State evaluation device, state evaluation method, and state evaluation program

Abstract

[Problem] To provide a state evaluation device with which it is possible to easily perform, on the basis of input of record information in a discretionary text format, quantitative evaluation without variation in the state of a subject. [Solution] This state evaluation device includes: a classification unit including a machine-learned language model that, in response to input of one or more items of record information for a prescribed period that are created in a discretionary text format and pertain to a subject, determines and outputs the applicability of a prescribed label for each item of record information; and an evaluation unit that evaluates the state of the subject on the basis of the result of determining the applicability for the prescribed period.

Description

Condition assessment device, condition assessment method, and condition assessment program

[0001] The present invention relates to a condition assessment device, a condition assessment method, and a condition assessment program.

[0002] Japan has seen a remarkable increase in life expectancy due to improvements in living standards, sanitary conditions, and medical standards that came with the rapid economic growth after the war. This, combined with a declining birth rate, has led to an aging society with a high aging rate. In such an aging society, it is expected that the number of people who require care, such as nursing care, due to illness, injury, and aging, will increase.

[0003] In nursing care facilities, multiple care staff work in shifts to provide care to one patient. Care staff prepares care records to enable each care staff member to understand the condition of each patient. Care records may be prepared in any written format. In some cases, the care records are required to include a quantitative assessment of each patient's condition to demonstrate the effectiveness of the care plan.

[0004] However, because the quantitative assessment of each subject's condition is performed independently by the care staff, there is a possibility that variations will occur and the reliability of the assessment results may be low.

[0005] The following prior art is disclosed in Patent Document 1 listed below: Fixed phrase information that sets fixed phrases to be used for entering care information is stored in a storage unit. Fixed phrases are displayed in the care record field of the care record input screen so that they can be selected using a pull-down menu. The fixed phrase selected by the caregiver is displayed in the care record field. Input is accepted into variable input fields included in the fixed phrase. Then, the care record in which input has been made into the variable input fields is stored in the care record information.

[0006] Patent Document 2 below discloses the following prior art: A medical ontology and input data entered in natural language format in a free-form comment field of a medical report are acquired. The medical ontology is a database that stores relationships between medical languages ​​and relationships between medical languages ​​and medical classifications. Natural language processing is performed on the input data to generate analysis data of the input data. Structured data in which the medical ontology is associated with the analysis data is generated. Confirmation data that displays the structured data in natural language format is generated. The confirmation data is then displayed on a display. The confirmation data may include an assessment of the possibility of cerebral infarction, etc.

[0007] JP 2022-103155 A JP 2022-54218 A

[0008] However, the prior art described in Patent Document 1 requires the nursing care record to be written using standard phrases, which may prevent flexible and appropriate nursing care records from being created. Furthermore, training costs for care staff are incurred due to changes in the method of writing nursing care records. Furthermore, past nursing care records cannot be effectively utilized, making it impossible to quantitatively evaluate the condition of the subject.

[0009] The prior art described in Cited Document 2 allows for the input of medical reports and other information in natural language format. However, it requires the preparation of a medical ontology. Creating a medical ontology requires high levels of technical skill in both medical and statistical fields, and there is a possibility that it may not be sufficiently comprehensive in terms of operation.

[0010] The present invention has been made to solve the above-mentioned problems, and aims to provide a condition assessment device, a condition assessment method, and a condition assessment program that can easily perform a consistent quantitative assessment of a subject's condition based on input of record information in any text format.

[0011] The above-mentioned problems of the present invention are solved by the following means.

[0012] (1) A condition evaluation device having: a classification unit including a machine-learned language model that receives input of one or more pieces of recorded information about a subject over a predetermined period, written in any sentence format, and determines whether each piece of recorded information corresponds to a predetermined label and outputs the determined result; and an evaluation unit that evaluates the condition of the subject based on the result of the determination for the predetermined period.

[0013] (2) The condition evaluation device described in (1) above, wherein the evaluation unit uses an approximate formula to evaluate the condition of the subject based on the result of the corresponding judgment.

[0014] (3) A condition evaluation device as described in (2) above, wherein the approximation formula is created from the relationship between the number of occurrences of the specified label in the specified period of the subject whose condition is known and the known condition.

[0015] (4) The evaluation unit outputs the condition of the subject in response to the input of the number of hits of the specified label in the specified period, and has a display unit that displays the number of hits of the specified label in the specified period along with the condition of the subject evaluated by the evaluation unit. The condition evaluation device described in (3) above has a display unit that displays the number of hits of the specified label in the specified period.

[0016] (5) A condition assessment device as described in (1) above, in which the language model is machine-learned based on the recorded information for the specified period written in any sentence format and the results of a known judgment of whether the specified label applies to the recorded information.

[0017] (6) The condition assessment device described in (1) above, wherein the record information is a care record and the condition is an assessment for each item of a dementia behavioral disorder scale.

[0018] (7) The state evaluation device according to (1) above, wherein the language model further outputs a likelihood for each predetermined label for each piece of record information in response to input of one or more pieces of record information for the predetermined period.

[0019] (8) The evaluation unit weights the result of the judgment of whether or not the subject is a candidate for the specified period of time based on the likelihood, and outputs the subject's condition in response to input of the weighted result of the judgment of whether or not the subject is a candidate for the specified period of time.

[0020] (9) A condition evaluation method comprising the steps of: (a) inputting one or more pieces of record information relating to a subject over a predetermined period, written in any sentence format, and using a machine-learned language model to make a judgment as to whether or not a predetermined label applies to each piece of record information and output the judgment; and (b) evaluating the condition of the subject based on the result of the judgment as to whether or not the record information applies to the predetermined period in step (a).

[0021] (10) A condition evaluation method according to (9) above, wherein in step (b), an approximate formula is used to evaluate the condition of the subject based on the result of the yes / no determination.

[0022] (11) A condition evaluation method as described in (10) above, in which the approximation formula is created from the relationship between the number of occurrences of the specified label in the specified period of the subject whose condition is known and the known condition.

[0023] (12) A condition evaluation method as described in (11) above, wherein in step (b), the condition of the subject is output in response to input of the number of hits of the specified label in the specified period, and the number of hits of the specified label in the specified period is displayed together with the condition of the subject evaluated in step (b).

[0024] (13) A condition assessment method as described in (9) above, in which the language model is machine-learned based on the recorded information for the specified period written in any sentence format and the results of a known judgment of whether the specified label applies to the recorded information.

[0025] (14) The condition assessment method described in (9) above, wherein the record information is a care record and the condition is an assessment for each item of the dementia behavioral disorder scale.

[0026] (15) The state evaluation method described in (9) above, wherein the language model further outputs a likelihood for each predetermined label for each piece of record information in response to input of one or more pieces of record information for the predetermined period.

[0027] (16) In the step (b), the result of the determination of whether or not the subject is a candidate for the specified period is weighted by the likelihood, and the condition of the subject is output in response to the input of the weighted result of the determination of whether or not the subject is a candidate for the specified period.

[0028] (17) A condition evaluation program for executing the condition evaluation method according to any one of (9) to (16) above by a computer.

[0029] Using a machine-learned language model, the system inputs record information for a specified period created in any sentence format, and outputs a judgment as to whether each record information corresponds to a specified label.The system then evaluates the subject's condition based on the results of the judgment.This makes it easy to perform a consistent quantitative evaluation of the subject's condition based on the input record information in any sentence format.

[0030] 1 is a diagram showing the overall configuration of an information processing system. FIG. 1 is a block diagram showing a schematic configuration of a server. FIG. 2 is a block diagram showing the functions of a control unit. FIG. 3 is an explanatory diagram showing information processing by a control unit. FIG. 4 is a diagram showing 13 types of items in DBD13. FIG. 5 is an explanatory diagram showing an example of a determination of applicability of an item in DBD13 for each care record. FIG. 6 is a diagram showing an example of a graph showing the relationship between the number of hits for one item in DBD13 and the known condition of a subject. FIG. 7 is a block diagram showing the schematic configuration of a mobile terminal. FIG. 8 is a diagram showing an example of a display screen displayed on a mobile terminal, which displays the condition of a subject in association with the number of hits for item 9 in DBD13 over a one-year period. FIG. 9 is a diagram showing another example of a display screen displayed on a mobile terminal, which displays the condition of a subject in association with the number of hits for item 9 in DBD13 over a one-year period. FIG. 10 is a flowchart showing the operation of a server. FIG. 11 is a diagram showing recorded information, predetermined labels, and the condition of a subject to be evaluated in Modification 1. FIG. 12 is a diagram showing recorded information, predetermined labels, and the condition of a subject to be evaluated in Modification 2. FIG. 13 is a diagram showing recorded information, predetermined labels, and the condition of a subject to be evaluated in Modification 3.

[0031] A condition assessment device, a condition assessment method, and a condition assessment program according to an embodiment of the present invention will be described below with reference to the drawings. In the drawings, identical elements are designated by the same reference numerals, and duplicate explanations will be omitted. Also, the dimensional proportions in the drawings are exaggerated for the sake of explanation, and may differ from the actual proportions.

[0032] In this specification, the subject broadly includes care recipients who receive care such as nursing care or caregiving, people with disabilities who do not require care, healthy people who have previously experienced an injury or surgery, etc. The subject includes, for example, people who require care, people who require assistance, and patients. For simplicity of explanation, the following description will be given taking the case where the subject is a person who requires care as an example.

[0033] First Embodiment FIG. 1 is a diagram showing the overall configuration of an information processing system 1. As shown in FIG. 1, the information processing system 1 includes a server 10, a fixed terminal 20, and one or more mobile terminals 30. These are connected to each other via a network 40, such as a local area network (LAN), a telephone network, or a data communication network, by wire or wirelessly, so that they can communicate with each other. The network 40 may include a repeater that relays communication signals. In the example shown in FIG. 1, the server 10, the mobile terminal 30, and the fixed terminal 20 are connected to each other via a network 40, such as a wireless LAN (e.g., a LAN conforming to the IEEE 802.11 standard) that includes an access point 41, so that they can communicate with each other.

[0034] The mobile terminal 30 can be carried by each of the care staff 50 who care for the subject.

[0035] The information processing system 1 may be configured only by the server 10. The information processing system 1 configures a condition evaluation device.

[0036] The server 10 and the fixed terminal 20 are preferably installed in a building such as a welfare facility for the elderly or a hospital. The server 10 may be configured as an on-premise server or a cloud server. The server 10 may also be configured as a standalone PC (Personal Computer).

[0037] (Server 10) Fig. 2 is a block diagram showing a schematic configuration of the server 10. The server 10 includes a control unit 11, a communication unit 12, and a storage unit 13. These components are interconnected by a bus. The server 10 constitutes an information processing device.

[0038] The control unit 11 is configured with a CPU (Central Processing Unit) and memories such as RAM (Random Access Memory) and ROM (Read Only Memory), and controls each part of the server 10 and performs arithmetic processing according to a program.

[0039] The communication unit 12 is an interface for communicating with other devices including, for example, the fixed terminal 20 and the mobile terminal 30 via the network 40, and may be, for example, a LAN card.

[0040] The storage unit 13 is configured by a HDD, an SSD, etc. The storage unit 13 stores various programs and various data.

[0041] Fig. 3 is a block diagram showing the functions of the control unit 11. Fig. 4 is an explanatory diagram showing information processing by the control unit 11. By executing a program, the control unit 11 functions as a recorded information receiving unit 111, a classification unit 112, an evaluation unit 113, a display control unit 114, a learning unit 115, and an approximate expression creation unit 116. The display control unit 114 constitutes the display unit.

[0042] The record information receiving unit 111 receives record information. The record information includes various records such as care records, diagnosis records (diagnosis reports), and follow-up observation records. For simplicity of explanation, the following description will be given taking the case where the record information is a care record as an example.

[0043] The record information receiving unit 111 receives care records created in any text format. That is, the record information receiving unit 111 receives care records created in non-standard sentences and standard sentences. The care records may be input by the care staff 50 into each mobile terminal 30 and received by the communication unit 12. The care records are input by the care staff 50 into the mobile terminal 30 for each subject at least once a day. That is, the record information receiving unit 111 receives one or more care records for one subject.

[0044] The classification unit 112 determines whether a predetermined label applies to each input care record and outputs the result. That is, the classification unit 112 classifies the care record into the corresponding label by determining whether a predetermined label applies to the care record. One care record can be classified into one or more labels. The classification unit 112 includes a machine-learned language model. Any language model can be used as long as it determines whether a predetermined label applies to each input care record and outputs the result. For example, BERT (Bidirectional Encoder Representations from Transformers) is used as the language model. The predetermined labels include any labels corresponding to the subject's evaluation. For simplicity of explanation, the following description will be given using an example in which the predetermined labels are the 13 types of items (evaluation items) of the Dementia Behavior Scale (DBD) 13. The DBD13, which is the Dementia Behavioral Disorders Scale, is an evaluation index that can concisely detect peripheral symptoms (behavioral and psychological symptoms) of dementia. As will be described later, the language model is machine-learned by the learning unit 115. Hereinafter, the machine-learned language model is also referred to as a "classification model."

[0045] FIG. 5 is a diagram showing 13 types of items in DBD13. The items in DBD13 include, for example, 4. "Getting up in the middle of the night for no particular reason" (nighttime waking up) and 7. "Walking around a lot" (walking around). Further items in DBD13 include 8. "Repeating the same action over and over" (repetitive behavior) and 9. "Using abusive language" (abusive language). Further items in DBD13 include 10. "Wearing inappropriate clothing that is out of place or not appropriate for the season" (inappropriate clothing).

[0046] The classification unit 112 determines whether or not "asking the same thing over and over again" applies to item 1 of the DBD13 for each care record and outputs the result. For example, if "asking the same thing over and over again" applies, the classification unit 112 outputs "1", and if not, the classification unit 112 outputs "0". The classification unit 112 also determines whether or not items 2 to 13 of the DBD13 apply in the same way and outputs the result.

[0047] FIG. 6 is an explanatory diagram showing an example of the determination of applicability of the items of the DBD 13 for each care record.

[0048] In the example of Figure 6, the care record includes the time the care record was entered, the record type, and the record text. The care record in the upper row of Figure 6 includes a non-standard record text: "The patient rested well except for going to the toilet until 11:00 PM, but from 11:00 PM, the patient wandered around the hallway and peeked out of the room door for about 30 minutes." The care record also includes a non-standard record text: "I tried to encourage the patient to fall asleep repeatedly, but the patient did not listen." Based on these record texts, the care record is classified as corresponding to items 4, 7, and 8 of DBD13. It is believed that the classification model classified the care record into items 4, 7, and 8 of DBD13 by particularly responding to the bolded parts of the record text shown in Figure 6. The care record in the lower row of Figure 6 includes a non-standard record text: "The patient tried to remove his jacket and pants while on the floor." The care record also includes a non-standard record text: "I watched over the patient to make sure he did not remove his clothes while on the floor." From these recorded sentences, the nursing care record is classified as corresponding to item 10 of DBD13. It is thought that the classification model classified the nursing care record into item 10 of DBD13 by particularly reacting to the parts written in bold in the recorded sentences shown in Figure 6.

[0049] The classification unit 112 determines whether each item in the DBD 13 applies to one or more care records for a predetermined period. The predetermined period is a period of fixed length, such as one year, but may also be an indefinite period of time. For simplicity of explanation, the following description will be given using an example in which the predetermined period is one year.

[0050] The classifying unit 112 can calculate and output the number of occurrences (number of occurrences) of each item in DBD 13 for each subject over a predetermined period of one year. That is, the classifying unit 112 calculates the number of occurrences of each item in DBD 13 for subject A over a one-year period, for example. The classifying unit 112 can output the number of occurrences of each item in DBD 13 for each subject over a one-year period as a result of the determination of whether or not the item applies.

[0051] The evaluation unit 113 evaluates the subject's condition based on the corresponding / non-corresponding determination result. The evaluation unit 113 can evaluate the subject's condition based on the corresponding / non-corresponding determination result using an approximation formula. The subject's condition is, for example, a score for each item on the DBD13. The score for each item on the DBD13 can be any value between 0 and 4. In this case, a score of "0" corresponds to an evaluation of "never." A score of "1" corresponds to an evaluation of "almost never." A score of "2" corresponds to an evaluation of "occasionally." A score of "3" corresponds to an evaluation of "often." A score of "4" corresponds to an evaluation of "always." The approximation formula can be a formula that specifies the relationship between the number of times an item on the DBD13 applies in a year (correspondence / non-correspondence determination result) and the score for that item (the subject's condition). As described below, the approximation formula is created by the approximation formula creation unit 116. The approximation formula is created from the relationship between the number of hits of a predetermined label for a subject whose condition is known over a predetermined period of one year and the condition of the known subject. In this embodiment, the approximation formula is created from the relationship between the number of hits of one item in DBD13 for a subject whose condition is known over a one-year period and the condition of the known subject. An approximation formula is created for each item in DBD13. Note that the calculation of the number of hits of each item in DBD13 for each subject over a one-year period by the classification unit 112 based on the output of the classification model, as described above, may be performed by the evaluation unit 113.

[0052] The display control unit 114 transmits the subject's condition evaluated by the evaluation unit 113 (evaluation result) to the mobile terminal 30 in association with information identifying the subject. The information identifying the subject may include, for example, the subject's name, room number, etc. As a result, the display control unit 114 causes the mobile terminal 30 to display the subject's condition for each subject. The display control unit 114 may transmit the number of occurrences of each item of DBD13 for a predetermined period of one year, along with the subject's condition evaluated by the evaluation unit 113, to the mobile terminal 30. As a result, the display control unit 114 can display the subject's condition for each subject on the mobile terminal 30 together with the number of occurrences of each item of DBD13 for a predetermined period of one year.

[0053] The learning unit 115 performs machine learning on a language model to generate a classification model for the classification unit 112. The learning unit 115 can machine learn a language model using a combination of care records of a plurality of subjects for a predetermined period of one year and applicable / inapplicable information for each item in the DBD13 of each care record corresponding to the care record.

[0054] The approximation formula creation unit 116 creates the above-mentioned approximation formula. The approximation formula creation unit 116 creates the approximation formula from the relationship between the number of hits of one item in the DBD 13 over a one-year period for multiple subjects whose conditions are known and the conditions of the known subjects. The approximation formula creation unit 116 can create an approximation formula for each item in the DBD 13. The approximation formula creation unit 116 can create an approximation formula by linear approximation.

[0055] FIG. 7 is a diagram illustrating an example of a graph showing the relationship between the number of hits for one item on DBD13 and the condition of a known subject. In the example of FIG. 7, a graph is shown showing the relationship between the number of hits (number of hits) for DBD13 item 9 (abusive language) over a one-year period and the evaluation results for DBD13 item 9 (abusive language). Multiple plots (dots) in the graph indicate the relationship between the number of hits for DBD13 item 9 over a one-year period for multiple subjects and the subject's condition. Based on this relationship (plot), an approximate equation can be created by linear approximation. The approximate equation is indicated by a solid line in the graph.

[0056] (Fixed terminal 20) The fixed terminal 20 is a PC (Personal Computer) and includes a control unit, a storage unit, a display unit, a communication unit, an input unit, etc. The functions of the control unit, the storage unit, and the communication unit are similar to the functions of the corresponding components of the server 10, and therefore description thereof will be omitted.

[0057] The input unit accepts input of various instructions and information from the administrator. The control unit performs various registrations in the server 10 based on the instructions and information accepted by the input unit. The control unit can register the subject in the server 10 in association with the care staff member 50 in charge of the subject.

[0058] (Mobile Terminal 30) Fig. 8 is a block diagram showing a schematic configuration of the mobile terminal 30. The mobile terminal 30 includes a control unit 31, a wireless communication unit 32, a storage unit 33, a display unit 34, an input unit 35, and an audio input / output unit 36. These components are connected to each other via a bus. The mobile terminal 30 may be configured by a communication terminal device that can be carried by the care staff 50, such as a tablet computer, a smartphone, or a mobile phone. The mobile terminal 30 may be replaced by a fixed terminal.

[0059] The control unit 31 has the same basic components as the control unit 11 of the server 10, such as a CPU, RAM, and ROM.

[0060] The wireless communication unit 32 has a function of performing wireless communication according to standards such as Wi-Fi and Bluetooth (registered trademark), and performs wireless communication with each device directly or via an access point 41 .

[0061] The storage unit 33 stores various data and programs and is configured by, for example, a flash memory.

[0062] The display unit 34 and the input unit 35 are touch panels, and a touch sensor as the input unit 35 is provided on the display surface of the display unit 34 which is made up of a liquid crystal or the like.

[0063] The voice input / output unit 36 ​​is configured with, for example, a speaker and a microphone, and enables voice communication between the care staff 50 and other mobile terminals 30 via the wireless communication unit 32.

[0064] The control unit 31 receives the status (evaluation result) of the subject associated with information identifying the subject from the server 10 via the wireless communication unit 32. The control unit 31 displays the status of each subject on a per subject basis via the display unit 34.

[0065] The control unit 31 may receive the number of occurrences of each item in the DBD 13 for a predetermined period of one year, together with the subject's condition, from the server 10. The control unit 31 may display, on the display unit 34, the number of occurrences of each item in the DBD 13 for one year, together with the subject's condition.

[0066] FIG. 9 is a diagram showing an example of a display screen displayed on the mobile terminal 30, which displays the subject's condition in association with the number of occurrences of item 9 of the DBD 13 over the course of one year.

[0067] 9, the graph showing the approximation formula displays the score for item 9 of DBD 13 of the subject as the evaluation result, which corresponds to the number of times item 9 of DBD 13 was met by the subject over the course of one year. This allows the care staff 50 to understand that the score for item 9 of DBD 13 of the subject is 3.9, and that the reason for this is that item 9 of DBD 13 was met 38 times over the course of one year.

[0068] FIG. 10 is a diagram showing another example of a display screen displayed on the mobile terminal 30, which displays the subject's condition in association with the number of occurrences of item 9 of the DBD 13 over the course of one year.

[0069] 10, the evaluation results are displayed in a table, with the scores for item 9 of DBD 13 of each subject corresponding to the number of times item 9 of DBD 13 was met for that subject over the course of one year. This allows the care staff 50 to understand the score for item 9 of DBD 13 of the subject under their care and the basis for that score, the number of times item 9 of DBD 13 was met for that subject over the course of one year.

[0070] The control unit 31 accepts the care record input to the input unit 35. The control unit 31 transmits the care record to the server 10 via the wireless communication unit 32.

[0071] 11 is a flowchart showing the operation of the server 10. This flowchart can be executed by the control unit 11 of the server 10 in accordance with a program.

[0072] The control unit 11 determines whether or not input of a care record has been accepted (S101). That is, the control unit 11 determines whether or not the care record input in the mobile terminal 30 has been received from the mobile terminal 30. If the control unit 11 determines that input of a care record has not been accepted (S101: NO), it executes step S101 again.

[0073] The control unit 11 registers the received care record by storing it in the storage unit 13 (S102).

[0074] The control unit 11 uses the classification model to determine whether each item of the DBD 13 is applicable to each care record (S103).

[0075] The control unit 11 calculates the number of times that each item in the DBD 13 applies to the subject over a predetermined period of one year (S104).

[0076] The control unit 11 uses an approximation formula to calculate the score for each item of the DBD 13 from the number of hits for each item of the DBD 13 over the course of one year (S105).

[0077] The control unit 11 displays the subject's score for each item in the DBD 13, as well as the number of times each item in the DBD 13 has been relevant over the past year (S106).

[0078] (Variation 1) A variation will be described. In the above-described embodiment, the recorded information is care information, the predetermined labels are the items on the DBD 13, and the subject's condition to be evaluated is the score for each item on the DBD 13. In this variation, the recorded information is a medical report, the predetermined labels are the rank of hospital visit frequency, and the subject's condition to be evaluated is the level of recovery of motor function after surgery.

[0079] 12 is a diagram showing the recorded information, the predetermined label, and the subject's condition to be evaluated in this modified example. In this modified example, the recorded information is a medical report, the predetermined label is a rank of the frequency of hospital visits, and the subject's condition to be evaluated is the level of recovery of motor function after surgery.

[0080] The record information receiving unit 111 receives, as record information, medical reports created in any text format. The medical reports may be input by a doctor to each mobile terminal 30 and received by the communication unit 12. The record information receiving unit 111 receives one or more medical reports for one subject.

[0081] The classification unit 112 determines whether each rank of hospital visit frequency applies to multiple medical reports of the subject after surgery and outputs the results. That is, the classification unit 112 determines the rank of hospital visit frequency based on the multiple medical reports of the subject. The rank of hospital visit frequency can be set arbitrarily. For example, the rank of hospital visit frequency can be set from 0 to 10. In this case, rank 0 can correspond to a hospital visit frequency of less than 0.1 times per six months. Rank 10 can correspond to a hospital visit frequency of 6 times or more per six months. Ranks 1 to 9 can be set according to the rank number, corresponding to a hospital visit frequency of 0.1 times or more to less than 6 times per six months. The classification unit 112 can use a language model such as BERT to determine whether each rank of hospital visit frequency applies to multiple medical reports of the subject.

[0082] The evaluation unit 113 evaluates the condition of the subject based on the result of the yes / no determination. Specifically, the evaluation unit 113 evaluates the post-operative motor function recovery level based on the yes / no determination for each rank of hospital visit frequency. The post-operative motor function recovery level can be arbitrarily set corresponding to each rank of hospital visit frequency. For example, the post-operative recovery level can be set to levels 0 to 5. In this case, level 5 can correspond to ranks 1 and 2 of hospital visit frequency. Level 0 can correspond to ranks 9 and 10 of hospital visit frequency. Levels 2 to 4 can be set according to the level number, corresponding to ranks 3 to 8 of hospital visit frequency. Post-operative recovery level 5 has the lowest frequency of hospital visits and is therefore considered to have a high post-operative motor function recovery level. Post-operative recovery level 0 has the highest frequency of hospital visits and is therefore considered to have a low post-operative motor function recovery level.

[0083] 13 is a diagram showing recorded information, predetermined labels, and the condition of the subject to be evaluated in this modified example. In this modified example, the recorded information is a medical report, the predetermined labels are ranks of motor function for each body part, and the condition of the subject to be evaluated is the disability level.

[0084] The record information receiving unit 111 receives, as record information, medical reports created in any text format. The medical reports may be input by a doctor into each mobile terminal 30 and received by the communication unit 12. The record information receiving unit 111 receives one or more medical reports for one subject. The medical report is considered to include details of treatment, test results, etc.

[0085] The classification unit 112 determines whether each rank of motor function for each body part is appropriate for the multiple medical reports of the subject and outputs the results. That is, the classification unit 112 determines each rank of motor function for each body part based on the multiple medical reports of the subject. Each rank of motor function for each body part can be set arbitrarily. For example, each rank of motor function for the arm can be set to ranks 0 to 3. In this case, rank 3 can correspond to normal motor function for the arm. Rank 0 can correspond to non-functioning motor function for the arm. Rank 2 can correspond to functioning motor function for the arm, but with some abnormality. The classification unit 112 can use a language model such as BERT to determine whether each rank of motor function for each body part is appropriate for the multiple medical reports of the subject.

[0086] The evaluation unit 113 evaluates the condition of the subject based on the result of the yes / no determination. Specifically, the evaluation unit 113 evaluates the disability level based on the yes / no determination of each rank of motor function for each body part. The disability level can be arbitrarily set corresponding to each rank of motor function for each body part. For example, the disability level can be set to levels 0 to 5. In this case, level 5 can correspond to the highest degree of disability. Level 0 can correspond to a state equivalent to that of a healthy person. Levels 2 to 4 can be set corresponding to each rank of motor function for each body part, as disability levels intermediate between levels 0 and 5, according to the level number. The evaluation unit 113 may evaluate the disability level from the sum of the ranks of motor function for each body part using a table in which the sum of the ranks of motor function for each body part is previously associated with the disability level. The evaluation unit 113 may evaluate the disability level for each body part from the rank of motor function for each body part using a table in which the ranks of motor function for each body part is previously associated with the disability level. The table can be flexibly modified in consideration of the validity of the evaluation.

[0087] 14 is a diagram showing the recorded information, predetermined labels, and the subject's condition to be evaluated in this modified example. In this modified example, the recorded information is the execution history of the practice menu, the predetermined labels are the ranks of accumulated physical strain, and the subject's condition to be evaluated is the risk of injury.

[0088] The record information accepting unit 111 accepts the implementation history of a practice menu created in any text format as record information. The implementation history of the practice menu may be input into each mobile terminal 30 by the trainer of the subject baseball player, for example, and received by the communication unit 12. In this case, the practice menu may be a baseball practice menu. The record information accepting unit 111 accepts implementation history of one or more practice menus for one subject.

[0089] The classification unit 112 determines whether each rank of accumulated physical strain is appropriate for the subject's implementation history of multiple practice menus and outputs the result. That is, the classification unit 112 determines each rank of accumulated physical strain based on the subject's implementation history of multiple practice menus. Each rank of accumulated physical strain can be set arbitrarily. For example, each rank of accumulated physical strain can be set from 0 to 5. In this case, rank 0 can correspond to no accumulated physical strain. Rank 5 can correspond to the most severe accumulated physical strain. Ranks 1 to 4 are intermediate ranks between rank 0 and rank 5, and can correspond to the degree of accumulated physical strain depending on the rank number. The classification unit 112 can use a language model such as BERT to determine whether each rank of accumulated physical strain is appropriate for the subject's implementation history of multiple practice menus.

[0090] The evaluation unit 113 evaluates the risk of injury of the subject based on the result of the corresponding determination. Specifically, the evaluation unit 113 evaluates the risk of injury of the subject based on the corresponding determination of each rank of accumulated physical strain. The risk of injury of the subject can be arbitrarily set corresponding to each rank of accumulated physical strain. For example, the risk of injury of the subject can be set as levels 0 to 5. In this case, level 5 may correspond to the highest risk of injury. Level 0 may correspond to the lowest risk of injury. Levels 2 to 4 may be set corresponding to each rank of accumulated physical strain as intermediate injury risk levels between level 0 and level 5. The evaluation unit 113 may evaluate the risk of injury from the rank of accumulated physical strain using a table in which each rank of accumulated physical strain is previously associated with the risk of injury. This table can be flexibly modified from the perspective of the validity of the evaluation.

[0091] (Second Embodiment) A second embodiment will be described. The present embodiment differs from the first embodiment in the following respects. In the first embodiment, in order to determine whether a predetermined label applies, a language model is used that outputs a determination of whether a predetermined label applies to each care record input. The language model determines whether a predetermined label applies to each care record input and outputs the determination. On the other hand, in the present embodiment, a language model is used that outputs a determination of whether a predetermined label applies to each care record input and a likelihood for each care record. The determination of whether a predetermined label applies is weighted by the likelihood, and the subject's condition is evaluated based on the weighted determination of whether a predetermined label applies. In other respects, this embodiment is similar to the first embodiment, and therefore redundant explanations will be omitted or simplified.

[0092] The classification unit 112 outputs a determination of whether a predetermined label applies to each care record and a likelihood for each care record in response to input of the care record. That is, the classification unit 112 classifies the care record into the corresponding label by determining whether a predetermined label applies to the care record, and calculates the likelihood corresponding to the classified label. The likelihood is a value indicating the degree of likelihood that the care record corresponds to the classified label, and is a value of 1 or less. The classification unit 112 includes a machine-learned language model. Any language model can be used as long as it is a model that outputs a determination of whether a predetermined label applies to each care record in response to input of any care record and a likelihood for each care record. For example, BERT is used as the language model.

[0093] The evaluation unit 113 weights the likelihood of each of the determinations made by the classification unit 112 for a predetermined period of one year. The evaluation unit 113 evaluates the subject's condition based on the weighted determinations. Specifically, the evaluation unit 113 evaluates the subject's condition, for example, as follows: Assume that a determination that one care record corresponds to item 9 of DBD13 is made and the likelihood is calculated to be 0.9. In this case, when calculating the number of occurrences of item 9 of DBD13 for the one care record during the predetermined period of one year, the number of occurrences of item 9 of DBD13 for the one care record is weighted by 0.9, resulting in a weight of 0.9. As a result, the number of occurrences of item 9 of DBD13 for the predetermined period of one year can be reduced by the weighting. This allows for a more accurate evaluation of the subject's condition.

[0094] The embodiment has the following advantages.

[0095] Using a machine-learned language model, the system inputs record information for a specified period created in any sentence format, and outputs a judgment as to whether each record information corresponds to a specified label.The system then evaluates the subject's condition based on the results of the judgment.This makes it easy to perform a consistent quantitative evaluation of the subject's condition based on the input record information in any sentence format.

[0096] Furthermore, the subject's condition is evaluated based on the result of the judgment using an approximate formula, which allows accurate and consistent quantitative evaluation of the subject's condition with a simple configuration.

[0097] Furthermore, an approximation formula is created from the relationship between the number of occurrences of a predetermined label in a predetermined period of time for a subject whose condition is known and the known condition, thereby enabling more accurate quantitative evaluation of the subject's condition without variation.

[0098] Furthermore, in response to the input of the number of hits of a specific label for a specific period, the system outputs the subject's condition and displays the number of hits of the specific label for the specific period, allowing users to confirm the basis for the evaluation results of the subject's condition.

[0099] Furthermore, the language model is trained by machine learning based on recorded information over a specified period of time, written in any sentence format, and the results of known classifications of specified labels for the recorded information, enabling more accurate and consistent quantitative assessment of the subject's condition.

[0100] The recorded information is treated as a care record, and the condition is evaluated for each item on the dementia behavioral disorder scale. This makes it easy to obtain a consistent evaluation for each item on the dementia behavioral disorder scale for the subject based on the care record input in any text format.

[0101] Furthermore, the language model is configured to output the likelihood of each predetermined label for each record information in response to input of one or more record information for a predetermined period, thereby enabling confirmation of the accuracy of the determination of whether or not the predetermined label applies.

[0102] Furthermore, the results of the suitability determination for a predetermined period are weighted by likelihood, and the subject's condition is output in response to the input of the weighted suitability determination result, thereby further improving the accuracy of the evaluation of the subject's condition.

[0103] The configuration of the information processing system 1 described above is a description of the main configuration in explaining the features of the above embodiment, but is not limited to the above configuration and can be modified in various ways within the scope of the claims. Furthermore, configurations that are included in general information processing systems are not excluded.

[0104] For example, the functions of the server 10 may be provided in the fixed terminal 20 or any of the mobile terminals 30 .

[0105] Furthermore, the server 10, the fixed terminal 20, and the mobile terminal 30 may each be configured by a plurality of devices, or any of a plurality of devices may be configured as a single device.

[0106] In addition, some steps may be omitted from the above-described flowcharts, other steps may be added, some of the steps may be executed simultaneously, or one step may be divided into multiple steps and executed.

[0107] The means and methods for performing various processes in the information processing system 1 described above can be realized by either dedicated hardware circuits or a programmed computer. The programs may be provided, for example, by a computer-readable recording medium such as a USB memory or a DVD (Digital Versatile Disc)-ROM, or may be provided online via a network such as the Internet. In this case, the programs recorded on the computer-readable recording medium are typically transferred and stored in a storage unit such as a hard disk. The programs may also be provided as standalone application software, or may be incorporated as a function into the software of a device such as a detection unit.

[0108] This application is based on a Japanese patent application (Patent Application No. 2024-002328) filed on January 11, 2024, the disclosure of which is incorporated herein by reference in its entirety.

[0109] 1 Information processing system, 10 Server, 11 Control unit, 111 Recorded information reception unit, 112 Classification unit, 113 Evaluation unit, 114 Display control unit, 115 Learning unit, 116 Approximation formula creation unit, 12 Communication unit, 13 Memory unit, 20 Fixed terminal, 30 Portable terminal, 31 Control unit, 32 Wireless communication unit, 33 Memory unit, 34 Display unit, 35 Input unit, 36 Voice input / output unit, 40 Network, 41 Access point, 50 Care staff.

Claims

1. A state evaluation device comprising: a classification unit including a machine-learned language model that performs a presence / absence determination of a predetermined label for each of one or more pieces of record information for a predetermined period created in an arbitrary text format regarding a target person and outputs the determination; and an evaluation unit that evaluates the state of the target person based on the presence / absence determination result for the predetermined period.

2. The state evaluation device according to claim 1, wherein the evaluation unit evaluates the state of the target person based on the presence / absence determination result using an approximation formula.

3. The state evaluation device according to claim 2, wherein the approximation formula is created from the relationship between the number of occurrences of the predetermined label in the predetermined period of the target person whose state is known and the known state.

4. The state evaluation device according to claim 3, wherein the evaluation unit outputs the state of the target person in response to an input of the number of occurrences of the predetermined label in the predetermined period, and has a display unit that displays the number of occurrences of the predetermined label in the predetermined period together with the state of the target person evaluated by the evaluation unit.

5. The state evaluation device according to claim 1, wherein the language model is machine-learned based on the record information for the predetermined period created in an arbitrary text format and the known presence / absence determination result of the predetermined label for the record information.

6. The state evaluation device according to claim 1, wherein the record information is a care record, and the state is an evaluation for each item of a dementia behavior disorder scale.

7. The state evaluation device according to claim 1, wherein the language model further outputs a likelihood for each predetermined label for each of the record information in response to an input of one or more pieces of record information for the predetermined period.

8. The state evaluation device according to claim 7, wherein the evaluation unit weights the presence / absence determination result for the predetermined period by the likelihood, and outputs the state of the target person in response to an input of the weighted presence / absence determination result.

9. A state evaluation method comprising: a step (a) of performing a presence / absence determination of a predetermined label for each of one or more pieces of record information for a predetermined period created in an arbitrary text format regarding a target person and outputting the determination using a machine-learned language model; and a step (b) of evaluating the state of the target person based on the presence / absence determination result for the predetermined period in step (a).

10. The state evaluation method according to claim 9, wherein in step (b), an approximate expression is used to evaluate the state of the subject based on the result of the determination of whether or not.

11. The state evaluation method according to claim 10, wherein the approximate expression is created from the relationship between the number of occurrences of the predetermined label in the predetermined period of the subject whose state is known and the known state.

12. The state evaluation method according to claim 11, wherein in step (b), for the input of the number of occurrences of the predetermined label in the predetermined period, the state of the subject is output, and step (c) of displaying the number of occurrences of the predetermined label in the predetermined period together with the state of the subject evaluated in step (b) is included.

13. The state evaluation method according to claim 9, wherein the language model is machine-learned based on the recorded information in the predetermined period created in an arbitrary text format and the result of the known determination of whether or not of the predetermined label for the recorded information.

14. The state evaluation method according to claim 9, wherein the recorded information is care record, and the state is an evaluation for each item of the dementia behavior disorder scale.

15. The state evaluation method according to claim 9, wherein the language model further outputs the likelihood for each predetermined label for each piece of recorded information for the input of one or more pieces of recorded information in the predetermined period.

16. The state evaluation method according to claim 15, wherein in step (b), the result of the determination of whether or not in the predetermined period is weighted by the likelihood, and for the input of the weighted result of the determination of whether or not, the state of the subject is output.

17. A state evaluation program for causing a computer to execute the state evaluation method according to any one of claims 9 to 16.

Images