A method for typing sepsis, a typing prediction model construction, an application method and device

Abstract

This application discloses a method for classifying sepsis, a model for predicting sepsis classification, an application method, and an apparatus, relating to the field of machine learning. The method includes obtaining the central venous oxygen saturation and peripheral perfusion index of several patients within a preset time period; calculating and obtaining the dynamic blood flow oxygen flow index trajectory of each patient within the preset time period based on the central venous oxygen saturation and peripheral perfusion index; evaluating trajectory similarity using a dynamic time warping algorithm and grouping the trajectories using a hierarchical clustering algorithm to obtain multiple sepsis classifications; and training a machine learning model using each sepsis classification and its corresponding patient clinical characteristics to obtain a sepsis classification prediction model. This application classifies sepsis based on longitudinal data of central venous oxygen saturation and peripheral perfusion index, and establishes prediction models for each classification based on clinical characteristics, ultimately achieving the classification and accurate prediction of the sepsis population.

Description

Technical Field

[0001] This application relates to the field of machine learning, and in particular to a method for classifying sepsis subtypes, construction of a subtype prediction model, application method, and apparatus. Background Technology

[0002] Sepsis is defined as a life-threatening organ dysfunction resulting from a dysregulated response to infection. Sepsis is typically accompanied by systemic inflammation, immune dysregulation, and metabolic disturbances. In severe cases, it can rapidly progress to septic shock, causing widespread organ failure and a high mortality rate. Although significant progress has been made in recent years in the definition, diagnostic criteria, and treatment of sepsis, it remains one of the deadliest acute diseases globally. The mortality rate of sepsis varies among different populations, but is generally estimated to be between 30% and 50%.

[0003] Because sepsis is a highly heterogeneous syndrome, identifying its different clinical subtypes is crucial for precision treatment and management. Early sepsis subtyping studies relied primarily on clinical characteristics; however, these clinically-based classifications often relied on cross-sectional data (a single point in time), failing to accurately reflect the dynamic evolution of sepsis. In recent years, researchers have begun exploring sepsis subtyping based on longitudinal data that reflects dynamic clinical evolution. For example, a sepsis subtyping classification characterized by body temperature dynamics has been proposed, as temperature change patterns may reflect different inflammatory responses, thus affecting the clinical presentation and mortality of sepsis. Furthermore, the trajectory of the Sequential Organ Failure Assessment (SOFA) score has also been used for sepsis subtyping analysis. Longitudinal data studies are beneficial for revealing the characteristics of sepsis, a disease that evolves rapidly over a short period; however, how to utilize longitudinal data to construct subtyping methods with good predictive power and clinical feasibility remains an active area of ​​research. Summary of the Invention

[0004] The purpose of this application is to provide a method for classifying sepsis, a model for constructing a classification prediction model, an application method and device, which can calculate and obtain the dynamic blood flow oxygen flow index based on dynamic central venous oxygen saturation and peripheral perfusion index, classify sepsis, and then use each sepsis classification and its corresponding patient clinical characteristics to train a machine learning model to achieve accurate prediction of sepsis classification.

[0005] To achieve the above objectives, this application provides the following solution:

[0006] Firstly, this application provides a method for classifying sepsis subtypes, including:

[0007] Obtain the central venous oxygen saturation and peripheral perfusion index of several patients within a preset time period;

[0008] The dynamic blood flow oxygen flow index trajectory for each patient within a preset time period is obtained based on the patient's central venous oxygen saturation and peripheral perfusion index; the blood flow oxygen flow index is calculated based on the central venous oxygen saturation and peripheral perfusion index.

[0009] The dynamic time warping algorithm was used to evaluate the trajectory similarity of the blood oxygen flow index of several patients;

[0010] Based on the trajectory similarity of the blood oxygen flow index, a hierarchical clustering algorithm was used to group the trajectories of the dynamic blood oxygen flow index, resulting in multiple sepsis subtypes.

[0011] Secondly, this application provides a method for constructing a sepsis classification prediction model, including:

[0012] A machine learning model is trained using each sepsis subtype and its corresponding patient clinical characteristics to derive a sepsis subtype prediction model. This sepsis subtype prediction model is used to predict the sepsis subtype based on the patient's clinical characteristics. Each sepsis subtype is determined based on the above-mentioned sepsis subtype classification method.

[0013] Thirdly, this application provides a method for applying a sepsis classification prediction model, including:

[0014] Obtain the patient's clinical characteristics within the target time period;

[0015] The patient's clinical characteristics within the target time period are input into the sepsis classification prediction model to obtain the sepsis classification; the sepsis classification prediction model is derived based on the above-mentioned sepsis classification prediction model construction method based on dynamic blood flow oxygenation.

[0016] Fourthly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described sepsis classification method, or the above-described sepsis classification prediction model construction method, or the above-described sepsis classification prediction model application method.

[0017] Fifthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described sepsis classification method, or the above-described sepsis classification prediction model construction method, or the above-described sepsis classification prediction model application method.

[0018] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0019] This application provides a method for sepsis classification, a classification prediction model construction, application method, and device. Based on central venous oxygen saturation and peripheral perfusion index, this application calculates the dynamic blood flow oxygen flow index. Then, using dynamic time warping and hierarchical clustering algorithms, it clusters the trajectory of the dynamic blood flow oxygen flow index to obtain multiple sepsis classifications. This demonstrates that the clinical outcomes of patients in each classification are different. Based on this, a machine learning model is trained to derive a sepsis classification prediction model. This application classifies sepsis based on longitudinal data of central venous oxygen saturation and peripheral perfusion index, and establishes a prediction model for this classification based on clinical characteristics, ultimately achieving the segmentation and accurate prediction of the sepsis population. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a diagram illustrating the application environment of a sepsis classification method according to an embodiment of this application.

[0022] Figure 2 A flowchart illustrating a sepsis classification method according to an embodiment of this application;

[0023] Figure 3 A schematic diagram illustrating the patient screening method required for sepsis classification according to an embodiment of this application;

[0024] Figure 4 A schematic diagram of the 72-hour dynamic blood oxygen flow index trajectory curve of the training set provided in an embodiment of this application;

[0025] Figure 5 A schematic diagram of the 72-hour dynamic blood oxygen flow index trajectory curve provided for an embodiment of this application;

[0026] Figure 6 The following is a proportional histogram of each group in the training and validation sets provided in an embodiment of this application;

[0027] Figure 7 A schematic diagram of the PISO exponential trajectory of the training set during clustering using the Time Warping Algorithm (HAC) + Hierarchical Clustering Algorithm (DTW) method provided in an embodiment of this application;

[0028] Figure 8 A schematic diagram of the PISO exponential trajectory of the training set when clustering using the group-based trajectory modeling (GBTM) method according to an embodiment of this application;

[0029] Figure 9 A schematic diagram illustrating the sensitivity analysis of the HAC+DTW method and GBTM method for classification according to an embodiment of this application;

[0030] Figure 10 A SHAP (SHapley Additive ex Planations) diagram corresponding to type 1 provided in an embodiment of this application;

[0031] Figure 11 The SHAP value corresponding to type 1 provided in one embodiment of this application;

[0032] Figure 12 This is a SHAP diagram corresponding to type 2 provided in one embodiment of this application;

[0033] Figure 13 The SHAP value corresponding to type 2 provided in one embodiment of this application;

[0034] Figure 14 This is a SHAP diagram corresponding to type 3 provided in an embodiment of this application;

[0035] Figure 15 The SHAP value corresponding to type 3 provided in one embodiment of this application;

[0036] Figure 16 A flowchart illustrating an application method for a sepsis classification prediction model provided in an embodiment of this application;

[0037] Figure 17 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0038] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0039] Central venous oxygen saturation (ScvO2) and peripheral perfusion index (PI) are two hemodynamic parameters that can be obtained in real time and are particularly crucial. Central venous oxygen saturation, measured by blood gas analysis of blood from the superior vena cava, reflects the balance between oxygen delivery and utilization in the body; peripheral perfusion index, obtained by monitors, reflects peripheral tissue perfusion. These two indicators have been widely used in clinical hemodynamic assessment, and numerous studies have shown that each indicator has significant assessment value for critically ill patients and can predict patient outcomes in sepsis patients. However, there are currently few studies on sepsis classification based on hemodynamic characteristics; no study has combined central venous oxygen saturation and peripheral perfusion index for sepsis classification; furthermore, many sepsis classification studies are based on cross-sectional data, which cannot reflect the rapid evolution of sepsis within a short period.

[0040] In response, this application proposes a method for classifying sepsis subtypes, a subtype prediction model construction, an application method, and a device. The method classifies sepsis subtypes based on the dynamic blood flow oxygen flow index calculated from central venous oxygen saturation and peripheral perfusion index. Then, it uses each sepsis subtype and its corresponding patient clinical characteristics to train a machine learning model to achieve accurate prediction of sepsis subtypes.

[0041] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] The sepsis classification method provided in this application is specifically a sepsis classification method based on dynamic blood flow oxygenation, which can be applied to, for example... Figure 1The application environment is shown. The terminal communicates with the server via a network. The data storage system stores the data the server needs to process. The data storage system can be set up independently, integrated into the server, or placed in the cloud or on another server. The terminal can send the central venous oxygen saturation and peripheral perfusion index of several patients within a preset time period to the server. After receiving the central venous oxygen saturation and peripheral perfusion index of several patients within the preset time period, the server calculates the trajectory of the dynamic blood flow oxygen flow index for each patient within the preset time period based on the central venous oxygen saturation and peripheral perfusion index. The blood flow oxygen flow index is calculated based on the central venous oxygen saturation and peripheral perfusion index. The similarity of the blood flow oxygen flow index trajectories of several patients is evaluated using a dynamic time warping algorithm. Based on the similarity of the blood flow oxygen flow index trajectory, a hierarchical clustering algorithm is used to group the blood flow oxygen flow index trajectories, resulting in multiple sepsis subtypes. In addition, in some embodiments, the sepsis classification method can also be implemented by the server or the terminal alone. For example, the terminal can directly determine the sepsis classification, or the server can obtain the patient's central venous oxygen saturation and peripheral perfusion index from the data storage system and classify the sepsis classification.

[0043] The terminal can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. The server can be a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0044] In one exemplary embodiment, such as Figure 2 As shown, a method for classifying sepsis is provided. This method is executed by a computer device, specifically a terminal or server, or both. In this embodiment, the method is applied to... Figure 1 The following steps, 101 to 105, are used as an example of the server in the example.

[0045] Step 101: Obtain the central venous oxygen saturation and peripheral perfusion index of several patients within a preset time period.

[0046] Step 102: Calculate and obtain the dynamic blood flow oxygen flow index trajectory for each patient within a preset time period based on the patient's central venous oxygen saturation and peripheral perfusion index within the preset time period; the blood flow oxygen flow index is calculated based on the central venous oxygen saturation and peripheral perfusion index.

[0047] Step 103: Calculate the trajectory similarity of the blood oxygen flow index of several patients using the dynamic time warping algorithm.

[0048] Step 104: Based on the trajectory similarity of the blood oxygen flow index, a hierarchical clustering algorithm is used to group the trajectories of the blood oxygen flow index to obtain multiple sepsis subtypes.

[0049] By implementing steps 101 to 104 above, and addressing the current scarcity of studies on sepsis classification based on hemodynamic characteristics, and the lack of research combining central venous oxygen saturation and peripheral perfusion index for sepsis classification, this application calculates and obtains the dynamic blood flow oxygen flow index based on central venous oxygen saturation and peripheral perfusion index. Then, it uses dynamic time warping algorithm and clustering algorithm to cluster the trajectory of the dynamic blood flow oxygen flow index, resulting in multiple sepsis classifications. This demonstrates that the clinical outcomes of patients in each classification are different, and that it can accurately classify various types of sepsis.

[0050] The data collection in step 101 came from the Intensive Care Unit (ICU) of Peking Union Medical College Hospital, specifically from structured medical records such as the patient's cover page, admission record, and progress notes. The included patients were admitted between January 1, 2015, and December 31, 2023. The inclusion criteria were patients aged ≥18 years. Exclusion criteria included: missing key demographic information (such as age or sex), or missing central venous oxygen saturation and peripheral perfusion index measurements (measured every 6 hours) for 12 consecutive time points within 72 hours of a sepsis diagnosis.

[0051] The total number of patients meeting the inclusion and exclusion criteria for this application was 12,975. Of these, 10,525 developed the defined infection after admission to the ICU, and 5,205 met the criteria for organ failure within the time window before and after infection onset. Ultimately, 720 patients were able to provide central venous oxygen saturation and peripheral perfusion index measurements at 12 consecutive time points (every 6 hours) within 72 hours of sepsis diagnosis. The specific enrollment process is as follows... Figure 3 As shown.

[0052] The criteria for diagnosing sepsis were defined, and sepsis patients were screened from the population based on these criteria to classify sepsis subtypes using relevant data from the screened patients. According to the Sepsis 3.0 criteria, sepsis is defined as an infection accompanied by organ failure. The specific definition and screening process is as follows:

[0053] (1) Infection Occurrence: The patient develops an infection after admission to the ICU and meets the following conditions within a specific time window: First, a pathogen culture order is issued (regardless of whether the culture result is negative or positive); second, the patient receives therapeutic antibiotic treatment. Therapeutic antibiotics are defined as continuous use of a single antibiotic for more than 48 hours. Regarding the specific time window, if the pathogen culture order is issued before the therapeutic antibiotic, the therapeutic antibiotic must be started within 72 hours after the pathogen culture order is issued; if the pathogen culture order is issued after the therapeutic antibiotic, the pathogen culture must be performed within 24 hours after the start of antibiotic use, and the time of start of therapeutic antibiotic use shall be used as the "infection onset time".

[0054] (2) Organ Failure: Within a defined time window before and after the onset of infection (48 hours before and 24 hours after the onset of infection), the patient's Sequential Organ Dysfunction Assessment (SOFA) score must increase by at least 2 points compared to the baseline score before ICU admission. The SOFA score is a scale used to assess the degree of organ failure in sepsis patients, covering sub-scores of six major organ systems: respiratory, coagulation, liver, cardiovascular, central nervous system, and kidney. Each sub-score is calculated every 6 hours based on the patient's data within the first 72 hours after ICU admission, using the worst variable value within that time period. Special attention should be paid to the following when calculating the SOFA score: For the assessment of the central nervous system, the lowest Glasgow Coma Scale (GCS) is used, assessed based on the patient's state before sedation. The baseline SOFA score is the patient's SOFA score before ICU admission; if no relevant biochemical tests were performed before ICU admission, the baseline SOFA score is considered to be 0.

[0055] In step 102, based on the connotations of central venous oxygen saturation and peripheral perfusion index, this application innovatively designed the PISO index (blood oxygen flow index) and its formula: ,in, This represents the critical value used to predict poor patient outcomes (death, mechanical ventilation, ICU admission, etc.) based on PI. This formula allows for simultaneous comparison of PI and ScvO2 levels, and determines the outcome based on the relationship between PI and the critical value. The direction confirms the positive or negative value.

[0056] In order to establish The PISO index can effectively predict poor patient outcomes. Based on the range of critical values ​​for central venous oxygen saturation and peripheral perfusion index reported in existing literature, it is assumed that... The values ​​were 0.7, 0.8, 0.9, 1.0, and 1.1, respectively, and then different... Under the given values, hierarchical clustering (HAC) and dynamic time warping (DTW) methods were used to cluster the PISO index trajectories of the population into three subtypes. Analysis of variance was used to compare the differences in 28-day mortality rates among the subtypes. By calculating F-values ​​and P-values, the influence of each parameter value on prognostic indicators after clustering was determined, revealing... When the value is 1.1, the three subtypes show the greatest prognostic difference, thus determining... The optimal value is 1.1, as shown in Table 1. In analysis of variance, the F-value and P-value are two core statistics used to determine whether differences between groups are significant. The F-value measures the degree of difference in means between different groups relative to the degree of variation within individuals in the group. The P-value is the probability of observing the current or more extreme F-value under the null hypothesis.

[0057] Table 1 differs Differences in various prognostic indicators under different value conditions

[0058]

[0059] In Table 1, the data for mechanical ventilation duration and ICU stay within 28 days are shown. The value before the brackets [] is the median, and the values ​​inside the brackets [] are the first and third quartiles, respectively.

[0060] In steps 103 and 104, after the PISO index calculation formula was established, the data for each included patient consisted of 12 PISO index vectors generated every 6 hours over 72 hours. The study population was randomly divided into training and validation sets at a 4:1 ratio to implement internal validation of the model. Subsequently, the similarity of PISO index trajectories was evaluated using the DTW method on the training set, and the PISO index trajectories were grouped according to similarity using the HAC method.

[0061] To determine the optimal number of clusters for each subtype, this application employed various clustering performance evaluation metrics, including the silhouette coefficient, Davies-Bouldin index, variance ratio, Bayesian Information Criterion (BIC), Within-Cluster Sum of Squares (WCSS), Hartigan's test, and Standard Deviation index (SD). Ultimately, the optimal number of clusters was determined to be 3, as shown in Table 2, representing 3 subtypes.

[0062] Table 2 Clustering Performance Evaluation

[0063]

[0064] Subsequently, the HAC+DTW method was repeated on the validation set, and consistent grouping results were found between the training and validation queues, such as... Figures 4 to 6 As shown, Figure 4 This shows the number of clusters in the training set. Figure 5 The number of clusters in the validation set is shown. Figure 6 The histogram of the number of people in each cluster between the two queues is shown.

[0065] Furthermore, to ensure the robustness of the derived classification, this application further employs a group-based trajectory modeling method (GBTM). GBTM is a latent class analysis method that calculates the probability of each patient belonging to different classifications through maximum likelihood estimation. Sensitivity analysis results of the HAC+DTW method and the GBTM method show that the two methods have good consistency, such as... Figures 7 to 9 As shown.

[0066] Observing the trajectory characteristics of each subtype, in the training set, subtype 1 patients were the fewest (n=140, 24.3%), and their trajectory was characterized by a continuous increase in the PISO index within the first 54 hours after sepsis diagnosis, followed by a decrease. Subtype 2 patients (n=246, 42.7%) showed a slow increase in the PISO index from the onset of sepsis to 72 hours, and this was the most common phenotype. Subtype 3 patients (n=190, 33.0%) were characterized by a slow decrease in the PISO index from a negative baseline within 72 hours. Similar trajectory trends were obtained for the above subtypes in the validation set. Here, n represents the number of patients in the corresponding subtype, and percentage represents the percentage of patients in that subtype out of the total number of patients.

[0067] In another exemplary embodiment of this application, a method for constructing a sepsis subtyping prediction model is provided, comprising:

[0068] A machine learning model is trained using each sepsis subtype and its corresponding patient clinical characteristics to derive a sepsis subtype prediction model. This sepsis subtype prediction model is used to predict the sepsis subtype a patient belongs to based on their clinical characteristics. Each sepsis subtype is determined based on the sepsis subtype classification method described in steps 101 to 104 above.

[0069] In this application, multiple sepsis subtypes are analyzed and derived, and a machine learning model is then trained based on these subtypes to obtain a sepsis subtype prediction model. This application classifies sepsis subtypes based on the blood flow oxygen flow index calculated from central venous oxygen saturation and peripheral perfusion index, which can leverage the advantages of real-time, longitudinal data and easy availability of central venous oxygen saturation and peripheral perfusion index to achieve accurate prediction of sepsis subtypes.

[0070] Based on the three PISO index-based classifications established in the preceding steps, the differences in prognosis after diagnosis of sepsis among patients of each classification were subsequently compared. Prognostic indicators included the incidence of acute kidney injury (AKI), ICU stay, mechanical ventilation time, and 28-day mortality. The results showed that patients in classification 3 had the worst prognosis, exhibiting significant differences compared to the other two classifications in AKI incidence, ICU stay, mechanical ventilation time, and 28-day mortality (P < 0.05), as shown in Table 3. Although patients in classification 1 showed a trend of being better than those in classification 2 in the above indicators, no difference in prognosis between patients in classification 1 and classification 2 was found. Therefore, this application identifies and demonstrates the clinical significance of sepsis classification.

[0071] Table 3. Differences in prognosis among patients with different subtypes after diagnosis of sepsis

[0072]

[0073] The clinical manifestations of the three sepsis subtypes differed significantly within 6 hours of diagnosis (baseline), as shown in Table 4. Based on these differences, several machine learning models were constructed and trained to predict the blood flow oxygenation subtype of patients using clinical characteristics at the onset of sepsis. 80% of the dataset, consisting of the clinical characteristics and corresponding subtypes of several sepsis patients, was used for training, and the remaining 20% ​​was used as the test set. The test sample sizes for subtypes 1, 2, and 3 were 35, 62, and 47, respectively. To evaluate the model performance, precision, recall, and F1 score were used, and the results are summarized in Table 5. The results show that the K-Nearest Neighbors (KNN) model performed the worst, exhibiting low recognition ability across all test samples. Support Vector Machines (SVM) performed slightly better than KNN, but overall still lagged behind other more complex models. Multilayer Perceptrons (MLP) showed improved performance compared to KNN and SVM, demonstrating better classification ability. Among various tree-based ensemble methods, XGBoost, Random Forest, and LightGBM showed relatively small performance differences, but Random Forest performed best among the three models. Specifically, Random Forest achieved F1 scores of 0.78, 0.82, and 0.93 for the three subtypes, demonstrating high accuracy and stability. Particularly in the identification of subtype 3, the F1 score reached as high as 0.93, exhibiting extremely strong predictive ability. Based on the classification results of the Random Forest model, SHAP plots were further used to illustrate the direction and importance of the Top 20 features in predicting the three subtypes (subtype 1, subtype 2, and subtype 3) based on the PISO index. Figures 10 to 15As shown in the SHAP plot, blue to red represents the magnitude of the feature value. The PI shows a distribution trend from the left to the right of the coordinate axis, from smaller feature values ​​(blue) to larger feature values ​​(red and purple), indicating that high-valued PI features have a positive impact on the model prediction results. Figure 10 , Figure 12 and Figure 14 The “SHAP value” in the text indicates the impact on the model output; Figure 11 , Figure 13 and Figure 15 "mean" in ")" indicates the average effect on the model output size.

[0074] Table 4. Statistical Analysis Results of Clinical Characteristics

[0075]

[0076]

[0077] In Table 4, mean (SD) represents the average (standard deviation); median [Q1, Q3] represents the median (first quartile and third quartile). The units for each clinical characteristic in Table 4 are: age (years), ScvO2 (%), and body mass index (BMI) (kg / m²). 2 Body temperature (°C), heart rate (beats / min), respiratory rate (breaths / min), mean arterial pressure (mmHg), systolic blood pressure (mmHg), diastolic blood pressure (mmHg), lymphocyte percentage (%), mean corpuscular hemoglobin (g / L), prothrombin time (s), indirect bilirubin ( Albumin concentration (g / L), blood urea nitrogen concentration (12.3 mmol / L), platelet concentration (10... 9 / L), white blood cell count (10 9 / L), hematocrit (%), fibrinogen (g / L), platelet distribution width (fl), mean platelet volume (fl), lymphocyte count (10 9 / L), activated partial thromboplastin time (seconds), absolute neutrophil count (10 9 / L), platelet count (%), activated partial thromboplastin time ratio (%), D-dimer concentration (mg / L), creatinine concentration ( Alanine aminotransferase (U / L), positive end-expiratory pressure (PEEP) (cmH2O), direct bilirubin ( ), total bilirubin ( ), Procalcitonin (ng / m l ).

[0078] Table 5. Performance results of each model on the test set.

[0079]

[0080] Therefore, in step 205, a machine learning model is trained using each sepsis subtype and its corresponding patient clinical characteristics to derive a sepsis subtype prediction model, specifically including:

[0081] (1) Use the sepsis subtypes and their corresponding patient clinical characteristics to train multiple machine learning models.

[0082] (2) Evaluate the performance of each trained machine learning model.

[0083] (3) Select the best-performing post-trained machine learning model as the sepsis subtyping prediction model.

[0084] Furthermore, the method for constructing the sepsis classification prediction model also includes:

[0085] (1) Based on the classification results of the best-performing trained machine learning model, the influence direction and importance of each clinical feature on the model output are determined by combining the SHAP diagram.

[0086] (2) Determine the key clinical features for predicting sepsis subtypes based on the direction and importance of each clinical feature's influence on the model output. Key clinical features refer to the top-ranked clinical features in the SHAP plot corresponding to each clinical feature.

[0087] Figure 7 All clinical features in the SHAP graph are used as model inputs to build the model and analyze the key clinical features that have a significant impact on the prediction results. In actual prediction scenarios, the final model input is selected based on the top few clinical features ranked by importance in the SHAP graph. The specific number can be determined manually.

[0088] This application classifies sepsis subtypes based on clinically available real-time central venous oxygen saturation and peripheral perfusion index. Specifically, it calculates the PISO index using these two metrics, then performs similarity and clustering calculations on the PISO index trajectories, ultimately identifying three subtypes. This constructs a sepsis subtyping prediction model capable of predicting sepsis subtypes based on the PISO index. This application provides a reliable sepsis subtyping method and sepsis subtyping prediction model based on readily available, real-time, and dynamic clinical data.

[0089] In another exemplary embodiment of this application, such as Figure 16 As shown, an application method for a sepsis classification prediction model is provided, including:

[0090] S1: Obtain the patient's clinical characteristics within the target time period. The clinical characteristics within the target time period refer to the top-ranked clinical characteristics in the SHAP graph corresponding to each clinical characteristic. For example, this could be obtaining clinical characteristics within 6 hours of sepsis diagnosis, or clinical characteristics from other time periods.

[0091] S2: Input the patient's clinical characteristics within the target time period into the sepsis classification prediction model to obtain the sepsis classification; the sepsis classification prediction model is derived based on the above-mentioned sepsis classification prediction model construction method.

[0092] This application also provides an application scenario that utilizes the aforementioned sepsis classification method, sepsis classification prediction model construction method, and sepsis classification prediction model application method. Specifically, it can be applied to a patient sepsis classification prediction scenario. This scenario includes a sepsis classification stage, a sepsis classification prediction model construction stage, and a patient sepsis classification prediction stage. The sepsis classification construction stage calculates the PISO index based on the central venous oxygen saturation and peripheral perfusion index of sepsis patients, and performs clustering to obtain multiple sepsis classifications based on the PISO index. The sepsis classification prediction model construction stage trains a machine learning model based on this classification to construct a sepsis classification prediction model. The patient sepsis classification prediction stage applies the sepsis classification prediction model to determine the sepsis classification based on the clinical characteristics of the sepsis patient.

[0093] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 17 As shown. The computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs in the non-volatile storage media to run. The database stores relevant data required for constructing sepsis classification and sepsis classification prediction models, as well as the final constructed sepsis classification prediction model. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the aforementioned sepsis classification method, or the sepsis classification prediction model construction method, or the sepsis classification prediction model application method.

[0094] Those skilled in the art will understand that Figure 17 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described sepsis classification method, or the sepsis classification prediction model construction method, or the sepsis classification prediction model application method.

[0095] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the above-described sepsis classification method, or the sepsis classification prediction model construction method, or the sepsis classification prediction model application method.

[0096] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0097] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0098] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for classifying sepsis subtypes, characterized in that, include: Obtain the central venous oxygen saturation and peripheral perfusion index of several patients within a preset time period; The dynamic blood flow oxygen flow index trajectory for each patient within a preset time period is calculated and obtained based on the patient's central venous oxygen saturation and peripheral perfusion index. The blood flow oxygen flow index is calculated based on the central venous oxygen saturation and peripheral perfusion index. The dynamic time warping algorithm was used to evaluate the trajectory similarity of the blood oxygen flow index of several patients; Based on the trajectory similarity of the blood oxygen flow index, a hierarchical clustering algorithm was used to group the trajectories of the blood oxygen flow index, resulting in three sepsis subtypes. The formula for calculating the blood oxygen flow index is as follows: PISO = (PI - ∂) × ScvO2; Wherein, PISO represents the blood oxygen flow index; PI represents the peripheral perfusion index; ScvO2 represents the central venous oxygen saturation; and ∂ represents the critical value for predicting poor patient prognosis based on the peripheral perfusion index PI.

2. A method for constructing a sepsis classification prediction model, characterized in that, include: A machine learning model is trained using each sepsis subtype and its corresponding patient clinical characteristics to derive a sepsis subtype prediction model; the sepsis subtype prediction model is used to predict the sepsis subtype to which the patient belongs based on the patient's clinical characteristics; each sepsis subtype is determined based on the sepsis subtype classification method described in claim 1.

3. The method for constructing a sepsis classification prediction model according to claim 2, characterized in that, A machine learning model was trained using various sepsis subtypes and their corresponding patient clinical characteristics to derive a sepsis subtype prediction model, which specifically includes: Multiple machine learning models were trained using various sepsis subtypes and their corresponding patient clinical characteristics. Evaluate the performance of each trained machine learning model. The best-performing trained machine learning model was selected as the sepsis subtyping prediction model.

4. The method for constructing a sepsis classification prediction model according to claim 3, characterized in that, The method for constructing the sepsis classification prediction model also includes: Based on the classification results of the best-performing trained machine learning model, the influence direction and importance of each clinical feature on the model output are determined by combining the SHAP plot. The key clinical features for predicting sepsis subtypes are determined based on the direction and importance of the influence of each clinical feature on the model output.

5. The method for constructing a sepsis classification prediction model according to claim 4, characterized in that, The clinical characteristics include: sex, age, body mass index, central venous oxygen saturation, peripheral perfusion index, body temperature, heart rate, respiratory rate, mean arterial pressure, systolic blood pressure, diastolic blood pressure, lymphocyte percentage, mean corpuscular hemoglobin, prothrombin time, mean corpuscular hemoglobin concentration, indirect bilirubin measurement, albumin concentration, blood urea nitrogen concentration, platelet concentration, white blood cell count, hematocrit, fibrinogen content measurement, platelet distribution width, mean platelet volume, lymphocyte count, activated partial thromboplastin time measurement, hemoglobin concentration, absolute neutrophil count, plateletcrit, SOFA total score, activated partial thromboplastin time ratio, vasoactive drug score, D-dimer concentration, creatinine concentration, alanine aminotransferase measurement, positive end-expiratory pressure, direct bilirubin measurement, total bilirubin measurement, and procalcitonin measurement.

6. The method for constructing a sepsis classification prediction model according to claim 5, characterized in that, The key clinical features refer to the top-ranked clinical features in the SHAP graph corresponding to each clinical feature.

7. A method for applying a sepsis classification prediction model, characterized in that, include: Obtain the patient's clinical characteristics within the target time period; The patient's clinical characteristics within the target time period are input into the sepsis classification prediction model to obtain the sepsis classification; the sepsis classification prediction model is derived based on the sepsis classification prediction model construction method described in any one of claims 2 to 6.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the sepsis classification method of claim 1, or the sepsis classification prediction model construction method of any one of claims 2-6, or the sepsis classification prediction model application method of claim 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the sepsis classification method of claim 1, or the sepsis classification prediction model construction method of any one of claims 2-6, or the sepsis classification prediction model application method of claim 7.

Images