Subtype characteristic and typing system for sepsis clotting disease

Abstract

The invention discloses a sepsis clotting disease subtype characteristic and typing system, which comprises a basic data acquisition module used for acquiring SIC patient characteristics; a potential category analysis module configured to identify potential groups of the SIC patients based on the SIC patient features; the K-means clustering module is configured to cluster the characteristics of the SIC patients and obtain a clustering result of the SIC patients; and the typing result output module is configured to output subtype features and typing results of the SIC patients based on the potential groups of the SIC patients and the clustering results of the SIC patients. The subtype classification of the SIC patient can be rapidly and accurately analyzed, and a direction is provided for subsequent treatment.

Description

Technical Field

[0001] The present invention relates to the field of medical devices, and particularly to a sepsis coagulopathy subtype characteristic and classification system. Background Art

[0002] Sepsis is a common critical illness in the intensive care unit (ICU). It is a highly heterogeneous disease with different pathogenesis and pathophysiological characteristics, often accompanied by multiple organ dysfunction, and has a high mortality and morbidity. The coagulation dysfunction caused by sepsis, also known as sepsis-induced coagulopathy (SIC), is mainly caused by infection and acute systemic inflammatory response leading to endothelial injury, and is defined by prothrombin time (PT) / international normalized ratio (INR), platelet count, and SOFA score. As shown in Table 1, when the total score reaches or exceeds 4 points, it is diagnosed as SIC.

[0003] Table 1

[0004]

[0005] There is controversy about the overall effect of anticoagulant therapy for SIC patients. Some studies have shown that anticoagulant therapy may only be beneficial to certain specific phenotypes, but may cause harm to other phenotypes. Therefore, determining different treatment phenotypes of SIC is crucial for targeted anticoagulant therapy. Summary of the Invention

[0006] In view of the above deficiencies in the prior art, the sepsis coagulopathy subtype characteristic and classification system provided by the present invention can quickly obtain the subtype characteristics and categories of SIC patients.

[0007] To achieve the above invention object, the technical solution adopted by the present invention is as follows:

[0008] Provide a sepsis coagulopathy subtype characteristic and classification system, which includes:

[0009] A basic data acquisition module, used to acquire the characteristics of SIC patients;

[0010] A latent class analysis module, configured to identify the latent groups of SIC patients based on the characteristics of SIC patients;

[0011] A K-means clustering module, configured to cluster the characteristics of SIC patients to obtain the clustering results of SIC patients;

[0012] A classification result output module, configured to output the subtype characteristics and classification results of SIC patients based on the latent groups of SIC patients and the clustering results of SIC patients.

[0013] Furthermore, the characteristics of SIC patients include age, heart rate, SOFA score, APS III score, fibrinogen, international normalized ratio, activated partial thromboplastin time, platelets, hemoglobin, mean corpuscular hemoglobin content, white blood cell count, red blood cell count, blood urea nitrogen, creatinine, alanine aminotransferase, pH, partial pressure of oxygen, and lactate.

[0014] Furthermore, the latent class analysis module includes:

[0015] An external index analysis unit configured to characterize the association of SIC patient characteristics through latent class variables;

[0016] A clinical feature type analysis unit configured to explore the latent structure behind the SIC patient characteristics through factor analysis, perform finite mixture modeling on the data based on probability, identify latent groups that are not measured or observed in the overall sample, and analyze different clinical feature types of SIC patients;

[0017] A latent subgroup classification unit configured to discover latent subgroups in the sample using a mixed effects model and assign SIC patients to different subtypes.

[0018] Furthermore, the number of clusters of the K-means clustering module is obtained by evaluating the consensus matrix heat map of the cumulative distribution function of the "elbow method" and the clustering consensus graph.

[0019] Furthermore, the sepsis coagulopathy subtypes include 3 subclasses, where:

[0020] The first subclass is characterized by blood cells below the threshold and clinically manifested as excessive blood loss;

[0021] The second subclass is characterized by coagulation disorders and multiple organ dysfunction;

[0022] The third subclass is characterized by age greater than the threshold, a higher proportion of comorbidities than the threshold, a higher fibrinogen concentration than the threshold, and a lower proportion of plasma and platelet transfusions than the threshold.

[0023] Furthermore, when the latent group of SIC patients is consistent with the clustering result of SIC patients, the latent group of SIC patients or the clustering result of SIC patients is directly output as the typing result; when the latent group of SIC patients is inconsistent with the clustering result of SIC patients, a warning message is output.

[0024] The beneficial effects of the present invention are as follows: The present invention can quickly and accurately analyze the subtype categories of SIC patients, providing a direction for subsequent treatment. Description of the Drawings

[0025] Figure 1 It is a schematic structural diagram of the system;

[0026] Figure 2 The clustering result obtained by the K-means clustering module in the embodiment;

[0027] Figure 3 The violin plot of age in the subcategory classification;

[0028] Figure 4 The violin plot of heart rate in the subcategory classification;

[0029] Figure 5 The violin plot of SOFA score in the subcategory classification;

[0030] Figure 6 The violin plot of APS III score in the subcategory classification;

[0031] Figure 7 The violin plot of fibrinogen in the subcategory classification;

[0032] Figure 8 The violin plot of international normalized ratio in the subcategory classification;

[0033] Figure 9 The violin plot of thrombin time (partial thromboplastin time) in the subcategory classification;

[0034] Figure 10 The violin plot of platelets in the subcategory classification;

[0035] Figure 11 The violin plot of hemoglobin in the subcategory classification;

[0036] Figure 12 The violin plot of mean corpuscular hemoglobin content in the subcategory classification;

[0037] Figure 13 The violin plot of white blood cell count in the subcategory classification;

[0038] Figure 14 The violin plot of red blood cell count in the subcategory classification;

[0039] Figure 15 The violin plot of blood urea nitrogen in the subcategory classification;

[0040] Figure 16 The violin plot of creatinine in the subcategory classification;

[0041] Figure 17 The violin plot of alanine aminotransferase (glutamic-pyruvic transaminase) in the subcategory classification;

[0042] Figure 18 The violin plot of pH (acidity and alkalinity) in the subcategory classification;

[0043] Figure 19It is a violin plot of partial pressure of oxygen in the subclass classification;

[0044] Figure 20 It is a violin plot of lactic acid in the subclass classification;

[0045] Figure 21 It is an elbow plot for recommending the best classification category in K-means;

[0046] Figure 22 It is the sample distribution after non-linear dimensionality reduction of each sample;

[0047] Figure 23 It is the recommended optimal number of clusters for each index of K-means. Detailed implementation manners

[0048] The following describes the detailed implementation manners of the present invention to facilitate those skilled in the art of this technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the detailed implementation manners. For those of ordinary skill in the art of this technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

[0049] As Figure 1 shown, the sepsis coagulopathy subtype feature and classification system includes:

[0050] A basic data acquisition module, configured to acquire SIC patient features;

[0051] A latent class analysis module, configured to identify the latent groups of SIC patients based on SIC patient features;

[0052] A K-means clustering module, configured to cluster SIC patient features to obtain the clustering results of SIC patients;

[0053] A classification result output module, configured to output the subtype features and classification results of SIC patients based on the latent groups of SIC patients and the clustering results of SIC patients.

[0054] SIC patient features include age, heart rate, SOFA score, APS III score, fibrinogen, international normalized ratio (INR), activated partial thromboplastin time (PTT), platelets, hemoglobin, mean corpuscular hemoglobin content (MCH), white blood cell count (WBC), red blood cell count (RBC), blood urea nitrogen (BUN), creatinine, alanine aminotransferase (ALT), pH, partial pressure of oxygen (PO2), and lactic acid. The violin plots of these 18 features are as Figure 3 , Figure 4 , Figure 5 , Figure 6 ,Figure 7 , Figure 8 , Figure 9 , Figure 10 , Figure 11 , Figure 12 , Figure 13 , Figure 14 , Figure 15 , Figure 16 , Figure 17 , Figure 18 , Figure 19 and Figure 20 as shown

[0055] In this embodiment, the latent class analysis module processes data based on the latent class analysis (LCA) method. LCA is a probabilistic finite mixture modeling algorithm that can identify latent groups that are not measured or observed in the overall sample. First, the association of external indicators (such as WBC, PT, etc.) is characterized by discontinuous latent variables, i.e., latent class variables. Factor analysis is used to explore the latent structure behind the characteristics of SIC patients, analyze different clinical characteristic types of patients, use a mixed effect model to discover latent subgroups in the sample, and assign individuals to different classes. Exploratory LCA is selected to determine the number of subtypes in the sample, and the poLCA package in the R language package is used to fit the specified latent class type. During the fitting process, maximum likelihood estimation and Bayesian estimation methods are used to estimate model parameters, and Bayesian information criterion (BIC), Akaike information criterion (AIC), likelihood ratio test (LRT), Bootstrap-based likelihood ratio test (BLRT), Entropy, etc. of each subtype are calculated to determine the best subtype.

[0056] Correspondingly, the latent class analysis module includes:

[0057] An external indicator analysis unit, configured to characterize the association of SIC patient characteristics through latent class variables for subsequent subtype classification;

[0058] A clinical characteristic type analysis unit, configured to explore the latent structure behind the characteristics of SIC patients through factor analysis, perform finite mixture modeling on the data based on probability, identify latent groups that are not measured or observed in the overall sample, and analyze different clinical characteristic types of SIC patients;

[0059] A latent subgroup classification unit, configured to use a mixed effect model to discover latent subgroups in the sample and assign SIC patients to different subtypes.

[0060] In the specific implementation process, factor analysis explores the probability distribution between categories through estimation, calculates the correlation between latent factors and features, and thus estimates or infers the existence and nature of latent variables. The mixed - effect model can use the model in the literature of Blum, M., McKendrick, K., Gelfman, L.P., Pinney, S.P., & Goldstein, N.E. (2023). Using Latent Class Analysis to Identify Different Clinical Profiles Among Patients With Advanced Heart Failure. Journal of pain and symptom management, 65(2), 111–119, which will not be elaborated here.

[0061] In this embodiment, the representation of the latent class variable is that when the external categories are insufficient for classification, the external categories are associated through latent categories and then represented by the latent categories to achieve classification. For example, some people in a certain country believe that there are more than two genders, male and female, but some other genders (such as A, B, C). These categories are only internal characters and values, etc., but the external manifestations are still male and female. Latent class analysis is to classify those people with the same internal characters and values.

[0062] Table 2 is a table showing the classification accuracy. The lower the AIC and BIC, the better, and the higher the Entropy, the better. BLRT_p is the difference compared with the previous category. Less than 0.05 means better than the previous category. Generally speaking, when k = 3 (sub - types are 3 categories), the classification is better. If there are more categories, the entropy will decrease and the proportion of each category will decrease, which has no clinical significance.

[0063] Table 2

[0064]

[0065] K - means clustering is an iterative clustering analysis algorithm. By iteratively calculating the distance from each sample to the pre - set K cluster centers, it assigns the sample to the subtype corresponding to the nearest cluster center. In this embodiment, the "elbow method" and the cumulative distribution function of the clustering consensus graph are used to evaluate the heat map of the consensus matrix to determine the optimal number of clusters.

[0066] The sepsis coagulopathy subtypes include 3 sub - classes, among which:

[0067] The first sub - class is characterized by blood cells below the threshold, and clinically manifested as excessive blood loss;

[0068] The second subclass is characterized by coagulation disorders and multiple organ dysfunction;

[0069] The third subclass is characterized by an age greater than the threshold, a proportion of comorbidities higher than the threshold, a fibrinogen concentration higher than the threshold, and a proportion of plasma and platelet transfusions lower than the threshold.

[0070] When the potential group of SIC patients is consistent with the clustering result of SIC patients, the potential group of SIC patients or the clustering result of SIC patients is directly output as the typing result; when the potential group of SIC patients is inconsistent with the clustering result of SIC patients, a warning message is output.

[0071] In this embodiment, the clustering results corresponding to the 18 SIC patient characteristics are as Figure 2 shown, and the 18 SIC patient characteristics are divided into 3 subclasses.

[0072] The characteristics of patients in Class 1 are lower blood cells (white blood cells, red blood cells, platelets, and hemoglobin), and clinically it can be manifested as excessive blood loss rather than coagulation disorders. The anticoagulant treatment for Class 1 patients has poor effects, and the anticoagulant treatment of this type of patients is associated with an increase in 28-day mortality and in-hospital mortality.

[0073] The characteristics of patients in Class 2 are severe coagulation disorders and multiple organ dysfunction. In addition, the INR, PT, PTT, total bilirubin concentration, creatinine, lactate, and SOFA and SAP III scores of Class 2 patients are the highest, and the proportion of patients receiving CRRT, vasopressin, and anticoagulant treatment is the highest, but the mortality rate of this type of patients is still the highest. This subtype is more likely to have severe coagulation dysfunction and organ function abnormalities and can benefit from anticoagulant treatment.

[0074] Patients in Class 3 are the largest category in the population, and their characteristics are older age and a higher proportion of comorbidities. The fibrinogen level of Class 3 patients is the highest. Fibrinogen is a positive acute-phase protein, and hyperfibrinogenemia during sepsis is caused by an increase in fibrinogen and has been considered a cause of thrombosis and vascular damage.

[0075] In this embodiment, the characteristics of 3 patients are used to exemplify the typing result, and some data and classification results are shown in Table 3.

[0076] Table 3

[0077]

[0078] In this embodiment, as Figure 21 , Figure 22 and Figure 23As shown, for each of the 26 indicators calculated using K-means, the indicator recommended for classification as "3" is the most numerous (19), followed by 5 categories (17), and then 2 categories (16). Similarly, Figure 4 The "elbow plot" also shows that when the classification category k = 3, the slope change is the largest, indicating that there will be a relatively obvious change in the classification category at k = 3. If the classification category continues to increase, it will affect the classification accuracy of the overall result.

[0079] In an embodiment of the present invention, MIMIC IV is a longitudinal single-center public database that includes data on more than 40,000 patients admitted to the intensive care unit (ICU) of Beth Israel Deaconess Medical Center from 2008 to 2019 and 11,263 patients with sepsis (defined by Sepsis 3.0). This embodiment uses MIMIC IV to validate the classification model.

[0080] For 4,993 patients in MIMIC IV who met the SIC diagnosis, both LCA and K-means cluster analysis accurately identified three subtypes of SIC. The first subtype of patients (n = 1,808) had the lowest blood cell counts (white blood cells, red blood cells, and platelets). The second subtype of patients (n = 1,157) had severe coagulation disorders, with the highest prothrombin time and international normalized ratio, multiple organ dysfunction, lactate levels, SOFA scores, and mortality. The third subtype (n = 2,028) of patients was older, had more comorbidities, higher fibrinogen concentrations, and lower plasma and platelet transfusion ratios. After adjusting for potential variables, heparin treatment only reduced the 28-day mortality (odds ratio, OR: 0.39, 0.30 - 0.49, P < 0.001) and in-hospital mortality (OR: 0.42, 0.33 - 0.53, P < 0.001) of the second subtype of patients.

[0081] In summary, the present invention can quickly and accurately analyze the subtype categories of SIC patients, providing a direction for subsequent treatment.

Claims

1. A sepsis coagulopathy subtype characterization and classification system, characterized in that Comprising: A basic data acquisition module for acquiring the characteristics of SIC patients; A latent class analysis module configured to identify the latent groups of SIC patients based on the characteristics of SIC patients; A K-means clustering module configured to cluster the characteristics of SIC patients to obtain the clustering results of SIC patients; A subtyping result output module configured to output the subtype characteristics and subtyping results of SIC patients based on the latent groups of SIC patients and the clustering results of SIC patients.

2. The system according to claim 1, characterized in that The characteristics of SIC patients include age, heart rate, SOFA score, APS III score, fibrinogen, international normalized ratio, activated clotting time, platelets, hemoglobin, mean corpuscular hemoglobin content, white blood cell count, red blood cell count, blood urea nitrogen, creatinine, alanine aminotransferase, pH, partial pressure of oxygen, and lactate.

3. The system according to claim 1, wherein The latent class analysis module includes: An external index analysis unit configured to characterize the association of the characteristics of SIC patients through latent class variables; A clinical feature type analysis unit configured to explore the latent structure behind the characteristics of SIC patients through factor analysis, perform finite mixture modeling on the data based on probability, identify the latent groups that are not measured or observed in the overall sample, and analyze the different clinical feature types of SIC patients; A latent subgroup classification unit configured to use a mixed effects model to discover the latent subgroups in the sample and assign SIC patients to different subtypes.

4. The system according to claim 1, wherein The number of clusters of the K-means clustering module is obtained by evaluating the consensus matrix heat map of the cumulative distribution function of the "elbow method" and the clustering consensus graph.

5. The system according to claim 1, wherein The subtypes of sepsis coagulopathy include 3 subcategories, among which: The first subcategory is characterized by blood cells below the threshold and clinically manifested as excessive blood loss; The second subcategory is characterized by coagulation disorders and multiple organ dysfunction; The third subcategory is characterized by age greater than the threshold, a proportion of comorbidities higher than the threshold, a fibrinogen concentration higher than the threshold, and a plasma and platelet transfusion ratio lower than the threshold.

6. The system according to claim 1, wherein When the latent groups of SIC patients are consistent with the clustering results of SIC patients, the latent groups of SIC patients or the clustering results of SIC patients are directly output as the subtyping results; when the latent groups of SIC patients are inconsistent with the clustering results of SIC patients, warning content is output.

Images