Pulmonary artery hypertension diagnosis and severity degree judgment prediction model based on pulmonary artery angiography, construction method and application

Abstract

The invention discloses a pulmonary arterial hypertension diagnosis and severity degree judgment prediction model based on pulmonary artery angiography, a construction method and application. By collecting and analyzing pulmonary artery angiography and inspection information of a pulmonary hypertension patient and a contrast, a clinical prediction model for diagnosis and severity judgment of the pulmonary hypertension is constructed, and the diagnosis model and the severity judgment model both comprise five independent prediction variables. After the model is constructed, visualization is carried out through a column diagram, efficiency evaluation is carried out through an ROC curve, a DCA curve, a calibration curve and the like, and inspection is carried out through a machine learning method, which all show that the model constructed by the method has good performance. In addition, through correlation analysis of continuous variables, a linear regression equation with the average pulmonary arterial pressure as a dependent variable is further constructed. The invention provides a new clue basis for noninvasive screening of pulmonary arterial hypertension, and has important significance for expanding disease screening crowds and early diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a prediction model, construction method and application for diagnosing and judging the severity of pulmonary hypertension based on pulmonary artery angiography. Background Art

[0002] Pulmonary hypertension (PH) is a disease caused by a variety of reasons and characterized by pulmonary vasoconstriction and pulmonary vascular remodeling. The global incidence is about 1%, and the incidence is as high as 10% in people over 65 years old. And the 7-year survival rate is only 49% from the date of right heart catheter diagnosis. Its high incidence and mortality make early diagnosis of PH a top priority in clinical work. At present, the gold standard for the diagnosis of PH is right heart floating catheter pressure measurement, but its invasiveness greatly limits its clinical application. Some critically ill patients and some remote areas do not have the conditions for right heart floating catheter examination. Therefore, it is urgent to find a new non-invasive diagnostic model for PH. Summary of the invention

[0003] Pulmonary artery angiography (CTA) is a simple and non-invasive chest imaging examination. In view of the shortcomings of the existing technology, the present invention finds that a new prediction model for clinical diagnosis and severity judgment of pulmonary hypertension can be constructed by combining patient CTA and test information, which will provide new clinical clues for non-invasive diagnosis of pulmonary hypertension. Therefore, the present invention provides a prediction model, construction method and application for diagnosis and severity judgment of pulmonary hypertension based on pulmonary artery angiography.

[0004] To achieve the above object, the technical solution of the present invention is as follows: The present invention collects pulmonary artery angiography (CTA) data and test information of patients with pulmonary hypertension and controls, screens out 21 variables that differ between the case group and the control group through Lasso regression, further screens out 5 variables that affect the outcome (whether or not pulmonary hypertension is present) through logistic regression, and constructs a clinical diagnosis prediction model for pulmonary hypertension: Albumin level: OR value 1.258, 95% CI 1.087-1.499, p = 0.005; Ascending aorta diameter: OR value 0.752, 95%CI 0.641-0.853, p<0.001; Mean platelet volume: OR value 2.193, 95%CI 1.359-3.856, p=0.003; Pulmonary artery width: OR value 1.387, 95%CI 1.204-1.68, p<0.001; Right cardiac output: OR value 0.739, 95%CI 0.585-0.903, p=0.005.

[0005] The area under the ROC curve (AUC) of this model is 0.92.

[0006] The present invention provides new clues for the non-invasive diagnosis of pulmonary hypertension. Through this model, only pulmonary artery CTA and blood tests can be used to predict patients with pulmonary hypertension, providing new clues for the non-invasive diagnosis of pulmonary hypertension.

[0007] Based on this, in a first aspect, the present invention provides a method for constructing a pulmonary hypertension diagnosis prediction model based on pulmonary artery angiography, comprising the following steps: S1: collecting the age, gender, pulmonary artery angiography parameter information and test information of the case group and the control group, using the age, gender, pulmonary artery angiography parameter information and test information of the case group and the control group as variables, constructing samples, and obtaining the first sample set; S2: Eliminate variables and samples with too many missing values ​​or abnormal data in the first sample set to obtain the second sample set. Specifically, eliminate variables and samples with missing values ​​greater than 30% in the sample, interpolate the remaining variables using the KMN method, and finally select useful variables for subsequent analysis.

[0008] S3: Multiple potential predictor variables were screened out in the second sample set through Lasso regression, and all the predictor variables of the case group and the control group constituted the sample to form the third sample set: A total of 21 potential predictor variables were screened out through Lasso regression analysis, namely: right cardiac output, pulmonary artery width, ascending aorta width, right ventricular enlargement, right atrium enlargement, absolute neutrophil value, carbon dioxide concentration, lipoprotein level, total bile acid level, direct bilirubin level, albumin level, creatinine level, age, gender, phosphorus level, high-density lipoprotein level, mean platelet volume, atrial septal defect, ventricular septal defect, phospholipid level, and calcium ion level.

[0009] S4: Through binary logistic regression, the key variables that affect the outcome are screened out in the third sample set and a prediction model is constructed. The outcome refers to whether or not pulmonary hypertension is present. The key variables that affect the outcome are the several independent predictive variables that are finally screened out. All the key variables of the case group and the control group constitute the sample to form the fourth sample set. The key variables of the fourth sample set are statistically analyzed to obtain the model parameters, which are visualized through a nomogram, specifically: The key variables that affect the outcome are the five independent predictive variables that were finally screened out, namely: albumin level, ascending aorta width, mean platelet volume, pulmonary artery width, and right ventricular output; The formula of the prediction model obtained by the binary logistic regression method is: the probability of a patient having pulmonary hypertension is p, logit(p) = -16.217 - 0.303×RCO+ 0.327×PA + 0.23×Alb + 0.785×PltVol- 0.285×AA; Wherein, logit(p) is the value of the logistic function; RCO is the right ventricular output; PA is the pulmonary artery width; Alb is the albumin level; PltVol is the mean platelet volume; AA is the ascending aorta diameter.

[0010] In the formula of the prediction model: Albumin level: OR value 1.258, 95% CI 1.087-1.499, p = 0.005; Ascending aorta width: OR value 0.752, 95%CI 0.641-0.853, p<0.001; Mean platelet volume: OR value 2.193, 95%CI 1.359-3.856, p=0.003; Pulmonary artery width: OR value 1.387, 95%CI 1.204-1.68, p<0.001; Right cardiac output: OR value 0.739, 95%CI 0.585-0.903, p=0.005.

[0011] S5: The clinical prediction ability of the prediction model was evaluated by the receiver characteristic curve, decision curve, and calibration curve; S6: The model is validated by random forest and specifically displayed by SHAP; S7: The prediction model was verified by 10 machine learning methods: decision tree, gradient boosted tree, LightGBM, Logistic classifier, random forest, naive Bayes, CatBoost, XGBoost, support vector machine and classification perceptron.

[0012] In a second aspect, the present invention provides a diagnostic prediction model for pulmonary hypertension based on pulmonary artery angiography, and the prediction model is constructed by the above-mentioned method; the prediction model is visualized by a nomogram, and the nomogram includes a score scale, a total score scale, and predictive variables; the score scale is 0-100 points, the total score scale is 0-240 points, and the predictive variables are: albumin level, ascending aorta width, mean platelet volume, pulmonary artery width, and right cardiac output.

[0013] In a third aspect, the present invention provides a prediction model for judging the severity of pulmonary hypertension based on pulmonary artery angiography, wherein the prediction model is constructed by polynomial Lasso regression and ordered multi-classification Logistic regression analysis, and its independent prediction variables are: right ventricular enlargement, left ventricular enlargement, hemoglobin content, atrial septal defect, and fibrinogen content; the prediction model is expressed as: In the fitting model, different values ​​are assigned to different degrees of pulmonary hypertension: mild pulmonary hypertension is assigned a value of 1, moderate pulmonary hypertension is assigned a value of 2, and severe pulmonary hypertension is assigned a value of 3; the formula of the fitting model is: logit(p 轻 )=-6.667-(0.028×Hb-0.012×FBC-1.577×RVE-3.860×LVE-2.392×DAT), Logit(p 轻 +p 中 )=-4.796-(0.028×Hb-0.012×FBC-1.577×RVE-3.860×LVE-2.392×DAT), p 重 =1-p 轻 -p 中 ; Among them, the probability that the sample is mild pulmonary hypertension is p 轻 , the probability that the sample has moderate pulmonary hypertension is p 中 , the probability that the sample is severe pulmonary hypertension is p 重 ; Hb is hemoglobin content (Hemoglobin), FBC is fibrinogen content (Fibrogen Content), RVE is right ventricular enlargment (Right Ventricular Enlargment), LVE is left ventricular enlargment (Left Ventricular Enlargment), DAT is decreased activity tolerance (Decreased Activity Tolerance).

[0014] In a fourth aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for constructing a prediction model as described above when executing the program.

[0015] The application of the prediction model of the present invention is not limited to the conventional clinical diagnosis of pulmonary hypertension, but can also include the following expanded applications: 1) Early screening for special populations: This model can be used to conduct early screening for people with risk factors for pulmonary hypertension, such as people who are in a chronic low-oxygen environment at high altitudes for a long time, and patients with chronic obstructive pulmonary disease, interstitial lung disease, etc. that cause chronic hypoxia in the body.

[0016] 2) Make a more comprehensive assessment of patients with pulmonary embolism: Pulmonary thromboembolism is a disease with relatively hidden symptoms but a serious condition. Chronic thromboembolism can cause thromboembolic pulmonary hypertension, and the diagnosis of pulmonary thromboembolism often relies on pulmonary angiography. Based on the diagnostic model constructed by the present invention, when performing pulmonary angiography on patients suspected of pulmonary thromboembolism, in addition to understanding the condition of pulmonary thromboembolism, this model can also be used to predict the probability of patients suffering from pulmonary hypertension, providing more clues for the subsequent diagnosis and treatment of patients.

[0017] 3) As a medical supplement: For people in remote areas who do not have the conditions to carry out right heart floating catheter examination, it is simple and easy to conduct pulmonary artery angiography and test information. This model can be used for diagnostic screening of pulmonary hypertension and expand the scope of diagnosis and treatment of pulmonary hypertension.

[0018] 4) Scientific research and education: The model can be used as a scientific research tool to find the incidence and outcome factors related to pulmonary hypertension, explore the pathogenesis of pulmonary hypertension, and the interaction between different factors.

[0019] In a fifth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for constructing a prediction model as described above.

[0020] The present invention discloses a diagnosis and prediction system for pulmonary arterial hypertension, which constructs the prediction model using any of the methods described above; through a computer program, when the program is executed by a processor, it realizes the application of any of the prediction models described above in the clinical diagnosis and severity prediction of pulmonary arterial hypertension.

[0021] The overall scheme of the present invention provides a method for constructing a clinical diagnosis prediction model for patients with pulmonary hypertension using pulmonary artery angiography information, and blood routine, liver and kidney function electrolyte and other test information. The model screens out variables that are different between the pulmonary hypertension group and the control group through multiple statistical methods such as Lasso regression analysis and Logistic regression analysis, screens out independent predictive variables for pulmonary hypertension, and constructs a clinical diagnosis model for pulmonary hypertension with high sensitivity and specificity.

[0022] It is particularly important to note that when using the model of the present invention, attention should be paid to data privacy protection and ethical review to ensure that all operations comply with relevant laws and regulations and medical ethics standards. At the same time, since the model is built based on data from a specific population, differences in factors such as region and race should be taken into account when promoting and applying it, and the model may need to be appropriately adjusted and verified.

[0023] The above model comprehensively utilizes pulmonary artery angiography information, and through statistical methods and machine learning technology, it screens out variables that have a significant impact on the diagnosis of pulmonary hypertension and constructs a highly accurate prediction model.

[0024] The technical principles of the present invention are as follows: In the clinical diagnostic model of pulmonary hypertension: First, the samples were divided into a case group and a control group based on whether they were sick or not, and the age, gender, pulmonary angiography-related parameters and various test indicators of the two groups were counted respectively. Whether or not the disease was the dependent variable (binary variable), and the statistical indicators were used as independent variables (including continuous variables and categorical variables). The independent variables that had the greatest impact on the dependent variable were screened out through Lasso regression. Lasso regression compresses the variable coefficients in the regression model by generating a penalty function to prevent overfitting and solve the problem of serious collinearity. The most relevant indicators and variables can be selected to establish an efficient prediction model. In the present invention, Lasso regression screened out 21 potential predictive variables from the 84 statistical variables.

[0025] Furthermore, we continued to use whether or not the patient was ill as the dependent variable (binary variable) and the 21 variables selected by Lasso regression as independent variables for binary logistic regression analysis. We selected the independent variable with the greatest impact on the dependent variable based on the size of the OR value (Odds Ratio): the larger the OR value, the greater the strength of the association between the independent variable and the dependent variable. At the same time, we selected the independent variable based on the size of the p value: p < 0.05 was considered to have a statistically significant impact on the dependent variable. Finally, logistic regression selected 5 independent variables with high OR values ​​and p < 0.05, and constructed an efficient prediction model. The prediction model constructed by the present invention is: the probability that the patient has pulmonary hypertension is p, logit(p) = -16.217 - 0.303×RCO+ 0.327×PA + 0.23×Alb + 0.785×PltVol- 0.285×AA; wherein, logit(p) is the value of the logistic function; RCO is the right cardiac output; PA is the pulmonary artery width; Alb is the albumin level; PltVol is the mean platelet volume; AA is the ascending aorta width.

[0026] Finally, the model is visualized through a nomogram. The nomogram assigns points to each value level of each independent variable according to the contribution of each independent variable to the dependent variable in the model (the size of the regression coefficient), and then adds up the scores of each independent variable to get the total score. Finally, the predicted value of the individual outcome event is calculated through the functional conversion relationship between the total score and the probability of the outcome event. The nomogram transforms the complex regression equation into a visual graph, making the results of the prediction model more readable and easier to apply in clinical practice.

[0027] After the model is built, the variables in the model are incorporated into the random forest machine learning model and visualized through SHAP to obtain parameter information such as the importance ranking of the variables to the model, and further understand the model. In addition, the variables in the model are verified through 10 machine learning methods, and the model performance is judged through ROC and other methods to conduct a comprehensive performance evaluation of the model.

[0028] In the pulmonary hypertension severity judgment model: In the construction of the logistic regression model, the samples of the pulmonary hypertension case group were analyzed. According to the international pulmonary hypertension severity classification standard, the case group was divided into mild pulmonary hypertension (20 mmHg<mPAP≤40 mmHg), moderate pulmonary hypertension (40 mmHg<mPAP≤55 mmHg), and severe pulmonary hypertension (55 mmHg<mPAP) according to the mean pulmonary artery pressure (mPAP) level. With the severity of pulmonary hypertension as the dependent variable, the variables correlated with the severity of pulmonary hypertension were screened out through correlation analysis, and then the potential outcome predictor variables were further screened out through lasso polynomial regression analysis. Finally, the independent predictor variables were screened out through ordered three-category logistic regression and a prediction model for the severity of pulmonary hypertension was constructed.

[0029] In the construction of the linear regression equation, the samples of the pulmonary hypertension case group were analyzed, and the correlation analysis between all continuous variables and the pulmonary hypertension level was performed. Six variables with high correlation were screened out according to p < 0.05 and correlation coefficient r > 0.3. The above six variables were subjected to stepwise regression (both)---linear regression, and the regression equation constructed by the two variables of pulmonary artery width and red blood cell level was finally screened out according to p < 0.05. This equation can predict the pulmonary artery pressure level by pulmonary artery width and red blood cell level.

[0030] The advantages and beneficial effects of the present invention are as follows: 1. Strict inclusion and exclusion criteria: Strict inclusion and exclusion criteria were set in the study design. First, all patients with pulmonary hypertension were diagnosed using the gold standard for pulmonary hypertension diagnosis: the average pulmonary artery pressure measured by the right heart floating catheter was >20 mmHg; the conclusions of all pulmonary artery angiography examinations included in the control group were: no obvious abnormalities were found in pulmonary artery CTA.

[0031] 2. Application of statistical methods: In the diagnosis and prediction model of pulmonary hypertension, Lasso regression eliminated variables with collinearity and screened out potential predictive variables for pulmonary hypertension, while the application of binary logistic regression screened out independent predictive variables for pulmonary hypertension. In the prediction model of the severity of pulmonary hypertension, the application of multinomial lasso regression and ordered three-classification logistic regression screened out independent predictive variables for judging the severity of pulmonary hypertension. The application of these statistical methods can effectively screen out key variables related to pulmonary hypertension, discover independent predictive factors of the disease, and build a high-precision prediction model, which is of great significance for early diagnosis and risk assessment.

[0032] 3. Construction of a clinical diagnosis prediction model for pulmonary hypertension: The prediction model finally constructed includes 5 independent prediction variables: right ventricular output, pulmonary artery width, albumin level, mean platelet volume, and ascending aorta width; and the prediction formula is calculated: the probability of a patient suffering from pulmonary hypertension is p, logit(P) = -16.217 - 0.303×RVCO+ 0.327×PA + 0.23×Alb + 0.785×PltVol- 0.285×AA; where logit(p) is the value of the logistic function; RVCO is right ventricular output; PA is pulmonary artery width; Alb is albumin level; PltVol is mean platelet volume; and AA is the inner diameter of the ascending aorta.

[0033] The prediction model is visualized through a nomogram, making it easy to use in clinical practice. Through multi-dimensional evaluations such as ROC (AUC is 0.92), calibration curve (mean square absolute error 0.049), DCA (prevalence has good benefits in the range of 0-1), it can be seen that the model has excellent predictive ability and has extremely high application value in clinical practice. It can help clinicians identify patients with pulmonary hypertension earlier and take timely intervention measures.

[0034] 4. Verification of machine learning methods: The present invention uses 10 different machine learning methods, including decision tree, gradient boosting tree, LightGBM, Logistic classifier, random forest, naive Bayes, CatBoost, XGBoost, support vector machine, and classification perceptron, to verify the model. The average AUC value of ROC of the 10 machine learning methods for model verification is 0.879, which means that the model shows good prediction performance under different algorithms, which enhances its confidence in practical applications.

[0035] 5. Construction of a prediction model for the severity of pulmonary hypertension: In the fitting model, different values ​​are assigned to different degrees of pulmonary hypertension: mild pulmonary hypertension is assigned a value of 1, moderate pulmonary hypertension is assigned a value of 2, and severe pulmonary hypertension is assigned a value of 3; the formula of the fitting model is: logit(p 轻 )=-6.667-(0.028×Hb-0.012×FBC-1.577×RVE-3.860×LVE-2.392×DAT), Logit(p 轻 +p 中 )=-4.796-(0.028×Hb-0.012×FBC-1.577×RVE-3.860×LVE-2.392×DAT), p重 =1-p 轻 -p 中 ; Among them, the probability that the sample is mild pulmonary hypertension is p 轻 , the probability that the sample has moderate pulmonary hypertension is p 中 , the probability that the sample is severe pulmonary hypertension is p 重 ; Hb is the hemoglobin content, FBC is the fibrinogen content, RVE is the right ventricular enlargement, LVE is the left ventricular enlargement, and DAT is the decreased activity tolerance.

[0036] The model constructs prediction formulas for mild, moderate and severe pulmonary hypertension, which can help clinical physicians quickly determine the severity of the disease in patients with pulmonary hypertension and provide an important basis for subsequent targeted diagnosis and treatment.

[0037] 6. Construction of the linear regression equation of mean pulmonary artery pressure: Through correlation analysis and linear regression analysis, a regression equation with mean pulmonary artery pressure as the dependent variable was constructed: y = -15.949 + 1.123× PA + 8.808× RBC +ε, where y is the mean pulmonary artery pressure level, PA is the pulmonary artery width, and RBC is the red blood cell level.

[0038] In summary, the present invention demonstrates the significant advantages and potential beneficial effects of pulmonary angiography and related test information in the diagnosis and prediction of pulmonary hypertension through strict inclusion and exclusion criteria, accurate application of statistical methods, effective clinical prediction model construction and multi-angle model verification. These achievements are expected to be widely used in clinical practice, thus providing a new tool for the diagnosis of pulmonary hypertension for patients with contraindications to right heart floating catheter examination and patients in remote areas who are unable to undergo right heart floating catheter examination, which is of great significance for the early diagnosis and treatment of pulmonary hypertension and improving the survival rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Construct a flow chart for the model of the present invention.

[0040] Figure 2 This is the lasso regression screening process of the present invention. Wherein: A: Lasso coefficient path diagram; B: Lasso regression analysis cross validation curve.

[0041] Figure 3 It is a forest diagram of the prediction model of the present invention.

[0042] Figure 4 It is the nomogram of the prediction model of the present invention.

[0043] Figure 5This is a multi-dimensional evaluation of the prediction model of the present invention, where A: ROC curve; B: DCA curve; C: calibration curve.

[0044] Figure 6 The present invention uses SHAP (SHapley Additive exPlanations) to visualize the prediction model verified by random forest; A: SHAP bee chart; B: SHAP importance ranking chart; C: SHAP heat map; D: SHAP waterfall chart; E: SHAP heat map.

[0045] Figure 7 This is a validation diagram of the 10 machine learning multi-prediction models of the present invention, wherein A: ROC of multiple machine learning; B: summary of ROC curves of multiple machine learning and their confidence intervals; C: precision and completeness curves of multiple machine learning; D: DCA curves of multiple machine learning.

[0046] Figure 8 This is the lasso polynomial regression screening process of the present invention. Wherein: A: Lasso coefficient path diagram; B: Lasso coefficient path diagram; C: Lasso coefficient path diagram; D: Lasso regression analysis cross validation curve.

[0047] Fig. 9 This is a validation diagram of the prediction model of the present invention, wherein A: a three-classification logistic regression heat map; B: ROC of the prediction model.

[0048] Fig.10 This is a correlation analysis of the pulmonary hypertension pressure level of the present invention, wherein A: correlation analysis between pulmonary artery width (PA) and mean pulmonary artery pressure, B: correlation analysis between hemoglobin level (Hemoglobin) and mean pulmonary artery pressure, C: correlation analysis between red blood cell level (RBC) and mean pulmonary artery pressure, D: correlation analysis between hematocrit (HCT) and mean pulmonary artery pressure, E: correlation analysis between increased body mass index (BMI) and mean pulmonary artery pressure, F: correlation analysis between fibrinogen content and mean pulmonary artery pressure. DETAILED DESCRIPTION

[0049] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0050] Example 1 Construction of a clinical prediction model for diagnosing pulmonary hypertension based on pulmonary artery angiography (CTA) and test information S1. Collect general information, CTA parameter information, and test information of patients with PH and control group who underwent pulmonary artery angiography (CTA) in our hospital from April 2022 to April 2024. All PH patients were diagnosed with PH (mean pulmonary artery pressure>20 mmHg) by right heart floating catheter, and the conclusions of all control CTA examinations were: no obvious abnormalities were found in pulmonary artery CTA.

[0051] General information collected included: gender, age; CTA information includes: right atrial enlargement, right ventricular enlargement, left atrial enlargement, left ventricular enlargement, atrial septal defect, ventricular septal defect, right ventricular transverse diameter, right ventricular wall thickness transverse diameter, right atrial superior-inferior diameter, right atrial left-right diameter, left atrial anterior-posterior diameter, left atrial superior-inferior diameter, left atrial left-right diameter, left ventricular transverse diameter, left ventricular end-diastolic volume (LEDV), left ventricular end-systolic volume (LESV), left ventricular stroke volume (LSV), left ventricular ejection fraction (LEF), left ventricular mass (Leftmass), left ventricular output (LeftCO), right ventricular end-diastolic volume (REDV), right ventricular end-systolic volume (RESV), right ventricular stroke volume (RSV), right ventricular ejection fraction (REF), right ventricular output (Right CO), pulmonary artery width (PA), left pulmonary artery diameter (LPA), right pulmonary artery diameter (RPA), McGoon ratio, aortic sinotubular junction diameter, ascending aorta diameter, descending aorta diameter, descending aorta diameter at diaphragm level, and total calcification score.

[0052] Inspection information includes: Routine blood test: white blood cells, red blood cells, hemoglobin, platelets, neutrophil percentage, lymphocyte percentage, monocyte percentage, eosinophil percentage, basophil percentage, absolute neutrophil value, absolute lymphocyte value, absolute monocyte value, absolute eosinophil value, absolute basophil value, hematocrit, mean corpuscular volume, mean hemoglobin content, mean hemoglobin concentration, red blood cell distribution width cv, red blood cell distribution width SD, mean platelet volume; Coagulation function: prothrombin time, international normalized ratio, prothrombin activity, activated partial thromboplastin time, thrombin time, fibrinogen content; D2 dimer; NTProBNP; Procalcitonin (PCT); Electrolytes for liver and kidney function: alanine aminotransferase, aspartate aminotransferase, ratio of alanine aminotransferase to aspartate aminotransferase, direct bilirubin, indirect bilirubin, total bilirubin, total protein, albumin, globulin, albumin-globulin ratio, γ-glutamyl transferase, alkaline phosphatase, total bile acid, creatinine, urea nitrogen, uric acid, CO2, serum cystatin C, potassium ion, sodium ion, chloride ion, calcium ion, magnesium ion, phosphorus ion, total cholesterol, triglycerides, high-density lipoprotein, low-density lipoprotein, small low-density lipoprotein, lipoprotein a, free fatty acids, phospholipids; Thyroid function: free T3, free T4, thyroid stimulating hormone; Immune function: antiphospholipid antibody IgM, antiphospholipid antibody IgG, antiphospholipid antibody IgA, complement C3, complement C4, immunoglobulin IgG, immunoglobulin IgA, immunoglobulin IgM, immunoglobulin IgE, antistreptolysin (ASO), rheumatoid factor (RF), rheumatoid factor antibody IgM, rheumatoid factor antibody IgG, rheumatoid factor antibody IgA; Blood gas analysis: pH value, oxygen partial pressure, oxygen saturation, CO2 partial pressure, body temperature, lactic acid, standard bicarbonate, actual base excess, standard base excess, anion gap; S2. Variables and samples with missing values ​​> 30% were eliminated, and the remaining variables were interpolated using the KMN method. Finally, there were 59 pulmonary hypertension cases and 43 controls, and 84 variables were included in the subsequent analysis, as shown in Table 1.

[0053] Table 1. Samples and variables included in the model

[0054]

[0055]

[0056] S3. Variables that differ between the case group and the control group were screened out through Lasso regression ( Figure 2 Middle A, Figure 2 B), and the variables with collinearity were eliminated. A total of 21 potential predictor variables were screened: right cardiac output, pulmonary artery width, ascending aorta diameter, right ventricular enlargement, right atrial enlargement, absolute neutrophil count, carbon dioxide concentration, lipoprotein, total bile acid, direct bilirubin, albumin, pulmonary artery width, age, gender, phosphorus level, high-density lipoprotein, mean platelet volume, presence of atrial septal defect, presence of ventricular septal defect, phospholipids, calcium ion level, etc.

[0057] S4. We further screened out the variables that affect the outcome (whether or not the patient is ill) through logistic regression and constructed a clinical prediction model. The predictive ability of the clinical model was determined by the area under the curve. Finally, five independent predictive factors were screened out (Table 2): albumin level, ascending aorta diameter, mean platelet volume, pulmonary artery width, and right heart output. See the forest plot for details ( Figure 3 ).

[0058] The probability that the sample suffers from pulmonary hypertension is p, logit(p) = -16.217 - 0.303×RCO+ 0.327×PA + 0.23×Alb + 0.785×PltVol- 0.285×AA; Where logit(p) is the value of the logistic function; RCO is right cardiac output; PA is pulmonary artery width; Alb is albumin level; PltVol is mean platelet volume; AA is the ascending aorta diameter. The model is visualized by a nomogram ( Figure 4 ).

[0059] Table 2. Five independent predictor variables selected by binary logistic regression

[0060] S5. Evaluate the clinical prediction ability, net benefit, and accuracy of the prediction model through ROC, DCA, and calibration curves. After the prediction model is constructed, evaluate the model through multi-dimensional parameter calculation. The area under the constructed ROC curve (ReceiverOperating Characteristic curve) is: 0.92 ( Figure 5 A in the middle), indicating that the prediction model has good accuracy. Figure 5 Middle B) shows that when the probability of patients suffering from pulmonary hypertension is in the range of 0-1, the model has a good net benefit. The prediction model was internally validated by computer simulation repeated sampling method (Bootstrap method). After repeated sampling 300 times, the calibration curve ( Figure 5 (C) The prediction effect of the model is evaluated. As can be seen in the figure, the calibration curve follows y=x, and the mean absolute error is 0.049, indicating that the predicted probability of the model is highly consistent with the actual observed probability.

[0061] S6. The model is verified by random forest and specifically displayed by SHAP. In this embodiment, the constructed model is verified by random forest and intuitively described by SHAP. Figure 6 A) and importance ranking diagram ( Figure 6 From Figure B, we can see that the variables that affect the outcome of pulmonary hypertension are ranked from largest to smallest: pulmonary artery width, right atrial size, E / E', mean platelet volume, left ventricular posterior wall thickness, and pulmonary valve velocity. According to the SHAP heat map ( Figure 6 Middle C) It can be seen that variables such as pulmonary artery width, right atrial size, and left ventricular posterior wall thickness have a negative impact on the deviation of SHAP value from the base value, while variables such as pulmonary valve flow velocity, E / E', and mean platelet volume have a positive impact; and right atrial size and left ventricular posterior wall thickness have the greatest impact on SHAP value ( Figure 6 D), and the larger the pulmonary artery width, the lower the SHAP value ( Figure 6 Middle E).

[0062] S7. The model is verified by 10 machine learning methods: decision tree, gradient boosting tree, LightGBM, Logistic classifier, random forest, naive Bayes, CatBoost, XGBoost, support vector machine, and classification perceptron. In this embodiment, the test set ratio is set to 0.3, and the random number 1 is used to split the training test set. The model is verified by 10 machine learning methods: decision tree, gradient boosting tree, LightGBM, Logistic classifier, random forest, naive Bayes, CatBoost, XGBoost, support vector machine, and classification perceptron. In this embodiment, the test set ratio is set to 0.3, and the random number 1 is used to split the training test set. The model is verified by 10 machine learning methods: decision tree, gradient boosting tree, LightGBM, Logistic classifier, random forest, naive Bayes, CatBoost, XGBoost, support vector machine, and classification perceptron. The constructed model was validated multiple times using 10 machine learning methods including Trees (Table 3). The results are shown in the figure. The model has good sensitivity and specificity under various machine learning methods ( Figure 7 The average area under the ROC curve of the 10 machine learning validation models is 0.879 ( Figure 7 Middle B), precision and completeness curve (PR curve) ( Figure 7 C) shows that this model has good precision and recall under 11 machine learning methods (Table 3), and the DCA curve ( Figure 7Middle D) shows that the application of this model has good net benefits. In summary, the verification results show that the pulmonary hypertension diagnosis and prediction model constructed by the present invention has good predictive ability.

[0063] Table 3. Validation of the model by various machine learning methods

[0064] Example 2: A prediction model for judging the severity of pulmonary hypertension was constructed based on pulmonary artery angiography and test information of a pulmonary hypertension case group.

[0065] S1. Extract and analyze the information of the pulmonary hypertension case group separately, extract the information of the pulmonary hypertension case group (n=59) sorted out in step S2 of Example 1, and divide the case group into mild pulmonary hypertension (20 mmHg<mPAP≤40 mmHg), moderate pulmonary hypertension (40 mmHg<mPAP≤55 mmHg), and severe pulmonary hypertension (55 mmHg<mPAP) according to the international pulmonary hypertension severity classification standard and the mean pulmonary artery pressure (mPAP) level; S2. The variables collected from pulmonary angiography and test data were used as the first part of the independent variables, and the following statistics were used as the second part of the independent variables: basal metabolic rate, height, weight, BMI (height-weight index), heart rate, symptom duration (days), shortness of breath after activity (binary variable), chest tightness (binary variable), dyspnea (binary variable), palpitation (binary variable), decreased activity tolerance (binary variable), dizziness (binary variable), fatigue (binary variable), edema of both lower limbs (binary variable), cough (binary variable), etc. were used as the second part of the independent variables. The first and second parts of the independent variables together constitute the independent variables of the entire sample. The outcome variable was used as the dependent variable (mild, moderate, and severe pulmonary hypertension). The variables that were correlated with the outcome variable were screened out through Spearman correlation analysis. p < 0.05 was considered to be correlated. Finally, 9 variables were obtained (the corresponding units of the variables are the same as in the table), as shown in Table 4.

[0066] Table 4. Variables associated with severity of pulmonary hypertension

[0067] S3. The 9 obtained variables were included in the Lasso polynomial regression analysis to screen out potential predictive variables ( Figure 8 A, B, C, D in the table), and the above variables were included in the ordered three-category Logistic regression model. According to p < 0.05, the prediction model constructed by the following five independent predictor variables (the corresponding units of the variables are in the same table) was finally obtained (Table 5): hemoglobin content, fibrinogen content, right ventricular enlargement, left ventricular enlargement, and decreased activity tolerance.

[0068] Table 5. Independent predictor variables and model parameters selected by ordered three-category logistic regression

[0069] In the fitting model, different values ​​are assigned to different degrees of pulmonary hypertension: mild pulmonary hypertension is assigned a value of 1, moderate pulmonary hypertension is assigned a value of 2, and severe pulmonary hypertension is assigned a value of 3; the formula of the fitting model is: logit(p 轻 )=-6.667-(0.028×Hb-0.012×FBC-1.577×RVE-3.860×LVE-2.392×DAT), Logit(p 轻 +p 中 )=-4.796-(0.028×Hb-0.012×FBC-1.577×RVE-3.860×LVE-2.392×DAT), p 重 =1-p 轻 -p 中 ; Among them, the probability that the sample is mild pulmonary hypertension is p 轻 , the probability that the sample has moderate pulmonary hypertension is p 中 , the probability that the sample is severe pulmonary hypertension is p 重 ; Hb is the hemoglobin content, FBC is the fibrinogen content, RVE is the right ventricular enlargement, LVE is the left ventricular enlargement, and DAT is the decreased activity tolerance.

[0070] According to the estimated value coefficient, the hemoglobin content is a positive coefficient, indicating that the higher the hemoglobin content, the more severe the pulmonary hypertension of the sample may be; while the logistic regression coefficients of fibrinogen content, right ventricular enlargement, left ventricular enlargement, and decreased activity tolerance are negative, indicating that the lower the fibrinogen content, the more severe the pulmonary hypertension of the sample may be if there is right ventricular enlargement, left ventricular enlargement, and decreased activity tolerance.

[0071] S4, heat map of the multi-classification logistic regression matrix ( Fig. 9 A) and ROC curve ( Fig. 9 (B) It can be seen that the area under the ROC curve for predicting mild pulmonary hypertension is 0.842, the area under the ROC curve for predicting moderate pulmonary hypertension is 0.55, and the area under the ROC curve for predicting severe pulmonary hypertension is 0.918, indicating that the model has relatively weak predictive ability for moderate pulmonary hypertension, but has good predictive ability for both mild and severe pulmonary hypertension.

[0072] Example 3: Construction of linear regression model S1. Among the variables counted in Example 2, all continuous variables were analyzed for correlation with mean pulmonary artery pressure, and six variables with high correlation were selected based on p < 0.05 and correlation coefficient r > 0.3: pulmonary artery width ( Fig.10 A), hemoglobin content ( Fig.10 Middle B), red blood cell level ( Fig.10 Middle C), hematocrit ( Fig.10 D), height and weight index ( Fig.10 E), fibrinogen content ( Fig.10 Middle F).

[0073] S2. The above six variables were subjected to stepwise regression (both) - linear regression. According to p < 0.05, the two variables PA and RBC were finally screened out, and the regression equation was constructed: y = -15.949 + 1.123× PA + 8.808× RBC + ε, where y is the mean pulmonary artery pressure level, PA is the pulmonary artery width, and RBC is the red blood cell level.

Claims

1. A method for constructing a diagnostic prediction model for pulmonary hypertension based on pulmonary artery angiography, characterized in that: The steps include: S1: collecting the age, gender, pulmonary artery angiography parameter information and test information of the case group and the control group, using the age, gender, pulmonary artery angiography parameter information and test information of the case group and the control group as variables, constructing samples, and obtaining the first sample set; S2: Remove variables and samples with too many missing values ​​or abnormal data from the first sample set to obtain the second sample set; S3: Multiple potential predictor variables were screened out in the second sample set through Lasso regression, and all the predictor variables of the case group and the control group constituted the sample to form the third sample set; S4: The key variables that affect the outcome are screened out in the third sample set by binary logistic regression and a prediction model is constructed. The outcome refers to whether or not pulmonary hypertension is present. The key variables that affect the outcome are the several independent predictive variables that are finally screened out. All the key variables of the case group and the control group constitute a sample to form a fourth sample set. The key variables of the fourth sample set are statistically analyzed to obtain model parameters, which are visualized by a nomogram. S5: The clinical prediction ability of the prediction model was evaluated by the receiver characteristic curve, decision curve, and calibration curve; S6: The model is validated by random forest and specifically displayed by SHAP; S7: The prediction model is verified by 10 machine learning methods: decision tree, gradient boosting tree, lightweight gradient boosting machine LightGBM, logistic classifier Logistic classifier, random forest, naive Bayes, classification boosting CatBoost, extreme gradient boosting XGBoost, support vector machine and classification perceptron.

2. The construction method according to claim 1, characterized in that: The step S2 specifically includes: removing variables and samples with missing values ​​greater than 30% in the sample, interpolating the remaining variables with missing values ​​using the KMN method, and finally screening useful variables for inclusion in subsequent analysis.

3. The construction method according to claim 2, characterized in that: In step S3: A total of 21 potential predictor variables were screened out through Lasso regression analysis, namely: right cardiac output, pulmonary artery width, ascending aorta width, right ventricular enlargement, right atrium enlargement, absolute neutrophil value, carbon dioxide concentration, lipoprotein level, total bile acid level, direct bilirubin level, albumin level, creatinine level, age, gender, phosphorus level, high-density lipoprotein level, mean platelet volume, atrial septal defect, ventricular septal defect, phospholipid level, and calcium ion level.

4. The construction method according to claim 3, characterized in that: In step S4: The key variables that affect the outcome are the five independent predictive variables that were finally screened out, namely: albumin level, ascending aorta width, mean platelet volume, pulmonary artery width, and right ventricular output; The formula of the prediction model obtained by the binary logistic regression method is: the probability of a patient having pulmonary hypertension is p, logit(p) = -16.217 - 0.303×RCO+ 0.327×PA + 0.23×Alb + 0.785×PltVol-0.285×AA; where logit(p) is the value of the logistic function; RCO is right cardiac output; PA is pulmonary artery width; Alb is albumin level; PltVol is mean platelet volume; and AA is the inner diameter of the ascending aorta.

5. The construction method according to claim 4, characterized in that: In the formula of the prediction model of step S4: Albumin level: OR value 1.258, 95% CI 1.087-1.499, p = 0.005; Ascending aorta diameter: OR value 0.752, 95%CI 0.641-0.853, p<0.001; Mean platelet volume: OR value 2.193, 95%CI 1.359-3.856, p=0.003; Pulmonary artery width: OR value 1.387, 95%CI 1.204-1.68, p<0.001; Right cardiac output: OR value 0.739, 95%CI 0.585-0.903, p=0.

005.

6. A diagnostic prediction model for pulmonary hypertension based on pulmonary artery angiography, characterized in that: The prediction model is constructed by the method described in claim 5; the prediction model is visualized by a nomogram, which includes a score scale, a total score scale, and predictive variables; the score scale is 0-100 points, the total score scale is 0-240 points, and the predictive variables are: albumin level, ascending aorta width, mean platelet volume, pulmonary artery width, and right heart output.

7. A prediction model for judging the severity of pulmonary hypertension based on pulmonary artery angiography, characterized by: The prediction model is constructed by multinomial Lasso regression and ordered multi-classification Logistic regression analysis, and its independent prediction variables are: right ventricular enlargement, left ventricular enlargement, hemoglobin content, atrial septal defect, and fibrinogen content; the prediction model is expressed as follows: in the fitting model, different degrees of pulmonary hypertension are given different values: mild pulmonary hypertension is assigned a value of 1, moderate pulmonary hypertension is assigned a value of 2, and severe pulmonary hypertension is assigned a value of 3; the formula of the fitting model is: logit(p 轻 )=-6.667-(0.028×Hb-0.012×FBC-1.577×RVE-3.860×LVE-2.392×DAT), Logit(p 轻 +p 中 )=-4.796-(0.028×Hb-0.012×FBC-1.577×RVE-3.860×LVE-2.392×DAT), p 重 =1-p 轻 -p 中 ; Among them, the probability that the sample is mild pulmonary hypertension is p 轻 , the probability that the sample has moderate pulmonary hypertension is p 中 , the probability that the sample is severe pulmonary hypertension is p 重 ; Hb is the hemoglobin content, FBC is the fibrinogen content, RVE is the right ventricular enlargement, LVE is the left ventricular enlargement, and DAT is the decreased activity tolerance.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for constructing a prediction model as described in any one of claims 1 to 5 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for constructing a prediction model as claimed in any one of claims 1 to 5 is implemented.

Images