A peritoneal dialysis patient cardiovascular complication prediction model, method and system based on glycation proteomics and machine learning

Abstract

This application discloses a predictive model, method, and system for cardiovascular complications in peritoneal dialysis patients based on glycated proteomics and machine learning. This application utilizes techniques such as glycated peptide enrichment, high-throughput glycated proteomics detection, targeted absolute quantification, and machine learning modeling to construct a predictive model suitable for assessing the risk of CVD complications in peritoneal dialysis patients. This model organically integrates glycated peptide biomarkers with clinical indicators, enabling early identification and stratified management of cardiovascular risk in peritoneal dialysis patients, and has good clinical application potential.

Description

Technical Field

[0001] This application relates to the fields of biomedical and disease diagnosis technology, specifically to a model, method, and system for predicting cardiovascular complications in peritoneal dialysis patients based on glycosylated proteomics and machine learning. Background Technology

[0002] Cardiovascular disease (CVD) has long been one of the leading complications and causes of death for dialysis patients. The 2024 annual report from the American Kidney Data System showed that over half of dialysis patient deaths were related to CVD, with arrhythmias, cardiac arrest, or sudden cardiac death being the primary causes. Although traditional CVD risk factors are prevalent in dialysis patients, the dialysis treatment itself and the complex pathophysiological changes caused by renal function loss further increase the risk of CVD. Therefore, the high incidence of cardiovascular complications in peritoneal dialysis patients has become a crucial clinical problem that urgently needs to be addressed. Currently, several biomarkers have been reported to be closely related to CVD complications in peritoneal dialysis patients, such as cardiac troponin, N-terminal pro-B-type natriuretic peptide, soluble matrix lysin-2, and angiopoietin-2, and some of these biomarkers have been applied in clinical practice. However, existing biomarkers generally suffer from limited sensitivity and specificity and are easily affected by renal function and dialysis status, making it difficult to meet the need for early and accurate prediction of CVD complications in peritoneal dialysis patients.

[0003] Protein glycation is a non-enzymatic pathological modification, referring to a series of non-enzymatic reactions between reducing sugars and protein amino groups, commonly seen in states such as high glucose exposure, aging, and metabolic disorders. This process typically consists of two stages: the first stage is the reversible formation of early glycation products (AGEs), and the second stage is the irreversible formation of advanced glycation end products (AGEs). In patients with chronic kidney disease (CHD), elevated AGE levels are associated with increased cardiovascular risk and mortality, even in the absence of significant hyperglycemia, consistent with their pro-inflammatory and pro-oxidative stress biological effects. However, AGEs are late, irreversible products of glycation, primarily reflecting long-term metabolic stress, thus limiting their application in early disease risk assessment. In contrast, early glycation products are more likely to reflect metabolic abnormalities in the early stages of disease development and are more suitable as early predictive biomarkers for cardiovascular disease (CVD). Existing research suggests that early glycation products derived from proteins such as albumin may serve as potential biomarkers for diabetic vascular complications. For peritoneal dialysis patients, the continuous high glucose exposure from the peritoneal dialysis fluid can further exacerbate protein glycation processes in the body, potentially promoting the occurrence and development of CVD. Although various CVD prediction models exist, those specifically designed for peritoneal dialysis patients remain limited, and most models only incorporate traditional clinical indicators, lacking disease-specific biomarkers.

[0004] Therefore, it is urgent to combine glycated proteomics and machine learning technologies to integrate glycated peptide biomarkers associated with CVD in peritoneal dialysis patients with traditional clinical risk factors, construct a CVD risk prediction model for peritoneal dialysis patients, and validate it in multiple patient cohorts, thereby achieving early risk identification and accurate assessment of cardiovascular complications in peritoneal dialysis patients. Summary of the Invention

[0005] The purpose of this application is to address the shortcomings of existing technologies by providing an early risk prediction model, method, and system for cardiovascular complications in peritoneal dialysis patients, so as to achieve early identification and stratified management of the risk of cardiovascular complications in peritoneal dialysis patients.

[0006] To achieve the above objectives, this application adopts the following technical solution:

[0007] In a first aspect, this application provides a method for constructing a predictive model for cardiovascular complications in peritoneal dialysis patients based on glycosylated proteomics and machine learning, including: S1: Collect clinical samples to obtain screening and modeling cohorts; S2: Screening for glycosylated peptide biomarkers in the screening queue; S3: The modeling queue is divided into a training set and a test set. In the training set, based on 56 clinical indicators, different feature selection methods and different machine learning algorithms are used to configure and evaluate the model performance. The model configuration of F-test feature selection + ENetLR classifier is determined to be the best. The clinical indicators included in the ENetLR classifier are determined to be the six clinical indicators to be included in the subsequent prediction model construction, specifically including: serum phosphorus, history of diabetes, total Kt / V, albumin, cardiac troponin T, and glycated albumin; among them, history of diabetes is a categorical variable, with a value of 1 for yes and 0 for no. S4: Use the test set to evaluate the predictive performance of the ENetLR classifier model constructed by incorporating glycated peptide biomarkers and / or six clinical indicators, determine the ENetLR classifier model constructed by combining glycated peptide biomarkers and six clinical indicators as the optimal predictive model, and verify the predictive performance of the optimal predictive model in the test set.

[0008] In some embodiments, the screening process for glycated peptide biomarkers in S2 includes: collecting serum samples, enriching glycated peptides, detecting glycated peptides using high-throughput glycated proteomics, screening differentially glycated peptides using the glmnet model, and finally selecting 55 glycated peptides by the glmnet model. Based on the importance of the features, the top two peptides are selected as candidate biomarkers, namely GEAFTLK(g)ATVLNYLPK derived from α-2-macroglobulin and K(g)QHLFVK derived from apolipoprotein B-100.

[0009] In some embodiments, the 56 clinical indicators in S3 include: gender, age, dialysis duration, height, weight, hypertension, history of diabetes, history of cardiovascular disease, history of antiplatelet therapy, history of anemia medication, history of lipid-lowering therapy, history of phosphorus-lowering therapy, blood pressure, red blood cell count, hemoglobin, platelet count, white blood cell count, neutrophil count, total bilirubin, direct bilirubin, total protein, albumin, fasting blood glucose, glycated hemoglobin, glycated albumin, alanine aminotransferase, aspartate aminotransferase, lactate dehydrogenase, serum creatinine, blood urea nitrogen, uric acid, estimated glomerular filtration rate, triglycerides, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, calcium, phosphorus, serum iron, transferrin saturation, ferritin, C-reactive protein, whole-segment parathyroid hormone, N... Terminal pro-B-type natriuretic peptide, cardiac troponin T, D-dimer, urine output, peritoneal Kt / V, renal Kt / V, total Kt / V, peritoneal creatinine clearance, renal creatinine clearance, total creatinine clearance, intima-media thickness of left and right carotid arteries, left ventricular ejection fraction, and left ventricular end-diastolic diameter.

[0010] In some implementations, in S3, the feature selection method is F-test (F), mutual information (MI), L1 regularization-based feature selection (L1LR), tree-based feature importance (L1SVC), or recursive feature elimination (FRE); the classification model used for modeling is logistic regression (LR), Elastic-Net Logistic Regression (ENetLR), Radial Basis Function Kernel Support Vector Machine (RBF-SVM), Random Forest (RF), Histogram-based Gradient Boosting Model (HGB), or Extreme Gradient Boosting Tree (XGBoost). Each classification model undergoes hyperparameter tuning, using grid search combined with cross-validation to determine the optimal parameters; each model configuration consists of a feature selection method and a classification model, or only a classification model.

[0011] In some implementations, in S3, the evaluation metrics for model performance include at least one of ROC-AUC, F1 score, balanced accuracy, PR-AUC, precision, and recall.

[0012] In some implementations, in S4, the evaluation and validation metrics for the prediction performance include at least one of ROC-AUC, F1 score, balanced accuracy, PR-AUC, precision, and recall.

[0013] Secondly, this application provides a method for predicting cardiovascular complications in peritoneal dialysis patients based on a prediction model constructed using the construction method described in the first aspect, comprising the following steps: Acquire clinical indicators and glycated peptide marker levels in peritoneal dialysis patients. The clinical data include serum phosphorus, history of diabetes, total Kt / V, albumin, cardiac troponin T, and glycated albumin. The glycated peptide markers are GEAFTLK(g)ATVLNYLPK and K(g)QHLFVK. The levels of the clinical indicators and glycated peptide markers are input into the prediction model, and the predicted probabilities are output.

[0014] Thirdly, this application provides a system for predicting cardiovascular complications in peritoneal dialysis patients, including: The data acquisition unit is used to acquire the clinical indicators and glycated peptide markers of peritoneal dialysis patients. The clinical data includes serum phosphorus, history of diabetes, total Kt / V, albumin, cardiac troponin T and glycated albumin. The glycated peptide markers are GEAFTLK(g)ATVLNYLPK and K(g)QHLFVK. A prediction unit is used to input the content of the clinical indicators and glycated peptide markers into a pre-constructed prediction model and output the prediction results; the prediction model is a prediction model constructed according to the construction method described in the first aspect.

[0015] Compared with the prior art, this application has the following beneficial effects: (1) Based on the screening and verification of early glycated peptides, this application has for the first time achieved earlier identification and risk assessment of CVD complications in peritoneal dialysis patients, which can provide a basis for early clinical intervention; (2) This application combines glycated peptide markers with clinical indicators and completes feature screening and model construction through machine learning. It has been validated in multiple patient cohorts, making the resulting model both biologically specific and clinically accessible, thus forming a more targeted, accurate and clinically valuable CVD risk prediction model for peritoneal dialysis patients. Attached Figure Description

[0016] Figure 1 Performance evaluation results for 34 model configurations.

[0017] Figure 2 Bar charts sorting features based on average absolute SHAP values.

[0018] Figure 3 Performance comparison of different models (biomarker model, clinical model and combined model) in a single-center training set.

[0019] Figure 4 Performance comparison of different models (biomarker models, clinical models, and combined models) in single-center test sets and external multi-center validation cohorts. Detailed Implementation

[0020] To make the technical solution of this application clearer and easier to understand, preferred embodiments are described in detail below with reference to the accompanying drawings.

[0021] Unless otherwise stated, all experimental materials (including but not limited to reagents and test kits) involved in the following examples are commercially available standards or reagents.

[0022] Example This embodiment first screens out glycated peptide biomarkers associated with CVD complications in peritoneal dialysis patients. Then, based on the screened glycated peptide biomarkers and clinical indicators, a predictive model is constructed and validated. The specific steps are as follows: 1. Patient cohort inclusion: This study included three cohorts of peritoneal dialysis patients. The inclusion criteria were as follows: (1) age 18-80 years; (2) receiving regular peritoneal dialysis treatment for more than 6 months. The exclusion criteria were as follows: (1) having undergone kidney transplantation before dialysis; (2) receiving combined peritoneal dialysis and hemodialysis treatment; (3) having malignant tumors or autoimmune diseases; (4) having chronic infections or infections occurring within 3 months prior to enrollment.

[0023] (1) Screening cohort: 48 pairs of peritoneal dialysis patients from the Peritoneal Dialysis Center of Zhongshan Hospital Affiliated to Fudan University were included. The two groups of patients were matched in terms of gender, age and dialysis duration. Based on whether CVD events occurred within 6 months prior to serum sample collection, patients were divided into a CVD group and a non-CVD group.

[0024] (2) Single-center modeling cohort: 301 peritoneal dialysis patients from the peritoneal dialysis center of Zhongshan Hospital affiliated to Fudan University were included. None of the patients had CVD events in the 6 months prior to enrollment and were followed up for 1 year. The occurrence of CVD was used as the endpoint of the follow-up.

[0025] (3) Multicenter validation cohort: 113 peritoneal dialysis patients were included, from Renji Hospital affiliated to Shanghai Jiao Tong University School of Medicine, Minhang Hospital affiliated to Fudan University, and Wusong Central Hospital of Baoshan District. All patients had no CVD events in the 6 months prior to enrollment and were followed up for 1 year, with CVD occurrence as the endpoint of follow-up.

[0026] 2. Serum sample processing, enrichment, and desalting: ① Collect serum samples from all cohorts, add 1.11 μL of 75 mmol / L sodium cyanoborohydride solution to 10 μL of serum sample, and react at 37℃ and 1100 rpm for 4 hours to reduce the protein. ② Take 500 μg of the reduced protein solution into a 10 kDa ultrafiltration tube, remove excess sodium cyanoborohydride solution with ammonium bicarbonate solution, then add 4.04 L of 1 mol / L dithiothreitol and react at 56 °C for 1 hour; then add 8.08 L of 1 mol / L iodoacetamide solution and react at room temperature and in the dark for 45 minutes; after the above reduction alkylation reaction is completed, remove unreacted reagents by ultrafiltration. ③ Add 0.5 μL of peptidase N-glycosidase F and react at 37°C and 150 rpm for 16 hours to remove N-glycan chains. After the reaction is complete, collect the protein filtrate. ④ Add trypsin to the protein filtrate at a mass ratio of 50:1 (serum protein:trypsin), and enzymatically hydrolyze for 16 hours at 37 ℃ and 1100 rpm. Collect the filtrate after the enzymatic hydrolysis is complete.

[0027] ⑤ Add the synthesized relabeled glycosylated peptide as a standard to the protease hydrolysate.

[0028] ⑥ Add the collected filtrate to 20 mg of boric acid material that has been balanced and activated, and react for 16 hours at 37 °C and 1100 rpm. ⑦ After saccharification and enrichment are completed, the supernatant is added to a Sep-Pak C18 solid-phase extraction column for desalting; after loading the sample 3 times and washing 2 times, the desalted eluent is collected and lyophilized for storage.

[0029] 3. Screening of glycated peptide biomarkers: Screening for glycated peptide biomarkers was conducted in a screening cohort comprising 48 matched peritoneal dialysis patients, including CVD and non-CVD groups, to identify CVD-related glycated peptide biomarkers. A PRM-LC-MS / MS targeted detection system was established using an OrbitrapExploris 480 high-resolution mass spectrometer to perform high-throughput glycated proteomics analysis, enabling qualitative identification and characterization of glycated peptides. Compared to the non-CVD group, 237 glycated peptides were significantly upregulated and 20 were significantly downregulated in the CVD group. The glmnet model was used to screen differentially glycated peptides, ultimately identifying 55 glycated peptides. Based on feature importance, the top two peptides were selected as candidate biomarkers: GEAFTLK(g)ATVLNYLPK derived from α-2-macroglobulin and K(g)QHLFVK derived from apolipoprotein B-100.

[0030] 4. Targeted quantification of glycated peptide biomarkers: After reconstitution of the lyophilized samples to a concentration of 0.5 g / L, PRM-LC–MS / MS analysis was performed using an Orbitrap Exploris 480 high-resolution mass spectrometer. Peptides were separated on a 40 cm column. The acquisition range was m / z 350–1600, with a resolution of 60,000 (m / z 200). HCD fragmentation (normalized collision energy 29%) was used, with an MS / MS resolution of 60,000 (m / z 200), an AGC target of 75%, and a maximum injection time of 120 ms. The optimal parent-daughter ion pair selection is shown in Table 1. The obtained data were imported into Skyline Daily, and the concentration of the target glycosylated peptide was calculated by the peak area ratio of the target peptide to the corresponding internal standard.

[0031] Table 1. Optimal mother-daughter ion pairs

[0032] 5. Construction of the prediction model: System benchmark tests were performed on 56 clinical indicators with 34 model configurations (feature selection + classifier combination) in the training set of a single-center modeling cohort.

[0033] To identify clinical predictive features and construct CVD prediction models, patients in a single-center modeling cohort were randomly divided into training and testing sets in a 7:3 ratio. Model development and feature selection were performed on the training set, followed by model performance evaluation on the testing set. To systematically compare candidate modeling strategies, we evaluated 34 model configurations based on 56 clinical indicators from the training set within a five-fold cross-validation framework. The 56 clinical indicators included: gender (male = 1, female = 0), age, dialysis duration, height, weight, hypertension, history of diabetes (yes = 1, no = 0), history of cardiovascular disease (yes = 1, no = 0), history of antiplatelet therapy (yes = 1, no = 0), history of anemia medication (yes = 1, no = 0), and history of lipid-lowering medication (yes = 1, no = 0). Value 0), history of phosphorus-lowering drug use (1 for yes, 0 for no), blood pressure, red blood cell count, hemoglobin, platelet count, white blood cell count, neutrophil count, total bilirubin, direct bilirubin, total protein, albumin, fasting blood glucose, glycated hemoglobin, glycated albumin, alanine aminotransferase, aspartate aminotransferase, lactate dehydrogenase, serum creatinine, blood urea nitrogen, uric acid, estimated glomerular filtration rate, triglycerides, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, calcium, phosphorus, serum iron, transferrin saturation, ferritin, C Reactive protein, whole-segment parathyroid hormone, N-terminal pro-B-type natriuretic peptide, cardiac troponin T, D-dimer, urine output, peritoneal Kt / V, renal Kt / V, total Kt / V, peritoneal creatinine clearance, renal creatinine clearance, total creatinine clearance, intima-media thickness of left and right carotid arteries, left ventricular ejection fraction, and left ventricular end-diastolic diameter. The 34 model configurations include the following feature selection methods: F-test (F), mutual information (MI), L1 regularization-based feature selection (L1LR), tree-based feature importance (L1SVC), and recursive feature elimination (FRE). The included classifiers are Logistic Regression (LR), Elastic-Net Logistic Regression (ENetLR), Radial Basis Function Kernel Support Vector Machine (RBF-SVM), Random Forest (RF), Histogram-based Gradient Boosting (HGB), and XGBoost. Each classifier algorithm undergoes hyperparameter tuning, using grid search combined with cross-validation to determine the optimal parameters. Each model configuration consists of one feature selection method and one classifier, or only a classifier, with specific combinations as follows: Figure 1As shown in the figure, None indicates that the model configuration has not undergone feature screening, and all 56 clinical indicators are included in the classifier model.

[0034] Based on benchmark comparisons and model simplicity, the number of clinical features retained in subsequent model development was fixed at six. Model performance was evaluated using ROC-AUC, F1 score, balanced accuracy, PR-AUC, precision, and recall. Ultimately, the best-performing model was selected as the final predictive model, such as... Figure 1 As shown, the model configuration combining F-test feature selection and the ENetLR classifier performed best. In the F-test feature selection, patients were divided into CVD and non-CVD groups based on CVD occurrence. One-way ANOVA was performed on each candidate clinical indicator to calculate its F-statistic and corresponding P-value. Subsequently, a Top-k strategy based on a pre-defined number of features to retain was used to select features for subsequent model construction. The final ENetLR classifier model incorporated six clinical indicators: serum phosphorus, diabetes, total Kt / V, albumin, cardiac troponin T, and glycated albumin.

[0035] The clinical indicators and screened glycated peptides included in the classifier model were ranked according to their feature importance based on their shap value, such as... Figure 2 As shown. To evaluate model performance, three ENetLR classifier models incorporating different indicators were compared in a single-center test set: (1) a biomarker model using only glycated peptide biomarkers; (2) a clinical model using only the six selected clinical indicators; and (3) a combined model integrating clinical indicators and glycated peptide biomarkers. Specifically, the elastic network logistic regression model uses whether cardiovascular complications occur during the 1-year follow-up of peritoneal dialysis patients as a binary outcome variable, and uses the indicators (glycated peptides, 6 clinical indicators, glycated peptides + 6 clinical indicators) in (1)-(3) as input features; the elastic network logistic regression model also includes L1 regularization and L2 regularization terms, the regularization mixing parameter α is set to 0.5, and the regularization strength parameter λ is determined by five-fold cross-validation and grid search, with a candidate range of λ ∈ {0.001, 0.01, 0.1, 1, 10}, the model uses binary cross-entropy as the loss function, and uses the coordinate descent method to complete the parameter solution optimization, and outputs the predicted probability of peritoneal dialysis patients developing cardiovascular complications. Model performance was evaluated using multiple complementary metrics, including specificity, sensitivity, ROC-AUC, PR-AUC, F1 score, balanced accuracy, and precision. Results showed that the joint model exhibited the best predictive performance, with an AUC of 0.78, precision of 0.75, and specificity of 0.71. Figure 3 As shown.

[0036] 6. Validation of the prediction model: In single-center and external multi-center validation sets, the combined model was compared with biomarker-only and clinical indicator-only models. The results showed that the combined model performed best in both, with AUC values ​​of 0.71 and 0.76, respectively. Figure 4 As shown.

[0037] The above description is merely a preferred embodiment of this application and is not intended to limit this application in any form or substance. It should be noted that those skilled in the art can make several improvements and additions without departing from this application, and these improvements and additions should also be considered within the scope of protection of this application.

Claims

1. A method for constructing a cardiovascular complication prediction model for peritoneal dialysis patients based on glycation proteomics and machine learning, characterized by, include: S1: Collect clinical samples to obtain screening and modeling cohorts; S2: Screening for glycosylated peptide biomarkers in the screening queue; S3: The modeling queue is divided into a training set and a test set. In the training set, based on 56 clinical indicators, different feature selection methods and different machine learning algorithms are used to configure and evaluate the model performance. The model configuration of F-test feature selection + ENetLR classifier is determined to be the best. The clinical indicators included in the ENetLR classifier are determined to be the six clinical indicators to be included in the subsequent prediction model construction, specifically including: serum phosphorus, history of diabetes, total Kt / V, albumin, cardiac troponin T, and glycated albumin; among them, history of diabetes is a categorical variable, with a value of 1 for yes and 0 for no. S4: Use the test set to evaluate the predictive performance of the ENetLR classifier model constructed by incorporating glycated peptide biomarkers and / or six clinical indicators, determine the ENetLR classifier model constructed by combining glycated peptide biomarkers and six clinical indicators as the optimal predictive model, and verify the predictive performance of the optimal predictive model in the test set.

2. The construction method according to claim 1, characterized in that, In S2, the screening process for glycated peptide biomarkers includes: collecting serum samples, enriching glycated peptides, detecting glycated peptides using high-throughput glycated proteomics, and using the glmnet model to screen differentially glycated peptides. Finally, the glmnet model screened out 55 glycated peptides. Based on the importance of the features, the top two peptides were selected as candidate biomarkers, namely GEAFTLK(g)ATVLNYLPK derived from α-2-macroglobulin and K(g)QHLFVK derived from apolipoprotein B-100.

3. The construction method of claim 1, wherein, The S3 specification includes 56 clinical indicators, including: gender, age, dialysis duration, height, weight, hypertension, history of diabetes, history of cardiovascular disease, history of antiplatelet therapy, history of anemia medication, history of lipid-lowering therapy, history of phosphorus-lowering therapy, blood pressure, red blood cell count, hemoglobin, platelet count, white blood cell count, neutrophil count, total bilirubin, direct bilirubin, total protein, albumin, fasting blood glucose, glycated hemoglobin, glycated albumin, alanine aminotransferase, aspartate aminotransferase, lactate dehydrogenase, serum creatinine, blood urea nitrogen, uric acid, estimated glomerular filtration rate, triglycerides, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, calcium, phosphorus, serum iron, transferrin saturation, ferritin, C-reactive protein, whole-segment parathyroid hormone, N-terminal B12, and total bilirubin levels. Type I natriuretic peptide propionidate, cardiac troponin T, D-dimer, urine output, peritoneal Kt / V, renal Kt / V, total Kt / V, peritoneal creatinine clearance, renal creatinine clearance, total creatinine clearance, intima-media thickness of left and right carotid arteries, left ventricular ejection fraction, and left ventricular end-diastolic diameter.

4. The construction method of claim 1, wherein, In S3, the feature selection method is F-test (F), mutual information (MI), L1 regularization-based feature selection (L1LR), tree-based feature importance (L1SVC), or recursive feature elimination (FRE); the classification model used for modeling is logistic regression (LR), Elastic-Net Logistic Regression (ENetLR), Radial Basis Function Kernel Support Vector Machine (RBF-SVM), Random Forest (RF), Histogram-based Gradient Boosting (HGB), or XGBoost. Each classification model undergoes hyperparameter tuning, and the optimal parameters are determined using grid search combined with cross-validation; each model configuration consists of one feature selection method and one classification model, or only a classification model.

5. The construction method of claim 1, wherein, In S3, the evaluation metrics for model performance include at least one of ROC-AUC, F1 score, balanced accuracy, PR-AUC, precision, and recall.

6. The construction method of claim 1, wherein, In step S4, the evaluation and validation metrics for the prediction performance include at least one of ROC-AUC, F1 score, balanced accuracy, PR-AUC, precision, and recall.

7. A method for predicting cardiovascular complications in peritoneal dialysis patients using a prediction model constructed based on the construction method described in any one of claims 1-6, characterized in that, Includes the following steps: Acquire clinical indicators and glycated peptide marker levels in peritoneal dialysis patients. The clinical data include serum phosphorus, history of diabetes, total Kt / V, albumin, cardiac troponin T, and glycated albumin. The glycated peptide markers are GEAFTLK(g)ATVLNYLPK and K(g)QHLFVK. The levels of the clinical indicators and glycated peptide markers are input into the prediction model, and the predicted probabilities are output.

8. A system for predicting cardiovascular complications in peritoneal dialysis patients, characterized in that, include: The data acquisition unit is used to acquire the clinical indicators and glycated peptide markers of peritoneal dialysis patients. The clinical data includes serum phosphorus, history of diabetes, total Kt / V, albumin, cardiac troponin T and glycated albumin. The glycated peptide markers are GEAFTLK(g)ATVLNYLPK and K(g)QHLFVK. A prediction unit is used to input the content of the clinical indicators and glycated peptide markers into a pre-constructed prediction model and output the prediction results; the prediction model is a prediction model constructed according to the construction method of any one of claims 1-6.

Images