A biomarker for predicting the prognosis of sepsis and its application

Abstract

The present invention discloses a biomarker and application for predicting the prognosis of sepsis, belonging to the field of biotechnology. The biomarkers include neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, monocyte-to-lymphocyte ratio, white blood cell-to-lymphocyte ratio, systemic inflammatory response index and peripheral inflammation value. The present invention utilizes neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, monocyte-to-lymphocyte ratio, white blood cell-to-lymphocyte ratio, systemic inflammatory response index and peripheral inflammation value to accurately stratify the prognosis risk of sepsis patients and predict the prognosis survival risk of sepsis patients, significantly improving the accuracy and timeliness of prognostic assessment of sepsis patients, providing a reliable guiding basis for clinical decision-making, and having important clinical application value and promotion significance.

Description

Technical Field

[0001] The present invention relates to the field of biotechnology, and in particular to a biomarker for predicting the prognosis of sepsis and its application. Background Art

[0002] Sepsis is a leading cause of death in intensive care units, with a mortality rate as high as 25-30%, and even as high as 40-50% for patients with septic shock. Despite significant advances in critical care medicine, the diagnosis and treatment of sepsis still face two fundamental challenges: the lack of accurate early risk stratification tools and insufficient understanding of the mechanisms of cellular dysfunction during disease progression.

[0003] The Sequential Organ Failure Assessment (SOFA) system, currently widely used in clinical practice, has significant limitations: its reliance on complex organ dysfunction parameter assessment often delays critical therapeutic interventions. Although recent studies have shown that inflammatory markers, particularly the neutrophil-to-lymphocyte ratio (NLR) and platelet-to-lymphocyte ratio (PLR), have potential prognostic value in various pathological conditions, the molecular mechanisms underlying these readily accessible clinical parameters remain unclear, limiting their application in guiding targeted therapeutic interventions.

[0004] In addition, current methods for stratifying sepsis patients either lack molecular mechanism explanations, require complex testing methods, and are not suitable for acute clinical decision-making. Although cellular dysfunction is the basis for driving the progression of sepsis, the molecular programs corresponding to elevated inflammatory markers remain undefined. The progression of sepsis involves the interaction of complex molecular mechanisms such as neutrophil dysfunction, platelet activation, and metabolic reprogramming, but existing assessment methods have difficulty in effectively integrating this information. Although single-cell technology has revealed unprecedented cellular heterogeneity in sepsis, how to translate these molecular insights into clinically usable stratification tools and therapeutic targets still faces major challenges. Therefore, there is an urgent need for a method that can bridge the gap between rapid patient stratification and mechanistic insights in order to achieve targeted intervention. Summary of the Invention

[0005] The purpose of the present invention is to provide a biomarker for predicting the prognosis of sepsis and its application to address the problems existing in the above-mentioned prior art. The present invention utilizes the neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, monocyte-to-lymphocyte ratio, white blood cell-to-lymphocyte ratio, systemic inflammatory response index, and peripheral inflammation value to accurately stratify the prognostic risk of sepsis patients and predict the prognostic survival risk of sepsis patients. This significantly improves the accuracy and timeliness of prognostic assessment for sepsis patients, provides a reliable guide for clinical decision-making, and has important clinical application value and promotion significance.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] The present invention provides a biomarker for predicting the prognosis of sepsis, wherein the biomarker includes a neutrophil-to-lymphocyte ratio, a platelet-to-lymphocyte ratio, a monocyte-to-lymphocyte ratio, a leukocyte-to-lymphocyte ratio, a systemic inflammatory response index, and a peripheral inflammation value;

[0008] The calculation formula of the systemic inflammatory response index is: systemic inflammatory response index = neutrophil count × monocyte count / lymphocyte count;

[0009] The calculation formula of the peripheral inflammation value is: peripheral inflammation value = platelet count × neutrophil count / lymphocyte count.

[0010] The present invention also provides the use of the above biomarkers in preparing products for predicting the prognosis of sepsis.

[0011] Optionally, the product for predicting the prognosis of sepsis includes a sepsis prognosis risk stratification system for classifying sepsis patients into high inflammation risk group and low inflammation risk group and a sepsis prognosis survival risk prediction system for predicting the in-hospital mortality rate and 28-day mortality rate of sepsis patients.

[0012] The present invention also provides a sepsis prognosis risk stratification system, which includes a risk stratification module;

[0013] The risk stratification module is used to classify patients into high inflammation risk group and low inflammation risk group based on the patient data of the biomarkers using a prediction model and output the risk stratification results;

[0014] The prediction model is constructed based on a machine learning algorithm using patient data of the biomarkers in the population.

[0015] Optionally, the machine learning algorithm can be selected from any one of a support vector machine classifier, a random forest algorithm, a gradient boosting decision tree, a deep neural network and a logistic regression model.

[0016] Furthermore, the classification method is to set a threshold value based on the median value of the group patient data to classify the patients.

[0017] Optionally, the sepsis prognostic risk stratification system further includes a data collection module; the data collection module is used to collect hematological examination data of the patient.

[0018] Optionally, the data collection module is also used to preprocess the patient's hematological examination data; the preprocessing process is: standardizing the acquired hematological examination data; eliminating outliers and missing values; performing data quality control; the standardization method can be selected from any one of central logarithmic ratio transformation, Min-Max standardization, quantile standardization and robust standardization.

[0019] The data collection module can be connected to the hospital information system (HIS), automatically extract blood routine test data, and establish a data cache and backup mechanism.

[0020] Optionally, the sepsis prognostic risk stratification system further includes a data calculation module; the data calculation module is configured to calculate the patient data of the biomarkers based on the hematological examination data.

[0021] Optionally, the data calculation module is further used to perform standardization processing on the calculated patient data; the process of the standardization processing is: performing central logarithmic ratio transformation on the calculated patient data; performing z-score standardization; and eliminating outliers and missing values.

[0022] Optionally, the sepsis prognosis risk stratification system further includes a dynamic monitoring module; the dynamic detection module is used to periodically and repeatedly measure the patient data of the biomarkers; record the changing trend of the patient data; and trigger an early warning signal when the change in the patient data exceeds a preset threshold.

[0023] Optionally, the warning signal includes three levels of warning signals: a first-level signal is triggered when the patient data of a single biomarker exceeds a preset threshold; a second-level signal is triggered when the patient data of two or more biomarkers exceeds a preset threshold; and a third-level signal is triggered when the patient data of multiple biomarkers continues to be abnormal or continuously deteriorates.

[0024] Optionally, the frequency of dynamic monitoring is determined according to the patient's risk level: the measurement data and patient prognostic risk stratification results of the high inflammation risk group are updated every 4 hours; the measurement data and patient prognostic risk stratification results of the low inflammation risk group are updated every 12 hours.

[0025] The dynamic monitoring and early warning module can be connected to a medical workstation to automatically push early warning signals.

[0026] Optionally, the sepsis prognosis risk stratification system further includes a treatment requirement prediction module; the treatment requirement prediction module is used to predict the probability of treatment requirement based on the patient's risk stratification results; the treatment requirements include mechanical ventilation requirements, airway management requirements, fluid therapy requirements, nutritional support requirements and renal replacement therapy requirements.

[0027] The present invention also provides a sepsis prognosis survival risk prediction system, which includes a survival risk prediction module;

[0028] The survival risk prediction module is used to predict the patient's in-hospital mortality rate and 28-day mortality rate using a prediction model based on the patient data of the biomarkers and the patient's basic clinical data, and output the prediction results;

[0029] The basic clinical data include anion gap, potassium ion concentration, chloride ion concentration, sodium ion concentration, body weight, albumin, bicarbonate, hemoglobin, alkaline phosphatase, alanine aminotransferase, aspartate aminotransferase, total bilirubin, mean corpuscular volume, age, blood urea nitrogen, international normalized ratio, prothrombin time and red blood cell volume distribution width;

[0030] The prediction model is constructed using a deep learning framework using patient data of biomarkers in the population and patient clinical basic data.

[0031] Optionally, the sepsis prognosis survival risk prediction system further includes a data collection module; the data collection module is used to collect the patient's basic clinical data and hematological examination data;

[0032] Optionally, the data collection module is also used to preprocess the patient's basic clinical data and hematological examination data; the preprocessing process is: standardizing the acquired basic clinical data and hematological examination data; eliminating outliers and missing values; performing data quality control; the standardization method can be selected from any one of central logarithmic ratio transformation, Min-Max standardization, quantile standardization and robust standardization.

[0033] Optionally, the sepsis prognosis survival risk prediction system further includes a data calculation module; the data calculation module is used to calculate the patient data of the biomarkers based on the hematological examination data.

[0034] Optionally, the data calculation module is further used to perform standardization processing on the calculated patient data; the process of the standardization processing is: performing central logarithmic ratio transformation on the calculated patient data; performing z-score standardization; and eliminating outliers and missing values.

[0035] The present invention also provides the application of the neutrophil-to-lymphocyte ratio and the platelet-to-lymphocyte ratio in constructing a septic shock risk prediction model.

[0036] The present invention discloses the following technical effects:

[0037] (1) Forecast accuracy significantly improved:

[0038] The prediction model constructed based on six inflammatory markers achieved excellent discriminative ability in predicting in-hospital mortality (AUC 0.77, 95% CI: 0.74-0.80), significantly outperforming the traditional SOFA score (AUC 0.70, 95% CI: 0.67-0.73; P < 0.001). The predictive performance of the proposed method was further improved in predicting 28-day mortality (AUC 0.79, 95% CI: 0.76-0.82), far exceeding the SOFA score (AUC 0.67, 95% CI: 0.64-0.70; P < 0.001). A support vector machine (SVM) model integrating the six inflammatory markers achieved extremely high predictive accuracy (AUC 0.86) and also demonstrated stable stratification in an external validation cohort.

[0039] (2) Early identification has significant effects:

[0040] The prediction model constructed based on six inflammatory indicators in this invention can achieve significant differentiation of survival curves within 72 hours of hospital admission, providing an important time window for early clinical intervention. It can accurately predict treatment needs. It has been verified that the high-inflammation risk group showed significantly increased treatment needs: mechanical ventilation needs (50.16% vs 34.52%, P < 0.001), airway management needs (32.70% vs 22.67%, P < 0.001), fluid therapy needs (49.96% vs 34.52%, P < 0.001), nutritional support needs (24.34% vs 16.39%, P < 0.001), and renal replacement therapy needs (8.32% vs 2.31%, P < 0.001).

[0041] (3) Obvious advantages in clinical practicality:

[0042] Easy to operate: Only routine blood test data is required, no additional equipment investment is required, and the test time is short, with results available in about 15-30 minutes.

[0043] Significant economic benefits: It can reduce testing costs, saving 50-70% of testing costs compared to traditional molecular marker testing; reduce hospitalization time, and shorten the average hospitalization time through early warning; optimize medical resource allocation: accurately predict treatment needs and improve resource utilization efficiency.

[0044] (4) Prediction stability:

[0045] Temporal stability: The differential pattern of inflammatory markers remained stable during the 14-day follow-up period, with non-survivors maintaining significantly elevated inflammatory marker levels (P<0.001).

[0046] Multicenter validation results: Stable predictive efficacy was demonstrated in both the MIMIC-IV database (n=14,350) and the external validation cohort (n=74), with consistent prediction accuracy across institutions.

[0047] (5) The value of individualized treatment guidance:

[0048] Precise stratification effect: Five patient types with different prognostic characteristics were successfully identified, and the 30-day survival rate showed significant differences among the types.

[0049] Prediction of treatment response: It can predict the potential benefits of different treatment methods and provide an objective basis for individualized adjustment of clinical treatment plans.

[0050] The present invention significantly improves the accuracy and timeliness of prognostic assessment for sepsis patients, provides a reliable guide for clinical decision-making, and has important clinical application value and promotion significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 Figure 1 shows the identification results of different sepsis types and their related clinical phenotypes in Example 1; A shows the clustering results of sepsis patients in the MIMIC-IV cohort; B shows the distribution violin plot of each inflammatory index and clinical parameter in the five types; C shows the distribution of hospital mortality (left) and 28-day mortality (right) in each type; D shows the Kaplan–Meier survival curves of different types (log-rank test, P < 0.0001);

[0053] Figure 2 The distribution of underlying diseases and complications of different types of sepsis in Example 1;

[0054] Figure 3 The following table shows the prediction results of the deep learning model in Example 2 for sepsis mortality; A is the comparison result of the ROC curve of the deep learning model and the ROC curve of the SOFA score; B is a schematic diagram of the Spearman correlation matrix between inflammatory indicators and clinical parameters: red indicates positive correlation, blue indicates negative correlation, and all related P values ​​are <0.001. Statistical significance was determined by comparing AUCs using the DeLong test and the Spearman rank correlation test;

[0055] Figure 4The results of the survival analysis of sepsis patients based on the six inflammatory indicators in Example 3 are shown; A is the distribution of the six inflammatory indicators between 28-day survivors and non-survivors in the MIMIC-IV cohort (n=14,350); B is the external validation result of the distribution of the six inflammatory indicators in the Longhua Hospital cohort (n=74);

[0056] Figure 5 The following are the results of sepsis prognostic risk stratification based on the six inflammatory indicators in Example 3; wherein, A is the Kaplan-Meier survival curve (log-rank test, P<0.0001) drawn after dividing patients into high inflammation group and low inflammation group according to the median of each inflammatory indicator at admission; B is the ROC curve of the SVM model based on the six inflammatory indicators in sepsis prognostic risk stratification; C is the comparison of mortality rates between high and low inflammatory risk groups in the Longhua Hospital cohort; D is the validation result of the SVM model in the Longhua Hospital cohort; E is the distribution of treatment interventions in high and low inflammatory risk groups in the MIMIC-IV cohort;

[0057] Figure 6 The longitudinal analysis results of inflammatory indicators in sepsis patients of different risk groups in Example 4;

[0058] Figure 7 The single-cell and deconvolution analysis results of inflammatory indicators in sepsis in Example 5; wherein A is the UMAP visualization result obtained based on peripheral blood single-cell RNA sequencing (GSE167363 dataset); B is the expression of classic marker genes annotated by cell type; C is the analysis result of cellular composition of healthy controls, sepsis survivors and non-survivors;

[0059] Figure 8 This is the deconvolution analysis based on the GSE154918 dataset (n=105) in Example 5, comparing the changes in cell type proportions in healthy controls (Hlty), infection (Inf1_P), sepsis (Seps_P), and septic shock (Shock_P) as the disease progresses;

[0060] Figure 9 The results of the analysis of inflammatory indicators during the progression of sepsis in Example 5 are shown; A represents the response pattern of the six inflammatory indicators during disease progression; B represents the receiver operating characteristic (ROC) curves of PLR and NLR for predicting the progression of septic shock (AUC>0.85);

[0061] Figure 10Figure 6 shows the results of single-cell transcriptome analysis of neutrophils during sepsis in Example 6; A is the UMAP visualization result; B is the expression of specific marker genes for each neutrophil subset; C is the analysis of neutrophil subset composition under each clinical outcome; D is the result of pseudo-time trajectory analysis, with the color gradient representing the pseudo-time process (dark blue: early stage; red: late stage); E is the neutrophil maturation characteristics of the disease progression stage (Hlty: healthy control; Inf1_P: infection; Seps_P: sepsis; Shock_P: septic shock) using single-sample gene set enrichment analysis (ssGSEA); F is the result of gene set enrichment analysis, with the circle size representing the number of genes and the color depth representing the statistical significance (corrected P value);

[0062] Figure 11 The single-cell transcriptome analysis results of neutrophils during sepsis in Example 6; wherein A is the enrichment curve analysis result; B is the effector molecule expression analysis result in different neutrophil subsets; C is the pathway-specific gene expression heat map;

[0063] Figure 12 The single-cell transcriptomic feature analysis results in Example 7; wherein A is the single-cell transcriptomic and UMAP dimensionality reduction analysis results of platelets; B is the violin plot of the differential expression of specific marker genes of each platelet subpopulation; C is the cell composition analysis result; D is the gene set enrichment analysis result; E is the pathway enrichment analysis result; F is the differential expression analysis result of key functional molecules;

[0064] Figure 13 The single-cell transcriptomic feature analysis results in Example 7; wherein A is the hierarchical clustering analysis result for pathway-specific transcripts; B is the Spearman rank correlation analysis result; and C is a similar Spearman rank correlation analysis result. DETAILED DESCRIPTION

[0065] Various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as limiting the present invention, but rather as a more detailed description of certain aspects, features, and embodiments of the present invention.

[0066] It should be understood that the terms described herein are intended only to describe particular embodiments and are not intended to limit the present invention. In addition, for numerical ranges herein, it should be understood that each intermediate value between the upper and lower limits of the range is also specifically disclosed. The intermediate value within any stated value or stated range, and each smaller range between any other stated value or intermediate value within the stated range, is also encompassed within the present invention. The upper and lower limits of these smaller ranges may be independently included or excluded within the scope.

[0067] Unless otherwise indicated, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art. Although only preferred methods and materials are described herein, any methods and materials similar or equivalent to those described herein may also be used in the practice or testing of the present invention. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods and / or materials associated with the documents. In the event of any conflict with any incorporated document, the contents of this specification shall prevail.

[0068] It will be apparent to those skilled in the art that various modifications and variations may be made to the specific embodiments described herein without departing from the scope or spirit of the invention. Other embodiments will be apparent to those skilled in the art from the description of the invention. The description and examples are intended to be illustrative only.

[0069] The words “include,” “including,” “have,” “contain,” etc. used in this document are open-ended terms, meaning including but not limited to.

[0070] The analysis cohort population in the following examples of the present invention was selected from the MIMIC-IV database, and a total of 50,914 ICU inpatients were extracted. After strict inclusion / exclusion criteria, patients with SOFA score <2 were excluded from preliminary screening, and patients who were not admitted for the first time and patients with short-term hospitalization <48 hours were subsequently excluded. In order to ensure the integrity of the data and minimize confounding variables, patients with potential monocyte and lymphocyte-related lesions, incomplete hematological data and excessive data loss were further excluded. An analysis cohort (MIMIC-IV cohort) comprising 14,350 patients with sepsis was eventually formed. In the following examples of the present invention, 74 patients with sepsis from Longhua Hospital (inclusion / exclusion criteria were the same as those for the MIMIC-IV cohort) were also included to form an independent cohort for external validation.

[0071] The calculation method of the inflammation index in the following embodiments of the present invention is as follows:

[0072] NLR (Neutrophil-to-Lymphocyte Ratio) = neutrophil count / lymphocyte count;

[0073] PLR (Platelet-to-Lymphocyte Ratio) = platelet count / lymphocyte count;

[0074] MLR (Monocyte-to-Lymphocyte Ratio) = monocyte count / lymphocyte count;

[0075] LLR (Leukocyte to Lymphocyte Ratio) = white blood cell count / lymphocyte count;

[0076] SIRI (Systemic Inflammation Response Index) = (neutrophil count × monocyte count) / lymphocyte count;

[0077] PIV (Peripheral Inflammation Value) = (platelet count × neutrophil count) / lymphocyte count.

[0078] Some of the method steps used in the following embodiments of the present invention are as follows:

[0079] Unsupervised clustering of patients in the MIMIC database: Patients in the MIMIC database were clustered based on 21 physiological and biochemical indicators and 6 inflammatory markers. Each indicator was logarithmically transformed and normalized, followed by PCA and identification of shared nearest neighbors (SNNs) between samples. At a resolution of 0.35, the SNN module was used to optimize the clustering algorithm and group the patients.

[0080] A support vector machine classifier (SVM) model was constructed based on six inflammatory markers. The dataset was divided into a training set and a validation set in an 8:2 ratio. The training data was subjected to feature normalization. A classification threshold was set to categorize patients into high- and low-inflammatory risk groups. The R package ROCR was used to plot receiver operating characteristic (ROC) curves based on the model's predicted scores and the true labels, and the area under the curve (AUC) was calculated.

[0081] The SVM model was validated using a comprehensive approach: 5-fold cross-validation in the MIMIC-IV cohort; external validation in the independent Longhua Hospital cohort; and temporal validation of predictive stability at different time points. For survival analysis, the median of six inflammatory markers was used as the threshold to stratify patients into high and low inflammation groups. The survival model was fitted using the survival package (version 3.2-13) in R, and Kaplan-Meier curves were plotted using the survminer package (version 0.4.9).

[0082] Deep Learning Model Construction: PyTorch (version 1.9.0) was used to implement the deep learning network, which consisted of a 24-node input layer, three hidden layers (128, 64, and 32 nodes, respectively) with ReLU activation and a dropout rate of 0.3, and a single-node sigmoid output layer for mortality prediction. The model was trained using the Adam optimizer (learning rate 0.001), a binary cross-entropy loss function, and mini-batches (batch size 64) for 100 epochs with early stopping (patience = 10). The dataset was split into training and testing sets with an 80% / 20% split. Model performance was evaluated using multiple metrics (AUROC, precision-recall curves, sensitivity / specificity at optimal thresholds, calibration curves, and net reclassification improvement compared to the Somatic Pathology (SOFA) score).

[0083] Single-cell RNA sequencing analysis method: Single-cell RNA sequencing data is derived from Gene Expression Omnibus (GEO), numbered GSE167363. Low-quality cells (UMI count <500, or UMI count >5000, or mitochondrial gene proportion >20%) were screened out, and 48,486 high-quality cells were retained for subsequent analysis. Specifically, standardization was performed according to the count matrix, and the 3,000 gene features with the highest variability were selected. Principal component analysis (PCA) was performed on the standardized expression of variable features. Subsequently, the Harmony method of R was used to integrate cells from different samples, and UMAP dimensionality reduction and neighboring cell analysis were performed based on the Harmony dimension. The clustering algorithm was optimized by the SNN (Shared Nearest Neighbor) module to identify cell populations at a resolution of 1. Differential gene expression analysis was used to identify marker genes for each cell population. Cell populations expressing known marker genes were annotated (B cells: CD79A, MS4A1; proliferating cells: MKI67, TOP2A; erythrocytes: HBB, HBA2; monocytes: CD14, LYZ; neutrophils: CSF3R, MMP9; NK cells: GNLY, NKG7; plasma cells: MZB1, JCHAIN; platelets: PPBP, PF4; T cells: CD3D, CD3E). Subpopulation analysis of neutrophils and platelets was performed similarly to the main population analysis described above. Cell trajectory analysis was performed using Monocle3.

[0084] Bulk RNA sequencing deconvolution analysis based on scRNA-seq data: Bulk RNA-seq data were obtained from GEO (GSE154918). Specifically, single-cell RNA sequencing (scRNA-seq) data were used as a reference, and the single-cell RNA quantity informed deconvolution (SQUID) method was used to estimate the proportions of individual cells in the bulk RNA-seq samples. Inflammatory markers were calculated based on relative cell proportions.

[0085] Statistical Analysis Methods: Statistical methods were selected based on data distribution characteristics. Continuous variables were presented as mean ± standard deviation or median (interquartile range) based on the results of the Shapiro-Wilk normality test. Intergroup comparisons were performed using the Student's t-test or Mann-Whitney U test (continuous variables) and the chi-square test or Fisher's exact test (categorical variables). Multiple-group comparisons were performed using one-way ANOVA or the Kruskal-Wallis test with Bonferroni correction. Correlation analyses included Spearman rank correlation analysis for nonnormal data, Pearson correlation analysis for normally distributed data, and partial correlation analysis with adjustment for confounding factors. Multivariate analyses included Cox proportional hazards regression (survival analysis), logistic regression (binary outcomes), and linear mixed models (longitudinal data evaluation).

[0086] Sample size assessment: Sample size adequacy was verified by comprehensive power analysis, ensuring 80% power to detect a hazard ratio ≥1.5 in survival analysis, a minimum cluster size ≥100 patients in cluster analysis, at least 10 events / variables in machine learning models, and 90% power to detect an area under the curve (AUC) difference of 0.1 in validation studies. All statistical analyses were performed in R software (version 4.1.0) with the use of survival (3.2-13), lme4 (1.1-27), stats (4.1.0), and rms (6.2-0) packages.

[0087] The methods involved in the following embodiments of the present invention, unless otherwise specified, are carried out according to conventional methods in the art.

[0088] Example 1

[0089] This example investigates the routine admission parameters and inflammatory markers of patients with sepsis and describes a stratification model for sepsis based on admission parameters. The analyzed data included 28 clinical variables (height, age, gender, SOFA score, weight, body fat index (BMI), albumin, alkaline phosphatase (ALK), alanine aminotransferase (ALT), anion gap (ANG), aspartate aminotransferase (AST), bicarbonate, blood urea nitrogen (BUN), calcium, chloride, C-reactive protein (CRP), total bilirubin (TB), high-density lipoprotein (HDL), low-density lipoprotein (LDL), hemoglobin, international normalized ratio (INR), mean corpuscular volume (MCV), platelet count (PLT), potassium, prothrombin time (PT), corpuscular volume distribution width (RDW), sodium, and cholesterol) and six inflammatory markers. Among them, the six inflammatory indicators are neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), monocyte-to-lymphocyte ratio (MLR), leukocyte-to-lymphocyte ratio (LLR), systemic inflammatory response index (SIRI) and peripheral inflammation value (PIV). After the central logarithmic ratio transformation of the data normalization, principal component analysis (PCA) and uniform manifold approximation and projection (UMAP) were used for dimensionality reduction. Five different patient types ( Figure 1 Cluster A) includes cluster 0 (excessive inflammation type), cluster 1 (early inflammatory response type), cluster 2 (physiological stability type), cluster 3 (late organ failure type), and cluster 4 (early organ dysfunction type).

[0090] There are different pathophysiological characteristics among these different types ( Figure 1B): Physiologically stable patients (cluster 2) exhibited optimal clinical parameters, characterized by stable body weight (75.2 ± 4.8 kg), albumin (0.75 ± 0.05 g / dL), bicarbonate (24.5 ± 2.1 mmol / L), and hemoglobin (12.8 ± 1.2 g / dL), as well as minimal inflammation (CRP: 28.5 ± 15.2 mg / L) and a low SOFA score (3.2 ± 1.1; all comparisons P < 0.001). The early inflammatory response patient (cluster 1) showed an initial transitional state with moderate electrolyte changes (chloride: 106.8 ± 3.2 mmol / L; sodium: 142.5 ± 2.8 mmol / L) while maintaining stable organ function (SOFA: 4.1 ± 1.3). In contrast, the hyperinflammatory phenotype (cluster 0) exhibited significant elevations in all inflammatory markers (mean fold change relative to the physiologically stable phenotype: LLR 2.8, NLR 3.1, SIRI 2.9, PIV 2.7, PLR 2.5, MLR 2.6 for all comparisons P < 0.001) and peak CRP levels (185.4 ± 42.3 mg / L), despite relatively stable organ function as evidenced by a moderate SOFA score (4.3 ± 1.4). The early organ dysfunction phenotype (cluster 4) exhibited emerging multi-organ dysfunction, characterized by significantly elevated renal and coagulation markers (BUN 38.5 ± 5.2 mg / dL, INR 1.8 ± 0.3, PT 18.2 ± 2.1 seconds, RDW 16.8 ± 1.4%; P < 0.001) and an increased SOFA score (6.8 ± 1.6). The advanced organ failure type (cluster 3) showed severe organ dysfunction, with significant liver dysfunction (ALK 285.4±45.6U / L, ALT 325.8±58.2U / L, AST 298.4±52.1U / L, total bilirubin 4.8±0.9mg / dL; P<0.001), the highest SOFA score (8.9±1.8), and profound metabolic dysfunction reflected by the lowest HDL level (22.4±4.2mg / dL).

[0091] Analysis of clinical outcomes revealed significant differences in mortality and survival patterns among the different types ( Figure 1CD): Kaplan-Meier analysis showed different 30-day survival trajectories (log-rank test, P<0.0001). The physiologically stable type showed a higher survival rate (95% at 30 days), while the hyperinflammation, advanced organ failure and early organ dysfunction types showed a significantly reduced survival rate (70-75%). The early inflammatory response type maintained a moderate survival curve (85%), which was significantly different from the physiologically stable (P=1.44e-25) and hyperinflammation (P=6.8e-46) types. The most significant survival difference was observed between the physiologically stable type and the advanced organ failure and early organ dysfunction types (P=6.78e-110 and P=2.12e-78, respectively). Compared with the types with lower inflammatory characteristics (early inflammatory response and physiological stability), the types showing higher inflammatory markers (hyperinflammation, advanced organ failure and early organ dysfunction) always showed significantly reduced survival.

[0092] Detailed analysis of the underlying conditions revealed different comorbidity distributions among the different types ( Figure 2 ). The physiologically stable type showed the smallest chronic disease burden, with low prevalence of autoimmune diseases (1.71%), malignancies (8.32%), and chronic liver diseases (1.89%), mainly manifested by cardiovascular diseases (78.5%) and metabolic diseases (45.8%). In contrast, the excessive inflammation and early organ dysfunction type showed the highest prevalence of chronic kidney disease (32.91% and 31.33%, respectively) and diabetes (22.90% and 21.42%, respectively), while the advanced organ failure type showed the highest frequency of chronic liver disease (18.07%). Analysis of sepsis-related complications revealed different patterns among different types. Patients characterized by elevated inflammatory markers (hyperinflammation, advanced organ failure, and early organ dysfunction) showed significantly higher rates of acute kidney injury (52.64%, 44.94%, and 42.39%, respectively), respiratory failure (44.36%, 41.82%, and 39.75%, respectively), and progression to septic shock (38.52%, 35.84%, and 34.98%, respectively; P < 0.001 for all comparisons). Notably, the advanced organ failure subtype exhibited the highest frequencies of liver dysfunction (18.07%) and acute liver injury (15.31%), consistent with its unique pattern of liver dysfunction. The physiologically stable subtype maintained the lowest complication rate of all subtypes, supporting its favorable prognostic profile.

[0093] The above results indicate that the six inflammatory indicators have potential value in predicting the prognosis risk of sepsis.

[0094] Example 2

[0095] This example integrates 18 clinical variables (anion gap, potassium ion concentration, chloride ion concentration, sodium ion concentration, weight, albumin, bicarbonate, hemoglobin, alkaline phosphatase, alanine aminotransferase, aspartate aminotransferase, total bilirubin, mean corpuscular volume, age, blood urea nitrogen, international normalized ratio, prothrombin time, and red blood cell volume distribution width) and 6 inflammatory indicators to design a deep learning model that includes admission parameters.

[0096] A deep learning model was implemented using PyTorch (version 1.9.0). It consisted of a 24-node input layer, three hidden layers (128, 64, and 32 nodes, respectively) with ReLU activation and a dropout rate of 0.3, and a single-node sigmoid output layer for mortality prediction. The model was trained using the Adam optimizer (learning rate 0.001), a binary cross-entropy loss function, and mini-batches (batch size 64) for 100 epochs with early stopping (patience = 10). The dataset consisted of patient data from the MIMIC-IV cohort (n = 14,350), split into 80% and 20% training and test sets. Data preprocessing included z-score normalization to standardize the parameter distribution. Model performance was evaluated using multiple metrics (AUROC, precision-recall curves, sensitivity / specificity at optimal thresholds, calibration curves, and net reclassification improvement compared to the Somatic Pathology (SOFA) score).

[0097] The performance evaluation results are as follows Figure 3 As shown in A, it can be seen that for the 28-day mortality rate, the above model has a higher discrimination degree (AUC=0.79 vs. 0.67, P<0.001), with a sensitivity of 71.38% and a specificity of 71.40% at a threshold of 0.2339; for the in-hospital mortality rate, the model still performs well (AUC=0.77 vs. 0.70, P<0.001), with a sensitivity of 69.75% and a specificity of 69.84% at a threshold of 0.1983.

[0098] Further correlation analysis was performed on the six inflammatory indicators and all 28 clinical parameters (same as in Example 1, including the 18 clinical parameters used to construct the above model). Figure 3As shown in Figure B, red indicates positive correlation, blue indicates negative correlation, and all correlation P values ​​were < 0.001. Inflammatory markers showed moderate correlations with systemic inflammatory indicators: C-reactive protein showed moderate positive correlations with LLR (r = 0.29, P < 0.001) and NLR (r = 0.28, P < 0.001), while serum albumin showed a significant negative correlation (LLR: r = -0.27, NLR: r = -0.26; P < 0.001). Organ dysfunction parameters showed varying degrees of correlation. SOFA scores showed weak correlations (r = 0.11-0.12, P < 0.001), while specific organ function markers showed moderate correlations. Blood urea nitrogen was weakly correlated with LLR and NLR (r = 0.20 and 0.19, respectively; P < 0.001), and the anion gap showed similar correlations with inflammatory markers (r = 0.18-0.20, P < 0.001). Serum bicarbonate showed a weak inverse correlation, most notably with LLR (r = -0.18, P < 0.001). Demographic factors showed minimal correlations (age: r = 0.05-0.10; sex: negligible), suggesting that inflammatory markers primarily reflect disease severity rather than baseline characteristics.

[0099] These results demonstrate that integrating six inflammatory markers with admission parameters enables earlier and more accurate risk assessment of sepsis patients. This model, based on which a sepsis prognostic risk assessment model is constructed, yields more accurate predictions than the SOFA score.

[0100] Example 3

[0101] This example systematically analyzed the inflammatory index data of patients in the MIMIC-IV cohort (n=14350) at admission. Figure 4 A) There were significant differences in inflammatory characteristics between 28-day survivors and non-survivors. Non-survivors showed a significantly elevated inflammatory state, with all measured indices significantly upregulated: NLR showed a 3-fold increase (median 15.0 vs. 5.0, P < 2.22e-16), PLR showed a 1.33-fold increase (median 200.0 vs. 150.0, P < 2.22e-16), MLR showed a 1.6-fold increase (median 0.8 vs. 0.5, P < 2.22e-16), SIRI showed a 5-fold increase (median 10.0 vs. 2.22), and PIV showed a 4-fold increase (median 2000.0 vs. 500.0, P < 2.22e-16). External validation was performed using an independent cohort from Longhua Hospital (n = 74). The results showed that ( Figure 4B), non-survivors still showed higher levels of inflammatory indicators: NLR (P = 0.003), MLR (P = 0.035), PLR (P = 0.022), SIRI (P = 0.002), PIV (P = 0.007) and LLR (P = 0.012), indicating that these inflammatory patterns are reproducible in different clinical settings.

[0102] According to the median of each inflammatory index at admission, the patients were divided into high inflammatory risk group and low inflammatory risk group, and the Kaplan-Meier survival curve was drawn. The results showed that ( Figure 5 A), the 30-day survival trajectories of the two groups were significantly different (log-rank test, P < 0.0001). NLR, SIRI, and PIV showed clear early differences in the survival curves between the high and low groups (log-rank test, all comparisons P < 0.0001). This early prognostic signal was particularly evident in the NLR stratification, in which elevated NLR was associated with increased mortality within the first 72 hours after admission.

[0103] This example further establishes a support vector machine classifier (SVM model) based on six inflammatory indicators of patients in the MIMIC-IV cohort (n=14350), thereby accurately stratifying the risk of sepsis patients. The results show that ( Figure 5 The area under the ROC curve (AUC) of the support vector machine (SVM) model based on six inflammatory indicators in mortality risk stratification was 0.86. An independent cohort (n = 74) from Longhua Hospital was used for external validation, and the results showed that ( Figure 5 D), the high-inflammation risk group and the low-inflammation risk group showed significant differences in inflammatory indicators: NLR (median 12.5 vs. 4.2, P < 0.001), PLR (median 285 vs. 142, P < 0.001), MLR (median 0.85 vs. 0.32, P < 0.001), SIRI (median 8.2 vs. 2.8, P < 0.001), PIV (median 1850 vs. 650, P < 0.001) and LLR (median 18.5 vs. 6.8, P < 0.001). In addition, the high-risk group in the independent cohort of Longhua Hospital showed a higher mortality rate ( Figure 5 C, Fisher's exact test, P = 0.01).

[0104] Based on the above results, we further analyzed the distribution of treatment interventions in the high and low groups in the MIMIC-IV cohort and found that ( Figure 5The high-risk group had a significantly increased need for intensive treatment support, including mechanical ventilation (50.16% vs. 34.52%, P < 0.001), airway management (32.70% vs. 22.67%, P < 0.001), fluid therapy (49.96% vs. 34.52%, P < 0.001), nutritional support (24.34% vs. 16.39%, P < 0.001), and renal replacement therapy (8.32% vs. 2.31%, P < 0.001). This suggests that the SVM model based on six inflammatory indicators can not only accurately stratify the risk of sepsis patients but also predict the treatment needs of sepsis patients, thereby establishing a stratified management mechanism for sepsis patients: data are repeatedly measured and assessed every 4 hours for the high-inflammatory risk group, and every 12 hours for the low-inflammatory risk group.

[0105] Example 4

[0106] This example validates the biological basis of the SVM model based on six inflammatory indices. A comprehensive longitudinal analysis of inflammatory indices (LLR, MLR, NLR, PIV, PLR, and SIRI) was performed within 14 days of admission to the MIMIC-IV cohort to describe the stability and changing trends of inflammatory patterns in patients with sepsis in different risk groups.

[0107] Systematic trajectory analysis revealed distinct and persistent inflammatory signatures between 28-day survivors and non-survivors ( Figure 6 ). 28-day survivors and non-survivors (n=14,350) maintained significant differences within 14 days of admission. LLR remained higher in non-survivors (20-22) than in survivors (12-14) throughout the first week (P<0.001, days 0-6). MLR maintained a relatively mild but statistically significant separation (non-survivors 0.9-1.1 vs. survivors 0.6-0.7; P<0.01, days 0-4). NLR showed a significant and stable difference between non-survivors (15-20) and survivors (10-12) (P<0.001, days 0-6). PIV showed the greatest difference in values ​​(non-survivors 3000-4000 vs. survivors 1500-2000; P<0.001, days 0-5). PLR showed significant differences early (days 0-3) (P<0.001), followed by some temporal fluctuations. SIRI showed greater amplitude differences between non-survivors (12-18) and survivors (8-10), but remained elevated throughout (P<0.001, days 0-6).

[0108] These results demonstrate that inflammatory indicators, as markers for distinguishing different sepsis patients, are stable, rather than transiently fluctuating. Regularly measuring clinical parameters, including six inflammatory indicators, in sepsis patients allows for dynamic monitoring of their physiological and pathological states. By connecting the model constructed in Example 2 or Example 3 to a nursing station or laboratory, and establishing early warning thresholds for inflammatory indicators and a warning device, personalized treatments and real-time monitoring can be implemented for sepsis patients, enabling timely medical intervention.

[0109] Example 5

[0110] This example uses single-cell and bulk RNA sequencing methods to perform comprehensive transcriptome analysis and establish the cellular basis of inflammatory markers.

[0111] Single-cell RNA sequencing data were obtained from Gene Expression Omnibus (GEO) with accession number GSE167363. The initial single-cell RNA sequencing analysis results of the GSE167363 dataset showed that ( Figure 7 A) , based on peripheral blood single-cell RNA sequencing, nine major cell populations were identified; cell type annotation using canonical markers clearly delineated key immune populations ( Figure 7 B): CD14 / LYZ of monocytes, CSF3R / MMP9 of neutrophils, CD3D / CD3E of T cells, CD79A / MS4A1 of B cells, GNLY / NKG7 of NK cells and MZB1 / JCHAIN ​​of plasma cells. The comparative analysis results of cell composition showed that ( Figure 7 C), nonsurvivors of sepsis exhibited significantly elevated neutrophil and platelet ratios compared with survivors and healthy controls (P < 0.001). This observation provides the first direct cellular evidence supporting the prognostic value of neutrophil- and platelet-based inflammatory indices in sepsis outcome.

[0112] The deconvolution algorithm was further used to analyze the large amount of RNA sequencing data from the GSE154918 dataset, including healthy controls (Hlty, n = 40), infection without organ dysfunction (Inf1_P, n = 12, SOFA < 2), sepsis with organ dysfunction (Seps_P, n = 20, SOFA ≥ 2), and septic shock requiring vasopressors (Shock_P, n = 19). This new analysis method can accurately quantify the proportion of immune cells and calculate the inflammatory index based on the existing transcriptome data. The results of cellular component analysis showed that ( Figure 8), the adaptive immune compartment showed significant depletion from healthy controls to the disease state, with a significant decrease in B cells (P = 7e-05) and T cells (P = 1.4e-05), indicating impaired adaptive immunity; NK cells gradually decreased (P = 9.5e-07), while the plasma cell population remained relatively stable; consistent with the results of single-cell studies, the bone marrow compartment showed progressive expansion, with neutrophils and platelets showing a step-by-step increase associated with disease severity.

[0113] In addition, by calculating the inflammatory indicators in the development stage of sepsis, the results showed that ( Figure 9 As the severity of the disease increases, all indices increase progressively, transitioning from the healthy control group to infection, sepsis, and septic shock. Neutrophil and platelet indices (NLR, LLR, and PLR) increase most significantly in patients with septic shock. Using ROC curve analysis, the results showed that ( Figure 9 B), PLR and NLR showed superior predictive ability (AUC > 0.85) compared with other inflammatory markers, establishing their specific use in identifying patients at risk for developing shock.

[0114] These results establish a direct mechanistic link between single-cell observations of elevated neutrophil-to-platelet ratios in nonsurvivors and the prognostic value of corresponding inflammatory indices in septic shock. Neutrophil and platelet indices exhibit particularly striking patterns of elevation in septic shock, establishing a direct mechanistic link between changes in cellular composition and the dynamics of inflammatory indices. This progressive elevation likely reflects the cumulative effects of acute myelopoiesis, altered cell trafficking, and dysregulated immune homeostasis in severe sepsis.

[0115] Example 6

[0116] Based on the results observed in Example 5, this example conducts in-depth single-cell transcriptional characterization of neutrophil heterogeneity to reveal potential pathogenic mechanisms. The results of unbiased cluster analysis of neutrophil populations show that ( Figure 10 AB), the neutrophil population was divided into three distinct subpopulations characterized by differential expression of established maturation markers: pre-neutrophils (PreNeu) expressing LYZ and PTMA, immature neutrophils (ImmNeu) marked by expression of MMP9 and LCN2, and mature neutrophils (MatNeu) marked by expression of PROK2 and ADM. Comparative analysis of cell composition showed that ( Figure 10 C), the proportion of immature neutrophils (ImmNeu) in sepsis non-survivors was significantly higher than that in survivors and healthy controls (P < 0.001, one-way ANOVA), suggesting their potential role in adverse outcomes. Pseudo-time trajectory analysis results ( Figure 10D) further reveals a developmental continuum from an immature to a mature state, providing insights into the dynamics of neutrophil maturation during sepsis. This observation is consistent with previous studies that have documented alterations in neutrophil maturation and function during sepsis, with accumulation of immature neutrophils associated with poor prognosis.

[0117] Single-sample gene set enrichment analysis (ssGSEA) was further used to quantify the characteristics of neutrophil subpopulations in large amounts of RNA sequencing data. The results showed that ( Figure 10 E) The progressive increase in immature neutrophil scores at each stage of the disease demonstrates that immature neutrophils increase with disease severity (all pairwise comparisons P < 0.001). This pattern suggests a mechanistic link between neutrophil immaturity and disease severity. To decipher the molecular program underlying the pathogenic potential of immature neutrophils, their transcriptional signatures were analyzed. Gene set enrichment analysis revealed ( Figure 10 F), pro-inflammatory and hypoxic adaptation pathways are activated simultaneously in immature neutrophils, and the enrichment curve shows that ( Figure 11 A), key pathogenic pathways in patients with septic shock were significantly upregulated, including hypoxia-inducible factor, neutrophil degranulation, and leukocyte transendothelial migration. Analysis of effector molecule expression in different neutrophil subsets ( Figure 11 B) shows that the immature subpopulation synergistically upregulated degranulation mediators (CD177, CD63, LCN2, MMP9), hypoxia response factors (LDHA, ASPH, HIF1A, PKM) and inflammatory signaling components (S100A8 / A9, HMGB1, TLR2). Pathway-specific gene expression heat map ( Figure 11 C) Demonstrates selective enrichment of pathogenic features between patients with septic shock and sepsis.

[0118] These results demonstrate that molecular markers of neutrophil dysfunction are significantly elevated in the transcriptome of patients with septic shock compared to patients with sepsis. These findings provide new mechanistic insights into neutrophil-based inflammatory indices in sepsis by demonstrating that disease progression is specifically associated with the accumulation of immature neutrophils that exhibit both proinflammatory and hypoxic phenotypes. This finding not only elucidates the underlying mechanisms of neutrophil elevation during the development of septic shock but also identifies immature neutrophils and their associated molecular pathways as potential therapeutic targets for septic shock intervention.

[0119] Example 7

[0120] Following the identification in Example 6 that elevated PLR is a predictive marker for mortality in septic shock, this example performed comprehensive single-cell transcriptome profiling to describe platelet heterogeneity during sepsis progression.

[0121] Based on single-cell transcriptomics and the UMAP dimensionality reduction method, platelets in peripheral blood (n=14350) were analyzed ( Figure 12 A), two subgroups P1 and P2 with different metabolic and immune regulatory characteristics were found. Violin plot ( Figure 12 B) shows the differential expression of specific marker genes in each subgroup: PFN1, CYBA, IFITM2, and IFITM3 were expressed more highly in the P1 subgroup, while PTCRA, ACRBP, RIPOR2, and RAB32 were expressed more highly in the P2 subgroup. The results of cell composition analysis showed that ( Figure 12 C), in sepsis non-survivors, the proportion of P1 platelets enriched with metabolic activation phenotype was significantly increased compared with survivors and healthy controls (P<0.001, one-way ANOVA), suggesting that it may have pathogenic significance. Gene set enrichment analysis ( Figure 12 D) confirmed the expression of P1 subgroup-specific transcriptional signatures in patients with septic shock: in the comparison of transcriptomes between shock group and sepsis group, P1 pattern genes were significantly enriched (corrected P = 0.0066, FDR < 0.05), confirming the correlation between P1 platelet activation and disease severity at the population level. Pathway enrichment analysis ( Figure 12 E) revealed that the P1 platelet phenotype simultaneously activates four functional modules: (i) energy adaptation (oxidative phosphorylation, respiratory chain electron transport); (ii) metabolic remodeling (glycolysis / gluconeogenesis); (iii) stress response (ROS detoxification); (iv) platelet activation (calcium-dependent signaling). Differential expression analysis of key functional molecules ( Figure 12 F) further demonstrated the upregulation of P1 subpopulation-specific proteins, including electron transport chain components (NDUFA4, COX7A2, COX6A1, NDUFB1), glycolytic enzymes (LDHA, TPI1, ENO1, PGK1), and platelet activation markers (PFN1, CD63, CD9, PROS1). Hierarchical clustering analysis of pathway-specific transcripts ( Figure 13 A) showed that patients with septic shock had synergistic differences in metabolic and activation characteristics compared with patients with sepsis. Spearman rank correlation analysis was used to explore the association between NLR and neutrophil-specific molecular determinants, and the results showed that ( Figure 13 B) It was significantly correlated with damage-associated molecular patterns (S100A9: r = 0.895, S100A8: r = 0.881; P < 0.001) and extracellular matrix remodeling factors (MMP9: r = 0.881, CD177: r = 0.837; P < 0.001), confirming the mechanistic connection between inflammatory indicators and neutrophil pathogenic pathways. Similar Spearman rank correlation analysis ( Figure 13C) found that PLR was closely related to key regulatory factors of platelet activation and energy adaptation, including granule secretion (CD63: r = 0.836, P < 0.001), oxidative phosphorylation mechanism (COX6A1: r = 0.815, P < 0.001), and glycolysis remodeling (LDHA: r = 0.792, ENO1: r = 0.500; P < 0.001).

[0122] These results indicate that P1 platelets are a metabolically overactive subpopulation associated with disease severity, and PLR can serve as a surrogate marker of platelet metabolic activation during sepsis progression.

[0123] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A sepsis prognostic risk stratification system, characterized in that: The sepsis prognostic risk stratification system includes a risk stratification module; The risk stratification module is used to classify patients into high-inflammation risk groups and low-inflammation risk groups based on patient data of biomarkers predicting sepsis prognosis using a prediction model, and output risk stratification results; The prediction model is constructed based on a machine learning algorithm using patient data of the biomarkers in the population; The biomarkers include neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, monocyte-to-lymphocyte ratio, leukocyte-to-lymphocyte ratio, systemic inflammatory response index, and peripheral inflammation value; The calculation formula of the systemic inflammatory response index is: systemic inflammatory response index = neutrophil count × monocyte count / lymphocyte count; The calculation formula of the peripheral inflammation value is: peripheral inflammation value = platelet count × neutrophil count / lymphocyte count.

2. The sepsis prognostic risk stratification system according to claim 1, characterized in that: The classification method is to set a threshold value according to the median value of the group patient data to classify the patients.

3. The sepsis prognostic risk stratification system according to claim 1, wherein: The sepsis prognosis risk stratification system further includes a data collection module; the data collection module is used to collect the patient's hematological examination data; It also includes a data calculation module; the data calculation module is used to calculate the patient data of the biomarker based on the hematological examination data.

4. The sepsis prognostic risk stratification system according to claim 1, wherein: The sepsis prognosis risk stratification system further includes a dynamic monitoring module; the dynamic monitoring module is used to periodically and repeatedly measure the patient data of the biomarkers; record the changing trend of the patient data; and trigger an early warning signal when the change in the patient data exceeds a preset threshold; The frequency of dynamic monitoring is determined according to the patient's risk level: the measurement data and patient prognostic risk stratification results are updated every 4 hours for the high inflammation risk group; and the measurement data and patient prognostic risk stratification results are updated every 12 hours for the low inflammation risk group; It also includes a treatment demand prediction module; the treatment demand prediction module is used to predict the probability of treatment demand based on the patient's risk stratification results; the treatment demand includes mechanical ventilation demand, airway management demand, fluid therapy demand, nutritional support demand and renal replacement therapy demand.

5. A sepsis prognosis survival risk prediction system, characterized in that: The sepsis prognosis survival risk prediction system includes a survival risk prediction module; The survival risk prediction module is used to predict the in-hospital mortality rate and 28-day mortality rate of patients using a prediction model based on the patient data of biomarkers for predicting sepsis prognosis and the patient's basic clinical data, and output the prediction results; The basic clinical data include anion gap, potassium ion concentration, chloride ion concentration, sodium ion concentration, body weight, albumin, bicarbonate, hemoglobin, alkaline phosphatase, alanine aminotransferase, aspartate aminotransferase, total bilirubin, mean corpuscular volume, age, blood urea nitrogen, international normalized ratio, prothrombin time and red blood cell volume distribution width; The prediction model is constructed using a deep learning framework using patient data such as the biomarkers and basic clinical data of the patients in the population; The biomarkers include neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, monocyte-to-lymphocyte ratio, leukocyte-to-lymphocyte ratio, systemic inflammatory response index, and peripheral inflammation value; The calculation formula of the systemic inflammatory response index is: systemic inflammatory response index = neutrophil count × monocyte count / lymphocyte count; The calculation formula of the peripheral inflammation value is: peripheral inflammation value = platelet count × neutrophil count / lymphocyte count.

6. The sepsis prognosis survival risk prediction system according to claim 5, characterized in that: The sepsis prognosis survival risk prediction system further includes a data collection module; the data collection module is used to collect the patient's basic clinical data and hematological examination data; It also includes a data calculation module; the data calculation module is used to calculate the patient data of the biomarker based on the hematological examination data.

Images