Medical information processing device, medical information processing method, and program

Abstract

To recommend the most suitable treatment from a large number of options, in line with the patient's wishes, without placing a burden on the patient. [Solution] The medical information processing device of the embodiment comprises a patient state estimation unit and an evaluation unit. The patient state estimation unit receives at least the patient's first state and treatment candidates as input, and estimates and outputs a second state of the patient that occurs after the transition from the first state to the patient by adapting the treatment candidate to the patient. The evaluation unit inputs a plurality of different treatment candidates to the patient state estimation unit, and calculates an evaluation value for each of the plurality of treatment candidates based on the plurality of second states output and the tolerance for parameters related to the loss the patient will suffer as a result of the treatment.

Description

Technical Field

[0001] The embodiments disclosed in this specification and the drawings relate to a medical information processing apparatus, a medical information processing method, and a program.

Background Art

[0002] In order to determine a treatment policy in a medical field, it is necessary to sufficiently confirm the intention of the patient. For example, when cancer chemotherapy is performed, a doctor needs to explain the antitumor effect and side effects to the patient and obtain the patient's understanding. If a cancer drug with cardiotoxicity is excessively administered to a patient, there is a risk of serious heart disorders causing physical restrictions and economic burdens in daily life. On the other hand, if treatment with low antitumor effect is selected due to excessive caution about side effects, a sufficient treatment effect cannot be obtained, and the prognosis of the patient may deteriorate. Therefore, it is desirable that the doctor and the patient correctly understand the merits and demerits (risks) of treatment options jointly, carefully consider the balance between the antitumor effect and side effects while considering the impact on daily life after the treatment, and select the optimal treatment (in line with the patient's intention) for the patient.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Traditionally, offline reinforcement learning methods have been proposed to optimize treatment based on past clinical data. These methods utilize a treatment recommendation model that takes the patient's condition as input and outputs treatment candidates that maximize antitumor effect with minimal side effects, and a state transition model that takes the patient's condition and treatment candidates as input and outputs the next patient state. However, this method presents a limited number of treatment options, making it difficult to select the optimal treatment for the patient. On the other hand, a method combining a treatment recommendation model trained with multiple rewards weighted differently for antitumor effect and side effect prevention, along with a state transition model, could be conceivable to present a diverse range of treatment options. However, if too many treatment options are presented, it can lead to information overload, making it difficult to understand the treatment. As a result, there is a risk that an inappropriate treatment may be selected, or that treatments too far removed from the patient's values ​​may be excessively presented, leading to increased patient distrust.

[0005] The problem that the embodiments disclosed in this specification and drawings aim to solve is to recommend the optimal treatment from among numerous treatment candidates, in accordance with the patient's wishes, without burdening the patient. However, the problem that the embodiments disclosed in this specification and drawings aim to solve is not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described later can also be positioned as other problems. [Means for solving the problem]

[0006] The medical information processing device of this embodiment includes a patient state estimation unit and an evaluation unit. The patient state estimation unit receives at least the patient's first state and treatment candidates as input, and estimates and outputs a second state of the patient that occurs after the transition from the first state to the patient by adapting the treatment candidate to the patient. The evaluation unit inputs a plurality of different treatment candidates to the patient state estimation unit, and calculates an evaluation value for each of the plurality of treatment candidates based on the plurality of second states output and the tolerance for parameters related to the loss the patient will suffer as a result of the treatment. [Brief explanation of the drawing]

[0007] [Figure 1] A diagram illustrating the overview of the treatment candidate recommendation process according to the embodiment. [Figure 2] A figure showing an example of a patient preference score according to the embodiment. [Figure 3] A figure showing an example of the treatment simulation results according to the embodiment. [Figure 4] A diagram showing an example of an evaluation value (degree of reflection of intent) according to the embodiment. [Figure 5] A diagram showing an example of the configuration of a medical information processing system S including a medical information processing device 1 according to an embodiment. [Figure 6] A diagram illustrating an example of input and output data for a treatment recommendation model according to the embodiment. [Figure 7] A diagram illustrating an example of input and output data for a state transition model according to an embodiment. [Figure 8] A diagram showing an example of a treatment candidate recommendation screen according to the embodiment. [Figure 9] A flowchart showing an example of the treatment candidate recommendation process by the medical information processing device 1 according to the embodiment. [Modes for carrying out the invention]

[0008] The medical information processing device, medical information processing method, and program according to the embodiment will be described below with reference to the drawings. The medical information processing device according to the embodiment presents treatment candidates to the patient (subject) and physician (healthcare professional) that have been selected based on evaluation axes (evaluation criteria) that the patient considers important. This makes it possible to efficiently recommend the optimal treatment candidate that aligns with the patient's wishes from among a large number of treatment candidates. For example, when using anticancer drug therapy, it is possible to prevent the selection of treatment candidates that underestimate or overestimate the side effects, and to select a treatment that aligns with the patient's wishes while also having a high survival rate.

[0009] [overview] Figure 1 is a diagram illustrating the overview of the treatment candidate recommendation process according to the embodiment. The treatment candidate recommendation process includes, for example, a first step (calculation of evaluation criteria), a second step (calculation of evaluation values), a third step (presentation of treatment candidates), a fourth step (response), and a fifth step (updating evaluation values).

[0010] <Step 1 (Calculation of Evaluation Criteria)> In the first step, the patient's preferences regarding treatment plan are confirmed based on patient information, and evaluation criteria indicating the patient's preferences are calculated. Patient information includes, for example, information obtained from electronic medical records (medical history, physician's free-form comments, etc.), the patient's responses to questionnaires administered before consultation, and the contents of the medical questionnaire written by the patient (free-form comments, etc.). For example, evaluation criteria can be calculated by applying natural language processing to this patient information.

[0011] The evaluation criteria are expressed, for example, as a patient preference score. Figure 2 shows an example of a patient preference score according to an embodiment. The patient preference score indicates the degree of the patient's willingness to accept or reject each treatment impact item. Treatment impact items are parameters related to the losses the patient will suffer as a result of the treatment. Treatment impact items include, for example, treatment costs, length of hospital stay, peripheral nerve numbness, nausea, cardiotoxicity, hair loss, etc. Treatment impact items may be set according to the disease being treated. Treatment impact items may include conditions other than the patient's physiological state. In Figure 2, for example, an example of a patient who will accept the treatment costs (they don't mind if the treatment costs are high) but will reject the length of hospital stay (they want to avoid a long hospital stay) is shown. Note that the evaluation criteria may be expressed in other forms.

[0012] <Step 2 (Calculation of Evaluation Value)> In the second step, treatment simulation results are calculated for each treatment candidate, and an evaluation value is calculated for each calculated treatment simulation result (each treatment candidate). Treatment candidates include, for example, in the case of anticancer drug treatment, the type of anticancer drug, the amount of anticancer drug, the type of protective agent, the amount of protective agent, and the content of examinations. Protective agents are drugs used to reduce side effects. For example, if suppression of side effects is important, the use of protective agents will be recommended. Figure 3 is a diagram showing an example of treatment simulation results according to the embodiment. The treatment simulation results show the degree of influence for each treatment influence item. Figure 3 shows the treatment simulation results for one treatment candidate. This treatment simulation result shows that the impact on treatment costs is large (treatment costs are high), but the impact on the length of hospital stay is small (the length of hospital stay is short). Note that the treatment simulation results may be displayed in other formats.

[0013] Figure 4 shows an example of an evaluation value (degree of reflection of intention) according to the embodiment. The evaluation value indicates the degree to which the patient's intention is reflected (hereinafter referred to as "degree of reflection of intention") for each treatment candidate. For example, by comparing the patient intention score with the treatment simulation results, calculating the similarity between the two for each treatment influence item, and calculating the average of the calculated similarities, the evaluation value (degree of reflection of intention) for the treatment candidate is calculated. For example, in the patient intention score shown in Figure 2, the treatment cost is acceptable, while other influence items such as the length of hospital stay are rejected. On the other hand, in the treatment simulation results shown in Figure 3, the treatment cost has a large influence, while other influence items such as the length of hospital stay have a small influence. Therefore, the similarity for each influence item in the two is high, and as a result, a high evaluation value is calculated. In Figure 4, for example, an example is shown in which treatment candidates A and E have high evaluation values, while treatment candidate F has a low evaluation value.

[0014] <Step 3 (Presenting treatment options)> In the third step, based on the calculated evaluation value, treatment candidates with a high evaluation value that reflect the patient's intention are presented to the patient and the doctor. In the example of FIG. 4, for example, treatment candidates A and E with relatively high evaluation values are presented to the patient and the doctor. The treatment candidates are displayed, for example, on the terminal devices of the doctor and the patient as a treatment candidate recommendation screen.

[0015] <Fourth Step (Response)> In the fourth step, among the presented treatment candidates, the patient's selection (response) of a treatment candidate is received. For example, based on the screen operation by the patient (or the doctor who confirmed the patient's intention) on the treatment candidate recommendation screen displayed on the terminal device, the patient's selection (response) of a treatment candidate is received. Here, it is possible that the patient may not be convinced by the presented treatment candidates. In this case, the patient (or the doctor who asked about the patient's intention) performs a screen operation on the treatment candidate recommendation screen and inputs information about the patient's further intention.

[0016] <Fifth Step (Update of Evaluation Value)> In the fifth step, based on the response result obtained in the above fourth step, the evaluation value is updated. Thereafter, the process returns to the above third step, and updated treatment candidates based on the updated evaluation value are presented to the patient and the doctor. By repeating these third to fifth steps until the patient is convinced, it becomes possible to finally select an optimal treatment in line with the patient's intention.

[0017] [System Configuration] Next, the configuration of the medical information processing system according to the embodiment will be described. FIG. 5 is a diagram showing an example of the configuration of a medical information processing system S including a medical information processing apparatus 1 according to the embodiment. The medical information processing system S includes, for example, a medical information processing apparatus 1 and at least one terminal device T. The medical information processing apparatus 1 and the terminal device T are connected so as to be capable of transmitting and receiving data via, for example, a communication network NW. The communication network NW includes, in addition to a wireless / wired LAN such as a hospital backbone LAN (Local Area Network) or the Internet, a telephone communication line network, an optical fiber communication network, a cable communication network, a satellite communication network, and the like.

[0018] <Medical information processing apparatus> [[ID=⑥]]The medical information processing apparatus 1 controls the overall operation of the medical information processing system S. The medical information processing apparatus 1 is an example of a "medical information processing apparatus". The medical information processing apparatus 1 may be, for example, a workstation, a server, or the like. The medical information processing apparatus 1 includes, for example, a processing circuit 10, a communication interface 20, an input interface 30, and a memory 40.

[0019] The processing circuit 10 includes, for example, an acquisition function 11, a treatment recommendation function 12, a patient state estimation function 13, a treatment candidate identification function 14, an evaluation function 15, a treatment optimization function 16, and a display control function 17. The processing circuit 10 realizes these functions by, for example, a hardware processor (computer) executing a program stored in the memory 40 (storage circuit).

[0020] It should be noted that in the original text, the numbering of some elements seems a bit irregular. In the translation, I have tried to follow the original structure as closely as possible while making the English text grammatically correct. Also, for the tags like , they are left unchanged as required.A hardware processor refers to circuits such as a CPU (Central Processing Unit), GPU (Graphics Processing Unit), Application Specific Integrated Circuit (ASIC), and programmable logic devices (e.g., Simple Programmable Logic Device (SPLD) or Complex Programmable Logic Device (CPLD), Field Programmable Gate Array (FPGA)). Instead of storing the program in memory 40, the hardware processor may be configured to directly embed the program within its circuitry. In this case, the hardware processor functions by reading and executing the program embedded within its circuitry.

[0021] The above program may be stored in memory 40 in advance, or it may be stored in a non-temporary storage medium such as a DVD or CD-ROM, and installed from the non-temporary storage medium to memory 40 when the non-temporary storage medium is mounted in the drive device (not shown) of the medical information processing device 1. The hardware processor is not limited to being configured as a single circuit, but may be configured as a single hardware processor by combining multiple independent circuits to realize each function. Alternatively, multiple components may be integrated into a single hardware processor to realize each function.

[0022] The acquisition function 11 acquires various information from external devices via a communication network NW. For example, the acquisition function 11 acquires instruction input from a doctor or patient from a terminal device T. The acquisition function 11 also acquires information from terminal device T regarding treatment impact items set based on the patient's wishes. Treatment impact items may be specified by the patient or doctor, or they may be predefined as a list of treatment impact items. The acquisition function 11 also acquires from external devices (not shown) such as various patient test data (measured values, images, etc.), electronic medical records, and response data to questionnaires. The acquisition function 11 is an example of an "acquisition unit".

[0023] The treatment recommendation function 12 takes at least the patient status and treatment impact items acquired by the acquisition function 11 as input and outputs treatment candidates. The treatment recommendation function 12 outputs treatment candidates using, for example, multiple treatment recommendation models M1 (reinforcement learning models) that have been trained with reward functions with different weightings that are pre-stored in memory 40. In other words, these multiple treatment recommendation models are prepared assuming different treatment policies. For example, the first treatment recommendation model M11 is trained with a reward weighting ratio of "cancer response: side effect improvement = 10:1" in accordance with a cancer treatment-focused treatment policy. The second treatment recommendation model M12 is trained with a reward weighting ratio of "cancer response: side effect improvement = 3:7" in accordance with a side effect avoidance treatment policy. The third treatment recommendation model M13 is trained with a reward weighting ratio of "cancer response: side effect improvement = 5:5" in accordance with a treatment policy that balances side effects and cancer treatment. The treatment recommendation function 12 is an example of a "treatment recommendation unit".

[0024] Figure 6 illustrates an example of input and output data for a treatment recommendation model M1 according to an embodiment. The treatment recommendation model M1 uses patient status as input data and treatment candidates as output data. Patient status includes information about the patient's disease, such as "cancer stage," and the influence of treatment-influencing items. For example, when designing the treatment recommendation model M1, if the input data includes information about one disease and the influence of five types of treatment-influencing items (e.g., treatment cost, length of hospital stay, peripheral nerve numbness, nausea, cardiotoxicity), the input data will have a total of six dimensions. The output data, the treatment candidates, indicates, for example, the probability of obtaining high future compensation when applying each treatment candidate. For example, when designing the treatment recommendation model M1, if the output data is set to N types of treatment candidates, the output data will have N dimensions. The output of the cancer-focused first treatment recommendation model M11 will be treatment candidates where cancer treatment is prioritized (the probability of treatment candidates with a high cancer response rate is higher). On the other hand, the output of the second treatment recommendation model M12, which prioritizes avoiding side effects, will be treatment candidates that prioritize avoiding side effects (the probability of treatment candidates with a high rate of side effect improvement will be higher). In this way, each of the multiple treatment recommendation models, trained with reward functions with different weightings, will output different treatment candidates from each other.

[0025] The degree of influence of treatment impact items is determined according to the content of the treatment candidate, the type of disease, the state of the disease, etc. The degree of influence of treatment impact items may be determined by the physician or by following guidelines. The degree of influence of treatment impact items may be expressed by a degree (numerical level), such as large, medium, or small. For example, if the treatment candidate set as the output of treatment recommendation model M1 is "anti-cancer drug treatment," the degree of influence of treatment costs may be "high," and the degree of influence of hospitalization days may be "medium," and so on, according to the trend of the treatment candidate. If the treatment impact items include items related to side effects, the degree of those side effects (e.g., severe nausea, mild nausea) may also be entered.

[0026] The patient state estimation function 13 takes at least the current patient state of the patient to be treated (also called the "first state") and the treatment candidates output by the treatment recommendation function 12 as input, and outputs the next patient state (also called the "second state of the patient, which is later in time than the first state" or the "future patient state") that the patient has transitioned to by applying the treatment candidates to the patient. The patient state estimation function 13 outputs the next patient state using, for example, a state transition model M2 that is pre-stored in memory 40. The state transition model M2 is, for example, a neural network or a regression model such as random forest regression. Figure 7 is a diagram illustrating an example of the input and output data of the state transition model M2 according to the embodiment. The input data of the state transition model M2 is the current patient state and the treatment candidates output by the treatment recommendation model M1, and the output data is the next patient state that the patient has transitioned to by treatment with these treatment candidates. For example, if the current patient state in the input data is "cancer stage", the next patient state in the output data is "next cancer stage". Furthermore, for example, if the input data is "the degree of peripheral system side effects" for the current patient state, the output data is "the degree of the next peripheral system side effects." Note that the state transition model M2 may be a single model, or it may be multiple models prepared for each side effect. The patient state estimation function 13 is an example of a "patient state estimation unit."

[0027] The treatment candidate identification function 14 identifies treatment candidates that are likely to improve the prognosis for the disease to be treated, from among the multiple treatment candidates output by the treatment recommendation function 12, based on the following patient conditions output by the patient condition estimation function 13. The treatment candidate identification function 14 controls the treatment recommendation function 12 and the patient condition estimation function 13 (treatment recommendation model M1 and state transition model M2) to simulate the patient's treatment progress. If the simulation results show that the disease to be treated has progressed to complete recovery, the treatment candidate identification function 14 saves that treatment candidate and the treatment simulation results (influence of treatment influence items) in memory 40. The influence of treatment influence items may be obtained at multiple states (timings). The influence of treatment influence items may be obtained, for example, at the final state, the worst state during treatment, or as an average value. The treatment candidate identification function 14 is an example of a "treatment candidate identification unit".

[0028] The evaluation function 15 calculates an evaluation value for each of the multiple treatment candidates based on the multiple next patient states output by inputting multiple different treatment candidates to the patient state estimation function 13, and the tolerance for treatment impact items. The evaluation function 15 calculates an evaluation value for each treatment candidate based on evaluation criteria that are important to the patient (from the patient's perspective). An example of a patient-perspective evaluation criterion is a patient preference score calculated based on patient information (see Figure 2). First, the evaluation function 15 calculates the similarity of the output results of the natural language processing model for each treatment impact item and patient information ("patient-related text," "preference information"), and uses this as the patient preference score. Similarity calculation methods include, for example, cosine similarity and mean squared error. The natural language processing model includes, for example, BERT (Bidirectional Encoder Representations from Transformers). For example, "length of hospital stay" and "participation in recent events" have a high correlation, so the similarity of the output results is calculated to be high. Next, the evaluation function 15 calculates the similarity between the treatment simulation results of the treatment candidates and the patient preference score, and uses this as the evaluation value. The calculated patient preference score is provided to the user (doctor, patient, patient's family, etc.) and may even be modified. Sets of text and treatment impact items with high evaluation values ​​may be recorded and provided via screen display. The evaluation function 15 is an example of an "evaluation unit." The patient preference score is an example of "acceptability." That is, the evaluation function 15 calculates the evaluation value for each of the multiple treatment candidates based on the similarity between the patient preference score (acceptability) for the calculated treatment impact items (parameters) and the next patient state (second state) estimated in correspondence with the treatment impact items (parameters).

[0029] Furthermore, the evaluation function 15 may calculate evaluation values ​​for treatment candidates based on evaluation criteria (degree of real-world reflection of treatment) that are important to the physician (from the physician's perspective). The evaluation function 15 calculates other evaluation values ​​for each of the multiple treatment candidates based on the treatment candidates based on the treatment criteria and each of the multiple treatment candidates output by the treatment recommendation function 12. For example, the evaluation function 15 quantifies how close each of the multiple treatment candidates is to the treatment criteria (calculates similarity). Treatment criteria include, for example, general treatments defined in guidelines, treatments included in past history data, most frequently occurring treatments, representative treatments, and treatments specified by the physician. Similarity is calculated based on, for example, the edit (Levenshtein) distance, the minimum number of steps required to transform one string into the other string by inserting, deleting, or replacing one character, etc. The evaluation function 15 replaces the elements included in the reference treatment and treatment candidates included in the treatment criteria with single English words that do not overlap. Then, the evaluation function 15 calculates the edit distance between the reference treatment and the treatment candidates. Furthermore, the similarity may be calculated, for example, using an autoencoder, or based on the cosine similarity of the intermediate layer (latent variable) of the autoencoder (compressed and restored using the treatment as input).

[0030] The treatment optimization function 16 identifies at least one treatment candidate to present to the patient based on the evaluation value for each of the multiple treatment candidates calculated by the evaluation function 15. The treatment optimization function 16 narrows down the treatment candidates through a cycle of presentation to the patient and selection by the patient. For example, if the treatment optimization function 16 uses the degree of preference reflection as the evaluation value, it selects N treatment candidates with the highest degree of preference reflection, presents the patient with a treatment candidate recommendation screen showing the selected N treatment candidates and the next patient condition, and allows the patient to select the treatment candidate they want. Next, the treatment optimization function 16 updates the degree of preference reflection based on the selection result. The treatment optimization function 16 updates the degree of preference reflection so that the degree of preference reflection increases for selected candidates and decreases for unselected candidates. The treatment optimization function 16 is an example of a "treatment optimization unit".

[0031] Figure 8 shows an example of a treatment candidate recommendation screen according to the embodiment. In the example shown in Figure 8, the treatment candidate recommendation screen includes three treatment candidates (A, B, C) identified based on evaluation values, and treatment candidate B is selected through an operation on this treatment candidate recommendation screen (e.g., mouse operation). The treatment impact item list column graphically displays information about the treatment simulation results (degree of impact of treatment impact items (large, small)) performed for the selected treatment candidate B. By referring to the degree of impact of these treatment impact items, the patient can intuitively understand the degree of impact of the treatment impact items for the selected treatment candidate. In addition, treatment impact items can be added and deleted by operating the "Add" and "Delete" buttons included in the treatment impact item list column. Furthermore, the description field displays text included in the patient information, such as "The patient strongly desires to return to work, ~". In addition, cautionary statements such as "Note: This treatment may affect administrative work after returning to work" are displayed in relation to the content of this description field. These cautionary statements are prepared in advance, for example, in template format. When the "Confirm" button, which is displayed in association with each treatment option, is pressed, the treatment option is finally decided.

[0032] The treatment optimization function 16 narrows down the treatment candidates by repeatedly presenting treatment candidates to the patient, allowing the patient to make a selection, and updating the evaluation value (degree of preference reflection), as described above. When the patient identifies a final treatment candidate, the processing of the treatment optimization function 16 ends. Regarding the treatment candidates presented, an element of exploration may be included, for example, by presenting three treatment candidates, two of which are the top two in evaluation value (degree of preference reflection), and one is presented randomly.

[0033] The display control function 17 controls the display of a treatment candidate recommendation screen, which shows the treatment candidates selected by the treatment optimization function 16, on the terminal device T. The display control function 17 is an example of a "display control unit".

[0034] The communication interface 20 communicates with external devices such as terminal devices T via the communication network NW. The communication interface 20 includes, for example, a communication interface such as a NIC (Network Interface Card).

[0035] The input interface 30 receives various input operations from the administrator of the medical information processing device 1, converts the received input operations into electrical signals, and outputs them to the processing circuit 10. The input interface 30 includes, for example, a mouse, keyboard, etc.

[0036] Memory 40 can be implemented by semiconductor memory elements such as RAM (Random Access Memory), flash memory, hard disks, or optical discs. These non-transient storage media may also be implemented by other storage devices connected via a communication network NW, such as NAS (Network Attached Storage) or external storage server devices. Memory 40 may also include non-transient storage media such as ROM (Read Only Memory) or registers. Memory 40 stores, for example, a treatment recommendation model M1, a state transition model M2, etc. In addition, memory 40 stores programs, parameter data, and other data used by the processing circuit 10. Memory 40 is an example of a "storage device".

[0037] <Medical Database> Medical database 3 is a storage device that stores various information about a patient. For example, medical database 3 stores various patient test data (measurements, images, etc.), electronic medical records, and response data to questionnaires. Medical database 3 can be implemented using semiconductor memory elements such as RAM and flash memory, hard disks, or optical discs. Medical database 3 may also be configured to be distributed across multiple devices.

[0038] <Terminal device> Terminal device T is a device for accessing various information provided by the medical information processing device 1. Terminal device T is operated, for example, by a doctor D or a patient P. Terminal device T is, for example, a personal computer, a tablet, a smartphone or other mobile device. Terminal device T is equipped with, for example, a communication function for data communication with other devices, an input interface function for receiving various instructions from doctor D or patient P, and a display function for displaying various information. In addition, another terminal device operated by patient P that has the same functions as terminal device T may be used instead of (or in addition to) terminal device T. For example, in online medical consultations where doctor D and patient P are in separate locations, both doctor D's terminal device and patient P's terminal device may be used. Terminal device T is an example of a "display device".

[0039] [Processing flow] Next, the sequence of steps in the treatment candidate recommendation process in the medical information processing device 1 will be described. Figure 9 is a diagram showing an example of the flow of the treatment candidate recommendation process by the medical information processing device 1 according to this embodiment. The treatment candidate recommendation process shown in Figure 9 is started, for example, when a doctor D who is treating patient P in a hospital examination room gives an instruction to start the treatment candidate recommendation process via the input interface of terminal device T.

[0040] First, the acquisition function 11 acquires patient status information from, for example, the medical database 3 or terminal device T (step S101). Patient status includes, for example, the patient's disease status such as "cancer stage" and the degree of influence of treatment-affecting items. For example, the acquisition function 11 uses a patient identifier (patient ID) that identifies the patient, entered by a physician D operating the terminal device T, as a key to acquire the patient's disease status from the medical database 3.

[0041] Next, the treatment recommendation function 12 generates multiple treatment candidates based on the patient status acquired by the acquisition function 11 (step S103). The treatment recommendation function 12 generates multiple treatment candidates, for example, using multiple treatment recommendation models M1 that have been trained with different weighted reward functions pre-stored in memory 40.

[0042] Next, the patient state estimation function 13 estimates the patient state after treatment (the next patient state) for each of the multiple treatment candidates based on the patient state acquired by the acquisition function 11 and the multiple treatment candidates generated by the treatment recommendation function 12 (step S105). The patient state estimation function 13 estimates the next patient state using, for example, a state transition model M2 that is pre-stored in memory 40.

[0043] Next, the treatment candidate identification function 14 narrows down the treatment candidates by identifying treatment candidates that are likely to improve the prognosis for the disease to be treated, based on the patient condition estimated by the patient condition estimation function 13 (step S107). The treatment candidate identification function 14 controls the treatment recommendation function 12 and the patient condition estimation function 13 (treatment recommendation model M1 and state transition model M2) to simulate the treatment progress of the target patient. If the simulation results show that the disease to be treated progresses to complete recovery, the treatment candidate identification function 14 saves the pair of that treatment candidate and the treatment simulation result (influence of treatment influence items) in memory 40.

[0044] Next, the evaluation function 15 calculates evaluation values ​​for the treatment candidates identified by the treatment candidate identification function 14 (step S109). For example, the evaluation function 15 calculates the similarity of the output results of the natural language processing model to each treatment impact item and patient information, and calculates this as a patient preference score. Next, the evaluation function 15 calculates the similarity between the treatment simulation results of the treatment candidates estimated by the patient condition estimation function 13 and the patient preference score, and uses this as an evaluation value.

[0045] Next, the treatment optimization function 16 identifies treatment candidates to present to the patient based on the evaluation value for each of the identified treatment candidates, and the display control function 17 outputs information of the treatment candidate recommendation screen, which includes the identified multiple treatment candidates, to the terminal device T (step S111). For example, the treatment optimization function 16 identifies the top N treatment candidates based on their evaluation values ​​as multiple treatment candidates to present to the patient. As a result, the terminal device T displays a treatment candidate recommendation screen that includes the identified multiple treatment candidates.

[0046] Next, the treatment optimization function 16 determines whether or not it has received an instruction to change the treatment candidates based on the patient's (or the doctor's) actions on the treatment candidate recommendation screen displayed on the terminal device T (step S113). Instructions to change the treatment candidates include, for example, instructions to change the presented treatment candidates, inputting evaluation results for each of the presented treatment candidates, and adding and deleting treatment influence items. If it is determined that an instruction to change the treatment candidates has been received (step S113; YES), the evaluation function 15 updates the evaluation values, and the treatment optimization function 16 changes or re-narrows the treatment candidates (step S117).

[0047] On the other hand, if the treatment optimization function 16 determines that it has not accepted the instruction to change the treatment candidate (step S113; NO), it accepts a treatment decision instruction (for example, pressing the "Confirm" button) based on the patient's (or the doctor who has confirmed the patient's intentions) operation on the treatment candidate recommendation screen (step S115), and the processing of this flowchart ends.

[0048] According to the embodiments described above, it is possible to recommend the optimal treatment from among numerous treatment options in accordance with the patient's wishes, without burdening the patient. For example, it is possible to prevent treatment selection that underestimates or overestimates side effects, and to select a treatment that is in line with the patient's wishes while also having a high survival rate.

[0049] While several embodiments have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These embodiments can be carried out in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents. [Explanation of Symbols]

[0050] 1. Medical Information Processing Device 3. Medical Databases 10 Processing Circuit 11. Acquisition function 12 Treatment recommendation function 13. Patient Status Estimation Function 14. Treatment candidate identification function 15. Rating function 16 Treatment Optimization Function 17 Display Control Function 20 Communication Interfaces 30 Input Interfaces 40 memory M1 Treatment Recommendation Model M2 State Transition Model M11 First Treatment Recommendation Model M12 Second Treatment Recommendation Model M13 Third Treatment Recommendation Model NW (Network Communication Network) S Medical Information Processing System T terminal device

Claims

1. A patient state estimation unit that takes at least the patient's first state and a treatment candidate as input, estimates and outputs a second state of the patient that occurs in a time period later than the first state, which is the state to which the patient has transitioned by applying the treatment candidate to the patient. An evaluation unit calculates an evaluation value for each of the multiple treatment candidates based on a plurality of second states output by inputting a plurality of different treatment candidates to the patient state estimation unit, and an tolerance for parameters related to the loss the patient will suffer as a result of the treatment. A medical information processing device equipped with [a specific feature].

2. The system further includes a treatment recommendation unit that inputs at least the first state and the parameters and outputs the treatment candidate, The patient condition estimation unit acquires the treatment candidates output by the treatment recommendation unit as input. The medical information processing device according to claim 1.

3. The system further includes a treatment candidate identification unit that identifies, from among a plurality of treatment candidates, a treatment candidate that is likely to improve the prognosis for the disease to be treated, based on the second state output by the patient state estimation unit. The medical information processing device according to claim 1.

4. The system further includes an acquisition unit that acquires intention information indicating the patient's intentions regarding treatment, The evaluation unit, Based on the acquired intention information, the tolerance for each parameter is calculated. Based on the tolerance for the calculated parameters and the similarity to the second state estimated in relation to the parameters, an evaluation value is calculated for each of the multiple treatment candidates. The medical information processing device according to claim 1.

5. The system further includes a treatment optimization unit that identifies a treatment candidate to present to the patient based on the evaluation value of each of the multiple treatment candidates. A medical information processing device according to any one of claims 1 to 4.

6. The system further includes a display control unit that causes a display device to display a treatment candidate recommendation screen, which includes treatment candidates to be presented to the identified patient. The medical information processing device according to claim 5.

7. The aforementioned treatment candidate recommendation screen includes information indicating the degree of influence of the parameters associated with the treatment candidates presented to the identified patient. The medical information processing device according to claim 6.

8. The patient state estimation unit estimates and outputs the second state using a regression model that has been trained to output the second state when the first state and the treatment candidate are input. A medical information processing device according to any one of claims 1 to 4.

9. The treatment recommendation unit outputs multiple treatment candidates using multiple reinforcement learning models trained with reward functions having different weights. The medical information processing device according to claim 2.

10. The evaluation unit calculates other evaluation values ​​for each of the multiple treatment candidates based on the treatment candidate based on the treatment criteria and each of the multiple treatment candidates. A medical information processing device according to any one of claims 1 to 4.

11. Computers The system obtains at least the patient's first state and a treatment candidate, and estimates and outputs the patient's second state, which occurs after the first state and is achieved by applying the treatment candidate to the patient. Based on the multiple second states output based on the multiple different treatment candidates, and the tolerance for parameters related to the loss the patient will suffer as a result of the treatment, an evaluation value is calculated for each of the multiple treatment candidates. Medical information processing method.

12. On the computer, The system obtains at least the patient's first state and a treatment candidate, and estimates and outputs the patient's second state, which occurs at a time later than the first state, resulting from applying the treatment candidate to the patient. Based on the multiple second states output based on the multiple different treatment candidates, and the tolerance for parameters related to the loss the patient will suffer as a result of the treatment, an evaluation value for each of the multiple treatment candidates is calculated. program.

Images