Method and system for predicting sepsis occurrence risk of ICU patient

Abstract

The invention discloses an ICU patient sepsis occurrence risk prediction method and system. The method comprises the following steps: collecting clinical time sequence data of an ICU patient; performing information extraction processing on the clinical time sequence data of the ICU patient based on a structured query language technology and a semantic analysis model; constructing a data set; constructing eight dichotomy machine learning models, and selecting the dichotomy machine learning model with the highest AUROC value as a sepsis prediction model; screening effective features; a Platt Scaling algorithm and the calibration set are adopted to carry out calibration processing on the sepsis prediction model, and a calibrated sepsis prediction model is obtained; and acquiring real-time clinical time sequence data of the target ICU patient, preprocessing the real-time clinical time sequence data of the target ICU patient, inputting the preprocessed real-time clinical time sequence data into the calibrated sepsis prediction model, and outputting and displaying a sepsis occurrence risk prediction result. According to the method, human intervention is not needed, the sepsis occurrence risk of the ICU patient can be monitored in real time, and important support is provided for early warning of sepsis.

Description

Technical Field

[0001] The present invention relates to the field of prediction of sepsis occurrence in ICU patients, and in particular to a method and system for predicting the risk of sepsis occurrence in ICU patients. Background Art

[0002] Sepsis is an inflammatory response syndrome, defined as life-threatening organ dysfunction, mainly caused by the dysregulation of the host's response to infection, and septic shock is the most severe form of sepsis. Sepsis is currently one of the diseases with high morbidity and mortality in the intensive care unit (ICU).

[0003] Sepsis is a critical illness in the ICU, characterized by rapid disease progression and high mortality. Despite the deepening of clinical understanding of the pathological mechanism of sepsis and the continuous advancement of medical level, this critical illness is still the main cause of death in the ICU. The high mortality rate of sepsis is directly related to delayed diagnosis and improper treatment. Early and comprehensive evaluation of patients with sepsis is crucial to provide rapid diagnosis, treatment and improve patient prognosis.

[0004] At present, the sepsis screening methods for ICU patients mainly include: systemic inflammatory response syndrome (SIRS) criteria, sequential organ failure assessment (SOFA) criteria or quick SOFA (qSOFA), etc. These methods have the following defects: (1) clinical methods are subjective, and their accuracy is related to the clinical experience and professionalism of doctors; (2) clinical methods cannot provide early warning of sepsis in patients, causing patients to miss the best time for early intervention.

[0005] Therefore, a method and system for predicting the risk of sepsis in ICU patients were proposed. Summary of the invention

[0006] Based on this, it is necessary to provide a method and system for predicting the risk of sepsis in ICU patients in response to the above technical problems.

[0007] According to a first aspect of the present invention, a method for predicting the risk of sepsis in ICU patients is provided, comprising: collecting clinical time series data of ICU patients according to a predefined first screening condition; performing information extraction processing on the clinical time series data of ICU patients based on structured query language technology and semantic analysis model, extracting basic information of ICU patients, clinical test feature data of baseline time just after entering the ICU, clinical test feature data of 6 hours after entering the ICU, clinical test feature data of 12 hours after entering the ICU, clinical test feature data of 18 hours after entering the ICU and previous history feature data, and calculating the maximum and minimum values ​​of various clinical test feature data of ICU patients within 18 hours after entering the ICU; the basic information of ICU patients, clinical test feature data of baseline time just after entering the ICU, clinical test feature data of 6 hours after entering the ICU, clinical test feature data of 12 hours after entering the ICU, clinical test feature data of 18 hours after entering the ICU, and previous history feature data 18 hours of clinical test feature data and historical feature data were preprocessed, and a data set was constructed; the data set was randomly divided into training set, test set and calibration set according to a preset ratio, and eight binary classification machine learning models for predicting the risk of sepsis in ICU patients were constructed. The eight binary classification machine learning models were trained and optimized through the training set and test set, six model evaluation indicators were calculated, and the binary classification machine learning model with the highest AUROC value was selected as the sepsis prediction model; the ROC curve of five-fold cross-validation was used to evaluate the performance of the sepsis prediction model, and combined with the feature correlation heat map, effective features were screened from the clinical test feature data; the sepsis prediction model was calibrated using the Platt Scaling algorithm and the calibration set, and the effect of the sepsis prediction model calibration was evaluated based on the calibration curve and Brier score to obtain the calibrated sepsis prediction model; the real-time clinical time series data of the target ICU patients were obtained, the real-time clinical time series data of the target ICU patients were preprocessed, and input into the calibrated sepsis prediction model, and the sepsis risk prediction results were output and displayed.

[0008] Optionally, collecting clinical time series data of ICU patients according to a predefined first screening condition includes: screening ICU patients who meet the first screening condition from the ICU database; acquiring clinical time series data of ICU patients based on the screened ICU patients; wherein, the first screening condition is patients who are both older than or equal to 18 years old and are entering the ICU for treatment for the first time; the screened ICU patients include ICU sepsis-positive patients and ICU sepsis-negative patients, wherein the ICU sepsis-positive patients are patients who develop sepsis within 18-24 hours after entering the ICU, and the ICU sepsis-negative patients are patients who enter the ICU for treatment within 18-24 hours.

[0009] Optionally, the basic information of ICU patients, clinical laboratory test characteristic data of the baseline time just entering the ICU, clinical laboratory test characteristic data 6 hours after entering the ICU, clinical laboratory test characteristic data 12 hours after entering the ICU, clinical laboratory test characteristic data 18 hours after entering the ICU, and previous history characteristic data are preprocessed to construct a data set, including: format unification processing, categorical variable coding conversion processing, missing value filling processing, outlier processing and normalization processing of the basic information of ICU patients, clinical laboratory test characteristic data of the baseline time just entering the ICU, clinical laboratory test characteristic data 6 hours after entering the ICU, clinical laboratory test characteristic data 12 hours after entering the ICU, clinical laboratory test characteristic data 18 hours after entering the ICU, and previous history characteristic data, and constructing a data set.

[0010] Optionally, the eight binary classification machine learning models for predicting the risk of sepsis in ICU patients include a naive Bayes model, a K nearest neighbor model, a support vector machine model, a logistic regression model, a decision tree model, an AdaBoost model, an XGBoost model and a random forest model.

[0011] Optionally, the six model evaluation indicators include accuracy, sensitivity, specificity, precision, F1 score and AUROC.

[0012] Optionally, the performance of the sepsis prediction model is evaluated using a five-fold cross-validated ROC curve, and effective features are screened out from the clinical test feature data in combination with a feature correlation heat map, including: using the sepsis prediction model to calculate the feature importance of the clinical test feature data, sorting the feature importance of the clinical test feature data, and evaluating the performance of the sepsis prediction model using a five-fold cross-validated ROC curve, combining the feature importance ranking of the clinical test feature data and the performance of the sepsis prediction model to screen out important features, draw a feature correlation heat map, and screen out effective features from the important features according to a predefined second screening condition.

[0013] Optionally, after outputting and displaying the sepsis occurrence risk prediction result, it also includes: setting an early warning threshold, and automatically triggering an early warning reminder message when the sepsis occurrence risk prediction result reaches or exceeds the early warning threshold.

[0014] According to a second aspect of the present invention, a system for predicting the risk of sepsis in ICU patients is provided, comprising: a data acquisition module, for collecting clinical time series data of ICU patients according to a predefined first screening condition; a data extraction module, for performing information extraction processing on the clinical time series data of ICU patients based on structured query language technology and semantic analysis model, extracting basic information of ICU patients, clinical test feature data of baseline time just after entering the ICU, clinical test feature data of 6 hours after entering the ICU, clinical test feature data of 12 hours after entering the ICU, clinical test feature data of 18 hours after entering the ICU and previous history feature data, and calculating the maximum and minimum values ​​of various clinical test feature data of ICU patients within 18 hours after entering the ICU; a data processing module, for performing information extraction processing on the clinical time series data of ICU patients based on structured query language technology and semantic analysis model, extracting basic information of ICU patients, clinical test feature data of baseline time just after entering the ICU, clinical test feature data of 6 hours after entering the ICU, clinical test feature data of 12 hours after entering the ICU, clinical test feature data of 18 hours after entering the ICU and previous history feature data The 18-hour clinical test feature data and historical feature data were preprocessed and a data set was constructed; the model training module was used to randomly divide the data set into training set, test set and calibration set according to a preset ratio, and eight binary classification machine learning models for predicting the risk of sepsis in ICU patients were constructed. The eight binary classification machine learning models were trained and optimized through the training set and test set, six model evaluation indicators were calculated, and the binary classification machine learning model with the highest AUROC value was selected as the sepsis prediction model; the feature selection module was used to evaluate the performance of the sepsis prediction model using the ROC curve of five-fold cross-validation, and to screen out effective features from the clinical test feature data in combination with the feature correlation heat map; the model calibration module was used to use Platt The scaling algorithm and calibration set were used to calibrate the sepsis prediction model, and the calibration effect of the sepsis prediction model was evaluated according to the calibration curve and Brier score to obtain the calibrated sepsis prediction model; the model prediction module was used to obtain the real-time clinical time series data of the target ICU patients, preprocess the real-time clinical time series data of the target ICU patients, and input them into the calibrated sepsis prediction model, and output and display the sepsis risk prediction results.

[0015] According to a third aspect of the present invention, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method when executing the computer program.

[0016] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the steps of the method when executed by a processor.

[0017] The advantages and beneficial effects of the present invention are as follows: a method and system for predicting the risk of sepsis in ICU patients provided by the present invention, the method screens out ICU patients who meet the preset conditions from the ICU database by setting a first screening condition, and simultaneously obtains the clinical time series data of the corresponding ICU patients, constructs a data set after preprocessing the clinical time series data, and randomly divides the data set into a training set, a test set and a calibration set according to a preset ratio, trains and optimizes the constructed eight binary classification machine learning models, calculates six model evaluation indicators, and selects the binary classification machine learning model with the highest AUROC value as the sepsis prediction model; then, the ROC curve of the five-fold cross-validation and the feature correlation heat map are used to perform feature selection on the screened model to improve the interpretability and prediction ability of the model; at the same time, the Platt The model is calibrated by the scaling algorithm and calibration set to improve the accuracy of the model's prediction results. In conjunction with the early warning mechanism, end-to-end prediction of the entire process can be achieved without human intervention. This can not only improve the accuracy of the prediction results of the risk of sepsis, but also monitor the risk of sepsis in ICU patients in real time to provide important support for the early warning of sepsis. Effective features can also be extracted through the sepsis prediction model to improve the interpretability and predictive ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is the overall flow chart of the present invention.

[0019] Figure 2 It is a system principle block diagram of the present invention.

[0020] Figure 3 A schematic diagram of an electronic device. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below through specific implementation methods in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0022] Embodiment 1

[0023] Figure 1 A flowchart of a method for predicting the risk of sepsis in ICU patients provided in Example 1 of the present invention.

[0024] Reference Figure 1 , the method comprises the following steps.

[0025] S1. Collect clinical time series data of ICU patients according to the predefined first screening condition.

[0026] In this embodiment, clinical time series data of ICU patients are collected according to a predefined first screening condition, including: screening ICU patients who meet the first screening condition from the ICU database; based on the screened ICU patients, obtaining clinical time series data of ICU patients; wherein, the first screening condition is patients who are both older than or equal to 18 years old and are entering the ICU for treatment for the first time; the screened ICU patients include ICU sepsis-positive patients and ICU sepsis-negative patients, wherein the ICU sepsis-positive patients are patients who develop sepsis within 18-24 hours after entering the ICU, and the ICU sepsis-negative patients are patients who enter the ICU for treatment within 18-24 hours.

[0027] Furthermore, clinical time series data include, but are not limited to, basic information of ICU patients, electronic medical record information, and clinical test characteristic data.

[0028] S2. Based on structured query language technology and semantic analysis model, information extraction and processing are performed on the clinical time series data of ICU patients to extract the basic information of ICU patients, clinical test feature data at the baseline time of entering the ICU, clinical test feature data 6 hours after entering the ICU, clinical test feature data 12 hours after entering the ICU, clinical test feature data 18 hours after entering the ICU, and historical feature data, and the maximum and minimum values ​​of various clinical test feature data within the 18-hour period of ICU patients entering the ICU are calculated.

[0029] In this embodiment, the present invention uses structured query language (SQL) technology to perform information extraction processing on clinical time series data of ICU patients, extracting basic information of ICU patients, clinical test feature data of baseline time just entering the ICU, clinical test feature data 6 hours after entering the ICU, clinical test feature data 12 hours after entering the ICU, and clinical test feature data 18 hours after entering the ICU.

[0030] Furthermore, the basic information of ICU patients includes gender, age, height, and weight.

[0031] Furthermore, the clinical test characteristic data include baseline_white blood cell count, baseline_neutrophil count, baseline_neutrophil percentage, baseline_monocyte count, baseline_platelet count, baseline_hemoglobin concentration, baseline_PT prothrombin time, baseline_APTT activated partial thromboplastin time, baseline_Fib fibrinogen quantification, baseline_D-dimer, baseline_international normalized ratio INR, baseline_albumin, baseline_ALT alanine aminotransferase, baseline_creatinine, baseline_total bilirubin, baseline_Glu glucose, baseline_sodium Na+, baseline_potassium K+, baseline_chloride Cl-, baseline_ionized calcium, baseline_body temperature, baseline_heart rate, baseline_respiratory rate, baseline_NBP non-invasive systolic blood pressure, baseline_NBP non-invasive diastolic blood pressure, baseline_NBP non-invasive mean arterial pressure, baseline_SPO2 Blood oxygen saturation, latest body temperature, latest heart rate, latest respiratory rate, latest NBP non-invasive systolic pressure, latest NBP non-invasive diastolic pressure, latest NBP non-invasive mean arterial pressure, latest SPO2 blood oxygen saturation, lowest heart rate, lowest respiratory rate, lowest NBP non-invasive systolic pressure, lowest NBP non-invasive diastolic pressure, lowest NBP non-invasive mean arterial pressure, lowest SPO2 blood oxygen saturation, highest heart rate, highest respiratory rate, highest NBP non-invasive systolic pressure, highest NBP non-invasive diastolic pressure, highest NBP non-invasive mean arterial pressure, highest SPO2 blood oxygen saturation, latest total intake, latest total output, latest urine output, baseline total lymphocyte count, latest respiratory support, latest immunosuppressants, latest hormone drugs, latest chemotherapy drugs, latest vasoconstrictor drugs.

[0032] In this embodiment, since the content of electronic medical record information is mostly descriptive text, the work of manually extracting the medical history information of ICU patients is rather cumbersome. Therefore, the present invention utilizes the semantic understanding and text generation capabilities of the semantic analysis model, and specifies the output format by optimizing the prompt words, so as to accurately identify and extract the medical history feature data of all ICU patients.

[0033] Furthermore, the past medical history characteristics include, but are not limited to, baseline_hypertension, baseline_myocardial infarction, baseline_heart function grade IV, baseline_chronic obstructive pulmonary disease, baseline_cor pulmonale, baseline_cirrhosis, baseline_chronic hepatitis, baseline_portal hypertension, baseline_esophageal varices, baseline_spontaneous peritonitis, baseline_diabetes, baseline_diabetic peripheral neuropathy, baseline_hyperosmolar glucose coma, baseline_ketoacidosis, baseline_chronic renal insufficiency, baseline_renal anemia, baseline_lymphatic Tumor, Baseline_Leukemia, Baseline_Plasmacytoma, Baseline_Myeloma, Baseline_Systemic Lupus Erythematosus, Baseline_Myositis, Baseline_Rheumatoid Arthritis, Baseline_Behcet's Disease, Baseline_Vasculitis, Baseline_Dementia, Baseline_Cerebral Hemorrhage, Baseline_Cerebral Infarction, Baseline_Cerebral Infarction, Baseline_Cerebral Thrombosis, Baseline_Solid Malignancies, Baseline_Peptic Ulcer, Baseline_Gastric Ulcer, Baseline_Duodenal Ulcer, Baseline_CRRT, Baseline_Chemotherapy, Baseline_Transplantation.

[0034] Furthermore, when the electronic medical record information is extracted, the patient's diagnosis conclusion is also extracted, that is, the outcome event is extracted as the occurrence of sepsis / non-occurrence of sepsis.

[0035] S3. Preprocess the basic information of ICU patients, clinical test characteristic data at the baseline time of entering the ICU, clinical test characteristic data 6 hours after entering the ICU, clinical test characteristic data 12 hours after entering the ICU, clinical test characteristic data 18 hours after entering the ICU, and historical characteristic data, and construct a data set.

[0036] In this embodiment, the basic information of ICU patients, clinical laboratory test characteristic data of the baseline time just entering the ICU, clinical laboratory test characteristic data 6 hours after entering the ICU, clinical laboratory test characteristic data 12 hours after entering the ICU, clinical laboratory test characteristic data 18 hours after entering the ICU, and historical characteristic data are preprocessed, and a data set is constructed, including: the basic information of ICU patients, clinical laboratory test characteristic data of the baseline time just entering the ICU, clinical laboratory test characteristic data 6 hours after entering the ICU, clinical laboratory test characteristic data 12 hours after entering the ICU, clinical laboratory test characteristic data 18 hours after entering the ICU, and historical characteristic data are format unified, categorical variable coding conversion processing, missing value filling processing, outlier processing and normalization processing are performed, and a data set is constructed.

[0037] Furthermore, the basic information of the ICU patients is processed in a unified format, and the specific steps are: the age data format of the ICU patients is unified into an integer.

[0038] Furthermore, the clinical test feature data and the past history feature data are subjected to categorical variable coding conversion processing, wherein the categorical variable coding conversion processing includes label coding or sequence coding.

[0039] Furthermore, missing values ​​are filled in the clinical test feature data. The specific steps are as follows: Impute.SimpleImputer in the preprocessing module of the python machine learning library Sklearn is used to fill the missing values ​​with the average.

[0040] Furthermore, the clinical test characteristic data are processed for outliers. The specific steps are as follows: first, the upper quartile and the lower quartile are used for judgment (the normal value range is between [Q1-1.5×IQR] and [Q3+1.5×IQR], where Q3 is the upper quartile, Q1 is the lower quartile, and IQR=Q3-Q1). In the position judged as an outlier, the average value corresponding to the indicator is used to replace and supplement.

[0041] Furthermore, the clinical test feature data is subjected to min-max normalization processing, and the data is mapped to the range [0, 1].

[0042] S4. Randomly divide the dataset into training set, test set and calibration set according to the preset ratio, construct eight binary classification machine learning models for predicting the risk of sepsis in ICU patients, train and optimize the eight binary classification machine learning models through the training set and test set, calculate six model evaluation indicators, and select the binary classification machine learning model with the highest AUROC value as the sepsis prediction model.

[0043] Furthermore, in order to build a sepsis prediction model, 80% of the samples were randomly selected from the entire dataset as the training set to train the model, 12% of the samples were used as the test set to evaluate the performance of the model, and the remaining 8% of the samples were used as the calibration set. To ensure the class balance between negative and positive samples in the training set, the training set was randomly downsampled using imblearn.under_sampling in the preprocessing module of the python machine learning library Sklearn.

[0044] Further, the present invention constructs eight different binary classification machine learning models: Naive Bayes model, K-Nearest Neighbors model, Support Vector Machines model, Logistic Regression model, Decision Tree model, AdaBoost model, XGBoost model and Random Forest model. In order to ensure the robustness of the model and prevent overfitting, a five-fold cross validation method is used, and six model evaluation indicators are calculated: Accuracy, Sensitivity, Specificity, Precision, F1 Score and Area Under the Curve (Area Under the Curve, AUC). The area under the receiver operating characteristic curve (Receiver Operating Characteristic, ROC) (Area Under the ROC Curve, AUROC) is used to test the predictive performance of the model. According to the AUROC value, the performance of the model is divided into five levels: 0.5-0.6 is poor, 0.6-0.7 is poor, 0.7-0.8 is good, 0.8-0.9 is good, and 0.9-1.0 is excellent. The results of six model evaluations of eight binary classification machine learning models are shown in Table 1 below.

[0045] Table 1 Model evaluation indicators of eight binary classification machine learning models

[0046] Model Accuracy Sensitivity Specificity Accuracy F1 score AUROC Naive Bayes Model 0.675 0.401 0.949 0.889 0.551 0.87 K nearest neighbor model 0.746 0.578 0.914 0.871 0.695 0.828 Support Vector Machine Model 0.842 0.777 0.907 0.893 0.831 0.913 Logistic regression model 0.817 0.768 0.866 0.852 0.808 0.888 Decision Tree Model 0.762 0.766 0.758 0.76 0.763 0.762 AdaBoost Model 0.853 0.833 0.873 0.868 0.85 0.928 XGBoost Model 0.867 0.853 0.881 0.877 0.865 0.942 Random Forest Model 0.858 0.837 0.88 0.875 0.855 0.93

[0047] As can be seen from Table 1 above, the AUC value of the XGBoost model is the highest, specifically 0.942. Therefore, the XGBoost model is selected as the sepsis prediction model.

[0048] S5. The ROC curve of five-fold cross-validation was used to evaluate the performance of the sepsis prediction model, and the feature correlation heat map was combined to screen out effective features from the clinical test feature data.

[0049] In this embodiment, the ROC curve of five-fold cross-validation is used to evaluate the performance of the sepsis prediction model, and combined with the feature correlation heat map, effective features are screened out from the clinical test feature data, including: using the sepsis prediction model to calculate the feature importance of the clinical test feature data, sorting the feature importance of the clinical test feature data, and using the ROC curve of five-fold cross-validation to evaluate the performance of the sepsis prediction model, combining the feature importance ranking of the clinical test feature data and the performance of the sepsis prediction model to screen out important features, draw a feature correlation heat map, and screen out effective features from the important features according to the predefined second screening condition.

[0050] Furthermore, the XGBoost model was used to calculate the feature importance of clinical test feature data, and the feature importance of clinical test feature data was ranked, and the performance of the sepsis prediction model was evaluated using the ROC curve of five-fold cross-validation. The XGBoost model tended to be stable in the ROC curve based on the five-fold cross-validation, so as to obtain important features; then, based on the important features, a feature correlation heat map was drawn, where the important features included many features with strong correlation, such as baseline_heart rate and maximum_heart rate, to prevent the two features from having a collinear relationship, retaining the more important features, and pre-defined second screening conditions, such as filtering out features with a correlation coefficient higher than 0.5, so as to screen out effective features from the important features.

[0051] Furthermore, the effective features screened out include baseline body temperature, baseline PT prothrombin time, lowest SPO2 blood oxygen saturation, latest total intake, latest vasoconstrictor, baseline APTT activated partial thromboplastin time, baseline chloride Cl-, highest respiratory rate, baseline albumin, latest immunosuppressant, highest heart rate, baseline creatinine, baseline Fib fibrinogen quantification, baseline solid malignancy, baseline cerebral hemorrhage, latest respiratory support, baseline myocardial infarction, baseline sodium Na+, baseline peptic ulcer, lowest NBP non-invasive mean arterial pressure, lowest NBP non-invasive systolic pressure, lowest NBP non-invasive diastolic pressure, baseline D-dimer, baseline ionized calcium, baseline Glu glucose, and baseline ALT alanine aminotransferase.

[0052] S6. The sepsis prediction model was calibrated using the Platt Scaling algorithm and the calibration set, and the calibration effect of the sepsis prediction model was evaluated based on the calibration curve and Brier score to obtain the calibrated sepsis prediction model.

[0053] Furthermore, in classification problems, many models will output a probability value, which indicates the confidence of the model in predicting that a sample belongs to a certain category; however, this probability value is not always consistent with the true probability; the goal of probability calibration is to make the probability output by the model as close as possible to the true classification probability.

[0054] Furthermore, the CalibratedClassifierCV provided by the python machine learning library Scikit-learn is used to build a probability calibration model. The probability calibration method adopts the Platt Scaling algorithm. Then, the probability calibration model is trained on the calibration set, and the test set is predicted using the probability calibration model. The effect of model calibration is evaluated using the calibration curve and Brier Score.

[0055] Furthermore, the Brier score calculated using brier_score_loss in the python machine learning library Scikit-learn is 0.002, and the calibration effect is very good.

[0056] S7. Acquire the real-time clinical time series data of the target ICU patients, preprocess the real-time clinical time series data of the target ICU patients, and input the data into the calibrated sepsis prediction model, and output and display the sepsis risk prediction results.

[0057] Furthermore, by acquiring real-time clinical time series data of ICU patients, the real-time clinical time series data is processed and input into the calibrated XGBoost model, and the sepsis risk prediction results are output to facilitate sepsis risk prediction. At the same time, the output sepsis risk prediction results are displayed in an intuitive form for easy viewing by doctors.

[0058] In this embodiment, after outputting and displaying the sepsis occurrence risk prediction result, it also includes: setting an early warning threshold, and when the sepsis occurrence risk prediction result reaches or exceeds the early warning threshold, the early warning reminder information is automatically triggered.

[0059] Furthermore, when the risk prediction result of sepsis is a high risk of sepsis, the preset warning threshold will automatically trigger a warning reminder message to remind the doctor to pay close attention to the patient and take appropriate early intervention according to the situation to reduce the patient's risk of disease.

[0060] Embodiment 2

[0061] This embodiment provides a system 200 for predicting the risk of sepsis in ICU patients based on the above-mentioned embodiment 1. Figure 2, implementing a method for predicting the risk of sepsis in ICU patients as described in Example 1, the system includes a data acquisition module 210, a data extraction module 220, a data processing module 230, a model training module 240, a feature selection module 250, a model calibration module 260 and a model prediction module 270.

[0062] In this embodiment, the data collection module 210 is used to collect clinical time series data of ICU patients according to a predefined first screening condition.

[0063] In this embodiment, the data extraction module 220 is used to perform information extraction processing on the clinical time series data of ICU patients based on structured query language technology and semantic analysis model, extract the basic information of ICU patients, clinical test feature data of the baseline time just entering the ICU, clinical test feature data 6 hours after entering the ICU, clinical test feature data 12 hours after entering the ICU, clinical test feature data 18 hours after entering the ICU and historical feature data, and calculate the maximum and minimum values ​​of various clinical test feature data of ICU patients within the 18-hour period after entering the ICU.

[0064] In this embodiment, the data processing module 230 is used to preprocess the basic information of ICU patients, the clinical test characteristic data of the baseline time just entering the ICU, the clinical test characteristic data 6 hours after entering the ICU, the clinical test characteristic data 12 hours after entering the ICU, the clinical test characteristic data 18 hours after entering the ICU, and the historical characteristic data, and construct a data set.

[0065] In this embodiment, the model training module 240 is used to randomly divide the data set into a training set, a test set, and a calibration set according to a preset ratio, construct eight binary classification machine learning models for predicting the risk of sepsis in ICU patients, train and optimize the eight binary classification machine learning models through the training set and the test set, calculate six model evaluation indicators, and select the binary classification machine learning model with the highest AUROC value as the sepsis prediction model.

[0066] In this embodiment, the feature selection module 250 is used to evaluate the performance of the sepsis prediction model using a five-fold cross-validation ROC curve, and to screen out effective features from the clinical test feature data in combination with the feature correlation heat map.

[0067] In this embodiment, the model calibration module 260 is used to calibrate the sepsis prediction model using the Platt Scaling algorithm and the calibration set, and evaluate the effect of the sepsis prediction model calibration according to the calibration curve and the Brier score to obtain the calibrated sepsis prediction model.

[0068] In this embodiment, the model prediction module 270 is used to obtain the real-time clinical time series data of the target ICU patient, preprocess the real-time clinical time series data of the target ICU patient, and input it into the calibrated sepsis prediction model to output and display the sepsis risk prediction results.

[0069] Embodiment 3

[0070] This embodiment further provides an electronic device based on the above embodiment 1. Figure 3 , Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0071] like Figure 3 As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. In RAM 303, various programs and data required for the operation of the electronic device are also stored. The processing device 301, ROM 302, and RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0072] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, etc., an output device 307 including, for example, a liquid crystal display (LCD), a speaker, etc., a storage device 308 including, for example, a magnetic tape, a hard disk, etc., and a communication device 309. The communication device 309 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Figure 3 An electronic device having various devices is shown, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0073] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0074] Embodiment 4

[0075] Based on the above-mentioned embodiment 1, this embodiment further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned method are implemented.

[0076] It should be noted that the computer-readable medium in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0077] In this embodiment, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperTextTransferProtocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an adhoc peer-to-peer network), as well as any currently known or future developed network.

[0078] The computer-readable medium may be included in the device, or may exist independently without being installed in the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains training data, converts the training data into initial data; determines an initial rule base based on the initial data, and optimizes the parameters of the initial rule base to obtain a target rule base; calculates the rules in the target rule base according to a preset activation weight calculation formula to obtain activation weights; determines abnormal information according to the test data and the activation weights.

[0079] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0080] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0081] The units described in some embodiments of the present disclosure may be implemented by software or hardware. The units described may also be provided in a processor, for example, may be described as: a processor including a data acquisition unit, a rule determination unit, a weight calculation unit, and an anomaly determination unit. The names of these units do not, in some cases, constitute limitations on the units themselves, for example, the data acquisition unit may also be described as a "unit for acquiring training data".

[0082] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), etc.

[0083] Obviously, it should be understood by those skilled in the art that the above-mentioned various steps of the present invention can be performed in a manner different from the present invention, and the simulation method and experimental equipment include but are not limited to the above description. The above-mentioned various steps of the present invention can be performed in an order different from that here in some cases, and the steps shown or described above can be performed separately. Therefore, the present invention is not limited to any specific combination of hardware and software.

[0084] The above contents are further detailed descriptions of the present invention in combination with specific implementation methods, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the scope of protection of the present invention.

Claims

1. A method for predicting the risk of sepsis in ICU patients, characterized in that: include: Clinical time series data of ICU patients were collected according to the predefined first screening criteria; Based on structured query language technology and semantic analysis model, information extraction and processing were performed on the clinical time series data of ICU patients. The basic information of ICU patients, clinical test feature data of baseline time just entering the ICU, clinical test feature data of 6 hours after entering the ICU, clinical test feature data of 12 hours after entering the ICU, clinical test feature data of 18 hours after entering the ICU, and historical feature data were extracted. The maximum and minimum values ​​of various clinical test feature data of ICU patients within 18 hours after entering the ICU were calculated. The basic information of ICU patients, clinical test characteristic data at baseline time of entering the ICU, clinical test characteristic data 6 hours after entering the ICU, clinical test characteristic data 12 hours after entering the ICU, clinical test characteristic data 18 hours after entering the ICU, and historical characteristic data were preprocessed and a data set was constructed; The data set was randomly divided into training set, test set and calibration set according to the preset ratio, and eight binary classification machine learning models for predicting the risk of sepsis in ICU patients were constructed. The eight binary classification machine learning models were trained and optimized through the training set and test set, and six model evaluation indicators were calculated. The binary classification machine learning model with the highest AUROC value was selected as the sepsis prediction model. The performance of the sepsis prediction model was evaluated using the ROC curve of five-fold cross-validation, and effective features were screened from the clinical test feature data in combination with the feature correlation heat map; The Platt Scaling algorithm and calibration set were used to calibrate the sepsis prediction model, and the calibration effect of the sepsis prediction model was evaluated based on the calibration curve and Brier score to obtain the calibrated sepsis prediction model. The real-time clinical time series data of the target ICU patients are obtained, preprocessed, and input into the calibrated sepsis prediction model, and the sepsis risk prediction results are output and displayed.

2. A method for predicting the risk of sepsis in ICU patients according to claim 1, characterized in that: The collecting of clinical time series data of ICU patients according to the predefined first screening condition includes: ICU patients who met the first screening criteria were screened from the ICU database; Based on the screened ICU patients, clinical time series data of ICU patients were obtained; Among them, the first screening condition is that the patients are both older than or equal to 18 years old and are entering the ICU for treatment for the first time; the screened ICU patients include ICU sepsis-positive patients and ICU sepsis-negative patients, among which the ICU sepsis-positive patients are patients who develop sepsis within 18-24 hours after entering the ICU, and the ICU sepsis-negative patients are patients who enter the ICU for treatment within 18-24 hours.

3. The method for predicting the risk of sepsis in ICU patients according to claim 1, characterized in that: The basic information of ICU patients, clinical test characteristic data of baseline time just entering ICU, clinical test characteristic data of 6 hours after entering ICU, clinical test characteristic data of 12 hours after entering ICU, clinical test characteristic data of 18 hours after entering ICU and past history characteristic data are preprocessed, and a data set is constructed, including: The basic information of ICU patients, clinical laboratory characteristic data at the baseline time of entering the ICU, clinical laboratory characteristic data 6 hours after entering the ICU, clinical laboratory characteristic data 12 hours after entering the ICU, clinical laboratory characteristic data 18 hours after entering the ICU, and historical characteristic data were unified in format, categorical variable coding was converted, missing values ​​were filled, outliers were processed, and normalized to construct a data set.

4. A method for predicting the risk of sepsis in ICU patients according to claim 1, characterized in that: The eight binary classification machine learning models for predicting the risk of sepsis in ICU patients include a naive Bayes model, a K nearest neighbor model, a support vector machine model, a logistic regression model, a decision tree model, an AdaBoost model, an XGBoost model and a random forest model.

5. The method for predicting the risk of sepsis in ICU patients according to claim 1, characterized in that: The six model evaluation indicators include accuracy, sensitivity, specificity, precision, F1 score and AUROC.

6. A method for predicting the risk of sepsis in ICU patients according to claim 1, characterized in that: The ROC curve of five-fold cross validation was used to evaluate the performance of the sepsis prediction model, and combined with the feature correlation heat map, effective features were screened from the clinical test feature data, including: The sepsis prediction model was used to calculate the feature importance of clinical test feature data, and the feature importance of clinical test feature data was ranked. The ROC curve of five-fold cross-validation was used to evaluate the performance of the sepsis prediction model. The important features were screened out by combining the feature importance ranking of clinical test feature data and the performance of the sepsis prediction model, and a feature correlation heat map was drawn. According to the predefined second screening condition, effective features were screened out from the important features.

7. A method for predicting the risk of sepsis in ICU patients according to claim 1, characterized in that: After outputting and displaying the sepsis risk prediction result, it also includes: setting an early warning threshold, and automatically triggering an early warning reminder message when the sepsis risk prediction result reaches or exceeds the early warning threshold.

8. A system for predicting the risk of sepsis in ICU patients, characterized in that: include: A data collection module, used for collecting clinical time series data of ICU patients according to a predefined first screening condition; The data extraction module is used to extract information from the clinical time series data of ICU patients based on structured query language technology and semantic analysis models, extract the basic information of ICU patients, clinical test feature data at the baseline time of entering the ICU, clinical test feature data 6 hours after entering the ICU, clinical test feature data 12 hours after entering the ICU, clinical test feature data 18 hours after entering the ICU, and past history feature data, and calculate the maximum and minimum values ​​of various clinical test feature data within the 18-hour period of ICU patients entering the ICU; The data processing module is used to pre-process the basic information of ICU patients, clinical test characteristic data of the baseline time just entering the ICU, clinical test characteristic data of 6 hours after entering the ICU, clinical test characteristic data of 12 hours after entering the ICU, clinical test characteristic data of 18 hours after entering the ICU, and past history characteristic data, and construct a data set; The model training module is used to randomly divide the data set into a training set, a test set, and a calibration set according to a preset ratio, build eight binary classification machine learning models for predicting the risk of sepsis in ICU patients, train and optimize the eight binary classification machine learning models through the training set and the test set, calculate six model evaluation indicators, and select the binary classification machine learning model with the highest AUROC value as the sepsis prediction model; The feature selection module is used to evaluate the performance of the sepsis prediction model using a five-fold cross-validation ROC curve and to select effective features from the clinical test feature data in combination with the feature correlation heat map; A model calibration module is used to calibrate the sepsis prediction model using the Platt Scaling algorithm and the calibration set, and evaluate the effect of the sepsis prediction model calibration according to the calibration curve and the Brier score to obtain the calibrated sepsis prediction model; The model prediction module is used to obtain the real-time clinical time series data of the target ICU patients, preprocess the real-time clinical time series data of the target ICU patients, and input them into the calibrated sepsis prediction model to output and display the sepsis risk prediction results.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images