Acute kidney injury prediction device, control method for acute kidney injury prediction device, and acute kidney injury prediction program

Abstract

To provide an acute kidney injury onset predicting device or the like, which can easily predict the onset of the future acute kidney injury in advance.SOLUTION: An acute kidney injury onset predicting device 10 includes: a predicted patient kidney injury related index 205 which is a kidney injury related index of a patient to be predicted; and the generated learned model information 114 based on the kidney injury related index 205 of a past patient who enters into an intensive care unit. Based on the learned model information and the predicted patient kidney injury related index, the onset of the future acute kidney injury of the patient to be predicted is predicted.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an acute kidney injury prediction device for predicting the onset of acute kidney injury, a control method for the acute kidney injury onset prediction device, and an acute kidney injury onset prediction program. [Background technology]

[0002] The incidence of acute kidney injury (hereinafter referred to as "AKI") has traditionally been high in hospitals, particularly in intensive care units (ICUs), with 30% to 40% of patients receiving treatment in the ICU said to develop the condition.

[0003] Here, AKI (Acute Kidney Injury) refers to a condition in which, for some reason, kidney function (the ability to filter blood and remove waste products) suddenly declines, making it impossible to maintain the balance of fluids in the body and the balance of components in the blood. Acute kidney injury can be divided into three types depending on the cause: The first is "prerenal," where blood flow to the kidneys is reduced. The second is "renal," where abnormalities occur in the kidneys themselves. And the third is "postrenal," where the path through which urine produced in the kidneys flows (the urinary tract: renal pelvis, ureter, bladder, and urethra) is blocked, damaging the kidneys. Serious complications include heart failure and hyperkalemia (high potassium levels in the blood).

[0004] As mentioned above, an ICU is also known as an intensive care unit, and is a facility that provides 24-hour management for patients with severe acute respiratory, circulatory, metabolic, and other dysfunctions in order to provide more effective treatment. Depending on the facility, the term intensive care unit may refer to a general intensive care unit (GICU), a circulatory care unit (CCU) that specializes in the circulatory system, an emergency intensive care unit (SCU) that cares for patients with high urgency, an emergency room (ER), etc. Once AKI develops in an intensive care unit, both the mortality rate and medical costs rise, so early detection of the onset of AKI is an important issue. Since the RIFLE criteria were published in Critical Care in 2004, the revised AKIN and KDIGO criteria, which combine the RIFLE criteria, have been published. Both criteria are defined based on changes in serum Cr and urine volume. There are issues with serum Cr levels, such as the lag time between when an injury event occurs and when they rise, and the amount of muscle mass and water in the body can also cause errors. In particular, because it takes time for serum Cr levels to rise, once serum Cr levels have risen, a certain amount of kidney damage has already occurred.

[0005] On the other hand, in order to detect kidney damage at an early stage, it has been proposed to use a kidney-specific biomarker such as NGAL (Neutrophil Gelatinase-Associated Lipocalin) (for example, Patent Document 1, etc.). [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Patent No. 4879993 Summary of the Invention [Problem to be solved by the invention]

[0007] However, these biomarkers for kidney damage cannot detect kidney damage until 6 to 8 hours after the occurrence of a kidney damage event, or until a lag time of several hours at the earliest has elapsed. In addition, several risk models for AKI after cardiac surgery have been proposed based on statistical analysis, but these models have the problem that they are difficult to use because they have many items to check.

[0008] Therefore, the present invention aims to provide an acute kidney injury (AKI) onset prediction device, a control method for the acute kidney injury onset prediction device, and an acute kidney injury onset prediction program that can easily predict the onset of AKI in an intensive care unit in advance. [Means for solving the problem]

[0009] The object of the present invention is to provide an acute kidney injury onset prediction device for predicting the onset of acute kidney injury, comprising: a prediction target patient renal injury-related index, which is a renal injury-related index of a patient to be predicted, and a prediction target patient renal injury-related index based on the past renal injury-related index of an intensive care unit. Raw and trained model information obtained by the training, the renal impairment-related index includes information on an acute renal impairment stage, and the information on the acute renal impairment stage is information obtained based on a plurality of acute renal impairments, such as acute renal impairment determination criteria information based on the serum creatinine information and urine volume information; This can be achieved by an acute kidney injury onset prediction device, which predicts the onset of acute kidney injury in the patient to be predicted based on the trained model information and the prediction target kidney injury-related index.

[0010] According to the above configuration, the future onset of acute kidney injury in a patient to be predicted is predicted based on the generated trained model information based on the past kidney damage-related indicators (e.g., age information, gender information, urine volume information, serum creatinine, etc.) of a patient admitted to an intensive care unit, and the predicted kidney damage-related indicators (e.g., age information, gender information, urinary tract information, etc.) of the patient to be predicted. Therefore, it is possible to easily predict in advance the occurrence of acute kidney injury, which is highly likely to occur in an intensive care unit.

[0011] Preferably, The information regarding the acute kidney injury stage is generated when the information regarding the presence or absence of AKI onset is not stored. It is characterized by:

[0013] Preferably, The predicted patient renal disorder-related index includes the predicted patient renal disorder-related index not only before admission to the intensive care unit but also at the time of admission to the intensive care unit and after admission to the intensive care unit,The present invention is characterized in that it predicts acute kidney injury in an intensive care unit of the patient to be predicted.

[0014] Preferably, the prediction of the onset of acute kidney injury in the intensive care unit of the patient to be predicted based on the trained model information is performed using a machine learning technique.

[0015] According to the above configuration, the prediction is performed using a machine learning technique, thereby improving the accuracy of predicting the onset of acute kidney injury.

[0016] Preferably, the acute kidney injury onset prediction device predicts the onset of acute kidney injury in the intensive care unit for a plurality of different time points in the future.

[0017] According to the above configuration, the onset of acute kidney injury in an intensive care unit can be detected at multiple times, making it possible to select a treatment that is suited to the patient's condition.

[0018] Preferably, the prediction target patient renal disorder related index is configured to be stored in an electronic medical record information storage unit or input from an external source.

[0019] According to the above configuration, when the renal impairment-related index of the patient to be predicted is stored in the electronic medical record information storage unit, a prediction of acute kidney injury is automatically performed, whereas when the renal impairment-related index of the patient to be predicted is input from outside, a prediction is performed based on the input information. Therefore, the acute kidney injury onset prediction device can be used according to the user's purpose and situation.

[0020] Preferably, the learning model is characterized in that it is possible to reacquire and regenerate the past renal damage-related indicators of the intensive care unit.

[0021] The object of the present invention is to provide a method for controlling an acute kidney injury onset prediction device for predicting the onset of acute kidney injury, comprising: acquiring a renal injury-related index of a patient to be predicted, the renal injury-related index being a renal injury-related index of a patient to be predicted; the renal impairment-related index includes information on an acute renal impairment stage, and the information on the acute renal impairment stage is information obtained based on a plurality of acute renal impairments, such as acute renal impairment determination criteria information based on the serum creatinine information and urine volume information; and generating trained model information based on the renal impairment-related index of a past patient admitted to an intensive care unit, and predicting the future onset of acute kidney injury in the prediction target patient based on the trained model information and the prediction target renal impairment-related index. Operation This is achieved by the method.

[0022] The object, according to the present invention, is to cause an acute kidney injury onset prediction device for predicting the onset of acute kidney injury to execute a function of acquiring a prediction target patient renal injury-related index, which is a renal injury-related index of a prediction target patient, a function of generating trained model information based on the renal injury-related index of past patients who have been admitted to an intensive care unit, and a function of predicting the future onset of acute kidney injury of the prediction target patient based on the trained model information and the prediction target renal injury-related index. The renal impairment-related index includes information on an acute renal impairment stage, and the information on the acute renal impairment stage is information obtained based on a plurality of acute renal impairments, such as information on criteria for acute renal impairment based on the serum creatinine information and urine volume information. This is achieved through a program to predict the onset of acute kidney injury. Effect of the Invention

[0023] The present invention has the advantage of providing an acute kidney injury (AKI) onset prediction device, a control method for the acute kidney injury onset prediction device, and an acute kidney injury onset prediction program that can easily predict the onset of AKI in an intensive care unit in advance. [Brief description of the drawings]

[0024] [Figure 1] FIG. 1 is a schematic diagram showing an AKI (acute kidney injury) onset prediction system 1. [Diagram 2] FIG. 2 is a schematic block diagram showing the main configuration of the electronic medical record management system 200 of FIG. [Diagram 3] 3 is a schematic diagram showing the electronic medical record information storage unit 205 of FIG. 2. [Figure 4] 2 is a schematic block diagram showing the main configuration of the AKI onset prediction device 10 of FIG. 1. [Diagram 5] 2 is a schematic block diagram showing the contents of a various information storage unit 110. FIG. [Figure 6]1 is a schematic flowchart showing the process of creating a "trained model" required for predicting the onset of AKI. [Figure 7] 7 is a schematic flowchart showing the process of predicting the onset of AKI in the ICU for patient A using the "trained model" in Figure 6. [Figure 8] FIG. 1 is a schematic explanatory diagram showing an example of a criterion for determining AKI. [Figure 9] FIG. 13 is a schematic diagram showing an example of a display of "AKI development possibility information." [Figure 10] 13A is an example of the display before input, and 13B is an example of the AKI development possibility information displayed after the user has input age information and the like. [Figure 11] 13 is an evaluation graph of AUROC for the results of "machine learning" in this embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0025] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that the embodiments described below are preferred specific examples of the present invention, and therefore various technically preferable limitations are imposed thereon. However, the scope of the present invention is not limited to these embodiments unless otherwise specified in the following description to the effect that the present invention is limited thereto.

[0026] (Configuration of AKI Onset Prediction System 1) FIG. 1 is a schematic diagram showing an AKI (acute kidney injury) onset prediction system 1. As shown in FIG. 1, the system 1 includes an acute kidney injury onset prediction device, for example, an "AKI onset prediction device 10," and an "electronic medical record management system 200" that manages medical record information of patients in a hospital, etc. These are capable of communicating with each other. For example, the AKI onset prediction device 10 is configured to be able to access the electronic medical record management system 200 as necessary and acquire electronic medical record information, etc. Furthermore, the AKI onset prediction device 10 is configured so that users such as administrators and medical professionals who have been given access rights can directly input values ​​as necessary.

[0027] In addition, the AKI onset prediction device 10 and the electronic medical record management system 200 in Figure 1 have a computer, which has a CPU (Central Processing Unit), RAM (Random Access Memory), ROM (Read Only Memory), etc. (not shown), and these are connected via a bus, etc.

[0028] (Configuration of the Electronic Medical Record Management System 200) FIG. 2 is a schematic block diagram showing the main components of the electronic medical record management system 200 of FIG. As shown in Figure 2, the electronic medical record information storage unit 205, which stores patient medical records, etc., is controlled via user terminals, and each user terminal can access the information accumulated in the information accumulation unit 206 via the network 207.

[0029] Medical devices 208 and 209 that monitor the condition of patients, an information integration device 210 that consolidates information on a plurality of medical devices, and an examination device 211 store acquired information in an information accumulation unit 206 via a network 207 to construct a database 212. In this case, even for the same type of information, for example, the same blood oxygen saturation (SpO2), the devices used may differ depending on the ward, department, or facility, and the specifications for accumulating the information will differ. The information accumulation unit 206 and the database 212 may use the facility's own server or may use a cloud.

[0030] Specifically, this refers to the number of digits of information, acquisition frequency, accuracy, and the input method, such as whether the information is automatically transferred to the information accumulation unit 206 or whether it is manually entered by a medical professional who has been granted access rights at a user terminal. Not all acquired information is recorded; large amounts of information may be processed or selected before being stored.

[0031] Fig. 3 is a schematic diagram showing the electronic medical record information storage unit 205 in Fig. 2. The electronic medical record information storage unit 205 in Fig. 3 also stores information obtained by testing and measuring the renal disorder-related index of the patient, such as patient background information such as age and sex, urine volume, serum creatine, arterial blood pH, etc., which will be described later. The user terminal display 203 shown in FIG. 2 is connected to a network 207 via a management system control unit 201, a management system side communication device 202, and the like.

[0032] (Configuration of AKI onset prediction device 10) FIG. 4 is a schematic block diagram showing the main components of the AKI onset prediction device 10 of FIG. As shown in FIG. 4, the AKI onset prediction device 10 has a “prediction device control unit 11”, which controls a “prediction device communication unit 12” and a “touch panel 13” that communicate with an electronic medical record management device 200, etc. The touch panel 13 is an electronic component that combines a display, which is a display unit, with a position input device, and is an input device that allows a user to input various pieces of information by touching the display. In this embodiment, for convenience of explanation, the touch panel 13 is taken as an example for explanation, but the present invention includes not only the touch panel 13 but also other input / display units and the like. Furthermore, the input unit and the display unit may be independent and linked to each other.

[0033] The control unit 11 also controls a various information storage unit 110 as shown in FIG. 5 is a schematic block diagram showing the contents of the various information storage unit 110. The contents of these components will be described later.

[0034] 6 and 7 are schematic flowcharts showing an example of the operation of the AKI onset prediction device 10, and Fig. 6 is a schematic flowchart showing the process of creating a "trained model" necessary for AKI onset prediction. Fig. 7 is a schematic flowchart showing the process of predicting AKI onset in the ICU of patient A, for example, using the "trained model" of Fig. 6.

[0035] Before describing the process of creating the "trained model" and the process of predicting the onset of AKI in the ICU for patient A using Figures 6 and 7, we will explain the five elements (renal impairment-related indices) used in this embodiment for predicting the onset of AKI.

[0036] As described above, the five elements are 1) age information, 2) sex information, 3) urine volume information, 4) serum Cr (creatinine) information, and 5) arterial blood pH information, and will be explained in order below.

[0037] (age) "Age information" is included as a factor in prediction because the older the patient, the higher the risk of developing AKI. Specifically, it is stored in the electronic medical record information storage unit 205 in FIG.

[0038] (sex) "Sex information" is used as a factor in predicting the onset of AKI because men are more likely to develop acute kidney injury and are treated as a risk factor for AKI. Specifically, it is stored in the electronic medical record information storage unit 205 in FIG.

[0039] (Urine volume) "Urine output (UO) information" is used as a prediction factor because it is an element used in the AKI diagnosis criteria in the AKI guidelines. Many cases of acute kidney injury result in oliguria. For example, in the KDIGO (Kindney Disease Improving Global Outcome) criteria, AKI is diagnosed when urine output decreases to less than 0.5 mL / kg / hr for six hours.

[0040] Specifically, the amount of urine of the patient accumulated in a urine bag via an indwelling bladder catheter is measured over time, handwritten notes are created, and the data is stored in the electronic medical record information storage unit 205 in FIG. 2 at any desired timing. In some cases, a memo is not created and is directly stored in the electronic medical record information storage unit 205. The creation of the memo, the creation method, and the timing of storing the memo in the electronic medical record information storage unit 205 vary depending on the facility and the medical staff in charge.

[0041] (Serum Creatinine) "Serum Creatinine (sCr) information" is used as an indicator of renal function and is therefore used as an indicator for prediction. Serum creatinine is one of the waste products produced in muscles, most of which is excreted from the glomeruli of the kidneys. Therefore, an increase in serum creatinine indicates a decrease in the glomerular filtration function.

[0042] For example, as shown in Figure 8, AKI is diagnosed when a) serum creatinine levels rise by 0.3 mg / dl or more within 48 hours, or b) serum creatinine levels increase by 1.5 times or more above the known or predicted baseline value within the previous 7 days. Specifically, blood collected from a patient is tested using a dedicated testing device, and the results are stored in the electronic medical record information storage unit 205 in FIG. The testing equipment, timing of the measurement, and the time it takes from blood collection to obtaining test results vary depending on the facility and medical professional in charge.

[0043] (arterial blood pH) "Arterial blood pH (Art pH) information" is used as a factor in prediction because it is an indicator that reflects whether the kidneys are functioning normally. Blood (extracellular fluid) is regulated to a very narrow range of pH = 7.40 ± 0.05, and this pH is regulated by the kidneys and lungs. In particular, the kidneys produce bicarbonate ions (HCO3) and have the role of neutralizing acid and excreting excess acid in urine, making them extremely important organs in regulating pH in the body and blood. For this reason, when kidneys are damaged and renal function declines, acid excretion becomes difficult, which can lead to a condition called acidosis. In addition, when the effects of drugs such as diuretics or the hyperfunctioning of the adrenal glands occur, the body can overproduce bicarbonate ions, resulting in a condition called alkalosis. In either case, the body becomes unwell, but it is thought that the kidneys are no longer functioning normally.

[0044] Medications that are frequently used in ICUs, such as diuretics, anticancer drugs, and antibiotics, that affect kidney function or are toxic to the kidneys, are also used in ICUs. Therefore, arterial blood pH is added as a factor in prediction.

[0045] Specifically, the patient's arterial blood is examined, and the results are stored in the electronic medical record information storage unit 205 in FIG. The testing equipment, timing of the measurement, and the time it takes from blood collection to obtaining test results vary depending on the facility and medical professional in charge.

[0046] (AKI incidence data) In this embodiment, as shown in the electronic medical record information storage unit 205 in FIG. 3, in addition to the above-mentioned five elements, data on patients who have developed AKI in the ICU is also stored. Specifically, for a patient whom a doctor has diagnosed as having AKI, that fact is entered into the electronic medical record information storage unit 205, and the severity (stage, etc.) of the AKI, such as mild, moderate, severe, etc., is also stored.

[0047] It is expected that the form of memory will vary depending on the physician and facility. When following the RIFLE, AKIN, and KDIGO criteria, it is expected that AKI stages will be classified as 1, 2, or 3. In addition, because diagnosing AKI requires comparing and calculating values ​​over time, it is thought that in many cases, patients do not bother to determine the stage of AKI or record whether or not it has developed.

[0048] Therefore, just because a patient has developed AKI does not necessarily mean that the fact that they have AKI will be recorded in their electronic medical record. Therefore, in an embodiment of the present invention, in such a case, the stage of AKI may be determined by comparing the reference serum creatinine and urine volume data with the AKI determination criteria shown in Figure 8 in the electronic medical record information storage unit 205, and the determined stage may be stored separately. FIG. 8 is a schematic diagram illustrating an example of the criteria for determining AKI.

[0049] (Criteria for determining AKI) As described above, when a medical professional determines that a patient has developed AKI, this information is entered into the electronic medical record information storage unit 205. However, cases in which AKI is determined include, for example, any of the following: However, as described above, there are also cases in which a medical professional who has been granted access rights to AKI does not enter this information. "An increase in serum creatinine of 0.3 mg / dl or more within 48 hours," or "an increase in serum creatinine of 1.5 times or more above baseline levels known or expected within the previous 7 days," or "a decrease in urine output of 0.5 ml / kg / hour for 6 hours." Here, we will assume that the standard complies with either the RIFLE standard, the AKIN standard, or the KDIGO standard. In addition, if new standards are proposed and become common in the future, we will comply with those standards.

[0050] (Trained model creation process) Based on the above five elements, the AKI onset prediction device 10 in FIG. 1 executes AKI onset prediction, and the process of creating the "trained model" that is the premise of the prediction will be described with reference to FIG.

[0051] First, in step (hereinafter referred to as "ST") 1 of Figure 6, the "dataset generation unit (program) 111" of Figure 4 operates to obtain five elements of patients who have been admitted to the ICU in the past from the electronic medical record information storage unit 205 of Figure 2.

[0052] Specifically, for example, five elements of data in a time series such as 24 hours before admission to the ICU, at the time of admission, and 6 hours, 12 hours, 24 hours, 48 ​​hours, and 72 hours after admission are obtained, and the time data on the occurrence of AKI is also obtained and stored as "dataset" information in the "dataset storage unit 112" in Figure 5.

[0053] Next, the process proceeds to ST2. In ST2, the "trained model creation unit (program) 113" in Fig. 5 operates. In ST2, the presence or absence of AKI onset after ICU stay for each patient is calculated from the acquired information, and stored. In addition, patients who develop AKI will be compared against the RIFLE criteria, AKIN criteria, or KDIGO criteria, and AKI stage information will also be stored. Next, in ST3, a dataset is created by combining the five acquired elements, the presence or absence of AKI onset, and stage information, and the dataset is stored in the dataset storage unit 112 in FIG. Next, in ST4, the stored dataset is shaped into a state in which machine learning or deep learning can be performed, a learning model is created by performing machine learning or deep learning, and the created learning model is stored in the learned model memory unit 114 of Figure 5.

[0054] The "machine learning" in this embodiment is classified as "supervised learning" and is divided into two stages, "learning" and "recognition / prediction", and in this process, "learning" is executed. That is, in the "learning" step, the rules and patterns of the data are learned based on the data set in the "dataset storage unit 112", and "trained model information" is created.

[0055] In addition, in the above-mentioned machine learning, for example, a cross-validation method is used. Specifically, for example, the obtained data set is randomly divided into five parts, and one part is first analyzed, and the remaining part is subjected to an analysis test, and the validity of the analysis itself is repeatedly verified and confirmed. In this way, the generalization performance can be verified.

[0056] This process generates "trained model information" for predicting the onset of AKI. This model information is generated using machine learning techniques, so it is highly accurate. This process may be performed on a facility-by-facility basis, or may be used with a trained model created based on previously acquired data stored.

[0057] (AKI prediction process) Next, the process of generating information predicting the future onset of AKI for an actual patient who is the subject of prediction, for example, patient A, will be described with reference to the flowchart in FIG. First, in ST11, the "AKI prediction information acquisition unit (program) 115" in FIG. 5 is operated to acquire data on patient A's age, sex, and three elements (urine volume (UO), serum creatinine (sCr), and arterial blood pH) from the electronic medical record information storage unit 205 in FIG. 2 and store them in the "AKI prediction information storage unit 116" in FIG. 5. These data for Patient A are examples of "kidney damage-related indicators to be predicted." It should be noted that this data may be input separately instead of being obtained from the electronic medical record information storage unit 205 .

[0058] Next, the process proceeds to ST12. In ST12, the "AKI onset confirmation unit (program) 117" in Fig. 5 operates to confirm whether or not AKI has occurred using the electronic medical record information storage unit 205 in Fig. 2. This step may be omitted.

[0059] Then, in ST12, the "AKI onset probability prediction unit (program) 118" in Figure 5 is operated, and based on the learned model information in the "learned model memory unit 114" in Figure 5 and the AKI prediction data in the "AKI prediction information memory unit 116" in Figure 5, information on the probability of AKI onset for multiple different times, for example, 6 hours, 12 hours, 24 hours, 48 ​​hours, and 72 hours after admission to the ICU, is generated. Additionally, prediction information is generated for AKI stages 1, 2, and 3, and this information is stored as "AKI onset possibility information" in the "AKI onset possibility information storage unit 119" in FIG. The possibility of developing AKI may also be expressed in terms of high, medium, low, or other likelihood.

[0060] For example, 24 hours after admission to the ICU, the risk of developing AKI is predicted to be "high," and AKI at that time is predicted to be stage 2, etc.

[0061] The "AKI onset possibility prediction unit 118" in Figure 5 executes the "recognition and prediction" part of the "machine learning", but when generating this "AKI onset possibility risk information", techniques such as regression and classification may be used.

[0062] Next, the process proceeds to ST 13. In ST 13, "AKI onset possibility information" is displayed on the touch panel 13 of the AKI onset prediction device 10 in FIG. Specifically, as shown in FIG. 9, a message is displayed on the touch panel 13 indicating that the risk of "AKI" in 24 hours is high. FIG. 9 is a schematic diagram showing an example of a display of "AKI onset possibility information."

[0063] As described above, in this embodiment, the AKI development risk of patient A in the ICU is displayed on the touch panel 13, so that users such as doctors can be notified of the AKI development risk and the like in advance. Furthermore, in this embodiment, judgments are made mainly based on five factors commonly used in ICUs (age, sex, urine volume, serum Cr, arterial blood pH, etc.), making predictions easy, and since these factors are closely related to the onset of AKI, the information is highly reliable. Furthermore, AKI prediction is performed using machine learning techniques, which can further improve reliability. In addition, since it predicts the risk of onset at several different times after admission to the ICU (6 hours, 12 hours, etc.), it will also provide useful information when considering treatment strategies, making it highly valuable information for medical professionals.

[0064] (Variations of this embodiment) In this embodiment, information on the possibility of AKI onset for patient A was generated using data from the "AKI prediction information storage unit 121" from 24 hours prior to patient A's admission to the ICU. However, by further adding subsequent data such as at the time of admission to the ICU and 6 hours later, the condition close to the timing of the chronological prediction is input, thereby improving the predictability.

[0065] In addition, in the present invention, as shown in Figure 9, the AKI onset prediction device 10 may not only automatically acquire age information, etc. from the "electronic medical record information storage unit 205" and generate AKI onset possibility information, but may also input the necessary age information, etc. from outside and generate and display AKI onset possibility information based on this input information. In addition, the display terminal may be linked to the electronic medical record system 200 via the network 207 in FIG. 2, and may display the electronic medical record on the screen of the user terminal display 203 used to operate the electronic medical record.

[0066] FIG. 10 is a schematic diagram showing an example of a display that displays AKI onset possibility information input from an external source. In the example of FIG. 10, (a) is an example of the display before input, and (b) is an example in which AKI development possibility information is displayed after age information and the like have been input by the user.

[0067] As shown in Figure 10(b), when a user inputs age information, etc., the "AKI onset probability prediction unit (program) 118" operates and, based on the learned model in the "learned model memory unit 114" and the input AKI prediction information data such as age information, risk information on the possibility of AKI onset 6 hours, 12 hours, 24 hours, 48 ​​hours, and 72 hours after admission to the ICU, can be generated and displayed as low, medium, or high.

[0068] In the example of Figure 10, the risk after 24 hours is represented as "low (AKI is dashed line)," the risk after 48 hours is represented as "medium (AKI is solid line)," and the risk after 72 hours is represented as "high (AKI is bold)." This is an expression to inform the user that the risk of AKI is high, and in addition to dashed lines / solid lines / bold, any effort can be made to convey the high risk by changing the color or font size. In addition, a specific date and time may be calculated and displayed. For example, in the case of a prediction result for a patient who was admitted to the ICU at 10:25 a.m. on October 10, 2020, the result may be expressed as 6h or 6 hours later, indicating 6 hours later, or as 16:25 p.m. on October 10, 2020. The method of expressing the date and time may be any method that generally expresses the date and time, and is not limited to the example shown here.

[0069] In this way, the present invention allows the user to use the device in a variety of ways according to the user's purpose, making the device easy for the user to use.

[0070] In addition, the "dataset" in the "dataset storage unit 112" that has been created may be updated based on new data of the patient. For example, new data of the patient is stored in the electronic medical record information storage unit 205, and based on this updated electronic medical record information, steps ST1 to ST4 in Fig. 6 are performed to update the dataset and the learning model. By configuring in this way, the prediction accuracy can be further improved.

[0071] (How to verify a trained model) Regarding the results of the machine learning in this embodiment, a "confusion matrix evaluation" was performed to evaluate the sensitivity, specificity, etc., and the results were shown in the Area Under Receiver Operating Characteristic Curve (AUROC), which was close to "1" as shown in Figure 11, verifying that the method is highly reliable. FIG. 11 is an evaluation graph of the AUROC for the results of "machine learning" in this embodiment.

[0072] A learning model was created using machine learning to predict whether or not a patient would develop AKI 6, 12, 24, 48, or 72 hours after ICU admission, using 329 patients who had the necessary information as input variables, including the patient's age, sex, and urine volume within 24 hours of ICU admission, serum creatinine, and arterial blood pH. Using data from the first day of ICU admission and the patient's age and gender, it is possible to predict with high accuracy the condition 72 hours after ICU admission. Although previous cases of predicting the onset of AKI have been reported, the majority of models required a considerable number of variables, and it has been shown that the future risk of developing AKI can be determined with high accuracy using only five factors routinely obtained in the ICU. Although the results are not shown here, we confirmed that the AUROC was particularly low when five elements were missing, and we determined that these five elements were ultimately necessary.

[0073] A confusion matrix is ​​a matrix that summarizes the class classification results output from a binary classification problem, and is used as an indicator to measure the performance of a binary classification machine learning model.

[0074] In the above-described embodiment, an example has been given of the case where the invention is realized as an apparatus, but the invention is not limited to this, and the invention may be distributed as a program that can be executed by a computer, stored on a storage medium such as a magnetic disk (e.g., a floppy (registered trademark) disk, a hard disk, etc.), an optical disk (e.g., a CD-ROM, a DVD), a magneto-optical disk (MO), or a semiconductor memory.

[0075] The storage medium may be any medium that can store a program and is computer readable. The storage format of the storage medium is not particularly limited.

[0076] In addition, an OS (operating system), database management software, network software, or other MW (middleware) running on a computer based on instructions from a program installed on the computer from a storage medium may execute some of the processes required to realize this embodiment, or the cloud may be used.

[0077] Furthermore, the storage medium in the present invention is not limited to a medium independent of a computer, but also includes a storage medium that stores or temporarily stores a program downloaded via a LAN, the Internet, or the like.

[0078] Furthermore, the computer in the present invention only needs to execute each process in this embodiment based on a program stored in a storage medium, and may be a device consisting of a single personal computer (PC) or the like, or may be a system in which multiple devices are connected to a network.

[0079] In addition, the computer in this invention is not limited to a personal computer, but also includes an arithmetic processing device included in an information processing device, a microcomputer, an arithmetic processing system on a cloud, etc., and is a general term for a device, apparatus, or system that can realize the functions of the present invention by a program.

[0080] Although the embodiment of the present invention has been described above, the present invention is not limited to the above embodiment, and various modifications can be made without departing from the scope of the claims. [Explanation of symbols]

[0081] 1...AKI onset prediction system, 10...AKI onset prediction device, 11...Prediction device control unit, 12...Prediction device communication unit, 13...Touch panel, 110...Various information storage unit, 111...Data set generation unit, 112...Data set storage unit, 113...Trained model creation unit, 114...Trained model storage unit, 115...AKI prediction information acquisition unit, 116...AKI prediction information storage unit, 117...AK I Onset confirmation unit, 118... AKI onset possibility prediction unit, 119... AKI onset possibility information storage unit, 200... Electronic medical record etc. management device, 201... Management system installation side control unit, 202... Management system side communication device, 203... User terminal display, 2205... Electronic medical record information storage unit, 207... Network, 208... Medical device 1, 209... Medical device 2, 210... Information integration device, 211... Examination device

Claims

1. An acute kidney injury onset prediction device for predicting the onset of acute kidney injury, comprising: a prediction subject patient's kidney injury-related index, which is a kidney injury-related index of a prediction subject patient; and trained model information generated based on the past renal impairment-related indicators of an intensive care unit; The renal impairment-related index includes information regarding an acute renal impairment stage, The information regarding the stage of acute kidney injury is information obtained based on a plurality of acute kidney injuries, such as information regarding criteria for acute kidney injury based on the serum creatinine information and urine volume information, An acute kidney injury onset prediction device characterized by predicting the onset of acute kidney injury in the patient to be predicted based on the trained model information and the prediction target kidney injury-related index.

2. The acute kidney injury onset prediction device described in Claim 1, characterized in that the information regarding the acute kidney injury stage is generated when information regarding whether or not AKI has occurred is not stored.

3. The acute kidney injury onset prediction device described in claim 1 or claim 2, characterized in that the renal damage-related indices of the predicted patient include not only the renal damage-related indices of the predicted patient before admission to the intensive care unit but also the renal damage-related indices of the predicted patient at the time of admission to the intensive care unit and after admission to the intensive care unit, and predicts acute kidney injury in the intensive care unit for the predicted patient.

4. An acute kidney injury onset prediction device as described in any one of claims 1 to 3, characterized in that the prediction of the onset of acute kidney injury in the intensive care unit of the patient to be predicted based on the learned model information is performed using a machine learning technique.

5. 5. The acute kidney injury onset prediction device according to claim 1, wherein the prediction of the onset of acute kidney injury in the intensive care unit is performed for a plurality of different future time points.

6. The acute kidney injury onset prediction device according to any one of claims 1 to 5, characterized in that the renal damage-related index of the patient to be predicted is stored in an electronic medical record information storage unit or is input from an external source.

7. The acute kidney injury onset prediction device according to any one of claims 1 to 6, characterized in that the trained model is capable of re-acquiring and re-generating the past renal injury-related indicators of the intensive care unit.

8. A method for controlling an acute kidney injury onset prediction device for predicting the onset of acute kidney injury, comprising: Obtain a prediction subject patient renal impairment-related index, which is a renal impairment-related index of a patient to be predicted; the renal impairment-related index includes information on an acute renal impairment stage, and the information on the acute renal impairment stage is information obtained based on a plurality of acute renal impairments, such as acute renal impairment determination criteria information based on the serum creatinine information and urine volume information; Generate trained model information based on the renal impairment-related indicators of past patients who have been admitted to an intensive care unit; A method for operating an acute renal injury onset prediction device, characterized in that it predicts the future onset of acute renal injury in the patient to be predicted based on the learned model information and the prediction target renal injury-related indicators.

9. A device for predicting the onset of acute kidney injury, A function for acquiring a renal impairment-related index of a patient to be predicted, which is a renal impairment-related index of a patient to be predicted; A function of generating trained model information based on the renal impairment-related indicators of past patients who have been admitted to an intensive care unit; The purpose is to execute a function of predicting future onset of acute kidney injury in the prediction target patient based on the trained model information and the prediction target renal injury related index, The renal impairment-related indicator includes information regarding the stage of acute renal impairment, and the information regarding the stage of acute renal impairment is information obtained based on a plurality of acute renal impairments, such as acute renal impairment determination criteria information based on the serum creatinine information and urine volume information.

Images