Intelligent auxiliary diagnosis system and method for unstable angina pectoris based on biomarker RETN

Abstract

The invention discloses an intelligent auxiliary diagnosis system and method for unstable angina pectoris based on a biomarker RETN, and belongs to the technical field of biological medicine. According to the present invention, the Olink high-throughput proteomics technology screening finds that the expression of RETN and other proteins in the UA patient serum is up-regulated, and after the ELISA verification, the RETN and other proteins are combined with the clinical conventional indexes, and the machine learning algorithm is adopted to construct the multi-dimensional diagnosis system; the prediction module adopts a double-layer architecture, a base module layer parallelly runs six algorithms to capture a nonlinear relationship, a meta module layer adopts logistic regression secondary modeling, and the generalization ability is improved. The system and the method can realize accurate recognition and dynamic monitoring of UA early lesion, break through the bottleneck of traditional detection, make up for the defect of insufficient specificity, and overcome the limitation of incapability of repeated monitoring. The AUC of the diagnosis model in an independent verification queue reaches 0.842 or above, the accuracy rate exceeds 76%, UA patients can be screened, the coronary artery stenosis degree can be evaluated, and personalized treatment can be assisted.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, and more specifically to an intelligent auxiliary diagnostic system and method for unstable angina based on the biomarker RETN. Background Technology

[0002] Coronary atherosclerotic heart disease (CAD) is a heart condition caused by narrowing or blockage of the coronary arteries, leading to insufficient blood supply to the myocardium and subsequently ischemia, hypoxia, and even necrosis. This disease encompasses two main types: stable coronary artery disease and acute coronary syndrome (ACS). Unstable angina (UA) is a typical phenotype within ACS. The clinical characteristics of UA fall between stable angina and acute myocardial infarction, primarily manifesting as resting chest pain that progressively worsens, accompanied by dynamic changes in the electrocardiogram (ECG). UA carries a high risk of sudden cardiac death, with a 3-5 times higher risk compared to patients with stable angina. Furthermore, UA rapidly progresses to myocardial infarction within a short period; the rate of conversion to myocardial infarction within 48 hours of onset can reach 20%–30%. Therefore, early diagnosis and accurate risk assessment of UA are crucial for timely and effective treatment and improved patient prognosis.

[0003] Currently, the clinical diagnosis of coronary artery stenosis mainly relies on coronary CT angiography, invasive coronary angiography, and intravascular imaging techniques (such as intravascular ultrasound IVUS and optical coherence tomography OCT). While these imaging methods can visually present the morphology and degree of stenosis of the coronary arteries, they have significant limitations. In the early stages of the disease, when the vascular morphology has not yet changed significantly, these imaging examinations often fail to accurately diagnose the disease, easily leading to missed or misdiagnosis, thus delaying the patient's optimal treatment time. Furthermore, these examinations have drawbacks such as requiring highly skilled operators, high costs, and some trauma to the patient, and are not suitable for large-scale screening and dynamic monitoring. Traditional diagnostic methods also have significant shortcomings in the application of biomarkers. For example, while cardiac troponin is a specific marker of myocardial cell damage, its level is affected by various factors, such as diabetes, kidney disease, and heart failure, and does not solely reflect myocardial ischemia. Therefore, relying solely on whether cardiac troponin levels exceed the upper limit of normal to differentiate between unstable angina and myocardial infarction is very crude and cannot meet the needs of early, accurate diagnosis. At the same time, there is currently a lack of specific biomarkers that can accurately evaluate unstable or vulnerable plaques, which greatly limits the ability to accurately assess the severity of patients' conditions in the early stages.

[0004] Given the numerous challenges in the diagnosis and treatment of coronary atherosclerotic heart disease, especially unstable angina, overcoming the limitations of existing technologies, exploring biomarkers that can accurately assess the severity of a patient's condition in the early stages, and constructing corresponding diagnostic models to achieve early diagnosis, risk assessment, and the development of personalized prevention and treatment strategies have become critical technical issues that urgently need to be addressed in the cardiovascular field. Summary of the Invention

[0005] In view of the current situation where traditional imaging is difficult to accurately diagnose unstable angina in the early stages, and is prone to missed or misdiagnosis; and that traditional biomarkers have poor specificity and lack biomarkers for evaluating unstable plaques, making it impossible to accurately assess the condition in the early stages, this invention aims to provide an intelligent auxiliary diagnostic system and method for unstable angina based on the biomarker RETN.

[0006] To achieve the above objectives, the present invention employs the following technical solution: This invention provides the application of a reagent for detecting the concentration of a biomarker associated with unstable angina in the preparation of products for diagnosing unstable angina, wherein the biomarker associated with unstable angina is serum resistin RETN.

[0007] The product is a kit, chip, or monitoring system for detecting the biomarker serum resistin RETN.

[0008] This invention provides a method for constructing a predictive model for unstable angina based on the biomarker RETN, comprising the following steps: S1. Select a sample set with complete serum test data from the physical examination database and divide it into a disease group and a control group according to the clinical diagnosis results; S2, RETN concentration and 26 routine clinical indicators were extracted as candidate variables, and feature variables were selected for inclusion in the model construction through multicollinearity analysis and machine learning algorithms; S3, an ensemble learning strategy is used to construct a prediction model, which includes two stages: training of the base module model and optimization of the meta module model; S4. Evaluate the model's discriminative power, calibration, and clinical applicability in the independent validation set to complete model construction.

[0009] In step S2, LASSO regression, random forest, and support vector machine recursive feature elimination are used to jointly screen variables, and multicollinearity analysis is combined to optimize feature selection. Then, machine learning algorithms are used to screen feature variables to be included in the model construction.

[0010] The feature variables included in the model construction include age, systolic blood pressure (SBP), aspartate aminotransferase (AST), total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), cystatin C, white blood cell count (WBC), hemoglobin (HGB), and serum resistin (RETN).

[0011] The base module model employs at least one of the following algorithms: XGBoost, Random Forest (RF), Logistic Regression (LR), LightGBM, Support Vector Machine (SVM), and Gaussian Naive Bayes (GNB).

[0012] Preferably, the base module model uses logical regression (LR).

[0013] The meta-module model is constructed through a stacking integration strategy. Its input is the predicted probability output by the base module model, and its output is the risk value of unstable angina.

[0014] This invention provides an intelligent auxiliary diagnostic system for unstable angina based on the biomarker RETN, the auxiliary diagnostic system comprising: The data acquisition module is used to collect patients' serum test data, which includes RETN concentration and routine clinical indicators. The model invocation module is used to invoke the constructed prediction model; The diagnostic output module is used to input the data collected by the data acquisition module into the prediction model called by the model calling module, and to generate a risk prediction report for unstable angina.

[0015] The present invention provides a computer device, comprising: a memory and a processor, wherein the memory is used to store program instructions; the processor is used to call the program instructions, and when the program instructions are executed, the following functions are implemented: (1) calling the prediction model construction method; (2) executing the data processing flow of the intelligent auxiliary diagnostic system; (3) outputting the risk prediction result of unstable angina. The present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it performs the following functions: (1) calling the prediction model construction method; (2) executing the data processing flow of the intelligent auxiliary diagnostic system; and (3) outputting the risk prediction result of unstable angina.

[0016] Compared with the prior art, the present invention achieves the following technical effects: This invention provides the first-ever manifestation of the association between RETN protein and the pathogenesis of unstable angina and the degree of coronary artery stenosis. Utilizing the high specificity and sensitivity of this biomarker, it overcomes the bottleneck of traditional imaging examinations in missing cases where early vascular morphology changes are not significant. By applying RETN detection products to the preparation of diagnostic reagents or predictive tools, non-invasive, low-cost large-scale screening is achieved, solving the key problem of the lack of specific biomarkers for evaluating unstable plaques in existing technologies.

[0017] The present invention provides a method for constructing a predictive model for unstable angina based on the biomarker RETN. This method ensures the data quality and correspondence with clinical diagnostic results by screening and grouping a sample set with complete serum data, providing a reliable foundation for model training. It simultaneously extracts RETN concentration and multidimensional conventional clinical indicators as candidate variables, eliminates redundant variable interference through multicollinearity analysis, and then uses machine learning algorithms to screen key features. This retains the core role of RETN while integrating the synergistic effects of other indicators, improving the model's comprehensive discriminative ability. An ensemble learning strategy is employed, achieving multi-algorithm fusion through base module model training, and then optimizing and integrating the predicted probabilities output by the base modules through meta-modules, enhancing the model's generalization performance and predictive stability. Finally, the model's discriminative power, calibration, and clinical applicability are evaluated on an independent validation set, ensuring its applicability in real-world scenarios, ultimately forming a predictive tool that is both scientifically sound and practical.

[0018] Furthermore, by combining the aforementioned three types of machine learning algorithms with statistical analysis methods, a complementary variable selection mechanism is formed. This mechanism retains the disease-strongly correlated biomarker RETN while integrating conventional clinical indicators such as age, blood pressure, and blood lipids, thus constructing a highly discriminative predictive model. This feature combination includes both traditional cardiovascular risk factors and the novel biomarker RETN. Through the systematic correlation of multi-dimensional parameters, it overcomes the limitations of single-indicator detection and significantly improves the model's early identification ability for unstable angina. This invention provides an intelligent auxiliary diagnostic system for unstable angina based on the biomarker RETN. By constructing an intelligent auxiliary diagnostic system with multiple interconnected modules, it achieves accurate diagnosis of unstable angina based on the RETN biomarker. RETN concentration and routine clinical indicators are directly obtained through serum testing, overcoming the limitations of traditional imaging methods that rely on morphological changes, and enabling the capture of early biomolecular abnormalities. The model invocation module employs an integrated machine learning prediction model to perform multi-dimensional correlation analysis between RETN and selected clinical indicators, solving the problem of insufficient specificity of single biomarkers, while simultaneously improving the model's generalization ability through algorithm optimization. The diagnostic output module transforms the model calculation results into a visualized risk prediction report, providing quantitative evidence for clinical decision-making and overcoming the limitation of traditional examinations in dynamically assessing disease progression. The synergistic effect of the three modules achieves full automation from data acquisition and model calculation to result output, significantly reducing human error and providing an efficient and low-invasive solution for early diagnosis and risk assessment. The diagnostic model constructed in this invention achieved an area under the receiver operating characteristic (AUC) of over 0.842 in an independent validation cohort, with an accuracy exceeding 76%, significantly outperforming the diagnostic efficacy of single traditional biomarkers. It holds promise for achieving early and non-invasive accurate diagnosis of UA. Attached Figure Description

[0019] Figure 1 Differential protein expression levels between the healthy (HC) group and the UA group, where A is a volcano plot of differential protein expression; B is a heatmap of differential protein expression.

[0020] Figure 2 Protein expression levels and diagnostic performance in the HC and UA groups: A shows the box plots of the top 10 proteins with the most significant differences; B shows the ROC curves.

[0021] Figure 3 : Correlation analysis of differentially expressed proteins between the HC group and the UA group, where A is the correlation clustering heatmap of differentially expressed proteins; B is the Spearman correlation analysis.

[0022] Figure 4 Bioinformatics analysis of differentially expressed proteins in the HC and UA groups, where A is GO enrichment analysis and B is KEGG enrichment analysis.

[0023] Figure 5 : Protein-protein interaction network diagram.

[0024] Figure 6ELISA was used to verify the expression levels of different biomarkers in the healthy control group (HC) and the disease group (UA). A represents the concentration distribution of AZU1 in the HC and UA groups; B represents the concentration distribution of MPO in the HC and UA groups; C represents the concentration distribution of RETN in the HC and UA groups; D represents the concentration distribution of PRTN3 in the HC and UA groups; E represents the concentration distribution of MMP-9 in the HC and UA groups; and F represents the receiver operating characteristic (ROC) curve and area under the curve (AUC) of RETN.

[0025] Figure 7 Features for predicting UA were selected using the Least Absolute Shrinkage and Selection Operator (LASSO) and the Random Forest (RF) algorithm. Here, A is the LASSO coefficient plot; B is the LASSO cross-validation plot; C is the final number of features at lambda.1se in the LASSO model; D is the relationship between the number of features selected by RF and the model accuracy; E is the RF feature importance ranking based on the average reduction in Gini impurity; and F is the Venn diagram of overlapping key features identified by the LASSO and RF algorithms.

[0026] Figure 8 The clinical diagnostic prediction model was constructed, interpreted, and validated. AB represents the receiver operating characteristic (ROC) curves of six machine learning algorithms on the training set and internal validation set; C represents the learning curve of the logistic regression model, and the convergence of the training score and cross-validation score indicates a good model fit; D represents the calibration of the model prediction probability and the observed event occurrence rate of all algorithms, with the dashed line representing perfect calibration. Figure 9 The clinical diagnostic prediction model was constructed, interpreted, and validated. A represents the feature importance ranking based on SHAP; B represents the feature attributes in SHAP; C represents the SHAP force-directed graphs of UA patients and healthy controls; D evaluates the clinical applicability of the model through decision curve analysis, demonstrating the net benefit under a series of clinically relevant probability thresholds; E validates the discriminative ability of the independent test cohort; and F validates the calibration accuracy of the independent test cohort.

[0027] Figure 10 Predictors of UA and serum RETN levels, where A is a forest plot of multivariate logistic regression analysis of independent predictors of UA, B is the serum RETN level grouped by the severity of coronary artery stenosis (assessed by Gensini score), and C is the serum RETN level grouped by hypertension status. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention. Where specific experimental steps or conditions are not specified in the embodiments, they can be performed according to the conventional experimental steps or conditions described in the literature in the art. Reagents or instruments used without specified manufacturers are all commercially available conventional reagent products.

[0029] This invention employs a three-stage prospective cohort design to systematically screen and validate UA-related biomarkers and construct a diagnostic model. A total of 528 participants were included, comprising 378 UA patients and 150 healthy controls.

[0030] The invention protocol was reviewed and approved by the Ethics Committee of the First Affiliated Hospital of Air Force Medical University (Approval No.: KY20252103-F-1). All participants signed informed consent forms.

[0031] Categorical variables were expressed as frequencies (percentages) and compared using the chi-square test. Continuous variables were assessed for normality using the Kolmogorov-Smirnov test: normally distributed variables were expressed as mean ± standard deviation and analyzed using Student's t-test (between groups) or paired t-test (within groups); non-normally distributed variables were expressed as median (interquartile range) and analyzed using the Mann-Whitney U test (between groups) or Wilcoxon signed-rank test (within groups). Comparisons among multiple groups were performed using ANOVA (for normally distributed variables) or the Kruskal-Wallis test (for non-normally distributed variables). Feature selection was performed using LASSO regression and random forest (RF) algorithms. Six machine learning algorithms (XGBoost, RF, LR, LightGBM, SVM, GNB) were used to construct diagnostic models, and model performance and interpretability were evaluated using learning curve, calibration curve, decision curve analysis (DCA), and Shapley sum of explanations (SHAP) values. A two-sided p-value < 0.05 was considered statistically significant.

[0032] Example 1 Serum Olink proteomics study of UA Study participants: This study included 30 UA patients and 28 healthy controls (age and sex matched) as the discovery cohort from October to December 2024.

[0033] Research methods: (1) Fasting fresh whole blood samples were collected from UA patients and healthy controls respectively. 5 mL of fasting venous blood was collected using a pyrogen-free and enzyme-free vacuum coagulation tube (BD Vacutainer SST). After the blood was fully coagulated at room temperature for 2 hours, it was centrifuged at 4℃ and 4000rpm for 10 minutes. The supernatant was collected into a 2 mL cryovial, and after being rapidly frozen in liquid nitrogen, it was transferred to a -80℃ freezer for storage.

[0034] (2) Thaw serum samples on ice, centrifuge at 4°C and 12,000 rpm for 5 minutes to remove precipitate, and take the supernatant; dilute the sample according to the Olink standard procedure (1:10 dilution). The Olink Cardiovascular II Panel uses the adjacent extension technology to link protein-specific antibodies with DNA-encoded tags. After antibody-protein binding, the DNA tag extends to form an amplifiable template. The protein information is converted into an amplifiable signal by high-sensitivity qPCR to obtain the abundance of key proteins. This invention uses 92 cardiovascular-related proteins for analysis.

[0035] (3) The Normalized Protein Expression (NPX) software OlinkNPX Manager was used for quality control and data normalization with reference to internal and external controls. NPX value is a relative quantitative method used to compare the expression level of individual proteins under different conditions. Proteins with significant changes in NPX value are considered differentially expressed proteins (DEPs) (p < 0.05 and |log2(FC)| > 0.1, marked as DEPs); bioinformatics analysis was performed, and the AUC value was calculated using the pROC package in R to evaluate the diagnostic efficacy of the protein; Spearman correlation analysis, GO / KEGG analysis (using the Org.Hs.eg.db database in the ClusterProfiler package, threshold p.adjust < 0.05, enriched entries for extracellular secretion and innate immune activation) and PPI network analysis (TRING database, confidence > 0.7, constructing a protein interaction network, Cytoscape, MCODE plugin, identifying core nodes CCL2 and MMP-9).

[0036] See appendix Figures 1-5Olink proteomics analysis initially revealed 32 differentially expressed proteins in UA patients compared to healthy controls, of which 28 were upregulated and 4 were downregulated. The top 10 proteins with the most significant differences were MMP-9, AZU1, MPO, RETN, PRTN3, proprotein convertase subtilisin-9 (PCSK9), urokinase-type plasminogen activator receptor (U-PAR), peptidoglycan recognition protein 1 (PGLYRP1), paraoxonase 3 (PON3), and tumor necrosis factor receptor 1 (TNF-R1) (p < 0.001). Figure 2 ROC analysis showed that MMP-9, AZU1, MPO, RETN, and PRTN3 had good diagnostic value for UA (AUC > 0.85). Figure 2 Spearman correlation analysis revealed 28 highly correlated protein combinations (B). Figure 3 A and B). GO and KEGG pathway enrichment analysis indicated that the pathogenesis of UA is closely related to extracellular secretory processes, membrane signaling dysregulation, and activation of the innate immune system. Figure 4 (A and B). PPI network analysis suggests that CCL2 and MMP-9 are core nodes, which may synergistically promote plaque instability through the "inflammation-matrix degradation" axis. Figure 5 ).

[0037] Example 2: ELISA Validation and Diagnostic Model Construction Study participants: This study included 44 UA patients and 44 healthy controls from March to June 2025 as the ELISA validation cohort and model construction cohort.

[0038] Research methods: (1) Fasting fresh whole blood samples were collected from UA patients and healthy controls. After being placed at room temperature for 2 hours, the samples were centrifuged at 4℃ and 4000rpm for 10 minutes. The supernatant was collected into 2mL cryovials and transferred to -80℃ for storage as soon as possible. (2) Five candidate proteins (MMP-9, AZU1, MPO, RETN, PRTN3) were validated using the Wuhan Yunclone ELISA kit according to the supplier's instructions. (3) Routine clinical test indicators were collected, and feature variables were screened using LASSO regression and recursive feature elimination (RFE) algorithms of random forest. (4) Six machine learning algorithms were used to construct UA diagnostic models: XGBoost, RF, LR, LightBGM, SVM, and GNB. To comprehensively evaluate the model's performance and reliability, learning curves were plotted. By observing the changing trends of training and validation set scores with increasing sample size, the model's fit and generalization ability were assessed, and it was determined whether overfitting or underfitting existed. Calibration curves were plotted, and the accuracy of the model's output probabilities was evaluated by comparing the consistency between the probabilities predicted by the model and the actual observation frequencies. The SHAP method was used for model interpretability analysis, calculating the marginal contribution of each feature to a single prediction result, thereby quantitatively identifying the key predictive factors affecting the model's decision-making.

[0039] Results: (1) ELISA results showed that the serum RETN and MPO concentrations in UA patients were significantly higher than those in the healthy control group (p<0.05), with RETN showing the most significant diagnostic efficacy (AUC=0.849). Figure 6 .

[0040] (2) Given the excellent single diagnostic performance of RETN, to further enhance its diagnostic potential, a machine learning model for UA diagnosis was constructed by integrating RETN concentration and 26 routine clinical indicators. Feature selection was performed using LASSO regression (λ=0.053, following a standard error rule), retaining 10 variables: age, SBP, AST, TC, TG, HDL-C, Cys C, WBC, HGB, and RETN. Coefficient analysis showed that age, SBP, AST, TG, Cys C, WBC, and RETN may be positively correlated with UA, while TC, HGB, and HDL-C may be negatively correlated with UA. To further evaluate the importance of the features, the RFE algorithm (10-fold cross-validation) was used to determine the most relevant predictors. This method ultimately selected 20 optimal features through the cross-validation process. The features were systematically ranked according to the average accuracy index, which quantifies the contribution of each variable to the model performance. Analysis showed that RETN was the most influential predictor, followed by other clinical indicators such as SBP, TC, TG, HDL-C, LDL-C, and Cys C. This hierarchical ranking of variables helps us understand their relative importance in the predictive model, with RETN showing the strongest association with the target outcome. Finally, by cross-filtering the features identified by the LASSO and random forest methods, nine core features were ultimately determined for model construction: RETN, SBP, TC, TG, HDL-C, Cys C, AST, HGB, and WBC. Figure 7 ).

[0041] (3) Six machine learning algorithms (XGBoost, RF, LR, LightBGM, SVM, GNB) were used for modeling. All algorithms demonstrated strong predictive capabilities, achieving excellent performance with AUC values ​​greater than 0.9 on both the training and validation sets. Among them, the LR model showed the best overall performance and calibration (Brier Score = 0.061). On the training set, its AUC was 0.977 (95% CI: 0.970-0.990), accuracy was 92.04%, sensitivity was 90.30%, and specificity was 93.83%. On the internal validation set, the AUC was 0.955 (95% CI: 0.914-0.998), accuracy was 87.32%, sensitivity was 88.18%, and specificity was 89.78%. Figure 8 SHAP analysis showed that RETN was the most influential positively correlated predictor (AD). Figure 9 The DCA curve shows that the model achieves a maximum net benefit of 0.3 when the disease probability reaches the 20% threshold, demonstrating a significant clinical net benefit. Figure 9 D).

[0042] Example 3: External Independent Validation of the Diagnostic Model 1. Research Subjects This study prospectively recruited 290 UA patients and 62 healthy controls from June to September 2025 as an independent sample set to externally validate the discriminative power and clinical applicability of the constructed model.

[0043] 2. Research Methods (1) Fasting fresh whole blood samples were collected from UA patients and healthy controls respectively. After being placed at room temperature for 2 hours, the samples were centrifuged at 4℃ and 4000rpm for 10 minutes. The supernatant was collected into 2mL cryovials and transferred to -80℃ freezer for storage as soon as possible.

[0044] (2) In an independent external validation set, the LR model was used for validation, and the consistency between its predicted probabilities and actual observations was evaluated by plotting calibration curves.

[0045] 3. Research Results In an independent external validation cohort, the model maintained excellent diagnostic efficacy (AUC=0.842, 95% CI: 0.821-0.862), with an accuracy of 76.53%, sensitivity of 75.52%, and specificity of 81.29%, and the calibration curves showed a high degree of consistency between the predicted and measured values. Figure 9 ,EF).

[0046] Example 4: Application of RETN in assessing the degree of coronary artery stenosis Study subjects: The ELISA validation cohort and the diagnostic model validation cohort were merged to form a comprehensive analysis cohort containing 106 healthy controls and 334 patients with UA, which will be used for subsequent analysis of UA risk factors and coronary artery stenosis.

[0047] Research Methods: (1) To identify independent risk factors for UA, we constructed a multivariate logistic regression model. The model used clinical diagnosis as the dependent variable and the independent variables included were the variables selected by the diagnostic model. The analysis results were expressed as odds ratio (OR) and its 95% confidence interval (95% CI) to assess the association strength and statistical significance of each factor with the risk of UA. (2) To explore the relationship between serum RETN concentration and the severity of coronary artery lesions, we quantitatively assessed the degree of coronary artery stenosis in all UA patients according to the Gensini scoring system and stratified UA patients (mild, moderate, and severe stenosis subgroups). The Kruskal-Wallis H test (for non-normally distributed data) was used to compare the differences in serum RETN concentration among different stenosis subgroups.

[0048] See appendix Figure 10 Multivariate logistic regression analysis confirmed that SBP (OR=1.028 (95% CI: 1.012-1.045, p=0.001)); AST (OR=1.053 (95% CI: 1.019-1.093, p=0.002)); TG (OR=3.255 (95% CI: 1.728-6.709, p<0.001)); WBC (OR=1.432 (95% CI: 1.175-1.775, p<0.001)); RETN (OR=1.172 (95% CI: 1.172, p<0.001)). CI: 1.069-1.306, p<0.001) was an independent risk factor for UA; while HDL-C (OR=0.083 (95% CI: 0.030-0.212, p<0.001)) and HGB (OR=0.959, 95% CI: 0.940-0.978, p<0.001) were protective factors.

[0049] After stratifying UA patients by coronary artery stenosis using the Gensini score, RETN concentrations were significantly elevated in all stenosis subgroups (mild, moderate, and severe) compared to the healthy control group (p < 0.01). The RETN concentrations were: 1.955 (95% CI: 1.410–4.135) in the healthy control group; 3.585 ng / ml (95% CI: 2.014–6.482) in the mild stenosis group (p < 0.0001); 3.052 ng / ml (95% CI: 1.745–5.558) in the moderate stenosis group (p = 0.0035); and 4.292 ng / ml (95% CI: 2.543–9.066) in the severe stenosis group (p < 0.0001). Notably, the RETN concentrations in the severe stenosis group were significantly higher than those in the mild (p = 0.0405) and moderate (p = 0.0011) groups. Furthermore, subgroup analysis showed that UA patients with hypertension had significantly higher RETN levels than UA patients with normal blood pressure, at 4.455 (95% CI: 2.495-9.028) and 2.736 (95% CI: 0.940-0.978, p < 0.001), respectively, suggesting a positive correlation between RETN levels and blood pressure.

[0050] In summary, this invention, based on Olink proteomics technology, systematically screened and verified for the first time the clinical application value of serum RETN protein biomarkers in the diagnosis of UA and the grading of coronary artery stenosis.

[0051] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. The application of products that detect biomarkers associated with unstable angina in the preparation of products for the early diagnosis or prediction of unstable angina, characterized in that, The biomarker associated with unstable angina is RETN.

2. The application according to claim 1, characterized in that, The product is a kit, chip, reagent, or monitoring system for detecting the biomarker RETN.

3. A method for constructing a predictive model for unstable angina based on the biomarker RETN, characterized in that, Includes the following steps: S1. Select a sample set with complete serum test data from the physical examination database and divide it into a disease group and a control group according to the clinical diagnosis results; S2, RETN concentration and 26 routine clinical indicators were extracted as candidate variables, and feature variables were selected for inclusion in the model construction through multicollinearity analysis and machine learning algorithms; S3, an ensemble learning strategy is used to construct a prediction model, which includes two stages: training of the base module model and optimization of the meta module model; S4. Evaluate the model's discriminative power, calibration, and clinical applicability in the independent validation set to complete model construction.

4. The method for constructing a predictive model for unstable angina based on the biomarker RETN according to claim 3, characterized in that, In step S2, LASSO regression, random forest, and support vector machine recursive feature elimination are used to jointly screen variables, and multicollinearity analysis is combined to optimize feature selection. Then, machine learning algorithms are used to screen feature variables to be included in the model construction.

5. The method for constructing a predictive model for unstable angina based on the biomarker RETN according to claim 4, characterized in that, The feature variables included in the model construction include age, systolic blood pressure (SBP), aspartate aminotransferase (AST), total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), cystatin C, white blood cell count (WBC), hemoglobin (HGB), and resistin (RETN).

6. The method for constructing a predictive model for unstable angina based on the biomarker RETN according to claim 3, characterized in that, The base module model employs at least one of the following algorithms: XGBoost, Random Forest (RF), Logistic Regression (LR), LightGBM, Support Vector Machine (SVM), and Gaussian Naive Bayes (GNB).

7. The method for constructing a predictive model for unstable angina based on the biomarker RETN according to claim 3, characterized in that, The meta-module model is constructed through a stacking integration strategy. Its input is the predicted probability output by the base module model, and its output is the risk value of unstable angina.

8. A smart auxiliary diagnostic system for unstable angina based on the biomarker RETN, characterized in that, The auxiliary diagnostic system includes: The data acquisition module is used to collect patients' serum test data, which includes RETN concentration and routine clinical indicators. The model invocation module is used to invoke the prediction model constructed according to any one of claims 3-7; The diagnostic output module is used to input the data collected by the data acquisition module into the prediction model called by the model calling module, and to generate a risk prediction report for unstable angina.

9. A computer device, comprising: The memory and processor are used to store program instructions; the processor is used to call the program instructions, and when the program instructions are executed, the following functions are implemented: (1) calling the prediction model construction method of any one of claims 3-7; (2) executing the data processing flow of the intelligent auxiliary diagnostic system of claim 8; (3) outputting the risk prediction result of unstable angina.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it performs the following functions: (1) calling the prediction model construction method of any one of claims 3-7; (2) executing the data processing flow of the intelligent auxiliary diagnostic system of claim 8; and (3) outputting the risk prediction result of unstable angina.

Images