Information processing device, information processing method, and computer program

Abstract

This information processing device is for predicting a prognosis of a subject patient affected by a disease, and comprises a model acquisition unit, a subject patient information acquisition unit, and a prognosis prediction execution unit. The model acquisition unit acquires a prognosis prediction model that is a machine learning model which uses, as an input, time-series information indicating time-series transition of factors of the disease, and as an output, the prognosis of the disease. The subject patient information acquisition unit acquires time-series information about the subject patient. The prognosis prediction execution unit executes prognosis prediction of the subject patient by using the time-series information about the subject patient and the prognosis prediction model, and outputs the result of the prognosis prediction.

Description

Information processing device, information processing method, and computer program

[0001] The technology disclosed in this specification relates to information processing for predicting the prognosis of a patient suffering from a disease.

[0002] Interstitial pneumonia is a general term for chronic, progressive fibrotic lung diseases. Acute exacerbation of interstitial pneumonia is a pathological condition characterized by a rapid deterioration of the condition within one month, and is associated with an extremely poor prognosis, with an in-hospital mortality rate of approximately 50%. If acute exacerbation of interstitial pneumonia could be predicted with high accuracy, it would be possible to suppress onset with antifibrotic drugs and improve prognosis through early diagnosis and therapeutic intervention, for example.

[0003] A clinical model has been proposed for predicting the risk of acute exacerbation in patients with idiopathic pulmonary fibrosis, a type of interstitial pneumonia (see, for example, Non-Patent Document 1).

[0004] Qi Wu and five others, "A Clinical Model for the Prediction of Acute Examination Risk in Patients with Idiopathic Pulmonary Fibrosis," BioMed Research International, Hindawi, 2020, pp. 1-6

[0005] The progression of interstitial pneumonia varies from patient to patient, and the patient's condition changes over time. The conventional prediction models described above do not take into account the chronological changes in disease factors, including the patient's condition, resulting in low prediction accuracy. This issue is not limited to prediction of acute exacerbation of interstitial pneumonia, but is common to disease prognosis prediction in general.

[0006] This specification discloses a technique that can solve the above-mentioned problems.

[0007] The technology disclosed in this specification can be realized, for example, in the following forms.

[0008] (1) The information processing device disclosed in this specification is a device for predicting the prognosis of a target patient suffering from a disease, and includes a model acquisition unit, a target patient information acquisition unit, and a prognosis prediction execution unit. The model acquisition unit acquires a prognosis prediction model, which is a machine learning model that receives time-series information indicating chronological changes in disease factors and outputs the prognosis of the disease. The target patient information acquisition unit acquires time-series information about the target patient. The prognosis prediction execution unit uses the time-series information about the target patient and the prognosis prediction model to predict the prognosis of the target patient and outputs the prognosis prediction result.

[0009] According to this information processing device, it is possible to predict the prognosis of a disease for each individual patient based on time-series information showing the time-series changes in disease factors, and to predict the prognosis of the disease with high accuracy.

[0010] (2) In the information processing device, the disease factors may include environmental factors. By adopting this configuration, the prognosis of a disease can be predicted with higher accuracy by using information indicating time-series changes in environmental factors that may have a significant impact on the prognosis of the disease.

[0011] (3) In the information processing device, the disease factors may include environmental factors of the patient's residence. By adopting this configuration, it is possible to predict the prognosis of the disease with higher accuracy by using information indicating time-series changes in factors related to the environment to which the patient is primarily exposed, compared to using environmental factors of the location of the hospital to which the patient is visiting, for example.

[0012] (4) In the information processing device, the environmental factors of the patient's residence may be environmental factors of a location within a straight-line distance of 200 km from the patient's current address. By using environmental factors of a location relatively close to the patient's current address, information that more accurately indicates the time-series changes in factors related to the environment to which the patient is primarily exposed can be used, enabling the prognosis of the disease to be predicted with extremely high accuracy.

[0013] (5) In the information processing device, the environmental factors may include at least one of the presence of environmental pollutants and meteorological parameters. By adopting this configuration, the prognosis of a disease can be predicted with higher accuracy by using information indicating time-series changes in environmental factors that may have a significant impact on the prognosis of the disease.

[0014] (6) In the information processing device, the time-series information may include information indicating a change in the environmental factor. By adopting this configuration, the prognosis of a disease can be predicted with higher accuracy by using the information indicating a change in the environmental factor that may have a significant impact on the prognosis of the disease.

[0015] (7) In the information processing device, the time-series information may be information that identifies the values ​​of the disease factors at regular time intervals. By adopting this configuration, it is possible to predict the prognosis of the disease with higher accuracy than when information that identifies the values ​​of the disease factors irregularly is used.

[0016] (8) In the information processing device, the time-series information may include information specifying at least monthly values ​​of the disease factors. By adopting this configuration, the prognosis of the disease can be predicted with higher accuracy by using the time-series information indicating monthly changes in the disease factors.

[0017] (9) In the information processing device, the prognosis of the disease may include the occurrence of multiple events that are in a competing risk relationship, and the prognosis prediction model may be a model trained using a machine learning algorithm corresponding to the multiple events that are in a competing risk relationship. By adopting this configuration, it is possible to predict the occurrence of multiple events that are in a competing risk relationship with high accuracy.

[0018] (10) In the information processing device, the prognosis prediction model may be configured to output, as the prognosis of the disease, an index value representing the likelihood of an event occurring for a plurality of events in a competing risk relationship. By adopting this configuration, it is possible to predict the occurrence of a plurality of events in a competing risk relationship with high accuracy.

[0019] (11) In the information processing device, the plurality of events having a competing risk relationship may include an acute exacerbation and death. By adopting this configuration, it is possible to predict with high accuracy the occurrence of each of the acute exacerbation and death having a competing risk relationship.

[0020] (12) In the information processing device, the disease may be a respiratory or circulatory disease. By adopting this configuration, it is possible to predict the prognosis of a respiratory or circulatory disease with high accuracy using time-series information that indicates a chronological change in factors of the respiratory or circulatory disease.

[0021] (13) In the information processing device, the prognosis prediction execution unit may be configured to execute a virtual prognosis prediction for the target patient using virtual information obtained by changing a part of the time-series information for the target patient and the prognosis prediction model, predict an effect of an intervention corresponding to the change based on the prediction result of the actual prognosis and the prediction result of the virtual prognosis, and output the prediction result of the effect of the intervention. By adopting this configuration, it is possible to determine whether or not to actually perform the intervention based on the output prediction result of the effect of the intervention.

[0022] The technology disclosed in this specification can be realized in various forms, such as an information processing device, an information processing method, a computer program that realizes those methods, a non-transitory recording medium on which that computer program is recorded, etc.

[0023] 1. An explanatory diagram conceptually showing the prognosis prediction model MO in this embodiment. 2. An explanatory diagram showing a method for predicting the effect of intervention based on the prognosis prediction result. 3. An explanatory diagram showing the general configuration of the information processing device 100. 4. A flowchart showing the process of acquiring the prognosis prediction model in this embodiment. 5. An explanatory diagram showing specific examples of factors (features) of interstitial pneumonia. 6. An explanatory diagram conceptually showing the interpolation process for the original information Io. 7. An explanatory diagram showing an example of learning data LD obtained through preprocessing. 8. An explanatory diagram conceptually showing a model using LSTM. FIG. 1 is another explanatory diagram showing the prediction accuracy of the prognosis prediction model MO of the embodiment. FIG. 2 is an explanatory diagram showing the prediction accuracy of the prognosis prediction model MO of the embodiment. FIG. 3 is an explanatory diagram showing an example of a prognosis prediction result by the prognosis prediction model MO of the embodiment. FIG. 4 is an explanatory diagram showing the prediction accuracy of the prognosis prediction model MO of another embodiment. FIG. 5 is an explanatory diagram showing an example of the importance of each disease factor in the embodiment. FIG. 6 is an explanatory diagram showing an example of the importance of each disease factor in the other embodiment. FIG. 7 is an explanatory diagram showing the relationship between the upper limit of the straight-line distance from the patient's current address to the measurement station and the prediction accuracy.

[0024] A. Embodiments: A-1. Overview of Prognosis Prediction Model MO: First, an overview of the prognosis prediction model MO in this embodiment will be described. FIG. 1 is an explanatory diagram conceptually showing the prognosis prediction model MO in this embodiment. The prognosis prediction model MO is a model for predicting the prognosis of a patient suffering from a disease. The prognosis prediction model MO is a machine learning model that inputs time-series information showing the time-series changes in disease factors (feature amounts) and outputs the prognosis of the disease. Note that in this specification, machine learning refers to a general term for techniques and methods for finding rules and patterns by using a computer to learn based on large amounts of data (i.e., data-driven), and includes deep learning.

[0025] In this embodiment, interstitial pneumonia is used as a specific example of a disease. Events of interstitial pneumonia include acute exacerbation and death. Because an acute exacerbation event does not occur after a death event occurs, the two events can be said to be in a competing risk relationship.

[0026] Factors (features) of interstitial pneumonia include, for example, patient background (smoking status, BMI, etc.), test findings (blood test, chest CT image, etc.), environmental factors (NO ２ , P.M. ２．５ These include environmental pollutants such as serotonin, meteorological parameters such as temperature, and treatment information (administration of antifibrotic agents, etc.). Time-series information showing the time-series changes in these factors is, for example, information that specifies the values ​​and amounts of change of these factors at regular time intervals (e.g., monthly) over a certain period (e.g., the period from the time of initial diagnosis to the Mth month).

[0027] As a prognosis prediction using the prognosis prediction model MO, for example, an index value representing the possibility of each event occurring is calculated. In the example shown in FIG. 1 , the probability of acute exacerbation and death occurring at a certain timing (e.g., M+N months) is calculated as a prognosis prediction. However, prognosis prediction may be performed in other manners. For example, as a prognosis prediction, a patient's predicted state (survival, acute exacerbation, death) may be classified based on the probability of acute exacerbation and death occurring at a certain timing.

[0028] By using the prognosis prediction model MO of this embodiment, it is possible to accurately predict the prognosis of a disease for an individual patient based on time-series information showing the time-series changes in disease factors. The prognosis prediction results can be used for various purposes. For example, for a patient predicted to have a high probability of developing acute exacerbation, it is possible to administer an anti-fibrotic drug to suppress the onset of the disease, or to perform early diagnosis and therapeutic intervention to improve the prognosis.

[0029] Furthermore, as shown in FIG. 2 , the effect of intervention can be predicted based on the prognosis prediction results. The upper part of FIG. 2 shows an example of a prediction result of the probability of occurrence of acute exacerbation and death based on actual time-series information. The lower part of FIG. 2 shows an example of a hypothetical prediction result of the probability of occurrence of acute exacerbation and death based on virtual information in which part of the time-series information is modified to correspond to an expected intervention (e.g., smoking cessation, medication, rehabilitation, nutritional therapy, etc.). Based on the prognosis prediction result based on the actual time-series information and the prognosis prediction result based on the virtual information, the effect of intervention corresponding to the above modification (e.g., a XX% reduction in the risk of acute exacerbation and a YY% reduction in the risk of death) can be predicted. Based on the output predicted result of the effect of intervention, a decision can be made as to whether or not to actually perform intervention.

[0030] A-2. Configuration of information processing device 100: Next, the configuration of the information processing device 100 for creating the prognosis prediction model MO and executing prognosis prediction using the prognosis prediction model MO will be described. Figure 3 is an explanatory diagram showing a schematic configuration of the information processing device 100. The information processing device 100 is configured by a computer (PC, server, etc.).

[0031] The information processing device 100 includes a control unit 110, a storage unit 120, a display unit 130, an operation input unit 140, and an interface unit 150. These units are connected to each other so as to be able to communicate with each other via a bus 190. The information processing device 100 may also include a speaker as an output means.

[0032] The display unit 130 of the information processing device 100 is configured, for example, by a liquid crystal display or the like, and displays various images and information. The operation input unit 140 is configured, for example, by a keyboard, mouse, buttons, a microphone, a trackpad, or the like, and accepts operations and instructions from an administrator. The display unit 130 may also function as the operation input unit 140 by being equipped with a touch panel. The interface unit 150 is configured, for example, by a LAN interface, a USB interface, or the like, and communicates with other devices via wired or wireless connections.

[0033] The storage unit 120 of the information processing device 100 is configured, for example, with a ROM, RAM, hard disk drive (HDD), etc., and is used to store various programs and data, and as a work area when executing various programs, and as a temporary storage area for data. For example, the storage unit 120 stores a prognosis prediction program CP, which is a computer program for executing the prognosis prediction model acquisition process and prognosis prediction process described below. The prognosis prediction program CP is provided in a state stored on a computer-readable recording medium (not shown), such as a CD-ROM, DVD-ROM, or USB memory, or is provided in a state that can be obtained from an external device (a server or other terminal device on a network) via the interface unit 150, and is stored in the storage unit 120 in a state that is operable on the information processing device 100.

[0034] Furthermore, learning data LD, prognosis prediction model MO, target patient information Ip, and prognosis prediction result data RD are stored in the storage unit 120 of the information processing device 100 in advance or during the execution of the prognosis prediction model acquisition process and prognosis prediction process described below. The contents of this information and data will be explained together with the explanation of the prognosis prediction model acquisition process and prognosis prediction process described below.

[0035] The control unit 110 of the information processing device 100 is configured with, for example, a CPU, and controls the operation of the information processing device 100 by executing a computer program read from the storage unit 120. For example, the control unit 110 reads a prognosis prediction program CP from the storage unit 120 and executes it, thereby functioning as a raw information acquisition unit 111, a learning data acquisition unit 112, a model acquisition unit 113, a target patient information acquisition unit 114, and a prognosis prediction execution unit 119 for executing the prognosis prediction model acquisition process and prognosis prediction process described below. The functions of each of these units will be described in conjunction with the prognosis prediction model acquisition process and prognosis prediction process described below.

[0036] A-3. Prognosis prediction model acquisition process: Next, the prognosis prediction model acquisition process executed by the information processing device 100 of this embodiment will be described. FIG. 4 is a flowchart showing the prognosis prediction model acquisition process in this embodiment. The prognosis prediction model acquisition process is a process for acquiring a prognosis prediction model MO, which is a machine learning model for predicting the prognosis of a patient suffering from a disease (interstitial pneumonia). In this embodiment, the information processing device 100 acquires the prognosis prediction model MO by creating the prognosis prediction model MO by using predetermined machine learning. The prognosis prediction model acquisition process is started in response to a start instruction being input by a user operating the operation input unit 140 of the information processing device 100.

[0037] First, the raw information acquisition unit 111 ( FIG. 3 ) of the information processing device 100 acquires information used to create the prognosis prediction model MO (hereinafter referred to as "raw information Io") (S110). The raw information Io is information that serves as the basis for the learning data LD used in training, verifying, and testing the prognosis prediction model MO. Specifically, the raw information Io is information that associates, for multiple patients suffering from interstitial pneumonia, time-series information that indicates the chronological changes in factors (features) of interstitial pneumonia with information indicating prognosis. The raw information Io is acquired via the interface unit 150 or the operation input unit 140.

[0038] 5 is an explanatory diagram showing specific examples of factors (feature amounts) of interstitial pneumonia. In this embodiment, 44 factors classified into four types, namely, patient background, test findings, environmental factors, and treatment information, are used as factors of interstitial pneumonia.

[0039] Patient background information includes, for example, the following 12 factors. Patient background information can be obtained, for example, by interview or examination: age, BMI, GAP score, current smoker, ex-smoker, IPF, PPFE, SSc, collagen disease, gender, CCI≧3, mMRC≧2

[0040] Test findings include, for example, the following 18 factors. Test finding information can be obtained, for example, from blood tests or chest CT images: LDH, BNP, WBC, neutrophils, lymphocytes, eosinophils, albumin, KL-6, SP-D, FVC (% pred), FEV １(% pred), DLco (% pred), 6-minute walk distance, PCO ２ , P.O. ２ , SpO ２ Minimum value, log CRP, CT image UIP pattern

[0041] The environmental factors are factors related to the environment of the patient's residence, and include, for example, the presence (e.g., concentration) of environmental pollutants and meteorological parameters. More specifically, the environmental factors include, for example, the following 10 factors: ２ , NO, SO ２ , P.M. ２．５ , SPM, precipitation, temperature, season (autumn, summer, winter)

[0042] Examples of environmental factors that can be used include representative values ​​for a certain period of time (e.g., monthly average values, daily average values), the amount of change for a certain period of time (e.g., the amount of change in the monthly average values, the amount of change in the daily average values), and / or the number of times that the standard value is exceeded for a certain period of time (e.g., the number of days per month that the environmental standard value is exceeded, the number of hours per day that the environmental standard value is exceeded).

[0043] The patient's place of residence may be the region to which the patient's current address belongs. In this case, information regarding environmental pollutants, which are environmental factors, can be obtained, for example, by referencing measurement data from a measurement station located in that region. Specifically, the measurement data from the measurement station closest in a straight line to the patient's current address is referenced. If no measurement station exists within a predetermined upper limit distance (e.g., 100 km) from the patient's current address, data is deemed missing for the environmental factor for which measurement data from the measurement station was referenced. Measurement station data can be obtained, for example, from the website of the National Institute for Environmental Studies. Furthermore, information regarding meteorological parameters (precipitation, temperature, season), which are environmental factors, can be obtained, for example, from the website of the Japan Meteorological Agency.

[0044] The patient's residence may also be the room where the patient lives (e.g., a clean room). In this case, at least some of the information on the environmental factors can be obtained by, for example, referring to measurement data from a sensor installed in the room or a sensor attached to the patient.

[0045] The information on environmental factors may be acquired by referring to measurements taken by artificial satellites or predicted values ​​obtained by weather simulations.

[0046] The treatment information includes, for example, the following four factors. The treatment information can be obtained, for example, by recording treatment results: Prednisolone, calcineurin inhibitors, immunosuppressants, anti-fibrotic agents

[0047] Next, the learning data acquisition unit 112 (FIG. 3) of the information processing device 100 performs preprocessing on the original information Io to acquire learning data LD (S120 in FIG. 4). The preprocessing may include, for example, interpolation, outlier removal, data augmentation, etc.

[0048] 6 is an explanatory diagram conceptually illustrating the interpolation process for the original information Io. As shown in the upper part of FIG. 6, before the interpolation process, the timing of the data for factors obtained from each test (e.g., pulmonary function test, blood test, chest CT image) varies depending on the patient, resulting in different timings of the data for factors obtained from each test. Therefore, in this embodiment, the interpolation process is performed so that data at regular time intervals is obtained for all factors. The same applies to factors other than those obtained by tests.

[0049] It is preferable to select and execute an interpolation process suited to the characteristics of each factor. For example, for factors obtained from similar test items, interpolation can be performed using multiple regression analysis based on a group of similar test items. For factors that are highly likely to be estimated from preceding and following values, linear interpolation (interpolation) or nearest neighbor interpolation (extrapolation) can be used. For factors that exhibit rapid fluctuations over a short period of time (e.g., CRP), nearest neighbor interpolation (interpolation, extrapolation) can be used. For categorical variables, nearest neighbor interpolation (interpolation, extrapolation) can be used. For factors with low correlation between time and data (e.g., environmental pollutants), interpolation can be omitted and data can be left missing. By performing such interpolation, data at regular time intervals can be obtained for each factor without compromising the characteristics of each factor.

[0050] FIG. 7 is an explanatory diagram showing an example of training data LD obtained after preprocessing. The training data LD is data in which time-series information (monthly data in the example of FIG. 7) showing the chronological changes of each factor (feature amount) is associated with information (correct label) showing the prognosis at each timing. FIG. 7 shows data of a patient who developed acute exacerbation t months after initial diagnosis. When predicting the onset of acute exacerbation within t-s months, in this patient's data, from the 1st month to the (s-1)th month, the "survival" label is "1" and the remaining labels are "0", and from the sth month onwards, the "acute exacerbation" label is "1" and the remaining labels are "0". Note that zero padding is used after the tth month.

[0051] Next, the model acquisition unit 113 ( FIG. 3 ) of the information processing device 100 creates a prognosis prediction model MO through machine learning using the learning data LD (S130 in FIG. 4 ). Various known machine learning algorithms can be used for the machine learning used to create the prognosis prediction model MO. For example, a long short term memory (LSTM) may be used to create the prognosis prediction model MO. FIG. 8 is an explanatory diagram conceptually illustrating a model using LSTM. The LSTM is an RNN (Recurrent Neural Network) that can handle time-series information, which has been improved to solve the gradient vanishing problem. As shown in FIG. 8 , in creating the prognosis prediction model MO using LSTM, the feature quantity X ｔ The output value y when ｔ and the correct label Y ｔ The model is updated so that the loss calculated from

[0052] Alternatively, the Dynamic-DeepHit model may be used to create the prognosis prediction model MO. The Dynamic-DeepHit model is a known machine learning algorithm that accommodates multiple events that are in a competing risk relationship. As described above, acute exacerbation and death, which are events of interstitial pneumonia, are in a competing risk relationship. Therefore, by using a machine learning algorithm that accommodates multiple events that are in a competing risk relationship, a prognosis prediction model MO with high prediction accuracy can be created. Details of the Dynamic-DeepHit model are described in, for example, the following literature. Lee Changhee and three others, "DeepFit: A Deep Learning Approach to Survival Analysis with Competing Risks," Proceedings of the 31st AAAI Conference on Artificial Intelligence, Association for the Advancement of Artificial Intelligence, 2018, pp. 2314-2321.

[0053] The prognosis prediction model MO created by machine learning is stored in the storage unit 120 of the information processing device 100. This completes the process of acquiring the prognosis prediction model MO ( FIG. 4 ). When creating the prognosis prediction model MO, for example, a portion of the learning data LD is used as training data for updating model parameters (weights, etc.), another portion of the learning data LD is used as verification data for setting hyperparameters, and another portion of the learning data LD is used as test data for checking the generalization performance of the model.

[0054] A-4. Prognosis Prediction Processing: Next, the prognosis prediction processing executed by the information processing device 100 of this embodiment will be described. FIG. 9 is a flowchart showing the prognosis prediction processing in this embodiment. The prognosis prediction processing is processing for predicting the prognosis (predicting the risk of acute exacerbation and death) of a patient suffering from interstitial pneumonia using the prognosis prediction model MO. The prognosis prediction processing is started in response to a start instruction being input by the user operating the operation input unit 140 of the information processing device 100.

[0055] First, the target patient information acquisition unit 114 ( FIG. 3 ) of the information processing device 100 acquires target patient information Ip (S310). The target patient information Ip is the above-mentioned time-series information about the patient who is the target of the prognosis prediction process. The target patient information Ip is acquired via the interface unit 150 or the operation input unit 140, and stored in the storage unit 120.

[0056] Next, the prognosis prediction execution unit 119 ( FIG. 3 ) of the information processing device 100 executes prognosis prediction for the subject patient using the subject patient information Ip and the prognosis prediction model MO (S320). That is, the prognosis prediction execution unit 119 inputs the subject patient information Ip into the prognosis prediction model MO, thereby obtaining a prognosis prediction result output from the prognosis prediction model MO. The prognosis prediction execution unit 119 generates prognosis prediction result data RD, which is information indicating the prognosis prediction result, and stores it in the storage unit 120 of the information processing device 100.

[0057] Next, the prognosis prediction execution unit 119 outputs the prognosis prediction result based on the prognosis prediction result data RD (S330). For example, the prognosis prediction execution unit 119 displays the prognosis prediction result on the display unit 130. This completes the prognosis prediction process.

[0058] For example, doctors can refer to the displayed prognosis prediction results and administer anti-fibrotic drugs to patients who are predicted to have a high probability of developing acute exacerbations to suppress the onset of the disease, or can perform early diagnosis and therapeutic intervention to improve the prognosis.

[0059] 2, the prognosis prediction execution unit 119 may execute a virtual prognosis prediction for the target patient using virtual information obtained by changing part of the time-series information for the target patient and the prognosis prediction model MO, predict the effect of an intervention corresponding to the change based on the predicted result of the actual prognosis and the predicted result of the virtual prognosis, and output the predicted result of the effect of the intervention. In this way, it is possible to determine whether or not to actually perform the intervention based on the output predicted result of the effect of the intervention.

[0060] A-5. Example: An example of the prognosis prediction model MO described above is described below. The prognosis prediction model MO of this example was created through a multi-institutional, retrospective study of patients with newly diagnosed interstitial pneumonia at two hospitals (Tosei Public Hospital and Hamamatsu University School of Medicine) between 2008 and 2015. Of the 839 cases from Tosei Public Hospital, 80% were used as training data for model construction, and the remaining 20% ​​were used as validation data for internal validity verification. In addition, 336 cases from Hamamatsu University School of Medicine were used as test data for external validity (generalization performance) verification.

[0061] FIG. 10 is an explanatory diagram showing the prediction accuracy of the prognosis prediction model MO of the embodiment. FIG. 10 shows the results of internal validity verification using verification data (C-index value) and external validity verification using test data (same) for the prognosis prediction model MO created using the Dynamic-DeepFit model described above. Note that the example in FIG. 10 shows the accuracy of the prognosis prediction results T months later (T = 6, 12, 24, 36) when the prediction time point is 12 months after the initial diagnosis. As shown in FIG. 10, a high C-index value of 0.85 or higher was obtained for both acute exacerbation and death, indicating that the prognosis prediction model MO of the embodiment generally achieved high prediction accuracy. Note that the C-index is an index of prediction accuracy, with a higher value (maximum value: 1) indicating better model performance.

[0062] 11 and 12 are other explanatory diagrams showing the prediction accuracy of the prognosis prediction model MO of the example. FIGS. 11 and 12 show the results of internal validity verification (C-index values) and external validity verification (same results) when monthly concentration or change amount is used as the environmental factor. FIG. 11 shows an example using data from the initial diagnosis up to 12 months later, while FIG. 12 shows an example using data from the initial diagnosis up to 24 months later. As shown in FIGS. 11 and 12 , the example using change amount as the environmental factor achieved prediction accuracy equal to or higher than that of the example using monthly concentration as the environmental factor.

[0063] Figure 13 is an explanatory diagram showing the predictive accuracy of the prognosis prediction model MO of the embodiment. Figure 13 shows the results of external validity verification of the prediction of acute exacerbation by a model that predicts only acute exacerbation (acute exacerbation model) created using the Dynamic-DeepFit model (C-index value), the results of external validity verification of the prediction of death by a model that predicts only death (death model) (same), and the results of external validity verification of the prediction of acute exacerbation and death by a model that predicts both acute exacerbation and death in a competing risk relationship (competitive model) (same). As shown in Figure 13, the predictive accuracy of the model that took competing risks into account was as high as the predictive accuracy of the model that took only acute exacerbation or only death into account. Therefore, it can be said that high predictive accuracy can also be achieved by a model that takes competing risks into account.

[0064] FIG. 14 is an explanatory diagram showing an example of a prognosis prediction result by the prognosis prediction model MO of the embodiment. FIG. 14 shows the results of predicting the cumulative occurrence probability of each event using test data from the initial diagnosis to the 24th month using the prognosis prediction model MO created using the Dynamic-DeepFit model. Column A of FIG. 14 shows an example using test data in which an acute exacerbation occurred 38 months after the initial diagnosis, column B of FIG. 14 shows an example using test data in which death occurred 54 months after the initial diagnosis, and column C of FIG. 14 shows an example using test data in which survival was terminated 81 months after the initial diagnosis. In the prediction results shown in column B of FIG. 14, the probability of death is consistently higher compared to the example shown in column A of FIG. 14. Furthermore, in the prediction results shown in column C of FIG. 14, the probability of both acute exacerbation and death is consistently lower compared to the examples shown in columns A and B of FIG. 14. Thus, it can be said that the prognosis prediction model MO of the embodiment generally achieves high prediction accuracy.

[0065] Fig. 15 is an explanatory diagram showing the prediction accuracy of a prognosis prediction model MO of another embodiment. Fig. 15 shows the results of internal validity verification using verification data (Balanced-Accuracy and F1 score values) for the prognosis prediction model MO created using the LSTM shown in Fig. 8. In an example where the cross-entropy weighting described below was not adjusted (the example of "k = 0" in Fig. 15), the values ​​of Balanced-Accuracy and F1 score were both around 0.6, which means that a reasonable degree of prediction accuracy was achieved.

[0066] As shown in FIG. 7, in the training data LD used to create the prognosis prediction model MO, the proportion of data whose correct answer label is "survival" is very high. In other words, the training data LD is imbalanced data. In order to correct the influence of such data imbalance, the cross-entropy weight W is calculated according to the following formula (1): ｊ The prediction accuracy of the model was confirmed by changing various parameters. ｊ = (N / (M × N ｊ )) ｋ ... (1) However, ・W ｊ: Label = weight at j N: total number of labels M: number of classes N ｊ : Label = number of j ・k: Hyperparameter

[0067] However, as shown in FIG. 15, the weight W of each label can be adjusted by adjusting the value of the hyperparameter k. ｊ Even when the parameter values ​​were changed, the prediction accuracy did not improve. Possible causes of this include, in addition to the data imbalance problem described above, problems with the definition of "predicted events," such as cases with different predicted probabilities of "acute exacerbation" being classified as the same "acute exacerbation," and the similarity between events, such as acute exacerbation being a fatal pathological condition. Based on the above, it can be said that it is preferable to use a machine learning algorithm (e.g., Dynamic-DeepFit) that supports multiple events in the competing risk relationship described above when creating a prognosis prediction model MO for predicting acute exacerbation and / or death in interstitial pneumonia.

[0068] FIG. 16 is an explanatory diagram showing an example of the importance of each factor of a disease in an example. FIG. 16 shows the importance of each factor (top 20) determined by examining the change in the risk of acute exacerbation when the value that each factor (feature amount) can take is changed, and assuming that the greater the change in risk, the higher the importance (contribution). As shown in FIG. 16, several environmental factors (SPM, NO ２ , P.M. ２．５ ) are more important than factors related to pulmonary function (FVC, DLco) and factors related to severity (GAP score). Therefore, it can be said that it is preferable to use environmental factors for predicting the prognosis of interstitial pneumonia.

[0069] FIG. 17 is an explanatory diagram showing an example of the importance of each factor of a disease in another embodiment. In FIG. 17, the importance of each factor is shown similarly to FIG. 16, but in the example of FIG. 17, in addition to the monthly average value, the amount of change from the previous month (the environmental factor name in FIG. 17 has "diff" added to the end) is used as the environmental factor. As shown in FIG. 17, several environmental factors (PM ２．５, NO, Ox) the importance of the amount of change from the previous month is high. Therefore, it can be said that it is preferable to use the amount of change per certain period in addition to or instead of the representative value per certain period as an environmental factor used in predicting the prognosis of interstitial pneumonia. Note that, from the results shown in FIG. 17, it can be said that for each environmental factor, whether the representative value per certain period or the amount of change is important differs. For example, the importance of the representative value per certain period is high for SPM, the amount of change is high for NO, and the importance of PM is high. ２．５ and Ox are both highly important.

[0070] 18 and 19 are explanatory diagrams showing the relationship between the upper limit of the straight-line distance from the patient's current address to the measurement station (upper limit distance R) and prediction accuracy. As described above, in this embodiment, if there is no measurement station within the predetermined upper limit distance R from the patient's current address, data is considered missing for the environmental factors that reference the measurement data of the measurement station. FIG. 18 shows the change in the number of data points (number of data points when there is no missing data: 68,261 (person-month)) for each environmental factor as the upper limit distance R changes. As shown in FIG. 18, the larger the value of the upper limit distance R, the fewer data points are missing for each environmental factor.

[0071] FIG. 19 shows the change in prediction accuracy of acute exacerbation and death with changes in upper limit distance R. Regardless of whether training data, validation data, or test data is used, the prediction accuracy is highest when the upper limit distance R is 50 km or 100 km. If the upper limit distance R is too large, data loss will be reduced, but the deviation from the actual environment to which the patient was exposed will increase, resulting in a decrease in prediction accuracy. From the results shown in FIG. 19, it can be said that the upper limit distance R is preferably 20 km or more and 100 km or less.

[0072] A-6. Effects of this embodiment: As described above, the information processing device 100 of this embodiment is a device for predicting the prognosis of a target patient suffering from a disease, and includes a model acquisition unit 113, a target patient information acquisition unit 114, and a prognosis prediction execution unit 119. The model acquisition unit 113 receives time-series information indicating chronological changes in disease factors as input and acquires a prognosis prediction model MO, which is a machine learning model that outputs the prognosis of the disease. The target patient information acquisition unit 114 acquires time-series information about the target patient. The prognosis prediction execution unit 119 uses the time-series information about the target patient and the prognosis prediction model MO to predict the prognosis of the target patient and outputs the prognosis prediction result.

[0073] In this way, the information processing device 100 of this embodiment can predict the prognosis of a disease for each individual patient based on time-series information that indicates the chronological changes in disease factors. Therefore, the information processing device 100 of this embodiment can predict the prognosis of a disease with high accuracy.

[0074] In this embodiment, the disease factors include environmental factors, and therefore, the information processing device 100 of this embodiment can predict the prognosis of a disease with higher accuracy by using information indicating the time-series changes in environmental factors that can have a significant impact on the prognosis of the disease.

[0075] In this embodiment, the disease factors include environmental factors of the patient's residence. Therefore, the information processing device 100 of this embodiment can predict the prognosis of the disease with higher accuracy by using information indicating time-series changes in factors related to the environment to which the patient is primarily exposed, compared to using environmental factors of the location of the hospital to which the patient visits, for example.

[0076] In this embodiment, the environmental factors include at least one of the presence of environmental pollutants and meteorological parameters. Therefore, the information processing device 100 of this embodiment can predict the prognosis of a disease with higher accuracy by using information indicating the time-series changes in environmental factors that can have a significant impact on the prognosis of the disease.

[0077] In this embodiment, the time-series information includes information indicating the amount of change in environmental factors, and therefore, according to the information processing device 100 of this embodiment, the prognosis of a disease can be predicted with even higher accuracy by using information indicating the amount of change in environmental factors that can have a significant impact on the prognosis of the disease.

[0078] In this embodiment, the time-series information includes information that identifies the values ​​of disease factors at regular time intervals, and therefore, the information processing device 100 of this embodiment can predict the prognosis of the disease with higher accuracy than when using information that irregularly identifies the values ​​of disease factors.

[0079] In this embodiment, the time-series information includes information specifying at least the values ​​of disease factors for each month. Therefore, according to the information processing device 100 of this embodiment, by using the time-series information showing the monthly changes in disease factors, it is possible to predict the prognosis of the disease with even higher accuracy.

[0080] In this embodiment, the prognosis of a disease includes the occurrence of multiple events that are in a competing risk relationship, and the prognosis prediction model MO is a model trained using a machine learning algorithm that corresponds to multiple events that are in a competing risk relationship. Therefore, the information processing device 100 of this embodiment can predict the occurrence of multiple events that are in a competing risk relationship with high accuracy.

[0081] In this embodiment, the prognosis prediction model MO is a model that outputs, as a prognosis of a disease, an index value that represents the probability of occurrence of multiple events that are in a competing risk relationship. Therefore, the information processing device 100 of this embodiment can predict the occurrence of multiple events that are in a competing risk relationship with high accuracy.

[0082] In this embodiment, the multiple events that are competing risks include acute exacerbation and death. Therefore, the information processing device 100 of this embodiment can accurately predict the occurrence of each of the competing risks, acute exacerbation and death.

[0083] In this embodiment, the disease is a disease of the respiratory system or the circulatory system. Therefore, according to the information processing device 100 of this embodiment, it is possible to predict the prognosis of the disease of the respiratory system or the circulatory system with high accuracy by using time-series information that indicates the time-series changes in the factors of the disease of the respiratory system or the circulatory system.

[0084] In addition, in this embodiment, the prognosis prediction execution unit 119 may execute a virtual prognosis prediction for the target patient using virtual information obtained by changing part of the time-series information for the target patient and the prognosis prediction model MO, predict the effect of an intervention corresponding to the change based on the prediction result of the actual prognosis and the prediction result of the virtual prognosis, and output the prediction result of the effect of the intervention. With this configuration, it is possible to determine whether or not to actually perform the intervention based on the output prediction result of the effect of the intervention.

[0085] B. Modifications: The technology disclosed in this specification is not limited to the above-described embodiment, and can be modified in various forms without departing from the spirit of the invention. For example, the following modifications are also possible.

[0086] The configuration of the information processing device 100 in the above embodiment is merely an example and can be modified in various ways. Furthermore, the contents of the prognosis prediction model acquisition process and the prognosis prediction process in the above embodiment are merely an example and can be modified in various ways. For example, in the above embodiment, the information processing device 100 acquires the prognosis prediction model MO by creating the prognosis prediction model MO, but the information processing device 100 may also acquire a prognosis prediction model MO generated by another device.

[0087] The disease factors (features) and machine learning algorithms used to create the prognosis prediction model MO in the above embodiment are merely examples and can be modified in various ways. For example, features other than those exemplified in the above embodiment may be used as features used to create the prognosis prediction model MO, or some of the features exemplified in the above embodiment may not be used. Furthermore, algorithms other than Dynamic-DeepFit and LSTM may be used as machine learning algorithms used to create the prognosis prediction model MO.

[0088] In the above embodiment, information processing for predicting the prognosis of a patient suffering from interstitial pneumonia is exemplified, but the technology disclosed in this specification is not limited to interstitial pneumonia and can be similarly applied to predicting the prognosis of a patient suffering from any disease. Note that, since the prognosis of a respiratory or circulatory system disease is thought to be significantly influenced by environmental factors, it is preferable that the disease factors used for predicting the prognosis of a respiratory or circulatory system disease include environmental factors.

[0089] In the above-described embodiment, a part of the configuration realized by hardware may be replaced by software, and conversely, a part of the configuration realized by software may be replaced by hardware.

[0090] 100: Information processing device 110: Control unit 111: Raw information acquisition unit 112: Learning data acquisition unit 113: Model acquisition unit 114: Target patient information acquisition unit 119: Prognosis prediction execution unit 120: Memory unit 130: Display unit 140: Operation input unit 150: Interface unit 190: Bus CP: Prognosis prediction program LD: Learning data MO: Prognosis prediction model RD: Prognosis prediction result data

Claims

1. An information processing apparatus for predicting the prognosis of a target patient suffering from a disease, comprising: a model acquisition unit that acquires a prognosis prediction model, which is a machine learning model that takes as input time-series information indicating the temporal change of factors of the disease and outputs the prognosis of the disease; a target patient information acquisition unit that acquires the time-series information about the target patient; a prognosis prediction execution unit that uses the time-series information about the target patient and the prognosis prediction model to execute a prediction of the prognosis of the target patient and outputs the prediction result of the prognosis; wherein the factors of the disease include environmental factors.

2. (Deleted)

3. The information processing apparatus according to claim 1, wherein the factors of the disease include environmental factors of the place of residence of the target patient.

4. The information processing apparatus according to claim 3, wherein the environmental factors of the place of residence of the target patient are environmental factors of a location within a straight-line distance of 200 km from the current address of the target patient.

5. The information processing apparatus according to claim 1 or claim 3, wherein the environmental factors include at least one of the presence status of environmental pollutants and meteorological parameters.

6. The information processing apparatus according to claim 1 or claim 3, wherein the time-series information includes information indicating the amount of change in the environmental factors.

7. The information processing apparatus according to claim 1, wherein the time-series information is information that specifies the values of the factors of the disease at regular time intervals.

8. The information processing apparatus according to claim 7, wherein the time-series information includes information that specifies the values of the factors of the disease at least monthly.

9. The information processing apparatus according to claim 1, wherein the prognosis of the disease includes the occurrence of a plurality of events related to competing risks, and the prognosis prediction model is a model learned using a machine learning algorithm corresponding to a plurality of events related to competing risks.

10. The information processing apparatus according to claim 9, wherein the prognosis prediction model is a model that outputs, as the prognosis of the disease, an index value representing the possibility of the occurrence of a plurality of events related to competing risks.

11. The information processing apparatus according to claim 9 or claim 10, wherein A plurality of events related to the competition risk, including acute exacerbation and death, are information processing devices.

12. An information processing device according to claim 1, wherein the disease is a respiratory or cardiovascular disease, is an information processing device.

13. An information processing device according to claim 1, wherein the prognosis prediction execution unit uses virtual information obtained by changing a part of the time-series information about the target patient and the prognosis prediction model to execute a prediction of the virtual prognosis of the target patient, and based on the prediction result of the actual prognosis and the prediction result of the virtual prognosis, predicts the effect of the intervention corresponding to the change, and outputs the prediction result of the effect of the intervention, is an information processing device.

14. An information processing method for predicting the prognosis of a target patient suffering from a disease, comprising the step of obtaining a prognosis prediction model, which is a machine learning model that takes as input time-series information indicating the temporal change of the factors of the disease and outputs the prognosis of the disease; the step of obtaining the time-series information about the target patient; the step of using the time-series information about the target patient and the prognosis prediction model to execute a prediction of the prognosis of the target patient and output the prediction result of the prognosis; and wherein the factors of the disease include environmental factors, is an information processing method.

15. A computer program for predicting the prognosis of a target patient suffering from a disease, causing a computer to perform a process of obtaining a prognosis prediction model, which is a machine learning model that takes as input time-series information indicating the temporal change of the factors of the disease and outputs the prognosis of the disease; a process of obtaining the time-series information about the target patient; a process of using the time-series information about the target patient and the prognosis prediction model to execute a prediction of the prognosis of the target patient and output the prediction result of the prognosis; and wherein the factors of the disease include environmental factors, is a computer program.