System for predicting clinical outcome of disease, program, and method

Abstract

To provide a technique capable of accurately predicting changes in physical conditions, including clinical outcomes.SOLUTION: The system for predicting clinical outcome of disease includes an information processing device that executes: a first program that causes an information processing device to execute a first information processing method of predicting the subsequent clinical outcome of a patient to be predicted from the examination data at the first visit; and a second program that causes the information processing device to execute a second information processing method of predicting the subsequent clinical outcome of the patient to be predicted from the examination data during hospitalization.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a system, program, and method for predicting clinical outcomes of diseases. [Background technology]

[0002] Populations are beginning to decline in 25 countries around the world. The total world population is predicted to begin declining by the middle of the 21st century. This means that the birthrate will decline and the population will age on a global scale. The tax burden used for the medical care and pensions of a large number of elderly people will make life difficult for young people. Attempting to solve the problem of disease through expensive, advanced medical care will strain national finances, creating a situation where only the economically well-off can receive treatment. To achieve sustainable medical care in which everyone can receive appropriate medical and welfare services, it is essential to develop technology that can individually predict and prevent the onset of disease (Non-Patent Document 1).

[0003] COVID-19, an acute respiratory disease caused by the SARS-CoV-2 virus, has been spreading steadily since its outbreak in December 2019. As of April 2021, more than 150 million people have been confirmed infected and over 3 million have died worldwide (Non-Patent Document 2). COVID-19 is characterized by a wide range of symptoms; while most infected individuals are asymptomatic or experience mild symptoms such as cough and fever, some individuals develop pneumonia that rapidly progresses and require mechanical ventilation (Non-Patent Document 3). With the ever-increasing influx of COVID-19 patients, medical institutions around the world are facing a serious shortage of medical resources, making it urgent to optimize the allocation of medical equipment such as ventilators and extracorporeal membrane oxygenation (ECMO), as well as medications and staff.

[0004] Because the severity of COVID-19 is closely related to the allocation of medical resources and staff, it is believed that if the severity prognosis and the timing of worsening could be predicted, optimal medical resources could be prepared in advance. Numerous prognostic scores and prognostic prediction models for COVID-19 severity and mortality have been proposed (Non-Patent Document 3). However, all of these prediction models predict prognosis using data at the time of hospitalization and do not take into account changes in the patient's condition over time, so their prediction accuracy is limited. [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Stein Emil Vollset, Emily Goren, Chun-Wei Yuan, Jackie Cao, Amanda E Smith, Thomas Hsiao, Catherine Bisignano, Gulrez S Azhar, Emma Castro, Julian Chalek, Andrew J Dolgert, Tahvi Frank, Kai Fukutaki, Simon I Hay, Rafael Lozano, Ali H Mokdad, Vishnu Nandakumar, Maxwell Pierce, Martin Pletcher, Toshana Robalik, Krista M Steuben, Han Yong Wunrow, Bianca S Zlavog, Christopher JL Murray (2020) Fertility, mortality, migration, and population scenarios for 195 countries and territories from 2017 to 2100:a forecasting analysis for the Global Burden of Disease Study Lancet 2020; 396: 1285-306 [Non-patent document 2] Dong E, Du H, Gardner L. An interactive web-based dashboard to track COVID-19 in real time. Lancet Infect Dis 2020;20(5):533-4. [Non-patent document 3] Menni C, Valdes AM, Freidin MB, et al. Real-time tracking of self-reported symptoms to predict potential COVID-19. Nat Med 2020;26(7):1037-40. [Non-patent document 4] Wynants L, Van Calster B, Collins GS, et al. Prediction models for diagnosis and prognosis of covid-19 infection: systematic review and critical appraisal. BMJ 2020;369:m1328. Summary of the Invention [Problem to be solved by the invention]

[0006] Due to the rapid increase in dementia and other geriatric diseases caused by the aging of society, the poleward migration of wild animals due to rapid global warming, and the destruction of the zoning between wild animals and humans and livestock due to deforestation, it is thought that emerging infectious diseases such as COVID-19 will become more frequent in the future. To achieve sustainable medical care amid such rapid changes in disease structure, it is necessary to expand intervention methods and timing beyond treatment after onset to include the detection of early signs and prevention. This will require highly accurate predictive technology that can identify signs that humans may not notice through conventional medical treatment processes and provide suggestions for intervention.

[0007] Predicting the progression of macroscopic conditions, such as the entire body, is essential for highly accurate, personalized prevention and preemptive medicine. However, no principles or methods exist for predicting such macroscopic changes. The knowledge system of biomedical sciences is based on microscopic molecular models based on causal relationships. Macroscopic state transitions based on this knowledge are explained using simulations and network models. However, these explanations approximate nonlinear, non-equilibrium open biological systems with linear models. Many existing scores and models are based on statistical models such as linear and additive models, and do not take into account the heterogeneity of individual patients or the interactions between explanatory variables, making them unsuitable for highly accurate prediction.

[0008] The present invention aims to provide a technology for predicting changes in physical conditions, including clinical outcomes, with high accuracy from previously unnoticed symptoms. [Means for solving the problem]

[0009] To address the above-mentioned challenges, the inventors considered it effective to understand the macro-states that a person can assume as a prerequisite for predicting physical conditions, including clinical outcomes. This list of possible states can be used to predict individual state transitions through inductive inference. A useful approach is to calculate state transition probabilities. Information geometry has demonstrated that in nonlinear systems in which components interact, possible states are confined within geometric constraints, rather than presupposed state space. Applying this concept to state identification is effective in identifying the constraints on possible states. This requires a classification method that converts different states into distances. Decision tree methods are suitable for this purpose because they not only determine distances based on the number of branches traversed but also distinguish nonlinear interactions. Meanwhile, to accurately predict changes in physical conditions, including clinical outcomes, in clinical settings, dynamic prediction is required, combining condition identification at the time of initial consultation with condition identification based on longitudinal data at the time of hospitalization or follow-up consultations. Therefore, the first step in solving this problem is to develop treatment and prevention plans based on condition identification at the time of initial consultation. In this case, we believe that methods such as unsupervised random forests and gradient boosting decision trees are effective, and by using gradient boosting decision trees, a type of ensemble machine learning, we developed a severity prognosis prediction model for COVID-19 based on test data obtained at the time of initial consultation.We then found that the subsequent clinical outcomes of the target patients can be predicted using the test data at the time of initial consultation of COVID-19 patients and this severity prognosis prediction model.

[0010] Next, by observing the progression of physical condition over time from the initial consultation, we need to accurately predict changes in physical condition, including clinical outcomes. This prediction is fundamentally different from predictions made at the initial consultation because interventions, such as treatment, occur during follow-up observation. We believe that the random survival forest method is effective for such predictions. We developed a dynamic mortality risk model by performing machine learning survival analysis based on longitudinal clinical data of hospitalized patients. We then found that by using the dynamic mortality risk model and laboratory data from hospitalized patients at multiple time points, we could accurately predict the COVID-19 mortality risk of the target patient. These new findings demonstrate that state classification using distance information using decision trees is effective in accurately predicting changes in physical condition, including clinical outcomes, in a variety of clinical situations. This method can be applied to any disease, leading to the completion of this invention.

[0011] In order to solve the above problems, one aspect of the present invention relates to the design of a method, system, or program for generating highly accurate clinical outcomes from patient data, characterized by incorporating two axes into a discrete physical state transition scheme: a predicted axis of state transition and a variation axis of actual state transition.

[0012] In order to solve the above-mentioned problems, a clinical outcome prediction system for a disease according to one aspect of the present invention includes: a first program that causes an information processing device to execute a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted based on test data at the time of the first visit; a second program that causes an information processing device to execute a second information processing method for predicting a subsequent clinical outcome of a patient who is a prediction target based on test data during hospitalization; an information processing device that executes The first information processing method includes: A first teacher data acquisition step of acquiring teacher data including one or more pairs of test data at the time of initial consultation of a patient suffering from a target disease and a clinical outcome; a first learning step of inputting test data at the time of the initial visit and learning a first prediction model that predicts the clinical outcome by referring to the training data; a first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the first test data acquisition step into the first prediction model; Including, The second information processing method comprises: A second teacher data acquisition step of acquiring teacher data including one or more pairs of test data during hospitalization of a patient suffering from a target disease and a clinical outcome; a second learning step of inputting test data at multiple time points during hospitalization and learning a second prediction model that predicts the clinical outcome by referring to the training data; a second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the second test data acquisition step into the second prediction model; It is a configuration that includes:

[0013] In order to solve the above-mentioned problems, a clinical outcome prediction system for a disease according to one aspect of the present invention includes: a first program that causes an information processing device to execute a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted based on test data at the time of the first visit; a second program that causes an information processing device to execute a second information processing method for predicting a subsequent clinical outcome of a patient who is a prediction target based on test data during hospitalization; an information processing device that executes The first information processing method includes: a first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting test data acquired in the test data acquisition step into a first prediction model that uses test data at the time of initial consultation as an input and predicts the clinical outcome; Including, The second information processing method comprises: a second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting test data acquired in the test data acquisition step into a second prediction model that uses test data at multiple time points during hospitalization as an input and predicts the clinical outcome; The configuration includes:

[0014] In order to solve the above-mentioned problems, a clinical outcome prediction program for a disease according to one embodiment of the present invention includes: a first program that causes an information processing device to execute a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted based on test data at the time of the first visit; a second program that causes an information processing device to execute a second information processing method for predicting a subsequent clinical outcome of a patient who is a prediction target based on test data during hospitalization; Including, The first information processing method includes: A first teacher data acquisition step of acquiring teacher data including one or more pairs of test data at the time of initial consultation of a patient suffering from a target disease and a clinical outcome; a first learning step of inputting test data at the time of the initial visit and learning a first prediction model that predicts the clinical outcome by referring to the training data; a first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the first test data acquisition step into the first prediction model; Including, The second information processing method comprises: A second teacher data acquisition step of acquiring teacher data including one or more pairs of test data during hospitalization of a patient suffering from a target disease and a clinical outcome; a second learning step of inputting test data at multiple time points during hospitalization and learning a second prediction model that predicts the clinical outcome by referring to the training data; a second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the second test data acquisition step into the second prediction model; The configuration includes:

[0015] In order to solve the above-mentioned problems, a clinical outcome prediction program for a disease according to one embodiment of the present invention includes: a first program that causes an information processing device to execute a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted based on test data at the time of the first visit; a second program that causes an information processing device to execute a second information processing method for predicting a subsequent clinical outcome of a patient who is a prediction target based on test data during hospitalization; Including, The first information processing method includes: a first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting test data acquired in the test data acquisition step into a first prediction model that uses test data at the time of initial consultation as an input and predicts the clinical outcome; Including, The second information processing method comprises: a second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting test data acquired in the second test data acquisition step into a second prediction model that uses test data acquired at multiple time points during hospitalization as an input and predicts the clinical outcome; The configuration includes:

[0016] In order to solve the above problems, one aspect of the present invention provides a method for predicting a clinical outcome of a disease, comprising: a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted based on test data at the time of initial diagnosis; A second information processing method for predicting a subsequent clinical outcome of the patient based on test data during hospitalization; Including, The first information processing method includes: A first teacher data acquisition step of acquiring teacher data including one or more pairs of test data at the time of initial consultation of a patient suffering from a target disease and a clinical outcome; a first learning step of inputting test data at the time of the initial visit and learning a first prediction model that predicts the clinical outcome by referring to the training data; a first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the first test data acquisition step into the first prediction model; Including, The second information processing method comprises: A second teacher data acquisition step of acquiring teacher data including one or more pairs of test data during hospitalization of a patient suffering from a target disease and a clinical outcome; a second learning step of inputting test data at multiple time points during hospitalization and learning a second prediction model that predicts the clinical outcome by referring to the training data; a second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the second test data acquisition step into the second prediction model; The configuration includes:

[0017] In order to solve the above problems, one aspect of the present invention provides a method for predicting a clinical outcome of a disease, comprising: a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted based on test data at the time of initial diagnosis; A second information processing method for predicting a subsequent clinical outcome of the patient based on test data during hospitalization; Including, The first information processing method includes: a first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting test data acquired in the test data acquisition step into a first prediction model that uses test data at the time of initial consultation as an input and predicts the clinical outcome; Including, The second information processing method comprises: a second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting test data acquired in the test data acquisition step into a second prediction model that uses test data at multiple time points during hospitalization as an input and predicts the clinical outcome; The configuration includes:

[0018] In order to solve the above-mentioned problems, a disease clinical outcome prediction program according to one embodiment of the present invention is a program that causes an information processing device to execute a second information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data during hospitalization, the program comprising: The second information processing method comprises: A training data acquisition step of acquiring training data including one or more pairs of test data during hospitalization of a patient suffering from a target disease and a clinical outcome; a learning step in which test data at multiple time points during hospitalization is input and a prediction model for predicting the clinical outcome is learned by referring to the training data; a test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the test data acquisition step into the prediction model; The configuration includes:

[0019] In order to solve the above-mentioned problems, a disease clinical outcome prediction program according to one embodiment of the present invention is a program that causes an information processing device to execute a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data at the time of an initial visit, the program comprising: The first information processing method includes: A teacher data acquisition step of acquiring teacher data including one or more pairs of test data at the time of the first visit of a patient suffering from a target disease and a clinical outcome; a learning step in which test data at the time of the initial visit is input and a prediction model for predicting the clinical outcome is learned by referring to the training data; an examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the test data acquisition step into the prediction model; The configuration includes:

[0020] In order to solve the above-mentioned problems, a disease clinical outcome prediction program according to one embodiment of the present invention is a program that causes an information processing device to execute a second information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data during hospitalization, the program comprising: The second information processing method comprises: a test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting test data acquired in the test data acquisition step into a prediction model that predicts the clinical outcome, using test data acquired at multiple time points during hospitalization as an input; The configuration includes:

[0021] In order to solve the above-mentioned problems, a disease clinical outcome prediction program according to one embodiment of the present invention is a program that causes an information processing device to execute a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data at the time of an initial visit, the program comprising: The first information processing method includes: an examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a prediction step of predicting a subsequent clinical outcome of the patient to be predicted by inputting test data at the time of the first visit into a prediction model that predicts the clinical outcome, the test data acquired in the test data acquisition step; The configuration includes:

[0022] The above-mentioned program and a computer-readable recording medium on which it is recorded also fall within the scope of the present invention. [Effects of the Invention]

[0023] According to one aspect of the present invention, it is possible to achieve an effect of being able to predict changes in physical conditions, including clinical outcomes, with high accuracy in various clinical situations. [Brief explanation of the drawings]

[0024] [Figure 1] FIG. 1 is a block diagram showing the configuration of a clinical outcome prediction system according to a first embodiment of the present invention. [Figure 2]1 is a flowchart showing the flow of an information processing method implemented by an information processing device included in the clinical outcome prediction system according to the first embodiment of the present invention. [Figure 3] 1 is a flowchart showing the flow of an information processing method implemented by an information processing device included in the clinical outcome prediction system according to the first embodiment of the present invention. [Figure 4] FIG. 10 is a block diagram showing the configuration of a clinical outcome prediction system according to a second embodiment of the present invention. [Figure 5] FIG. 10 is a flowchart showing the flow of an information processing method implemented by an information processing device included in a clinical outcome prediction system according to a second embodiment of the present invention. [Figure 6] FIG. 10 is a block diagram showing the configuration of an information processing device included in a clinical outcome prediction system according to a third embodiment of the present invention. [Figure 7] FIG. 11 is a flowchart showing the flow of an information processing method implemented by an information processing device included in a clinical outcome prediction system according to a third embodiment of the present invention. [Figure 8] FIG. 10 is a block diagram showing the configuration of a clinical outcome prediction system according to a fourth embodiment of the present invention. [Figure 9] FIG. 10 is a flowchart showing the flow of an information processing method implemented by an information processing device included in a clinical outcome prediction system according to a fourth embodiment of the present invention. [Figure 10] FIG. 10 is a block diagram showing the configuration of a clinical outcome prediction system according to a fifth embodiment of the present invention. [Figure 11] FIG. 2 is a block diagram showing the hardware configuration of an information processing device included in the clinical outcome prediction system according to the first to fifth embodiments of the present invention. [Figure 12] FIG. 1 shows the results of screening for severity prognosis at the time of initial consultation. [Figure 13] This figure shows the results of evaluating the importance of variables in predicting each severity level of COVID-19 using SHAP values. [Figure 14] FIG. 1 depicts box and jitter plots showing the distribution of significant continuous variables by severity group of COVID-19 patients. [Figure 15]Figure 1 shows the change in severity status during hospitalization by severity outcome group of COVID-19 patients. [Figure 16] This figure shows the results of assessing the risk of death in COVID-19 patients using the 7-day cumulative hazard function (CHF) for mortality calculated using RSF as an index. [Figure 17] FIG. 1 shows the results of analyzing variables that contribute to RSF prediction. [Figure 18] FIG. 1 shows the results of confirming the agreement between RSF prediction and actual clinical course in severe cases. [Figure 19] FIG. 10 shows the results of a comparative example, in which the risk of death was evaluated based on LDH, lymphocytes, and CRP identified by the NMI model. [Figure 20] FIG. 10 is a diagram showing the results of comparing the accuracy of the NMI model (comparative example) with the accuracy of the RSF model of this embodiment. [Figure 21] This figure shows the results of examining the effectiveness of the RSF model in predicting mortality and hospital discharge for non-COVID-19 cases. [Figure 22] FIG. 2 is a diagram illustrating a functional scheme for predicting a clinical outcome from disease data that forms the basis of a prediction program of an information processing device provided in the clinical outcome prediction system according to the first to fifth embodiments of the present invention. [Figure 23] FIG. 1 is a conceptual diagram showing a concept for inference. DETAILED DESCRIPTION OF THE INVENTION

[0025] [Outline of the Functional Method and Problems to be Solved According to One Aspect of the Present Invention] With reference to Figure 22, a functional scheme (method, system, or program) for predicting clinical outcomes from disease data that forms the basis of a prediction program in an information processing device included in a clinical outcome prediction system according to embodiments 1 to 5 of the present invention will be described. Hospitalized patients undergo tests starting with an initial examination and continuing over time. Predictions of the patient's prognosis and outcome can be divided into those based on test data from the initial examination (first information processing method) and those based on test data obtained over time (second information processing method).

[0026] In one aspect of the present invention, a clinical outcome prediction program is executed by combining programs (first program and second program) for executing both information processing methods, thereby enabling highly accurate prediction of changes in physical condition, including clinical outcomes, in various clinical situations. When prediction is made from test data at the initial visit, the physical condition at the initial visit is considered to be one state. When prediction is made from longitudinal test data during hospitalization, the period is considered to be one state. In either case, differences in conditions are identified based on the distance between states. In one aspect of the present invention, the first information processing method using test data at the initial visit uses a first prediction model trained using a gradient boosting decision tree. In another aspect of the present invention, the second information processing method using longitudinal data during hospitalization uses a second prediction model trained using a random survival forest.

[0027] FIG. 23 is a conceptual diagram illustrating the concept of inference by an information processing device included in the clinical outcome prediction system according to embodiments 1 to 5 of the present invention. FIG. 23 illustrates a problem to be solved by one aspect of the present invention. Knowledge in biomedical science is a directed graph integrating universal causal knowledge obtained through controlled trials using linear algebra. To predict the state of an individual and its changes using this knowledge, it is necessary to comprehensively measure the molecular network between cells and construct a prediction model. However, in reality, such measurements are not possible, nor is it possible to create a prediction model from the measured values. To solve this problem, the clinical outcome of an affected patient is predicted using a prediction system that identifies the state of an individual and its transitions from various test data using the method of the present invention.

[0028] On the other hand, the states (i.e., clinical outcomes) and state transitions identified by an information processing method according to one embodiment of the present invention can be used to select drugs or therapeutic interventions that reflect differences in state. A specific network expressed algebraically corresponds to each identified state, and by inferring the structure of the network from the identified state using a reverse method, it is possible to determine which elements to focus on in drug development or therapeutic intervention. In other words, one embodiment of the present invention can be used to realize personalized medicine.

[0029] [Embodiment 1] An embodiment of the present invention will be described in detail below. A clinical outcome prediction system 1000 for a disease according to embodiment 1 (hereinafter simply referred to as the "clinical outcome prediction system 1000") will be described with reference to FIG. 1. FIG. 1 is a block diagram showing the configuration of the clinical outcome prediction system 1000 according to embodiment 1 of the present invention. The clinical outcome prediction system 1000 according to embodiment 1 includes an information processing device 1 that executes a first program that causes an information processing device to execute a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted based on test data at the time of the first visit, and a second program that causes the information processing device to execute a second information processing method for predicting a subsequent clinical outcome of a patient to be predicted based on test data during hospitalization. The clinical outcome prediction system 1000 according to embodiment 1 further includes a database 2 connected to the information processing device 1.

[0030] (Configuration of information processing device 1) The configuration of an information processing device 1 according to the first embodiment will be described with reference to FIG.

[0031] As shown in FIG. 1, the information processing device 1 includes a control unit 11, a storage unit 12, an operation reception unit 13, a display unit 14, a communication unit 15, and a memory 16.

[0032] The communication unit 15 communicates with the database 2 via the network N. More specifically, the communication unit 15 receives test data of the patient who is the target of prediction from the database 2. The communication unit 15 supplies the test data of the patient who is the target of prediction received from the database 2 to the storage unit 12. Note that the specific configuration of the network N does not limit the present embodiment, and as an example, a wireless LAN (Local Area Network), a wired LAN, a WAN (Wide Area Network), a public line network, a mobile data communication network, or a combination of these networks can be used.

[0033] The memory unit 12 stores various types of data referenced by the control unit 11. As an example, the memory unit 12 stores teacher data TD referenced by the learning unit 112 described below. The memory unit 12 also stores model parameters MP that define the prediction model learned by the learning unit 112. The memory unit 12 also stores test data of the patient to be predicted that has been received from the database 2. The memory unit 12 also stores a first program and a second program described below.

[0034] Various applications for realizing the functions of each unit of the control unit 11 are loaded in the memory 16. The control unit 11 writes data to the memory 16 and reads data from the memory 16 to realize the functions of each unit.

[0035] 1, the control unit 11 includes a data acquisition unit 111, a learning unit 112, a prediction unit 113, and an output unit 114. The control unit 11 may be expressed as functioning as the data acquisition unit 111, the learning unit 112, the prediction unit 113, and the output unit 114.

[0036] As an example, control unit 11 executes a first program to execute a first information processing method. Also, as an example, control unit 11 executes a second program to execute a second information processing method. In other words, the first program is a program that causes control unit 11 to execute the first information processing method, and the second program is a program that causes control unit 11 to execute the second information processing method.

[0037] The information processing device 1 functions as a machine learning device in the learning stage and as a prediction device in the prediction stage. The machine learning device and the prediction device will be described below.

[0038] <Machine Learning Stage> The information processing device 1 will be described as a machine learning device that learns a prediction model. Note that in this specification, the expression "learning a model" may also be expressed as "training a model."

[0039] The data acquisition unit 111 acquires teacher data TD including one or more pairs of test data at the time of the first visit of a patient suffering from the target disease and the clinical outcome, or teacher data TD including one or more pairs of test data during hospitalization of a patient suffering from the target disease and the clinical outcome from the storage unit 12. The test data and teacher data TD will be described in detail later.

[0040] The learning unit 112 learns a prediction model by referring to the training data TD acquired by the data acquisition unit 111. Then, the learning unit 112 stores model parameters MP that define the learned prediction model in the storage unit 12. The prediction model will be described in detail later.

[0041] The information processing device 1 configured as described above can input test data from the initial consultation or test data collected at multiple points during hospitalization, and can effectively train a prediction model to predict the clinical outcome. Note that "outcome" refers to the progress or results of symptoms in the treatment of a disease or the like.

[0042] The test data, training data TD, and prediction model will be explained below.

[0043] (Items included in the test data) Test data refers to a dataset that includes at least multiple items of data, such as symptoms; patient background information (age, gender, etc.); vital information (medical record information, clinical test data, liquid biopsy (proteome, metabolome, exosome), single-cell profile (transcriptome), diagnostic imaging data (CT, X-ray), physiological function (electrocardiogram, heart rate, EEG), psychological characteristics, genomic and epigenomic information, microbiome information, etc.); blood test data; and urine test data. Data for each item included in the test data is available from the electronic medical records of patients suffering from the target disease.

[0044] (Teacher data)

[0045] A first example of the training data TD is training data including one or more pairs of test data from the initial visit of a patient suffering from a target disease and a clinical outcome. In the following description, this training data will be referred to as "first training data TD1." While the method of generating the first training data TD1 is not limited to this embodiment, as an example, a pair of test data from the initial visit of a patient suffering from a target disease and a clinical outcome is created by associating the test data from the initial visit of the patient suffering from the target disease with the clinical outcome of the patient. By creating such pairs for multiple patients suffering from the target disease, first training data TD1 including multiple pairs of test data from the initial visit of patients suffering from the target disease and a clinical outcome can be generated.

[0046] A second example of the training data TD is training data including one or more pairs of test data and clinical outcomes for a patient suffering from a target disease during hospitalization. In the following description, this training data is referred to as "second training data TD2." While the method for generating the second training data TD2 is not limited to this embodiment, as an example, first, a pair of test data and clinical outcomes for a patient suffering from a target disease during hospitalization is created by associating the test data at a certain time point during the patient's hospitalization with the patient's clinical outcome. Similarly, a pair of test data and clinical outcomes for a patient suffering from a target disease during hospitalization is created by associating the test data at each of multiple other time points during the patient's hospitalization with the patient's clinical outcome. By creating such pairs for multiple patients suffering from the target disease, second training data TD2 including multiple pairs of test data and clinical outcomes for patients suffering from the target disease during hospitalization can be generated.

[0047] In addition, the information processing device 1 may be configured to generate the first teacher data TD1 or the second teacher data TD2 within the device itself and store the generated first teacher data TD1 or second teacher data TD2 in the memory unit 12, or may be configured to acquire the first teacher data TD1 or the second teacher data TD2 generated outside the information processing device 1 via the communication unit 15 and store the acquired first teacher data TD1 or second teacher data TD2 in the memory unit 12.

[0048] (Prediction model) The specific configuration of the prediction model is not limited to this embodiment, but may be, for example, a configuration using a gradient boosting decision tree, a random forest, a random survival forest, an energy landscape model, a time-series state change model, or the like. Furthermore, when data on a scale of tens of thousands to hundreds of thousands of people is acquired, a configuration using a neural network such as a convolutional neural network (CNN) or a recurrent neural network (RNN) may be used in addition to or instead of the above-mentioned model. The specific configuration of the prediction model can be appropriately selected depending on the purpose of the prediction process by the prediction unit 113.

[0049] A first example of a prediction model is a prediction model that uses test data from an initial consultation as input and predicts the clinical outcome. In the following description, this prediction model will be referred to as the "first prediction model." The learning unit 112 can train the first prediction model by referring to the first teacher data TD1. A gradient boosting decision tree is a machine learning technique that combines three techniques: gradient descent, ensemble learning, and decision tree, and is the most powerful technique in supervised learning. It is preferable that the first prediction model be trained using a gradient boosting decision tree, because it does not require scale conversion, it is possible to prevent overfitting by adjusting the depth of the decision tree, and it is possible to identify variables important for classification and recognize them as existing knowledge.

[0050] A second example of a prediction model is a prediction model that uses test data from multiple time points during hospitalization as input and predicts the clinical outcome. In the following description, this prediction model is referred to as the "second prediction model." The learning unit 112 can train the second prediction model by referring to the second training data TD2. The second prediction model is preferably trained using a random survival forest, because using multiple decision trees improves prediction accuracy, adjusting the decision trees can adjust bias and variance to prevent overfitting, and variables that play important roles in prediction can be identified and compared with existing knowledge.

[0051] (Flow of machine learning method M1 by information processing device 1) Next, the flow of the machine learning method (information processing method) M1 by the information processing device 1 will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of the machine learning method M1 performed by the information processing device 1. Below, the first machine learning method M1 (first information processing method) will be described.

[0052] (Step M11: First teacher data acquisition step) First, in step M11, the data acquisition unit 111 acquires the first teacher data TD1.

[0053] (Step M12: First learning step) Next, in step M12, the learning unit 112 trains a first prediction model by referring to the first teacher data TD1 acquired by the data acquisition unit 111. Here, the first prediction model trained by the learning unit 112 is, as described above, a first prediction model that predicts the clinical outcome using test data from the first visit of the patient to be predicted as input.

[0054] As an example, the learning unit 112 inputs test data included in the first teacher data TD1 acquired by the data acquisition unit 111 into the first prediction model. Then, the learning unit 112 updates the parameters of the first prediction model so as to reduce the difference between the clinical outcome associated with the test data and the clinical outcome output by the first prediction model. The learning unit 112 trains the first prediction model by repeatedly performing this update process. The parameters of the first prediction model after training are stored in the storage unit 12 as the above-mentioned model parameters MP1.

[0055] According to the processing by the information processing device 1 as described above, the test data at the first visit of the patient to be predicted is input, and the first prediction model for predicting the clinical outcome can be suitably trained.

[0056] In step M11 of the above-described machine learning method M1, a second machine learning method M1' (second information processing method) can be performed by replacing the teacher data acquired by the data acquisition unit 111 with second teacher data TD2. In the second machine learning method M1', in step M11 (second teacher data acquisition step), the data acquisition unit 111 acquires second teacher data TD2. In step M12 (second learning step), the learning unit 112 learns the second prediction model by referring to the second teacher data TD2. According to such processing by the information processing device 1, it is possible to preferably learn a second prediction model that predicts the clinical outcome using test data from multiple time points during the hospitalization of the patient to be predicted as input.

[0057] <Prediction stage> An information processing device 1 as a prediction device that predicts the subsequent clinical outcome of a patient to be predicted from test data during hospitalization of the patient or test data at the time of the first visit will be described with reference to FIG.

[0058] The data acquisition unit 111 acquires, from the storage unit 12, test data D of the patient to be predicted.

[0059] The prediction unit 113 executes a prediction process for predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data D of the patient to be predicted, acquired by the data acquisition unit 111, into the prediction model trained by the learning unit 112. The test data D and the prediction results will be described in detail later.

[0060] The output unit 114 outputs the prediction result by the prediction unit 113 to the display unit 14.

[0061] The display unit 14 presents a clinical outcome based on the prediction result output by the output unit 114. The display unit 14 may be provided with a display panel and configured to present the clinical outcome based on the prediction result by the prediction unit 113 by image. The display unit 14 may also be provided with a speaker and configured to present the clinical outcome based on the prediction result by the prediction unit 113 by sound. According to the information processing device 1 including the display unit 14 as described above, the clinical outcome based on the prediction result by the prediction unit 113 can be suitably notified to the operator.

[0062] In this way, the information processing device 1 predicts the subsequent clinical outcome of the patient to be predicted using the test data of the patient to be predicted and both the first prediction model and the second prediction model for the target disease, thereby making it possible to predict changes in physical condition, including clinical outcomes, with high accuracy in various clinical situations.

[0063] The following describes the test data D used by the prediction unit 113 in the prediction process and the prediction results generated by the prediction unit 113 through the prediction process.

[0064] (Test data D) The items included in the test data D are as described above. A first example of the test data D is test data from the first visit of the patient to be predicted. In the following description, this test data will be referred to as "first test data D1." Data for each item included in the test data D1 can be obtained from the electronic medical record of the patient to be predicted at the time of the first visit. A combination of data for necessary items can be selected depending on the purpose of the prediction process by the prediction unit 113, and input as test data D1 to the first prediction model described above. For example, test data including at least data for 84 items such as background information and blood and urine test data shown in Table 1 can be used as the test data D1. [Table 1] JPEG2023108532000003.jpg213163JPEG2023108532000004.jpg217162JPEG2023108532000005.jpg220163JPEG20231085320 00006.jpg218163JPEG2023108532000007.jpg218163JPEG2023108532000008.jpg219163JPEG2023108532000009.jpg161162

[0065] A second example of test data D is test data collected at multiple points during the hospitalization of a patient to be predicted. In the following description, this test data will be referred to as "second test data D2." Data for each item included in test data D2 can be obtained from the electronic medical record of the patient to be predicted at any time during hospitalization. A combination of data for required items can be selected according to the purpose of the prediction process by prediction unit 113 and input as test data D2 to the second prediction model described above. For example, test data including data on 93 items such as blood tests and vital information shown in Table 2 can be used as test data D2. [Table 2] JPEG2023108532000011.jpg210162JPEG2023108532000012.jpg204162JPEG202 3108532000013.jpg207162JPEG2023108532000014.jpg208164JPEG2023108532 000015.jpg211162JPEG2023108532000016.jpg206163JPEG2023108532000017. jpg207163JPEG2023108532000018.jpg210162JPEG2023108532000019.jpg90162

[0066] It is rare that all of the items listed above are measured in all patients, and the types of items tested vary depending on the patient. In such cases, variable selection and imputation of missing values ​​(missing value imputation) can be performed to handle missing data.

[0067] In the missing value imputation process, a standard may be set for a certain percentage of a measurement item (e.g., 20% or more), and if the percentage of missing data for a measurement item is greater than the predetermined percentage, the item can be excluded from the test data. Then, for the remaining items, the missing data can be filled in using a known imputation method. Here, the known imputation method may be missForest, a single imputation method using a random forest, or it may be combined with a multiple imputation method in which imputation is repeated multiple times. The complete test data thus created, with no missing data for any items, can be used for prediction processing by the prediction unit 113.

[0068] (Prediction results) The prediction result by the prediction unit 113 includes at least information indicating the predicted outcome of which of the multiple clinical outcomes of the target disease the patient to be predicted will end up. Here, the "multiple clinical outcomes of the target disease" may include at least two or more clinical outcomes. While not limiting this embodiment, as an example, the multiple clinical outcomes of the target disease may include two types of clinical outcomes corresponding to recovery and death. As another example, as shown in the examples described below, the multiple clinical outcomes of the target disease may include six types of clinical outcomes classified based on the oxygen support status score.

[0069] The type of the "target disease" is not particularly limited and can be appropriately selected depending on the purpose. The clinical outcome prediction system 1000 according to the first embodiment is applicable to infectious diseases, non-communicable diseases in general, and psychiatric diseases.

[0070] "Infectious diseases" as used herein include acute respiratory disease caused by the SARS-CoV-2 virus (COVID-19), bacterial pneumonia, bacterial prostatitis, urinary tract infections, Ebola hemorrhagic fever, Crimean-Congo hemorrhagic fever, smallpox, South American hemorrhagic fever, Marburg disease, Lassa fever, plague, polio, tuberculosis, diphtheria, severe acute respiratory syndrome (SARS), Middle East respiratory syndrome (MERS), avian influenza (including, but not limited to, H5N1, H7N9, etc.), cholera, enterohemorrhagic Escherichia coli infection, bacterial dysentery, Typhoid, paratyphoid, hepatitis E, hepatitis A, echinococcosis, yellow fever, psittacosis, relapsing fever, Q fever, coccidioidomycosis, Zika virus infection, tick-borne encephalitis, anthrax, scrub typhus, botulism, malaria, tularemia, Lyme disease, Rift Valley fever, melioidosis, Legionnaires' disease, West Nile fever, rabies, severe fever with thrombocytopenia syndrome (SFTS), chikungunya, dengue fever, Japanese spotted fever, Japanese encephalitis, hantavirus pulmonary syndrome, brucellosis, leptospirosis, Rocky Mountain spotted fever, amoebic dysentery, viral hepatitis, carbapenem-resistant Enterobacteriaceae Bacterial infections, acute encephalitis, cryptosporidiosis, Creutzfeldt-Jakob disease, severe hemolytic streptococcal infection, acquired immunodeficiency syndrome (HIV / AIDS), giardiasis, invasive Haemophilus influenzae infection, invasive meningococcal infection, invasive pneumococcal infection, congenital rubella syndrome (CRS), syphilis, disseminated cryptococcosis, tetanus, vancomycin-resistant Staphylococcus aureus infection, vancomycin-resistant enterococcus infection, whooping cough, rubella, measles, drug-resistant Acinetobacter infection, respiratory syncytial virus infection, pharyngoconjunctival fever (swimming pool fever), group A hemolysis The infection may be any infectious disease such as streptococcal pharyngitis, infectious gastroenteritis (including, but not limited to, norovirus and rotavirus), chickenpox, hand, foot and mouth disease, erythema infectiosum, exanthema subitum, herpangina, mumps, influenza, acute hemorrhagic conjunctivitis, epidemic keratoconjunctivitis, chlamydial pneumonia, bacterial meningitis, mycoplasma pneumonia, aseptic meningitis, genital chlamydial infection, genital herpes virus infection, genital warts, gonorrhea, penicillin-resistant pneumococcus infection, methicillin-resistant Staphylococcus aureus infection, or drug-resistant Pseudomonas aeruginosa infection.In addition to these known infectious diseases, emerging infectious diseases caused by currently unidentified or non-existent viruses or bacteria are also included in the category of infectious diseases in this specification. The "infectious disease" is preferably an acute infectious disease.

[0071] In this specification, "lifestyle-related diseases" refers to a group of diseases in which lifestyle habits such as eating habits, exercise habits, rest, smoking, and drinking are involved in the onset and progression of such diseases, and examples thereof include, but are not limited to, cancer (including malignant neoplasms and intraepithelial neoplasms), heart disease, cerebrovascular disease, diabetes, hypertensive diseases, cirrhosis of the liver, and chronic renal failure.

[0072] As used herein, the term "mental illness" refers to a disease caused by functional impairment or organic problems in the brain, and can also be referred to as a mental disorder. Examples include, but are not limited to, mood disorders such as depression, schizophrenia, and dementia.

[0073] Although the information included in the prediction result by the prediction unit 113 does not limit this embodiment, the prediction result may include a probability that the outcome of the patient to be predicted will be the clinical outcome indicated by the prediction result. Here, the probability may be expressed as accuracy.

[0074] (Flow of clinical outcome prediction method M2 by information processing device 1) Next, the flow of the clinical outcome prediction method (information processing method) M2 by the information processing device 1 will be described with reference to Fig. 3. Fig. 3 is a flow diagram showing the flow of the clinical outcome prediction method M2 by the information processing device 1. The information processing device 1 executes the clinical outcome prediction method M2. The second clinical outcome prediction method M2 (second information processing method) will be described below.

[0075] The second clinical outcome prediction method M2 includes a second test data acquisition step M21 of acquiring test data D2 at multiple time points during the hospitalization of the patient to be predicted, a second prediction step M22 of predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data D2 acquired in the second test data acquisition step into the second prediction model, and a second output step M23 of outputting the prediction result from the second prediction step.

[0076] (Step M21: Second inspection data acquisition step) First, in step M21, the data acquisition unit 111 acquires the test data D2 stored in the storage unit 12 and acquired at multiple time points during the hospitalization of the patient who is the target of prediction.

[0077] The test data D2 may be obtained at multiple time points during the follow-up period (during hospitalization), and the timing of obtaining the test data at multiple time points is not particularly limited. However, there is a discrepancy between the timing of obtaining the test data and the timing of onset or worsening of the disease, and each patient follows a different course of the disease. In addition, there are cases where a patient who showed no signs of worsening at the time of hospitalization suddenly becomes seriously ill and dies during hospitalization. The risk of worsening or death varies dynamically even within an individual. Therefore, the shorter the time interval until new test data is obtained, the more accurately the outcome can be predicted. However, taking into account the effort required to obtain test data and the burden on the patient, test data should typically be obtained every three days, preferably every day.

[0078] (Step M22: Second prediction step) Next, in step M22, the prediction unit 113 performs a prediction process to predict the subsequent clinical outcome of the patient to be predicted by inputting test data D2 acquired by the data acquisition unit 111 at multiple points during the patient's hospitalization into the second prediction model trained by the learning unit 112.

[0079] (Step M23: Second output step) Subsequently, in step M23, the output unit 114 outputs the prediction result by the prediction unit 113 to the display unit 14. The prediction result obtained by the prediction process indicates the mortality risk (clinical outcome) at multiple time points, which makes it possible to dynamically evaluate the mortality risk.

[0080] In this way, according to processing by the information processing device 1, by inputting test data D2 at multiple points during the hospitalization of the patient to be predicted into the second prediction model for the target disease, the subsequent clinical outcome of the patient to be predicted is predicted, so that the clinical outcome of the target disease for the patient to be predicted can be predicted with high accuracy.

[0081] In step M21 of the clinical outcome prediction method M2 described above, a first clinical outcome prediction method M2' (first information processing method) can be performed by replacing the test data acquired by the data acquisition unit 111 with the first test data D1. In the first clinical outcome prediction method M2', in step M21 (first test data acquisition step), the data acquisition unit 111 acquires test data D1 from the first visit of the patient to be predicted, stored in the storage unit 12. In step M22 (first prediction step), the prediction unit 113 inputs the test data D1 from the first visit of the patient to be predicted, acquired by the data acquisition unit 111, into the second prediction model trained by the learning unit 112, thereby performing a prediction process to predict the subsequent clinical outcome of the patient to be predicted. The prediction result obtained by the prediction process indicates the severity prognosis at the first visit. This makes it possible to perform severity screening at the first visit based on the prediction result. This makes it possible to predict outcomes, such as whether the patient will be hospitalized and whether the condition will worsen after hospitalization, as early as at the first visit, thereby enabling screening to be performed to estimate the allocation of focused clinical resources.

[0082] Processing by the information processing device 1 makes it possible to predict changes in physical conditions, including clinical outcomes, with high accuracy in various clinical situations, which is expected to lead to appropriate intervention for each patient and advance preparation of optimal medical resources. This will contribute to achieving "Goal 3: Good health and well-being" of the Sustainable Development Goals (SDGs).

[0083] (Additional Notes to the First Embodiment) In the above description, an example has been given of a configuration in which the information processing device 1 executes a clinical outcome prediction method by combining both the first information processing method and the second information processing method, but this does not limit the present embodiment. The information processing device 1 may be configured to execute at least one of the first information processing method or the second information processing method.

[0084] [Embodiment 2] Other embodiments of the present invention will be described below. For ease of explanation, the same reference numerals will be used to designate components having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

[0085] 4 is a block diagram showing the configuration of a clinical outcome prediction system 1000a according to the present embodiment. The clinical outcome prediction system 1000a according to the second embodiment includes an information processing device 1a. The clinical outcome prediction system 1000a according to the second embodiment further includes a database 2 connected to the information processing device 1a.

[0086] 4, the information processing device 1a includes an identifying unit 115 and a proposing unit 116 in addition to the components included in the information processing device 1 according to embodiment 1. Of the components included in the information processing device 1a, the components other than the identifying unit 115 and the proposing unit 116 are the same as the components included in the information processing device 1 according to embodiment 1, and therefore will not be described here.

[0087] The identification unit 115 is configured to identify, in the process of predicting the clinical outcome in the prediction step, an item that has a greater contribution to the process of predicting the clinical outcome among multiple items contained in the test data.

[0088] The proposing unit 116 is configured to be able to propose candidate target molecules for therapeutic drugs for the target disease based on the identification results identified by the identifying unit 115.

[0089] (Flow of identification and proposal method M3 by information processing device 1a) Next, the flow of the identification and proposal method (information processing method) M3 by the information processing device 1a will be described with reference to Fig. 5. Fig. 5 is a flow diagram showing the flow of the identification and proposal method M3 by the information processing device 1a. The first information processing method and the second information processing method include the identification and proposal method M3. The information processing device 1a executes the identification and proposal method M3. The second identification and proposal method M3 (second information processing method) will be described below.

[0090] The second identification / proposition method M3 includes an identification step M31 for identifying, among multiple items contained in the test data, an item that has a greater contribution to the prediction process of the clinical outcome in the prediction step M22, and a proposal step M32 for proposing candidate target molecules for a therapeutic drug for the target disease based on the identification results of the identification step M31.

[0091] (Step M31: Second Identification Step) First, in step M31, the identification unit 115 performs an identification process to identify, among the multiple items contained in the test data, the item that has the greatest contribution to the clinical outcome prediction process by the prediction unit 113.

[0092] The identification process by the identification unit 115 is a process of selecting items corresponding to relatively heavily weighted features from among the features calculated by the second prediction model. While the specific method of the identification process is not limited to this embodiment, as an example, features with relatively heavy weights can be identified in the second prediction model by evaluating the variable importance in the second prediction model using a Shapley Additive exPlanations (SHAP) value. In step M31, the principles that influence clinical outcomes can be identified from the second prediction model, thereby ensuring the validity (reliability) of predictions made by the information processing device 1a according to this embodiment.

[0093] (Step M32: Second proposal step) Next, in step M32, the proposing unit 116 generates proposal information that proposes candidate target molecules for therapeutic drugs for the target disease based on the identification results identified by the identifying unit 115. The specific content of the proposal information is not limited to this embodiment, but as an example, it can be a list of target molecules with high variable importance in the second prediction model among the identification results identified by the identifying unit 115, sorted in descending order of variable importance. This allows the user to use the proposal information to search for target molecules for therapeutic drugs for the target disease.

[0094] In this way, the processing by the information processing device 1a makes it possible to identify the principles that affect clinical outcomes from the second prediction model, thereby ensuring the validity (reliability) of the predictions made by the information processing device 1a according to this embodiment. Furthermore, by using the proposed information from the proposal unit 116, the user can search for target molecules of therapeutic drugs for the target disease, which can lead to the discovery of target molecules of therapeutic drugs.

[0095] Similarly, the first information processing method includes a first identification / proposal method M3'. The first identification / proposal method M3' may perform the same processing as the second identification / proposal method M3 described above, but differs from the second identification / proposal method M3 in that in step M31 (first identification step), a principle that influences clinical outcomes is identified from the first prediction model. In step M32 (first proposal step), the proposal unit 116 generates proposal information that proposes candidate target molecules for a therapeutic drug for the target disease based on the identification results identified by the identification unit 115. In this way, the processing by the information processing device 1a makes it possible to identify the principle that influences clinical outcomes from the first prediction model.

[0096] (Additional Notes to Embodiment 2) In the above explanation, an example was given of a configuration in which the control unit 11a is equipped with both the identification unit 115 and the proposal unit 116, but this does not limit this embodiment, and the control unit 11a may be configured not to be equipped with the proposal unit 116.

[0097] In such a configuration, the identification result identified by the identification unit 115 is directly supplied to the output unit 114. Then, the output unit 114 outputs the identification result by the identification unit 115 to the display unit 14. Since the identification unit 115 can identify the principle that influences the clinical outcome from the first prediction model or the second prediction model, the validity (reliability) of the prediction by the information processing device 1a according to this embodiment is guaranteed.

[0098] [Embodiment 3] Other embodiments of the present invention will be described below. For ease of explanation, the same reference numerals will be used to designate components having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

[0099] 6 is a block diagram showing the configuration of a clinical outcome prediction system 1000b according to the present embodiment. The clinical outcome prediction system 1000b according to the third embodiment includes an information processing device 1b. The clinical outcome prediction system 1000b according to the third embodiment further includes a database 2 connected to the information processing device 1b.

[0100] 6, the information processing device 1b includes an identification unit 115 and an extraction unit 117 in addition to the components included in the information processing device 1 according to embodiment 1. Of the components included in the information processing device 1a, the components other than the identification unit 115 and the extraction unit 117 are the same as the components included in the information processing device 1 according to embodiment 1, and therefore a description thereof will be omitted here. Furthermore, the configuration of the identification unit 115 is the same as the identification unit 115 included in the information processing device 1b according to embodiment 2, and therefore a description thereof will be omitted here.

[0101] The extraction unit 117 is configured to extract test data to be used in the prediction step from the test data acquired by the data acquisition unit 111 based on the identification result by the identification unit 115.

[0102] (Flow of identification / extraction method M4 by information processing device 1b) Next, the flow of the identification / extraction method (information processing method) M4 by the information processing device 1b will be described with reference to Fig. 7. Fig. 7 is a flow diagram showing the flow of the identification / extraction method M4 by the information processing device 1b. The first information processing method and the second information processing method include the identification / extraction method M4. The information processing device 1b executes the identification / extraction method M4. The second identification / extraction method M4 (second information processing method) will be described below.

[0103] The second identification / extraction method M4 includes an identification step M41 for identifying, from among multiple items contained in the test data, an item that has a greater contribution to the prediction process of the clinical outcome in the prediction step M22, and a test data extraction step M42 for extracting test data items to be used in the prediction step M22 from the test data acquired in the test data acquisition step M21 based on the identification results from the identification step M41.

[0104] (Step M41: Second Identification Step) Step M41 is the same as step M31 described above, so a description thereof will be omitted.

[0105] (Step M42: Second inspection data extraction step) Next, in step M42, the extraction unit 117 performs an extraction process to extract test data items to be used by the prediction unit 113 in the prediction step M22 from the test data acquired by the data acquisition unit 111 based on the identification results identified by the identification unit 115.

[0106] The test data extraction process by the extraction unit 117 is a process of extracting, from the test data used by the prediction unit 113, items corresponding to feature quantities with relatively large weights among the feature quantities calculated by the second prediction model described above. The specific method of the identification process is not limited to this embodiment, but as an example, an arbitrary number of important items with high variable importance in the second prediction model among the identification results identified by the identification unit 115 may be extracted from the test data in descending order of variable importance, and the test data including these important items may be used by the prediction unit 113 in the prediction step M22. As a result, in the prediction step M22, a clinical outcome is predicted based on test data including items with higher variable importance, thereby enabling more accurate prediction.

[0107] Similarly, the first information processing method includes a first identification / extraction method M4'. The first identification / extraction method M4' may perform the same processing as the second identification / extraction method M4 described above, but differs from the second identification / extraction method M4 in that in step M42 (first test data extraction step), items corresponding to feature quantities with relatively large weights among feature quantities calculated by the first prediction model are extracted from the test data used by the prediction unit 113. In this way, the processing by the information processing device 1b makes it possible to extract important items with high variable importance in the first prediction model.

[0108] In step M42, the number of items extracted by the extraction unit 117 from the test data is not particularly limited, but from the viewpoint of capturing diverse characteristics of diseases, it is preferable that the number of items extracted by the extraction unit 117 from the test data be 10 or more. On the other hand, from the viewpoint of preventing overlearning due to an excessive increase in variables and reducing the burden of measuring a large number of items in clinical practice, it is preferable that the number of items extracted by the extraction unit 117 from the test data be 20 or less. Therefore, in step M42, it is preferable that the number of items extracted by the extraction unit 117 from the test data be the top 10 to 20 items in descending order of variable importance in the first prediction model or the second prediction model.

[0109] The extraction unit 117 may extract test data or items corresponding to feature quantities with relatively large weights among feature quantities calculated by the first prediction model or the second prediction model from the test data (training data) used in the update process (retraining) of the first prediction model or the second prediction model, and use these as the test data or items used by the learning unit 112 in the learning step M12. This allows the first prediction model or the second prediction model to be retrained based on variables with higher variable importance in the learning step M12, thereby enabling more accurate predictions.

[0110] In this way, the processing by the information processing device 1b allows the principles that affect clinical outcomes to be identified from the first prediction model or the second prediction model, thereby ensuring the validity (reliability) of the prediction by the information processing device 1b according to this embodiment. Furthermore, the prediction unit 113 performs prediction processing using the test data extracted by the extraction unit 117, thereby enabling more accurate prediction. [Embodiment 4] Other embodiments of the present invention will be described below. For ease of explanation, the same reference numerals will be used to designate components having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

[0111] 8 is a block diagram showing the configuration of a clinical outcome prediction system 1000c according to the fourth embodiment. The clinical outcome prediction system 1000c according to the fourth embodiment includes an information processing device 1c. The clinical outcome prediction system 1000c according to the fourth embodiment further includes a database 2 connected to the information processing device 1c.

[0112] 8, the information processing device 1c includes a proposal unit 116 in addition to the components included in the information processing device 1 according to embodiment 1. Of the components included in the information processing device 1c, the components other than the proposal unit 116 are the same as the components included in the information processing device 1 according to embodiment 1, and therefore, description thereof will be omitted here.

[0113] In the information processing device 1c, the proposal unit 116 is configured to be able to propose a drug or therapeutic intervention method for the patient being predicted based on the first prediction result R1 and the second prediction result R2 predicted by the prediction unit 113 by executing the first information processing method and the second information processing method.

[0114] The first prediction result R1 is a prediction result obtained by the prediction unit 113 performing a first information processing method for predicting the subsequent clinical outcome of the patient to be predicted from the test data at the time of the first visit. In other words, the first prediction result R1 indicates the clinical outcome of the patient to be predicted based on the test data at the time of the first visit.

[0115] The second prediction result R2 is a prediction result obtained by the prediction unit 113 performing a second information processing method for predicting the subsequent clinical outcome of the patient to be predicted from test data during hospitalization. That is, the second prediction result R2 indicates the clinical outcome of the patient to be predicted based on test data during hospitalization. The first prediction result R1 and the second prediction result R2 are stored in the memory unit 12a.

[0116] (Flow of third information processing method M5 by information processing device 1c) Next, the flow of a third information processing method M5 performed by the information processing device 1c will be described with reference to FIG. 9. FIG. 9 is a flow chart showing the flow of the third information processing method M5 performed by the information processing device 1c. A disease clinical outcome prediction method according to one embodiment of the present invention includes the third information processing method M5 in addition to the first and second information processing methods described above. The information processing device 1c executes a third program that executes the third information processing method M5. That is, the clinical outcome prediction program according to one embodiment of the present invention includes a first program, a second program, and a third program. The third information processing method M5 includes a prediction result acquisition step M51 in which the data acquisition unit 111 acquires a first prediction result R1 and a second prediction result R2 (the patient's clinical outcome predicted by the first and second information processing methods) stored in the storage unit 12a, a proposal step M52 in which a drug or therapeutic intervention method is proposed based on the first prediction result and the second prediction result, and an output step M53 in which the proposal result is output.

[0117] (Step M51) First, in step M51, the data acquisition unit 111 acquires the first prediction result R1 and the second prediction result R2 stored in the storage unit 12a.

[0118] (Step M52: Proposal Step) Next, in step M52, the proposing unit 116 performs a proposing process to propose a drug or therapeutic intervention method for the patient to be predicted, by referring to the first prediction result R1 and the second prediction result R2 acquired by the data acquiring unit 111.

[0119] The proposal process by the proposal unit 116 is a process of identifying differences in the condition (clinical outcome) of the patient to be predicted by referring to the first prediction result R1 and the second prediction result R2, and selecting and proposing appropriate drugs and therapeutic intervention methods based on the differences in the condition and the transition of the condition. This makes it possible to optimize the effect of drugs or therapeutic intervention methods on the patient to be predicted by referring to the first prediction result R1 and the second prediction result R2 predicted with high accuracy by the prediction unit 113.

[0120] There is no particular limitation on the type of drug or therapeutic intervention method proposed by the proposing unit 116. The proposing unit 116 can propose a drug or therapeutic intervention method that is most suitable for the condition (clinical outcome) of the patient to be predicted from among all possible drug or therapeutic intervention method options.

[0121] (Step M53) Subsequently, in step M53, the output unit 114 outputs the proposal result by the proposal unit 116 to the display unit 14.

[0122] In this way, the processing by the information processing device 1c can refer to the first prediction result R1 and the second prediction result R2 predicted with high accuracy by the prediction unit 113 to suggest a drug or therapeutic intervention method that is most suitable for the condition (clinical outcome) of the patient to be predicted, thereby optimizing the effect of the drug or therapeutic intervention method on the patient to be predicted. In other words, the clinical outcome prediction system 1000c according to this embodiment can be applied to the development of personalized medicine.

[0123] [Embodiment 5] Other embodiments of the present invention will be described below. For ease of explanation, the same reference numerals will be used to designate components having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

[0124] 10 is a block diagram showing the configuration of a clinical outcome prediction system 1000d according to the fifth embodiment. The clinical outcome prediction system 1000d according to the fifth embodiment includes an information processing device 1d. The clinical outcome prediction system 1000d according to the fifth embodiment further includes a database 2 connected to the information processing device 1d.

[0125] 10, the information processing device 1d does not include the learning unit 112 among the components included in the information processing device 1 according to embodiment 1. The components included in the information processing device 1d are similar to the components included in the information processing device 1 according to embodiment 1, and therefore, description thereof will be omitted here.

[0126] The information processing device 1d configured as described above functions as a prediction device. The information processing device 1d predicts the subsequent clinical outcome of the patient to be predicted using test data D1 from the patient's initial visit and the first prediction model described above for the target disease. The information processing device 1d also predicts the subsequent clinical outcome of the patient to be predicted using test data D2 from multiple time points during the patient's hospitalization and the second prediction model described above for the target disease. By combining both of these programs, the information processing device 1d can predict changes in physical condition, including clinical outcomes, with high accuracy in various clinical situations.

[0127] (Additional Notes to the Fifth Embodiment) The information processing device 1d may perform a third information processing method including a step of proposing a drug or a therapeutic intervention method based on the clinical outcome of the patient predicted by the first information processing method and the second information processing method. For this purpose, the information processing device 1d may be configured to further include a proposing unit 116. This allows the information processing device 1d to refer to the first prediction result R1 and the second prediction result R2 to identify differences in the condition (clinical outcome) of the patient to be predicted, and to select and propose an appropriate drug or therapeutic intervention method based on the difference in condition and the progression of the condition.

[0128] [Software implementation example] The functions of the information processing devices 1, 1a, 1b, 1c and 1d (hereinafter referred to as "devices") can be realized by a program for causing a computer to function as the device, and a program for causing a computer to function as each control block of the device (particularly each part included in the control units 11, 11a, 11b, 11c and 11d).

[0129] The devices 1, 1a, 1b, 1c, and 1d execute a first program to perform a first information processing method, and execute a second program to perform a second information processing method.

[0130] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The control device and storage device execute the program, thereby realizing the functions described in each of the above embodiments.

[0131] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0132] Furthermore, some or all of the functions of the control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of the control blocks can also be realized by, for example, a quantum computer.

[0133] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

[0134] 〔summary〕 A disease clinical outcome prediction system according to a first aspect of the present invention includes an information processing device that executes a first program that causes the information processing device to execute a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data at the time of initial consultation, and a second program that causes the information processing device to execute a second information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data during hospitalization, wherein the first information processing method includes a first teacher data acquisition step of acquiring teacher data including one or more pairs of test data at the time of initial consultation and clinical outcome of a patient suffering from a target disease, a first learning step of using the test data at the time of initial consultation as input and training a first prediction model that predicts the clinical outcome by referring to the teacher data, a first test data acquisition step of acquiring test data at the time of initial consultation of the patient to be predicted, and a first prediction step of predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the first test data acquisition step into the first prediction model, the second information processing method includes: a second training data acquisition step of acquiring training data including one or more pairs of test data during hospitalization of a patient suffering from a target disease and a clinical outcome; a second learning step of inputting test data at multiple time points during hospitalization and training a second prediction model that predicts the clinical outcome by referring to the training data; a second test data acquisition step of acquiring test data at multiple time points during hospitalization of the patient to be predicted; and a second prediction step of predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the second test data acquisition step into the second prediction model.

[0135] A clinical outcome prediction system for a disease according to a second aspect of the present invention includes an information processing device that executes a first program that causes the information processing device to execute a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data at the time of the patient's initial visit, and a second program that causes the information processing device to execute a second information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data obtained during hospitalization, wherein the first information processing method includes a first test data acquisition step of acquiring test data at the time of the patient to be predicted from the first visit, and a first prediction step of predicting a subsequent clinical outcome of the patient to be predicted from a first prediction model that predicts the clinical outcome, using the test data at the time of the patient's initial visit as input; and a second information processing method includes a second test data acquisition step of acquiring test data at multiple time points during the patient's hospitalization, and a second prediction step of predicting a subsequent clinical outcome of the patient to be predicted from a second prediction model that predicts the clinical outcome, using the test data at the multiple time points during hospitalization as input.

[0136] In the disease clinical outcome prediction system according to aspect 3 of the present invention, in aspect 1 or 2, it is preferable that the second prediction model is trained using a random survival forest.

[0137] In the disease clinical outcome prediction system according to aspect 4 of the present invention, in aspect 1 or 2, the first prediction model is preferably one trained using a gradient boosting decision tree.

[0138] A disease clinical outcome prediction system according to aspect 5 of the present invention may be any one of aspects 1 to 4, wherein the first information processing method further includes a first identification step of identifying, in the clinical outcome prediction process in the first prediction step, an item from among a plurality of items contained in the test data that has a greater contribution to the clinical outcome prediction process, and the second information processing method further includes a second identification step of identifying, in the clinical outcome prediction process in the second prediction step, an item from among a plurality of items contained in the test data that has a greater contribution to the clinical outcome prediction process.

[0139] A clinical outcome prediction system for diseases according to aspect 6 of the present invention may be such that, in aspect 5, the first information processing method further includes a first proposal step of proposing candidate target molecules for therapeutic drugs for the target disease based on the identification results of the first identification step, and the second information processing method further includes a second proposal step of proposing candidate target molecules for therapeutic drugs for the target disease based on the identification results of the second identification step.

[0140] A disease clinical outcome prediction system according to aspect 7 of the present invention may be such that, in aspect 5, the first information processing method further includes a first test data extraction step of extracting test data to be used in the first prediction step from the test data acquired in the first test data acquisition step based on the identification result from the first identification step, and the second information processing method further includes a second test data extraction step of extracting test data to be used in the second prediction step from the test data acquired in the second test data acquisition step based on the identification result from the second identification step.

[0141] In the disease clinical outcome prediction system of aspect 8 of the present invention, in aspect 7, it is preferable that in the first test data extraction step and the second test data extraction step, the top 10 to 20 items are extracted from the test data in order of increasing variable importance in the first prediction model or the second prediction model.

[0142] A disease clinical outcome prediction system according to aspect 9 of the present invention may be configured such that, in any one of aspects 1 to 8, the information processing device further executes a third program that causes the information processing device to execute a third information processing method, which includes a step of suggesting a drug or therapeutic intervention method based on the clinical outcome of the patient predicted by the first information processing method and the second information processing method.

[0143] A disease clinical outcome prediction program according to aspect 10 of the present invention includes a first program that causes an information processing device to execute a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data at the time of initial consultation, and a second program that causes an information processing device to execute a second information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data during hospitalization, wherein the first information processing method includes a first teacher data acquisition step of acquiring teacher data including one or more pairs of test data at the time of initial consultation and clinical outcome of a patient suffering from a target disease, a first learning step of using the test data at the time of initial consultation as input and training a first prediction model that predicts the clinical outcome by referring to the teacher data, a first test data acquisition step of acquiring test data at the time of initial consultation of the patient to be predicted, and the second information processing method includes a program comprising: a first prediction step of predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in a training step; a second training data acquisition step of acquiring training data including one or more pairs of test data and clinical outcomes of a patient suffering from a target disease during hospitalization; a second learning step of inputting test data from multiple time points during hospitalization and training a second prediction model that predicts the clinical outcome by referring to the training data; a second test data acquisition step of acquiring test data from multiple time points during hospitalization of the patient to be predicted; and a second prediction step of predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the second test data acquisition step into the second prediction model.

[0144] A disease clinical outcome prediction program according to aspect 11 of the present invention includes: a first program that causes an information processing device to execute a first information processing method for predicting the subsequent clinical outcome of a patient to be predicted from test data at the time of initial consultation; and a second program that causes an information processing device to execute a second information processing method for predicting the subsequent clinical outcome of a patient to be predicted from test data during hospitalization, wherein the first information processing method includes a first test data acquisition step for acquiring test data at the time of initial consultation of the patient to be predicted; and a first prediction step for predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the test data acquisition step into a first prediction model that predicts the clinical outcome, and the second information processing method includes a second test data acquisition step for acquiring test data at multiple time points during the patient's hospitalization, and a second prediction step for predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data at the multiple time points during hospitalization into a second prediction model that predicts the clinical outcome.

[0145] A method for predicting a clinical outcome of a disease according to aspect 12 of the present invention includes a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data at the time of initial consultation, and a second information processing method for predicting a subsequent clinical outcome of a patient to be predicted from test data during hospitalization, wherein the first information processing method includes a first teacher data acquisition step for acquiring teacher data including one or more pairs of test data at the time of initial consultation and clinical outcome of a patient suffering from a target disease, a first learning step for using the test data at the time of initial consultation as input and training a first prediction model that predicts the clinical outcome by referring to the teacher data, a first test data acquisition step for acquiring test data at the time of initial consultation of the patient to be predicted, and inputting the test data acquired in the first test data acquisition step into the first prediction model. Therefore, the second information processing method is a method for predicting the clinical outcome of a disease, comprising: a first prediction step of predicting the subsequent clinical outcome of the patient to be predicted; a second training data acquisition step of acquiring training data including one or more pairs of test data from the patient suffering from the target disease during hospitalization and the clinical outcome; a second learning step of inputting test data from multiple time points during hospitalization and training a second prediction model that predicts the clinical outcome by referring to the training data; a second test data acquisition step of acquiring test data from multiple time points during hospitalization of the patient to be predicted; and a second prediction step of predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the second test data acquisition step into the second prediction model.

[0146] A thirteenth aspect of the present invention provides a method for predicting a clinical outcome of a disease, comprising: a first information processing method for predicting a subsequent clinical outcome of a patient to be predicted based on test data at the time of the patient's initial visit; and a second information processing method for predicting a subsequent clinical outcome of the patient to be predicted based on test data during hospitalization, wherein the first information processing method comprises: a first test data acquisition step for acquiring test data at the time of the patient to be predicted based on test data at the time of the patient's initial visit; and a first prediction step for predicting a subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the test data acquisition step into a first prediction model that predicts the clinical outcome, using the test data at the time of the patient's initial visit as an input; and the second information processing method comprises: a second test data acquisition step for acquiring test data at multiple time points during the patient's hospitalization, and a second prediction step for predicting a subsequent clinical outcome of the patient to be predicted based on the test data acquired in the test data acquisition step, using the test data at the multiple time points during hospitalization as an input into a second prediction model that predicts the clinical outcome.

[0147] The disease clinical outcome prediction program according to aspect 14 of the present invention is a program that causes an information processing device to execute a second information processing method for predicting the subsequent clinical outcome of a patient to be predicted from test data during hospitalization, wherein the second information processing method includes: a teacher data acquisition step for acquiring teacher data including one or more pairs of test data during hospitalization of a patient suffering from the target disease and clinical outcomes; a learning step for inputting test data from multiple time points during hospitalization and training a prediction model that predicts the clinical outcome by referring to the teacher data; a test data acquisition step for acquiring test data from multiple time points during hospitalization of the patient to be predicted; and a prediction step for predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the test data acquisition step into the prediction model.

[0148] A disease clinical outcome prediction program according to aspect 15 of the present invention is a program that causes an information processing device to execute a first information processing method for predicting the subsequent clinical outcome of a patient to be predicted from test data at the time of initial consultation, wherein the first information processing method includes: a teacher data acquisition step for acquiring teacher data including one or more pairs of test data at the time of initial consultation and clinical outcome of a patient suffering from the target disease; a learning step for using the test data at the time of initial consultation as input and training a prediction model that predicts the clinical outcome by referring to the teacher data; a test data acquisition step for acquiring test data at the time of initial consultation of the patient to be predicted; and a prediction step for predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data acquired in the test data acquisition step into the prediction model.

[0149] A disease clinical outcome prediction program according to aspect 16 of the present invention is a program that causes an information processing device to execute a second information processing method for predicting the subsequent clinical outcome of a patient to be predicted from test data obtained during hospitalization, the second information processing method including: a test data acquisition step for acquiring test data obtained at multiple time points during the hospitalization of the patient to be predicted; and a prediction step for predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data obtained in the test data acquisition step into a prediction model that predicts the clinical outcome, using the test data obtained at multiple time points during hospitalization as input.

[0150] A disease clinical outcome prediction program according to aspect 17 of the present invention is a program that causes an information processing device to execute a first information processing method for predicting the subsequent clinical outcome of a patient to be predicted from test data at the time of the first visit, the first information processing method including: a test data acquisition step for acquiring test data at the time of the first visit of the patient to be predicted; and a prediction step for predicting the subsequent clinical outcome of the patient to be predicted by inputting the test data at the time of the first visit into a prediction model that predicts the clinical outcome, using the test data at the time of the first visit as input, and inputting the test data acquired in the test data acquisition step into a prediction model that predicts the clinical outcome.

[0151] The disease clinical outcome prediction program of aspect 18 of the present invention may be configured in aspect 10 or 11 to further include a third program that causes an information processing device to execute a third information processing method, which includes a step of suggesting a drug or therapeutic intervention method based on the clinical outcome of the patient predicted by the first information processing method and the second information processing method.

[0152] The method for predicting a clinical outcome of a disease according to aspect 18 of the present invention may be configured in accordance with aspect 12 or 13, further comprising a third information processing method including a step of suggesting a drug or therapeutic intervention method based on the clinical outcome of the patient predicted by the first information processing method and the second information processing method. [Example]

[0153] Study design and setting This study was conducted as a retrospective cohort study of 382 COVID-19 cases treated at the Department of Respiratory Medicine, Tokyo Shinagawa Hospital, between January and September 2020. A total of 201 outpatients and 181 inpatients were included as COVID-19 cases. To verify whether the dynamic mortality risk model is also effective for non-COVID-19 cases, data from 146 non-COVID-19 cases was also used. Note that the test data and severity prediction results based on oxygen support status in the following examples were acquired, predicted, or output by the various components of the control unit 11 of the information processing device 1, as an example.

[0154] Data were extracted from electronic medical records, including 84 variables (see Table 1) for prognostic screening at initial consultation, such as symptoms, background information, and blood and urine test data, and 93 variables (see Table 2) for dynamic risk assessment during hospitalization, such as blood test and vital information. Given that the severity and mortality of COVID-19 have been reported to be influenced by various factors, all variables with missing data of less than 15% were used in the analysis. A total of 265 blood and urine tests were performed at initial consultation, but most were not measured in non-hospitalized patients. Tests measured in more than half of non-hospitalized patients were selected to enable comparison by severity, resulting in 53 test items and an overall missingness criterion of 15%.

[0155] For prognostic screening at the initial visit, blood and urine test data obtained from 4 days before to 1 day after the initial visit were used. For dynamic risk assessment during hospitalization, the most recent test was used until the next test was performed. Missing values ​​at the initial visit and for dynamic risk assessment were imputed using the missForest package in the statistical analysis software R (Reference 1: Stekhoven DJ, Buhlmann P. MissForest—non-parametric missing value imputation for mixed-type data. Bioinformatics. 2012 Jan 1;28(1):112-8.).

[0156] [Exclusion criteria and data division] One patient under the age of 18 and one patient with unknown changes in severity during hospitalization or unknown clinical outcomes were excluded from the analysis. These included cases whose prognosis was unknown due to transfer or discontinuation of observation, and cases in which changes in severity could not be sufficiently extracted from the electronic medical records. Hospitalized patients with a difference of more than four days between the date of initial diagnosis and the date of admission were excluded from prognosis screening but were included in the dynamic risk assessment. The included cases were randomly divided into training and validation datasets in a 2:1 ratio to preserve the distribution of severity outcomes. The data division was consistent between the prediction at initial consultation and the dynamic risk prediction during hospitalization.

[0157] [Severity prediction] We adopted the oxygen support status score proposed by Grein et al. (Reference 2: Grein J, Ohmagari N, Shin D, et al. Compassionate Use of Remdesivir for Patients with Severe Covid-19. N Engl J Med 2020; 382: 2327-36.) to predict patient severity. The score was an ordinal scale of 1 to 6 based on the type of oxygenation level: 1 indicated no discharge or hospitalization, 2 indicated room air, 3 indicated low-flow oxygen support, 4 indicated noninvasive interventions including nasal high-flow oxygen therapy and / or noninvasive positive pressure ventilation (NIPPV), 5 indicated invasive interventions including invasive mechanical ventilation and / or ECMO, and 6 indicated death. In addition to the original score of 1 indicating hospital discharge status, outpatients were also assigned a score of 1.

[0158] [Machine learning model for predicting outcomes at initial consultation] Gradient-boosting decision trees (GBDTs) (Reference 3: Ke G, Meng Q, Finley T, et al. LightGBM: A Highly Efficient Gradient Boosting Decision Tree. In: Guyon I, Luxburg UV, Bengio S, et al., eds. Advances in Neural Information Processing Systems. Curran Associates, Inc., 2017.) were used to predict outcomes at initial consultation. GBDTs are an extension of random forests that utilize boosting techniques to improve predictive performance. They have the advantage of being able to handle missing data without imputation.

[0159] In the training phase, we used data imputed by missForest, but predictions were made on validation data containing missing values. Among various implementations of the GBDT algorithm, we used the Python package LightGBM due to its superior computing speed. As a preprocessing step, we excluded variables in the training data with an absolute Spearman correlation coefficient of ≥ 0.85, leaving only those variables showing the strongest correlation with severity. We repeated hyperparameter tuning by Bayesian optimization with cross-validation using the Optuna framework and training with the optimal parameters 100 times. The ensemble average of the models obtained from the iterations was used as the final prediction model. To evaluate the importance of variables in prediction, we used the Shapley Additive exPlanations (SHAP) value on the validation data.

[0160] [Random Survival Forest] We assessed the risk of hospitalization mortality using random survival forests (RSF) (Reference 4: Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS. Random survival forests. Ann. Appl. Stat. 2008;2:841-60). RSF calculates a final hazard function as an ensemble of hazard functions estimated by survival forests. Like random forests, RSF is robust to outliers and enables accurate risk assessment of event occurrence. For the analysis, we used the Python package scikit-survival, which includes an RSF implementation. To assess the dynamic risk of mortality, a 7-day cumulative hazard function, representing the risk of an event occurring within 7 days, was calculated for each patient for each day of hospitalization. The variable importance of the RSF model was calculated on the training data using the permutation importance function implemented in the Python package eli5. The final prediction model was created by retraining using only the top 10 important variables.

[0161] [ethics] This study was approved by the local Institutional Review Boards of the RIKEN Institute and Tokyo Shinagawa Hospital (approval number 20-A-06). It was conducted in accordance with the ethical standards of the 1964 Declaration of Helsinki and its later amendments. The requirement for informed consent to participate was waived due to the retrospective nature of this study.

[0162] [result] The median age of the 382 COVID-19 patients was 39 years; 233 were male and 149 were female (Table 3). Of these, 201 were outpatients who did not require hospitalization, and the remaining 181 were hospitalized. The median length of hospitalization for hospitalized patients was 11 days. Of the 51 hospitalized patients who required oxygen support, 30 required low-flow oxygen, 3 required high-flow oxygen, 8 were treated with invasive ventilation / ECMO, and 10 died.

[0163] [Table 3]

[0164] (Screening for severity and prognosis of COVID-19 at initial consultation) To examine whether the outcome is already determined at the time of initial consultation, we constructed a prognostic model for COVID-19 severity based on information available at the time of initial consultation (symptoms, background information, blood and urine markers).

[0165] Specifically, by inputting test data of the patient to be predicted at the time of the first visit and executing a prediction process using the information processing device 1 shown in Fig. 1, a prediction result for severity prediction based on the oxygen support state level proposed by Grein et al. was obtained (prediction step). Based on this prediction result, the maximum level of severity that the patient would reach (i.e., severity outcome) was predicted.

[0166] Figure 12 shows the results of screening for severity prognosis at initial visit. Data 1201 in Figure 12 is the receiver operating characteristic (ROC) curve and its area under the curve (AUC) for a prediction model of severity outcome based on the level of oxygen support status. The 95% confidence interval for the AUC is shown in parentheses. Data 1202 in Figure 12 is a box and jitter plot showing the predicted probability that a validation case will be determined to have a severity above the threshold indicated by the dashed line.

[0167] The area under the receiver operating characteristic curve (AUC) was 0.717 for severity outcomes ≥2 (hospitalization), 0.878 for severity outcomes ≥3 (oxygen administration), 0.951 for severity outcomes ≥4 (noninvasive ventilation), 0.952 for severity outcomes ≥5 (invasive ventilation), and 0.970 for severity outcomes ≥6 (death) (data 1201 in Figure 12).

[0168] We investigated the agreement between predicted probabilities and actual severity outcomes. Results showed that most patients with an actual severity rating of 3 or higher had a high probability of achieving a rating of ≥2, whereas patients with actual severity ratings of 1 and 2 showed a wide distribution of predicted probabilities, ranging from low to high, with no clear distinction between them (top panel of data 1202 in Figure 12). For other predictions, there was generally good agreement between predicted probabilities and actual severity, but there were some cases in which the risk of severity was not properly determined, such as patients with mild symptoms but a high probability of severe symptoms (data 1202 in Figure 12).

[0169] Next, we evaluated the variable importance of each prediction. Figure 13 shows the results of evaluating variable importance in predicting each severity level of COVID-19 using SHAP values. Data 1301 in Figure 13 shows the variable importance for each prediction model evaluated by SHAP values. The average absolute SHAP values ​​of individual patients in the validation dataset are shown as a bar graph. Data 1302 in Figure 13 is a heat map showing symptoms observed at initial consultation.

[0170] The presence of pneumonia was identified as a significant predictor of whether a severity rating of ≥2 was achieved (Figure 13, data 1301, leftmost panel). Most patients admitted to the hospital with a severity rating of ≥2 had pneumonia at initial consultation, and all patients with a severity rating of ≥3 had pneumonia (Figure 13, data 1302). Conversely, as the target severity level of prediction increased, the importance of symptoms decreased, but blood test parameters such as lymphocyte count (Lympho-Count), prothrombin time (PT), C-reactive protein (CRP), creatinine (Cre), and amylase (AMY) remained significant predictors (Figure 13, data 1301). Body mass index (BMI) was also a significant predictor of whether a severity rating of ≥4 or ≥5 was achieved. Age was significant for all severity levels.

[0171] Among the important predictors, we examined the distribution of age and blood test results according to the actual severity prognosis. Figure 14 shows box and jitter plots showing the distribution of significant continuous variables by severity group. As shown in Figure 14, age and BMI were higher in patients with severity level 3 or higher, and blood test results such as PT, CRP, creatinine (Cre), red blood cell distribution width (RDW), and blood glucose (GLU) were higher in patients with severity level 4 or higher. These blood test results, including lymphocyte count (Lympho-Count), estimated glomerular filtration rate (eGFR) minus creatinine (eGFR-Cre), and albumin (Alb), were lower in more severe cases, with platelet counts being particularly low in fatal cases. Amylase (AMY) showed a unique distribution, being low in severe patients with severity levels 4, 5, and 6, but extremely elevated in some patients with severity level 6.

[0172] (Dynamic mortality risk assessment based on longitudinal hospitalization data) Outcome screening based on information from the initial visit was highly accurate but incomplete. Therefore, we hypothesized that a patient's prognosis is not completely determined at the initial visit and that prognosis may change depending on the course and treatment after admission. Based on this hypothesis, we examined changes in severity status after admission. Figure 15 shows the changes in severity status during hospitalization by severity outcome group, showing the trend in the number of patients receiving treatments such as ventilation and oxygen administration at specific time points after admission. As shown in Figure 15, while oxygen administration and noninvasive ventilation were initiated within five days of admission, invasive ventilation was often introduced more than five days after admission, and death occurred more than 20 days after admission. Furthermore, many deaths occurred without the application of invasive ventilation (top panel of Figure 15). This is because most fatal cases were elderly patients who were not eligible for invasive ventilation or ECMO even if their condition worsened. These observations suggest that during the weeks between admission and death or discharge, patients undergo changes in their condition that cannot be adequately assessed by oxygen support status.

[0173] Next, we used random survival forest (RSF) (Reference 4) to assess the risk of death for hospitalized patients. The analysis included 93 variables, including background factors such as age and BMI, as well as blood test and vital signs measured over time during hospitalization. The risk of death was assessed using the 7-day cumulative hazard function (CHF) calculated using RSF as an index.

[0174] Specifically, test data of the patient measured over time during hospitalization at multiple points in time was used as input, and prediction processing was performed using the information processing device 1 shown in Figure 1 to obtain prediction results for severity prediction based on the oxygen support status level proposed by Grein et al. (prediction step). The risk of death was evaluated using a 7-day cumulative hazard function (CHF) calculated using this prediction result as an index.

[0175] Figure 16 shows the results of assessing the risk of death in COVID-19 patients using the 7-day cumulative hazard function (CHF) for mortality calculated using RSF as an index. As shown in Figure 16, the risk of death for each patient changed with each day of hospitalization. The risk of death is shown as the 7-day cumulative hazard function (CHF) calculated using RSF, i.e., the risk of death within 7 days. The filled circles for fatal cases indicate the last observation date. For all four fatal cases in the validation database, an increase in the risk of death was observed approximately one week after hospitalization, with CHF at the time of death being approximately 0.4. Conversely, in patients who recovered from invasive or noninvasive ventilation, CHF increased approximately one week after hospitalization, similar to the fatal cases, but then decreased, with the maximum CHF never exceeding 0.2. In mild cases that progressed with oxygen administration or room air, there was little increase in CHF.

[0176] The results of analyzing the variables that contribute to RSF prediction are shown in Figure 17. In Figure 17, the important variables that contributed to the RSF prediction of mortality were used to calculate the variable importance using permutation importance for the training data. As shown in Figure 17, factors related to blood coagulation, such as platelets (Plt) and prothrombin time (PT), were the most important in predicting mortality.

[0177] Figure 18 shows the results of confirming the agreement between RSF predictions and actual clinical progress in severe cases. Figure 18 shows the changes in RSF mortality risk and important predictors during hospitalization for six representative severe cases from the validation data, including two ECMO-cured cases and four deaths. Oxygen support status is shown between the CHF plot and the plot of important RSF variables. The shaded area in the graph represents the reference interval.

[0178] As shown in Figure 18, different patients had different combinations of factors associated with changes in CHF. For example, in patient #2, amylase was associated with elevated CHF levels, and in patient #4, elevated PT levels, along with amylase, occurred along with elevated CHF levels (Figure 18). Decreased platelets were observed in patients #4 and #6, and elevated β-D-glucan levels were observed in patients #3 and #6. Decreased blood calcium levels, increased neutrophil counts (Neutro%), and decreased monocyte counts (Mono%) were observed as changes common to all critically ill patients.

[0179] (Verification of the RSF model for dynamic mortality risk assessment) A machine learning model based on blood test data for assessing the risk of death from COVID-19 was proposed in Nature Machine Intelligence (NMI) 2020 (Non-Patent Document 3: Yan L, Zhang HT, Goncalves J, et al. An interpretable mortality prediction model for COVID-19 patients. Nature Machine Intelligence 2020; 2: 283-8.). This model (the NMI model) was constructed using only samples taken immediately before death or hospital discharge, and identified lactate dehydrogenase (LDH), lymphocytes, and CRP as key features. Therefore, as a comparative example, we assessed the risk of death based on these features identified by the NMI model.

[0180] Figure 19 shows the results of a comparative example, evaluating the risk of death from COVID-19 based on the LDH, lymphocytes, and CRP identified by the NMI model. As with Figure 18, we evaluated the changes in three predictors and the resulting risk of death for six severe cases in the validation data, including two ECMO-cured cases and four deaths. The dashed lines in the graph in Figure 19 represent the thresholds for the predictors proposed by the NMI model, and the shaded areas represent the reference intervals. The bar above the graph indicates the mortality prediction based on the NMI model.

[0181] As shown in Figure 19, in fatal cases, LDH levels were elevated immediately before death, indicating a high risk. However, even in fatal cases, there were cases in which LDH levels did not increase except immediately before death. In addition, patient #2 had elevated LDH levels, and was always considered to be at high risk of death, even though he was an ECMO survivor.

[0182] The CRP threshold proposed by the NMI model (comparison example) was very high (41.2 mg / L), and no cases in this study exceeded this threshold. On the other hand, in severe cases, lymphocytes were always below the threshold of the NMI model (14.7%) from the start of hospitalization.

[0183] The accuracy of the NMI model (comparative example) was compared with that of the RSF model of this example, and the results are shown in Figure 20. Figure 20 shows the results of comparing the performance of the RSF model of this example with the NMI model (comparative example). For 61 COVID-19 cases in the validation data, the F1 scores for discriminating between cured and fatal cases are shown by days until outcome. The bar plot shows the actual number of patients used to evaluate performance, which decreases as the number of days until outcome increases because some patients were not yet hospitalized.

[0184] As shown in Figure 20, one day before the outcome, both models perfectly predicted mortality with an F1-score of 1.0, but two or more days before the outcome, the accuracy of the NMI model (comparison example) dropped sharply to an F1-score of approximately 0.5. Conversely, the accuracy of the RSF model of this example did not drop until two weeks before the outcome, suggesting that the RSF model of this example can distinguish between fatal and cured cases well before the outcome.

[0185] Finally, we tested whether the RSF model for predicting mortality is also effective for non-COVID-19 cases. Figure 21 shows the results of examining the effectiveness of the RSF model for predicting mortality and hospital discharge for non-COVID-19 cases. The graph shows the change in mortality risk for 143 non-COVID-19 cases, including 9 deaths and 134 recovery cases. The filled circle for the deaths indicates the last observation date.

[0186] As shown in Figure 21, even in non-COVID-19 cases, CHF levels were elevated in approximately half of fatal cases, exceeding 0.2 at the time of death, whereas the remaining half of fatal cases did not show significant elevations. Among non-COVID-19 cured patients, several showed transiently elevated CHF levels early in hospitalization, but most were discharged without their CHF levels exceeding 0.2. Of the 19 non-COVID-19 cases with CHF levels of ≥ 0.2, 17 were due to infectious diseases such as bacterial pneumonia, bacterial prostatitis, and urinary tract infection, which were significantly more frequent than other diseases (odds ratio = 14.2, P value by Fisher's exact test = 0.00070, Table 4).

[0187] [Table 4]

[0188] [Consideration] The prognostic screening model at initial visit yielded different performance and predictors for each prognostic outcome. Pneumonia was the most important predictor of hospitalization, suggesting that hospitalization is determined by physical examination. Pneumonia has been reported as a risk factor for worsening of the condition, and in the study conducted in this example, all moderate to severe cases requiring oxygen administration had pneumonia at initial visit (data 1302 in Figure 13). Although the prognostic model cannot adequately distinguish between patients with severity level 1 and patients with severity level 2, physical examination is considered to be useful as a screening tool for potentially severe cases.

[0189] Among the important factors predicting oxygen administration, CRP was included in the 4C Mortality Score (Reference 5: Knight SR, Ho A, Pius R, et al. Risk stratification of patients admitted to hospital with COVID-19 using the ISARIC WHO Clinical Characterization Protocol: development and validation of the 4C Mortality Score. BMJ 2020; 370: m3339.). CRP has been reported to be associated with the extent of lung disease, supporting its importance for early identification of moderate to severe cases that may require oxygen administration. However, CRP is not particularly important in distinguishing between severity outcomes 4, 5, and 6 (data 1301 in Figure 13). While a clear difference in CRP levels was observed between severity levels 3 and 4, there was little difference between severity levels 4 and above (Figure 14). This suggests that inflammation reflected by CRP is an entry point for severity, but that progression is primarily influenced by other factors.

[0190] Creatinine and amylase levels have been identified as important predictors of severe disease. COVID-19 has been reported to cause kidney damage, in addition to chronic kidney disease, which has been reported as a risk factor for severe disease. Amylase is primarily secreted by the pancreas and salivary glands, and previous studies have reported elevated blood amylase levels in severe COVID-19. Elevated blood amylase levels indicate damage to the tissues producing amylase or kidney damage related to clearance. Damage to tissues or organs, such as the kidneys, may be an important factor in the progression of COVID-19 severity.

[0191] For screening of deaths at first visit, RDW, albumin, and platelets were identified as the top predictors (data 1301 in Figure 13). RDW has been reported to be associated with COVID-19 mortality. RDW is elevated in cardiovascular disease, cancer, and infectious diseases, and it has been reported that red blood cell production slows as white blood cell and platelet production increases. Decreased blood albumin levels have also been reported to be a predictor of COVID-19 mortality. Coagulation disorders, including thrombocytopenia, are common in COVID-19, and disseminated intravascular coagulation (DIC) has been observed in many COVID-19 deaths. Therefore, these findings suggest abnormalities in hematopoiesis and blood coagulation as early signs of COVID-19 mortality.

[0192] An information processing device according to one embodiment of the present invention can perform risk assessment for COVID-19 mortality prediction, taking into account the time from measurement to event occurrence, using machine learning-based survival analysis, enabling accurate estimation of mortality risk two weeks before the outcome. Several RSF predictors that fluctuate early in mortality were identified. Platelet count and prothrombin time were the top two early predictors in the RSF model for COVID-19 mortality. As previously mentioned, COVID-19 patients are prone to both arterial and venous thrombosis, and in severe cases, DIC can complicate the patient's condition and lead to death. While DIC was observed in only 0.6% of COVID-19 survivors, it occurred very frequently in 71.4% of fatal cases. The results of this example support the idea that coagulation abnormalities that can lead to DIC are specific early predictors of COVID-19 mortality. Predictors in dynamic risk assessment can partially explain the validity of the RSF model for non-COVID-19 cases. Most of the non-COVID-19 cases used for comparison in this example were caused by respiratory tract bacterial infections, and severe cases developed sepsis and DIC. Thus, as shown in this example, platelets and prothrombin time may equally predict severity and mortality in non-COVID-19 infections.

[0193] The two models constructed in this study will be utilized in various aspects of COVID-19 treatment. A highly sensitive prognostic model at initial consultation, which screens potentially severe cases, could be applied to estimate the number of hospital beds and the required number of medical devices, such as ventilators and ECMO devices. Furthermore, due to the recent shortage of hospital beds, the number of patients receiving home care has increased, raising concerns about sudden exacerbations during home care. Using a prognostic screening model at initial consultation may reduce the number of severe cases receiving home care and ensure safer home care. The RSF model of mortality may be useful for early detection and intervention of patients at risk of death. Coagulation factors, which are important predictors of mortality in the RSF model, are potent targets for early intervention. Heparin administration has been reported to improve the prognosis of severe DIC cases in COVID-19. The RSF model can be used to determine the appropriate timing of anticoagulation therapy and evaluate the therapeutic effect over time. [Industrial Applicability]

[0194] The present invention can be used in various aspects of the treatment of any disease, such as early detection and intervention of patients at risk of death, advance preparation of medical resources, etc. It is also expected that the present invention will contribute to the development of more efficient treatments and therapeutic strategies to reduce mortality from COVID-19 and other infectious diseases, including emerging infectious diseases that may arise in the post-COVID-19 era. [Explanation of symbols]

[0195] 1, 1a, 1b, 1c, 1d Information processing device 11, 11a, 11b, 11c, 11d Control unit 111 Data Acquisition Unit 112 Learning Department 113 Prediction Department 114 Output section 115 Specific section 116 Proposal Department 117 Extraction part 1000,1000a,1000b,1000c,1000d Disease Clinical Outcome Prediction System M1,M2,M3,M4,M5 Information processing method M11 Teacher data acquisition step M12 Learning Steps M21 Inspection data acquisition step M22 prediction step M31 Specific Step M32 Proposed Step M41 Specific Step M42 Extraction Step M51 Prediction result acquisition step M52 Proposed Step

Claims

1. a first program for causing an information processing device to execute a first information processing method for predicting a clinical outcome of a patient to be predicted from test data at the time of the first visit; a second program for causing an information processing device to execute a second information processing method for predicting a clinical outcome of a patient to be predicted from test data during hospitalization; An information processing device that executes The first information processing method includes: A first teacher data acquisition step of acquiring teacher data including a plurality of pairs of test data at the first visit of a patient suffering from a target disease and a clinical outcome; a first learning step of learning a first prediction model that predicts the clinical outcome by inputting test data at the first visit and referring to the training data; A first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a clinical outcome of the patient to be predicted by inputting the test data acquired in the first test data acquisition step into the first prediction model; Including, The second information processing method includes: A second teacher data acquisition step of acquiring teacher data including a plurality of pairs of test data during hospitalization of a patient suffering from a target disease and a clinical outcome; a second learning step of learning a second prediction model that predicts the clinical outcome using test data at multiple time points during hospitalization as an input, with reference to the training data; A second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a clinical outcome of the patient to be predicted by inputting the test data acquired in the second test data acquisition step into the second prediction model; Contains A system for predicting clinical outcomes of disease.

2. a first program for causing an information processing device to execute a first information processing method for predicting a clinical outcome of a patient to be predicted from test data at the time of the first visit; a second program for causing an information processing device to execute a second information processing method for predicting a clinical outcome of a patient to be predicted from test data during hospitalization; An information processing device that executes The first information processing method includes: A first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a clinical outcome of the patient to be predicted by inputting test data at the time of the first visit into a first prediction model that predicts the clinical outcome, the test data acquired in the first test data acquisition step; Including, The second information processing method includes: A second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a clinical outcome of the patient to be predicted by inputting test data acquired in the second test data acquisition step into a second prediction model that inputs test data at multiple time points during hospitalization and predicts the clinical outcome; Contains A disease clinical outcome prediction system.

3. The disease clinical outcome prediction system according to claim 1 or 2, wherein the second prediction model is trained using a random survival forest.

4. The disease clinical outcome prediction system according to claim 1 or 2, wherein the first prediction model is trained using a gradient boosting decision tree.

5. The first information processing method includes: a first identification step of identifying an item that has a greater contribution to the prediction process of the clinical outcome among a plurality of items included in the test data in the first prediction step; Further comprising: The second information processing method includes: a second identification step of identifying an item that has a greater contribution to the prediction process of the clinical outcome among a plurality of items included in the test data in the second prediction step; The clinical outcome prediction system for a disease according to any one of claims 1 to 4, further comprising:

6. The first information processing method includes: A first proposal step of proposing target molecule candidates for a therapeutic drug for the target disease based on the identification result in the first identification step. Further comprising: The second information processing method includes: A second proposal step of proposing candidate target molecules for a therapeutic drug for the target disease based on the identification results from the second identification step. The disease clinical outcome prediction system according to claim 5, further comprising:

7. The first information processing method includes: a first inspection data extraction step of extracting inspection data to be used in the first prediction step from the inspection data acquired in the first inspection data acquisition step based on a result of the identification in the first identification step; Further comprising: The second information processing method includes: a second inspection data extraction step of extracting inspection data to be used in the second prediction step from the inspection data acquired in the second inspection data acquisition step based on the identification result in the second identification step; The disease clinical outcome prediction system according to claim 5, further comprising:

8. 8. The disease clinical outcome prediction system of claim 7, wherein in the first test data extraction step and the second test data extraction step, the top 10 to 20 items are extracted from the test data in order of variable importance in the first prediction model or the second prediction model.

9. The disease clinical outcome prediction system according to any one of claims 1 to 8, wherein the information processing device further executes a third program that causes the information processing device to execute a third information processing method, the third information processing method including a step of suggesting a drug or therapeutic intervention method based on the clinical outcome of the patient predicted by the first information processing method and the second information processing method.

10. a first program for causing an information processing device to execute a first information processing method for predicting a clinical outcome of a patient to be predicted from test data at the time of the first visit; a second program for causing an information processing device to execute a second information processing method for predicting a clinical outcome of a patient to be predicted from test data during hospitalization; Including, The first information processing method includes: A first teacher data acquisition step of acquiring teacher data including a plurality of pairs of test data at the first visit of a patient suffering from a target disease and a clinical outcome; a first learning step of learning a first prediction model that predicts the clinical outcome by inputting test data at the first visit and referring to the training data; A first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a clinical outcome of the patient to be predicted by inputting the test data acquired in the first test data acquisition step into the first prediction model; Including, The second information processing method includes: A second teacher data acquisition step of acquiring teacher data including a plurality of pairs of test data during hospitalization of a patient suffering from a target disease and a clinical outcome; a second learning step of learning a second prediction model that predicts the clinical outcome using test data at multiple time points during hospitalization as an input, with reference to the training data; A second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a clinical outcome of the patient to be predicted by inputting the test data acquired in the second test data acquisition step into the second prediction model; Contains A program to predict clinical outcomes of disease.

11. a first program for causing an information processing device to execute a first information processing method for predicting a clinical outcome of a patient to be predicted from test data at the time of the first visit; a second program for causing an information processing device to execute a second information processing method for predicting a clinical outcome of a patient to be predicted from test data during hospitalization; Including, The first information processing method includes: A first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a clinical outcome of the patient to be predicted by inputting test data at the time of the first visit into a first prediction model that predicts the clinical outcome, the test data acquired in the first test data acquisition step; Including, The second information processing method includes: A second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a clinical outcome of the patient to be predicted by inputting test data acquired in the second test data acquisition step into a second prediction model that inputs test data at multiple time points during hospitalization and predicts the clinical outcome; Contains A program to predict clinical outcomes of disease.

12. A first information processing method for predicting a clinical outcome of a patient based on test data at the time of the first visit; A second information processing method for predicting a clinical outcome of a patient based on test data during hospitalization; Including, The first information processing method includes: A first teacher data acquisition step of acquiring teacher data including a plurality of pairs of test data at the first visit of a patient suffering from a target disease and a clinical outcome; a first learning step of learning a first prediction model that predicts the clinical outcome by inputting test data at the first visit and referring to the training data; A first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a clinical outcome of the patient to be predicted by inputting the test data acquired in the first test data acquisition step into the first prediction model; Including, The second information processing method includes: A second teacher data acquisition step of acquiring teacher data including a plurality of pairs of test data during hospitalization of a patient suffering from a target disease and a clinical outcome; a second learning step of learning a second prediction model that predicts the clinical outcome using test data at multiple time points during hospitalization as an input, with reference to the training data; A second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a clinical outcome of the patient to be predicted by inputting the test data acquired in the second test data acquisition step into the second prediction model; Contains Methods for predicting clinical outcomes of disease.

13. A first information processing method for predicting a clinical outcome of a patient based on test data at the time of the first visit; A second information processing method for predicting a clinical outcome of a patient based on test data during hospitalization; Including, The first information processing method includes: A first examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a first prediction step of predicting a clinical outcome of the patient to be predicted by inputting test data at the time of the first visit into a first prediction model that predicts the clinical outcome, the test data acquired in the first test data acquisition step; Including, The second information processing method includes: A second test data acquisition step of acquiring test data at a plurality of time points during hospitalization of the patient to be predicted; a second prediction step of predicting a clinical outcome of the patient to be predicted by inputting test data acquired in the second test data acquisition step into a second prediction model that inputs test data at multiple time points during hospitalization and predicts the clinical outcome; Contains Methods for predicting clinical outcomes of disease.

14. A program for causing an information processing device to execute a second information processing method for predicting a clinical outcome of a patient to be predicted from test data during hospitalization, The second information processing method includes: A teacher data acquisition step of acquiring teacher data including a plurality of pairs of test data during hospitalization of a patient suffering from a target disease and a clinical outcome; A learning step of inputting test data at multiple time points during hospitalization and learning a prediction model for predicting the clinical outcome by referring to the training data; A test data acquisition step of acquiring test data at multiple time points during hospitalization of the patient to be predicted; a prediction step of predicting a clinical outcome of the patient to be predicted by inputting the test data acquired in the test data acquisition step into the prediction model; A disease clinical outcome prediction program, including:

15. A program for causing an information processing device to execute a first information processing method for predicting a clinical outcome of a patient to be predicted from test data at the time of an initial visit, The first information processing method includes: A teacher data acquisition step of acquiring teacher data including a plurality of pairs of test data at the first visit of a patient suffering from a target disease and a clinical outcome; A learning step of inputting test data at the first visit and learning a prediction model for predicting the clinical outcome by referring to the training data; An examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a prediction step of predicting a clinical outcome of the patient to be predicted by inputting the test data acquired in the test data acquisition step into the prediction model; A disease clinical outcome prediction program, including:

16. A program for causing an information processing device to execute a second information processing method for predicting a clinical outcome of a patient to be predicted from test data during hospitalization, The second information processing method includes: A test data acquisition step of acquiring test data at multiple time points during hospitalization of the patient to be predicted; A prediction step of predicting a clinical outcome of the patient to be predicted by inputting test data acquired in the test data acquisition step into a prediction model that predicts the clinical outcome, using test data acquired at multiple time points during hospitalization as an input; A disease clinical outcome prediction program, including:

17. A program for causing an information processing device to execute a first information processing method for predicting a clinical outcome of a patient to be predicted from test data at the time of an initial visit, The first information processing method includes: An examination data acquisition step of acquiring examination data at the first visit of the patient to be predicted; a prediction step of predicting a clinical outcome of the patient to be predicted by inputting test data at the first visit into a prediction model for predicting the clinical outcome, the test data acquired in the test data acquisition step; A disease clinical outcome prediction program, including:

18. The disease clinical outcome prediction program according to claim 10 or 11, further comprising a third program for causing an information processing device to execute a third information processing method, the third information processing method including a step of proposing a drug or therapeutic intervention method based on the clinical outcome of the patient predicted by the first information processing method and the second information processing method.

19. The method for predicting a clinical outcome of a disease according to claim 12 or 13, further comprising a third information processing method including a step of suggesting a drug or therapeutic intervention method based on the clinical outcome of the patient predicted by the first information processing method and the second information processing method.

Images