Multi-parameter fusion-based septicopyemia early warning system

Abstract

The invention discloses a septicopyemia early warning system based on multi-parameter fusion. Comprising the following steps: S1, data preparation and preprocessing; extracting clinical data from the intensive care medicine database; according to the method, two web version operation interfaces are respectively developed through R language core algorithm codes optimized by a shock risk model and a death risk model, and when the method is used by a clinician, 21 index data such as SOFA and systolic pressure are input aiming at shock risk prediction; for death risk prediction, 15 index data such as SOFA and GCS are input, the model automatically calculates the shock or death probability of a patient and an operation interface which is convenient for clinicians to operate and can predict the sepsis shock and death occurrence rate, and the risk prediction model can be applied to clinical treatment practice and has a wide application prospect. The risk prediction of shock and death of the sepsis patient is realized, clinical doctors are assisted in evaluating the condition of the patient, early intervention and treatment are carried out, and the overall control of the clinical doctors on the condition progress of the sepsis patient is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine learning algorithm technology, and particularly relates to a sepsis early warning system based on multi-parameter fusion. BACKGROUND

[0002] Sepsis is a life-threatening organ dysfunction caused by a dysregulated host response to infection. Common signs of sepsis include fever, rapid heart rate, rapid breathing, confusion, and pain in the body. It can lead to septic shock, multiple organ failure, and death. Sepsis can affect anyone, but older adults, young children, pregnant women, or people with other health problems are at higher risk. Sepsis has become one of the most common causes of death worldwide and one of the main causes of death in intensive care unit patients.

[0003] Early treatment of sepsis is the most effective, but its pathogenesis is complex, early clinical symptoms are not clear, and blood culture time is too long. Currently, some biomarkers of sepsis have been studied, and some studies have found that IL-18, CD64 and lactic acid are biomarkers of sepsis shock. With the widespread application of machine learning in early management of diseases, it is easier for clinical application to build prediction models based on risk factors. There is still a lack of a system for early warning of sepsis based on machine learning. SUMMARY

[0004] The purpose of the present application is to timely remind clinicians of early intervention for sepsis patients to shock and death, and to cause doctors to pay attention to patients in different degrees by predicting the quantitative value of the risk of shock and death, thereby greatly reducing the probability of sepsis patients to shock and death. At the same time, by comparing various machine learning algorithms, the optimal algorithm model is found to improve the specificity and sensitivity of the disease prediction model, and the sepsis early warning system based on multi-parameter fusion is proposed.

[0005] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0006] The sepsis early warning system based on multi-parameter fusion comprises the following steps:

[0007] S1: data preparation and preprocessing; extracting clinical data from a critical medical database, including sepsis patients after screening by inclusion and exclusion criteria, and dividing into a training set and a test set in a ratio of 7:3;

[0008] S2: feature selection and model construction preparation; setting shock and death as outcome indicators, including 30 demographic and examination and test indicators; using statistical software to perform single factor analysis (χ2 test for categorical variables and t test for continuous variables) and correlation analysis (Spearman correlation analysis for categorical variables and Pearson correlation analysis for continuous variables) to select features; using the selected features to build a machine learning model, and using the test set to evaluate the model. 2Test, continuous variables were selected by T-test or Mann-Whitney U test according to normal distribution and multivariate (LASSO regression) analysis; use random forest (RF), XGBoost, GBDT, CatBoost and Logistic model to evaluate the model performance by AUC, ACC, SEN, SPE;

[0009] S3: Determine the shock influencing factors; through single factor analysis of the training set, 26 indicators with significant differences between the shock group and the non-shock group were screened out (P<0.05);

[0010] S4: Construct shock risk model; 21 key indicators were screened out from 26 indicators by LASSO regression, which were included in multiple machine learning model training, and RF was determined as the optimal model (combined with ACC, SEN, SPE) after test set verification, and the importance of indicators was sorted by GINI coefficient;

[0011] S5: Determine the death influencing factors; through single factor analysis of septic death group, 28 indicators with significant differences between the death group and the non-death group were screened out (P<0.05);

[0012] S6: Construct death risk model; 15 key indicators were screened out from 28 indicators by LASSO regression, which were included in multiple machine learning model training, and RF was determined as the optimal model (combined with ACC, SEN, SPE) after test set verification, and the importance of indicators was sorted by GINI coefficient;

[0013] S7: Generate web operation interface; based on the core algorithm of the optimized shock and death risk model, develop web version interface to support input corresponding risk factor indicators to automatically calculate the shock or death probability of patients.

[0014] Preferably: the inclusion and exclusion criteria in step S1 include: the inclusion criteria are: (1) age > 18 years old; (2) first entering intensive care unit; (3) no shock and survival within 24 hours after entering ICU;

[0015] The exclusion criteria are: (1) missing clinical information; (2) missing shock / death outcome indicators; (3) ICU stay time < 24 hours.

[0016] Further: the training set in step S1 contains 9149 patients (15.2% died, 37.9% had shock), and the test set contains 3922 patients (15.8% died, 38.4% had shock), and there is no statistical difference in demographic and clinical indicators between the two groups (P>0.05).

[0017] Further, the statistical analysis software in step S2 includes SPSS 25.0 and R 4.3.3, and the evaluation indexes include area under curve (AUC), accuracy (ACC), sensitivity (SEN) and specificity (SPE) (test level α=0.05).

[0018] As a preferred scheme of the present application, the 26 shock-related indexes screened in step S3 include SOFA, GCS, age, heart rate, systolic pressure, diastolic pressure, mean blood pressure, respiratory rate, blood oxygen saturation, base excess, carbon dioxide partial pressure, oxygen partial pressure, hematocrit, hemoglobin, platelet count, white blood cell count, urea nitrogen, blood sodium, blood calcium, blood chloride, blood potassium, blood creatinine, prothrombin time, normalized ratio, lactic acid and fluid infusion volume.

[0019] As a further scheme of the present application, the 21 shock key indexes screened in step S4 include SOFA, GCS, age, heart rate, systolic pressure, diastolic pressure, respiratory rate, base excess, carbon dioxide partial pressure, oxygen partial pressure, hematocrit, hemoglobin, platelet count, white blood cell count, blood sodium, blood calcium, blood chloride, blood potassium, blood creatinine, lactic acid and fluid infusion volume; the AUC of the RF model in the test set is 0.876 (95% CI: 0.865-0.888), the ACC is 0.816, the SEN is 0.743, and the SPE is 0.862.

[0020] As a still further scheme of the present application, the 28 death-related indexes screened in step S5 include SOFA, GCS, gender, age, body weight, height, heart rate, systolic pressure, diastolic pressure, mean blood pressure, respiratory rate, body temperature, blood oxygen saturation, base excess, carbon dioxide partial pressure, oxygen partial pressure, hematocrit, hemoglobin, platelet count, white blood cell count, urea nitrogen, blood calcium, blood chloride, blood potassium, blood creatinine, prothrombin time, normalized ratio and lactic acid.

[0021] On the basis of the foregoing scheme, the 15 death key indexes screened in step S6 include SOFA, GCS, age, body weight, heart rate, systolic pressure, respiratory rate, body temperature, blood oxygen saturation, oxygen partial pressure, hemoglobin, urea nitrogen, blood chloride, prothrombin time and lactic acid; the AUC of the RF model in the test set is 0.853 (95% CI: 0.839-0.868), the ACC is 0.716, the SEN is 0.871, and the SPE is 0.687.

[0022] On the basis of the foregoing scheme, the webpage interface in step S7 includes a shock risk prediction module (inputting 21 indexes to calculate shock probability) and a death risk prediction module (inputting 15 indexes to calculate death probability), which are used for the clinician to quantitatively evaluate the patient's disease progression and intervention opportunity.

[0023] On the basis of the foregoing scheme: in steps S4 and S6, RF is taken as the optimal model (comprehensive evaluation of ACC, SEN and SPE).

[0024] The beneficial effects of the present application are:

[0025] 1. The sepsis early warning system based on multi-parameter fusion, through the R language core algorithm code optimized by the shock risk model and the death risk model, two web page operation interfaces are developed, when used by clinical doctors, 21 index data such as SOFA and systolic pressure are input for shock risk prediction, 15 index data such as SOFA and GCS are input for death risk prediction, the model automatically calculates the probability of shock or death of the patient, the operation interface which can predict the incidence of septic shock and death and is convenient for clinical doctors to operate, the risk prediction model can be applied to clinical treatment practice, the risk prediction of septic shock and death of the patient is realized, the clinical doctors are assisted to evaluate the patient's condition, early intervention and treatment are carried out, and the overall control of the clinical doctors on the progress of the patient's condition is improved.

[0026] 2. The sepsis early warning system based on multi-parameter fusion, through the incorporation of demographic indicators and examination indicators, the key risk factors are screened out through single factor analysis and LASSO regression, then five kinds of machine learning algorithms such as random forest are used to construct models, the optimal model is selected after comparison of AUC, ACC and other indicators, the shock risk model (RF) AUC reaches 0.876, the death risk model (RF) AUC reaches 0.853, both are more than 0.8, which indicates that the model can effectively quantify the risk of the patient and provide reliable prediction basis for the clinic.

[0027] 3. The sepsis early warning system based on multi-parameter fusion, through data analysis, it is determined that 10 indicators such as SOFA, GCS and age are common risk factors of sepsis shock and death, the clinical doctors focus on these factors and manage them in time (such as controlling SOFA score and maintaining GCS level) during treatment, which can reduce the risk of shock and death, realize the intervention effect of 'half the effort and twice the result', help to improve the survival rate of the patient and reduce long-term complications such as post-sepsis syndrome. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is the construction process schematic diagram of the sepsis early warning system based on multi-parameter fusion proposed in the present application;

[0029] Figure 2 is the variable importance ranking schematic diagram in the sepsis shock RF model of the sepsis early warning system based on multi-parameter fusion proposed in the present application;

[0030] Figure 3is a variable importance ranking diagram in a sepsis death RF model of a sepsis early warning system based on multi-parameter fusion provided by the application;

[0031] Figure 4 is a sepsis shock risk prediction web page operation interface of a sepsis early warning system based on multi-parameter fusion provided by the application.

[0032] Figure 5 is a sepsis death risk prediction web page operation interface of a sepsis early warning system based on multi-parameter fusion provided by the application. DETAILED DESCRIPTION

[0033] The technical solutions of the patent will be further described in detail below in combination with specific embodiments.

[0034] The embodiments of the patent are described in detail below, and examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the patent, and cannot be understood as a limitation on the patent.

[0035] Embodiment 1:

[0036] The sepsis early warning system based on multi-parameter fusion includes the following steps, as shown in the figure: Figure 1

[0037] S1: data preparation and data preprocessing; the data is selected from identifiable clinical data extracted from a severe medical database (MIMIC-IV); after strict inclusion and exclusion criteria, a total of 13071 sepsis patients were included, among which 4972 patients developed septic shock, and 2010 patients developed sepsis death; subsequently, the above patients were divided into a training set and a test set according to a 70% and 30% ratio according to a random number table; wherein the training set contains 9149 patients, and the test set contains 3922 patients;

[0038] The inclusion criteria in the inclusion and exclusion criteria are as follows: (1) the patient is older than 18 years old;

[0039] (2) the patient first enters the intensive care unit;

[0040] (3) the patient does not develop shock and survives within 24 hours after entering the ICU;

[0041] The exclusion criteria are as follows: (1) the patient's clinical information is missing;

[0042] (2) the patient's shock and death outcome indicators are missing;

[0043] ​(3) The patient's time in the ICU is less than 24 hours;

[0044] S2: Feature selection on patient data; To build the risk prediction model of septic shock and septic death, the outcome indicators are set as shock and death, and 6 demographic indicators including sequential organ failure score (SOFA), Glasgow scale score (GCS), gender, age, weight, height, and 24 examination and test indicators including heart rate, systolic pressure, diastolic pressure, mean blood pressure, respiratory rate, body temperature, blood oxygen saturation, base excess, carbon dioxide partial pressure, oxygen partial pressure, hematocrit, hemoglobin, platelet count, white blood cell count, urea nitrogen, blood sodium, blood calcium, blood chloride, blood potassium, blood creatinine, prothrombin time, standard ratio, lactic acid, fluid infusion volume are also included;

[0045] SPSS25.0 and R 4.3.3 software are used for statistical analysis of patient data; the number of cases and percentage n(%) are used to show the classification variables, and the median and interquartile range M(P25, P75) are used to show the continuous variables; for continuous variables that do not conform to normal distribution, Mann-Whitney U test is used, and for continuous variables that conform to normal distribution, T test is used; χ2test is used for single factor analysis of classification variables; LASSO regression is used for multiple factor analysis;

[0046] And four machine learning algorithms, Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Gradient Boosting Decision Tree (GBDT), and Categorical Boosting (CatBoost), are used to establish risk prediction models, with Logistic model as the control; area under the curve (AUC), accuracy (ACC), sensitivity (SEN), and specificity (SPE) are used to evaluate the predictive ability of the model; test level a = 0.05;

[0047] The results of comparative analysis of baseline data of the research sample showed that among the 9149 sepsis patients included in the training set, 1389 patients (15.2%) died, and 3466 patients (37.9%) developed shock; among the 3922 sepsis patients included in the test set, 621 patients (15.8%) died, and 1506 patients (38.4%) developed shock; we compared the demographic information and clinical indicators of the training set and the test set and found that there was no statistically significant difference (P>0.05), indicating that the training set and the test set were comparable;

[0048] S3: Determine the factors affecting septic shock; At the same time, single factor analysis was carried out on the training set, and the results showed that SOFA, GCS, age, heart rate, systolic pressure, diastolic pressure, mean blood pressure, respiratory rate, oxygen saturation, base excess, carbon dioxide partial pressure, oxygen partial pressure, hematocrit, hemoglobin, platelet count, white blood cell count, urea nitrogen, blood sodium, blood calcium, blood chloride, blood potassium, blood creatinine, prothrombin time, standard ratio, lactic acid, fluid volume These 26 indicators have significant statistical differences (P<0.05) between the shock group and the non-shock group, and the factors affecting septic shock are determined;

[0049] S4: Construct the risk model of septic shock; First, the above 26 indicators are further reduced using LASSO regression to further reduce the number of risk factor indicators, and the results show that there are 21 indicators that have good correlation for predicting septic shock; They are SOFA, GCS, age, heart rate, systolic pressure, diastolic pressure, respiratory rate, base excess, carbon dioxide partial pressure, oxygen partial pressure, hematocrit, hemoglobin, platelet count, white blood cell count, blood sodium, blood calcium, blood chloride, blood potassium, blood creatinine, lactic acid, and fluid volume;

[0050] Subsequently, these 21 indicators were included in RF, XGBoost, GBDT, CatBoost, and Logistic machine learning models, and the model was trained and parameterized through the training set data to construct the risk model of septic shock in patients with sepsis. In the test set, it was found that the areas under the curves of RF, XGBoost, GBDT, CatBoost, and Logistic were 0.876, 0.854, 0.874, 0.867, and 0.875, respectively;

[0051] Model index construction table of septic shock

[0052]

[0053] After combining ACC, SEN, SPE and other evaluation indicators, it was found that RF was the most optimal machine learning model for diagnosis, and according to the GINI coefficient in RF, the importance was sorted as SOFA, systolic pressure, fluid volume, base excess, oxygen partial pressure, lactic acid, blood creatinine, respiratory rate, white blood cell count, blood chloride, GCS, blood calcium, platelet count, diastolic pressure, blood sodium, hemoglobin, carbon dioxide partial pressure, hematocrit, potassium, heart rate, and age;

[0054] S5: Determine the factors affecting septic death, and perform single factor analysis on the indicators of septic death to determine that SOFA, GCS, gender, age, weight, height, heart rate, systolic pressure, diastolic pressure, mean blood pressure, respiratory rate, body temperature, blood oxygen saturation, base excess, carbon dioxide partial pressure, oxygen partial pressure, hematocrit, hemoglobin, platelet count, white blood cell count, urea nitrogen, blood calcium, blood chloride, blood potassium, blood creatinine, prothrombin time, standard ratio, lactic acid, and 28 other indicators have significant statistical differences (P<0.05) between the death group and the non-death group;

[0055] S6: Construct a risk model for septic death; further use LASSO regression on the 28 indicators in step S5 to determine 15 indicators that have good correlation for predicting septic death, which are SOFA, GCS, age, weight, heart rate, systolic pressure, respiratory rate, body temperature, blood oxygen saturation, oxygen partial pressure, hemoglobin, urea nitrogen, blood chloride, prothrombin time, and lactic acid; then based on these 15 indicators, RF, XGBoost, GBDT, CatBoost, and Logistic machine learning models are included, and through training set data, the model is trained and the parameters are optimized to construct a risk model for septic death. In the test set, it is found that the areas under the curves of RF, XGBoost, GBDT, CatBoost, and Logistic are 0.853, 0.819, 0.845, 0.842, and 0.833, respectively; combined with ACC, SEN, SPE, and other evaluation indicators, it is determined that the diagnostic performance of RF is the best, and then according to the GINI coefficient in RF, the importance is sorted, which are SOFA, GCS, urea nitrogen, oxygen partial pressure, respiratory rate, prothrombin time, lactic acid, systolic pressure, blood oxygen saturation, age, hemoglobin, blood chloride, weight, heart rate, and body temperature;

[0056] Table of model indicators for septic death

[0057]

[0058] S7: Generate a web-based operation interface according to the program code of the prediction model;

[0059] The core algorithm after parameter optimization in the S4 step is the prediction model of R language program code, and the program code of the prediction model is the algorithm core. A web-based operation interface based on the shock risk factors of sepsis patients is designed and developed. Clinicians input the above 21 risk factor indicators of sepsis patients into the corresponding risk factor blanks, and the prediction model automatically calculates the probability of shock in the patient, which can be more intuitive and convenient for clinicians to use, and can help clinicians have a more quantitative and intuitive understanding of the progress of sepsis patients, and timely and reasonably call medical resources for early intervention and treatment of patients;

[0060] At the same time, the program code of the prediction model after parameter optimization in the S6 step is the algorithm core, and a web-based operation interface based on the death risk factors of sepsis patients is designed and developed. Clinicians input the above 15 risk factor indicators of sepsis patients into the corresponding risk factor blanks, and the prediction model automatically calculates the probability of death in the patient, which can provide auxiliary support for clinicians to assess the condition of sepsis patients and can well remind clinicians to pay attention to clinical indicators and control the risk of death to a low level.

[0061] In the process of building the system, it is found that SOFA, GCS and other 21 risk factors are risk factors for septic shock, and SOFA, GCS and other 15 risk factors are risk factors for septic death. Among them, SOFA, GCS, age, heart rate, systolic blood pressure, respiratory rate, oxygen partial pressure, hemoglobin, blood chloride, and lactic acid are common risk factors for septic shock and septic death. This is conducive to clinicians paying particular attention to the above 10 risk factors during the treatment of sepsis patients, and timely managing and controlling them, while reducing the incidence of shock and death, achieving a multiplier effect.

[0062] In the process of building the system, it is found that the shock rate and mortality rate of sepsis patients in ICU based on MIMIC-IV database are as high as 38.04% (4972 / 13071) and 15.38% (2010 / 13071), respectively, which indicates that the treatment of sepsis patients needs to be continuously monitored by clinicians. The clinical risk prediction model of septic shock and septic death based on machine learning algorithm can provide real-time quantitative risk assessment for clinicians, assist doctors in timely intervention, and provide etiological treatment and supportive treatment for patients according to the "Sepsis Diagnosis and Treatment Guidelines (2023)" released by the Chinese Research Hospital Association, reduce the risk factor level of patients, and improve the survival rate of patients;

[0063] Among them, the management of septic shock is more important, septic shock is the result of imbalance between pathogens and hosts; timely and appropriate management is a key factor to improve the prognosis of this life-threatening syndrome; early identification and assessment of septic shock is crucial to maximize the chances of survival, clinicians collect clinical and laboratory indicators from septic patients, and through risk prediction models, the probability of patients developing shock is obtained, and patients are managed and intervened in a timely manner; the triad of septic shock management is resuscitation / antimicrobial therapy / infection source control, to reduce functional impairment in patients and improve the survival ability of patients after recovery, to avoid many patients suffering from long-term neurocognitive impairment, new functional disability and mental illness related to post-sepsis syndrome or post-intensive care syndrome after shock;

[0064] In the system, based on the identified risk factors, five machine learning algorithms are used to successfully establish risk prediction models for septic shock and septic death, and the AUC of all models exceeds 0.80, indicating that the model performance is good, among them, the prediction ability of the random forest model is the best, and the area under the curve reaches 0.876 and 0.853 respectively, and based on the prediction model program algorithm, an operation interface is designed for clinicians to predict the incidence of septic shock and death, and the risk prediction model can be applied to clinical treatment practice to realize the risk prediction of septic shock and death in septic patients, assist clinicians in assessing the patient's condition, and perform early intervention and treatment to improve the overall control of the patient's condition by clinicians.

[0065] The above describes the preferred embodiment of the present application, and the protection scope of the present application is not limited thereto, any modification, equivalent replacement and improvement made by those skilled in the art within the technical scope disclosed by the present application or public common knowledge, within the spirit and principles of the present application, should be covered within the protection scope of the present application.

Claims

1. A sepsis early warning system based on multi-parameter fusion, characterized in that, Comprising the following steps: S1: data preparation and preprocessing; clinical data was extracted from the critical care database, and patients with sepsis were included after screening by inclusion and exclusion criteria, and were divided into training set and test set in the ratio of 7:3; S2: Feature selection and model construction preparation; shock and death were set as outcome indicators, and 30 indicators were included, including demographic and examination results; univariate statistical software was used for categorical variables (χ² was used for categorical variables). 2 For continuous variables, the T-test or Mann-Whitney U test was used according to the normal distribution, and multivariate (LASSO regression) analysis was performed. Random forest (RF), XGBoost, GBDT, CatBoost and Logistic models were used, and the model performance was evaluated by AUC, ACC, SEN and SPE. S3: determine the shock influencing factors; 26 indicators significantly different between the shock group and the non-shock group were screened out by single factor analysis of the training set (P<0.05); S4: build a shock risk model; 21 key indicators were screened out from the 26 indicators by LASSO regression, and were included in multiple machine learning model training, and RF was determined as the optimal model after test set verification (combined with ACC, SEN, and SPE), and the importance of the indicators was sorted by GINI coefficient; S5: determine the death influencing factors; 28 indicators significantly different between the death group and the non-death group were screened out by single factor analysis of the sepsis death group (P<0.05); S6: build a death risk model; 15 key indicators were screened out from the 28 indicators by LASSO regression, and were included in multiple machine learning model training, and RF was determined as the optimal model after test set verification (combined with ACC, SEN, and SPE), and the importance of the indicators was sorted by GINI coefficient; S7: generate a web operation interface; Based on the optimized shock and death risk model core algorithm, a web version interface was developed to support automatic calculation of patient shock or death probability by inputting corresponding risk factor indicators.

2. The multi-parameter fusion based sepsis early warning system according to claim 1, wherein, The inclusion and exclusion criteria in step S1 include: inclusion criteria are: (1) age > 18 years old; (2) first entering the intensive care unit; (3) no shock and survival within 24 hours after entering the ICU; Exclusion criteria are: (1) missing clinical information; (2) missing shock / death outcome indicators; (3) ICU stay time < 24 hours.

3. The multi-parameter fusion based sepsis early warning system according to claim 2, wherein, The training set in step S1 contains 9149 patients (15.2% died, 37.9% had shock), and the test set contains 3922 patients (15.8% died, 38.4% had shock), and there is no statistical difference in demographic and clinical indicators between the two groups (P>0.05).

4. The multi-parameter fusion based early warning system of sepsis according to claim 1, characterized in that, The statistical analysis software in step S2 includes SPSS 25.0 and R 4.3.3, and the evaluation indicators include area under the curve (AUC), accuracy (ACC), sensitivity (SEN), and specificity (SPE) (test level α=0.05).

5. The multi-parameter fusion based early warning system of sepsis according to claim 1, wherein, The 26 shock-related indicators screened in step S3 include SOFA, GCS, age, heart rate, systolic blood pressure, diastolic blood pressure, mean blood pressure, respiratory rate, blood oxygen saturation, base excess, carbon dioxide partial pressure, oxygen partial pressure, hematocrit, hemoglobin, platelet count, white blood cell count, urea nitrogen, blood sodium, blood calcium, blood chloride, blood potassium, blood creatinine, prothrombin time, standardized ratio, lactic acid, and fluid volume.

6. The multi-parameter fusion based early warning system of sepsis according to claim 1, wherein, The 21 shock key indicators screened in step S4 include SOFA, GCS, age, heart rate, systolic pressure, diastolic pressure, respiratory rate, base excess, carbon dioxide partial pressure, oxygen partial pressure, hematocrit, hemoglobin, platelet count, white blood cell count, blood sodium, blood calcium, blood chloride, blood potassium, blood creatinine, lactate, and fluid infusion volume. The AUC of the RF model in the test set is 0.876 (95% CI: 0.865-0.888), the ACC is 0.816, the SEN is 0.743, and the SPE is 0.

862.

7. The multi-parameter fusion based early warning system of sepsis according to claim 1, wherein, The 28 death-related indicators screened in step S5 include SOFA, GCS, gender, age, weight, height, heart rate, systolic pressure, diastolic pressure, mean blood pressure, respiratory rate, body temperature, blood oxygen saturation, base excess, carbon dioxide partial pressure, oxygen partial pressure, hematocrit, hemoglobin, platelet count, white blood cell count, urea nitrogen, blood calcium, blood chloride, blood potassium, blood creatinine, prothrombin time, normalized ratio, and lactate.

8. The multi-parameter fusion based sepsis early warning system according to claim 6, wherein, The 15 death key indicators screened in step S6 include SOFA, GCS, age, weight, heart rate, systolic pressure, respiratory rate, body temperature, blood oxygen saturation, oxygen partial pressure, hemoglobin, urea nitrogen, blood chloride, prothrombin time, and lactate. The AUC of the RF model in the test set is 0.853 (95% CI: 0.839-0.868), the ACC is 0.716, the SEN is 0.871, and the SPE is 0.

687.

9. The multi-parameter fusion based early warning system of sepsis according to claim 1, wherein, The webpage interface in step S7 includes a shock risk prediction module (inputting 21 indicators to calculate shock probability) and a death risk prediction module (inputting 15 indicators to calculate death probability), which are used for clinicians to quantitatively evaluate patient condition progression and intervention timing.

10. The multi-parameter fusion based sepsis early warning system according to claim 8, wherein, In steps S4 and S6, the RF is the optimal model (comprehensive evaluation of ACC, SEN, and SPE).

Images