Assistance method, assistance device, and program

Abstract

An assistance method for assisting a user in understanding the health condition of a subject, the assistance method comprising: acquiring a care record of the subject including text data in a first period (S11); determining the presence or absence of a physical condition abnormality in the subject from the text data using a natural language processing model (S12); and outputting presentation information, based on the result of determination of the presence or absence of the physical condition abnormality, to an information terminal of the user (S17).

Description

Support method, support device and program

[0001] The present disclosure relates to an assistance method, an assistance device, and a program for assisting a user in understanding the health condition of a subject.

[0002] The 2025 problem, which is a problem caused by an aging society in which all 8 million members of the so-called "baby boomer generation" will be over 75 years old and one in four people will be over 75 years old, will lead to a labor shortage due to increasing demand for medical care and nursing care.

[0003] Therefore, it is desirable to reduce the burden on users, such as medical and care workers, who are in charge of subjects including those receiving care and nursing care. For example, Patent Literature 1 discloses a system that automatically generates a care diary (care record) from voice.

[0004] Japanese Patent Application Laid-Open No. 2022-120752

[0005] However, the system in Patent Document 1 only automatically generates care records, and the user must check the care records to understand the health condition of the subject, so its effectiveness in reducing the burden on the user is limited.

[0006] Therefore, the present disclosure provides an assistance method, an assistance device, and a program that can assist a user in understanding the health condition of a subject.

[0007] A support method according to one aspect of the present disclosure is a support method for supporting a user in understanding the health condition of a subject, which obtains care records of the subject including text data for a first period, determines whether or not the subject has an abnormal health condition from the text data using a natural language processing model, and outputs presentation information based on the determination result of whether or not the subject has an abnormal health condition to the user's information terminal.

[0008] An assistance device according to one aspect of the present disclosure is an assistance device for assisting a user in understanding the health condition of a subject, and includes an acquisition unit that acquires care records of the subject including text data for a first period, a determination unit that determines whether or not the subject has an abnormal health condition from the text data using a natural language processing model, and an output unit that outputs presentation information based on the determination result of whether or not the subject has an abnormal health condition to the user's information terminal.

[0009] A program according to one aspect of the present disclosure is a program for causing a computer to execute the above-described assistance method.

[0010] According to one aspect of the present disclosure, it is possible to realize a support method or the like that can support a user in understanding the health condition of a subject.

[0011] FIG. 1 is a diagram illustrating an example of the configuration of a physical condition detection system according to an embodiment. FIG. 2 is a block diagram illustrating an example of the functional configuration of an information management server according to an embodiment. FIG. 3 is a diagram illustrating an example of a care record according to an embodiment. FIG. 4 is a diagram for explaining processing by a proofreading processing unit according to an embodiment. FIG. 5 is a diagram illustrating a condition classification result by a condition classification unit according to an embodiment. FIG. 6 is a flowchart illustrating a first operation of an information management server according to an embodiment. FIG. 7 is a diagram illustrating an original text of a care record according to an embodiment. FIG. 8 is a diagram illustrating attention values ​​per word according to an embodiment. FIG. 9 is a diagram schematically illustrating a calculation result of attention values ​​for each text data according to an embodiment. FIG. 10A is a flowchart illustrating a first example of detailed operation of step S13 shown in FIG. 6. FIG. 10B is a flowchart illustrating a second example of detailed operation of step S13 shown in FIG. 6. FIG. 11 is a diagram illustrating attention values ​​for each sentence according to an embodiment. FIG. 12 is a flowchart illustrating a detailed operation of step S15 shown in FIG. 6. FIG. 13 is a diagram illustrating an example of a five-level scale score and its conditions according to an embodiment. FIG. 14 is a diagram illustrating an example of a daily summary including a score according to an embodiment. FIG. 15 is a diagram illustrating an example of a weekly summary according to an embodiment. FIG. 16 is a diagram showing an example of a weekly summary including activity data according to an embodiment. FIG. 17 is a flowchart showing a second operation of the information management server according to an embodiment. FIG. 18 is a diagram showing a first example of aggregated information according to an embodiment. FIG. 19 is a diagram showing a second example of aggregated information according to an embodiment. FIG. 20 is a diagram showing a third example of aggregated information according to an embodiment. FIG. 21 is a block diagram showing an example of the functional configuration of a model generation device according to a modified example of an embodiment. FIG. 22 is a flowchart showing an operation of generating a machine learning model according to a modified example of an embodiment.

[0012] A support method according to a first aspect of the present disclosure is a support method for supporting a user in understanding the health condition of a subject, which obtains care records of the subject including text data for a first period, determines whether or not the subject has an abnormal health condition from the text data using a natural language processing model, and outputs presentation information based on the determination result of whether or not the subject has an abnormal health condition to the user's information terminal.

[0013] This automatically determines whether the subject has any abnormalities in physical condition and outputs the information to the user's information terminal, allowing the user to understand the subject's health condition simply by looking at the information presented on the information terminal. For example, the user can understand the subject's health condition without checking the care records. Therefore, the support method disclosed herein can support the user in understanding the subject's health condition.

[0014] Furthermore, for example, the support method according to the second aspect may be the support method according to the first aspect, and if the subject's physical condition is abnormal, the natural language processing model may be further used to determine the type of abnormal physical condition from the text data, and the presented information may include information based on the determined type of abnormal physical condition.

[0015] This automatically determines the type of abnormal physical condition of the subject, allowing the user to understand the subject's health condition in more detail simply by looking at the information presented on the information terminal. Thus, the support method of the present disclosure can better support the user in understanding the subject's health condition.

[0016] Furthermore, for example, a support method according to a third aspect may be a support method according to the first or second aspect, and may use a learning model to extract from the care records at least one of normal portions, which are descriptions indicating that the subject is normal, and abnormal portions, which are descriptions indicating that the subject is abnormal, and generate a first summary sentence indicating the subject's health condition during the first period based on at least one of the extracted normal portions and abnormal portions, and the presented information may include the first summary sentence.

[0017] This allows the user to understand the health condition during the first period in detail just by reading the first summary, that is, without reading the care record.

[0018] Furthermore, for example, the support method according to the fourth aspect may be the support method according to the third aspect, in which the text data is broken down into a plurality of sentences, an attention value indicating the degree to which each of the plurality of sentences contributed to the determination result of the type of physical abnormality is calculated for each of the plurality of sentences, and the first summary sentence is generated based on the attention value for each of the plurality of sentences.

[0019] This allows the generation of a first summary sentence according to the determination result of whether or not there is an abnormal physical condition.

[0020] Also, for example, the support method according to the fifth aspect may be the support method according to the fourth aspect, in which one or more sentences having the attention value of a first predetermined value are extracted from the plurality of sentences, and the first summary sentence is generated based on the extracted one or more sentences.

[0021] In this way, if the subject's physical condition is abnormal, a first summary sentence including a sentence related to the abnormality can be generated. By presenting such a first summary sentence, the user can more efficiently understand the subject's health condition.

[0022] Furthermore, for example, the support method according to the sixth aspect may be the support method according to the fifth aspect, which acquires activity data of the subject during the first period, calculates a feature based on the acquired activity data, and acquires an abnormality score indicating the degree of abnormal physical condition during the first period based on the feature, and the presented information may include information in which a score based on the abnormality score is associated with the first summary sentence.

[0023] This allows the meaning of the score to be determined by the first summary. For example, if only the score is presented, the user may mistakenly determine that a high score is incorrect when the subject's condition appears normal. In contrast, in the present disclosure, the first summary is presented along with the score, allowing the user to understand the reason for the score from the first summary, thereby preventing such misjudgments from occurring.

[0024] Further, for example, a support method according to a seventh aspect is a support method according to the second aspect, wherein the second period is composed of two or more of the first periods, and a learning model is used to extract from the care records at least one of normal portions, which are descriptions indicating that the subject is normal, and abnormal portions, which are descriptions indicating that the subject is abnormal, and a first summary sentence indicating the subject's health condition during the first period is generated based on at least one of the extracted normal portions and abnormal portions, and a second summary sentence indicating the subject's health condition during the second period is generated based on two or more first summary sentences for each of the two or more first periods that constitute the second period and attention values ​​of each of the two or more first summary sentences, which attention values ​​indicate the degree to which the first summary sentences contributed to the diagnosis of abnormal physical condition, and the presented information may include the second summary sentences.

[0025] This allows the user to understand the health condition for the second period by simply reading the second summary, that is, without having to read the care records that include a large amount of records for the second period.

[0026] Also, for example, the support method of the eighth aspect may be the support method of the seventh aspect, in which one or more sentences of two or more of the first summary sentences whose attention value is a second predetermined value are extracted, and the second summary sentence is generated based on the extracted one or more sentences.

[0027] This allows the second summary to be generated according to the determination result of whether or not there is an abnormal physical condition.

[0028] Furthermore, for example, the support method according to the ninth aspect may be the support method according to the seventh or eighth aspect, in which activity data of the subject is obtained by sensing the subject during the second period, and the time series data of the activity data is associated with the second summary sentence and output to the information terminal of the user.

[0029] This allows the user to refer to the activity data to understand the health condition, and therefore the user can easily understand the health condition of the subject by checking the activity data.

[0030] Furthermore, for example, a support method according to a tenth aspect may be a support method according to any one of the first to ninth aspects, in which a record of a specific abnormality among a plurality of abnormalities associated with the subject is collected, and the collected dates and times of occurrence of two or more abnormalities are tallied, and the presented information may include information based on the tallied results.

[0031] This helps the user to understand trends regarding the date and time when a particular abnormality occurred.

[0032] Also, for example, the support method according to the eleventh aspect is the support method according to the tenth aspect, and the information based on the aggregation results may include information indicating the influence of season or time of day on the occurrence of the specific abnormality.

[0033] This can assist the user in understanding the influence of season or time of day on the occurrence of a particular anomaly.

[0034] Furthermore, for example, a support method according to a twelfth aspect may be a support method according to any one of the first to eleventh aspects, in which descriptions relating to at least one of a specified item, a specified behavior of the subject, and a report on medical visits excluding regular visits are extracted from the care record, and the text data may include the extracted descriptions.

[0035] This eliminates descriptions that are unnecessary for determining whether or not there is an abnormality in physical condition, thereby improving the accuracy of the determination results obtained by the natural language processing model.

[0036] A support device according to a thirteenth aspect of the present disclosure is a support device for supporting a user in understanding the health condition of a subject, and includes an acquisition unit that acquires care records of the subject including text data for a first period, a determination unit that determines whether or not the subject has an abnormal physical condition from the text data using a natural language processing model, and an output unit that outputs presentation information based on the determination result of the presence or absence of the abnormal physical condition to the information terminal of the user. Also, a program according to a fourteenth aspect of the present disclosure is a program for causing a computer to execute the support method according to any one of the first to twelfth aspects.

[0037] This provides the same effect as the above-mentioned support method.

[0038] These general or specific aspects may be realized as a system, a method, an integrated circuit, a computer program, or a non-transitory recording medium such as a computer-readable CD-ROM, or as any combination of the system, method, integrated circuit, computer program, or recording medium. The program may be pre-stored in the recording medium, or may be supplied to the recording medium via a wide area communication network including the Internet.

[0039] Hereinafter, the embodiments will be specifically described with reference to the drawings.

[0040] The embodiments described below are all comprehensive or specific examples. The numerical values, components, component placement and connection configurations, steps, and step order shown in the following embodiments are merely examples and are not intended to limit the present disclosure. Furthermore, among the components in the following embodiments, components not described in independent claims are described as optional components.

[0041] Furthermore, each figure is a schematic diagram and is not necessarily an exact illustration. Therefore, for example, the scales of the figures do not necessarily match. Furthermore, in each figure, substantially the same components are given the same reference numerals, and redundant explanations are omitted or simplified.

[0042] Furthermore, in this specification, numerical values ​​and numerical ranges are not expressions that express only the strict meaning, but are expressions that mean that they also include a substantially equivalent range, for example, a difference of about several percent (or about 10%).

[0043] Furthermore, in this specification, ordinal numbers such as "first" and "second" do not refer to the number or order of components unless otherwise specified, but are used for the purpose of avoiding confusion and distinguishing between components of the same type.

[0044] (Embodiment) Hereinafter, a support method and the like according to the present embodiment will be described with reference to FIGS.

[0045] [1. Configuration of the physical condition detection system] First, the configuration of a physical condition detection system including an information management server that executes the support method according to this embodiment will be described with reference to Figures 1 to 5. Figure 1 is a diagram showing an example of the configuration of a physical condition detection system 100 according to this embodiment.

[0046] The physical condition detection system 100 according to this embodiment is a system configured such that an information management server 10 assists a user in understanding the health condition of a person 50 receiving care or nursing care.

[0047] As shown in Fig. 1, the physical condition detection system 100 includes an information management server 10, a sensing unit 20, a care record collection unit 25, and a display terminal unit 30. These are communicatively connected to each other via a communication network 40. The communication network 40 may be a wired network, a wireless network, or both a wired network and a wireless network. Fig. 1 also shows a subject 50 receiving nursing or care, a user 60 who is a field staff member such as a medical professional who provides nursing or care for the subject 50, and a user 61 who is a field staff member such as a monitor of the subject 50 who can check the display terminal unit 30.

[0048] Note that Figure 1 shows an example in which the physical condition detection system 100 has one sensing unit 20, but this is not limited to this, and the system may have as many sensing units 20 as the number of subjects 50 being cared for or looked after.

[0049] The sensing unit 20 acquires activity data of the subject 50 over a predetermined period by sensing the data. The activity data includes data related to the activities of the subject 50. The activity data includes at least one of heart rate, respiratory rate, whether the subject 50 is in bed or not, body temperature, the number of times the subject turns over in bed, and the amount of food eaten, and may include, for example, at least the heart rate. The activity data may also include at least two of heart rate, respiratory rate, whether the subject 50 is in bed or not, body temperature, the number of times the subject 50 turns over in bed, and the amount of food eaten. For example, the sensing unit 20 may acquire data such as the heart rate, respiratory rate, and body movement (hereinafter also referred to as sensing data) every second while the subject 50 is in bed. The sensing unit 20 may also sense whether the subject 50 is in bed or not depending on whether the heart rate, respiratory rate, body movement, and the like can be sensed.

[0050] The interval for acquiring sensing data such as heart rate, respiratory rate, and body movement is not limited to one second, but may be two seconds, for example, as long as it is an interval that allows changes in the sensing data of the subject 50 to be determined. Furthermore, the sensing unit 20 may further sense life rhythms such as sleep state depending on whether or not it is possible to sense the heart rate, respiratory rate, body movement, and the like.

[0051] The sensing unit 20 may also include a sensor that performs sensing related to excretion, for example, and may sense the shape and color of excretion, the time required for excretion, and the like.

[0052] The sensing unit 20 may be an imaging device such as a camera, and is configured to capture an image of the subject 50 in bed or eating. The camera may be a thermal camera that detects the body temperature of the subject 50, or a regular camera (e.g., a CCD (Charge Coupled Device) camera). The amount of food consumed by the subject 50 in the morning, afternoon, and evening may be obtained by image analysis of the image. The body temperature and amount of food consumed may be obtained by input by the user 60 to the care record collection unit 25.

[0053] In the following, a case where the activity data includes the respiratory rate and the heart rate will be mainly described.

[0054] The care record collection unit 25 collects care records from the user 60. The care records are records of the content of nursing or care provided by the user 60 to the subject 50, changes in the condition of the subject 50, living conditions, etc. (see, for example, FIG. 3 described later). The care records include free-form text data describing the state of the subject 50 when the user 60 provided nursing or care for the subject 50.

[0055] The care record collection unit 25 includes an input unit for receiving care records from the user 60 and a display unit for displaying a screen for inputting the care records. The input unit is, for example, but not limited to, a touch panel, a keyboard, or a sound collection device (e.g., a microphone). The input unit is, for example, but not limited to, a display device such as a display. The care record collection unit 25 may be realized by a portable terminal device such as a smartphone or a tablet, or by a stationary terminal device such as a PC (personal computer).

[0056] The information management server 10 is realized by, for example, a computer including a processor (microprocessor), memory, a communication interface, etc. The information management server 10 may operate with a portion of its configuration included in a cloud server. The information management server 10 is a device for assisting the user 61 in understanding the health status of the subject 50 being cared for or cared for. In other words, the information management server 10 assists the user 61 or the like in monitoring the subject 50. Furthermore, the information management server 10 may be a device for further assisting the user 61 or the like in not overlooking small changes in the subject 50's physical condition (i.e., signs of physical abnormalities) that could lead to physical abnormalities.

[0057] FIG. 2 is a block diagram showing an example of the functional configuration of the information management server 10 according to the present embodiment.

[0058] 2, the information management server 10 includes a transmitter / receiver 11, an information recording unit 12, a feature calculation unit 13, a score calculation unit 14, and a support processing unit 15. For example, a support device is configured including at least the support processing unit 15. Note that the support device may be realized as a standalone device.

[0059] The transmitter / receiver 11 includes, for example, a communication interface, and transmits and receives various information to and from the sensing unit 20 or the display terminal unit 30 via the communication network 40. For example, the transmitter / receiver 11 acquires activity data including the respiratory rate and heart rate of the subject 50 over a predetermined period. Here, the activity data may include at least the respiratory rate and heart rate of the subject 50 over the predetermined period, for example.

[0060] In this embodiment, the transmitting / receiving unit 11 acquires sensing data such as the heart rate, respiratory rate, and body movements per second while the subject 50 is in bed at predetermined intervals, for example, every minute, from the sensing unit 20 via the communication network 40. In this way, the transmitting / receiving unit 11 acquires activity data including sensing data, which is obtained daily on-site, via the communication network 40. The transmitting / receiving unit 11 is an example of an acquiring unit.

[0061] The information recording unit 12 records information transmitted and received by the transmitting / receiving unit 11. The information recording unit 12 is a recording medium capable of recording information, and is configured, for example, by a rewritable nonvolatile memory such as a hard disk drive or a solid state drive. Note that the information recording unit 12 may record a plurality of feature amounts calculated by the feature amount calculation unit 13.

[0062] The feature calculation unit 13 includes, for example, a computer including a memory and a processor (microprocessor), and realizes a function of calculating multiple feature amounts by causing the processor to execute a control program stored in the memory. The feature calculation unit 13 calculates multiple feature amounts based on activity data including the respiratory rate and heart rate of the subject 50 acquired by the transmission / reception unit 11. For example, the feature calculation unit 13 acquires sensing data for a time period including a target date and time for physical condition detection from the activity data acquired by the transmission / reception unit 11 or recorded in the information recording unit 12, and calculates feature amounts for each hour of the sensing data, such as the respiratory rate.

[0063] For example, the feature calculation unit 13 may calculate at least the average and maximum values ​​of the subject's 50 respiratory rate and the average and maximum values ​​of the subject's 50 heart rate as multiple feature amounts per hour. From at least the respiratory rate and heart rate, the feature calculation unit 13 calculates the average and maximum values ​​of the respiratory rate and heart rate from the respiratory rate, respiratory rate difference data, heart rate, and the average, maximum, standard deviation, skewness, kurtosis, and impulse factor of the heart rate difference data as multiple feature amounts. Here, the impulse factor is obtained by subtracting the average value from the maximum value. In this way, the feature calculation unit 13 performs statistical processing on the activity data to calculate multiple feature amounts.

[0064] More specifically, the feature calculation unit 13 calculates, for example, feature amounts related to the respiration rate and the heart rate of the subject 50 on an hourly basis. For example, the feature calculation unit 13 acquires sensing data indicating the respiration rate of the subject 50 in a time period including the target date and time of physical condition detection from the activity data recorded in the information recording unit 12 or sensing data acquired from the sensing unit 20, and calculates statistical feature amounts for each hour of the time period.

[0065] More specifically, the feature calculation unit 13 acquires respiration rate data from the activity data, for example, where the respiration rate is not zero for a certain hour, and calculates statistical features such as the mean value, maximum value, minimum value, standard deviation, skewness, kurtosis, and impulse factor for that hour from the acquired respiration rate data. Here, the impulse factor can be calculated from the difference between the maximum value and the mean value of the respiration rate data for that hour (maximum value - mean value). Furthermore, the feature calculation unit 13 calculates statistical features such as the mean value, maximum value, minimum value, standard deviation, skewness, kurtosis, and impulse factor for that hour from the difference data of the acquired respiration rate data. The difference data of the acquired respiration rate data is, for example, data indicating the difference between the respiration rate at time t and the respiration rate at time t+1, one second after time t, i.e., data indicating the difference between the respiration rate data per second. Note that the feature calculation unit 13 may calculate at least the mean value and maximum value for that hour from the acquired respiration rate data as statistical features.

[0066] Furthermore, for example, the feature calculation unit 13 obtains heart rate data indicating the heart rate of the subject 50 during a time period including the target date and time for physical condition detection from the activity data recorded in the information recording unit 12 or the sensor data obtained from the sensing unit 20, and calculates statistical features for each hour of the time period.

[0067] More specifically, the feature calculation unit 13 acquires heart rate data from the activity data, for example, where the heart rate is not zero for a certain hour, and calculates statistical features such as the average value, maximum value, minimum value, standard deviation, skewness, kurtosis, impulse factor, etc. for that hour from the acquired heart rate data. The feature calculation unit 13 also calculates statistical features such as the average value, maximum value, minimum value, standard deviation, skewness, kurtosis, impulse factor, etc. for that hour from the difference data of the acquired heart rate data. The difference data of the acquired heart rate data, like the difference data of the respiratory rate data, is data indicating the difference between the heart rate at time t and the heart rate at time t+1, one second after time t, i.e., the difference between the heart rate data per second. Note that the feature calculation unit 13 may calculate at least the average value and maximum value for that hour from the acquired heart rate data as statistical features.

[0068] The feature amount calculation unit 13 may calculate the amount of food eaten by the subject 50 and the rate of being away from bed as one of the plurality of feature amounts.

[0069] The score calculation unit 14 obtains an abnormality score indicating the degree of abnormality in physical condition per predetermined period based on the feature amounts calculated by the feature calculation unit 13. The score calculation unit 14 obtains the abnormality score, for example, by inputting the multiple feature amounts calculated by the feature calculation unit 13 into a trained model (supervised model) generated by supervised learning.

[0070] Next, based on the acquired abnormality score, the score calculation unit 14 calculates a graded score for indicating in stages the degree of abnormal physical condition of the subject 50. The graded score is an example of a score based on the abnormality score.

[0071] In this embodiment, the score calculation unit 14 calculates an average value of the abnormality scores on a daily basis from the abnormality scores on an hourly basis on the target day of physical condition detection. Similarly, the score calculation unit 14 calculates an average value of the abnormality scores on a daily basis on the day before and the day before that target day of physical condition detection from the abnormality scores on an hourly basis on the day before and the day before that target day. The score calculation unit 14 calculates a three-day total score by summing the average daily abnormality scores on the target day, the day before that target day, and the day before that target day. Note that the three-day total score is one example of a calculation method for accurately calculating a graded score, and is not limited to this. It is sufficient to calculate a range of one-day to five-day total scores.

[0072] The score calculation unit 14 calculates a threshold value for the staging score (sometimes referred to as a staging threshold value) from a group of three-day total scores for approximately the past 90 days up to the target day. More specifically, the score calculation unit 14 calculates the staging threshold value by calculating the average and standard deviation of the group of three-day total scores for approximately the past 90 days.

[0073] If the scaled score is, for example, on a five-point scale, and the calculated scaled score value is 4 or 5, the score calculation unit 14 may output the calculated scaled score to the display terminal unit 30 via the communication network 40.

[0074] In this way, the score calculation unit 14 scores changes in the subject's 50 physical condition and presents the score to the user 61, thereby preventing the subject's 50 condition from becoming more severe due to overlooking changes in physical condition.

[0075] The support processing unit 15 includes a computer including, for example, a memory and a processor (microprocessor), and by the processor executing a control program stored in the memory, realizes the function of generating data to support the user 61 or the like in monitoring the subject 50.

[0076] The support processing unit 15 includes a transmission / reception unit 151, an information recording unit 152, a calibration processing unit 153, a state classification unit 154, a normal point extraction unit 155a, an abnormal point extraction unit 155b, a first summary generation unit 156, a memory unit 157, a second summary generation unit 158, and a calculation unit 159.

[0077] The transmission / reception unit 151 includes, for example, a communication interface, and transmits and receives various information between the care record collection unit 25 and the display terminal unit 30 via the communication network 40. For example, the transmission / reception unit 151 acquires the care records of the subject 50 for a predetermined period.

[0078] In this embodiment, the transmitting / receiving unit 151 acquires, on an hourly or daily basis, nursing care records including text data indicating details of nursing or care provided to the subject 50 by the user 60, who is a field staff member as shown in FIG. 1 , and the state of the subject 50, from the nursing care record collection unit 25 via the communication network 40. In this way, the transmitting / receiving unit 151 acquires, via the communication network 40, nursing care records including text data that are obtained daily on-site. The transmitting / receiving unit 151 functions as an acquisition unit.

[0079] The interval at which the transmitting / receiving unit 151 acquires the care records is not limited to one hour or one day.

[0080] Furthermore, the transmitting / receiving unit 151 outputs information (e.g., presentation information) generated by the first summary generating unit 156, the second summary generating unit 158, and the tallying unit 159 to the display terminal unit 30 of the user 61. The transmitting / receiving unit 151 functions as an output unit.

[0081] Here, the care record acquired by the transmitting / receiving unit 151 will be described with reference to Fig. 3. Fig. 3 is a diagram showing an example of the care record according to the present embodiment.

[0082] As shown in Figure 3, the care record includes the date, time, item, main, secondary, fluid, urine, stool, blood pressure (up and down), pulse, SPO2, bathing, record content, and recorder. The care record includes numerical values ​​such as body temperature and text data such as the record content. The record content is free-form text data.

[0083] In the present embodiment, the support processing unit 15 uses only the text data of the numerical values ​​and text data included in the care record when generating the presentation information. In other words, the support processing unit 15 does not use the numerical values ​​when generating the presentation information.

[0084] 2 again, the information recording unit 152 records information transmitted and received by the transmitting / receiving unit 151. The information recording unit 152 is a recording medium capable of recording information, and is configured, for example, by a rewritable nonvolatile memory such as a hard disk drive or a solid state drive. Note that the information recording unit 152 may also record various models used by each processing unit, such as the calibration processing unit 153, the state classification unit 154, the normal point extraction unit 155a, and the abnormal point extraction unit 155b.

[0085] The proofreading processing unit 153 is a processing unit for performing character proofreading of text data included in the care record. Since the care record may contain typos and omissions, the proofreading processing unit 153 corrects typos and omissions. This makes it possible to prevent a decrease in the accuracy of processing such as condition classification due to typos and omissions.

[0086] The proofreading processor 153 performs character proofreading of the text data using a language model, such as an n-gram language model.

[0087] FIG. 4 is a diagram for explaining the processing of the calibration processing unit 153 according to this embodiment.

[0088] As shown in Fig. 4, the proofreading processing unit 153 performs character proofreading using a plurality of language models. In this embodiment, the proofreading processing unit 153 uses four language models, namely, a first proofreading model to a fourth proofreading model, which are different from one another. The proofreading processing unit 153 inputs the same text data to each of the four language models and obtains proofread text data. The example in Fig. 4 shows an example in which "yori tsuite" is output from the first proofreading model, "yoro tsuite" is output from the second proofreading model and the third proofreading model, and "yoro zu tsuite" is output from the fourth proofreading model.

[0089] The proofreading processor 153 determines the proofread text data by taking a majority vote of the outputs of the four language models. In the example of Figure 4, "yorotsuite" is output most frequently, so the proofreading processor 153 determines that "yorotsuite" is the correct expression for the corresponding part of the original text data and corrects it accordingly.

[0090] The number of language models used by the proofreading processing unit 153 may be one or more. For example, the proofreading processing unit 153 may input the same text data multiple times into one language model and determine the proofread text data by majority vote of the outputs obtained three or more times. Furthermore, the same text data may be input multiple times into at least one of the first to fourth proofreading models shown in FIG. 4 .

[0091] Referring again to FIG. 2 , the condition classification unit 154 uses a natural language processing model to determine whether or not the subject 50 has an abnormal physical condition for at least a first period from text data included in the care record of the subject 50. The natural language processing model is a model that outputs whether or not the subject 50 has an abnormal physical condition for the first period when text data of the subject 50 for the first period is input. The first period is, for example, one day, but is not limited to this and may be one hour, for example. The first period may be set by the user 61, for example. The condition classification unit 154 is an example of a determination unit, and the natural language processing model is an example of a condition classification model.

[0092] In this embodiment, the condition classification unit 154 further outputs an abnormality classification result. The classification result includes an abnormality label that is the output of the natural language processing model. The abnormality label indicating the abnormality classification (type of abnormality) includes restlessness, pain, dizziness, trauma, fever, gastrointestinal symptoms, deterioration of movement, other incidents, respiratory system, poor physical condition, edema, deterioration of swallowing, drowsiness, intravenous drip, cold symptoms, other infections, bladder system, etc. Note that hereinafter, the abnormality label will also be referred to as a label, an inference label, etc.

[0093] When text data is input, the state classification unit 154 outputs a classification result of normal or abnormal if the data is abnormal. The natural language processing model is trained in advance by machine learning using the text data as input data and the classification result of normal or abnormal if the data is abnormal as correct answer data.

[0094] FIG. 5 is a diagram showing the state classification results obtained by the state classification unit 154 according to this embodiment.

[0095] As shown in FIG. 5, the state classification result includes a number ("num" in FIG. 5), text data input to the natural language processing model ("input_text" in FIG. 5), and the classification result of the natural language processing model ("inference label" in FIG. 5).

[0096] 5, for example, when the text data numbered "10" is input to the natural language processing model, "uneasy" is output as the classification result, and when the text data numbered "14" is input to the natural language processing model, "cold" is output as the classification result. Also, for example, when the text data numbered "39" is input to the natural language processing model, "normal" is output as the classification result.

[0097] If the natural language processing model only determines whether or not there is an abnormality in the physical condition, all the inference labels in FIG. 5 other than "normal" will be "abnormal."

[0098] The state classification unit 154 uses a natural language processing model generated in advance by supervised learning. As the natural language processing model, an existing model corresponding to the language used in the facility (e.g., a nursing home or a medical facility) where the natural language processing model is used. For example, as the natural language processing model, any of UTH-BERT, BERT pre-trained in Japanese (Japanese version of BERT), and RoBERTa (A Robustly Optimized BERT Pretraining Approach) can be used, but is not limited to these.

[0099] In this embodiment, the Japanese version of BERT is used as the natural language processing model. The Japanese version of BERT is generated by, for example, using "Wikipedia Cirrussearch" data (approximately 17 million pieces) as of August 31, 2020, masking words, and training the model to predict the hidden words. Examples of import methods include, but are not limited to, "BertForSequenceClassification.from_pretrained(cl-tohoku / bert-base-japanese)" and "AutoTokenizer.from_pretrained(cl-tohoku / bert-base-japanese)." The model structure of the Japanese version of BERT adopts the structure used in general BERT models, for example, "12 layers, 768 hidden layer dimensions, 12 attention heads."

[0100] In this way, the condition classification unit 154 is configured to be able to automatically determine whether the subject 50 is normal or, if abnormal, to classify the subject 50 as abnormal from the text data using natural language processing.

[0101] Referring again to FIG. 2 , the normal part extraction unit 155a receives a plurality of sentences determined to be normal by the state classification unit 154 and summarizes the input sentences. The normal part extraction unit 155a uses a machine learning model (normal summary output model) that, when a plurality of sentences are input, outputs a normal summary that summarizes the plurality of sentences. The machine learning model has been trained in advance. Note that the machine learning model may be an existing model.

[0102] A normal summary sentence may be a sentence that does not seem abnormal to the subject 50. For example, a normal summary sentence may include information that indicates that the subject appears to be having fun.

[0103] As the normal summary output model, for example, mt5_summarize_japanese is used, but the present invention is not limited to this, and for example, BART, BERT, LexRank, etc. may also be used.

[0104] The abnormal part extraction unit 155b receives a plurality of sentences determined to be abnormal by the state classification unit 154 and summarizes the input sentences. When a plurality of sentences are input, the abnormal part extraction unit 155b uses a machine learning model (anomaly part summary model) that extracts sentences related to the abnormality from the plurality of sentences and outputs them as an abnormality summary sentence. The machine learning model has been trained in advance. Note that the machine learning model may be an existing model.

[0105] The abnormality summary sentence may be a sentence containing content related to an abnormality in the subject 50. For example, the abnormality summary sentence includes information related to the assigned abnormality label. The abnormality location summary model is configured to identify a sentence indicating a serious abnormality among multiple sentences and to output an abnormality summary sentence including the identified sentence. A sentence indicating a serious abnormality is a sentence with a high attention value.

[0106] In this way, the support processing unit 15 is configured to use a machine learning model to generate at least one of a normal summary sentence containing a sentence indicating that the subject 50 is normal and an abnormal summary sentence containing a sentence indicating that the subject 50 is abnormal from the care record.

[0107] The first summary generation unit 156 generates a summary (summary) indicating the health condition of the subject 50 during a first period based on at least one of the output from the normal portion extraction unit 155a and the output from the abnormal portion extraction unit 155b. The summary may include content that is likely to lead to the subject's physical condition or a change in physical condition. The summary also indicates the subject's physical condition as a whole. The first summary generation unit 156 generates a summary for each first period, which is a relatively short period. In this embodiment, the first summary generation unit 156 generates a summary for each day (daily summary). The summary (e.g., daily summary) generated by the first summary generation unit 156 is an example of a first summary.

[0108] The first summary generator 156 generates a summary including the normal summary output from the normal part extractor 155a and the abnormal summary output from the abnormal part extractor 155b.

[0109] The first summary generation unit 156 may use a machine learning model that, when two or more sentences are input, generates a sentence by connecting the two or more sentences. The machine learning model is trained in advance. The machine learning model may be a language model. The machine learning model is, for example, mt5_summarize_japanese, but is not limited to this, and may be, for example, BART, BERT, LexRank, etc.

[0110] The storage unit 157 is a storage device that stores the summary generated by the first summary generation unit 156. The storage unit 157 is a recording medium that can record information, and is configured, for example, by a rewritable non-volatile memory such as a hard disk drive or a solid state drive.

[0111] The second summary generation unit 158 ​​generates a summary for each second period, which is longer than the first period, based on the two or more summary sentences generated by the first summary generation unit 156. The second period is, for example, composed of a plurality of single periods. For example, the second summary generation unit 158 ​​generates a second summary indicating the health state of the subject 50 during the second period based on two or more first summary sentences for each of the two or more first periods that make up the second period and attention values ​​(information indicating the level of attention) for each of the two or more first summary sentences. The attention value will be described later.

[0112] In this embodiment, the second summary generation unit 158 ​​generates a summary sentence (weekly summary sentence) for each week. The second summary generation unit 158 ​​generates the weekly summary sentence by extracting sentences or summaries having an attention value equal to or greater than a predetermined value from the daily summaries for one week.

[0113] The second summary generation unit 158 ​​uses a machine learning model that, when two or more sentences are input, generates a sentence by connecting the two or more sentences. The machine learning model is trained in advance. The machine learning model may be a language model. The machine learning model is, for example, mt5_summarize_japanese, but is not limited to this, and may be, for example, BART, BERT, LexRank, etc.

[0114] Furthermore, the second summary generating section 158 may include the activity data of the subject 50 acquired from the sensing section 20 in the weekly summary.

[0115] The counting unit 159 counts the abnormal labels based on the abnormal label history. The counting unit 159 collects records of an abnormal label that is the same as an abnormal label of interest (a label of interest) from the abnormal label history. The record includes information indicating at least the date and time when the abnormality indicated by the abnormal label was detected. For example, a date and time based on a care record may be associated as the date and time when the abnormal label was detected. The label of interest is input by the user 61 or the like.

[0116] The counting unit 159 counts the date and time of occurrence of the abnormality indicated by the label of interest based on the counted records. The counting unit 159 counts the number of occurrences for a predetermined period, such as by time of day (e.g., morning, afternoon, or evening), by month, or by season. The counting is performed for each subject 50. This makes it possible to confirm the occurrence trend of the abnormality indicated by the label of interest in the subject 50. Presenting such information to the user 61 can lead to the discovery of, for example, information about the physical condition of the subject 50 that has not been discovered until now and has been buried in the care records.

[0117] The display terminal unit 30 is realized by a computer including a processor (microprocessor), memory, a communication interface, a user interface, etc. The display terminal unit 30 is a terminal of a user 61 such as a monitor of the subject 50, and is a portable terminal device such as a tablet or smartphone, but may also be a mobile personal computer or a stationary personal computer connected to a display (stationary terminal device). The display terminal unit 30 is an example of an information terminal.

[0118] The display terminal unit 30 is connected to the communication network 40, and when presentation information is acquired from the information management server 10, the display terminal unit 30 causes the user interface to display the presentation information. The user interface includes, for example, a display device such as a liquid crystal display, but is not limited to this.

[0119] 2. Operation of Information Management Server Next, the operation of the information management server 10 configured as described above will be described with reference to Figs. 6 to 20. First, the operation of generating a summary regarding the physical condition of the subject 50 will be described with reference to Figs. 6 to 16. Fig. 6 is a flowchart showing a first operation (assistance method) of the information management server 10 according to this embodiment.

[0120] 6 , the transmitting / receiving unit 151 of the support processing unit 15 acquires the care records of the subject 50 from the care record collection unit 25 (S11). The transmitting / receiving unit 151 acquires, for example, the care records for a first period. For example, the transmitting / receiving unit 151 periodically acquires the care records of the subject 50, but the acquisition timing is not particularly limited.

[0121] Next, the condition classification unit 154 classifies the condition of the subject 50 based on the text data included in the care record (S12). In step S12, a natural language processing model may be used to determine whether the subject 50 has an abnormal physical condition from the text data, and if the subject 50 has an abnormal physical condition, a natural language processing model may be further used to determine the type of abnormal physical condition from the text data. It can also be said that the condition classification unit 154 classifies the type of abnormality from the text data.

[0122] In addition, the state classification unit 154 acquires the attention value assigned (calculated) within the natural language processing model.

[0123] For example, the condition classification unit 154 decomposes text based on characters or character strings that separate meanings, and extracts words from the nursing care records. A word is a structural unit of language composed of one or more morphemes, and is also called a word. A word is, for example, the smallest unit of meaningful expression, and may be each individual group of phonemes extracted by dividing it to the point where further decomposition would no longer be meaningful, or may be the smallest unit that can be pronounced independently. A collection of words forms phrases, clauses, and sentences. Note that any known method may be used for document decomposition.

[0124] Fig. 7 is a diagram showing the original text of a care record according to this embodiment, and Fig. 8 is a diagram showing attention values ​​for each word according to this embodiment.

[0125] Fig. 7 shows the original text of the nursing care record before it is decomposed, and the left column of Fig. 8 shows the results of decomposing the original text shown in Fig. 7. Fig. 8 shows an example in which the original text shown in Fig. 7 has been decomposed into 21 words.

[0126] The state classification unit 154 acquires the attention value for each word. The attention value is a numerical value indicating the contribution of each word to the classification result (e.g., a normal label or an abnormal label such as disturbing) by the state classification unit 154. Note that the attention value may be a normalized value between 0 and 1.

[0127] The numerical values ​​in the right column of Fig. 8 indicate the normalized attention value of each word. For example, if the state classification unit 154 determines that the subject 50 is agitated, the attention value shown in Fig. 8 indicates the degree to which each word influenced the determination that the subject 50 is agitated.

[0128] FIG. 9 is a diagram schematically illustrating the calculation results of the attention value for each text data according to this embodiment. The inference labels shown in FIG. 9 indicate the classification results (normal label or abnormal label) obtained by classifying the text data shown in input_text using the state classification unit 154. For example, No. ("num" in FIG. 9) 10 indicates an example in which the output obtained by inputting "Awake...the same thing over and over" into the natural language processing model is "disturbing." Note that hereinafter, the inference labels will also be referred to as state labels.

[0129] In addition, in FIG. 9, in the text data indicated by input_text, the higher the attention value, the darker the hatching, and the lower the attention value, the lighter the hatching (or no hatching).

[0130] For example, if No. ("num" in FIG. 9) is 10, this indicates that the contributions (e.g., attention values) of "repeat," "forever," and "keeps saying" are highest in the order of "repeat," "forever," and "keeps saying," from the input text data. In this case, the state classification unit 154 may extract a sentence (a decomposed sentence) containing at least one of "repeat," "forever," and "keeps saying," or a sentence (e.g., a sentence separated by a period) containing at least one of these from the input text data, and output the extracted sentence to the abnormal portion extraction unit 155b.

[0131] In addition, since nursing care records include parts that are not text records, such as numerical values, and text parts that are not necessary for condition classification, the condition classification unit 154 may extract only the text parts necessary for condition classification from the text data and input the extracted text parts into the natural language processing model.

[0132] In addition, for example, descriptions following title words such as "near-accident" or "condition observation" that frequently appear in care records, or fixed phrases, may be excluded from the text data. For example, the condition classification unit 154 may extract descriptions that are likely to contain records related to abnormalities from the care records, and input the extracted descriptions to the natural language processing model.

[0133] The condition classification unit 154 may extract from the care record certain items, such as special notes in the care record, observation details, condition observations, response records, today's situation, emergency medication, consultation contact, accident records, or descriptions following a title word, as items indicating the condition of the subject 50, and input the extracted descriptions into a natural language processing model.

[0134] Furthermore, the condition classification unit 154 may extract descriptions of predetermined behaviors of the subject 50, such as eating, bathing, and excretion, that may indicate an abnormality, and input the extracted descriptions into the natural language processing model. Since eating, bathing, excretion, and the like are often recorded in care records even when no abnormality is present, it is preferable to extract descriptions of changes in condition, discontinuation, and the like.

[0135] Furthermore, the condition classification unit 154 may input text data related to medical visits, excluding scheduled visits, into the natural language processing model. The text data may include descriptions in medical visit reports for urgent visits or unexpected visits, or descriptions extracted from medical visit reports as possibly indicating an abnormality. The process of extracting necessary descriptions from such text data is also referred to as extraction processing.

[0136] 6 again, next, normal part extraction unit 155a generates and outputs a normal summary, which is a summary regarding normality, based on normal data, which is a sentence determined to be normal by condition classification unit 154, and abnormal part extraction unit 155b generates and outputs an abnormal summary, which is a summary regarding abnormality, based on abnormal data, which is a sentence determined to be abnormal by condition classification unit 154 (S13). When first summary generation unit 156 generates a daily summary, in step S13, the normal summary and the abnormal summary may be generated based on one day's worth of care records.

[0137] 10A and 10B are flowcharts showing an example of detailed operations (assistance method) of step S13 shown in Fig. 6. Fig. 10A shows operations performed by the normal point extraction unit 155a in step S13, and Fig. 10B shows operations performed by the abnormal point extraction unit 155b in step S13.

[0138] 10A , the normal part extraction unit 155a collects normal data (S131a). The normal part extraction unit 155a may collect the normal data from, for example, the state classification unit 154. In step S131a, multiple sentences determined to be normal by the state classification unit 154 are extracted.

[0139] Next, the normal part extraction unit 155a applies a normal summary output model (S132a). The normal part extraction unit 155a inputs and summarizes the multiple sentences determined to be normal by the state classification unit 154 using the normal summary output model. As a result, a normal summary is generated that includes one or more sentences from the multiple sentences determined to be normal.

[0140] Next, the normal part extraction unit 155a outputs the generated normal summary to the first summary generation unit 156 (S133a).

[0141] 10B , the abnormal part extraction unit 155b collects various types of abnormal data (S131b). The abnormal part extraction unit 155b collects abnormal data, which is a sentence determined to be abnormal, regardless of the type of abnormality. The abnormal part extraction unit 155b may collect the abnormal data from, for example, the state classification unit 154. In step S131b, multiple sentences determined to be abnormal by the state classification unit 154 are extracted.

[0142] Next, the abnormal part extraction unit 155b refers to the attention values ​​assigned in the classification model (in this embodiment, the natural language processing model) (S132b) and calculates the sum of the attention values ​​of each sentence (S133b). The abnormal part extraction unit 155b acquires the attention value for each word as shown in Fig. 8, for example, and calculates the sum of the attention values ​​of each sentence. In other words, the abnormal part extraction unit 155b calculates the attention value for the sentence.

[0143] FIG. 11 is a diagram showing the attention value of each sentence according to this embodiment.

[0144] The left column of Fig. 11 shows the results of decomposing the original text shown in Fig. 7 into sentences. Fig. 11 shows an example in which the original text is decomposed into seven sentences.

[0145] The numbers on the right side of Figure 11 indicate the attention value of each sentence. Figure 11 shows an example in which the original sentence "Please guide me to my room and lie down" contributes most to the anomalous label obtained by inputting the original sentence into the natural language processing model. This sentence can also be said to be a sentence that receives a high level of attention for the anomalous label.

[0146] The attention value here is a numerical value that indicates to what extent each sentence is receiving attention (to what extent it has influenced the judgment result) in natural language processing using a natural language processing model, and can also be said to indicate the importance of the subject 50's abnormal physical condition.

[0147] In the above, it was explained that the abnormal part extraction unit 155b calculates the attention value of a sentence by summing the attention values ​​of one or more words contained in the sentence. However, the attention value of a sentence may also be calculated, for example, from the average, median, or most frequent value of the attention values ​​of one or more words contained in the sentence.

[0148] 10B , next, the abnormal part extraction unit 155b extracts sentences with high attention values ​​from the plurality of sentences (S134b). The abnormal part extraction unit 155b may extract sentences with attention values ​​equal to or greater than a predetermined value from the plurality of sentences, or may extract a predetermined number of sentences with the highest attention values. The extracted sentences with high attention values ​​are an example of information based on the type of abnormal physical condition.

[0149] Next, the abnormal part extraction unit 155b outputs the extracted sentences with high attention values ​​as abnormal summary sentences to the first summary generation unit 156 (S135b).

[0150] 6 , the first summary generation unit 156 generates a summary based on the acquired normal summary and abnormal summary (S14). For example, the first summary generation unit 156 supplements multiple sentences with high attention values ​​included in the abnormal summary with particles or the like to generate a summary that includes a normal summary and connected sentences that resemble Japanese. The summary generated in step S14 is an example of a first summary.

[0151] The first summary generation unit 156 may store the generated summary in the storage unit 157. The first summary generation unit 156 may also transmit the generated summary to the display terminal unit 30 via the communication network 40. This allows a summary summarizing the condition of the subject 50 on that day to be presented to the user 61. The user 61 can get an overview of the condition of the subject 50 simply by checking the summary, without having to check the care records which include a large amount of text data. This makes it possible to prevent the user 61 from overlooking important items, for example.

[0152] In this way, the support processing unit 15 breaks down the text data into a plurality of sentences, calculates the attention value for each of the plurality of sentences, and generates the first summary sentence based on the attention value for each of the plurality of sentences.

[0153] Next, the support processing unit 15 calculates a graded score for the subject 50 based on the activity data of the subject 50 acquired by the sensing unit 20 (S15).

[0154] FIG. 12 is a flowchart showing the detailed operation (assistance method) of step S15 shown in FIG.

[0155] 12, the transmitter / receiver 11 acquires activity data (e.g., activity data including at least one of a respiratory rate and a heart rate) of the subject 50 for a predetermined period (S151). The predetermined period is, for example, a first period.

[0156] Next, the feature amount calculation unit 13 calculates feature amounts based on the activity data acquired in step S151 (S152). The feature amount calculation unit 13 calculates, for example, feature amounts per hour. For example, the feature amount calculation unit 13 may calculate multiple feature amounts per hour on a target day for detecting the subject's 50's physical condition based on activity data including at least the subject's 50's respiratory rate and heart rate acquired by the transmission / reception unit 11. Note that the number of feature amounts calculated is not particularly limited and may be one or multiple.

[0157] Next, the score calculation unit 14 inputs the feature amounts calculated in step S152 into a previously trained anomaly detection model to obtain an anomaly score for a predetermined period (S153). The score calculation unit 14 calculates an anomaly score per hour, for example, from multiple feature amounts per hour. The score calculation unit 14 inputs the feature amounts calculated by the feature calculation unit 13 into a previously generated anomaly detection model to obtain an anomaly score indicating the degree of abnormal physical condition per hour during a predetermined period including the target day. If the predetermined period is one day, the anomaly score for that day is calculated based on 24 anomaly scores. The anomaly score for that day is, for example, but is not limited to, the average, median, or mode of the 24 anomaly scores.

[0158] Next, the score calculation unit 14 calculates a graded score for indicating the degree of abnormality in the physical condition of the subject 50 in a graded manner based on the abnormality score acquired in step S153 (S154). The score calculation unit 14 calculates, for example, an average value of the abnormality scores on a daily basis from the abnormality scores per hour. In the present embodiment, the score calculation unit 14 calculates an average value of the abnormality scores on a daily basis from the abnormality scores per hour for a predetermined period including the target day for detecting the physical condition of the subject 50.

[0159] Furthermore, the score calculation unit 14 calculates the average and standard deviation of the past abnormality scores for the target day (for example, abnormality scores for the past 90 days or so) to calculate the tiered threshold.

[0160] FIG. 13 is a diagram showing an example of five-level scale scores and their conditions according to this embodiment.

[0161] 13, when calculating a scaled score on a five-level scale, the score calculation unit 14 can calculate a threshold value from the average and the standard deviation. For example, the threshold value for a scaled score of 1 is equal to or less than the average based on the conditions shown in FIG. 13, and the threshold value for a scaled score of 2 is equal to the average plus half the standard deviation.

[0162] The score calculation unit 14 then calculates the tiered score by applying the calculated tiered threshold to the average value of the anomaly scores for the target day. More specifically, the score calculation unit 14 calculates the value of the tiered score by determining the average value of the anomaly scores for the target day using the threshold calculated based on the conditions shown in FIG. 13 .

[0163] Furthermore, the score calculation unit 14 may further check whether the gradation score calculated in step S154 is a value of 4 or 5, i.e., whether a value indicating abnormal physical condition has been calculated. If the gradation score is a value of 4 or 5, a factor analysis may be performed on each of the elements of the heart rate, respiratory rate, bed absence rate, body temperature, and food intake included in the activity data used to calculate the feature amount.

[0164] By calculating the score as described above, it is possible to detect small changes in the subject's 50 physical condition that may lead to an abnormality in the subject's physical condition (i.e., a sign of an abnormality in the physical condition).

[0165] 6 , the first summary generator 156 generates a daily summary including the graded score, the state label (inference label), and the summary (S16) and transmits the daily summary to the display terminal 30 via the communication network 40 (S17). Note that the summary generated in step S14 may be transmitted as the daily summary. Step S17 is an example of outputting presentation information based on the determination result of the presence or absence of a physical condition abnormality. Information indicating the daily summary is also an example of presentation information.

[0166] FIG. 14 is a diagram showing an example of a daily summary including a score according to this embodiment. FIG. 14 shows an example of a daily summary generated in step S16. The daily summary may include at least the date and the situation on the previous day. The situation on the previous day corresponds to the summary generated in step S14. For example, the presented information may include at least the summary generated in step S14. Furthermore, for example, the presented information may include at least the summary and the condition classification result (the factors (records) shown in FIG. 14) generated in step S14. Note that FIG. 14 shows the daily summary of subject 50 for one week from February 12 to February 18.

[0167] 14, the daily summary includes a date, a score (here, a graded score), a factor (sensor), a factor (record), and the state of the previous day. For example, the table shown in FIG. 14 is presented on the display terminal unit 30.

[0168] The date is the date on which the score etc. is calculated.

[0169] The score is the graded score calculated in step S154, but may be, for example, the abnormality score acquired in step S153. The graded score and the abnormality score are examples of a score based on the abnormality score.

[0170] The factor (sensor) indicates activity data of the subject 50 acquired by the sensing unit 20. In the example of FIG. 14, the bed presence / absence rate is shown. The value of the factor (sensor) may be entered only on days when the score is equal to or greater than a predetermined value (3 or greater in the example of FIG. 14). Furthermore, the factor (sensor) may include information indicating factors that result in a high score, which are considered from sensing data when the score is equal to or greater than a predetermined value.

[0171] The cause (record) indicates a state label, i.e., the cause (record) indicates the classification result of the state classification unit 154.

[0172] The previous day's condition is extracted from the previous day's care record and is likely to lead to changes in the current day. Note that the current day's condition may also be recorded regardless of the previous day. The previous day's condition is the information that forms the basis for calculating the score.

[0173] In this way, by presenting both the score and the basis for the score calculation, the "meaning of the score" can be made concrete. Presenting such a daily summary allows the user 61 to correctly understand the score and use it as a hint for the next action. For example, by presenting factors (records) indicating the state of the care record and a summary generated from the text data of the care record (in the example of FIG. 14 ) in chronological order along with the score, the user 61 can be encouraged to change their behavior more than when only the score is displayed, thereby realizing proactive care. Furthermore, by presenting the score and other information in chronological order, the subject 50's condition can be understood, including the condition of the subject 50 shortly before, which can contribute to preventing the subject 50's physical condition from being overlooked.

[0174] Next, the first summary generator 156 stores the generated daily summary in the storage unit 157 (S18). As a result, the daily summary is accumulated in the storage unit 157.

[0175] Steps S11 to S18 are repeatedly executed for each first period (for example, each day).

[0176] Next, the second summary generation unit 158 ​​generates a summary for a second period longer than the first period, based on the daily summaries stored in the storage unit 157. The summary for the second period is generated based on the summary for the first period. In this embodiment, the second summary generation unit 158 ​​generates a weekly summary based on one week's worth of daily summaries (S19).

[0177] The second summary generation unit 158 ​​extracts, for example, descriptions that are of particular interest from one week's worth of daily summaries and generates a weekly summary that includes the extracted descriptions. The second summary generation unit 158 ​​generates a weekly summary based on, for example, the attention value of each sentence in one week's worth of daily summaries. The second summary generation unit 158 ​​extracts, for example, from one week's worth of daily summaries, sentences with attention values ​​equal to or greater than a predetermined value, or a predetermined number of sentences with the highest attention values, and generates a weekly summary based on the extracted one or more sentences. Information indicating the weekly summary is an example of presentation information.

[0178] FIG. 15 is a diagram showing an example of a weekly summary according to the present embodiment.

[0179] As shown in FIG. 15, the weekly summary includes the number of days on which each state (abnormality label) occurred, the details of the main abnormality, and the main normal record.

[0180] The number of days each condition occurred indicates the number of times (number of days) each abnormality occurred in a week.

[0181] The main anomaly content is the anomaly content extracted from the daily summary sentence based on the attention value.

[0182] The main normal record is the normal content extracted from the daily summary sentence.

[0183] By presenting such a weekly summary to the user 61, the user 61 can get an overview of the condition of the subject 50 for one week and can use the weekly summary to plan care for the following week. Furthermore, because the necessary content is extracted and presented from the large volume of records, i.e., one week's worth of care records, the user 61 can easily understand the health condition of the subject 50 for one week. Furthermore, because the weekly summary summarizes the condition of the subject 50 for one week, it is possible to reduce the time required by the user 61 to prepare a report on the user's condition, prepare a report for family members, etc.

[0184] Furthermore, when the sensing unit 20 acquires activity data obtained by sensing the subject 50 for that one week, the second summary generation unit 158 ​​may further include the activity data for that one week in the weekly summary. For example, the second summary generation unit 158 ​​may include the activity data for that one week in the weekly summary, or may include the activity data for a day on which an abnormality occurred in the subject 50 in the weekly summary.

[0185] 16 is a diagram showing an example of a weekly summary including activity data according to the present embodiment. Information indicating the weekly summary including activity data shown in FIG. 16 is an example of presentation information.

[0186] As shown in Figure 16, the weekly summary may include a week's worth of activity data in addition to the summary. The activity data is displayed as a graph showing changes over time, but may also be displayed numerically, for example. The activity data may include, for example, sleep time, respiratory rate, and heart rate, but may also include other activity data.

[0187] The weekly summary may also include a graph showing the change over time in the score for one week (for example, the scaled score calculated in step S15).

[0188] In addition, the graph may highlight data on days when an abnormality occurred. The day on which an abnormality occurred refers to, for example, the day on which an abnormality described in the weekly summary occurred. For example, if multiple abnormalities occurred in a week, the data on days on which a particularly significant abnormality occurred may be highlighted.

[0189] Next, the second summary generator 158 transmits the weekly summary generated in step S19 to the display terminal unit 30 via the communication network 40 (S20). This allows the weekly summary to be presented to the user 61. Step S20 is an example of outputting presentation information based on the determination result of the presence or absence of an abnormality in physical condition.

[0190] Next, the operation of counting abnormal labels for the subject 50 will be described with reference to FIGS. 17 to 20. FIG. 17 is a flowchart showing a second operation (assistance method) of the information management server 10 according to this embodiment. Steps S31 and S32 shown in FIG. 17 are similar to steps S11 and S12 shown in FIG. 6, and therefore will not be described again.

[0191] 17 , the tallying unit 159 collects records of a specific abnormality (label of interest) from among the multiple abnormalities associated with the subject 50 (S33). The specific abnormality may be set by, for example, the user 61. Furthermore, the records of each of the multiple abnormalities associated with the subject 50 may be stored in the storage unit 157, and in step S33, the tallying unit 159 may read out the record of the specific abnormality from the storage unit 157.

[0192] Next, the counting unit 159 counts the number of occurrences for each date and time of occurrence of anomalies included in the collected anomaly records, performs analysis (S34), and outputs the analysis results (counting results) to the display terminal unit 30 (S35). Counting the number of occurrences for each date and time of occurrence of anomalies is an example of counting the dates and times of occurrence of anomalies. Furthermore, information based on the counting results is the information shown in the following Figures 18 to 20, and is an example of presentation information.

[0193] 18 to 20 are diagrams showing examples of tabulated information according to this embodiment. FIG. 18 shows, for example, the results of an analysis of abnormality trends in the entire facility (e.g., a nursing home). Each room may contain one subject 50, or multiple subjects 50. FIGS. 19 and 20 show the results of an analysis of the tabulated results for the subjects 50.

[0194] As shown in FIG. 18, the analysis results include the abnormality trend, the number of rooms, and the trend content.

[0195] The trend indicates whether an abnormality occurs locally or globally within the facility where the system is installed. In other words, the trend indicates the occurrence status of the abnormality at that time. Note that local occurrence means that the abnormality tends to occur only in a small number of rooms (i.e., a small number of people) within the facility where the system is installed, meaning that there is a bias in the occurrence of the abnormality.

[0196] The number of rooms indicates the number of rooms in which a certain abnormality occurs.

[0197] The trend content indicates the tendency of abnormality occurrence status when viewed over time.

[0198] For example, the top row of the three rows shows that there are eight rooms in which anomalies (labeled anomalies) that tend to occur locally within the facility where the system is installed, and that these anomalies tend to occur more frequently in the summer.The second row of the three rows shows that there are two rooms in which anomalies (labeled anomalies) that tend to occur globally, and that these anomalies tend to increase after the summer.

[0199] FIG. 19 shows the results of tallying the number of anomalies that occur by month. In the example of FIG. 19, many specific anomalies occur in September. This shows that the subject 50 tends to have specific anomalies more easily in September. By presenting (visualizing) the graph shown in FIG. 19 on the display terminal unit 30 of the user 61, the user 61 can be informed of the analysis results of which anomalies are more likely to occur in which seasons for the subject 50. The graph shown in FIG. 19 is an example of information showing the influence of seasons on the occurrence of specific anomalies.

[0200] FIG. 20 shows the results of tallying the number of abnormalities by month and by time period. The time periods include morning, afternoon, evening, night, and unknown time period. This shows that the subject 50 tends to experience specific abnormalities more frequently in winter mornings. By presenting (visualizing) the graph shown in FIG. 20 on the display terminal unit 30 of the user 61, the user 61 can be informed of the analysis results of which abnormalities are more likely to occur in which time periods for the subject 50. The graph shown in FIG. 20 is an example of information showing the influence of time on the occurrence of specific abnormalities.

[0201] In this way, since the occurrence tendency of abnormalities that differ for each subject 50 can be known, the user 61 can efficiently deal with the subject 50.

[0202] The counting shown in FIGS. 19 and 20 is performed for each subject 50.

[0203] It is sufficient that at least one of the tables or graphs shown in FIGS. 18 to 20 is presented to the user 61 as presented information.

[0204] (Modification of the Embodiment) In this modification, generation of the machine learning model used in the embodiment will be described with reference to FIGS. 21 and 22. FIG.

[0205] FIG. 21 is a block diagram showing an example of the functional configuration of a model generating device 200 according to this modification.

[0206] 21 , the model generating device 200 includes a transmitting / receiving unit 210, an extracting unit 220, and a training unit 230. The model generating device 200 is realized by, for example, a computer including a processor (microprocessor), a memory, a communication interface, etc.

[0207] The transmission / reception unit 210 includes, for example, a communication interface, and transmits and receives various information between the care record collection unit 25 and the information management server 10 via the communication network 40. For example, the transmission / reception unit 210 acquires care records of the subject 50 for a predetermined period from the care record collection unit 25. Furthermore, for example, the transmission / reception unit 210 acquires correct answer data (teacher data) indicating normality or abnormality (e.g., type of abnormality) for text data included in the acquired care records from the display terminal unit 30 or another terminal device. Furthermore, for example, the transmission / reception unit 210 outputs the machine learning model generated by the model generation device 200 to the information management server 10.

[0208] The extraction unit 220 extracts text portions to be used for training from text data included in the care records. The extraction unit 220 extracts only the text portions necessary for training from the text data. For example, the extraction unit 220 may exclude from the text data descriptions following title words such as "near-accident" or "condition observation" that frequently appear in care records, or fixed phrases. Furthermore, for example, the extraction unit 220 may extract descriptions from the care records that are likely to contain records related to abnormalities.

[0209] The extraction unit 220 may extract from the nursing record specific items, such as special notes in the nursing record, observation details, condition observations, response records, today's situation, emergency medication, consultation contact, accident records, or descriptions following a title word, or descriptions before and after it, as items indicating the condition of the subject 50.

[0210] Furthermore, the extraction unit 220 may extract descriptions of the subject 50's predetermined behavior, such as eating, bathing, excretion, etc., that may indicate an abnormality. Since eating, bathing, excretion, etc. are often recorded in the care record even if no abnormality is present, it is preferable to extract descriptions of changes in condition, discontinuation, etc.

[0211] The training unit 230 trains (learns) the machine learning model using the extracted text portion and the correct answer data. The training method is not particularly limited, and any known method may be used. For example, the training unit 230 may update the parameters of the machine learning model using backpropagation based on the prediction error between the text portion and the correct answer data that serves as the true value. In other words, the training unit 230 performs training to update the parameters so as to minimize the error (difference) between the text portion and the correct answer data that serves as the true value. In this way, the training unit 230 performs training by performing backpropagation learning between the text portion and the correct answer data.

[0212] Next, a model generation method in the model generation device 200 configured as described above will be described with reference to Fig. 22. Fig. 22 is a flowchart showing the operation (assistance method) of generating a machine learning model according to this modification. Fig. 22 explains the generation of a natural language processing model.

[0213] As shown in FIG. 22, the transmitting / receiving unit 210 acquires a care record including text data and correct answer data indicating the result of the determination of the presence or absence of abnormal physical condition (S41).

[0214] Next, the extraction unit 220 extracts text portions to be used for training from the text data included in the acquired care records (S42). For example, the extraction unit 220 may extract, from the text data, descriptions that are likely to contain records related to abnormalities as text portions. Step S42 is a preprocessing step prior to training the machine learning model.

[0215] Next, the training unit 230 uses the extracted text portion and the correct answer data to train a natural language processing model, which is an example of a machine learning model (S43).

[0216] Next, the transmitting / receiving unit 210 outputs the natural language processing model generated by the training unit 230 to the information management server 10 (S44).

[0217] The process of step S42 may be omitted.

[0218] (Other Embodiments) While the support methods etc. according to one or more aspects have been described above based on the embodiments etc., the present disclosure is not limited to these embodiments etc. As long as they do not deviate from the spirit of the present disclosure, various modifications conceivable by a person skilled in the art to the present embodiments and embodiments constructed by combining components of different embodiments may also be included in the present disclosure.

[0219] For example, the machine learning model described in the above embodiments may be a neural network (NN), such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a long-short term memory (LSTM).

[0220] In the above-described embodiments, the normal summary output model and the abnormal part summary model are machine learning models, but the present invention is not limited to this and may be realized by a learning model other than a machine learning model. The normal summary output model and the abnormal part summary model are examples of learning models.

[0221] Furthermore, for example, although some of the machine learning models in the above embodiments and the like are supervised models, they may be unsupervised models.

[0222] In addition, although the above-described embodiments and the like have described examples in which the information management server and the model generation device are separate entities, they may be realized as an integrated device. For example, the information management server may have at least one of the functions (e.g., all of the functions) of the model generation device.

[0223] In the above embodiments, each component may be configured with dedicated hardware, or may be realized by executing a software program suitable for each component. Each component may be realized by a program execution unit such as a CPU or processor reading and executing a software program recorded on a recording medium such as a hard disk or semiconductor memory.

[0224] The order in which the steps in the flowchart are executed is merely an example for specifically explaining the present disclosure, and other orders may be used. Some of the steps may be executed simultaneously (in parallel) with other steps, or some of the steps may not be executed.

[0225] The division of functional blocks in the block diagram is an example, and multiple functional blocks may be realized as a single functional block, one functional block may be divided into multiple blocks, or some functions may be moved to another functional block.Furthermore, the functions of multiple functional blocks having similar functions may be processed in parallel or time-shared by a single piece of hardware or software.

[0226] Furthermore, the information management server according to the above-described embodiments may be realized as a single device or may be realized by multiple devices. When the information management server is realized by multiple devices, the components of the information management server may be distributed among the multiple devices in any manner. When the information management server is realized by multiple devices, the communication method between the multiple devices is not particularly limited, and may be wireless communication or wired communication. Furthermore, wireless communication and wired communication may be combined between the devices.

[0227] Furthermore, each component described in the above embodiments may be implemented as software or, typically, as an LSI, which is an integrated circuit. These components may be individually integrated into a single chip, or some or all of them may be integrated into a single chip. Here, the term "LSI" is used, but depending on the level of integration, it may also be referred to as an IC, system LSI, super LSI, or ultra LSI. Furthermore, the integrated circuit implementation method is not limited to LSI, and may be implemented using a dedicated circuit (a general-purpose circuit that executes a dedicated program) or a general-purpose processor. After LSI fabrication, a field programmable gate array (FPGA) that can be programmed or a reconfigurable processor that can reconfigure the connections or settings of circuit cells within the LSI may also be used. Furthermore, if an integrated circuit technology that replaces LSI emerges due to advances in semiconductor technology or a derivative technology, that technology may naturally be used to integrate the components.

[0228] A system LSI is an ultra-multifunctional LSI manufactured by integrating multiple processing units on a single chip. Specifically, it is a computer system that includes a microprocessor, ROM (Read Only Memory), RAM (Random Access Memory), etc. Computer programs are stored in the ROM. The system LSI achieves its functions when the microprocessor operates in accordance with the computer program.

[0229] Furthermore, one aspect of the present disclosure may be a computer program that causes a computer to execute each of the characteristic steps included in the support method shown in any of Figures 6, 7, 12, 17, and 22.

[0230] Furthermore, for example, the program may be a program to be executed by a computer. Another aspect of the present disclosure may be a computer-readable non-transitory recording medium on which such a program is recorded. For example, such a program may be recorded on a recording medium and distributed or circulated. For example, the distributed program may be installed in a device having another processor, and the program may be executed by the processor, thereby causing the device to perform each of the above processes.

[0231] The present disclosure is useful for a support method for supporting a user in understanding the health condition of a subject based on care records.

[0232] DESCRIPTION OF SYMBOLS 10 Information management server 11, 210 Transmission / reception unit 12, 152 Information recording unit 13 Feature calculation unit 14 Score calculation unit 15 Support processing unit 20 Sensing unit 25 Care record collection unit 30 Display terminal unit (information terminal) 40 Communication network 50 Subject 60, 61 User 100 Physical condition detection system 151 Transmission / reception unit (acquisition unit, output unit) 153 Calibration processing unit 154 State classification unit (determination unit) 155a Normal part extraction unit 155b Abnormal part extraction unit 156 First summary generation unit 157 Storage unit 158 ​​Second summary generation unit 159 Aggregation unit 200 Model generation device 220 Extraction unit 230 Training unit

Claims

1. A method for assisting a user in understanding the health condition of a subject, comprising the steps of: acquiring care records of the subject, including text data, for a first time period; using a natural language processing model, determining whether or not the subject has an abnormality in health from the text data; and outputting presentation information based on the determination result of whether or not the subject has an abnormality in health to an information terminal of the user.

2. The support method according to claim 1, further comprising: if the subject's physical condition is abnormal, determining a type of abnormal physical condition from the text data using the natural language processing model; and the presented information includes information based on the determined type of abnormal physical condition.

3. The support method described in claim 2, further comprising: using a learning model to extract from the nursing records at least one of normal points which are descriptions indicating that the subject is normal and abnormal points which are descriptions indicating that the subject is abnormal; generating a first summary sentence indicating the subject's health condition during the first period based on the extracted normal points and / or abnormal points; and the presented information including the first summary sentence.

4. The support method described in claim 3, further comprising: dividing the text data into a plurality of sentences; calculating an attention value for each of the plurality of sentences indicating the degree to which the sentence contributed to the determination result of the type of physical abnormality; and generating the first summary sentence based on the attention value for each of the plurality of sentences.

5. The support method according to claim 4, further comprising extracting one or more sentences from the plurality of sentences, the one or more sentences having the attention value of a first predetermined value, and generating the first summary sentence based on the extracted one or more sentences.

6. The support method described in claim 5, further comprising: acquiring activity data of the subject during the first period; calculating a feature based on the acquired activity data; acquiring an abnormality score indicating the degree of abnormality in physical condition during the first period based on the feature; and the presented information including information in which a score based on the abnormality score is associated with the first summary sentence.

7. The support method described in claim 2, wherein a second period is composed of two or more of the first periods; using a learning model to extract from the care records at least one of normal parts which are descriptions indicating that the subject is normal and abnormal parts which are descriptions indicating that the subject is abnormal; generating a first summary sentence indicating the subject's health condition during the first period based on at least one of the extracted normal parts and abnormal parts; generating a second summary sentence indicating the subject's health condition during the second period based on two or more of the first summary sentences for each of the two or more first periods constituting the second period and an attention value for each of the two or more first summary sentences which indicates the degree to which they contributed to the result of the abnormal physical condition judgment; and the presented information includes the second summary sentence.

8. The support method described in claim 7, further comprising extracting one or more sentences from the two or more first summary sentences, the attention value of which is a second predetermined value, and generating the second summary sentence based on the extracted one or more sentences.

9. The support method according to claim 7 or 8, further comprising acquiring activity data of the subject obtained by sensing the subject during the second period, and outputting time series data of the activity data to the information terminal of the user in association with the second summary sentence.

10. A support method according to any one of claims 1 to 8, further comprising: collecting records of a specific abnormality from among a plurality of abnormalities associated with the subject; aggregating the dates and times of occurrence of two or more of the collected abnormalities; and the presented information including information based on the aggregation results.

11. The support method according to claim 10, wherein the information based on the aggregation results includes information indicating the influence of season or time of day on the occurrence of the specific anomaly.

12. A support method according to any one of claims 1 to 8, further comprising extracting from the nursing care records descriptions relating to at least one of specified items, specified behavior of the subject, and reports on medical visits excluding regular visits, and the text data includes the extracted descriptions.

13. An assistance device for assisting a user in understanding the health condition of a subject, comprising: an acquisition unit for acquiring nursing care records of the subject including text data for a first period; a determination unit for determining whether or not the subject has an abnormal physical condition from the text data using a natural language processing model; and an output unit for outputting presentation information based on the determination result of the presence or absence of the abnormal physical condition to an information terminal of the user.

14. A program for causing a computer to execute the support method according to any one of claims 1 to 8.

Images