Prognosis prediction system, prognosis prediction model generation device and prognosis prediction device

Abstract

To support selection of a treatment method suitable for individual patient.SOLUTION: A prognosis prediction system comprises: a prognosis prediction model generation device comprising, a learning basic data acquisition part for acquiring learning basic data including, information related to states of diseases of patients including an object patient, a feature amount extraction part for extracting a feature amount, on the basis of the learning basic data, and a model generation part for generating by machine learning, a prognosis prediction model for predicting a future state of the object patient; and a prognosis prediction device for predicting a future state of the object patient, using the generated prognosis prediction model, the prognosis prediction device comprising, a basic data acquisition part for acquiring basic data including, information related to a previous state and / or current state of the disease of the object patient, a feature amount calculation part for calculating a feature amount on the basis of the basic data, an input condition reception part for receiving designation of a prediction item, a prognosis prediction part for predicting a future state on the basis of the calculated feature amount, and a display part for displaying a prediction result indicating, the future state of the object patient.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a prognosis prediction system for predicting the prognosis of a patient, a prognosis prediction model generating device, and a prognosis prediction device. [Background technology]

[0002] The number of patients with cardiovascular disease continues to increase due to the aging of the population. For example, the number of patients with heart failure in Japan has already exceeded one million.

[0003] Cardiovascular disease is treated on an outpatient basis unless it is in an acute or severe state. (1)Outpatient treatment In conventional outpatient care, regular examinations are performed even during follow-up, but these are mainly simple tests that are minimally invasive and have little impact on the patient, such as blood pressure measurement, blood tests, and electrocardiograms. Since there are only a limited number of facilities that can perform detailed examinations and there is a limit to the amount of tests that can be accepted, detailed examinations are mainly recommended when there is a change in subjective symptoms. In addition, patients tend to become complacent, especially if there is no change in their subjective symptoms, which can lead to a decrease in the frequency of outpatient visits and compliance with oral medication. Given this background, there is a problem with current outpatient care in that it is difficult to detect asymptomatic cases of cardiac dysfunction early on. When this condition is discovered after the patient has reached a critical condition, not only does it cause a sudden and major blow to the patient's quality of life, but it also places a heavy burden on the medical system, including emergency medical care.

[0004] Another problem with outpatient drug therapy is that it is difficult to feel the effects of treatment, and it is extremely difficult to predict the effects and adjust the prescription. It is not uncommon for chronic cardiovascular disease to be treated over a period of years, and the main goal of treatment is to prevent the condition from worsening rather than dramatically improving it. As a result, the effects of the medication are difficult to see, and there are high risks involved in frequently changing the prescription during outpatient treatment every few months, so the prescription tends to lean toward the majority opinion. In reality, it is difficult to find prescriptions that are tailored to each individual patient.

[0005] Furthermore, there is a problem with the conventional method in that there is not enough way to predict the prognosis. Therefore, in most cases, active intervention from the medical side is only possible after the deterioration of the disease is detected. For elderly patients who do not have sufficient reserve physical strength, even a momentary delay can be fatal.

[0006] (2) Inpatient care, especially intensive care units There are no clear standards for admission to the intensive care unit (CCU) for patients with cardiovascular disease, but patients are eligible for multidisciplinary treatment in the CCU when their hemodynamics have deteriorated significantly and they are in a life-threatening condition, or when there is an extremely high possibility that they will fall into that condition. In such a situation, treatment is provided while keeping track of the patient's condition, which changes from moment to moment, based on a wide variety of medical data. CCU patients often have multiple underlying diseases, so treatment selection is always difficult and requires the utmost care. Doctors on the scene must continue to choose from a variety of options at the right time under tense circumstances.

[0007] The problem with the conventional system is that the decision on the treatment policy depends 100% on the ability and experience of the doctor on the scene. When multiple seriously ill patients are present at the same time, the time that can be spent on each patient is limited. Although there is always a risk that the demand for treatment will exceed the qualitative or quantitative supply, the only solution is to rely on the efforts of the personnel on the scene, and this operational situation continues worldwide. In addition, conventional treatment methods have the problem that it is difficult to fully reflect the individual differences of patients in the treatment. Conventionally, the choice of treatment has been determined based on a large amount of evidence and treatment experience. However, these clinical rules of thumb have tended to be adopted by prioritizing the results of the majority. It is a matter of theory, and does not necessarily fit perfectly to each individual patient. Although individual differences in age and sex are taken into consideration, it cannot be said that differences in race or genetic polymorphisms are fully reflected.

[0008] Furthermore, in order to support the selection of such treatment methods, various techniques using machine learning have been proposed (see Patent Documents 1 to 3). [Prior art documents] [Patent documents]

[0009] [Patent Document 1] JP 2022-11833 A [Patent Document 2] JP 2021-166037 A [Patent Document 3] JP 2020-13555 A Summary of the Invention [Problem to be solved by the invention]

[0010] However, this has not been sufficient from the viewpoint of selecting a treatment method suited to each individual patient. The present invention has been made in consideration of the above problems, and has an object to provide a technique for supporting the selection of a treatment method suitable for an individual patient. [Means for solving the problem]

[0011] To solve the above problems, the present invention provides: A learning basic data acquisition unit that acquires learning basic data including information on the disease state of patients, including a target patient who is a specific individual; A feature extraction unit that extracts features based on the learning base data; a model construction unit that constructs a prognosis prediction model by machine learning to predict a future state of the patient using the feature amount; A prognosis prediction model generating device comprising: A prognosis prediction device that predicts a future condition of the subject patient using the prognosis prediction model constructed by the prognosis prediction model generation device, A basic data acquisition unit that acquires basic data including information regarding the past and / or current state of the disease of the subject patient; a feature amount calculation unit that calculates the feature amount based on the basic data; an input condition receiving unit that receives a designation of an input condition including a prediction item indicating a type of the prediction; a prognosis prediction unit that predicts a future state of the subject patient related to the prediction item based on the calculated feature amount; A display unit that displays a prediction result indicating a predicted future condition of the subject patient; A prognosis prediction device comprising: The prognosis prediction system is characterized by comprising:

[0012] According to this, the prognosis of a target patient is predicted using a prognosis prediction model constructed by machine learning feature amounts extracted based on learning base data including information on the disease state of patients, including a target patient who is a specific individual, so that it is possible to predict the prognosis of the target patient with higher accuracy. In addition, when predicting a prognosis using a prognosis prediction model, input conditions including prediction items indicating the type of prediction can be specified, so that it is possible to provide a type of prediction result that is useful for selecting a treatment method suitable for each patient. The types of predictions that are prediction items include, but are not limited to, the state of the subject patient at any point or time in the future and / or what state the subject patient will be in. Here, the time when information on the state of a patient with a disease is obtained is not strictly limited to the time when the learning base data is obtained, but is regarded as the time when the information is substantially obtained in relation to the process of generating a prognosis prediction model. The present in the past and / or present state of a disease is not strictly limited to the time when the basic data was acquired, but includes a period that can be substantially regarded as the present in relation to the processing of prognosis prediction. The time when the learning basic data was acquired may extend to a period that can be substantially regarded as the present in relation to the processing of prognosis prediction.

[0013] In the present invention, The input condition receiving unit Information regarding future treatment for the subject patient may be received.

[0014] This allows for a prediction of the future condition of the target patient based on future treatment being considered, providing useful information that will aid in the consideration of what type of treatment, including medication, should be performed.

[0015] In the present invention, The prediction items may include predictions of at least one of the target patient's condition the next day, condition after 30 days, and condition after 90 days.

[0016] This makes it possible to obtain prediction results by specifying prediction items assuming various scenarios, such as when the target patient may suddenly change and be admitted to an intensive care unit, or when they may be transferred to a general ward, be discharged from the hospital, or be treated after discharge, making it possible to consider treatment methods that assume a variety of possible progressions.

[0017] In the present invention, The learning basis data may include a history of treatment of the disease for the patient.

[0018] In this way, by using the history of treatment of a patient's disease in the machine learning of a prognosis prediction model, it is possible to more accurately predict the future condition of the target patient. In addition, when future treatment is used as an input condition for the prognosis prediction model, it is possible to more accurately predict the effect of treatment on the condition of the target patient.

[0019] In the present invention, The input condition change receiving unit may be configured to receive a change to the input condition based on the prediction result displayed on the display unit.

[0020] According to this, by referring to the prediction results and changing the input conditions, a more appropriate treatment method for a desired prognosis can be considered.

[0021] The present invention also provides a method for producing a method for manufacturing a semiconductor device comprising the steps of: A learning basic data acquisition unit that acquires learning basic data including information on the disease state of patients, including a target patient who is a specific individual; A feature extraction unit that extracts features based on the learning base data; a model construction unit that constructs a prognosis prediction model by machine learning to predict a future state of the patient using the feature amount; The present invention is a prognosis prediction model generating device comprising:

[0022] According to this, a prognosis prediction model is constructed by machine learning feature quantities extracted based on learning base data including information on the patient's condition, thereby providing a prognosis prediction model that can more accurately predict the prognosis of the target patient. Here, the information on the state of the patient is not strictly limited to the current information at the time of learning the learning base data, but is considered to be substantially current in relation to the process of generating the prognosis prediction model. This includes the period during which it can be done.

[0023] In the present invention, The learning basis data may include a history of treatment of the disease for the patient.

[0024] In this way, by using the treatment history of a patient's disease in the machine learning of a prognosis prediction model, it is possible to construct a prognosis prediction model that can make more accurate predictions about the future condition of a target patient.

[0025] In the present invention, The prognosis prediction model may predict at least one of the patient's next day condition, 30 day condition, and 90 day condition.

[0026] This makes it possible to obtain prediction results by specifying prediction items assuming various scenarios, such as when a patient may suddenly change their condition and be admitted to an intensive care unit, or when they may be transferred to a general ward, be discharged from the hospital, or be treated after discharge, thereby providing a prognosis prediction model that is useful for considering treatment methods assuming various progressions.

[0027] The present invention also provides a method for producing a method for manufacturing a semiconductor device comprising the steps of: A prognosis prediction device for predicting a future state of a disease in a target patient, the prognosis prediction model being constructed by machine learning of learning base data including information on a disease state of a patient, the target patient being a specific individual, the prognosis prediction device comprising: A basic data acquisition unit that acquires basic data including information regarding the past and / or current state of the disease of the subject patient; a feature amount calculation unit that calculates the feature amount based on the basic data; an input condition receiving unit that receives a designation of an input condition including a prediction item indicating a type of the prediction; a prognosis prediction unit that predicts a future state of the subject patient related to the prediction item based on the calculated feature amount; A display unit that displays a prediction result indicating a predicted future condition of the subject patient; The prognosis prediction device is characterized by comprising:

[0028] According to this, the prognosis of the target patient is predicted using a prognosis prediction model constructed by machine learning information on the disease state of patients including the target patient, so that it is possible to predict the prognosis of the target patient with higher accuracy. In addition, when predicting the prognosis using the prognosis prediction model, input conditions including prediction items indicating the type of prediction can be specified, so that it is possible to provide a type of prediction result that is useful for selecting a treatment method suitable for each patient. The types of predictions that are prediction items include, but are not limited to, the state of the subject patient at any point or time in the future and / or what state the subject patient will be in. Here, the information on the state of a patient with a disease is not strictly limited to that at the time of machine learning, but includes information on a period that can be substantially regarded as the time of machine learning in relation to the processing of constructing a prognosis prediction model. The present in the past and / or present state of a disease is not strictly limited to the present time at the time of acquisition of basic data, but is intended to include a period that can substantially be regarded as the present in relation to the processing of prognosis prediction. The time of machine learning may extend to a period that can substantially be regarded as the present in relation to the processing of prognosis prediction.

[0029] In the present invention, The input condition receiving unit Information regarding future treatment for the subject patient may be received.

[0030] According to this, the future condition of the target patient is predicted assuming future treatment being considered. The results obtained can provide useful information that will aid in the consideration of what kind of treatment, including medication, should be performed.

[0031] In the present invention, The prediction items may include predictions of at least one of the target patient's condition the next day, condition after 30 days, and condition after 90 days.

[0032] This makes it possible to obtain prediction results by specifying prediction items assuming various scenarios, such as when the target patient may suddenly change and be admitted to an intensive care unit, or when they may be transferred to a general ward, be discharged from the hospital, or be treated after discharge, making it possible to consider treatment methods that assume a variety of possible progressions.

[0033] In the present invention, The learning basis data may include a history of treatment of the disease for the patient.

[0034] In this way, by using the history of treatment of a patient's disease in the machine learning of a prognosis prediction model, it is possible to more accurately predict the future condition of the target patient. In addition, when future treatment is used as an input condition for the prognosis prediction model, it is possible to more accurately predict the effect of treatment on the condition of the target patient.

[0035] In the present invention, The input condition change receiving unit may be configured to receive a change to the input condition based on the prediction result displayed on the display unit.

[0036] According to this, by referring to the prediction results and changing the input conditions, a more appropriate treatment method for a desired prognosis can be considered.

[0037] The present invention also provides a method for producing a method for manufacturing a semiconductor device comprising the steps of: A step of acquiring learning basis data including information on the disease state of patients, including a target patient who is a specific individual; Extracting features based on the learning base data; constructing a prognosis prediction model by machine learning to predict a future condition of the patient using the feature amount; The present invention relates to a method for generating a prognosis prediction model, the method including:

[0038] According to this, a prognosis prediction model is constructed by machine learning feature quantities extracted based on learning base data including information on the patient's condition, thereby providing a prognosis prediction model that can more accurately predict the prognosis of the target patient. Here, the time when information regarding the condition of a patient with a disease is obtained is not strictly limited to the present time when the learning base data is obtained, but includes a period that can be essentially regarded as the present in relation to the process of generating a prognosis prediction model.

[0039] The present invention also provides a method for producing a method for manufacturing a semiconductor device comprising the steps of: A prognosis prediction model generating program for causing a computer to execute the prognosis prediction model generating method. In addition, the prognosis prediction model generating program may be stored in a storage medium that can be read by a computer or other device or machine.

[0040] The present invention also provides a method for producing a method for manufacturing a semiconductor device comprising the steps of: Learning base data including information on the disease status of patients, including target patients who are specific individuals A prognosis prediction method for predicting a future state of the disease of a subject patient using a prognosis prediction model for predicting a future state of the patient constructed by machine learning, obtaining baseline data including information regarding the past and / or current status of the disease in the subject patient; calculating the feature amount based on the basic data; receiving a designation of a prediction item indicating a type of the prediction; predicting a future state of the subject patient related to the prediction item based on the calculated feature amount; displaying a prediction result indicating a predicted future condition of the subject patient; Includes.

[0041] According to this, the prognosis of the target patient is predicted using a prognosis prediction model constructed by machine learning information on the disease state of patients including the target patient, so that it is possible to predict the prognosis of the target patient with higher accuracy. In addition, when predicting the prognosis using the prognosis prediction model, input conditions including prediction items indicating the type of prediction can be specified, so that it is possible to provide a type of prediction result that is useful for selecting a treatment method suitable for each patient. The types of predictions that are prediction items include, but are not limited to, the state of the subject patient at any point or time in the future and / or what state the subject patient will be in. Here, the information on the state of a patient with a disease is not strictly limited to that at the time of machine learning, but includes information on a period that can be substantially regarded as the time of machine learning in relation to the processing of constructing a prognosis prediction model. The present in the past and / or present state of a disease is not strictly limited to the present time at the time of acquisition of basic data, but is intended to include a period that can substantially be regarded as the present in relation to the processing of prognosis prediction. The time of machine learning may extend to a period that can substantially be regarded as the present in relation to the processing of prognosis prediction.

[0042] The present invention also provides a method for producing a method for manufacturing a semiconductor device comprising the steps of: The present invention relates to a prognosis prediction program for causing a computer to execute the prognosis prediction method. The prognosis prediction program may be stored in a storage medium that can be read by a computer or other device or machine. Effect of the Invention

[0043] According to the present invention, it is possible to provide a technique for supporting the selection of a treatment method suitable for an individual patient. [Brief description of the drawings]

[0044] [Figure 1] FIG. 1 is a diagram showing a schematic configuration of a prognosis prediction system according to a first embodiment of the present invention. [Diagram 2] 1 is a diagram illustrating a hardware configuration of a prognosis prediction model generating device according to a first embodiment of the present invention. [Diagram 3] FIG. 1 is a functional block diagram of a prognosis prediction model generating device according to a first embodiment of the present invention. [Figure 4] 1 is a flowchart illustrating a processing procedure of a prognosis prediction model generating method according to Example 1 of the present invention. [Diagram 5] FIG. 4 is a diagram showing an example of basic data according to the first embodiment of the present invention. [Figure 6] 1 is a diagram illustrating a hardware configuration of a prognosis prediction device according to a first embodiment of the present invention. [Figure 7] FIG. 1 is a functional block diagram of a prognosis prediction device according to a first embodiment of the present invention. [Figure 8] 1 is a flowchart illustrating a processing procedure of a prognosis prediction method according to Example 1 of the present invention. [Figure 9] FIG. 2 is a diagram showing an example of an input reception screen of the prognosis prediction device according to the first embodiment of the present invention. [Figure 10] FIG. 2 is a diagram showing an example of an input condition receiving screen of the prognosis prediction device according to the first embodiment of the present invention. [Figure 11] FIG. 2 is a diagram showing an example of a prediction result display screen of the prognosis prediction device according to the first embodiment of the present invention. [Figure 12] FIG. 11 is a diagram illustrating a hardware configuration of a prognosis prediction device according to a second embodiment of the present invention. [Figure 13] FIG. 11 is a functional block diagram of a prognosis prediction device according to a second embodiment of the present invention. [Figure 14] FIG. 11 is a diagram showing the configuration of a prognosis prediction system according to Example 3 of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0045] <Example 1> Hereinafter, the configuration of a prognosis prediction system 1 according to a first embodiment of the present invention will be described with reference to the drawings. However, the configuration of the device and system described in this embodiment should be appropriately changed depending on various conditions. In other words, it is not intended to limit the scope of the present invention to the following embodiment. FIG. 1 shows an overall configuration diagram of the prognosis prediction system 1. Here, a prognosis prediction model generating device 100, a prognosis prediction device 200, and a prognosis prediction device 300 are communicably connected via a network NT. The type of the network NT is not particularly limited, and can be appropriately selected from the Internet, a wireless communication network, a mobile communication network, a telephone network, a dedicated network, and the like. The prognosis prediction model generating device 100 is a device that generates a model that predicts the prognosis of a patient with a cardiovascular disease or the like. The prognosis prediction device 200 and the prognosis prediction device 300 are devices that execute prognosis prediction of a patient using the prognosis prediction model generated in the prognosis prediction model generating device 100. In the following, cardiovascular disease will be described as an example of a disease, but diseases to which the prognosis prediction system 1 can be applied are not limited to this (the same applies to the following examples).

[0046] As shown in FIG. 1, the prognosis prediction model may be provided from prognosis prediction model generating device 100 to prognosis prediction device 200 and prognosis prediction device 300 via a network NT, or the generated prognosis prediction model may be stored in a storage medium in prognosis prediction model generating device 100, and prognosis prediction device 200 and prognosis prediction device 300 may acquire the prognosis prediction model via this storage medium.

[0047] (Prognostic prediction model generation device) FIG. 2 is a hardware configuration diagram of the prognosis prediction model generating device 100 according to this embodiment. The prognosis prediction model generating device 100 is a computer device including a processor 11, a main memory unit 12, an auxiliary memory unit 13, an input unit 14, an output unit 15, an external interface 16, a communication interface 17, and a bus 18.

[0048] The processor 11 is a CPU, a DSP, or the like. The main memory unit 12 is composed of a ROM (Read Only Memory) and a RAM (Random Access Memory). The auxiliary storage unit 13 is an EPROM (Erasable Programmable ROM), a hard disk drive, This includes hard disk drives (HDDs), removable media, etc. The media may be, for example, a flash memory such as a USB memory or an SD card, or a disk recording medium such as a CD-ROM, a DVD disk, or a Blu-ray disk. An operating system (OS), various programs, various tables, etc. are stored in the auxiliary memory unit 13, and the programs stored therein are loaded into the working area of ​​the main memory unit 12 and executed by the processor 11. By controlling each component through the execution of the programs, each functional unit that serves a specific purpose, as described below, can be realized. However, some or all of the functional units may be implemented using ASICs (Application Specific Integrated Circuits) or FPGs. Even if it is realized by a hardware circuit such as A (Field Programmable Gate Array), However, the prognosis prediction model generating device 100 does not necessarily have to be realized by a single physical configuration, and may be realized by a plurality of computers linked to each other. In the following, the main memory unit 12 and a predetermined program loaded into the working area of ​​the main memory unit 12 are described. The processor 11 that executes this is also referred to as the control unit 10 .

[0049] The input unit 14 includes a keyboard, a mouse, a microphone, etc., and accepts input operations from the user. The output unit 15 includes a display, a speaker, etc., and provides information to the user. An external interface (denoted as I / F in the figure) 16 is an interface for connecting to various external devices. The communication interface 17 is an interface for connecting the prognosis prediction model generating device 100 to the network NT. The communication interface 17 can adopt an appropriate configuration depending on the method of connection with the network NT. The bus 18 is a signal transmission path that connects each part of the prognosis prediction model generating device 100 .

[0050] Fig. 3 shows a functional block diagram of the prognosis prediction model generating device 100. Fig. 4 shows a flowchart explaining the processing procedure of the prognosis prediction method executed in the prognosis prediction model generating device 100. The control unit 10 of the prognosis prediction model generating device 100 deploys a program, including a program for generating a prognosis prediction model described later, stored in the auxiliary storage unit 13 into the main storage unit 12, and realizes each functional unit by interpreting and executing the program by the processor 11. As a result, as shown in Fig. 3, the prognosis prediction model generating device 100 functions as a computer including a learning basic data acquiring unit 101, a feature extracting unit 102, a model constructing unit 103, a basic data storage unit 131, and a learned model storage unit 132.

[0051] The learning basic data acquisition unit 101 acquires basic data for training a prognosis prediction model described below (basic data for training a prognosis prediction model is sometimes referred to as learning basic data) (step S101). FIG. 4 shows an example of basic data. The basic data includes the medical history, medical history, and treatment history of a group of patients with cardiovascular disease, vital signs at the time of reception (blood pressure, pulse, weight, oxygen saturation, respiratory rate, heart sounds), monitor electrocardiograms, test results (X-rays, 12-lead electrocardiograms, blood sampling data), administered medications, and the like. The basic data shown in FIG. 5 is an example and is not limited to these. Time information related to each piece of basic data is also acquired. The time information is, in the case of a medical history, when the patient has been afflicted; in the case of a medical history, when the patient has been afflicted; in the case of a medical history, when the patient has been afflicted; in the case of a medical history, when the patient has been afflicted or when the patient has been afflicted; in the case of a medical history, when the patient has been afflicted or when the patient has been afflicted; in the case of a vital sign, a monitor electrocardiogram, or a test result, the time of acquisition thereof; in the case of an administered drug, when the patient has been afflicted or ... afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when the patient has afflicted or when

[0052] The control unit 10 can acquire the above-mentioned learning basic data by reading it from the basic data storage unit 131 provided in the auxiliary storage unit 13. The control unit 10 may acquire the learning basic data from the basic data storage unit 131 provided outside the prognosis prediction model generating device 100 via the communication I / F 17. Such a basic data storage unit 131 may be, for example, an electronic medical record of a patient provided outside the prognosis prediction model generating device 100, or a database provided in another university, hospital, research institute, or the like. Furthermore, the basic data may be acquired from an external device such as a monitor electrocardiograph via the external I / F 16 or the communication I / F 17. The control unit 10 may receive the learning basic data as an input from a user via the input unit 14.

[0053] The feature extraction unit 102 extracts features constituting learning data for machine learning of the prognosis prediction model from the learning basic data acquired by the learning basic data acquisition unit 101. (Step S102). The method of extracting the feature amount is not particularly limited. It is preferable to extract the feature amount from time series data such as a monitor electrocardiogram among the learning base data by using a recurrent neural network (RNN). It is preferable to extract features from image data such as X-rays and blood sampling data using a Convolutional Neural Network (CNN). The extraction method is not limited to these. Furthermore, the learning base data itself may be used as the feature amount.

[0054] The model construction unit 103 constructs a trained prognosis prediction model (hereinafter, the trained prognosis prediction model is also simply referred to as a prognosis prediction model) by machine learning the learning data consisting of the features extracted by the feature extraction unit 102 into a prognosis prediction model (step S103). An artificial intelligence architecture is used to construct the prognosis prediction model. A prognosis prediction model that evaluates the current condition of a patient based on the features is constructed using a Bayesian network. A Bayesian network is an acyclic directed graph with random variables as nodes, and is a type of graphical modeling that is expressed by a conditional probability table between parent nodes and child nodes. A known method can be used to construct a Bayesian network. For example, the Greedy Search algorithm, Stingy Structural learning algorithms such as the Search algorithm and the Full Search method can be used. Also, AIC (Akaike's Information Criterion), C4.5, CHM (Cooper Herskovits Measure), MDL (Minimum Description Length), ML (Maximum Likelihood), etc. can be used as evaluation criteria for the Bayesian network. Also, pairwise methods, listwise methods, etc. can be used as processing methods when the learning data used to build the Bayesian network contains missing values.

[0055] As described above, the output of the prognosis prediction model constructed by the model construction unit 103 is a probability value representing how the possibility of A's condition worsening will change at a predetermined time (e.g., the next day, the next 30 days, and / or the next 90 days) if A is observed as is, or if some treatment (e.g., a change in medication or a change in dosage) is performed. For example, the probability of hospitalization for heart failure, myocardial infarction, cerebral infarction, hospitalization for cardiovascular treatment, and cardiac death occurring in the next day, 30 days, and / or 90 days. The output of the prognosis prediction model is not limited to these. In particular, in the case of cardiovascular disease, there is a high possibility that a serious change in the condition that is life-threatening will occur in a short period of time, so it is very significant to be able to predict the possibility of the condition worsening the next day.

[0056] The prognosis prediction model constructed by the model construction unit 103 is stored together with the parameters in the learned model storage unit 132 provided in the auxiliary storage unit 13 (step S104).

[0057] As the target patient A continues to be hospitalized or receives repeated visits, the learning basic data acquisition unit 101 acquires the basic data that is generated successively as learning data, updates the data in the basic data storage unit 131, and the model construction unit 103 can re-learn the prognosis prediction model based on the updated learning basic data.

[0058] (Prognosis prediction device) Next, a prognosis prediction device 200 that implements the prognosis prediction model constructed by the prognosis prediction model generating device 100 and performs prognosis prediction will be described. Such prognosis prediction devices 200 and 300 can be installed in hospitals that examine and treat patients, such as core hospitals and local clinics associated with core hospitals. Since the configuration and functions of the prognosis prediction device 300 are similar to those of the prognosis prediction device 200, the prognosis prediction device 200 will be described below, and a separate description will be omitted as the same description applies to the prognosis prediction device 300.

[0059] FIG. 6 is a diagram showing the hardware configuration of the prognosis prediction device 200. As shown in FIG. The prognosis prediction device 200 is a computer device including a processor 211, a main memory 212, an auxiliary memory 213, an input unit 214, an output unit 215, an external interface 216, a communication interface 217, and a bus 218. Here, the main memory 212 and the processor 211 that executes a predetermined program loaded into the working area of ​​the main memory 212 are also collectively referred to as a control unit 210. The processor 211, the main memory 212, the auxiliary memory 213, the input unit 214, the output unit 215, the external interface 216, the communication interface 217, and the bus 218 have the same configurations and functions as the processor 11, the main memory 12, the auxiliary memory 13, the input unit 14, the output unit 15, the external interface 16, the communication interface 17, and the bus 218 of the prognosis prediction model generating device 100, respectively, and therefore detailed description thereof will be omitted.

[0060] Fig. 7 shows a functional block diagram of the prognosis prediction device 200. Fig. 8 is a flowchart explaining the processing procedure of the prognosis prediction method executed in the prognosis prediction device 200. The control unit 210 of the prognosis prediction device 200 deploys a program, including a program for executing prognosis prediction using a prognosis prediction model, stored in the auxiliary storage unit 213, in the main storage unit 212, and realizes each functional unit by interpreting and executing the program by the processor 211. As a result, as shown in Fig. 5, the prognosis prediction device 200 functions as a computer including a basic data acquisition unit 201, an input condition acquisition unit 202, a feature calculation unit 203, and a prognosis prediction unit 204.

[0061] The basic data acquisition unit 201 acquires basic data of the target patient A for performing prognosis prediction (step S201). The basic data is the data exemplified in FIG. 5 or a part thereof. The basic data may be input by the operator of the prognosis prediction device 200 using the input unit 214, or may read out basic data stored in the basic data storage unit 213a of the auxiliary storage unit 213 of the prognosis prediction device 200. The basic data storage unit 213a may be, for example, an electronic medical record of the target patient A stored in an in-hospital database provided outside the prognosis prediction device 200, or may be a database provided in another university, hospital, research institute, or the like. The basic data acquisition unit 201 may acquire the basic data included in such an electronic medical record via the external interface 216 or the communication interface 217.

[0062] The input condition acquisition unit 202 acquires input conditions for performing prognosis prediction based on the basic data acquired by the basic data acquisition unit 201 (step S202). The input conditions are information on what kind of prognosis prediction is to be performed (prediction item), i.e., what kind of prediction result is desired to be obtained, and information on a broad-sense treatment for the target patient A (variable item), which is a condition for performing the prognosis prediction. The broad-sense treatment includes, for example, continuing to monitor the progress as is and performing some kind of treatment. The input conditions can be input by the operator of the prognosis prediction device 200 using the input unit 214. The input condition acquisition unit 202 corresponds to an input condition reception unit.

[0063] The feature value calculation unit 203 calculates the value of the feature value based on the basic data and the input conditions. This feature value is the feature value described for the feature value extraction unit 102 of the prognosis prediction model generating device 100.

[0064] The prognosis prediction unit 204 outputs a prognosis prediction result by inputting the feature amount calculated based on the input information into the prognosis prediction model generated by the prognosis prediction model generating device 100. The prognosis prediction output by the prognosis prediction unit 204 is a probability value indicating the possibility of the condition of the target patient A worsening at a predetermined time in the future, acquired as a prediction item. For example, it is the probability of hospitalization for heart failure, myocardial infarction, cerebral infarction, cardiovascular treatment during hospitalization, and cardiac death occurring within 30 days and / or 90 days. The output of the prognosis prediction model is not limited to these. Prognosis The prognosis prediction result output by the prediction unit 204 is displayed, for example, on the display unit 215a.

[0065] If the prognosis prediction result displayed on the display unit 215a is not desirable, or if the user wishes to know the prognosis prediction result when a different treatment is performed, the user can change the variable items using the input unit 214. Here, the input condition acquisition unit 202 functions as an input condition change acceptance unit.

[0066] Furthermore, by using this prognosis prediction model, the prognosis prediction unit 204 can also simulate the condition of a hospitalized cardiovascular disease patient, particularly a cardiovascular patient hospitalized in an intensive care unit, the next day. By inputting information on a broad range of treatments for the target patient A, the condition of the patient the next day can be recognized based on vital signs 24 hours later, the presence or absence of an artificial respirator, drug demand, the possibility of leaving the intensive care unit, the probability of re-admission to the intensive care unit if the patient leaves the intensive care unit, and the like.

[0067] (Example of use) An example of how the prognosis prediction device 200 is used will be described with reference to Figures 9, 10, and 11. Here, a patient A who is admitted to an intensive care unit will be described as an example. 9 is a basic data reception screen 410 displayed on the display unit 215a included in the output unit 215 of the prognosis prediction device 200. Here, the basic data illustrated below regarding the target patient A is automatically input from the system of the intensive care unit connected to the prognosis prediction device 200 via the external interface 216 or the communication interface 217. At this time, data such as date of birth and age: April 12, 1956, 67 years old, sex: female, height: 167 cm, weight: 65 kg, body temperature 37.5°C, blood pressure: 118 / 65 mmHg, heart rate: 112 bpm, oxygen saturation: 94%, oxygen administration amount: 8 L / min, artificial respirator attached: yes, ECMO attached: no, blood test results: NTproBNP: xxx ng / L, BUN: xx mg / dL, Cre: xx mg / dL, etc. are input. The basic data reception screen 410 displays the above-mentioned automatically input basic data as the current state. A message 412 is displayed at the bottom of the basic data reception screen 410, asking "Is this OK?" to prompt the operator (e.g., a doctor) of the prognosis prediction device 200 to confirm. The operator checks whether there is a problem with the data displayed as basic data for prognosis prediction, and if there is no problem, selects ENTER 413 using the keyboard 214a, mouse 214b, or the like included in the input unit 214, and the displayed basic data is confirmed as input data for prognosis prediction. If the operator determines that there is a problem with the displayed data, the operator may change the value of the basic data displayed on the basic data reception screen 410 as necessary, or may input additional basic data using the keyboard 214a, or the like. When ENTER 413 is selected, the screen of the display unit 215a transitions to an input condition reception screen 420, which will be described later.

[0068] 10 shows an input condition acceptance screen 420 displayed on the display unit 215a of the prognosis prediction device 200. The input condition acceptance screen 420 displays a prediction item designation field 421 and a variable item designation field 422. The prediction item designation field 421 accepts designation of a prediction item that is an item to be predicted by a prognosis prediction model. The variable item designation field 422 accepts designation of a variable item that is an item that can be changed according to the prediction item designated in the prediction item designation field 421.

[0069] In the prediction item specification field 421, specification of a desired prediction item from among a plurality of prediction items displayed as a drop-down list 423 is accepted by the keyboard 214a, the mouse 214b, etc. In the input condition acceptance screen 420 shown in Fig. 10, "Possibility of needing artificial respirator in 24 hours" 424a, "Possibility of leaving intensive care unit in 24 hours" 424b, "Blood oxygen concentration in 24 hours" 424c, and "Blood pressure and heart rate in 24 hours" 424d are selected from the drop-down list. 423. These forecast items are illustrative and not limiting.

[0070] Similarly, the variable item designation field 422 accepts designation of a desired variable item from among a plurality of variable items displayed in a drop-down list format by the keyboard 214a, the mouse 214b, etc. The variable item designation field 422 shown in Fig. 10 displays a dosage designation field 425, a treatment plan 1 designation field 426, and a treatment plan 2 designation field 427. The number of variable item designation fields 422 is not limited to this. When "possibility of needing artificial respirator in 24 hours" 424a is specified in the prediction item specification field 421, for example, it is possible to specify that "drug A" should be "increased by XX mg" within "one hour" from now in the dosage specification field 425 of the variable item specification field 422. The part of this text displayed in the dosage specification field 425 shown in quotation marks is a drop-down list, and it is possible to change it to a desired content by operating a triangular spinner with the mouse 214b or the like. Also, it is possible to specify that "artificial respirator" should be "used" within "one hour" from now in the treatment plan 1 specification field 426 of the variable item specification field 422. The part of this text displayed in the treatment plan 1 specification field 426 shown in quotation marks is a drop-down list, and it is possible to change it to a desired content by operating a triangular spinner with the mouse 214b or the like.

[0071] Furthermore, when "Possibility of leaving the intensive care unit in 24 hours" 424b is specified in the prediction item specification field 421, for example, it is possible to specify that "Drug B" be "increased by XX mg" within "one hour" from now in the dosage specification field 425 of the variable item specification field 422. Also, it is possible to specify that "weaning" from the "artificial ventilator" within "one hour" from now in the treatment policy 1 specification field 426 of the variable item specification field 422.

[0072] A message 428 is also displayed at the bottom of the input condition reception screen 420, asking the operator of the prognosis prediction device 200 to confirm, saying, "Is this OK?" If the operator determines that there is no problem with the specified contents of the prediction items and variable items, he or she selects ENTER 429 using the keyboard 214a or mouse 214b included in the input unit 214, and the displayed prediction items and variable items are confirmed as the specified output contents for prognosis prediction. When ENTER 429 is selected, the screen of the display unit 215a transitions to a prediction result display screen 430 described later. Here, one type of variable item is specified corresponding to one prediction item in the input condition reception screen 420, but multiple types of variable items may be specified corresponding to one prediction item, and the probability values ​​of the prediction items for each type of variable item may be displayed in parallel in the prediction result display screen 430 described later. The prediction results for multiple treatment policies can be compared at a glance, and the effects of multiple treatment methods can be compared more efficiently.

[0073] 11 shows a prediction result display screen 430 displayed on the display unit 215a of the prognosis prediction device 200. Here, an example is shown in which the possibility of needing an artificial respirator after 24 hours is specified on the above-mentioned input condition receiving screen 420. The prediction result display screen 430 displays a title "Prediction result", a prediction item 432 "Possibility of needing an artificial respirator" specified on the input condition receiving screen 420, and a prediction result column 433. In the prediction result column 433, the probability value of the prediction item is displayed as a percentage, "20%", as the prediction result.

[0074] Furthermore, the prediction result display screen 430 displays input condition fields 434-436 similar to those of the input condition reception screen 420. In the dosage specification field 434 of these input condition fields 434-436, the text "Increase the amount of "Drug A" by XX mg" within the next "1 hour" is displayed by default, which is the content specified in the variable item specification field 422 of the input condition reception screen 420. Here again, the part of this text shown in quotation marks can be changed by dragging it. The input condition columns 434 to 436 are drop-down lists, and can be changed by operating a triangular spinner with the mouse 214b or the like. In addition, the treatment policy 1 specification column 435 of the input condition columns 434 to 436 displays the text "Use" an "artificial ventilator" within "one hour" from now, which is the content specified in the variable item specification column 422 of the input condition reception screen 420, by default. Here, too, the part of this text shown in quotation marks is a drop-down list, and can be changed by operating a triangular spinner with the mouse 214b or the like. By changing the items displayed in the drop-down format of the input condition columns 434 to 436, the probability value in the prediction result column 433 also changes. That is, by accepting changes to the specified content in the input condition columns 434 to 436 of the prediction result display screen 430 in this way, the operator can change the input conditions while referring to the probability value in the prediction result column 433, and compare the therapeutic effects of multiple treatment methods.

[0075] In this way, by using the prognosis prediction system 1, when considering a treatment plan for a cardiovascular patient in outpatient care, a doctor can input treatment options such as "observation" and "additional oral administration of drug A" into the prognosis prediction device 200, and perform a virtual experiment to predict the condition for 30 or 90 days from that point. In addition, when treating a hospitalized patient, a doctor can input treatment options such as "continue current treatment without change," "addition of drug," and "discharge from intensive care unit" into the prognosis prediction device 200, and perform a virtual experiment to predict the condition on the next day or re-aggravation in the near future. This helps the doctor decide whether to adopt a treatment method that he or she considers as an option and the timing, and supports rapid decision-making. In addition, it is possible to realize a customized treatment based on the data of each individual patient, not on general trends.

[0076] Moreover, such a prognosis prediction system 1 can also be configured by installing the prognosis prediction device 200 in a flagship hospital and installing the prognosis prediction device 300 in a local clinic or the like connected to the flagship hospital. For example, a doctor in charge of the ward of the target patient A at a core hospital can use the prognosis prediction device 200 to make a discharge decision for the hospitalized target patient A and to decide on a treatment plan after discharge. This can support the doctor's decision-making, reducing the burden on the doctor and ensuring the quality of the decision-making.

[0077] In local clinics and the like associated with the above-mentioned flagship hospital, the prognosis prediction device 300 can be used to provide information on the predicted future course of the patient A, propose a specific treatment plan, and suggest referral to a specialized hospital when the patient A visits the clinic after being discharged from the flagship hospital. This allows the clinic to strengthen medical treatment in a medical department (e.g., cardiology) without significantly changing the current medical treatment format. The clinic can also provide more high-quality medical treatment in a short period of time.

[0078] Specifically, when the target patient A is discharged from the hospital, the prognosis prediction device 200 can be used to predict the condition of the target patient A one month and three months later, and a future treatment plan can be determined based on the simulation results of the predicted future course, frequency of outpatient visits, follow-up methods such as medication, etc. By sharing such information with connected local clinics, it is also possible to strengthen regional cooperation.

[0079] <Example 2> 12 is a diagram showing a hardware configuration of a prognosis prediction device 500 according to a second embodiment of the present invention. The prognosis prediction device 500 has a prognosis prediction model generating function and a prognosis prediction function according to the first embodiment.

[0080] The prognosis prediction device 500 includes a processor 511, a main memory unit 512, an auxiliary memory unit 513, and an input The prognosis prediction device 200 is a computer device including a processing unit 514, an output unit 515, an external interface 516, a communication interface 517, and a bus 518. Here, the main memory unit 512 and the processor 511 that executes a predetermined program loaded into the working area of ​​the main memory unit 512 are also referred to as a control unit 510. Each of the components is similar to that described for the prognosis prediction model generating device 100 and the prognosis prediction device 200 of the first embodiment, and therefore a detailed description thereof will be omitted.

[0081] Fig. 13 shows a functional block diagram of the prognosis prediction device 500. The control unit 510 of the prognosis prediction device 500 deploys a program, including a program for generating a prognosis prediction model and a program for executing prognosis prediction, stored in the auxiliary storage unit 513, in the main storage unit 512, and interprets and executes the program by the processor 511, thereby realizing each functional unit. As a result, as shown in Fig. 13, the prognosis prediction device 500 functions as a computer including a learning basic data acquisition unit 501, a feature extraction unit 502, a model construction unit 503, a basic data storage unit 531, and a learned model storage unit 532. The functions of each functional unit are the same as those described for the prognosis prediction model generation device 100 and the prognosis prediction device 200 according to the first embodiment, and therefore detailed description thereof will be omitted.

[0082] In the prognosis prediction device 500, as the basic data of the target patient A is updated and stored in the basic data memory unit 531, the prognosis prediction model can be re-learned, and when performing prognosis prediction, the re-learned prognosis prediction model stored in the learned model memory unit 532 can be used, making it possible to make highly accurate predictions.

[0083] <Example 3> FIG. 14 shows an overall configuration diagram of a prognosis prediction system 3 according to a third embodiment of the present invention. In the prognosis prediction system 3, a prognosis prediction server 600 and a user terminal 700 are communicably connected via a network NT. The prognosis prediction server 600 has a configuration similar to that of the prognosis prediction device 500, and receives input conditions from the user terminal 700, or input of basic data and input conditions, executes prognosis prediction, transmits the prediction result to the user terminal 700, and displays it on its display unit. In FIG. 14, only one user terminal 700 is shown, but the number of user terminals 700 can be appropriately provided. The prognosis prediction system 3 can also be a system in which a user terminal 700 is provided in each of a core hospital and a local clinic or the like that cooperates with the core hospital. The prognosis prediction server 600 may be configured as a single computer, or may be configured as a plurality of computers that cooperate with each other.

[0084] The prognosis prediction server 600 has a configuration similar to that of the prognosis prediction device 200, and is provided separately with a prognosis prediction model generating device 100, and may perform prognosis prediction using a prognosis prediction model generated by the prognosis prediction model generating device 100.

[0085] In this manner, a user who obtains prognosis prediction results via the user terminal 700 can be provided with highly accurate prediction results using a prognosis prediction model that has been re-learned in response to updates to the basic data of the target patient A. [Explanation of symbols]

[0086] 1. Prognostic prediction system 101: Learning Basic Data Acquisition Section 102 Feature extraction unit 103...Model Construction Section 100. Prognostic prediction model generator 200 Prognostic prediction device 201 Basic Data Acquisition Section 202 Input condition acquisition unit 203 Feature Calculation Unit 204... Prognostic prediction section

Claims

1. A learning base data acquisition unit acquires learning base data that includes information about the disease status of patients, including a specific individual patient. A feature extraction unit that extracts features based on the aforementioned basic learning data, A model building unit constructs a prognosis prediction model using machine learning to predict the future condition of the patient using the aforementioned features, A prognostic prediction model generation device equipped with, A prognosis prediction device that predicts the future condition of the target patient using the prognosis prediction model constructed by the prognosis prediction model generation device, A basic data acquisition unit that acquires basic data including information about the past and / or current state of the disease of the target patient, A feature calculation unit that calculates the feature quantities based on the aforementioned basic data, An input condition receiving unit that accepts the specification of input conditions including a prediction item indicating the type of prediction, A prognosis prediction unit that predicts the future condition of the target patient related to the prediction item based on the calculated feature quantities, A display unit that displays prediction results showing the predicted future condition of the target patient, A prognostic prediction device equipped with, A prognosis prediction system characterized by including [the following].

2. The aforementioned input condition receiving unit is: The prognosis prediction system according to claim 1, characterized in that it receives information regarding future treatment for the aforementioned target patient.

3. The prognosis prediction system according to claim 1 or 2, characterized in that the prediction items include predictions of at least one of the patient's condition the following day, 30 days later, and 90 days later.

4. The prognosis prediction system according to claim 1 or 2, characterized in that the learning base data includes the treatment history of the disease for the patient.

5. The prognosis prediction system according to claim 1 or 2, further comprising an input condition change acceptance unit that accepts changes to the input conditions based on the prediction results displayed on the display unit.

6. A learning base data acquisition unit acquires learning base data that includes information about the disease status of patients, including a specific individual patient, and A feature extraction unit that extracts features based on the aforementioned basic learning data, A model building unit constructs a prognosis prediction model using machine learning to predict the future condition of the patient using the aforementioned features, A prognostic model generation device equipped with the following features.

7. The prognosis prediction model generation device according to claim 6, characterized in that the learning base data includes the treatment history of the disease for the patient.

8. The prognosis prediction model generating device according to claim 6 or 7, characterized in that the prognosis prediction model predicts at least one of the patient's condition the following day, 30 days later, and 90 days later.

9. A prognosis prediction device that predicts the future state of a patient's disease using a prognosis prediction model constructed by machine learning on learning base data including information about the disease state of a patient, including a specific individual, A basic data acquisition unit that acquires basic data including information about the past and / or current state of the disease of the target patient, A feature calculation unit that calculates the feature quantities based on the aforementioned basic data, An input condition receiving unit that accepts the specification of input conditions including a prediction item indicating the type of prediction, A prognosis prediction unit that predicts the future condition of the target patient related to the prediction item based on the calculated feature quantities, A display unit that displays prediction results showing the predicted future condition of the target patient, A prognosis prediction device characterized by being equipped with the following features.

10. The aforementioned input condition receiving unit is: The prognosis prediction device according to claim 9, characterized in that it receives information regarding future treatment for the aforementioned target patient.

11. The prognosis prediction device according to claim 9 or 10, characterized in that the prediction items include predictions of at least one of the patient's condition the following day, 30 days later, and 90 days later.

12. The prognosis prediction device according to claim 9 or 10, characterized in that the learning base data includes the treatment history of the disease for the patient.

13. The prognosis prediction device according to claim 9 or 10, further comprising an input condition change receiving unit that accepts changes to the input conditions based on the prediction results displayed on the display unit.

14. The steps include obtaining learning base data that includes information about the disease status of patients, including a specific individual patient, and The steps include extracting features based on the aforementioned training data, The steps include: constructing a prognosis prediction model using machine learning to predict the future condition of the patient using the aforementioned features; A method for generating a prognosis prediction model that includes the following.

15. A prognosis prediction model generation program that causes a computer to execute the prognosis prediction model generation method described in claim 14.

16. A method for predicting the future state of a patient's disease, using a prognosis model constructed by machine learning on learning base data including information about the disease state of a patient, including a specific individual patient, A step of obtaining basic data including information about the past and / or present state of the disease of the subject patient, A step of calculating the feature quantities based on the aforementioned basic data, A step of receiving the specification of a prediction item indicating the type of prediction, A step of predicting the future condition of the target patient related to the prediction item based on the calculated feature quantities, The steps include displaying a prediction result showing the predicted future condition of the patient, A method for predicting prognosis, characterized by including the following:

17. A prognosis prediction program that causes a computer to execute the prognosis prediction method described in claim 16.

Images