the common phenomenon of underestimation and underestimation of predicted ages. Internal validation yields R0. 2 The value increased from 0.57 to 0.73. External validation was performed on this assimilated model, and the external validation R value was [value missing]. 2 It reached 0.75.
[0123] Module 5, which resamples data to fit the prediction interval, is based on the assimilation prediction age of the pan-vascular non-disease population. It supplements the sample size for each age, fits a linear function for the predicted age, and the 95% prediction interval.
[0124] Due to insufficient sample size for each actual age group in the dataset, the Bootstrap method is used to resample the predicted ages for each actual age group, supplementing the sample size to a level that sufficiently reflects the actual population situation. For example, the 2.5, 50, and 97.5 percentiles of 1000 samples at each actual age are calculated, and the differences between the median and the 2.5 and 97.5 percentiles are calculated. The median for each actual age group is fitted to a straight line y1 = 1.0031x - 1.0038. The mean of the differences between the median and the 2.5 and 97.5 percentiles for each age group is fitted to a quadratic curve y2 = 0.002(x - 20). 2 -0.0462(x-20)+0.8361. With the actual age as x, the line y1 is the median of the prediction interval, y1+y2 is the upper limit of the prediction interval, and y1-y2 is the lower limit of the prediction interval.
[0125] The Pan-vascular Health Index Construction Module 6 classifies risk levels based on a linear function of predicted age and a 95% prediction interval. It divides all populations into three categories: low risk, medium risk, and high risk. Low risk represents a predicted age lower than the median of the fitted line of 1000 samples, medium risk represents a predicted age slightly higher than the median of the fitted line of 1000 samples but not exceeding the upper limit of the prediction interval, and high risk represents a predicted age exceeding the upper limit of the prediction interval. The prediction interval can be set to [55, 65].
[0126] In another implementation method, all people are divided into three categories: low-risk, medium-risk, and high-risk. Low-risk means that the predicted age is lower than the actual age, medium-risk means that the predicted age is slightly higher than the actual age (e.g., less than 3.5 years higher than the actual age), and high-risk means that the predicted age is significantly higher than the actual age (e.g., more than 3.5 years higher than the actual age). When 3.5 years is used as the boundary, the number of medium- and high-risk people is roughly the same.
[0127] Based on this, a pan-vascular health index was constructed. The difference between predicted and actual ages was statistically analyzed for patients and healthy individuals in different age groups, and the model was evaluated through descriptive analysis. This section statistically analyzed the population distribution, mean actual age, mean predicted age, and the difference between mean predicted and actual age for both healthy and diseased individuals in three age groups: under 40 years old, 40-59 years old, and 60 years and above. Overall, the predicted age for healthy individuals was roughly equal to or slightly lower than their actual age, while the predicted age for diseased individuals was slightly higher than their actual age. Figure 6-8 As shown in Table 2, which is a descriptive analysis table of internal data for healthy individuals, and Table 3, which is a descriptive analysis table of internal data for patients, are presented. Table 4 is a descriptive analysis table of external data for healthy individuals, and Table 5 is a descriptive analysis table of external data for patients.
[0128] Table 2. Internal Data: Descriptive Analysis of Healthy Individuals
[0129]
[0130]
[0131] Table 3. Descriptive Analysis of Patient Population Based on Internal Data
[0132]
[0133] Table 4. Descriptive Analysis of External Data on Healthy Individuals
[0134]
[0135] Table 5. Descriptive Analysis of the Patient Population Based on External Data
[0136]
[0137] It is evident that the rate of non-compliance is higher among the sick population than among the healthy population, and the panvascular index shows high risk > medium risk > low risk, which is in line with expectations.
[0138] Preferably, the failure rate of each individual health indicator is calculated for patients and healthy individuals under different risk levels to evaluate the risk classification of the index. For example, systolic blood pressure is judged to be above 130; patients with a blood pressure above 130 are judged to be unqualified for this indicator, while other patients are judged to be qualified. There is no qualified standard for height and weight; BMI is used as the standard. Statistical analysis of the failure rates of each indicator in low-risk, medium-risk, and high-risk groups within both the patient and healthy populations clearly shows that the failure rate in the patient population is higher than that in the healthy population, and the failure rate in the high-risk population is higher than that in the medium- and low-risk populations. This demonstrates that the risk classification of the index is reasonable.
[0139] Example 3
[0140] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the panvascular health index model construction method of Embodiment 1.
[0141] Embodiment 3 of this disclosure is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this disclosure.
[0142] Electronic devices can take the form of general-purpose computing devices, such as server devices. Components of an electronic device may include, but are not limited to: at least one processor, at least one memory, and buses connecting different system components (including memory and processor).
[0143] The bus includes a data bus, an address bus, and a control bus.
[0144] The memory may include volatile memory, such as random access memory (RAM) and / or cache memory, and may further include read-only memory (ROM).
[0145] The memory may also include program tools having a set (at least one) of program modules, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.
[0146] The processor performs various functional applications and data processing by running computer programs stored in memory.
[0147] Electronic devices can also communicate with one or more external devices (such as keyboards, pointing devices, etc.). This communication can be achieved through input / output (I / O) interfaces. Furthermore, electronic devices can communicate with one or more networks (such as local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via network adapters. The network adapter communicates with other modules of the electronic device via a bus. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.
[0148] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.
[0149] Example 4
[0150] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the panvascular health index model construction method in Embodiment 1.
[0151] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.
[0152] In a possible implementation, this disclosure can also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps of implementing the panvascular health index model construction method described in Embodiment 1.
[0153] The program code for executing this disclosure can be written in any combination of one or more programming languages, and the program code can be executed entirely on a user device, partially on a user device, as a stand-alone software package, partially on a user device and partially on a remote device, or entirely on a remote device.
[0154] Although embodiments of the present disclosure have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present disclosure, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for constructing a panvascular health index model, characterized in that, Includes the following steps: Acquire detection data from both vascular disease patients and non-vascular disease patients, and divide the detection data into a modeling dataset and an internal validation dataset; The modeling dataset is input into the Boruta model, with age in the detection data as the dependent variable and other detection data in the detection data as independent variables to form a labeled dataset; Based on the labeled dataset, a preliminary model is constructed using multiple machine learning models. Optimal parameters, primarily based on the number of iterations, are adjusted. The preliminary model is validated using the highest area under the receiver operating characteristic (ROC) curve as the criterion. The preliminary model with the highest ROC curve area during internal validation is selected as the optimal model. The steps for constructing the preliminary model using multiple machine learning models include: using at least two of the following models—LightGBM, Random Forest, XGBoost, and CatBoost—to construct preliminary models based on the modeling dataset. Adjusting the optimal parameters, primarily based on the number of iterations, includes: using grid-based parameter tuning to adjust the model's optimal parameters, primarily based on the number of iterations. Selecting the preliminary model with the highest ROC curve area during internal validation includes: using the ROC curve (plotted with true positive rate on the y-axis and false positive rate on the x-axis) as a performance indicator of the preliminary model, and selecting the preliminary model with the highest ROC curve area during internal validation as the optimal model. The process involves calculating the difference between the predicted age and the actual age of a patient based on the optimal model, and then correcting the predicted age based on the fitting relationship between the difference and the actual age to obtain the assimilated predicted age. This includes the following steps: calculating the difference between the predicted age and the actual age of each patient in the optimal model, then plotting a scatter plot with the difference as the ordinate and the actual age as the abscissa, fitting the scatter points in the scatter plot to obtain a fitted line, obtaining a fitting equation with the actual age as the independent variable and the predicted age as the dependent variable based on the slope and intercept of the fitted line, and correcting the predicted age based on the fitted equation. Based on the assimilation prediction age of the pan-vascular non-disease population, the sample size of each age is supplemented, and a linear function of the predicted age and the 95% prediction interval are fitted. Risk levels are determined based on a linear function of predicted age and a 95% prediction interval. All populations are divided into three categories: low risk, medium risk, and high risk. Low risk represents a predicted age lower than the median of the fitted line of the sample, medium risk represents a predicted age slightly higher than the median of the fitted line of the sample but not exceeding the upper limit of the prediction interval, and high risk represents a predicted age exceeding the upper limit of the prediction interval. This leads to the construction of a pan-vascular health index.
2. The method for constructing a panvascular health index model as described in claim 1, characterized in that, The detection data includes at least three of the following parameters: Age, sex, systolic blood pressure, diastolic blood pressure, height, weight, waist circumference, total cholesterol, triglycerides, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, lipoprotein a, non-high-density lipoprotein cholesterol, apolipoprotein B, glycated hemoglobin, fasting glucose, creatinine, blood urea nitrogen, albumin, hemoglobin, D-dimer, cystatin C, homocysteine, C-reactive protein, troponin, BNP / NT-proBNP, carotid intima-media thickness, left pulse wave velocity, right pulse wave velocity, left ankle-brachial index, right ankle-brachial index, ST-T changes on ECG, cardiac enlargement on chest X-ray, left ventricular ejection fraction, left ventricular posterior wall thickness at end-diastole, interventricular septum thickness at end-diastole, left ventricular diameter at end-diastole, left atrial volume, E / e'.
3. The method for constructing a panvascular health index model as described in claim 1 or 2, characterized in that, After obtaining detection data from both pan-vascular disease patients and non-disease patients, and before dividing the detection data into a modeling dataset and an internal validation dataset, the following steps are also included: Null values and / or outliers in the detection data are removed, and the detection data is divided into a disease population dataset and a non-disease population dataset.
4. The method for constructing a panvascular health index model as described in claim 3, characterized in that, The detection data is divided into a modeling dataset and an internal validation dataset, including the following steps: The datasets of the sick population and the non-sick population are divided into internal datasets and external datasets based on time, and the external datasets are used for external validation of the model. The internal dataset is further divided into a modeling dataset for building the model and an internal validation dataset for internal validation using a random 8 / 2 partitioning method.
5. A system for constructing a panvascular health index model, characterized in that, include: The detection dataset acquisition module acquires detection data from panvascular disease patients and non-disease patients, and divides the detection data into a modeling dataset and an internal validation dataset. The dataset labeling module inputs the modeling dataset into the Boruta model, using age in the detection data as the dependent variable and other detection data in the detection data as independent variables to filter variables and form a labeled dataset. The optimal model selection module, based on the labeled dataset, constructs a preliminary model using multiple machine learning models, adjusts the optimal parameters (primarily based on the number of iterations), and validates the preliminary model using the highest area under the receiver operating characteristic (ROC) curve as the criterion. The preliminary model with the highest ROC curve area during internal validation is selected as the optimal model. The steps of constructing the preliminary model using multiple machine learning models include: using at least two of the following models—LightGBM, Random Forest, XGBoost, and CatBoost—to construct a preliminary model based on the modeling dataset; adjusting the optimal parameters (primarily based on the number of iterations) includes: using grid-based parameter tuning to adjust the optimal parameters of the model (primarily based on the number of iterations); and selecting the preliminary model with the highest ROC curve area during internal validation includes: using the ROC curve (a curve plotted with the true positive rate on the y-axis and the false positive rate on the x-axis) as a performance indicator of the preliminary model, and selecting the preliminary model with the highest ROC curve area during internal validation as the optimal model. The predicted age correction module calculates the difference between the patient's predicted age and actual age based on the optimal model, and then corrects the predicted age to obtain the assimilated predicted age based on the fitting relationship between the difference and the actual age. The module includes the following steps: calculating the difference between the predicted age and the actual age for each patient in the optimal model, then plotting a scatter plot with the difference as the ordinate and the actual age as the abscissa, fitting the scatter points in the scatter plot to obtain a fitted line, obtaining a fitting equation with the actual age as the independent variable and the predicted age as the dependent variable based on the slope and intercept of the fitted line, and correcting the predicted age based on the fitted equation. The data resampling and fitting prediction interval module is based on the assimilation prediction age of the pan-vascular non-disease population, supplementing the sample size for each age, fitting a linear function for the predicted age and the 95% prediction interval. The pan-vascular health index construction module classifies risk levels based on a linear function of predicted age and a 95% prediction interval, dividing all populations into three categories: low risk, medium risk, and high risk. Low risk represents a predicted age lower than the median level of the sample fitted line, medium risk represents a predicted age slightly higher than the median level of the sample fitted line but not exceeding the upper limit of the prediction interval, and high risk represents a predicted age exceeding the upper limit of the prediction interval, thereby constructing the pan-vascular health index.
6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the panvascular health index model construction method according to any one of claims 1 to 4.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the panvascular health index model construction method according to any one of claims 1 to 4.
Citation Information
Patent Citations
Old people nutrition and health state assessment and risk prediction system based on machine learning
CN114974570A
College faculty worker hypertension risk assessment model construction method based on health physical fitness
CN115775625A
