KR20200114975A - Method for predicting of bacteremia risk and device for predicting of bacteremia risk using the same - Google Patents

Abstract

The present specification relates to a method for predicting a risk of occurrence of bacteremia implemented by a processor and a device for predicting a risk of occurrence of bacteremia using the same. The method comprises the steps of: receiving biological test data for a subject; and predicting a risk of occurrence of bacteremia in the subject using a bacteremia prediction model configured to predict the risk of the occurrence of bacteremia in the subject based on the biological test data. The biological test data include at least one of a creatine level, an albumin level, a creactive protein (CRP) level, a white blood cell (WBC) minimum level, a WBC maximum level, platelet count and prothrombin time, and an alkaline phosphatase (ALP) level.

Description

A method for predicting the risk of bacteremia and a device for predicting the risk of bacteremia using the same {METHOD FOR PREDICTING OF BACTEREMIA RISK AND DEVICE FOR PREDICTING OF BACTEREMIA RISK USING THE SAME}

The present invention relates to a method and device for predicting the risk of developing bacteremia, and more particularly, to a method configured to predict and provide the risk of developing bacteremia based on various clinical data acquired from an individual, and a device using the same.

Bacteremia can refer to a clinical condition in which bacteria enter the blood and circulate throughout the body. These bacteremia can occur when bacteria multiply at a rate that exceeds the ability of the reticulum endothelial system to clear.

In this case, bacteremia can be classified into temporary bacteremia, intermittent bacteremia, and persistent bacteremia according to clinical aspects. More specifically, transient bacteremia is bacteremia occurring at the beginning of systemic or local infection, and may be accompanied by meningitis, pneumonia, purulent arthritis, osteomyelitis, gonococcal or meningococcal infection. In addition, intermittent bacteremia can occur when there is an abscess in the abdominal cavity, pelvic cavity, peri-renal, liver, prostate, etc., and persistent bacteremia can occur when there is bacterial endocarditis and infection of intravascular catheter. It may accompany cellosis.

On the other hand, as a treatment method for bacteremia, antibiotics may be used, and the empirical use of appropriate antibiotics at an early stage can prevent the progression from bacteremia to sepsis and septic shock.

Therefore, predicting bacteremia before the onset of bacteremia can be of great clinical importance.

Accordingly, there is a constant demand for the development of a new bacteremia risk prediction system that can predict and take quick action before bacteremia occurs.

The technology that is the background of the present invention has been prepared to facilitate understanding of the present invention. It should not be understood as an admission that the matters described in the technology behind the invention exist as prior art.

On the other hand, the inventors of the present invention have noted that changes in biological signals will precede the physiological response reaction of the human body before a situation in which a disease such as bacteremia is clinically suspected.

More specifically, the inventors of the present invention have noted that changes in various clinical data that can be obtained in a general hospital room or intensive care unit may have an association with potential diseases such as bacteremia.

In particular, the inventors of the present invention have noted that bacteremia is a disease that can be secondaryly caused by an external environment, and not only biological test data of the analysis results of biological samples such as blood isolated from patients, but also various biological signal data However, it was possible to recognize that data related to medical treatment were associated with the onset of bacteremia.

Further, the inventors of the present invention attempted to explore a biological feature that increases the accuracy of prediction of bacteremia, which is distinguished from sepsis, which causes intoxication symptoms such as fever and hypotension by proliferating bacteria in the blood.

As a result, the inventors of the present invention were able to find that among the biological test data obtained from individuals, platelet counts, especially ALP (Alkaline phosphatase) levels, are associated with the onset of bacteremia.

Accordingly, the inventors of the present invention can predict the risk of bacteremia with high reliability and accuracy by further considering biological test data including ALP (Alkaline phosphatase) level and/or platelet level, as well as biosignal data and clinical characteristic data. It has come to develop a system for predicting the risk of developing bacteremia.

On the other hand, the inventors of the present invention were able to apply a predictive model learned to predict bacteremia based on biological test data, further biosignal data, and clinical feature data to a new bacteremia risk prediction system.

In relation to this, the inventors of the present invention performed an evaluation to determine clinical data having a large influence on predicting bacteremia among various clinical data or clinical data having a large influence on predicting normal, not bacteremia, and this I tried to apply it to learning. Through this, the inventors of the present invention could expect to improve the diagnostic ability of predicting bacteremia of the predictive model.

As a result, a new bacteremia risk prediction system based on a bacteremia prediction model could provide diagnostic information for bacteremia with high accuracy and reliability for an individual.

Furthermore, the inventors of the present invention expect that through the development of such a bacteremia risk prediction system, the patient's condition can be quickly detected and the risk of bacteremia incidence is recognized in advance, so that the treatment time for the patient can be accelerated to improve the treatment outcome of bacteremia. Could Furthermore, the inventors of the present invention could further expect that the development of a system for predicting the risk of developing bacteremia can provide an effect of increasing patient survival, preventing complications, and reducing treatment costs.

In addition, the inventors of the present invention, so that the guardian or medical staff can more easily recognize the high-risk group of bacteremia for an individual requiring continuous monitoring, such as a critically ill patient, when bacteremia is predicted by a predictive model, an alarm It was configured to inform the risk of developing bacteremia.

Accordingly, the problem to be solved by the present invention is to use a bacteremia prediction model configured to predict the risk of bacteremia on the basis of the received biological test data, further biosignal data, and clinical feature data, for the individual. To provide a method for predicting the risk of developing bacteremia, configured to predict the risk of developing bacteremia.

Another problem to be solved by the present invention is a receiving unit configured to receive biological test data for an individual, further biosignal data, and clinical feature data, and a risk of developing bacteremia in an individual based on a predictive model connected to communicate with the receiving unit. It is to provide a device for predicting the risk of developing bacteremia, including a processor configured to predict.

The problems of the present invention are not limited to the problems mentioned above, and other problems that are not mentioned will be clearly understood by those skilled in the art from the following description.

In order to solve the above problems, a method of predicting the risk of developing bacteremia according to an embodiment of the present invention is provided. At this time, the method for predicting bacteremia of the present invention is configured to predict the risk of developing bacteremia implemented by the processor, receiving biological test data for the individual, and predicting the risk of developing bacteremia for the individual based on the biological test data. And predicting the risk of developing bacteremia in the individual using the constructed bacteremia prediction model.

At this time, the biological test data includes creatine level, albumin level, CRP (Creactive protein) level, WBC (white blood cell) minimum level, WBC maximum level, platelet count, prothrombin time, and Includes Alkaline phosphatase (ALP) levels.

According to a feature of the present invention, the method for predicting bacteremia of the present invention may further include receiving bio-signal data for an individual, and the bacteremia-prediction model calculates the risk of developing bacteremia based on biological test data and bio-signal data. It can be further configured to predict. At this time, the biosignal data may be at least one of systolic blood pressure (SBP), diastolic blood pressure (DBP), maximum body temperature, minimum body temperature, heart rate, and respiratory rate.

According to another feature of the present invention, the method for predicting bacteremia of the present invention further comprises receiving clinical characteristic data for an individual, and the bacteremia predicting model is based on biological test data and clinical characteristic data, and the risk of developing bacteremia May be further configured to predict. At this time, the clinical characteristic data may be at least one of sex, age, blood culture period, whether or not to be treated with an intensive care unit (ICU), whether to insert a central venous catheter, whether to be treated with steroids, or whether to be treated with antibiotics.

According to another feature of the present invention, the biological test data may include creatine levels, albumin levels, CRP levels, WBC minimum levels, WBC peak levels, prothrombin time, and platelet count, ALP levels.

According to another feature of the present invention, biological test data include ALT (alanine aminotransferase) level, thromboplastin coagulation time, AST (aspartate aminotransferase) level, CRP (C-reactive protein) level, erythrocyte sedimentation rate, FERR (ferritin) level, HMG (hemoglobin) level, PTINR (Prothrombin Time International Normalized Ratio), PTPER (Prothrombin Time (%)), PTSEC (Prothrombin Time (sec)), red blood cell count, TBIL (total bilirubin) level, TCO ₂ (total carbon dioxide) and white blood cell count

According to another feature of the present invention, the bacteremia may be acute severe bacteraemia.

According to another feature of the present invention, the bacteremia prediction model includes an ensemble model composed of a combination of a multi-layer perceptron (MLP), a support vector machine (SVM), a random forest (RF), and a plurality of models, pre-trained ( It may be at least one of a pre-trained) model and an Xgboost model.

According to another feature of the present invention, the model for predicting bacteremia may be or an Xgboost model. In this case, the MLP model may include a single hidden layer having 128 nodes or 256 nodes, or may include a plurality of hidden layers in which a plurality of single hidden layers exist.

According to another feature of the present invention, the bacteremia prediction model includes receiving data for learning consisting of biological experimental data, biological signal data, and clinical feature data for bacteremia-onset sample individuals and normal sample individuals, and training data. As a basis, it may be a model learned through the step of predicting bacteremia or normality. In this case, the normal sample individual may be an individual clinically evaluated as not having bacteremia and not bacteremia.

According to another feature of the present invention, the method for predicting bacteremia of the present invention may further include evaluating the learning data performed after the step of receiving the learning data. In this case, the step of evaluating the learning data includes calculating a relevance score for bacteremia with respect to the learning data, and determining the learning data related to bacteremia within a predetermined ranking based on the relevance score. Can include.

According to another feature of the present invention, the step of evaluating the learning data includes: one-out search (OOS), Layer-wise Relevance Propagation (LRP), gradient input (GI), integrated gradients (IG), and saliency (SM). map) based on at least one algorithm, calculating a relevance score for bacteremia with respect to the learning data, and determining bacteremia-related learning data within a predetermined ranking based on the relevance score. I can.

According to another feature of the present invention, the method for predicting bacteremia of the present invention may further include providing a predicted risk of developing bacteremia for an individual.

According to another feature of the present invention, the step of providing the risk of developing bacteremia may include providing a risk notification when the risk of developing bacteremia for an individual is predicted by the bacteremia prediction model.

According to another feature of the present invention, providing a risk of developing bacteremia may include calculating a risk probability of developing bacteremia for an individual by a bacteremia prediction model, and providing a risk probability of developing bacteremia.

In order to solve the above-described problem, a device for predicting the risk of developing bacteremia according to another embodiment of the present invention is provided. At this time, the device for predicting bacteremia of the present invention includes a receiving unit configured to receive biological test data for an individual, and a processor connected to the receiving unit. At this time, the processor is configured to predict the risk of developing bacteremia of the individual using a bacteremia prediction model configured to predict the risk of developing bacteremia for the individual based on the biological test data. Further, biological test data include creatine levels, albumin levels, CRP (Creactive protein) levels, WBC (white blood cell) minimum levels, WBC peak levels, platelet counts, prothrombin time. And ALP (Alkaline phosphatase) levels.

According to a feature of the present invention, the receiving unit is further configured to receive biosignal data for an individual, and the bacteremia layer prediction model may be further configured to predict a risk of developing bacteremia based on biological test data and biosignal data. . At this time, the biosignal data may be at least one of systolic blood pressure (SBP), diastolic blood pressure (DBP), maximum body temperature, minimum body temperature, heart rate, and respiratory rate.

According to another feature of the present invention, the receiving unit is further configured to receive clinical feature data for the individual, and the bacteremia prediction model may be further configured to predict the risk of developing bacteremia based on biological test data and clinical feature data. have. At this time, the clinical characteristic data may be at least one of sex, age, blood culture period, whether or not to be treated with an intensive care unit (ICU), whether to insert a central venous catheter, whether to be treated with steroids, or whether to be treated with antibiotics.

According to another feature of the present invention, the biological test data may include creatine levels, albumin levels, CRP levels, WBC minimum levels, WBC peak levels, prothrombin time, and platelet count, ALP levels.

According to another feature of the present invention, the biological test data includes ALT level, thromboplastin coagulation time, AST level, CRP level, erythrocyte sedimentation rate, ferritin level, hemoglobin level, PTINR, PTPER, PTSEC, red blood cells. It may further include at least one of a level, a total bilirubin level, a TCO ₂ level, and a white blood cell level.

According to another feature of the present invention, the bacteremia may be acute severe bacteraemia.

According to another feature of the present invention, the bacteremia prediction model includes an ensemble model composed of a combination of a multi-layer perceptron (MLP), a support vector machine (SVM), a random forest (RF), and a plurality of models, pre-trained ( It may be at least one of a pre-trained) model and an Xgboost model.

According to another feature of the present invention, the bacteremia prediction model may be an MLP model or an Xgboost model. In this case, the MLP model may include a single hidden layer having 128 nodes or 256 nodes, or may include a plurality of hidden layers in which a plurality of single hidden layers exist.

According to another feature of the present invention, the bacteremia prediction model receives data for learning consisting of biological experimental data, biological signal data, and clinical feature data for bacteremia-onset sample individuals and normal sample individuals, and based on the training data. , It may be a model learned through the step of predicting bacteremia or normality. In this case, the normal sample individual may be an individual clinically evaluated as not having bacteremia and not bacteremia.

According to another feature of the present invention, it may further include an output unit configured to provide a predicted risk of developing bacteremia for the individual.

According to another feature of the present invention, the output unit may be configured to provide a risk notification when the risk of developing bacteremia for an individual is predicted by the bacteremia prediction model.

According to another feature of the present invention, the bacteremia prediction model may be further configured to calculate a probability of developing bacteremia for an individual. Furthermore, the output unit may be further configured to provide a probability of developing bacteremia.

Hereinafter, the present invention will be described in more detail through examples. However, these examples are for illustrative purposes only, and the scope of the present invention should not be construed as being limited by these examples.

The present invention receives pharmacologic test data, further biosignal data, and clinical characteristic data for an individual, and provides a system for predicting the risk of bacteremia based on the received data, thereby rapidly detecting the risk of developing bacteremia in the individual, and You can provide relevant information.

In addition, the present invention provides a prediction system to which a predictive model learned to predict bacteremia based on biological test data, further biosignal data, and clinical feature data is applied, thereby providing information on predicting the onset of bacteremia with high accuracy and reliability. I can.

More specifically, the present invention performed an evaluation to determine clinical data having a large influence on predicting bacteremia among various clinical data or clinical data having a large influence on predicting a normal state other than bacteremia, and this By applying it to learning, it is possible to provide a prediction system with improved diagnostic ability for bacteremia prediction.

Accordingly, the present invention has the effect of predicting the risk of onset with high reliability and accuracy for bacteremia that is distinct from sepsis or different from sepsis.

In addition, the present invention is configured to notify the risk of developing bacteremia by providing an alarm when bacteremia is predicted by a predictive model for an individual, so that a guardian or a medical staff is easier for individuals requiring continuous monitoring, such as critically ill patients. It is possible to recognize high-risk groups for developing bacteremia.

Accordingly, the present invention can provide a good prognosis for treatment by advancing the time point of treatment for bacteremia.

Furthermore, the present invention can provide an effect of increasing the survival rate of an individual with bacteremia, preventing complications, and reducing treatment costs.

The effects according to the present invention are not limited by the contents exemplified above, and more various effects are included in the present specification.

1A is an exemplary diagram illustrating a system for predicting the risk of developing bacteremia based on a method and a device for predicting the risk of developing bacteremia according to an embodiment of the present invention.
1B is an exemplary configuration of a device for predicting the risk of developing bacteremia according to an embodiment of the present invention.
2 is an exemplary diagram illustrating a procedure of a method for predicting the risk of developing bacteremia according to an embodiment of the present invention.
3A and 3B exemplarily show training data of a bacteremia prediction model used in various embodiments of the present invention.
3C is an exemplary configuration diagram of a model for predicting bacteremia used in various embodiments of the present invention.
3D is an exemplary illustration of a method of evaluating a model for predicting bacteremia used in various embodiments of the present invention.
4A and 4B illustrate a first evaluation result of a model for predicting bacteremia used in various embodiments of the present invention.
4C and 4D illustrate clinical data according to the influence of bacteremia prediction for a bacteremia prediction model used in various embodiments of the present invention.
4E and 4F illustrate performance changes according to changes in clinical data according to the influence of bacteremia prediction for a bacteremia prediction model used in various embodiments of the present invention.
5A and 5B illustrate training data and evaluation data for a second evaluation of a bacteremia prediction model used in various embodiments of the present invention.
5C to 5I illustrate a second evaluation result for a bacteremia prediction model used in various embodiments of the present invention.

Advantages and features of the present invention, and a method of achieving them will become apparent with reference to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but will be implemented in a variety of different forms, only these embodiments make the disclosure of the present invention complete, and common knowledge in the technical field to which the present invention pertains. It is provided to completely inform the scope of the invention to those who have, and the invention is only defined by the scope of the claims.

The shapes, sizes, ratios, angles, numbers, etc. disclosed in the drawings for explaining the embodiments of the present invention are exemplary, and the present invention is not limited to the illustrated matters. In addition, in describing the present invention, when it is determined that a detailed description of related known technologies may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted. When'include','have','consists of' and the like mentioned in the present specification are used, other parts may be added unless'only' is used. In the case of expressing the constituent elements in the singular, it includes the case of including the plural unless specifically stated otherwise.

In interpreting the components, it is interpreted as including an error range even if there is no explicit description.

Each of the features of the various embodiments of the present invention can be partially or entirely combined or combined with each other, and as a person skilled in the art can fully understand, technically various interlocking and driving are possible, and each of the embodiments may be independently implemented with respect to each other. It may be possible to do it together in a related relationship.

For clarity of interpretation of the present specification, terms used in the present specification will be defined below.

As used herein, the term "bacteremia" refers to a clinical condition in which pathogens live and circulate in the bloodstream. At this time, bacteremia is different from'sepsis', which causes poisoning symptoms such as fever and hypotension by proliferating bacteria in the blood, and sepsis and bacteremia are classified in the present specification.

As used herein, the term "individual" may mean a target for predicting the risk of developing bacteremia. Meanwhile, in the present specification, an individual may be a'patient' or a'critical patient', but is not limited thereto and may include all subjects for which the risk of developing bacteremia is predicted.

At this time, the bacteremia-onset individuals include Bacillus spp. , Viridans group streptococci, CNS (coagulase-negative staphylococci), and other organisms in the bloodstream. Can be different.

As used herein, the term "biological test data" may mean clinical data on a biological sample, for example blood, isolated from an individual.

More specifically, the biological test data disclosed in the present specification includes creatine (Creatinine) levels, albumin (Albumin) levels, CRP (Creactive protein) levels, WBC (white blood cell) minimum levels, WBC peak levels, and platelet counts. , Prothrombin time and Alkaline phosphatase (ALP) levels.

Preferably, the biological test data may be an ALP level or a platelet level, but is not limited thereto.

As used herein, the term "biosignal data" refers to vital signs or vital signs, and may mean data related to the state of an individual. Such biosignal data may be associated with an individual's risk of developing bacteremia.

In this case, the biosignal data may be body temperature, pulse, oxygen saturation, systolic blood pressure (SBP), diastolic blood pressure (DBP), average blood pressure, and respiratory rate of the individual measured by the biosignal measuring equipment. Preferably, the biosignal data may be at least one of SBP, maximum body temperature, minimum body temperature, heart rate, and respiratory rate. However, the present invention is not limited thereto, and the biosignal data may include various measurement data related to a health state of an individual.

As used herein, the term "clinical characteristic data" may include data on whether or not to administer drugs, medical treatment, along with characteristics of an individual, for example, sex and age.

On the other hand, the clinical characteristic data disclosed in the present specification is at least one of sex, age, blood culture period, intensive care unit (ICU) treatment, central venous catheter insertion, steroid treatment, and antibiotic treatment. Can be However, the clinical characteristic data is not limited thereto, and may include a variety of data on drugs administered to an individual and medical treatment for the individual.

For example, the clinical characteristic data is data on the medical treatment of at least one of Catheter, Foley, Mechanical ventilator, Restraint, and Drainage. , Furthermore, Ultraset, Midazolam, Ultiva, Popol, Ativan, Fentanyl, Precedex, IR codon, Percussion 20/10 (TARGIN 20/10), peridol, lisperdal, zyprexa, seroquel, abilify, petidine, durogesic patch, It may further include administration data of morphine and mypol.

As used herein, the term "bacteremia prediction model" may be a model trained to predict the risk of developing bacteremia based on clinical data of at least one of biological test data, biological signal data, and clinical characteristic data.

For example, the bacteremia prediction model may be a model trained to predict the risk of developing bacteremia based on clinical data obtained at each predetermined time from a sample individual with or without bacteremia.

In this case, the bacteremia prediction model may be a model trained to predict the risk of developing bacteremia based on all clinical data of biological test data, biological signal data, and clinical feature data. However, the data used for learning is not limited thereto, and a combination of more diverse clinical data may be used for learning the predictive model.

On the other hand, the bacteremia prediction model of the present invention is an ensemble model composed of a combination of MLP (Multi-Layer Perceptron), SVM (support vector machine), RF (random forest), and MLP, SVM, RF, and pre-trained (pre- trained) model and the Xgboost model may be based on at least one model.

Preferably, the bacteremia prediction model of the present invention may be a prediction model based on one MLP (Multi-Layer Perceptron). In this case, the bacteremia prediction model of the present invention based on the MLP model may include a hidden layer having 128 nodes or 256 nodes.

For example, the bacteremia prediction model of the present invention includes an input layer into which biological test data, biosignal data, and clinical feature data are input, and an output layer that predicts normal, not bacteremia or bacteremia, and 128 nodes or 256 nodes between these layers. It may be a prediction model of a multi-layer structure in which a hidden layer having nodes exists.

More specifically, the bacteremia prediction model of the present invention may be configured to use'rmsprop' as an optimization function for updating a parameter in learning, and as a function that determines the intensity at which input values of various clinical data are transferred to an output value. It can be configured to use the'relu' function.

However, the configuration of the bacteremia prediction model of the present invention is not limited thereto.

According to another embodiment of the present invention, the bacteremia prediction model of the present invention may be an MLP model having a single node of 128 nodes. Furthermore, the bacteremia prediction model of the present invention may be an MLP model having two layers of hidden layers having a hidden layer having 128 nodes.

According to another embodiment of the present invention, the bacteremia prediction model of the present invention may be an Xgbooxt model having a gbtree layer.

However, the types of the bacteremia prediction model of the present invention are not limited to those described above. For example, the prediction model is a more diverse ANN (Artificial Neural Network) model, or DNN (Deep Neural Network), CNN Convolutional Neural Network), RNN (Recurrent Neural Network), DCNN (Deep Convolutional Neural Network), RBM ( Restricted Boltzmann Machine), DBN (Deep Belief Network), SSD (Single Shot Detector) model, or U-net-based prediction model.

Meanwhile, in the learning of the bacteremia prediction model of the present invention, an evaluation in which clinical data having a large influence on predicting normal, not bacteremia, may be determined. In this case, the influence evaluation is performed based on at least one algorithm among OOS (one-out search), LRP (Layer-wise Relevance Propagation), GI (gradient input), IG (integrated gradients), and SM (saliency map). I can.

For example, clinical data having a large influence on predicting bacteremia among various clinical data of biological test data, biosignal data, and clinical feature data by the at least one algorithm or in predicting normal, not bacteremia Large clinical data can be determined.

As clinical data having a high influence can be applied as input data to the learning of the predictive model, the predictive model may be a model with improved bacteremia predictive ability than other models.

Hereinafter, a device for predicting the risk of bacteremia and a system for predicting the risk of bacteremia using the same according to an embodiment of the present invention will be described in detail with reference to FIGS. 1A and 1B.

1A is an exemplary diagram illustrating a system for predicting the risk of developing bacteremia based on a method and a device for predicting the risk of developing bacteremia according to an embodiment of the present invention. 1B is an exemplary configuration of a device for predicting the risk of developing bacteremia according to an embodiment of the present invention.

Referring to FIG. 1A, the system 1000 for predicting the risk of developing bacteremia according to an embodiment of the present invention includes a device 100 for predicting the risk of developing bacteremia, a biological test data 310 obtained for an individual 200, It is composed of bio-signal data 320, clinical data 300 including clinical feature data 330, medical treatment instrument 400, and medical staff device 500.

In this case, the clinical data 300 may be data evaluated from the individual 200 at an arbitrary time point. However, it is not limited thereto.

More specifically, the device 100 for predicting the risk of bacteremia in the bacteremia risk prediction system 1000 receives various clinical data 300 measured or evaluated for the individual 200, and the risk of developing bacteremia based on this Can be configured to predict

At this time, the medical treatment device 400 may provide at least one biosignal data 320 from the group consisting of temperature, pulse, oxygen saturation, systolic blood pressure, diastolic blood pressure, average blood pressure, and respiratory rate to the individual 200. It may be a bio-signal measuring device.

Meanwhile, the biological test data 310 and further clinical feature data 330 in the bacteremia risk prediction system 1000 may be obtained from an external system such as an Electronic Medical Record (EMR) system.

More specifically, referring to FIG. 1B, the device 100 for predicting the risk of bacteremia incidence includes a receiving unit 110, an input unit 120, an output unit 130, a storage unit 140, and a processor 150.

In more detail, the receiving unit 110 may be configured to receive clinical data 300 including biological test data 310, biological signal data 320, and clinical feature data 330 for the individual 200.

According to a feature of the present invention, the reception unit 110 may be further configured to transmit a predicted result for the entity 200 determined by the processor 150 to be described later to the medical staff device 500.

The input unit 120 is not limited, such as a keyboard, a mouse, and a touch screen panel. The input unit 120 may set the device 100 for predicting the risk of developing bacteremia and instruct the operation of the device 100 for predicting the risk of developing bacteremia.

Meanwhile, the output unit 130 may display various clinical data 300 received by the receiving unit 110. Furthermore, the output unit 130 may display information related to the diagnosis of bacteremia predicted by the processor 150 on a display. Further, the output unit 130 may be further configured to output a notification sound when it is determined by the processor 150 that the risk of developing bacteremia is high.

The storage unit 140 stores various clinical data 300 for the individual 200 received through the receiving unit 110, and stores an instruction of the device 100 for predicting the risk of bacteremia incidence set through the input unit 120 Can be configured to Further, the storage unit 140 is configured to store bacteremia prediction information in the individual 200 generated by the processor 150 to be described later. However, it is not limited to the above, and the storage unit 140 may store various pieces of information determined by the processor 150 to predict the risk of bacteremia.

The processor 150 may be a component for providing an accurate prediction result of the device 110 for predicting the risk of developing bacteremia. In this case, in order to accurately predict the risk of developing bacteremia, the processor 150 may be based on a prediction model configured to predict the risk of developing bacteremia based on various clinical data 300.

Meanwhile, the processor 150 may be configured to provide a notification to the medical staff device 500 when the individual 200 is predicted to be a high risk group for developing bacteremia by the bacteremia prediction model. Thus, the medical staff can take quick action on the subject 200.

As described above, the system 1000 for predicting the risk of bacteremia incidence of the present invention can be applied to various medical systems as it can provide early diagnosis of bacteremia and early treatment of bacteremia for the individual 200.

Hereinafter, a method of predicting the risk of developing bacteremia according to an embodiment of the present invention will be described in detail with reference to FIG. 2. 2 is a diagram illustrating a procedure of a method for predicting the risk of developing bacteremia according to an embodiment of the present invention.

2, the method for predicting the risk of developing bacteremia according to an embodiment of the present invention includes first receiving biological test data from an individual (S210), and predicting the risk of developing bacteremia based on the biological test data. Using the model, the risk of developing bacteremia is predicted for the individual (S220). Finally, the predicted risk of developing bacteremia is provided for the individual (S230).

More specifically, in the step of receiving data (S210), creatine (Creatinine) level, albumin (Albumin) level, CRP (Creactive protein) level, WBC (white blood cell) minimum level, WBC maximum level, platelet (Platelet) Biological test data of at least one of water, prothrombin time, and Alkaline phosphatase (ALP) levels can be received.

According to a feature of the present invention, in the step of receiving data (S210), at least one biosignal data from among systolic blood pressure (SBP), maximum body temperature, minimum body temperature, heart rate, and respiration rate may be further received.

According to another feature of the present invention, in the step of receiving data (S210), sex, age, blood culture period, whether or not to be treated with an intensive care unit (ICU), whether to insert a central venous catheter, whether to be treated with steroids, and At least one clinical characteristic data of whether or not antibiotic treatment may be further received.

According to a feature of the present invention, the biosignal data received in the step of receiving data (S210) may be an ALP value and a platelet count, but is not limited thereto.

Next, in the step of predicting the risk of developing bacteremia (S220), the individual may be determined to be normal, not bacteremia or bacteremia, by a bacteremia prediction model configured to predict the risk of developing bacteremia.

At this time, the bacteremia prediction model is trained to predict the risk of bacteremia based on at least one clinical data of biological test data, biosignal data, and clinical feature data received in the step S210 of receiving the data. It can be a model.

For example, the bacteremia prediction model receives data for learning at least one of biological test data, biosignal data, and clinical evaluation data for bacteremia-onset sample individuals and normal sample individuals, and based on the training data, bacteremia or It may be a model learned through the step of predicting or classifying the normal.

Furthermore, the bacteremia prediction model is for learning related to bacteremia, which is determined through the step of calculating a relevance score for bacteremia with respect to the learning data, and determining learning data related to bacteremia within a predetermined ranking based on the relevance score. Based on the data, it may be a model configured to predict the risk of developing bacteremia.

According to another feature of the present invention, the bacteremia prediction model of the present invention comprises a multi-layer perceptron (MLP), a support vector machine (SVM), a random forest (RF), and a combination of the MLP, the SVM, and the RF. It may be at least one of the ensemble models.

According to another feature of the present invention, the bacteremia prediction model may be an MLP model composed of a hidden layer having 128 nodes or 256 nodes, but is not limited thereto.

Finally, in the step of providing the risk of developing bacteremia (S230), information regarding the diagnosis of bacteremia predicted in the step of predicting the risk of developing bacteremia (S220) may be provided.

According to another feature of the present invention, in the step of providing the risk of developing bacteremia (S230), when the risk of developing bacteremia is predicted by the bacteremia prediction model in the step (S220) of predicting the risk of developing bacteremia, A notification of the risk of developing bacteremia may be provided.

On the other hand, in the step (S230) of providing the risk of developing bacteremia, more various information may be provided in addition to the above. For example, in the step of providing the risk of developing bacteremia (S230), additional treatment information that may be effective in correcting bacteremia may be further provided.

According to the above procedure, the method for predicting the risk of developing bacteremia according to an embodiment of the present invention is to determine whether to develop bacteremia in real time or predict the risk of developing bacteremia in advance, information on the risk of developing bacteremia, and further bacteremia. It can provide an alarm for high-risk groups. Accordingly, the medical staff may be able to diagnose bacteremia early in the individual. Furthermore, medical staff can take quick action for high-risk groups for bacteremia.

Hereinafter, a model for predicting bacteremia used in various embodiments of the present invention will be described in detail with reference to FIGS. 3A to 3D.

3A and 3B exemplarily show training data of a bacteremia prediction model used in various embodiments of the present invention.

Referring to FIG. 3A, 22332 blood culture episodes obtained from 13402 patients obtained from Gangnam Severance Hospital were collected for learning of the bacteremia prediction model used in various embodiments of the present invention, and 1260 of them Episodes for bacteremia individuals and episodes for 1260 normal individuals were set as learning data. Accordingly, the bacteremia prediction model of the present invention was trained to predict bacteremia based on the positive result and negative result data of a 1:1 ratio.

Referring to FIG. 3B, data for learning obtained from sample individuals of bacteremia-onset individuals and normal individuals are composed of biosignal data, biological test data, and clinical feature data. The values of each of these data appear to differ at a statistically significant level for individuals with bacteremia and for normal individuals.

More specifically, the learning biological signal data may be composed of SBP, maximum body temperature, minimum body temperature, heart rate, and respiratory rate. The study biological test data may consist of creatine levels, albumin levels, CRP levels, minimum WBC levels, maximum WBC levels, platelet counts, prothrombin time, and ALP levels. Furthermore, the clinical characteristic data may consist of blood culture period, ICU treatment, central venous catheter, steroid treatment, antibiotic treatment, age and sex.

At this time, among the clinical characteristic data, whether ICU treatment and whether central venous catheter is treated may be data evaluated at the time of blood culture, and whether antibiotic treatment may be data evaluated 3 weeks before blood culture, and whether or not steroid treatment is blood It may be data evaluated 2 weeks before incubation. In this case, the biosignal data and biological test data may be data evaluated 72 hours before blood culture.

The bacteremia prediction model of the present invention may be trained to determine whether or not bacteremia occurs in a sample individual based on the learning data as described above. FIG. 3C illustrates the configuration of a bacteremia prediction model used in various embodiments of the present invention. It is shown as an enemy.

3C, the bacteremia prediction model of the present invention may be a prediction model based on the MLP algorithm, which is a multilayer artificial neural network.

More specifically, the bacteremia prediction model of the present invention is an input layer into which biological test data such as ALP level or platelet count, biosignal data such as maximum body temperature, clinical characteristic data, and normal, not bacteremia or bacteremia, are predicted. It may be a multi-layered prediction model in which an output layer and one hidden layer exist between these layers.

At this time, the bacteremia prediction model of the present invention may be configured to include one hidden layer composed of 128 or 256 nodes. Furthermore, in the prediction model, a learning rate value, which may be a parameter for finding a weight that minimizes an error in prediction, may be set to 0.1 in predictive learning of bacteremia. In addition, a momentum value, which is a parameter value for minimizing prediction errors and increasing the learning speed, may be set to 0.95. In addition, the bacteremia prediction model of the present invention may be configured to use'adadelta' as an optimization function for updating a parameter in learning, and as a function that determines the intensity at which input values of various clinical data are transferred to the output value, ' It can be configured to use the tanh' function.

However, the type of the bacteremia prediction model of the present invention is not limited thereto. For example, the prediction model may be at least one of an ensemble model composed of a support vector machine (SVM), a random forest (RF), and a combination of MLP, SVM, and RF, and DNN, CNN, RNN, DCNN , RBM, DBN, SSD model, or may be a prediction model based on U-net.

Meanwhile, in the learning of the bacteremia prediction model of the present invention, an evaluation in which clinical data having a large influence on predicting normal, not bacteremia, may be determined. In this case, the influence evaluation is performed based on at least one algorithm among OOS (one-out search), LRP (Layer-wise Relevance Propagation), GI (gradient input), IG (integrated gradients), and SM (saliency map). I can.

3D is an exemplary illustration of a method of evaluating a model for predicting bacteremia used in various embodiments of the present invention.

In this case, the evaluation of clinical data based on the LRP algorithm will be described as an example, but the evaluation algorithm may be performed based on more various algorithms as described above.

Referring to FIG. 3D, when a bacteremia prediction result is output by a bacteremia prediction model based on various clinical data, an evaluation based on an LRP algorithm may be performed.

More specifically, in this evaluation, by the LRP algorithm, ALP level, biological test data such as platelet count, biosignal data such as peak body temperature, and further input values of various clinical data of clinical characteristic data and normal non-bacteremia or bacteremia The relevance of the output value to can be estimated.

More specifically, the relationship between the input value of various clinical data and the output value to normal, not bacteremia or bacteremia may be calculated by the following [Equation 1].

[Equation 1]

). 나아가, b _j 는 j번째 노드의 편향값을 의미할 수 있다.Here, L denotes an output layer value, and Z _ji may denote a product of the weight of the l + 1 th layer and the l th layer, and an input value of the l th layer (

). Further, b _j May mean a bias value of the j-th node.

According to the calculated relevance score, among various clinical data, clinical data having a high relevance to predicting bacteremia different from sepsis, or clinical data having a large influence on predicting normal, not bacteremia, can be determined.

As clinical data having a large influence can be applied as input data to the learning of the predictive model, the predictive model of the present invention may be superior to other models in predicting bacteremia in predicting bacteremia.

Example 1: First evaluation of bacteremia prediction model according to various embodiments of the present invention

Hereinafter, a first evaluation result of the bacteremia prediction model used in various embodiments of the present invention will be described in detail with reference to 4a to 4f. 4A and 4B illustrate a first evaluation result of a model for predicting bacteremia used in various embodiments of the present invention.

First, referring to FIG. 4A, an evaluation result of a bacteremia prediction model configured to predict normal, not bacteremia or bacteremia, based on biological test data, biological signal data, and data is shown.

ve Bayesian이 본 평가에 함께 이용되었다.At this time, the bacteremia prediction model of the present invention is an MLP model consisting of a hidden layer having 128 nodes (the bacteremia prediction model (A) of the present invention), an MLP model consisting of a hidden layer having 256 nodes (predicting bacteremia of the present invention) Model (B)), classification model of random forest (bacteremia prediction model (C) of the present invention), classification model of SVM (bacteremia prediction model (D) of the present invention), and MLP model consisting of a hidden layer having 128 nodes , An MLP model composed of a hidden layer having 256 nodes and an ensemble model in which a random forest classification model is combined (the bacteremia prediction model (E) of the present invention). Furthermore, each variable is stratified and divided into sections, and the conventional predictive model Na is configured to analyze the risk for bacteremia.

ve Bayesian was used together in this evaluation.

Referring to FIG. 4A, the sensitivity of the bacteremia prediction model (A) of the present invention and the bacteremia prediction model (B) of the present invention based on MLP is 0.81, which is the highest compared to the other models. Furthermore, the specificity of the bacteremia prediction model (C) of the present invention is 0.655, which is the highest compared to the other models.

ve Bayesian에 비하여 높은 것으로 나타난다. 이때, AUC 값은 진단 능력과 연관이 있을 수 있다. 따라서, 본 발명의 다양한 실시예에 따른 균혈증 예측 모델은, 균혈증 진단 시스템에 적용될 경우, 개체에 대하여 균혈증의 발병 위험도를 높은 정확도로 미리 예측할 수 있다.In particular, referring to FIG. 4B together, the AUC value that can mean the diagnostic ability of the bacteremia prediction models (A) to (E) of the present invention is an average of about 0.729, which is a conventional prediction model Na

ve appears to be higher than Bayesian. At this time, the AUC value may be related to diagnostic ability. Accordingly, when the bacteremia prediction model according to various embodiments of the present invention is applied to a bacteremia diagnosis system, the risk of developing bacteremia in an individual can be predicted in advance with high accuracy.

According to the above results, it appears that the bacteremia prediction model according to various embodiments of the present invention predicts the risk of bacteremia with higher reliability than the conventional prediction model. In particular, the MLP-based bacteremia prediction model appears to have a higher AUC value while maintaining higher sensitivity than other models. Accordingly, the bacteremia prediction model of the present invention may be an MLP-based prediction model, but is not limited thereto.

4C and 4D illustrate clinical data according to the influence of bacteremia prediction for a bacteremia prediction model used in various embodiments of the present invention.

Referring to FIG. 4C, results of evaluating the influence (relevance) of clinical data (biological test data, biosignal data, and clinical feature data) used in various embodiments of the present invention are shown. In this evaluation, the ranking of variables with high influence on the prediction of bacteremia based on'One-out search' and'Gini Importance' was evaluated. More specifically, according to the ranking evaluation results of One-out search that can be applied to the MLP-based predictive model, the ALP level and platelet count among the biological test data, and the highest body temperature among the biological signal data are in predicting the risk of bacteremia. It appears to have high influence. On the other hand, according to the ranking evaluation result of Gini Importance, which can be used in the random forest-based prediction model, it is slightly different from the ranking evaluation result of One-out search, but ALP level, platelet, evaluated to have high influence by one-out search. Number and peak body temperature appear to have a high impact on classifying bacteremia.

4D, a variable having a high influence on the prediction of bacteremia based on a gradient-based method based on Layer-wise Relevance Propagation (LRP), gradient input (GI), integrated gradients (IG), and saliency map (SM) The results of evaluating their ranking are shown. More specifically, according to the ranking evaluation results of LRP, GI and IG, similar to the ranking evaluation results of the above-described one-out search, ALP levels appear to have a high influence on predicting the risk of invention of bacteremia. In addition, platelet counts appear to have a high influence on predicting the risk of developing bacteremia. Also in the results of ranking evaluation according to SM, it appears that platelet count and ALP level have a high influence on predicting the risk of invention of bacteremia.

These results may mean that in predicting bacteremia based on clinical data for an individual, the ALP level and platelet count among the biological test data, and further, the highest body temperature among the biological signal data, provides a great influence in the prediction of bacteremia. have. That is, the bacteremia prediction model of the present invention can be learned to predict whether or not bacteremia occurs based on the ALP level and platelet count, and further, the highest body temperature, so that the risk of bacteremia in an individual can be predicted with excellent diagnostic ability.

4E and 4F illustrate performance changes according to changes in clinical data according to the influence of bacteremia prediction for a bacteremia prediction model used in various embodiments of the present invention.

Referring to (a) of FIG. 4E, after applying Occlusion of LRP, SM, OOS (one-out search), and disturbance-based method for evaluating variables for the ANN machine learning algorithm of MLP, it is evaluated that the influence is high. The AUC values when the variables are accumulated and excluded in the order of influence are shown. Furthermore, for the evaluation of the variable based on the RF-based prediction model, the AUC value obtained when variables evaluated to have high influence are accumulated and excluded in the order of influence by the Gini Importance method.

More specifically, it appears that the AUC value for predicting bacteremia decreases as the upper variables evaluated to be highly influential in predicting bacteremia (e.g., ALP level, platelet count, peak body temperature, etc.) are excluded. . This may mean a decrease in the performance of predicting bacteremia due to the exclusion of highly influential clinical data.

Referring to (b) of FIG. 4E, after applying Occlusion of LRP, SM, OOS (one-out search), and disturbance-based method for evaluating variables for the ANN machine learning algorithm of MLP, it is evaluated that the influence is high. The AUC values obtained by accumulating and adding the variables in the order of influence are shown. Furthermore, the AUC value obtained by accumulating and adding variables evaluated to have high influence in the order of influence by the Gini Importance method for evaluating the variable for the RF-based prediction model is also shown.

More specifically, it appears that the AUC value for predicting bacteremia increases as the top variables evaluated to be highly influential in predicting bacteremia (e.g., ALP level, platelet count, peak body temperature, etc.) are added. . At this time, after adding the top 10 variables, the change in AUC value appears to be insignificant. These results may mean that the top 10 clinical data are highly influential in predicting bacteremia. On the other hand, in the case of OOS, a significant increase in AUC value appears due to the addition of variables than other evaluation models. Accordingly, in the method for predicting the risk of bacteremia according to various embodiments of the present invention, the OOS model may be applied to the evaluation of various clinical variables, that is, biological test data, biosignal data, and clinical feature data, but is limited thereto. It is not.

Referring to (a) of FIG. 4F, after applying the Gini Importance method or OOS for the evaluation of variables for RF and SVM non-ANN machine learning algorithms, variables evaluated to have high influence are accumulated and excluded in the order of influence. The AUC value when done is shown.

More specifically, it appears that the AUC value for predicting bacteremia decreases as the upper variables evaluated to be highly influential in predicting bacteremia (e.g., ALP level, platelet count, peak body temperature, etc.) are excluded. . This may mean a decrease in the performance of predicting bacteremia due to the exclusion of highly influential clinical data.

Referring to (b) of FIG. 4F, after applying the Gini Importance method or OOS for evaluating variables for RF and SVM non-ANN machine learning algorithms, variables evaluated to have high influence are accumulated and added in order of influence. The AUC value when done is shown.

More specifically, it appears that the AUC value for predicting bacteremia increases as the top variables evaluated to be highly influential in predicting bacteremia (e.g., ALP level, platelet count, peak body temperature, etc.) are added. . At this time, after adding the top 10 variables, the change in AUC value appears to be insignificant. These results may mean that the top 10 clinical data are highly influential in predicting bacteremia. On the other hand, in the case of OOS, a significant increase in AUC value appears due to the addition of variables than other evaluation models. Accordingly, in the method for predicting the risk of bacteremia according to various embodiments of the present invention, the OOS model may be applied to the evaluation of various clinical variables, that is, biological test data, biological signal data, and clinical characteristic data, It is not.

Meanwhile, the bacteremia prediction model used in various embodiments of the present invention may be configured to use clinical data having a large influence in predicting bacteremia as learning data. Accordingly, the bacteremia prediction model may provide a bacteremia prediction result with improved diagnostic ability than other prediction models in predicting bacteremia. As a result of Example 1 above, it was confirmed that the bacteremia prediction model used in various examples of the present invention predicts the risk of developing bacteremia with high accuracy. Furthermore, according to the results shown to have a high AUC value, it was confirmed that the predictive model of the present invention has excellent diagnostic ability.

Accordingly, the present invention can provide a good prognosis for treatment by advancing the time point of treatment for bacteremia. Furthermore, the present invention has the effect of predicting the risk of onset with high reliability and accuracy for bacteremia different from sepsis.

In particular, the present invention provides a system for predicting the risk of developing bacteremia based on clinical data that can be quickly obtained from an individual, thereby providing early diagnosis of the onset of bacteremia in an individual.

Accordingly, the present invention can provide an effect of increasing the survival rate of an individual with bacteremia, preventing complications, and reducing treatment costs.

In addition, the present invention is configured to notify the risk of developing bacteremia by providing an alarm when bacteremia is predicted by a predictive model for an individual, so that a guardian or a medical staff is easier for individuals requiring continuous monitoring, such as critically ill patients. It is possible to recognize the high risk group for developing bacteremia.

Example 2: Second evaluation of bacteremia prediction model according to various embodiments of the present invention

Hereinafter, a second evaluation result of the bacteremia prediction model used in various embodiments of the present invention will be described in detail with reference to 5a to 5i. 5A and 5B illustrate training data and evaluation data for a second evaluation of a bacteremia prediction model used in various embodiments of the present invention. 5C to 5I illustrate a second evaluation result for a bacteremia prediction model used in various embodiments of the present invention.

At this time, referring to FIG. 5A, in order to predict bacteremia in this evaluation, age and sex were set as characteristic parameters as clinical characteristic data, and systolic blood pressure (SBP), diastolic blood pressure (DBP), and highest as biosignal data. Body temperature, minimum body temperature, heart rate and respiratory rate were set as characteristic parameters. Furthermore, as biological test data, albumin level, ALT (alanine aminotransferase) level, thromboplastin coagulation time, AST (aspartate aminotransferase) level, CRP (C-reactive protein) level, red blood cell sedimentation rate, FERR (ferritin) Level, HMG (hemoglobin) level, PTINR (Prothrombin Time International Normalized Ratio), PTPER (Prothrombin Time (%)), PTSEC (Prothrombin Time (sec)), red blood cell count, TBIL (Total bilirubin) level, TCO ₂ (total carbon) dioxide) and white blood cells were set as characteristic parameters.

In this case, the biosignal data and biological test data may include a minimum value, a maximum value, or a measured value measured within 3 days.

More specifically, referring to FIG. 5B, the configuration of data sets A, B and C used for the second evaluation of the bacteremia prediction model used in various embodiments of the present invention is shown.

The A data set is constructed such that bacteremia positive and bacteremia negative data have a ratio of 1:1 for training and validation, and bacteremia positive and bacteremia negative data have a ratio of about 1:10 for testing. At this time, data set A is bacteremia positive and bacteremia negative data sampled between 2007 and 2018 at Gangnam Severance Hospital and Shinchon Severance Hospital, and data from 2007 to 2015 for training and verification, 2016 for testing. Data from year to 2018 can be used.

The B data set is configured such that bacteremia positive and bacteremia negative data have a ratio of 1:1 for training and validation, and bacteremia positive and bacteremia negative data have a ratio of about 1:7 for testing. In this case, the C data set may be randomly divided data after merging bacteremia positive and bacteremia negative data sampled between 2007 and 2018 at Gangnam Severance Hospital and Shinchon Severance Hospital by year.

The C data set is constructed such that bacteremia positive and bacteremia negative data have a ratio of 1:1 for training and validation, and bacteremia positive and bacteremia negative data have a ratio of about 1:7 for testing. At this time, the C data set is randomly divided after merging the bacteremia positive and bacteremia negative data sampled between 2007 and 2018 at Gangnam Severance Hospital and Shinchon Severance Hospital by year, and then randomly divide the same patient sample for each set. Separated, if there are positive/negative samples of the same patient at the same time, the negative samples may be removed.

Referring to FIG. 5C, an evaluation result of a model for predicting bacteremia learned based on data A is shown. At this time, for a bacteremia prediction model consisting of a plurality of layers of 4 layers consisting of 128 nodes, 64 nodes, 25 nodes, and 128 nodes, evaluation of the bacteremia predictive diagnosis performance according to prior learning was performed. Meanwhile, in the pre-learned bacteremia prediction model, weight initializing was applied to an AutoEncoder that learns label values by itself.

More specifically, in the case of the pre-learned bacteremia prediction model (pret) of the present invention, the accuracy of predicting positive or negative bacteremia is 0.759, the sensitivity is 0.762, and the accuracy (0.729) and sensitivity (0.75) of the model without appear. In particular, the AUC value that can mean the diagnostic ability of the pre-learned bacteremia predictive model (pret) of the present invention is 0.840, which is higher than the AUC value of 0.818 of the model that does not.

According to the above results, the bacteremia prediction model according to various embodiments of the present invention may be a model to which weight initializing is applied to an Otter encoder having a high AUC value while maintaining high sensitivity. However, it is not limited thereto.

Referring to FIGS. 5D and 5E, evaluation results according to layer sizes of a model for predicting bacteremia learned based on data B and C described above are shown.

First, referring to FIG. 5D, a prediction model consisting of a plurality of layers (64-128-64) of three layers consisting of 64 nodes, 128 nodes and 64 nodes, respectively, 64 nodes, 128 nodes, 128 nodes, and A performance evaluation result using the above-described B data is shown for a prediction model composed of a plurality of layers (64-128-128-64) of 4 layers each composed of 64 nodes.

At this time, for each of the 64-128-64 prediction model and the 64-128-128-64 prediction model, do (drop-out) or bn (batch normalization) of 0.5 probability was applied.

More specifically, the accuracy of bacteremia prediction for each of the 64-128-64 prediction model and the 64-128-128-64 prediction model to which bn is applied is 0.735 and 0.738, which appears to have similar values for changes in the layer size. In addition, the AUC values for bacteremia prediction for the 64-128-64 prediction model and the 64-128-128-64 prediction model to which bn was applied were 0.847 and 0.859, indicating that they have similar values in the change of the layer size.

Meanwhile, referring to FIG. 5E further, a prediction model composed of a single layer 128 composed of 128 nodes, a plurality of layers of three layers composed of 128 nodes of 64 nodes and 64 nodes respectively (64-128-64) Performance using the aforementioned C data for a prediction model consisting of 64 nodes, 128 nodes, 128 nodes, and multiple layers of 4 layers (64-128-128-64) each consisting of 64 nodes. The evaluation results are shown.

In this case, do of 0.5 probability was applied to the 128 prediction models, and do and/or bn of 0.5 probability were applied to each of the 64-128-64 prediction models and 64-128-128-64 prediction models.

More specifically, in the case of a single-layer 128 prediction model, the AUC value is 0.751, which appears to be higher than that of the 64-128-64 prediction model and the 64-128-128-64 prediction model having multiple layers.

At this time, in the case of the 64-128-64 prediction model and the 64-128-128-64 prediction model, it appears that the accuracy, sensitivity and AUC values of bacteremia prediction when bn are applied are high similar to the result of FIG. 5D described above. .

According to the above results, it is shown that the bacteremia prediction model according to various embodiments of the present invention has high sensitivity, accuracy, and AUC values even when the layer size is changed. Thus, the bacteremia prediction model according to various embodiments of the present invention has a single layer composed of 128 nodes, a plurality of layers of three layers of 64-128-64, or a plurality of layers of 64-128-128-64. It may be a model having a plurality of four layers and to which bn is applied. However, the structure of the bacteremia prediction model of the present invention is not limited thereto.

Referring to FIGS. 5F and 5G, evaluation results of a stacking ensemble model learned based on the above-described B data and C data are shown.

First, referring to FIG. 5F, an ensemble of 10, 15, and 20 stacking models has a plurality of layers of 3 layers (64-128-64) consisting of 64 nodes, 128 nodes and 64 nodes, respectively. For the model, the performance evaluation results using the above-described B data are shown.

More specifically, when the B data set is applied, the AUC value for the ensemble model with 15 stacking models is 0.887, the AUC value of the ensemble model with 10 stacking models (0.865), and the ensemble with 20 stacking models It appears to be higher than the model's AUC value (0.873).

Meanwhile, referring to FIG. 5G, a performance evaluation result using the aforementioned C data for an ensemble model having a single layer composed of 128 nodes and having 10, 15 and 20 stacking models is shown.

More specifically, when the C data set is applied, the AUC value for the ensemble model with 15 and 20 stacking models is 0.752, which is higher than the AUC value (0.750) of the ensemble model with 10 stacking models.

According to the above results, the bacteremia prediction model according to various embodiments of the present disclosure may be an ensemble model in which the number of stacking models is 15 to 20. However, the structure of the bacteremia prediction model of the present invention is not limited thereto.

5H and 5I, evaluation results of an MLP-based prediction model and an Xgboost-based prediction model learned based on the above-described B data and C data are shown.

First, referring to FIG. 5H, it has a plurality of layers of 3 layers (64-128-64) consisting of 64 nodes, 128 nodes and 64 nodes, respectively, and consists of an MLP model with 15 stacking models and a gbtree layer. For the Xgboost model, a performance evaluation result using the above-described B data is shown.

More specifically, when the B data set is applied, the AUC value of the predictive model based on the Xgboost model is 0.842 and the sensitivity is 0.745, which is similar to the AUC value (0.887) and the sensitivity (0.801) of the MLP-based predictive model. Appears to be.

Meanwhile, referring to FIG. 5I, a performance evaluation result using the aforementioned C data for an Xgboost model having a single layer composed of 128 nodes and composed of a 20-person MLP model of a stacking model and a gbtree layer is shown.

More specifically, when the C data set is applied, the AUC value of the predictive model based on the Xgboost model is 0.745 and the sensitivity is 0.635, and the AUC value (0.752) and the sensitivity (0.664) of the MLP-based predictive model are similar. Appears to be.

According to the above results, the bacteremia prediction model according to various embodiments of the present invention may be an MLP-based prediction model or an Xgboost-based prediction model. However, the model for predicting bacteremia of the present invention is not limited thereto, and may be a model configured to predict positive or negative bacteremia based on more various predictive models.

As a result of Example 2 above, it was confirmed that the bacteremia prediction model used in various examples of the present invention predicts the risk of developing bacteremia with high accuracy. Furthermore, according to the results shown to have a high AUC value, it was confirmed that the predictive model of the present invention has excellent diagnostic ability.

Accordingly, the present invention can provide a good prognosis for treatment by advancing the time point of treatment for bacteremia. Furthermore, the present invention has the effect of predicting the risk of onset with high reliability and accuracy for bacteremia different from sepsis.

In particular, the present invention provides a system for predicting the risk of developing bacteremia based on clinical data that can be quickly obtained from an individual, thereby providing early diagnosis of the onset of bacteremia in an individual.

Accordingly, the present invention can provide an effect of increasing the survival rate of an individual with bacteremia, preventing complications, and reducing treatment costs.

In addition, the present invention is configured to notify the risk of developing bacteremia by providing an alarm when bacteremia is predicted by a predictive model for an individual, so that a guardian or a medical staff is easier for individuals requiring continuous monitoring, such as critically ill patients. It is possible to recognize the high risk group for developing bacteremia.

Each of the features of the various embodiments of the present invention can be partially or totally combined or combined with each other, and as a person skilled in the art can fully understand, technically various interlocking and driving are possible, and each embodiment may be independently implemented with respect to each other. And it may be possible to do it together in a relationship

The embodiments of the present invention have been described in more detail with reference to the accompanying drawings, but the present invention is not necessarily limited to these embodiments, and various modifications may be made without departing from the spirit of the present invention. . Accordingly, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention, but to explain the technical idea, and the scope of the technical idea of the present invention is not limited by these embodiments. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not limiting. The scope of protection of the present invention should be interpreted by the following claims, and all technical ideas within the scope equivalent thereto should be interpreted as being included in the scope of the present invention.

100: device for predicting the risk of bacteremia
110: receiver
120: input
130: output
140: storage unit
150: processor
200: object
300: clinical data
310: biological test data
320: biosignal data
330: clinical characteristic data
400: medical treatment device
500: medical staff device
1000: risk prediction system

Claims (26)

In the method for predicting the risk of developing bacteremia implemented by a processor,
Receiving biological test data for the subject;
Predicting the risk of developing bacteremia of the individual using a bacteremia prediction model configured to predict the risk of developing bacteremia in the individual based on the biological test data,
The biological test data,
At least one of Creatinine, Albumin, CRP (Creactive Protein), WBC (white blood cell) minimum, WBC peak, platelet count, Prothrombin time, and ALP (Alkaline) phosphatase), a method for predicting the risk of developing bacteremia. The method of claim 1,
Receiving bio-signal data for the individual, further comprising,
The bacterial blood layer prediction model,
Further configured to predict a risk of developing bacteremia based on the biological test data and the biological signal data,
Singgi biosignal data,
A method for predicting the risk of developing bacteremia, which is at least one of systolic blood pressure (SBP), diastolic blood pressure (DBP), maximum body temperature, minimum body temperature, heart rate, and respiratory rate. The method of claim 1,
Receiving clinical feature data for the individual, further comprising,
The bacterial blood layer prediction model,
Further configured to predict a risk of developing bacteremia based on the biological test data and the clinical characteristic data,
Singgi clinical characteristic data,
A method for predicting the risk of bacteremia, which is at least one of sex, age, blood culture period, intensive care unit (ICU) treatment, central venous catheter insertion, steroid treatment, and antibiotic treatment. The method of claim 1,
The biological test data,
The creatine level, the albumin level, the CRP level, the WBC minimum level, the WBC maximum level, at least one of the prothrombin time, and the platelet count, including the ALP level, a method for predicting the risk of bacteremia. The method of claim 1,
The biological test data,
ALT (alanine aminotransferase) level, thromboplastin coagulation time, AST (aspartate aminotransferase) level, CRP (C-reactive protein) level, erythrocyte sedimentation rate, ferritin (FERR) level, hemoglobin (HMG) level, PTINR (Prothrombin Time International Normalized Ratio), PTPER (Prothrombin time (%)), PTSEC (Prothrombin time (sec)), red blood cell count, TBIL (total bilirubin) level, TCO ₂ (total carbon dioxide) level, and white blood cell count A method for predicting the risk of developing bacteremia further comprising a. The method of claim 1,
The bacteremia is acute severe bacteraemia, a method for predicting the risk of developing bacteremia. The method of claim 1,
The bacteremia prediction model,
MLP (multi-layer perceptron), SVM (support vector machine), RF (random forest), and an ensemble model consisting of a combination of a plurality of models, a pre-trained model, at least one of the Xgboost model , How to predict the risk of developing bacteremia. The method of claim 7,
The bacteremia prediction model,
The MLP model or the Xgboost model,
The MLP model includes a single hidden layer having 128 nodes or 256 nodes,
A method for predicting a risk of developing bacteremia, comprising a plurality of hidden layers in which a plurality of the single hidden layers are present. The method of claim 1,
The bacteremia prediction model,
Receiving data for learning consisting of biological experimental data, biosignal data, and clinical feature data for bacteremia-onset sample individuals and normal sample individuals, and
Based on the learning data, it is a model learned through the step of predicting bacteremia or normal,
The normal sample individual,
A method of predicting the risk of developing bacteremia, which is an individual that has not been clinically evaluated as bacteremia and not bacteremia. The method of claim 9,
Performed after the step of receiving the learning data,
Further comprising the step of evaluating the learning data,
Evaluating the learning data,
Calculating a relevance score for bacteremia with respect to the learning data, and
A method for predicting a risk of developing bacteremia, comprising the step of determining learning data related to bacteremia within a predetermined ranking based on the relevance score. The method of claim 10,
Evaluating the learning data,
Based on at least one algorithm among OOS (one-out search), LRP (Layer-wise Relevance Propagation), GI (gradient input), IG (integrated gradients), and SM (saliency map),
Calculating a score for a degree of association with bacteremia with respect to the learning data, and
A method for predicting a risk of developing bacteremia, comprising the step of determining learning data related to bacteremia within a predetermined ranking based on the relevance score. The method of claim 1,
The method of predicting the risk of developing bacteremia further comprising the step of providing the predicted risk of developing bacteremia for the individual. The method of claim 12,
The step of providing the risk of developing bacteremia,
When the risk of developing bacteremia in an individual is predicted by the bacteremia prediction model,
A method of predicting a risk of developing bacteremia, comprising the step of providing a risk notification. The method of claim 12,
The step of providing the risk of developing bacteremia,
Calculating a risk of developing bacteremia for an individual by the bacteremia prediction model, and
A method for predicting the risk of developing bacteremia, comprising the step of providing a probability of developing the risk of bacteremia. A receiving unit configured to receive biological test data for an individual, and
Including a processor connected to the receiving unit,
The processor is configured to predict the risk of developing bacteremia of the subject using a bacteremia prediction model configured to predict the risk of developing bacteremia for the subject based on the biological test data,
The biological test data,
At least one of Creatinine, Albumin, CRP (Creactive Protein), WBC (white blood cell) minimum, WBC peak, platelet count, Prothrombin time, and ALP (Alkaline) A device for predicting the risk of developing bacteremia, including phosphatase) levels. The method of claim 15,
The receiving unit is further configured to receive biometric signal data for the object,
The bacterial blood layer prediction model,
Further configured to predict a risk of developing bacteremia based on the biological test data and the biological signal data,
Singgi biosignal data,
SBP (systolic blood pressure), DBP (diastolic blood pressure), at least one of the highest body temperature, the lowest body temperature, heart rate and respiratory rate, a device for predicting the risk of bacteremia. The method of claim 15,
The receiving unit is further configured to receive clinical feature data for the individual,
The bacterial blood layer prediction model,
Further configured to predict a risk of developing bacteremia based on the biological test data and the clinical characteristic data,
Singgi clinical characteristic data,
A device for predicting the risk of bacteremia, which is at least one of sex, age, blood culture period, intensive care unit (ICU) treatment, central venous catheter insertion, steroid treatment, and antibiotic treatment. The method of claim 15,
The biological test data,
The device for predicting the risk of bacteremia, including at least one of the creatine level, the albumin level, the CRP level, the WBC minimum level, the WBC maximum level, and the prothrombin time, and the platelet count and the ALP level. The method of claim 15,
The biological test data,
ALT (alanine aminotransferase) level, thromboplastin coagulation time, AST (aspartate aminotransferase) level, CRP (C-reactive protein) level, erythrocyte sedimentation rate, ferritin (FERR) level, hemoglobin (HMG) level, PTINR (Prothrombin Time International Normalized Ratio), PTPER (Prothrombin time (%)), PTSEC (Prothrombin time (sec)), red blood cell count, TBIL (total bilirubin) level, TCO ₂ (total carbon dioxide) level, and white blood cell count A device for predicting the risk of developing bacteremia further comprising a. The method of claim 15,
The bacteremia is acute severe bacteraemia, a device for predicting the risk of developing bacteremia. The method of claim 15,
The bacteremia prediction model,
MLP (multi-layer perceptron), SVM (support vector machine), RF (random forest), and an ensemble model consisting of a combination of a plurality of models, a pre-trained model, at least one of the Xgboost model , A device for predicting the risk of bacteremia. The method of claim 21,
The bacteremia prediction model,
The MLP model or the Xgboost model,
The MLP model includes a single hidden layer having 128 nodes or 256 nodes,
A device for predicting the risk of bacteremia, including a plurality of hidden layers in which a plurality of the single hidden layers are present. The method of claim 15,
The bacteremia prediction model,
A model trained by receiving data for learning consisting of biological experimental data, biosignal data, and clinical feature data for bacteremia-onset sample individuals and normal sample individuals, and predicting bacteremia or normality based on the training data ego,
The normal sample individual,
A device for predicting the risk of developing bacteremia, which is an individual that has not been clinically evaluated as bacteremia and not bacteremia. The method of claim 15,
A device for predicting the risk of developing bacteremia further comprising an output unit configured to provide the predicted risk of developing bacteremia for the individual. The method of claim 24,
The output unit,
When the risk of developing bacteremia in an individual is predicted by the bacteremia prediction model,
A device for predicting the risk of developing bacteremia, configured to provide a risk notification. The method of claim 24,
The bacteremia prediction model,
Further configured to calculate a probability of developing bacteremia for the subject,
The output unit,
A device for predicting the risk of developing bacteremia, further configured to provide a risk probability of developing bacteremia.

Images