Method and system for predicting resuscitation status of septic shock patient
By using a gradient boosting ensemble classification model (GBC) to process multidimensional clinical information, this method addresses the insufficient accuracy and stability of predicting the resuscitation status of septic shock patients in existing technologies. It enables accurate prediction of the probability of achieving overall resuscitation goals by the 6th hour within 0-2 hours of ICU admission, supporting timely intervention and improving patient treatment outcomes.
Patent Information
- Application Number
- CN202610265342.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-05
- Publication Date
- 2026-06-23
Smart Images
Figure CN122266710A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical artificial intelligence technology, specifically to a method and system for predicting the resuscitation status of patients with septic shock. Background Technology
[0002] Septic shock (SS) is an acute and critical illness caused by a systemic inflammatory response triggered by infection, leading to circulatory, cellular, and metabolic abnormalities, significantly impaired effective circulatory perfusion, and consequently, tissue hypoperfusion and organ dysfunction. This disease is characterized by rapid onset, rapid progression, and high mortality, and is one of the leading causes of early death in severely infected patients. Studies have shown that the first 6 hours after admission to the intensive care unit are a critical time window for resuscitation; whether effective correction of circulatory and metabolic function is achieved during this period is closely related to subsequent prognosis.
[0003] Currently, there are various traditional scoring systems and monitoring indicators for the clinical assessment of patients with septic shock. Common methods include the Acute Physiology and Chronic Health Evaluation II (APACHE II), Sequential Organ Failure Assessment (SOFA), and rapid SOFA (qSOFA). These are combined with parameters such as mean arterial pressure (MAP), lactate and its clearance rate, and capillary refill time (CRT) for comprehensive interpretation and are used for risk identification in the early stages of ICU admission.
[0004] While these traditional methods have some clinical guiding significance, they share common limitations:
[0005] First, the scoring is mostly based on fixed thresholds or linear weighting, making it difficult to quantify the joint effects and interactions between continuous variables. Second, the coverage of the synergistic state of multiple domains of circulation and metabolism is insufficient, resulting in an inadequate characterization of complex, physiologically coupled processes. Consequently, the accuracy and stability of the predictions are limited, making it difficult to meet the needs of resuscitation scenarios with high timeliness and high heterogeneity.
[0006] In recent years, to improve assessment efficiency and prediction accuracy, some studies have introduced machine learning for risk modeling, employing algorithms such as random forests, gradient boosting (e.g., XGBoost / LightGBM), support vector machines, or neural networks, achieving high discriminative power on specific datasets. However, these models often target endpoints such as disease severity or long-term mortality, making it difficult to cover key stages in septic shock resuscitation management—especially the composite, procedural endpoint of achieving overall resuscitation goals by the 6th hour after ICU admission. Consequently, they lack direct support for early intervention triggering and process management.
[0007] Existing methods have significant shortcomings in both "combined effects / interaction modeling capabilities" and "coverage of short-term process endpoints." Therefore, it is necessary to develop a predictive approach for early resuscitation management of septic shock to address these issues. Summary of the Invention
[0008] Technical problems to be solved
[0009] To address the shortcomings of existing technologies, this invention provides a method and system for predicting the resuscitation status of patients with septic shock. It can prospectively output the overall probability of achieving resuscitation targets at the 6th hour by utilizing readily available multidimensional clinical information within 0 to 2 hours after admission to the ICU.
[0010] (II) Technical Solution
[0011] According to one aspect of the present invention, a method for predicting the resuscitation status of a patient with septic shock is provided, comprising:
[0012] Obtain the status parameters of patients with septic shock;
[0013] The acquired state parameters are preprocessed to obtain the input parameters;
[0014] Based on the critical illness database, training data is obtained, and the prediction model is trained using the training data.
[0015] The input parameters are input into the trained prediction model to obtain the first prediction probability;
[0016] The first predicted probability is divided into two categories to obtain the final prediction result.
[0017] As a further technical solution, the state parameters include target parameters, which include multiple variables such as mean arterial pressure, arterial blood lactate, base excess, pH, bicarbonate, and anion gap.
[0018] As a further technical solution, the acquired state parameters are preprocessed, including:
[0019] Data cleaning and outlier handling: physiological boundaries and rationality checks are performed on continuous variables, and the data is truncated according to quantile intervals (e.g., 1% to 99th percentile).
[0020] For missing values, multiple imputation (MICE) or equivalent methods are used for continuous variables, and mode imputation is used for categorical variables.
[0021] Standardization of continuous variables: The above continuous variables are standardized using the following formula:
[0022] ;
[0023] in, For the target parameter, the first Input features corresponding to continuous variables This is the original value of the variable. and These are the mean and standard deviation of the variable in the training set data of the prediction model, respectively.
[0024] As a further technical solution, preprocessing the acquired state parameters also includes:
[0025] Categorical variable encoding: When the state parameter includes a categorical or binary indicator variable, the variable is one-hot encoded.
[0026] As a further technical solution, training data is obtained based on a critical illness database. This training data is then used to train the prediction model, including:
[0027] Obtain the sample set used in this iteration. The sample set includes samples with multiple input parameters and the true label of each sample.
[0028] Each sample is input into the prediction model to obtain the second prediction probability corresponding to each sample output by the prediction model.
[0029] Calculate the cross-entropy loss for the sample set based on each true label and each second prediction probability;
[0030] Calculate the negative gradient of the cross-entropy loss with respect to the current model output score for each sample, i.e., the second predicted probability, and use the negative gradient as the pseudo residual for each sample.
[0031] The pseudo-residual fitting regression tree-based learner is used to determine the leaf node output value of the regression tree-based learner along the negative gradient direction with a preset learning rate.
[0032] The trained regression tree-based learner is added to the additive scoring function to update the ensemble structure of the prediction model.
[0033] As a further technical solution, the first predicted probability is binary classified to obtain the final prediction result, including:
[0034] If the first predicted probability is above the set classification threshold, then the first predicted result is output.
[0035] If the first predicted probability is less than the set classification threshold, then the second predicted result is output.
[0036] According to one aspect of the present invention, a system for predicting the resuscitation status of a patient with septic shock is provided, comprising:
[0037] The acquisition module is used to acquire the status parameters of patients with septic shock.
[0038] The conversion module is used to preprocess the state parameters to obtain the input parameters;
[0039] The training module is used to obtain training data based on the critical illness database, and to train the prediction model using the training data.
[0040] The prediction module is used to input the input parameters into the trained prediction model and obtain the first prediction probability output by the prediction model.
[0041] The output module is used to perform binary classification on the first predicted probability to obtain the final prediction result.
[0042] According to one aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for predicting the resuscitation status of a patient with septic shock.
[0043] According to one aspect of the present invention, a non-transitory computer read storage medium is provided, the non-transitory computer read storage medium storing computer instructions that cause the computer to perform the steps of the method for predicting the resuscitation status of a patient with septic shock.
[0044] According to one aspect of the present invention, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the steps of the method for predicting the resuscitation status of a patient with septic shock.
[0045] (III) Beneficial Effects
[0046] Compared with the prior art, the present invention provides a method and system for predicting the resuscitation status of patients with septic shock, which has the following beneficial effects:
[0047] This method and system for predicting the resuscitation status of patients with septic shock can prospectively output the overall probability of achieving resuscitation targets at the 6th hour using routinely available multidimensional clinical information within 0 to 2 hours after admission to the ICU, which helps to intervene and treat patients in a timely and accurate manner. Attached Figure Description
[0048] Figure 1 This is a flowchart illustrating the method for predicting the resuscitation status of patients with septic shock provided by the present invention.
[0049] Figure 2 This is a schematic diagram illustrating the performance of the prediction model provided by this invention on the training and test sets.
[0050] Figure 3 This is a schematic diagram of the calibration curves of the prediction model provided by this invention on the training set and the test set.
[0051] Figure 4 This is a schematic diagram of the SHAP values of the various parameters provided by this invention.
[0052] Figure 5 This is a schematic diagram of the structure of the septic shock patient status prediction system provided by the present invention.
[0053] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0054] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0055] This invention provides a method for predicting the resuscitation status of patients with septic shock, such as... Figure 1 As shown, it includes:
[0056] Obtain the status parameters of patients with septic shock;
[0057] The acquired state parameters and training data are preprocessed to obtain the input parameters;
[0058] Based on the critical illness database, training data is obtained, and the prediction model is trained using the training data.
[0059] Input the input parameters into the trained prediction model to obtain the first prediction probability, which is the probability that a septic shock patient will achieve overall resuscitation within 6 hours after being admitted to the intensive care unit (ICU).
[0060] The first predicted probability is divided into two categories to obtain the final prediction result.
[0061] In this embodiment of the invention, the status parameters of patients with septic shock are derived from basic data obtained in the intensive care unit from 0 to 2 hours, and do not depend on imaging results.
[0062] The status parameters include target parameters, which are multidimensional clinical variables obtained within 0 to 2 hours of admission to the ICU. The target parameters include 24 variables such as mean arterial pressure, arterial blood lactate, base excess, pH, bicarbonate, and anion gap.
[0063] Specifically, the parameters include 24 input variables: age, weight, Charlson Comorbidity Index (CCI), red blood cell count (RBC), platelet count, red blood cell distribution width (RDW), serum sodium, serum potassium, total calcium, chloride, blood glucose, anion gap, activated partial thromboplastin time (aPTT), international normalized ratio (INR), total bilirubin, alanine aminotransferase (ALT), blood urea nitrogen (BUN), creatinine, lactate dehydrogenase (LDH), vasopressin use, albumin use, mean arterial pressure (MAP), lactate, and base excess (BE).
[0064] In this embodiment of the invention, the acquired state parameters are preprocessed, including the following steps:
[0065] 1. Data cleaning and outlier handling: Perform physiological boundary and rationality checks on continuous variables, and truncate them according to quantile intervals (e.g., 1% to 99th percentile) to reduce the impact of extreme outliers on model training.
[0066] For example, for the continuous variable of lactate, its physiologically reasonable range can be preset. When the initial lactate value is less than 0 mmol / L, it is adjusted to 0 mmol / L; when the initial lactate value is greater than 20 mmol / L, it is adjusted to 20 mmol / L. Based on this, the first percentile of lactate is obtained statistically from the training set of the prediction model. and the 99th percentile , will be less than The lactate value cutoff is , will be greater than The lactate value cutoff is .
[0067] 2. Handling missing values: For continuous variables, use multiple imputation (MICE) or equivalent methods; for categorical variables, use mode imputation.
[0068] 3. Standardization of continuous variables: Standardize continuous variables; for the k-th continuous variable in the target parameters, its corresponding input feature is obtained by subtracting the mean of the variable in the training set data of the prediction model from the original value of the variable, and then dividing by the standard deviation of the variable.
[0069] The above continuous variables are standardized using the following formula:
[0070] ;
[0071] in, For the target parameter, the first Input features corresponding to continuous variables This is the original value of the variable. and These are the mean and standard deviation of the variable in the training set data of the prediction model, respectively.
[0072] It is expressed as follows:
[0073] Obtain the raw values of individual parameters in the target parameters, and the mean and standard deviation of individual parameters in the training set of the prediction model;
[0074] The input parameters corresponding to a single parameter are determined based on the original value, mean, and standard deviation.
[0075] 4. Categorical variable encoding: When the state parameter includes categorical or binary indicator variables, the variable is encoded using one-hot encoding.
[0076] For example, if vasopressin use and albumin use are binary variables, "yes" can be encoded as (1, 0) and "no" as (0, 1), as shown below:
[0077] When the binary indicator variable takes the value "yes", it is encoded as the input parameter (1, 0).
[0078] When the binary indicator variable takes the value "no", it is encoded as the input parameter (0, 1).
[0079] This step can be omitted if the state parameters do not include categorical variables.
[0080] After the above processing, the input feature vector x formed by the mapping of the above state parameters is obtained, and it is used as the input parameter of the prediction model.
[0081] In this embodiment of the invention, the prediction model is a gradient boosting ensemble classification model (GBC). The input feature vector is composed of 24 state parameters after preprocessing and standardization in step S2. ,in Input feature values for the following status parameters: age, weight, Charlson comorbidity index, red blood cell count, platelet count, red blood cell distribution width, serum sodium, serum potassium, total calcium, chloride ion, blood glucose, anion gap, activated partial thromboplastin time, international normalized ratio, total bilirubin, alanine aminotransferase, blood urea nitrogen, creatinine, lactate dehydrogenase, vasopressin usage indicator, albumin usage indicator, mean arterial pressure, lactate, and base excess.
[0082] The expression for the prediction model is as follows:
[0083] ;
[0084] in, Input vector with state parameters The first predicted probability obtained from the independent variable is (0,1); For input vector It is an additive score function for the independent variable.
[0085] Model training uses log loss (cross-entropy), in which... The expression on the sample set of samples can be represented as:
[0086] ;
[0087] in, Let be the number of samples in the sample set. For the first The true label of a sample when the overall recovery of that sample reaches the target at 6 hours. = 1, otherwise =0; To predict the model's performance on the current iteration of the th iteration The first predicted probability of each sample output The input feature vector is formed by preprocessing 24 status parameters obtained from the sample within 0 to 2 hours of admission to the ICU.
[0088] The second iteration can calculate the pseudo-residual based on the above logarithmic loss function and perform an additive update: First, the logarithmic loss function... Regarding the current model output score Taking the partial derivative, we get the first... In the nth iteration The negative gradient (pseudo residual) of each sample.
[0089] ,
[0090] in, For the first The predicted probability at the next iteration; subsequently, based on sample pairs... Fitting a regression tree-based learner to the training data and in accordance with The additive form updates the additive score function, where, For the first Round-regression tree-based learners, whose leaf node output values are determined by the direction... Minimize the log loss function Sure, For the first The learning rate or step size of each iteration.
[0091] Understandably, a predictive model needs to be trained before it can be used to predict a patient's resuscitation status. In this invention, the data used during the training phase can be derived from a publicly available critical care database (MIMIC-IV). The training data is randomly stratified into training and test sets in a 7:3 ratio. The training set is used for model training and cross-validation, while the test set is used for performance evaluation. To further improve the model's generalization ability, a 5-fold cross-validation strategy is used in the training set, randomly dividing the training data into five subsets. Four subsets are used for training in rotation, and one subset is used for validation, repeated five times. The final average validation performance is used as the basis for model selection and parameter tuning. Cross-validation is only performed within the training set and does not include the test set.
[0092] Training the prediction model requires multiple iterations until the model converges (e.g., the log loss on the validation set no longer decreases or its change is less than 1%). Or the number of iterations reaches the maximum number (e.g., 1000). For each iteration, in this embodiment of the invention, training data is obtained based on the critical illness database, and the prediction model is trained using the training data, including:
[0093] Obtain the sample set used in this iteration. The sample set includes samples with multiple input parameters and the true label of each sample.
[0094] Each sample is input into the prediction model to obtain the second prediction probability corresponding to each sample output by the prediction model.
[0095] Calculate the cross-entropy loss for the sample set based on each true label and each second prediction probability;
[0096] Calculate the negative gradient of the cross-entropy loss with respect to the current model output score for each sample, i.e., the second predicted probability, and use the negative gradient as the pseudo residual for each sample.
[0097] The regression tree-based learner is fitted based on pseudo-residuals, and the leaf node output value of the regression tree-based learner is determined along the negative gradient direction with a preset learning rate.
[0098] The trained regression tree-based learner is added to the additive scoring function to update the ensemble structure of the prediction model.
[0099] Furthermore, if the prediction model still fails to converge after 1000 training iterations, the following strategies can be used for optimization: reduce the learning rate (e.g., from 0.01 to 0.005), moderately increase the number or depth of base learners, and adjust the subsampling / column sampling ratio to improve robustness; if necessary, probabilistic calibration (ordination-preserving regression or Platt scaling) can be performed to improve calibration consistency.
[0100] In this embodiment of the invention, the input feature vector of the patient to be predicted is... Input the trained prediction model to obtain the first prediction probability. The first predictive probability characterizes the probability that a septic shock patient will achieve overall resuscitation goals by the 6th hour after admission to the intensive care unit (ICU). The value range is (0,1).
[0101] "Overall recovery achieved" is the composite endpoint, satisfying:
[0102] Circulatory target achieved: mean arterial pressure (MAP) ≥ 65 mmHg within the T6 window and no increase in vasopressor dosage;
[0103] Metabolic / acid-base domain targets: At least one of the following targets must be met: ① Lactate domain target (6 h lactate ≤ 2 mmol / L, or clearance rate according to lactate stratification upon admission); ② Acid-base domain target (BE > −2 mmol / L, or pH ≥ 7.35, or HCO3⁻ ≥ 20 mmol / L).
[0104] The final qualification criteria are: the cyclic domain is qualified and (either the lactate domain or the acid-base domain is qualified). Any variable used for T6 composite endpoint determination is not used as a model input feature, but only for label generation to avoid information leakage.
[0105] The predictive model is one that predicts the probability of achieving overall resuscitation goals at 6 hours based on the status parameters of septic shock patients, preferably a Gradient Boosting Classifier (GBC). The determination time can be 6 hours (T6) and can be adjusted in conjunction with clinical pathways. In other words, this invention provides a core predictive model for outputting the probability of achieving overall resuscitation goals at 6 hours, and multiple working thresholds (such as conservative / balanced / aggressive) can be configured in different deployment scenarios to adapt to different intervention strategies.
[0106] If the first predicted probability is high, the patient is more likely to achieve overall resuscitation goals by the 6th hour. Predicting the first predicted probability in this way allows for early optimization of fluid and vasopressor regimens, retesting plans, and monitoring intensity. If the first predicted probability is low, there is a higher risk of failure to achieve goals, facilitating earlier, more intensive monitoring and individualized intervention, reducing the risk of missing critical treatment opportunities. As can be seen from the above, the resuscitation status prediction method for septic shock patients of this invention can prospectively provide the probability of achieving overall resuscitation goals by the 6th hour based on the patient's early status parameters, achieving risk stratification and intervention triggering within the critical time window, which is beneficial for patient treatment and prognosis improvement.
[0107] In this embodiment of the invention, the first predicted probability Perform binary classification to obtain the final prediction results, including:
[0108] If the first predicted probability is above the set classification threshold, then the first predicted result is output.
[0109] If the first predicted probability is less than the set classification threshold, then the second predicted result is output.
[0110] Specifically, if the first predicted probability is above the set classification threshold, the first predicted result is output; if the first predicted probability is less than the set classification threshold, the second predicted result is output. The set classification threshold can be 0.5, the first predicted result can be 1, and the second predicted result can be 0; alternatively, the operating point with better net benefit can be selected based on validation set performance and decision curve analysis (DCA). Under this judgment, a first predicted result of 1 represents a predicted result of "overall recovery achieved by the 6th hour," and a first predicted result of 0 represents a predicted result of "not achieved." Thus, the final predicted result can be obtained by binary classification of the first predicted probability, specifically expressed as:
[0111] ;
[0112] This invention constructs a single prediction model based on the Gradient Boosting Classifier (GBC) ensemble classification model. For example... Figure 2The following figures show the performance of the model on the training and test sets: ROC (TrainingSet) for Gradient Boosting Classifier is the ROC curve of GBC on the training set, False Positive Rate is the false positive rate, and True Positive Rate is the true positive rate; PR Curve (Train Set) for Gradient Boosting Classifier is the precision-recall curve on the training set, Recall is the recall rate, and Precision is the precision rate; ROC (Test Set) for Gradient Boosting Classifier is the ROC curve of GBC on the test set; PR Curve (Test Set) for Gradient Boosting Classifier is the precision-recall curve on the test set.
[0113] like Figure 3 The diagram shows the calibration curves of the model on the training and test sets: In the figure, Calibration Curve (Train Set) for Gradient Boosting Classifier is the calibration curve for the training set, Predictedprobability is the predicted probability, and True probability is the true probability; Calibration Curve (Test Set) for Gradient Boosting Classifier is the calibration curve for the test set, and the orange dashed line represents the ideal calibration line.
[0114] This invention also interprets GBC using the SHAP (SHapley Additive exPlanations) method, with the following results: Figure 4 As shown in the figure: SHAP value (impact on model output) is the SHAP value (the impact on model output), and feature value is the parameter value; Figure 4 A is the global Beeswarm summary graph. Figure 4 B represents a dependency / interaction relationship based on characteristics such as age. Figure 4 C is a bar chart of the average absolute SHAP value. Figure 4 D is a SHAP waterfall plot of a representative sample.
[0115] This invention provides a system for predicting the resuscitation status of patients with septic shock, such as... Figure 5 As shown, it includes:
[0116] The acquisition module is used to acquire the status parameters of patients with septic shock.
[0117] The conversion module is used to preprocess the state parameters to obtain the input parameters;
[0118] The training module is used to obtain training data based on the critical illness database, and then use the training data to train the prediction model.
[0119] The prediction module is used to input the input parameters into the prediction model and obtain the first prediction probability output by the prediction model.
[0120] The output module is used to perform binary classification on the first predicted probability to obtain the final prediction result.
[0121] The resuscitation status prediction system for septic shock patients provided in this embodiment of the invention employs... Figure 5 Several modules within the system can prospectively output the overall resuscitation target probability at 6 hours using readily available multidimensional clinical information within 0-2 hours after admission to the ICU.
[0122] It should be noted that the system embodiments provided by this invention, in addition to implementing the methods in the above method embodiments, are also used to implement the methods in other method embodiments provided by this invention. The difference lies only in setting corresponding functional modules, and their principles are basically the same as those of the above system embodiments provided by this invention. As long as those skilled in the art, based on the above system embodiments and referring to the specific technical solutions in other method embodiments, obtain corresponding technical means and technical solutions composed of these technical means by combining technical features, and improve the modules in the above system embodiments while ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments. For example:
[0123] Based on the above system embodiments, as a preferred embodiment, the resuscitation status prediction system for septic shock patients provided in this embodiment of the invention includes target parameters, which include multiple variables such as mean arterial pressure, arterial blood lactate, base excess, pH, bicarbonate, and anion gap.
[0124] Based on the above system embodiments, as a preferred embodiment, the resuscitation status prediction system for septic shock patients provided in this invention preprocesses the acquired status parameters, including:
[0125] Data cleaning and outlier handling: physiological boundaries and rationality checks are performed on continuous variables, and the data is truncated according to quantile intervals;
[0126] For missing values, multiple imputation or equivalent methods are used for continuous variables, and mode imputation is used for categorical variables.
[0127] Standardization of continuous variables: The above continuous variables are standardized using the following formula:
[0128] ;
[0129] in, For the target parameter, the first Input features corresponding to continuous variables This is the original value of the variable. and These are the mean and standard deviation of the variable in the training set data of the prediction model, respectively.
[0130] Based on the above system embodiments, as a preferred embodiment, the resuscitation status prediction system for septic shock patients provided in this invention further includes preprocessing the acquired status parameters and:
[0131] Categorical variable encoding: When the state parameter includes a categorical or binary indicator variable, the variable is one-hot encoded.
[0132] Based on the above system embodiments, as a preferred embodiment, the resuscitation status prediction system for septic shock patients provided in this embodiment of the invention obtains training data based on a critical care database, and trains a prediction model using the training data, including:
[0133] Obtain the sample set used in this iteration. The sample set includes samples with multiple input parameters and the true label of each sample.
[0134] Each sample is input into the prediction model to obtain the second prediction probability corresponding to each sample output by the prediction model.
[0135] Calculate the cross-entropy loss for the sample set based on each true label and each second prediction probability;
[0136] Calculate the negative gradient of the cross-entropy loss with respect to the current model output score for each sample, i.e., the second predicted probability, and use the negative gradient as the pseudo residual for each sample.
[0137] The regression tree-based learner is fitted based on pseudo-residuals, and the leaf node output value of the regression tree-based learner is determined along the negative gradient direction with a preset learning rate.
[0138] The trained regression tree-based learner is added to the additive scoring function to update the ensemble structure of the prediction model.
[0139] Based on the above system embodiments, as a preferred embodiment, the resuscitation status prediction system for septic shock patients provided in this invention performs binary classification on the first prediction probability to obtain the final prediction result, including:
[0140] If the first predicted probability is above the set classification threshold, then the first predicted result is output.
[0141] If the first predicted probability is less than the set classification threshold, then the second predicted result is output.
[0142] like Figure 6 As shown, the present invention also provides an electronic device, including a processor, a communications interface, a memory, and a communication bus, wherein the processor, the communications interface, and the memory communicate with each other through the communication bus. The processor can call logical instructions in the memory to execute a method for predicting the resuscitation status of patients with septic shock.
[0143] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0144] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the resuscitation status prediction method for septic shock patients provided in the above embodiments, as shown below:
[0145] Obtain the status parameters of patients with septic shock;
[0146] The acquired state parameters are preprocessed to obtain the input parameters;
[0147] Based on the critical illness database, training data is obtained, and the prediction model is trained using the training data.
[0148] Input the input parameters into the trained prediction model to obtain the first prediction probability;
[0149] The first predicted probability is divided into two categories to obtain the final prediction result.
[0150] The present invention also provides a computer program product, comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, enable the computer to execute the resuscitation status prediction method for septic shock patients provided in the above embodiments, as shown below:
[0151] Obtain the status parameters of patients with septic shock;
[0152] The acquired state parameters are preprocessed to obtain the input parameters;
[0153] Based on the critical illness database, training data is obtained, and the prediction model is trained using the training data.
[0154] Input the input parameters into the trained prediction model to obtain the first prediction probability;
[0155] The first predicted probability is divided into two categories to obtain the final prediction result.
[0156] In summary, given that septic shock patients are prone to adverse outcomes due to inadequate resuscitation, this invention can predict the probability of whether a septic shock patient has achieved overall resuscitation goals by the 6th hour after admission to the intensive care unit based on the patient's resuscitation status parameters. This facilitates timely and accurate intervention and treatment of patients.
[0157] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0158] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0159] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for predicting the resuscitation status of patients with septic shock, characterized in that, include: Obtain the status parameters of patients with septic shock; The acquired state parameters are preprocessed to obtain the input parameters; Based on the critical illness database, training data is obtained, and the prediction model is trained using the training data. The input parameters are input into the trained prediction model to obtain the first prediction probability; The first predicted probability is divided into two categories to obtain the final prediction result.
2. The method for predicting the resuscitation status of patients with septic shock according to claim 1, characterized in that, The state parameters include target parameters, which include multiple variables such as mean arterial pressure, arterial blood lactate, base excess, pH, bicarbonate, and anion gap.
3. The method for predicting the resuscitation status of patients with septic shock according to claim 2, characterized in that, The acquired state parameters are preprocessed, including: Data cleaning and outlier handling: physiological boundaries and rationality checks are performed on continuous variables, and the data is truncated according to quantile intervals; For missing values, multiple imputation or equivalent methods are used for continuous variables, and mode imputation is used for categorical variables. Standardization of continuous variables: The above continuous variables are standardized using the following formula: ； in, For the target parameter, the first Input features corresponding to continuous variables This is the original value of the variable. and These are the mean and standard deviation of the variable in the training set data of the prediction model, respectively.
4. The method for predicting the resuscitation status of patients with septic shock according to claim 3, characterized in that, Preprocessing the acquired state parameters also includes: Categorical variable encoding: When the state parameter includes a categorical or binary indicator variable, the variable is one-hot encoded.
5. The method for predicting the resuscitation status of patients with septic shock according to claim 1, characterized in that, Based on the critical illness database, training data was obtained. Using this training data, the prediction model was trained, including: Obtain the sample set used in this iteration. The sample set includes samples with multiple input parameters and the true label of each sample. Each sample is input into the prediction model to obtain the second prediction probability corresponding to each sample output by the prediction model. Calculate the cross-entropy loss for the sample set based on each true label and each second prediction probability; Calculate the negative gradient of the cross-entropy loss with respect to the current model output score for each sample, i.e., the second predicted probability, and use the negative gradient as the pseudo residual for each sample. The pseudo-residual fitting regression tree-based learner is used to determine the leaf node output value of the regression tree-based learner along the negative gradient direction with a preset learning rate. The trained regression tree-based learner is added to the additive scoring function to update the ensemble structure of the prediction model.
6. The method for predicting the resuscitation status of patients with septic shock according to claim 1, characterized in that, The first predicted probability is subjected to binary classification to obtain the final prediction result, including: If the first predicted probability is above the set classification threshold, then the first predicted result is output. If the first predicted probability is less than the set classification threshold, then the second predicted result is output.
7. A system for predicting the resuscitation status of patients with septic shock, characterized in that, include: The acquisition module is used to acquire the status parameters of patients with septic shock. The conversion module is used to preprocess the state parameters to obtain the input parameters; The training module is used to obtain training data based on the critical illness database, and to train the prediction model using the training data. The prediction module is used to input the input parameters into the trained prediction model and obtain the first prediction probability output by the prediction model. The output module is used to perform binary classification on the first predicted probability to obtain the final prediction result.
8. 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, it implements the steps of the method for predicting the resuscitation status of a patient with septic shock as described in any one of claims 1 to 6.
9. A non-transitory computer read storage medium, characterized in that, The non-transitory computer read storage medium stores computer instructions, which cause the computer to perform the steps of the method for predicting the resuscitation status of a septic shock patient as described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method for predicting the resuscitation status of patients with septic shock as described in any one of claims 1 to 6.