A sepsis cardiomyopathy prognosis risk stratification system and a construction method thereof

Abstract

The application provides a sepsis cardiomyopathy prognosis risk stratification system and a construction method thereof. The method first classifies SIC patients through multi-modal cardiac ultrasound, and then adopts a machine learning algorithm to screen out key ultrasound variables: left ventricular overall longitudinal strain (GLS) and tricuspid annular plane systolic excursion (TAPSE) which are strongly related to prognosis. Based on these variables, cluster analysis is performed, and SIC risk subphenotypes with significant prognosis difference are successfully constructed, and the mortality of the high-risk subtype is significantly higher. The study found that a single indicator, TAPSE, has extremely high predictive efficiency in identifying high-risk subphenotypes and can achieve rapid bedside assessment. The application first establishes a prognosis risk stratification system for SIC, simplifies the evaluation process, realizes early identification of high-risk patients, and provides key technical support for clinical precise intervention and improvement of prognosis.

Description

Technical Field

[0001] This invention belongs to the field of medical diagnostic technology, specifically relating to a prognostic risk stratification system for sepsis cardiomyopathy and its construction method. Background Technology

[0002] Sepsis is a life-threatening organ dysfunction caused by a dysregulated response to infection, with the heart being one of the most frequently affected target organs. Sepsis-induced cardiac dysfunction is called sepsis-induced cardiomyopathy (SIC), characterized by a potentially reversible suppression of overall cardiac function. SIC has a high incidence in sepsis patients and is associated with a very high mortality rate, making it a key factor contributing to poor patient outcomes.

[0003] Bedside echocardiography, due to its non-invasive, convenient, and repeatable characteristics, has become a core tool for clinical assessment of cardiac function in patients with spontaneous cardiac arrest (SIC). Current research uses ultrasound to classify SIC into different ultrasound phenotypes, such as left ventricular systolic dysfunction, left ventricular diastolic dysfunction, right ventricular dysfunction, and mixed types. These different phenotypes exhibit significant differences in incidence, hemodynamic changes, and prognostic impact.

[0004] However, current clinical practice and research still have significant limitations. On the one hand, most studies focus only on a single type of cardiac dysfunction, lacking a systematic comparative and integrative analysis of the relationship between different ultrasound phenotypes and patient prognosis (such as mortality). On the other hand, a practical system based on multimodal ultrasound indicators for rapid risk stratification has not yet been established clinically, leading to difficulties and delays in identifying high-risk patients for SIC, thus affecting the implementation of early targeted interventions.

[0005] Therefore, there is an urgent need for a risk stratification method that can systematically integrate multimodal ultrasound information and effectively differentiate the prognoses of patients with symptomatic intravascular coagulation (SIC). Establishing such a system to enable early and rapid bedside identification of high-risk patients has significant practical implications for improving the clinical outcomes of SIC patients. Summary of the Invention

[0006] This invention provides a prognostic risk stratification system for septic cardiomyopathy (SIC) and its construction method. SIC patients are classified using multimodal echocardiography, and recursive feature elimination (RFE) and the Boruta algorithm are applied to screen for key ultrasound variables most relevant to prognosis: global longitudinal strain (GLS) of the left ventricle and tricuspid annular systolic displacement (TAPSE). Based on this, cluster analysis is used to construct SIC risk subphenotypes with significant prognostic differences, among which the high-risk subtype has a significantly higher 28-day mortality rate. This invention further confirms that the single indicator TAPSE has excellent predictive power for identifying high-risk subtypes (AUC=0.982). This method establishes a prognostic risk stratification system for SIC for the first time, enabling early and rapid identification of high-risk patients through a single bedside ultrasound examination, providing a key technical solution for precise clinical intervention.

[0007] On the one hand, the present invention provides a method for constructing a prognostic risk stratification system for septic cardiomyopathy, which adopts the following technical solution: A method for constructing a prognostic risk stratification system for septic cardiomyopathy includes the following steps: S1. Patient classification: Obtain multimodal echocardiography data of sepsis patients and classify patients diagnosed with septic cardiomyopathy into different echocardiographic phenotypes based on cardiac dysfunction criteria; S2. Screening of key variables: Using machine learning algorithms, at least one key ultrasound variable most relevant to the patient's prognostic endpoint was screened from multimodal echocardiography data; key ultrasound variables included global longitudinal strain of the left ventricle and / or systolic displacement of the tricuspid annulus; S3. Risk stratification model construction: Based on the selected key ultrasound variables and combined with ultrasound phenotypic information, patients are divided into at least two subphenotypes with different prognostic risks through cluster analysis, one of which is a high-risk prognostic subphenotype.

[0008] Preferably, the machine learning algorithm in step S2 includes the recursive feature elimination algorithm and the Boruta fully correlated feature selection algorithm.

[0009] Preferably, in step S3, a hierarchical clustering method is used for cluster analysis, and the optimal number of clusters is determined by statistical methods.

[0010] Preferably, the high-risk subphenotypes identified in step S3 have a significantly higher 28-day mortality rate than other subphenotypes.

[0011] On the other hand, the present invention also provides a prognostic risk stratification system for septic cardiomyopathy, which adopts the following technical solution: A prognostic risk stratification system for septic cardiomyopathy, based on ultrasound assessment of patients, classifies patients into different prognostic risk subphenotypes; Among them, the prognostic risk subphenotype includes at least one high-risk prognostic subphenotype; The criteria for classifying patients into high-risk prognostic subphenotypes are based on at least one key ultrasound variable, including global left ventricular longitudinal strain and / or tricuspid annular systolic displacement, combined with the patient’s ultrasound phenotype information, and determined by predetermined classification rules. The classification rules are constructed through the following steps: collecting sample data from patients with septic cardiomyopathy, applying machine learning algorithms to screen out the key ultrasound variables most relevant to the prognostic endpoint, and establishing the classification rules based on the key ultrasound variables and ultrasound phenotypic information through cluster analysis.

[0012] In summary, the beneficial effects of the present invention are as follows: This invention integrates multimodal ultrasound indicators with machine learning algorithms to construct, for the first time, a systematic prognostic risk stratification system for septic cardiomyopathy (SIC), enabling the scientific identification and objective description of SIC patients, especially high-risk subtypes, and solving the long-standing problem of a lack of effective early warning tools in clinical practice.

[0013] The core benefit of this invention lies in its outstanding clinical translational value. Studies have found that the ultrasound index TAPSE has near-perfect predictive ability for high-risk subphenotypes (AUC=0.982), which simplifies the complex risk stratification system into a single ultrasound measurement that can be rapidly performed at the bedside. This method significantly reduces the operational threshold and time cost of clinically identifying high-risk patients, providing crucial and practical technical support for achieving early warning, timely intervention, and ultimately improving adverse clinical outcomes in SIC patients. Attached Figure Description

[0014] Figure 1 Venn diagram for echocardiographic patterns of septic cardiomyopathy (SIC); Figure 2 Cluster diagram of SIC prognosis; Figure 3 A graph comparing 28-day mortality rates of different risk subphenotypes of septic cardiomyopathy; Figure 4 Forest plot of multivariate logistic regression analysis for independent risk factors of high-risk subphenotypes in SIC prognosis; Figure 5 In the middle A, the receiver operating characteristic curve (ROC) of the ultrasound index TAPSE for identifying high-risk subphenotypes of SIC prognosis is shown. Figure 5 The receiver operating characteristic curve (ROC) of the clinical risk factor model for identifying high-risk subphenotypes of SIC prognosis is shown in Figure B. Figure 5 The middle C represents the receiver operating characteristic curve (ROC) for identifying high-risk subphenotypes of SIC using the ultrasound index GLS. Detailed Implementation

[0015] The present invention will be further described in detail below with reference to the embodiments.

[0016] Example Example 1 The specific methods for patient screening and SIC ultrasound phenotypic classification are as follows: Multimodal echocardiography was performed on 181 patients with sepsis. Patients with acute cardiac dysfunction who were excluded from coronary artery disease were considered to have septic cardiomyopathy (SIC), and SIC was classified into 4 echocardiographic phenotypes according to the definition of cardiac dysfunction in the guidelines.

[0017] like Figure 1 As shown, among the SIC patients, 11 (10.00%) had isolated left ventricular systolic dysfunction (LVSDF), 40 (36.36%) had isolated left ventricular diastolic dysfunction (LVDDF), 11 (10.00%) had isolated right ventricular dysfunction (RVDF), and 48 (43.63%) had mixed ventricular dysfunction.

[0018] Example 2 The specific methods for key variable selection and risk model construction are as follows: S1. Collect multimodal ultrasound data and clinical prognostic data from all patients diagnosed with septic cardiomyopathy. Standardize continuous ultrasound variables (GLS, TAPSE) to eliminate the influence of dimensions. Label and encode categorical variables. Divide the dataset into training and validation sets according to a preset ratio to ensure that there are no significant differences in baseline characteristics between the two groups, for use in subsequent feature selection and model construction.

[0019] S2. Two complementary feature selection methods were used to screen the ultrasound variables most associated with 28-day mortality in SIC patients: Recursive Feature Elimination (RFE): First, a base classifier (such as logistic regression or random forest) is selected, and the model is fitted using the training set data. By recursively removing the least important features (based on model coefficients or feature importance) and evaluating model performance (such as AUC under cross-validation), a minimum optimal subset of features is finally determined that maintains or optimizes the model's predictive performance.

[0020] Boruta's fully relevant feature selection algorithm is a wrapper around a random forest. It creates shaded copies (randomly shuffled) of the original features and runs the random forest model iteratively, comparing the importance of the original and shaded features. Finally, it counts the number of times the original feature's importance is significantly higher than its best shaded copy, determining whether the feature is "relevant" (significantly related to the prognosis) or "unimportant." This method effectively identifies all relevant features, avoiding omissions.

[0021] Combining the results of the two methods, the intersection or consensus was used to screen out a few core ultrasound variables that were consistently identified as strongly correlated with prognosis. According to this invention, these were ultimately determined to be global longitudinal strain (GLS) of the left ventricle and systolic displacement of the tricuspid annulus (TAPSE).

[0022] S3. Integrate the key continuous variables (GLS, TAPSE) selected in the previous step with the coded SIC ultrasound phenotypic classification variables to form a feature matrix for cluster analysis. Using statistical packages such as NbClust, based on the training set data, evaluate the compactness and separation of the model at different numbers of clusters (k) by calculating various internal indices (such as silhouette coefficient, Calinski-Harabasz index, Davies-Bouldin index, etc.), and determine the optimal number of clusters using statistical methods. Use hierarchical clustering methods to cluster the patients in the training set based on the above feature matrix. According to the determined optimal number of clusters (k=2), divide the patients into two mutually exclusive clusters.

[0023] The two resulting clusters were defined as different SIC prognostic risk subphenotypes (subphenotype 1 and subphenotype 2). The same cluster centers or rules were applied to the validation set for subphenotype partitioning. Subsequently, the 28-day case fatality rates of the two subphenotypes on the validation set were compared, and statistical methods such as the chi-square test were used to verify whether there was a significant difference in prognosis. The subphenotype with a significantly higher case fatality rate was defined as the "high-risk prognostic subphenotype," and the other was defined as the low-risk or intermediate-risk subphenotype.

[0024] S4. On the validation set, evaluate the ability of this risk stratification model to distinguish patient mortality risk. Simultaneously, ROC curves of the relationship between key continuous variables and subphenotype attribution can be plotted separately to preliminarily explore its potential optimal cutoff value for rapid bedside stratification, laying the foundation for subsequent simplified applications.

[0025] like Figure 2 As shown, the data includes the multimodal ultrasound variables GLS and TAPSE associated with 28-day mortality in SIC patients, and the two subtypes obtained by hierarchical cluster analysis of the labeled SIC ultrasound phenotypes.

[0026] like Figure 3As shown, the 28-day mortality rates of subphenotype 1 (n = 59, 53.6%) and subphenotype 2 (n = 51, 46.4%) were 49.2% and 21.6%, respectively, with a significant difference between the two (P = 0.005). Both subphenotypes were also significantly different from the 28-day mortality rate of sepsis (both P < 0.05). Subphenotype 1 is a high-risk subphenotype for SIC prognosis.

[0027] Example 3 Multivariate logistic regression analysis, the specific method is as follows: S1. To identify clinical factors independently associated with high-risk subphenotypes of SIC prognosis, variable preparation was first performed: whether the patient belongs to the high-risk subtype was used as a binary outcome variable, and a wide range of candidate independent variables were selected, including demographic characteristics, vital signs, disease severity scores, biomarkers, treatment methods, and multimodal ultrasound indicators. After data preprocessing, univariate logistic regression analysis was performed on each variable, and variables significantly associated with the high-risk subtype were selected based on the pre-set significance level for subsequent modeling.

[0028] S2. Subsequently, all variables that showed significant initial screening were included in the initial multivariate logistic regression model. A stepwise regression method based on the likelihood ratio test was used for variable selection, ultimately retaining a set of independently related risk factors. The Hosmer-Lemeshow test was used to assess the model's goodness of fit, and the regression coefficients, adjusted odds ratios, confidence intervals, and p-values ​​of each factor in the final model were interpreted. The results are often visually presented using forest plots, clearly demonstrating that factors such as SOFA score, cTnI level, and invasive mechanical ventilation are independently positively correlated factors for the high-risk subtype of SIC.

[0029] like Figure 4 As shown, multivariate logistic regression analysis showed that heart rate, SOFA score, cTnI, and mode of mechanical ventilation were independently positively correlated with the high-risk subphenotype of SIC prognosis. P <0.05).

[0030] Example 4 The predictive power of each indicator for the high-risk subphenotypes of SIC was evaluated using the following method: S1. Using this high-risk subtype as the binary outcome, the ultrasound indicators TAPSE, GLS, and the predicted probabilities of clinical risk factors calculated based on a multivariate regression model were used as predictive factors to be tested. Subsequently, three corresponding receiver operating characteristic (ROC) curves were plotted, with the vertical axis representing sensitivity and the horizontal axis representing 1-specificity.

[0031] S2. Predictive accuracy was quantitatively compared by calculating the area under each curve and its 95% confidence interval. The DeLong test and other methods were used to perform pairwise statistical comparisons of AUC to verify whether the discriminative performance of TAPSE was significantly superior to that of GLS and the combination of clinical risk factors.

[0032] S3. Based on this, for the indicator with the best performance, determine its optimal diagnostic cutoff value according to the Youden index maximization principle, and report the corresponding diagnostic indicators such as sensitivity and specificity. Finally, visualize the three ROC curves and their AUC values ​​in the same graph to intuitively show the differences in predictive efficacy.

[0033] like Figure 5 As shown, TAPSE exhibits excellent predictive power in identifying high-risk subphenotypes of SIC prognosis, with an AUC of 0.982 (95% CI: 0.96–1.00). Figure 5 A), the AUC of clinical risk factors was 0.825, with a 95% CI of (0.747 - 0.902). Figure 5 B), AUC of GLS = 0.803, 95% CI (0.72 - 0.885) ( Figure 5 C).

[0034] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for constructing a prognostic risk stratification system for septic cardiomyopathy, characterized in that, Includes the following steps: S1. Patient classification: Obtain multimodal echocardiography data of sepsis patients and classify patients diagnosed with septic cardiomyopathy into different echocardiographic phenotypes based on cardiac dysfunction criteria; S2. Screening of key variables: Using machine learning algorithms, at least one key ultrasound variable most relevant to the patient's prognostic endpoint is screened from the multimodal echocardiography data; the key ultrasound variable includes global longitudinal strain of the left ventricle and / or systolic displacement of the tricuspid annulus; S3. Risk stratification model construction: Based on the selected key ultrasound variables and combined with the ultrasound phenotype information, patients are divided into at least two subphenotypes with different prognostic risks through cluster analysis, one of which is a high-risk prognostic subphenotype.

2. The method for constructing a prognostic risk stratification system for septic cardiomyopathy according to claim 1, characterized in that, The machine learning algorithms in step S2 include the recursive feature elimination algorithm and the Boruta fully correlated feature selection algorithm.

3. The method for constructing a prognostic risk stratification system for septic cardiomyopathy according to claim 1, characterized in that, In step S3, the hierarchical clustering method is used to perform the cluster analysis, and the optimal number of clusters is determined by statistical methods.

4. A method for constructing a prognostic risk stratification system for septic cardiomyopathy according to any one of claims 1 to 3, characterized in that, The high-risk prognostic subphenotype identified in step S3 has a significantly higher 28-day mortality rate than other subphenotypes.

5. A prognostic risk stratification system for septic cardiomyopathy according to any one of claims 1 to 4, characterized in that, The system categorizes patients into different prognostic risk subphenotypes based on ultrasound assessment results. The prognostic risk subphenotype includes at least one high-risk prognostic subphenotype; The criteria for classifying patients into the high-risk prognostic subphenotype are based on at least one key ultrasound variable including global left ventricular longitudinal strain and / or tricuspid annular systolic displacement, combined with the patient’s ultrasound phenotype information, and determined by a predetermined classification rule. The classification rules are constructed through the following steps: collecting sample data from patients with septic cardiomyopathy, applying machine learning algorithms to screen out the key ultrasound variables most relevant to the prognostic endpoint, and establishing the classification rules based on the key ultrasound variables and ultrasound phenotypic information through cluster analysis.

Images