Treatment policy determination support device and recording medium

The treatment policy decision support device uses a machine learning model to classify patients and predict prognosis, addressing the challenge of unclear disease causes by providing timely and effective treatment strategies.

WO2026235113A1PCT designated stage Publication Date: 2026-11-12THE UNIV OF TOKYO
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2026/017302
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-07
Filing Date
2026-04-30
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

Existing treatment policies for diseases like pneumonia are often empirically selected before definitive diagnosis, lacking clear causative agents, and susceptibility tests require 3-5 days, hindering timely empiric treatment decisions.

Method used

A treatment policy decision support device utilizing a machine learning model that processes clinical, test, and medical history information to classify patients into risk groups, predicting prognosis and guiding treatment plans based on known information.

Benefits of technology

Enables rapid and informed treatment strategy determination even when the disease cause is unclear, supporting evidence-based decisions for diseases like pneumonia.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2026017302_12112026_PF_FP_ABST
    Figure JP2026017302_12112026_PF_FP_ABST
Patent Text Reader

Abstract

This treatment policy determination support device employs a machine learning model that has been machine-trained to output estimated prognosis information using main factor information as input information, wherein: prognosis information corresponding to the main factor information for a plurality of sample patients is acquired using the machine learning model; the main factor information for the sample patients is classified into any of a plurality of background groups on the basis of the acquired prognosis information; the background group to which the main factor information of a target patient for whom a treatment policy is to be determined belongs is identified using the machine learning model; prognosis information for each treatment policy for the sample patients belonging to the identified group to which the target patient belongs is acquired; and the treatment policy is guided on the basis of the acquired prognosis information.
Need to check novelty before this filing date? Find Prior Art

Description

Treatment Policy Decision Support Device, and Recording Medium

[0001] The present invention relates to a treatment policy decision support device that predicts the prognosis of a disease to support decision-making of a treatment policy, and a recording medium.

[0002] Conventionally, treatment policies before definitive diagnosis (empiric treatment) have been selected empirically based on patient attributes such as the patient's age, underlying diseases, symptoms, and findings. For example, for a patient suspected of having an infectious disease, before a definitive diagnosis is made, the causative microorganism is estimated from the patient's attributes, symptoms, findings, etc., a therapeutic drug is selected and administration is started, and thereafter, the causative microorganism is narrowed down based on findings obtained from tests and the like, a therapeutic drug is further selected, and treatment is generally performed (Non-Patent Document 1).

[0003] Japanese Respiratory Society Adult Pneumonia Diagnosis and Treatment Guideline 2024 Editorial Committee, Adult Pneumonia Diagnosis and Treatment Guideline 2024, Japanese Respiratory Society, April 9, 2024

[0004] However, in the case of pneumonia for example, a susceptibility test or the like is performed to examine which drug is effective, but this test usually requires a test period of 3 to 5 days, and thus does not help in determining the empiric treatment policy.

[0005] The present invention has been made in view of the above circumstances, and one of its objects is to provide a treatment policy decision support device and a recording medium that can support the decision of a treatment policy even for diseases such as pneumonia whose cause is not necessarily clear.

[0006] One aspect of the present invention, which solves the problems of the above-mentioned conventional examples, is a treatment strategy decision support device, comprising: a holding means for holding a machine learning model that has been trained to output estimated prognosis information using a set of known information including clinical information, test result information, medical history information, drug information used, information identifying bacteria or viruses causing the disease, and prognosis information, with the aforementioned major factor information as input information; and a means for acquiring prognosis information corresponding to the major factor information for multiple sample patients using the machine learning model, and based on the acquired prognosis information, the major factor information for the multiple sample patients is processed by multiple back The system includes: a classification means for classifying the multiple sample patients into groups by classifying them into one of the background groups; a classification and identification means for receiving input of key factor information concerning the target patient for whom a treatment plan has been determined, acquiring prognosis information corresponding to the input key factor information of the target patient using the machine learning model, identifying the background group to which the key factor information for the target patient belongs based on the acquired prognosis information, and identifying the group to which the target patient belongs; and a guidance means for acquiring prognosis information for each treatment plan for the sample patients belonging to the identified group to which the target patient belongs, and guiding the treatment plan based on the acquired prognosis information.

[0007] According to the present invention, it is possible to support the determination of treatment strategies.

[0008] This is a block diagram showing an example of a treatment strategy decision support device according to an embodiment of the present invention. flowchart showing an example of processing by the treatment strategy decision support device according to an embodiment of the present invention. This is a flowchart showing an example of processing for guiding treatment strategies by the treatment strategy decision support device according to an embodiment of the present invention. This is an explanatory diagram showing an example of prognosis information generated by the treatment strategy decision support device according to an embodiment of the present invention. This is an explanatory diagram showing another example of prognosis information generated by the treatment strategy decision support device according to an embodiment of the present invention.

[0009] Embodiments of the present invention will be described with reference to the drawings. An example of a treatment strategy decision support device 1 according to an embodiment of the present invention is a general computer device comprising a control unit 11, a storage unit 12, an operation unit 13, a display unit 14, and a communication unit 15, as illustrated in Figure 1.

[0010] In the following explanation, factor information, major factor information, and prognostic information refer to the following types of information: Factor information... Information including clinical information, test result information, medical history information, information identifying the bacteria or viruses causing the disease, and treatment-related information such as information on medications used. Major factor information... Information of at least some types of factor information selected from the factor information, including at least treatment-related information. Prognostic information... Probability information representing the patient's future state, such as the probability of discharge per day from admission, and the probability of survival (probability of survival per day from the time of admission or initiation of treatment). Furthermore, the information included in the above factor information is, as an example, clinical information... Age, sex, height, weight, BMI, number of hospitalizations of the subject, as well as vital information at the time of admission (or initiation of treatment), such as PS (performance). Information on test results including status, body temperature, blood pressure, oxygenation (e.g., oxygen concentration (P / F))... Information on blood tests including white blood cell count (WBC), hemoglobin (HB), platelet count (PLt), amounts related to nutritional status (e.g., albumin (Alb) value), amounts related to renal function (e.g., uric acid (BUN) and creatinine (Cre) values, estimated glomerular filtration rate (eGFR)), amounts related to liver function (GOT, GPT), amounts related to the presence or absence of inflammation or infection (e.g., C-reactive protein (CRP) value), total bilirubin (T-bil)... Information on past medical history... Information on the presence or absence of hypertension, diabetes, cardiovascular disease, chronic heart failure, cerebral infarction, respiratory disease, sepsis, cancer, etc.... Information on medications used: Information that identifies the type, group, and dosage of antibiotics being used... Information that identifies the bacteria or viruses causing the disease... Information such as the name of the bacteria or viruses causing the disease, and the presence or absence of causative bacteria and resistant bacteria.

[0011] Here, the control unit 11 is a program control device such as a CPU, and operates according to a program stored in the memory unit 12. In this embodiment, the control unit 11 uses a machine learning model that has been trained to take a predetermined weighted set of major factor information and prognosis information as training data and test data, and outputs prognosis information.

[0012] The control unit 11 obtains prognosis information estimated by the machine learning model based on the major factor information of multiple actual patients (hereinafter referred to as sample patients for distinction). Then, based on the prognosis information for each sample patient, the control unit 11 classifies the patient's major factor information into multiple background groups.

[0013] The control unit 11 also receives input from the user for key factor information about the patient whose treatment plan is to be determined (referred to as the target patient), and uses this input key factor information as input to the machine learning model to obtain prognosis information for the target patient estimated by the machine learning model. The control unit 11 then refers to the prognosis information for the target patient and identifies the background group to which the key factor information for the target patient belongs from among the background groups.

[0014] The control unit 11 acquires prognosis information for each treatment plan for sample patients belonging to the background group identified as to which the target patient's major factor information belongs, and guides the patient on the treatment plan based on the acquired prognosis information. The detailed operation of these control units 11 will be described later.

[0015] The storage unit 12 is a memory device, a disk device, etc., and holds the program executed by the control unit 11. This storage unit 12 also operates as the work memory of the control unit 11. This program may be provided on a computer-readable, non-temporary recording medium and copied to this storage unit 12.

[0016] In one example of this embodiment, the storage unit 12 holds the data of the machine learning model that has been trained. In other words, in one example of this embodiment, the storage unit 12 functions as a storage means of the present invention.

[0017] The operation unit 13 is a mouse, keyboard, etc., which accepts user input and outputs information representing the content of that operation to the control unit 11. The display unit 14 is a display, etc., which displays and outputs information according to instructions input from the control unit 11.

[0018] The communication unit 15 is a network interface, etc., and sends and receives various data via the network according to instructions input from the control unit 11.

[0019] Next, the operation of the control unit 11 in this embodiment will be described. In this embodiment, the control unit 11 performs machine learning processing and treatment plan prediction processing using the results of the machine learning processing. Functionally, the control unit 11 includes a machine learning processing unit 20 and an estimation processing unit 30, as illustrated in Figure 2. Here, the machine learning processing unit 20 includes an information collection unit 21, a major factor selection unit 22, and a machine learning unit 23, and the estimation processing unit 30 includes an information input unit 31, a classification processing unit 32, a classification identification unit 33, and a guidance processing unit 34.

[0020] The information gathering unit 21 of the machine learning processing unit 20 receives multiple sets of known factor information and corresponding prognosis information (sets of known information), and stores and retains these sets of information in the storage unit 12.

[0021] Specifically, the information collected by this information gathering unit 21 is a set of multiple pieces of information about past patients whose prognosis is known, and the factor information includes, as already mentioned, at least one of the following types of information: clinical information, test result information, medical history information, medication information, information identifying the bacteria or virus causing the disease, and initial response information.

[0022] Furthermore, the prognostic information acquired by the information gathering unit 21 here is assumed to be survival probability information (survival probability for each day since the start of treatment).

[0023] The major factor selection unit 22 selects at least a portion of the factor information contained in the information collected by the information collection unit 21 as major factor information. The major factor selection unit 22 also obtains weight information corresponding to each of the selected major factor information. Here, the selection of major factor information and the values ​​of the corresponding weights may be performed by, for example, receiving information that identifies which major factor information it is and information about the weights in advance, and then selecting at least a portion of the factor information identified by the pre-received information and setting the corresponding weights. Alternatively, as will be explained later, the control unit 11 may perform preliminary machine learning processing (preliminary processing) to select the major factor information and obtain the values ​​of the corresponding weights.

[0024] The machine learning unit 23 extracts major factor information selected by the major factor selection unit 22 from the factor information contained in each set of known information collected by the information collection unit 21, and uses this as input information. The machine learning unit 23 also uses the prognosis information (contained in the same set) corresponding to the extracted major factor information as the target variable.

[0025] The machine learning unit 23 then uses the extracted major factor information as explanatory variables and outputs the corresponding prognosis information as the target variable by machine learning a model such as a decision tree, a random forest, or a neural network. Here, the machine learning processing method for decision trees and random forests may be a widely known method such as C4.5, and when using a neural network model, the machine learning processing may be performed using a method such as backpropagation.

[0026] Furthermore, the information input unit 31 of the estimation processing unit 30 accepts input from the user of key factor information about the patient (referred to as the target patient) for whom the treatment plan has been determined.

[0027] The classification processing unit 32 performs the following processing using pre-prepared major factor information for multiple patients (sample patients) (N people, where N is 2 or more, a statistically sufficiently large number). The major factor information for these sample patients may be clinically accumulated information. Here, the classification processing unit 32 excludes information related to treatment from the major factor information (hereinafter, the information processed in this way will be referred to as background major factor information for distinction) and inputs the obtained background major factor information into the machine learning model obtained by the machine learning processing of the machine learning processing unit 20, and obtains correspondingly estimated prognosis information. The classification processing unit 32 treats any missing information among the factor information input into the machine learning model as missing information. The classification processing unit 32 may also multiply each of the background major factor information input into the machine learning model by a corresponding weight before inputting it. In other words, each of the major factor information may be assigned a predetermined weight.

[0028] The classification processing unit 32 classifies the N prognosis information obtained here into multiple groups (referred to as background groups) using a predetermined method. In one example of this embodiment, the classification by the classification processing unit 32 is performed, for example, using the survival probability represented by the prognosis information and the survival probability after a predetermined number of days (for example, 90 days from the start of treatment).

[0029] Specifically, the classification processing unit 32 uses at least one threshold (cutoff value) to classify each of the background major factor information for N individuals into either a background group exceeding the cutoff value or a background group below the cutoff value. The classification processing unit 32 then classifies the N sample patients into groups. If a machine learning model such as a decision tree or random forest is used here, the machine learning processing method may be a widely known method such as C4.5.

[0030] The classification and identification unit 33 inputs the major factor information about the target patient (information related to treatment may be excluded here as well) received from the user by the information input unit 31 into the machine learning model obtained by the machine learning processing unit 20, and obtains correspondingly estimated prognosis information. The classification and identification unit 33 may also multiply each of the background major factor information input to the machine learning model by the corresponding weight before inputting it.

[0031] The classification unit 33 then compares the obtained prognosis information with the cutoff value used by the classification processing unit 32 to classify the background major factor information of the target patient into one of the above background groups. At this point, the background group to which the background major factor information of the target patient should belong is identified, and the group of sample patients to which the target patient should belong (a group of sample patients whose corresponding background major factor information is classified into the same background group) is identified.

[0032] The guidance processing unit 34 acquires prognosis information for each treatment plan for sample patients belonging to the group to which the target patient should belong, as identified by the classification identification unit 33. Specifically, the guidance processing unit 34 acquires major factor information (including information related to treatment) for sample patients belonging to the background group to which the target patient should belong.

[0033] The guidance processing unit 34 inputs the acquired major factor information into a machine learning model obtained by the machine learning processing unit 20, and obtains correspondingly estimated prognosis information. The guidance processing unit 34 analyzes the estimated prognosis information for each major factor information where the treatment-related information differs from one another (for example, the types of drugs administered differ from one another), and obtains treatment-related information (for example, information identifying the administered drugs) that meets the desired conditions (for example, the highest average survival probability or conditions exceeding a predetermined threshold). The guidance processing unit 34 then presents the acquired treatment-related information to the user and guides them on the treatment plan.

[0034] [Operation] This embodiment has the above configuration and operates as follows. The treatment strategy decision support device 1 of this embodiment uses known factor information and corresponding prognosis information (a set of known information) to obtain a trained machine learning model, for example, a decision tree or a random forest, which is used to output the major factor information extracted from the factor information as explanatory variables by assigning predetermined weights, and the corresponding prognosis information (hereinafter, assumed to be the survival probability for each day elapsed since the start of treatment) as the target variable. Here, when using a decision tree or a random forest as an example of this trained machine learning model, the machine learning processing method may be a widely known method such as C4.5.

[0035] The selection of key factor information from among the factor information, and the weights assigned to this key factor information, may be determined empirically, statistically, or by machine learning methods. Specifically, the weights of the key factor information may be determined based on the p-values ​​obtained after performing the preliminary processing exemplified in Japanese Patent No. 7503860. For example, these weights may be set such that the larger the p-value corresponding to the key factor information, the greater the weight, as this indicates that the key factor information is significant. However, this method of determining weights is just one example; for example, a neural network may be used to determine the weights of the relationship between each key factor information and prognosis.

[0036] Furthermore, the treatment strategy decision support device 1 of this embodiment obtains key factor information from multiple sample patients, for example from clinical information, estimates the prognosis information of the sample patients using the key factor information and the machine learning model that has been trained (for example, a random forest model), and classifies the sample patients into multiple groups based on the prognosis information.

[0037] Specifically, the control unit 11 of the treatment strategy decision support device 1, as illustrated in Figure 3, removes information related to treatment from each of the major factor information prepared for the sample patient to obtain background major factor information (S11), and inputs the background major factor information into a machine learning model that has been trained (S12). The control unit 11 treats any missing information among the factor information to be input into the machine learning model as missing information. In step S12, the control unit 11 may also multiply each of the background major factor information to be input into the machine learning model by the corresponding weight before inputting it.

[0038] The control unit 11 obtains prognosis information estimated as the output of the machine learning model, and from this prognosis information, it compares, for example, the survival probability value R after a predetermined number of days (e.g., 90 days) with a predetermined cutoff value, and classifies the background major factor information input in step S12 into either a first background group (relatively low-risk group; B1) that exceeds the cutoff value or a second background group (relatively high-risk group; B2) that falls below the cutoff value (S13).

[0039] The control unit 11 repeats the processing in steps S11 to S13 for each sample patient's major factor information, classifying each sample patient's major factor information into either the first background group (B1) or the second background group (B2). In other words, through this processing, the control unit 11 classifies the sample patients into two groups: one group of sample patients whose major factor information belongs to the first background group (B1) (referred to as the first group for convenience) and another group of sample patients whose major factor information belongs to the second background group (B2) (referred to as the second group).

[0040] The treatment decision support device 1 stores classification information for each sample patient, relating it to information that identifies the sample patient (it does not need to be possible to identify the individual; information such as P1, P2, etc. is acceptable as long as the individual pieces of information can be distinguished), major factor information, and classification results (information indicating which group the patient belongs to).

[0041] Further, the control unit 11 of the treatment policy decision support device 1 accepts main factor information related to a target patient from a user, starts the process illustrated in Figure 4, excludes information related to treatment from the accepted main factor information to obtain background main factor information (S21), and inputs the background main factor information into a machine learning model that has undergone machine learning (S22). Here too, among the factor information input to the machine learning model, missing information is treated as missing information. Also in this step S22, the control unit 11 may multiply each piece of the background main factor information to be input to the machine learning model by a corresponding weight before inputting it.

[0042] Then, the control unit 11 obtains prognostic information estimated corresponding to the input background main factor information (S23). The control unit 11 compares the prognostic information obtained in step S23 with the cutoff value used when classifying sample patients (the process in step S13 above), and classifies the background main factor information input in step S22 into either the first background group (B1) that exceeds the cutoff value or the second background group (B2) that falls below the cutoff value (S24).

[0043] The control unit 11 identifies the group of sample patients to which the target patient should belong based on the classification in step S24 (S25). Specifically, here, when the control unit 11 determines that the background main factor information input in step S22 belongs to the first background group (B1), the control unit 11 determines that the target patient belongs to the first group. When the control unit 11 determines that the background main factor information input in step S22 belongs to the second background group (B2), the control unit 11 determines that the target patient belongs to the second group.

[0044] Through this process, the treatment policy decision support device 1 of the present embodiment classifies the background of the target patient into one of the groups of patients for whom clinical information has been obtained in the past (for example, groups such as a low-risk group and a high-risk group).

[0045] The control unit 11 acquires the main factor information of sample patients belonging to the group identified in step S25 from the classification information, and sequentially inputs the acquired information into a machine learning model that has undergone machine learning (S26). Also in this step S26, the control unit 11 may multiply each piece of background main factor information to be input to the machine learning model by a corresponding weight before inputting it.

[0046] Then, the control unit 11 obtains estimated prognosis information for the sample patients belonging to the group identified in step S25 (S27). The control unit 11 analyzes (for example, calculates the average of) the obtained prognosis information for each piece of treatment-related information (for example, information on administered drugs) included in the input main factor information.

[0047] As an example, in this step S27, the control unit 11 performs analysis such that the average value of the prognosis information estimated corresponding to the main factor information including the treatment-related information indicating that drug α was administered (the average value of the 90-day survival probability) is 0.6, and the average value of the prognosis information estimated corresponding to the main factor information including the treatment-related information indicating that drug β was administered is 0.4...

[0048] Then, the control unit 11 selects treatment-related information that satisfies a preset desired condition (for example, a condition such as "the average survival probability is the highest"), here, the treatment-related information of administering drug α (S28), and presents it to the user of the treatment policy determination support apparatus 1 (S29). The user can use this presented information for determining a treatment policy such as empirical treatment.

[0049] [Multiple Cutoff Values] In the description up to this point, the treatment policy determination support apparatus 1 is configured such that the number of cutoff values for prognosis information used to classify sample patients and target patients is one, and the sample patients and the like are classified into two groups, but the present embodiment is not limited to this.

[0050] For example, the treatment decision support device 1 may classify sample patients or target patients into more (three or more) groups by setting multiple cutoff values. In this example as well, the treatment decision support device 1 inputs the major factor information of sample patients belonging to the group to which the target patient belongs into a machine learning model that has been trained (multiplying by the corresponding weights if weights are set), and analyzes the prognosis information obtained as the output of the machine learning model for each treatment-related information (for example, information on administered drugs) included in the input major factor information.

[0051] The treatment decision support device 1 then presents the user with information related to treatments that meet the desired conditions (for example, the treatment with the highest average survival probability) based on the results of the analysis.

[0052] [Information regarding treatment] In addition, in step S26 of Figure 4, the major factor information of sample patients belonging to the group identified in step S25 is obtained from the classification information and input into the machine learning model that has been trained. However, the control unit 11 of the treatment strategy decision support device 1 may include different information regarding treatment in the major factor information input into the machine learning model, instead of clinically obtained information.

[0053] As an example, the control unit 11 may sequentially acquire key factor information of sample patients belonging to the group identified in step S25 from the classification information, and may also set information regarding treatment to be included in the acquired key factor information by randomly selecting from among the expected candidates (for example, candidates for drugs that may be administered).

[0054] [Modification] In the description so far, the machine learning model generated by the control unit 11 operating as the machine learning unit 23 is assumed to be a general decision tree or random forest. However, as already mentioned, the machine learning model used in this embodiment is not limited to these examples. XGBoost or Light GBM (Gradient Boosting) may also be used, or other machine learning models, such as deep learning models, may also be used.

[0055] [Effects of the Embodiment] According to this embodiment, even in situations where factors affecting the prognosis of a disease are unknown, such as pneumonia in the elderly for which clinical trials have not been conducted, it is possible to support the determination of treatment strategies for the target disease using real-world data such as information obtained in clinical practice.

[0056] [Example] Next, an example of providing information on actual treatment policies using the treatment policy decision support device 1 of this embodiment will be described. The following example concerns pneumonia in elderly patients. In the treatment policy decision support device 1 of this embodiment, pairs of known factor information and corresponding prognosis information (pairs of known information) for elderly patients suffering from pneumonia are obtained in advance. From this factor information, major factor information is extracted using methods such as the preliminary processing operation exemplified in Japanese Patent No. 7503860, and the weights of these major factor information are determined. These major factor information is used as explanatory variables, and a random forest model is trained to output the corresponding prognosis information as the target variable when the major factor information is input multiplied by the corresponding weights, thereby obtaining a trained machine learning model.

[0057] Furthermore, this treatment strategy decision support device 1 uses information from multiple elderly pneumonia cases accumulated as clinical information, and uses each patient included in these cases as a sample patient. It inputs the major factor information (background major factor information, excluding treatment information) for each sample patient into the trained machine learning model (multiplying by the corresponding weights if weights are set) to obtain corresponding prognosis information. Here, the corresponding prognosis information is the survival probability 90 days after the start of treatment.

[0058] The treatment decision support device 1 accepts input of at least one cutoff value from the user. This cutoff value may be determined by statistical or machine learning methods, or it may be determined experimentally or empirically. In this example, the user inputs two values ​​as cutoff values: 0.3 and 0.7.

[0059] The treatment strategy decision support device 1 classifies the major factor information for each sample patient into three background groups by comparing the correspondingly obtained prognosis information R with the above cutoff value: R < 0.3: poor prognosis group, 0.3 ≤ R < 0.7: intermediate group, and 0.7 ≤ R: good prognosis group.

[0060] Furthermore, the treatment strategy decision support device 1 receives major factor information (background major factor information, excluding treatment-related information) about the target patient for which the treatment strategy is to be decided from the user, inputs this background major factor information into the trained machine learning model (multiplying it by the corresponding weights if weights are set) to obtain the corresponding prognosis information. Here, for example, let's assume that the obtained prognosis information was 0.2.

[0061] The treatment strategy decision support device 1 compares the above cutoff value with the prognosis information obtained here and identifies the major factor information of the target patient as belonging to the poor prognosis group. The treatment strategy decision support device 1 then sequentially inputs the major factor information (including information on treatment) belonging to the poor prognosis group into the above-trained machine learning model (multiplying by the corresponding weights if weights are set) to obtain the corresponding prognosis information (Figure 5).

[0062] Here, for example, we assume that prognosis information is available for cases where one of the antibiotics α, β, or γ is administered as treatment information. As illustrated in Figure 5, if major factor information including treatment information that drug α was administered is input, the prognosis information output by the machine learning model will show that the survival probability decreases daily, reaching approximately 0.5 on day 90. Furthermore, for the output of the machine learning model corresponding to major factor information including treatment information that drug β was administered, the survival probability will decrease daily, reaching approximately 0.4 on day 90. In addition, for the output of the machine learning model corresponding to major factor information including treatment information that drug γ was administered, the survival probability will decrease daily, reaching approximately 0.1 on day 90.

[0063] In this example, the treatment decision support device 1 presents the user with guidance information, for example, that recommends administering drug α, which has the highest survival probability after 90 days. Alternatively, the treatment decision support device 1 may present the user with the graph exemplified in Figure 5.

[0064] On the other hand, when the treatment decision support device 1 inputs major factor information (background major factor information excluding treatment-related information) for another target patient, which is input by the user, into the trained machine learning model (multiplying by the corresponding weights if weights are set), if the obtained prognosis information is 0.6, it identifies this major factor information as belonging to the intermediate group.

[0065] The treatment strategy decision support device 1 then sequentially inputs the major factor information (including information about treatment) belonging to the intermediate group into the trained machine learning model (multiplying by the corresponding weights if weights are set) to obtain the corresponding prognosis information (Figure 6).

[0066] Here, for example, if we assume that information regarding the prognosis when one of the antibiotics α, β, or γ is administered is available as treatment information, then, as illustrated in Figure 6, if major factor information including treatment information that drug α was administered is input, the prognosis information output by the machine learning model will be that the survival probability decreases day by day, reaching approximately 0.4 on day 90. Furthermore, in the output of the machine learning model corresponding to major factor information including treatment information that drug β was administered, the survival probability will decrease day by day, reaching approximately 0.6 on day 90. In addition, in the output of the machine learning model corresponding to major factor information including treatment information that drug γ was administered, the survival probability will decrease day by day, reaching slightly lower than 0.6 on day 90.

[0067] In this example, the treatment decision support device 1 presents the user with guidance information, for example, that recommends administering drug β, which has the highest survival probability after 90 days. Alternatively, the treatment decision support device 1 may present the user with the graph exemplified in Figure 6.

[0068] 1 Treatment plan decision support device, 11 Control unit, 12 Storage unit, 13 Operation unit, 14 Display unit, 15 Communication unit, 20 Machine learning processing unit, 21 Information collection unit, 22 Major factor selection unit, 23 Machine learning unit, 30 Estimation processing unit, 31 Information input unit, 32 Classification processing unit, 33 Classification identification unit, 34 Guidance processing unit.

Claims

1. A holding means for holding a machine learning model that has been trained to output estimated prognosis information using a set of known information consisting of major factor information selected and weighted from factor information including clinical information, test result information, medical history information, drug information, and information identifying the bacteria or viruses causing the disease, and prognosis information, with the major factor information as input information; a classification means for acquiring prognosis information corresponding to the major factor information for multiple sample patients using the machine learning model, and classifying the major factor information for the multiple sample patients into one of multiple background groups based on the acquired prognosis information, thereby classifying the multiple sample patients into groups; a classification and identification means for receiving input of the major factor information for a target patient who is the subject of treatment policy determination, acquiring prognosis information corresponding to the input target patient's major factor information using the machine learning model, and identifying the background group to which the target patient's major factor information belongs based on the acquired prognosis information, thereby identifying the group to which the target patient belongs; A treatment strategy decision support device including: a guidance means for acquiring prognosis information for each treatment strategy for sample patients belonging to the group to which the identified target patient belongs, and for guiding treatment strategies based on the acquired prognosis information.

2. A treatment strategy decision support device according to claim 1, wherein the guidance means acquires prognosis information for each drug to be administered as a treatment strategy for sample patients belonging to the group to which the identified target patient belongs, and guides a treatment strategy that includes information to identify the drug to be administered based on the acquired prognosis information.

3. A treatment strategy decision support device according to claim 1, wherein the prognosis information is information relating to the probability of survival after a predetermined number of days, and the classification means classifies the major factor information for the plurality of sample patients into one of a plurality of background groups by comparing the prognosis information obtained using the machine learning model for each sample patient with a predetermined cutoff value.

4. A computer-readable, non-temporary recording medium, comprising: a holding means for holding a machine learning model that has been trained to output estimated prognosis information using a set of known information consisting of major factor information selected from factor information including clinical information, test result information, medical history information, drug information used, and information identifying bacteria or viruses causing the disease, and weighted accordingly, and prognosis information, with the major factor information as input information; a classification means for acquiring prognosis information corresponding to the major factor information for multiple sample patients using the machine learning model, and classifying the major factor information for the multiple sample patients into one of multiple background groups based on the acquired prognosis information, thereby classifying the multiple sample patients into groups; a classification and identification means for receiving input of the major factor information for a target patient for whom a treatment plan has been decided, acquiring prognosis information corresponding to the input major factor information for the target patient using the machine learning model, and identifying the background group to which the major factor information for the target patient belongs based on the acquired prognosis information, thereby identifying the group to which the target patient belongs; A recording medium containing a program that causes a computer to function as a guidance means for acquiring prognosis information for each treatment plan for sample patients belonging to the group to which the identified target patient belongs, and for guiding treatment plans based on the acquired prognosis information.