Health condition detection method, health condition detection device, and program

Abstract

This physical condition detection method performed by a computer involves: acquiring activity data including the respiratory rate and heart rate of a subject over a predetermined period of time; calculating a plurality of feature amounts, on the basis of the acquired activity data; acquiring an abnormality score indicating the degree of physical abnormality per predetermined period by inputting the plurality of calculated feature amounts into a model that has trained the normality or abnormality in an activity data group; calculating a graded score for indicating the degree of abnormality of the subject's physical condition in stages, on the basis of the acquired abnormality score; and outputting the calculated graded score.

Description

Technical Field

[0001] The present disclosure relates to a physical condition detection method, a physical condition detection device, and a program.

Background Art

[0002] As a problem due to the aging society in 2025, that is, when all 8 million people of the so-called "baby boom generation" reach the age of 75 or older and one in four nationals is 75 or older, there is a problem of shortage of manpower due to the increasing demand for medical care and nursing care.

[0003] Therefore, the number of care recipients and care recipients in charge of medical and care workers increases, and there may be cases where small changes in physical condition that lead to health abnormalities are overlooked. And if small changes in physical condition are overlooked, there is a risk that the condition of the subject will deteriorate.

[0004] On the other hand, for example, in Patent Document 1, a technique for notifying an appropriate notification destination of the abnormality of a monitored person when the monitored person is determined to be abnormal is disclosed. Thereby, it is possible to notify an appropriate monitor of the abnormality of the monitored person according to the abnormal state of the monitored person.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] However, the above Patent Document 1 only discloses a technique for notifying when the vital information of the monitored person acquired from the sensor is an abnormal value, and cannot detect a small change in physical condition that leads to the health abnormality of the monitored person, that is, a sign of health abnormality.

[0007] This disclosure is made in light of the circumstances described above, and aims to provide a method for detecting signs of health abnormalities in a subject. [Means for solving the problem]

[0008] A physical condition detection method according to one aspect of the present disclosure is a computer-based physical condition detection method that acquires activity data including the respiratory rate and heart rate of a subject over a predetermined period, calculates a plurality of features based on the acquired activity data, inputs the calculated plurality of features into a model that has learned the normality or abnormality of the activity data set consisting of the plurality of features, thereby acquiring an abnormality score indicating the degree of physical abnormality per predetermined period, calculates a graded score for indicating the degree of physical abnormality of the subject in stages based on the acquired abnormality score, and outputs the calculated graded score.

[0009] These general or specific embodiments may be implemented as systems, devices, methods, integrated circuits, computer programs, or recording media such as computer-readable CD-ROMs, or as any combination of systems, devices, methods, integrated circuits, computer programs, and recording media. [Effects of the Invention]

[0010] According to the health condition detection method described herein, it is possible to detect signs of health abnormalities in the subject, that is, small changes in the subject's physical condition that could lead to health abnormalities. [Brief explanation of the drawing]

[0011] [Figure 1] Figure 1 shows an example of the configuration of a health condition detection system according to an embodiment. [Figure 2] Figure 2 is a block diagram showing an example of a specific configuration of an information management server according to the embodiment. [Figure 3] Figure 3 shows an example of the distribution of abnormal data and normal data in the activity data group according to the embodiment. [Figure 4]Figure 4 is a conceptual diagram showing a model according to an embodiment. [Figure 5] Figure 5 shows an example of a five-stage scoring system and its conditions according to the embodiment. [Figure 6] Figure 6 shows an example of a display for responding to abnormal physical conditions of a subject according to the embodiment. [Figure 7] Figure 7 shows an example of a display for responding to abnormal physical conditions of a subject according to the embodiment. [Figure 8] Figure 8 shows an example of a display for responding to abnormal physical conditions of a subject according to the embodiment. [Figure 9] Figure 9 is a flowchart illustrating the overview of the operation of the information management server according to the embodiment. [Figure 10] Figure 10 is a flowchart showing an example of the operation of the information management server according to the embodiment. [Figure 11A] Figure 11A is a diagram that conceptually explains anomaly detection when the model related to the comparative example is operated over a long period of time. [Figure 11B] Figure 11B is a diagram that conceptually explains anomaly detection when the model related to the comparative example is operated over a long period of time. [Figure 12] Figure 12 is a diagram that conceptually explains anomaly detection when the model according to the embodiment is operated over a long period of time. [Figure 13] Figure 13 is a conceptual diagram illustrating the performance improvement achieved by updating the model according to the embodiment. [Figure 14] Figure 14 shows an example of a linked display shown by the display terminal unit according to Embodiment 1. [Figure 15] Figure 15 shows Incident Discovery Example 1 related to Example 2. [Figure 16] Figure 16 shows Incident Discovery Example 2 related to Example 2. [Figure 17] Figure 17 shows Incident Discovery Example 3 related to Example 2. [Modes for carrying out the invention]

[0012] A physical condition detection method according to one aspect of the present disclosure is a physical condition detection method performed by a computer. The method includes obtaining activity data including the respiration rate and heart rate of a subject over a predetermined period, calculating a plurality of feature amounts based on the obtained activity data, and inputting the calculated plurality of feature amounts into a model that has learned the normality or abnormality in an activity data group composed of the plurality of feature amounts, thereby obtaining an abnormality score indicating the degree of physical condition abnormality per the predetermined period. Based on the obtained abnormality score, a staged score for gradually indicating the degree of physical condition abnormality of the subject is calculated, and the calculated staged score is output.

[0013] According to this aspect, it is possible to detect a sign of a health abnormality of a subject, that is, a small change in the physical condition leading to the health abnormality of the subject.

[0014] More specifically, a plurality of feature amounts are calculated from the activity data, and the calculated plurality of feature amounts are input into a model in which the normality or abnormality in the activity data group has been learned. Based on the abnormality score obtained as a result, a staged score for gradually evaluating the degree of physical condition abnormality of the subject is calculated.

[0015] Thereby, it is possible to know (detect) a sign of a health abnormality of the subject from the staged score.

[0016] Therefore, it is possible to detect a sign of a health abnormality of the subject that may not be noticed even by medical staff who care for or nurse the subject.

[0017] Furthermore, since the degree of abnormality can be easily grasped on-site by the staged score, it becomes easier to respond to various small changes in physical condition leading to health abnormalities in the daily life of the subject, that is, signs of health abnormalities.

[0018] Furthermore, for example, when calculating the grading score, if the grading score is equal to or greater than a predetermined value, a factor analysis may be performed to analyze whether or not each element included in the activity data is a factor, and when outputting the grading score, the elements that were analyzed as factors by the factor analysis and the grading score may be output.

[0019] In this way, if the graded score indicates a threshold above a certain value that requires attention to the signs of a health problem, the system will also notify the user of factors such as whether the small changes in physical condition that could lead to the health problem are related to heart rate or respiratory rate.

[0020] Therefore, healthcare professionals who care for or nurse the individuals concerned can use the notified factors as a clue to respond early and appropriately to signs of health abnormalities.

[0021] Here, for example, the activity data may include at least the respiratory rate and heart rate from among the subject's food intake, respiratory rate, heart rate, and bed absence rate during the predetermined period.

[0022] Thus, the activity data consists of daily activity data obtained on-site, including at least respiratory rate and heart rate, from among the subject's food intake, respiratory rate, heart rate, and bed absence rate. By calculating multiple features from this daily activity data obtained on-site, it becomes possible to detect early signs of health abnormalities with greater accuracy.

[0023] Furthermore, for example, when calculating the multiple features, at least the mean and maximum values ​​of the respiratory rate and heart rate from the respiratory rate, the difference data of the respiratory rate, the heart rate, and the mean, maximum, standard deviation, skewness, kurtosis, and impulse factor obtained by subtracting the mean from the maximum value may be calculated as the multiple features.

[0024] In this way, activity data is subjected to statistical processing to calculate multiple features. This allows for the acquisition of more accurate anomaly scores from these calculated features using a pre-trained model.

[0025] Furthermore, for example, the model may learn the normality or abnormality of the activity data set through unsupervised learning using the activity data set.

[0026] In this way, by using activity data obtained daily in the field and performing unsupervised learning on the model, a trained model capable of detecting early signs of health abnormalities can be obtained. Therefore, when obtaining a trained model capable of detecting early signs of health abnormalities using the activity data of the subjects, the model can be obtained without burdening the on-site staff such as healthcare workers of the subjects.

[0027] Furthermore, for example, the aforementioned model is one that separates outliers based on a decision tree.

[0028] In activity data sets, abnormal data occurs less frequently and is located in a different distribution than normal data. Models that separate outliers based on decision trees can utilize this property to detect early signs of health abnormalities from a subject's activity data.

[0029] Here, for example, the aforementioned model may be an Isolation Forest model.

[0030] Furthermore, for example, the model may be updated periodically using the acquired activity data.

[0031] This allows for repeated model updates using activity data that includes activity data accumulated after model creation, in addition to the activity data used during model creation. Therefore, the model can detect early signs of health abnormalities while also responding to medium- to long-term fluctuations due to disease or environmental influences in the subjects.

[0032] Furthermore, for example, when outputting the graded score, the calculated graded score may be output to the terminal of the person's monitor, and the user interface of the terminal may be displayed to allow the monitor to take action regarding any abnormalities in the person's physical condition.

[0033] This allows for the display of information to indicate any changes in the subject's physical condition, making it easier for on-site staff to identify any health problems and respond appropriately. In other words, on-site staff can more easily respond to various small changes in the subject's physical condition that could lead to health problems in their daily lives—that is, early signs of health problems.

[0034] A physical condition detection device according to one aspect of the present disclosure includes: a transmitting and receiving unit that acquires activity data including the respiratory rate and heart rate of a subject over a predetermined period; a feature calculation unit that calculates a plurality of features based on the acquired activity data; an abnormality score calculation unit that acquires an abnormality score indicating the degree of physical abnormality per predetermined period by inputting the calculated plurality of features into a model that has been learned by a model creation unit to determine the normality or abnormality of the activity data group consisting of the plurality of features; and a gradation score calculation unit that calculates a gradation score to indicate the degree of physical abnormality of the subject in stages based on the acquired abnormality score, wherein the transmitting and receiving unit outputs the calculated gradation score.

[0035] Furthermore, a program according to one aspect of this disclosure causes a computer to perform the following actions: acquire activity data including the respiratory rate and heart rate of a subject over a predetermined period; calculate a plurality of features based on the acquired activity data; input the calculated plurality of features into a model (model creation unit) that has learned the normality or abnormality of the activity data set consisting of the plurality of features; acquire an abnormality score indicating the degree of physical abnormality per predetermined period; calculate a graded score to indicate the degree of physical abnormality of the subject in stages based on the acquired abnormality score; and output the calculated graded score.

[0036] These comprehensive or specific embodiments may be implemented as systems, devices, methods, integrated circuits, computer programs, or recording media such as computer-readable CD-ROMs, or as any combination of systems, devices, methods, integrated circuits, computer programs, and recording media.

[0037] The embodiments of this disclosure will be described below with reference to the drawings. Each embodiment described below is a specific example of this disclosure. The numerical values, shapes, components, steps, and order of steps shown in the following embodiments are examples only and are not intended to limit this disclosure. Furthermore, any components in the following embodiments that are not described in an independent claim will be described as optional components. Also, in all embodiments, the contents of each can be combined.

[0038] (Embodiment) The following will explain the physical condition detection method and other related aspects according to this embodiment, with reference to the drawings.

[0039] [1 Health Condition Detection System 100] Figure 1 shows an example of the configuration of the health condition detection system 100 according to this embodiment.

[0040] The physical condition detection system 100 according to this embodiment is a system configured in which the information management server 10 detects minor changes in the physical condition of a person being nursed or cared for that could lead to a health problem (i.e., a precursor to a health problem).

[0041] As shown in Figure 1, the health condition detection system 100 comprises an information management server 10, a sensing unit 20, and a display terminal unit 30. These are connected by a communication network 40. The communication network 40 may be a wired network, a wireless network, or both. Figure 1 also shows a person 50 receiving nursing or care, a user 60 who is a medical professional or other on-site staff member providing nursing or care to the person 50, a user 61 who is a monitor of the person 50 and can view the display terminal unit 30, and recorded data 25 that records the details of the nursing or care provided by the user 60 to the person 50. The recorded data 25 includes, for example, the amount of food consumed by the person 50 in the morning, noon, and evening, as entered by the on-site staff user 60.

[0042] Although Figure 1 shows an example where the health condition detection system 100 has one sensing unit 20, it is not limited to this, and it is sufficient to have as many sensing units 20 as there are people 50 receiving nursing or care.

[0043] [1.1 Sensing Unit 20] The sensing unit 20 acquires activity data, including the respiratory rate and heart rate of the subject 50, by sensing it over a predetermined period. In this embodiment, the sensing unit 20 acquires data such as heart rate, respiratory rate, and body movement (hereinafter also referred to as sensor data) every second while the subject 50 is in bed.

[0044] Furthermore, the interval for acquiring sensor data such as heart rate, respiratory rate, and body movement is not limited to, for example, 1 second, but can be 2 seconds or any other interval that allows changes in the sensor data of the subject 50 to be detected. In addition, the sensing unit 20 may sense whether the subject 50 is in or out of bed, depending on whether it can sense heart rate, respiratory rate, body movement, etc., and may also sense lifestyle rhythms such as sleep status.

[0045] Furthermore, the sensing unit 20 may be a sensor device having, for example, a pressure sensor, and may be installed on the bed to sense the subject 50 every second. In this case, the sensing unit 20 may output a value of 1 every second as sensor data indicating that the subject 50 is out of bed. Alternatively, the sensing unit 20 may output sensor data values ​​such as the subject 50's respiratory rate every second.

[0046] [1.2 Information Management Server 10] Figure 2 is a block diagram showing an example of a specific configuration of the information management server 10 according to this embodiment.

[0047] The information management server 10 is implemented as a computer equipped with, for example, a processor (microprocessor), memory, a communication interface, etc. The information management server 10 may also operate with some components included in a cloud server. The information management server 10 is an example of a health condition detection device and detects minor changes in the health of 50 subjects that could lead to health problems (i.e., precursors to health problems).

[0048] In this embodiment, the information management server 10 includes a transmitting / receiving unit 11, an information recording unit 12, a feature calculation unit 13, a model creation unit 14, a model update unit 15, and a health condition detection unit 16, as shown in Figure 2.

[0049] [1.2.1 Transmitter / Receiver Unit 11] The transmitting / receiving unit 11, for example, is equipped with 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 transmitting / receiving unit 11 acquires activity data, including the respiratory rate and heart rate of the subject 50 over a predetermined period. Here, the activity data includes, as described above, at least the respiratory rate and heart rate from the subject 50's food intake, respiratory rate, heart rate, and bed absence rate over the predetermined period. The transmitting / receiving unit 11 also outputs the graded score calculated by the health condition detection unit 16 to the terminal of a user 61, such as the subject 50's monitor.

[0050] In this embodiment, the transmitting / receiving unit 11 acquires sensor data such as heart rate, respiratory rate, and body movement per second while the subject 50 is in bed from the sensing unit 20 via the communication network 40 at predetermined intervals, for example, every minute. The transmitting / receiving unit 11 also acquires recorded data 25, which records the details of nursing or care provided to the subject 50 by a user 60, who is a field staff member, as shown in Figure 1. In this way, the transmitting / receiving unit 11 acquires activity data, including sensor data and recorded data 25, which is activity data obtained daily at the site, via the communication network 40. The transmitting / receiving unit 11 also transmits the graded score calculated by the physical condition detection unit 16 to the display terminal unit 30 via the communication network 40. The transmitting / receiving unit 11 may also transmit information for the display interface of the display terminal unit 30, such as a graded score display, vital sign fluctuation graph display, or risk group display, which are used to enable the user 61 to respond to abnormalities in the physical condition of the subject 50.

[0051] [1.2.2 Information Recording Unit 12] The information recording unit 12 records the information transmitted and received by the transmitting and receiving unit 11. The information recording unit 12 is a recording medium capable of recording information, and is composed of a rewritable non-volatile memory such as a hard disk drive or a solid-state drive. The information recording unit 12 may also record multiple feature quantities calculated by the feature quantity calculation unit 13.

[0052] [1.2.3 Feature Calculation Unit 13] The feature calculation unit 13 includes, for example, a computer including memory and a processor (microprocessor), and the processor executes a control program stored in memory to realize the function of calculating multiple features. The feature calculation unit 13 calculates multiple features based on activity data, including the respiratory rate and heart rate of the subject 50, acquired by the transmitting / receiving unit 11. For example, the feature calculation unit 13 acquires sensor data for a time period including the target date and time for health condition detection from the activity data acquired by the transmitting / receiving unit 11 or recorded in the information recording unit 12, and calculates hourly features for each sensor data such as respiratory rate. Here, the feature calculation unit 13 calculates at least the average and maximum values ​​of the subject 50's respiratory rate and the average and maximum values ​​of the subject 50's heart rate as multiple hourly features.

[0053] In this embodiment, the feature calculation unit 13 calculates multiple features from at least the respiratory rate and heart rate, including the respiratory rate, the difference data of the respiratory rate, the heart rate, and the mean, maximum value, standard deviation, skewness, kurtosis, and impulse factor of the difference data of the heart rate. Here, the impulse factor is obtained by subtracting the mean from the maximum value. In this way, the feature calculation unit 13 performs statistical processing on the activity data and calculates multiple features.

[0054] More specifically, the feature calculation unit 13 calculates, for example, respiratory rate-related and heart rate-related features of the subject 50 on an hourly basis.

[0055] For example, the feature calculation unit 13 obtains sensor data indicating the respiratory rate of the subject 50 during the time period including the target date and time for health condition detection, from the activity data recorded in the information recording unit 12 or sensor data obtained from the sensing unit 20, and calculates hourly statistical features for that time period.

[0056] More specifically, the feature calculation unit 13 obtains non-zero respiratory rate data from the activity data, for example, for a given hour, and calculates statistical features such as the mean, maximum, minimum, standard deviation, skewness, kurtosis, and impulse factor from the obtained respiratory rate data for that hour. Here, the impulse factor can be calculated from the difference between the maximum and mean values ​​(maximum value - mean value) of the respiratory rate data for that hour. The feature calculation unit 13 also calculates statistical features such as the mean, maximum, minimum, standard deviation, skewness, kurtosis, and impulse factor from the difference data of the obtained respiratory rate data for that hour. The difference data of the obtained respiratory rate data is, for example, the difference between the respiratory rate at time t and the respiratory rate at time t+1, one second after time t, i.e., data showing the difference in respiratory rate data every second. Note that the feature calculation unit 13 only needs to calculate the mean and maximum values ​​for that hour as statistical features from the obtained respiratory rate data.

[0057] Furthermore, for example, the feature calculation unit 13 obtains heart rate data indicating the heart rate of the subject 50 during the time period including the target date and time for health condition detection from the activity data recorded in the information recording unit 12 or sensor data acquired from the sensing unit 20, and calculates hourly statistical features for that time period.

[0058] Here, the feature calculation unit 13 obtains heart rate data from the activity data, for example, where the heart rate is not zero during a given hour, and calculates the mean, maximum, minimum, standard deviation, skewness, kurtosis, and impulse factor as statistical features from the obtained heart rate data for that hour. The feature calculation unit 13 also calculates the mean, maximum, minimum, standard deviation, skewness, kurtosis, and impulse factor as statistical features from the difference data of the obtained heart rate data for that hour. The difference data of the obtained heart rate data is, similar to the difference data of respiratory rate data, for example, the difference between the respiratory rate at time t and the heart rate at time t+1, which is 1 second after time t, i.e., data showing the difference of the heart rate data every second. Note that the feature calculation unit 13 only needs to calculate the mean and maximum values ​​for that hour as statistical features from the obtained heart rate data.

[0059] The feature calculation unit 13 may also calculate the amount of food consumed and the bed absence rate of the subjects 50 as one of several features.

[0060] In other words, for example, the feature calculation unit 13 may calculate the amount of food consumed by the subject 50 as one of several features from the recorded data 25 included in the activity data. In this case, the feature calculation unit 13 calculates the total amount of food consumed over the past day from the recorded data 25, and then calculates the sum of the amount of food consumed during the time period including the target date and time for health condition detection. Here, if the target date and time is the time period of morning, noon, or night, the feature calculation unit 13 may calculate, for example, the sum of the amount of food consumed between the morning of the day before the target date for health condition detection and the morning of the current day, between noon of the day before and noon of the current day, and between the evening of the day before and the evening of the current day.

[0061] Alternatively, for example, the feature calculation unit 13 may calculate the bed occupancy rate as one of several features from the activity data acquired by the transmitting / receiving unit 11 and recorded in the information recording unit 12. In this case, the feature calculation unit 13 can acquire occupancy data indicating the presence or absence of beds for the subject 50 during the time period including the target date and time for health condition detection from the activity data recorded in the information recording unit 12 or sensor data acquired from the sensing unit 20, and calculate the bed occupancy rate in one-hour units for that time period. More specifically, the feature calculation unit 13 can calculate the bed occupancy rate for a given hour by, for example, counting the number of values ​​of 1 indicating out of bed during a given hour and dividing by the total number during that hour (i.e., the sum of the number of values ​​of 1 indicating out of bed and the number of values ​​of 0 indicating in bed during that hour).

[0062] [1.2.4 Model Creation Section 14] The model creation unit 14 creates a model that learns normality or abnormality in a set of activity data consisting of multiple features. More specifically, the model creation unit 14 creates a model that learns normality or abnormality in a set of activity data consisting of multiple features by performing unsupervised learning using the set of activity data.

[0063] In this embodiment, the model creation unit 14 includes, for example, a computer including memory and a processor (microprocessor), and various functions are realized by the processor executing a control program stored in memory. The model creation unit 14 acquires activity data for the learning period from activity data recorded in the information recording unit 12 or sensor data acquired from the sensing unit 20. The model creation unit 14 may also acquire recorded data 25 for the learning period and include it in the activity data for the learning period.

[0064] Furthermore, the model creation unit 14 causes the feature calculation unit 13 to calculate hourly features based on the activity data for the learning period. The model creation unit 14 uses the hourly features for the learning period to perform unsupervised training on a model, thereby creating a model that learns normality or abnormality in the activity data set. Here, the trained model is a model that separates outliers based on a decision tree, for example, an Isolation Forest model. Alternatively, the model creation unit 14 may create a model that learns normality or abnormality in the activity data set by performing unsupervised training on a k-means model using the hourly features for the learning period.

[0065] The following describes in detail how to create a model such as an Isolation Forest model according to this embodiment.

[0066] Figure 3 shows an example of the distribution of abnormal and normal data groups in the activity data group according to this embodiment. In Figure 3, the vertical axis shows the average heart rate, and the horizontal axis shows the average respiratory rate. The activity data group consists of multiple sensor data for heart rate and respiratory rate, and contains a mixture of abnormal and normal data.

[0067] As shown in Figure 3, the number of normal data points is greater than the number of abnormal data points, and the distribution of normal data points is somewhat concentrated into a single distribution; that is, the distribution of normal data points does not diverge but is concentrated. On the other hand, since the abnormal data points are mostly contained in the region indicated as the abnormal data group in Figure 3, it can be seen that the number of abnormal data points is less than the number of normal data points, and that the abnormal data points are located in a different distribution location from where the normal data points are concentrated.

[0068] Figure 4 is a conceptual diagram illustrating the model according to this embodiment. The model shown in Figure 4 is an Isolation Forest model.

[0069] The model creation unit 14 utilizes the premise that abnormal data occurs less frequently and has a different distribution than normal data, and creates a model that divides the activity data set using hourly features during the learning period, i.e., the activity data set during the learning period. More specifically, the model creation unit 14 creates a model that divides the activity data set by repeatedly selecting features and thresholds randomly to divide the data, creating multiple decision trees, and separating outliers from other values ​​when creating the decision trees. In this way, the model creation unit 14 can create a model that divides abnormal data, such as the abnormal data set shown in Figure 3, at an early stage in the decision tree because it is far removed from the distribution of normal data. As a result, the model created by the model creation unit 14 can calculate an anomaly score, which indicates at what stage of the decision tree the data was divided (distance from the root node). The earlier the division occurs in the decision tree (i.e., the smaller the distance from the root node at the divided node), the higher the anomaly score calculated by this model. Since this model creates multiple decision trees, the final anomaly score output will be the average of the anomaly scores obtained from the partitioning depth calculated from how the partitions are performed in the multiple decision trees.

[0070] For example, as shown in Figure 4(a), the model created by the model creation unit 14 separates (splits) outlier data at an early stage, as shown by the node x in Figure 4, because the outlier data deviates significantly from the values ​​of the normal data in the normal data group. Therefore, the model calculates a high anomaly score for outlier data.

[0071] On the other hand, as shown in Figure 4(b), for example, the model created by the model creation unit 14, if it is normal data, is in the normal data group and therefore cannot be easily separated (divided) as shown by the node y in Figure 4. For this reason, the model will calculate a low anomaly score if it is normal data.

[0072] [1.2.5 Model Update Section 15] The model update unit 15 periodically updates the model created by the model creation unit 14 using activity data acquired by the transmission / reception unit 11 after the model was created.

[0073] The frequency at which the model update unit 15 updates the model can be, for example, every two weeks to one month. The model update unit 15 may update the model frequently, such as every two weeks, for a certain period after it has been created by the model creation unit 14, and then update it monthly thereafter.

[0074] In this embodiment, the model update unit 15 includes, for example, a computer including memory and a processor (microprocessor), and the model update function is realized by the processor executing a control program stored in memory. The model update unit 15 updates the model by updating the structure or conditions of multiple decision trees using activity data acquired by the transmission / reception unit 11 after the model is created. This makes it possible to repeatedly perform model updates using activity data that adds activity data accumulated after the model is created to the activity data used when the model was created.

[0075] [1.2.6 Health condition detection unit 16] The health condition detection unit 16 is implemented as a computer equipped with, for example, a processor (microprocessor), memory, and a communication interface. The processor executes a control program stored in memory to perform various functions. The health condition detection unit 16 uses the model created by the model creation unit 14 and multiple features calculated by the feature calculation unit 13 to detect health abnormalities in the subject 50.

[0076] As shown in Figure 2, the physical condition detection unit 16 includes an abnormality score calculation unit 161, a calculation result recording unit 162, a graded score calculation unit 163, and a factor analysis unit 164.

[0077] [1.2.6.1 Anomaly Score Calculation Unit 161] The abnormality score calculation unit 161 inputs multiple features calculated by the feature calculation unit 13 into a model that has learned the normality or abnormality of the activity data set, thereby obtaining an abnormality score that indicates the degree of physical abnormality per predetermined period.

[0078] In this embodiment, the anomaly score calculation unit 161 inputs multiple hourly features calculated by the feature calculation unit 13 for the target day of health detection for the subject 50 into the model created by the model creation unit 14. The anomaly score calculation unit 161 calculates the division depth from the division of multiple decision trees that constitute the model shown in Figure 4, for example, and calculates the anomaly score by averaging the values ​​of the multiple division depths. The anomaly score calculation unit 161 records the calculated hourly anomaly score for the target day of health detection for the subject 50 in the calculation result recording unit 162.

[0079] [1.2.6.2 Calculation Result Recording Unit 162] The calculation result recording unit 162 is a recording medium capable of recording calculation results, and is composed of a rewritable non-volatile memory such as a hard disk drive or a solid-state drive. In this embodiment, the calculation result recording unit 162 records the abnormal score calculated by the abnormal score calculation unit 161, the stepped score calculated by the stepped score calculation unit 163, and the like as calculation results. The calculation result recording unit 162 may also record the factors analyzed by the factor analysis unit 164 as calculation results.

[0080] [1.2.6.3 Graded Score Calculation Unit 163] The graded score calculation unit 163 calculates graded scores to indicate the degree of physical abnormality of the subjects 50 in stages, based on the abnormal score calculated by the abnormal score calculation unit 161.

[0081] In this embodiment, the graded score calculation unit 163 calculates the average daily abnormal score from the hourly abnormal scores on the target day for health detection, which are recorded in the calculation result recording unit 162 or calculated by the abnormal score calculation unit 161. Similarly, the graded score calculation unit 163 calculates the average daily abnormal score for the day before and two days before the target day for health detection, from the hourly abnormal scores for the day before and two days before the target day for health detection, which are recorded in the calculation result recording unit 162. The graded score calculation unit 163 sums the average daily abnormal scores for the target day, the day before, and two days before to calculate a 3-day total score. Note that the 3-day total score is just one example of a method for accurately calculating the graded score, and is not limited to this. It is sufficient to calculate it within the range of a 1-day total score to a 5-day total score.

[0082] The grading score calculation unit 163 calculates a threshold for the grading score (sometimes referred to as the grading threshold) from a group of 3-day total scores for the past 90 days or so recorded in the calculation result recording unit 162 for the target day. More specifically, the grading score calculation unit 163 calculates the grading threshold by calculating the mean and standard deviation of the group of 3-day total scores for the past 90 days or so.

[0083] Figure 5 shows an example of a five-stage scoring system and its conditions according to the embodiment.

[0084] When the graded score calculation unit 163 calculates a graded score of 5 levels as shown in Figure 5, it can calculate a threshold from the mean and standard deviation, as shown in Figure 5. For example, the threshold for a graded score of 1 is, from the conditions shown in Figure 5, above the mean, and the threshold for a graded score of 2 is the mean minus half the standard deviation minus the mean.

[0085] The graded score calculation unit 163 then calculates the graded score by applying the threshold calculated in this way to the total score for the three days of the target day. More specifically, the graded score calculation unit 163 calculates the value of the graded score by making a determination on the total score for the three days of the target day using the threshold calculated from the conditions shown in Figure 5.

[0086] The graded score calculation unit 163 outputs the calculated graded score value to the calculation result recording unit 162. Furthermore, if the calculated graded score value is between 1 and 3, the graded score calculation unit 163 may output the calculated graded score to the display terminal unit 30 via the communication network 40.

[0087] Note that while Figure 5 shows an example where the graded score calculation unit 163 calculates a graded score of 5 levels, it is not limited to this. It may also calculate a graded score of 2 to 4 levels.

[0088] Here, using Figure 5, we will explain an example of a method for calculating thresholds for grading scores.

[0089] [1.2.6.4 Factor Analysis Section 164] The factor analysis unit 164 performs a factor analysis to determine whether each element included in the activity data is a factor, if the grading score is above a predetermined value. Here, the elements are the amount of food consumed, respiratory rate, heart rate, or bed absence rate of the subjects 50 during a predetermined period. The predetermined value is a value that requires action to be taken as a precursor to health abnormalities. For example, if the grading score is on a 5-point scale, the value can be set to 4 or 5; if it is on a 3-point scale, it can be set to 3; if it is on a 2-point scale, it can be set to 2, and so on.

[0090] In this embodiment, if the graded score calculated by the graded score calculation unit 163 is 4 or 5, the factor analysis unit 164 performs factor analysis on each element of heart rate, respiratory rate, bed absence rate, and food intake included in the activity data used to calculate the features. If the activity data used to calculate the features includes only heart rate and respiratory rate, then factor analysis should be performed on the heart rate and respiratory rate elements.

[0091] The factor analysis unit 164 converts, for example, the multiple features of each element over the entire period used to calculate the grading score into data of multiple features for each element on a daily basis, and calculates the mean and standard deviation over the entire period used to calculate the grading score. In this embodiment, the factor analysis unit 164 converts the multiple features of each element for three days into data of multiple features for each element on a daily basis, and calculates the mean and standard deviation of each element over three days.

[0092] The factor analysis unit 164 then analyzes that an element is not a factor if the following equation 1 is true, and analyzes that an element is a factor if the following equation 1 is not true.

[0093] (Mean - 2 * Standard Deviation) ≤ (Feature Value of that element at the target date and time) ≤ (Mean + 2 * Standard Deviation) ... (Equation 1)

[0094] Note that (Equation 1) utilizes the property of standard deviation, which states that 95.45% of all data are distributed within the range of mean ± twice the standard deviation.

[0095] The factor analysis unit 164 outputs the elements identified as factors and their graded scores to the calculation result recording unit 162. Alternatively, the factor analysis unit 164 may output the elements identified as factors and their graded scores to the display terminal unit 30 via the communication network 40.

[0096] [1.3 Display terminal unit 30] The display terminal unit 30 is implemented as a computer equipped with a processor (microprocessor), memory, communication interface, user interface, etc. The display terminal unit 30 is the terminal of a user 61, such as a monitor of the subject 50, and may be, for example, a tablet or a smartphone. The display terminal unit 30 may also be a mobile PC or a stationary PC connected to a display.

[0097] In this embodiment, the display terminal unit 30 can be viewed by a user 61, such as a monitor of the subject 50. The display terminal unit 30 is connected to a communication network 40, and when it obtains a graded score, etc., from the information management server 10, it displays information on the user interface to allow the user 61 to take action regarding any abnormalities in the subject 50's physical condition. The user interface can display information on the display in response to input from the user 61, etc.

[0098] Figures 6 to 8 show examples of displays used to address abnormal physical conditions in the 50 subjects according to this embodiment.

[0099] Figure 6 shows an example of a graded score display screen 301 according to the embodiment. More specifically, Figure 6 shows an example of an application screen viewed by user 60 or user 61, who is a field staff member, and displays a graded score display screen 301 that displays the graded score of the subject 50 in order to detect and respond to health abnormalities of the subject 50 at an early stage. This graded score display screen 301 is displayed by the display terminal unit 30 when it is selected by touching the menu screen when the application is launched on the display terminal unit 30. In Figure 6, one subject 50 is residing in each of the room numbers 201 to 211, and the graded score of each subject 50 is shown. Also, Figure 6 shows which of the following elements—heart rate, respiratory rate, food intake, and bed occupancy rate—is the cause when the graded score is 4 or 5, and the element that is the cause is hatched. The area indicated by a in Figure 6 shows an input field for input by user 60, who is a field staff member such as a nurse. This input field is used to enter information such as the validity of the graded score values.

[0100] Figure 7 shows an example of a vital sign fluctuation graph display screen 302 according to the embodiment. More specifically, Figure 7 shows another example of an application screen viewed by a field staff member, user 60 or user 61, and shows a vital sign fluctuation graph display screen 302 that displays a vital sign fluctuation graph of a specific subject 50 in order to detect and respond to health abnormalities in that subject 50 at an early stage. This vital sign fluctuation graph display screen 302 is also displayed by the display terminal unit 30 when it is selected by touching the menu screen when the application is launched on the display terminal unit 30. In Figure 7, the range of vital information when the grading score in the vital sign fluctuation graph of a specific subject 50 is, for example, 5 is shown. This makes it possible to investigate the cause at an early stage when the grading score of a specific subject 50 is, for example, 5.

[0101] Figure 8 shows an example of a risk group display screen 303 according to an embodiment. More specifically, Figure 8 shows another example of an application screen viewed by user 60 or user 61, who are on-site staff, and shows a risk group display screen 303 that displays the risk group for the entire facility in which the subject 50 resides, in order to detect and respond to health abnormalities of the subject 50 at an early stage. This risk group display screen 303 is also displayed by the display terminal unit 30 when it is selected by touching the menu screen when the application is launched on the display terminal unit 30. In Figure 8, the risk distribution, which is the proportion of risk groups, is shown in a pie chart for the entire facility, which consists of multiple rooms such as room 201. In the example shown in Figure 8, it is shown that 20% are high risk with a grading score of 5, 10% are medium risk with a grading score of 4, 30% are low risk with a grading score of 3, and 40% are standard (no risk) with a grading score of 1 or 2. In this way, the risk distribution across the entire facility becomes visible and can be viewed from a broad perspective. For example, experienced and highly specialized nurses can be assigned to rooms with a medium risk, and even more experienced and specialized nurses can be assigned to rooms with a high risk. Similarly, experienced caregivers can be assigned to rooms with a low risk, and caregivers can be assigned to rooms with a standard risk. In short, because the risk distribution across the entire facility becomes visible and can be viewed from a broad perspective, the limited number of on-site staff can be assigned to the right positions throughout the facility. This allows for a quicker and more appropriate response to any health abnormalities or signs of health abnormalities in the 50 target individuals.

[0102] [2. Operation of Information Management Server 10] Next, we will explain the operation of the information management server 10 configured as described above.

[0103] Figure 9 is a flowchart illustrating the overview of the operation of the health condition detection device according to this embodiment. The health condition detection device according to this embodiment is, for example, an information management server 10, but it is sufficient if it includes at least the following components of the information management server 10 described above: a transmitting / receiving unit 11, a feature quantity calculation unit 13, an abnormal score calculation unit 161, and a stepped score calculation unit 163.

[0104] First, in the physical condition detection device according to this embodiment, the transmitting / receiving unit 11 acquires activity data including the respiratory rate and heart rate of the subject 50 over a predetermined period (S11). Next, the feature calculation unit 13 calculates a plurality of features based on the activity data acquired in step S11 (S12). Next, the abnormality score calculation unit 161 acquires an abnormality score per predetermined period by inputting the plurality of features calculated in step S12 into a model that has been previously trained on the normality or abnormality of the activity data group (S13). Next, the gradation score calculation unit 163 calculates a gradation score to indicate the degree of physical abnormality of the subject 50 in stages based on the abnormality score acquired in step S13 (S14). Finally, the gradation score calculation unit 163 outputs the gradation score calculated in step S14 (S15).

[0105] Next, we will describe an example of the operation of the information management server 10 according to this embodiment, as an example of the operation of the health condition detection device explained in Figure 9.

[0106] Figure 10 is a flowchart showing an example of the operation of the information management server 10 according to this embodiment.

[0107] In the information management server 10, first, the transmitting / receiving unit 11 acquires sensor data and recorded data 25 (S101). In this embodiment, the transmitting / receiving unit 11 acquires activity data that includes at least the respiratory rate and heart rate of the subject 50 over a predetermined period.

[0108] Next, the feature calculation unit 13 calculates multiple hourly features from the sensor data and recorded data 25 acquired in step S101 (S102). In this embodiment, the feature calculation unit 13 calculates multiple hourly features for the target day of physical condition detection of the subject 50 based on activity data that includes at least the respiratory rate and heart rate of the subject 50 acquired by the transmitting / receiving unit 11.

[0109] Next, the health condition detection unit 16 calculates an abnormality score per hour from multiple features calculated per hour in step S102 using a pre-trained model (S103). In this embodiment, the abnormality score calculation unit 161 inputs multiple features calculated by the feature calculation unit 13 into the model created by the model creation unit 14 to obtain an abnormality score indicating the degree of health abnormality per hour during a predetermined period including the target day.

[0110] Next, the health condition detection unit 16 calculates the average daily abnormal score from the abnormal score per hour calculated in step S103 (S104). In this embodiment, the graded score calculation unit 163 calculates the average daily abnormal score from the abnormal score per hour during a predetermined period including the target day for health condition detection of the subject 50.

[0111] Next, the health condition detection unit 16 totals the data for the target day, the day before, and the day before that for the target person 50, and calculates a 3-day total score (S105). In this embodiment, the graded score calculation unit 163 calculates the average value of the daily abnormal scores for the day before and the day before that of the target day, and calculates a 3-day total score by summing the averages of the daily abnormal scores for the target day, the day before, and the day before that.

[0112] Next, the physical condition detection unit 16 calculates a graded score (S106) by applying a graded threshold calculated from a group of 3-day total scores for the past 90 days to the 3-day total score for the target day of physical condition detection calculated in step S105. In this embodiment, the graded score calculation unit 163 calculates the graded threshold by calculating the mean and standard deviation of the group of 3-day total scores for the past 90 days for the target day. The graded score calculation unit 163 then calculates the graded score by applying the calculated graded threshold to the 3-day total score for the target day. Here, the graded score represents one of five values ​​from 1 to 5.

[0113] Next, the health condition detection unit 16 checks whether a graded score of 4 or 5 was calculated in step S106, that is, a value indicating an abnormality (S107).

[0114] In step S107, if the graded score is 4 or 5 (Yes in S107), the physical condition detection unit 16 performs factor analysis on each element of food intake, respiratory rate, heart rate, or bed absence rate (S108). In this embodiment, if the calculated graded score is 4 or 5, the factor analysis unit 164 performs factor analysis on each element of heart rate, respiratory rate, bed absence rate, and food intake included in the activity data used to calculate the features.

[0115] Next, the physical condition detection unit 16 outputs the elements that have been analyzed as factors by the factor analysis, along with a graded score (S109). In this embodiment, the factor analysis unit 164 outputs the elements that have been analyzed as factors by the factor analysis, along with a graded score.

[0116] Furthermore, if the graded score in step S107 is not a value of 4 or 5 (No in S107), the physical condition detection unit 16 outputs the graded score in step S109.

[0117] [3 Effects, etc.] As described above, the physical condition detection device according to this embodiment can detect signs of health abnormalities in the subject 50, that is, small changes in the subject 50's physical condition that could lead to health abnormalities. More specifically, the physical condition detection device according to this embodiment calculates multiple features from activity data and inputs the calculated multiple features into a model that can detect signs of health abnormalities by learning the normality or abnormality of the activity data set. The physical condition detection device according to this embodiment calculates a graded score that evaluates the degree of the subject 50's physical abnormality in stages based on the abnormality score obtained by inputting the calculated multiple features into the model.

[0118] This allows for the detection of early signs of health abnormalities in the 50 subjects based on their graded scores. Therefore, it is possible to detect early signs of health abnormalities in the 50 subjects that may go unnoticed even by healthcare professionals caring for or nursing them.

[0119] Furthermore, the graded score makes it easier for on-site staff, such as healthcare professionals, who provide care or nursing to the 50 subjects to understand the degree of abnormality, thus making it easier to respond to various small changes in the subjects' physical condition that could lead to health problems in their daily lives, i.e., early signs of health problems.

[0120] Furthermore, if the physical condition detection device according to this embodiment exceeds a predetermined value that requires action to be taken regarding signs of a health abnormality, factor analysis may be performed to analyze whether each element included in the activity data is a contributing factor. If the graded score exceeds a predetermined value that requires action to be taken regarding signs of a health abnormality, the device will also notify the user whether the cause of the small change in physical condition that leads to a health abnormality is, for example, heart rate or respiratory rate.

[0121] This allows healthcare professionals caring for or nursing the 50 individuals to use the notified factors as a clue to respond early and appropriately to signs of health abnormalities.

[0122] Here, the activity data refers to daily activity data obtained on-site, including at least respiratory rate and heart rate, from among the food intake, respiratory rate, heart rate, and bed absence rate of 50 subjects over a predetermined period. By calculating multiple features from the daily activity data obtained on-site, it becomes possible to detect early signs of health abnormalities with greater accuracy.

[0123] Furthermore, the multiple features according to this embodiment include the respiratory rate of the subject 50, the difference data of respiratory rates, the heart rate of the subject 50, and the mean, maximum, standard deviation, skewness, kurtosis, and impulse factor obtained by subtracting the mean from the maximum value in the difference data of the heart rate. In this embodiment, at least the mean and maximum values ​​of the respiratory rate and heart rate of the subject 50 are calculated as multiple features. In this way, the physical condition detection device, etc. according to this embodiment performs statistical processing on activity data and calculates multiple features. As a result, a more accurate anomaly score can be obtained from the multiple features calculated using a trained model.

[0124] Furthermore, the model trained in this embodiment (also referred to as the pre-trained model) learns normality or abnormality in the activity data set through unsupervised learning using the activity data set. In other words, in this embodiment, a pre-trained model capable of detecting signs of health abnormalities is created by performing unsupervised learning on the model using activity data obtained daily in the field. Therefore, by performing unsupervised learning using the activity data of 50 subjects, a pre-trained model capable of detecting signs of health abnormalities can be created without burdening the on-site staff, such as medical professionals, of the 50 subjects.

[0125] Here, we will explain the techniques used when creating the model according to this embodiment.

[0126] Figures 11A and 11B are diagrams that conceptually explain anomaly detection when the model according to the comparative example is operated over a long period of time. Figure 12 is a diagram that conceptually explains anomaly detection when the model according to this embodiment is operated over a long period of time. In Figures 11A to 12, the horizontal axis represents time, and the vertical axis represents activity data such as heart rate. Figures 11A to 12 conceptually show the period of activity data used to train the model and the intervals in which health abnormalities for the 50 subjects are to be detected.

[0127] The model in the comparative example is, for example, the model described in Patent Document 1, and is a single model generated by learning using activity data from a short learning period.

[0128] Therefore, as shown in Figure 11A, for example, if the period between the learning period and the interval to be detected as a health abnormality is relatively short, the model related to the comparative example can be used to detect health abnormalities. On the other hand, as shown in Figure 11B, for example, if the period between the learning period and the interval to be detected as a health abnormality is relatively long, the model related to the comparative example will be affected by medium- to long-term fluctuations in the activity data of the 50 subjects, and will misdetect health abnormalities. This is because if the 50 subjects have a disease, their activity data will fluctuate in the medium to long term due to seasonal and environmental influences. In other words, the model related to the comparative example is created using activity data from a relatively short initial learning period, so it is highly likely that it will not be able to cope with medium- to long-term fluctuations in the activity data of the 50 subjects when used in long-term operation. To put it another way, the model related to the comparative example has a small amount of activity data used during learning, and is highly likely to fail to detect small changes in physical condition that lead to health abnormalities, which may appear in multiple or complex patterns.

[0129] In contrast, in the model according to this embodiment, the model is repeatedly updated using activity data accumulated, for example, in units of two weeks to one month. That is, the model according to this embodiment is repeatedly updated using activity data which is the activity data used when the model was created plus the activity data accumulated after the model was created.

[0130] Therefore, as shown in Figure 12, for example, the model according to this embodiment is trained or updated with activity data up to the immediate vicinity of the interval in which health abnormalities are to be detected, and can detect health abnormalities in the interval in which health abnormalities are to be detected. In other words, the model according to this embodiment can detect signs of health abnormalities while also responding to medium- to long-term fluctuations in the subject 50 due to disease or medium- to long-term fluctuations due to environmental influences.

[0131] Figure 13 is a conceptual diagram illustrating the performance improvement achieved by updating the model according to this embodiment.

[0132] In Figure 13, the horizontal axis represents the success rate of detecting health abnormalities, and the vertical axis represents the number of days of accumulated activity data. As shown in Figure 13, the success rate of detection improves as the number of days of accumulated activity data used to train and update the model increases, and it can be seen that the success rate saturates after a certain amount of activity data.

[0133] In other words, the model according to this embodiment shows that, during medium- to long-term operation, the detection performance improves as the accumulated activity data increases.

[0134] Furthermore, the model relating to the comparative example, i.e., the model relating to Patent Document 1, is created using activity data from the period when the subject 50 is in a normal physical condition as training data. Therefore, it is necessary for on-site staff to refer to caregiving record data 25 to determine whether the activity data for the relevant period represents only normal conditions or includes abnormal conditions, which places a burden on the on-site staff.

[0135] In contrast, the model according to this embodiment is created by performing unsupervised learning using activity data, taking advantage of the fact that in activity data where abnormal and normal data are mixed, abnormal data occurs less frequently and is located in a different distribution position than normal data. This makes it possible to create a model using the activity data of 50 subjects without burdening the medical professionals or other on-site personnel of the 50 subjects. Furthermore, by utilizing this property, the model according to this embodiment can be created to determine how far the activity data targeted for health condition detection is from the area where the distribution of normal data is concentrated. For example, models that separate outliers based on decision trees, such as the Isolation Forest model, can be created using unsupervised learning by utilizing this property, and can therefore be used as the model according to this embodiment.

[0136] (Example 1) In the above embodiment, the display terminal 30 was described as displaying information such as that shown in Figures 6 to 8, i.e., information to respond to abnormal physical conditions of the subject 50, but it is not limited to this.

[0137] The display terminal 30 may perform a linked display that combines the graded score calculated by the physical condition detection unit 16 with the recorded data 25 that records the details of nursing or care provided to the subject 50.

[0138] Figure 14 shows an example of a linked display shown by the display terminal unit 30 according to Embodiment 1.

[0139] Figure 14 shows an example of a linked display that shows care records detailing the care provided to 50 subjects, along with a graded score. The linked display screen shown in Figure 14 is a screen that can be viewed by on-site staff providing care services.

[0140] In the collaborative display in Figure 14, checkmarks are placed in the care records indicating observations made by on-site staff and judgments made by doctors and nurses. In addition, in the collaborative display in Figure 14, a dotted line is superimposed on the care records indicating a period where the grading score is 5, meaning that the subject's health abnormality requires attention at a predetermined level of 50. Furthermore, the collaborative display in Figure 14 also displays the date and time of sleep disturbances, the date and time of excretion and urination, and the date and time of meals.

[0141] By looking at the linked display shown in Figure 14, on-site staff can quickly investigate the cause of the early signs of health abnormalities based on the lifestyle of the 50 subjects when the grading score is 5.

[0142] (Example 2) Example 2 describes an incident detection case in which a precursor to a health problem in 50 subjects, i.e., a small change in the subject's physical condition that could lead to a health problem, was detected.

[0143] Figure 15 shows Incident Discovery Case 1 related to Example 2. Figure 15 shows an example in which 50 subjects were hospitalized on February 27 due to suspected pneumonia with a high degree of sudden onset.

[0144] In Figure 15, the calculated graded score is presented every morning at 8:00 AM, and caregivers' observations regarding the health of the 50 subjects are entered daily at 10:00 AM or 3:00 PM. Furthermore, when the graded score is 4 or higher, nurses check the subject's condition after 6:00 PM and enter whether the graded score presentation was accurate or not (OK or NG).

[0145] As shown in Figure 15, on February 21st and 22nd, when the grading score indicated by a was 4, the nurse made an OK judgment, and on February 24th, when the grading score indicated by b was also 4, the caregiver did not notice anything unusual at 10:00. Furthermore, after 11:00 on February 24th, subject 50 developed a fever, which has not subsided since, but on February 25th, the caregiver did not notice anything unusual (i.e., there was no problem). However, on February 27th, subject 50 was hospitalized on suspicion of pneumonia with a high degree of sudden onset.

[0146] In contrast, the calculation results of the graded score show that on February 21st and 22nd, when the graded score shown in a was 4, and on February 24th, when the graded score shown in b was also 4, signs of health abnormalities in the subjects, i.e., small changes in their physical condition that could lead to health abnormalities, were detected.

[0147] Figure 16 shows Incident Discovery Case 2 related to Example 2. Figure 16 shows an example where another subject 50 was hospitalized on March 12 due to fever. In Figure 16, the nurse's judgment and the caregiver's observations are the same as in Figure 15, so the explanation is omitted.

[0148] As shown in Figure 16, on March 7th, when the grading score indicated by a is 4, the caregiver noticed nothing (i.e., no problem). Also, on March 8th, when the grading score indicated by a is 5, the caregiver noticed nothing at 10am, but at 3pm noticed that the patient seemed lethargic, was moving poorly, and had cold-like pain. Furthermore, on March 8th, when the grading score indicated by a is 5, the nurse made an OK judgment. Subsequently, the caregiver noticed something, and on March 12th, patient 50 was hospitalized because they seemed lethargic, moved poorly, and had cold-like symptoms.

[0149] In contrast, the calculation results of the graded score show that on March 7th, the graded score was 4 as shown in a, and on March 8th, the graded score was 5 as shown in b, and remained at 5 thereafter. In other words, on March 7th, when the graded score was 4, it can be seen that the system detected signs of health abnormalities in the 50 subjects, that is, small changes in their physical condition that could lead to health abnormalities.

[0150] Figure 17 shows Incident Discovery Case 3 related to Example 2. Figure 17 shows another case in which subject 50 choked on mochi (rice cake) at 10:00 on February 20th and was hospitalized late at night on February 24th with suspected aspiration pneumonia. The nurse's judgment and the caregiver's awareness in Figure 17 are the same as in Figure 15, so the explanation is omitted.

[0151] As shown in Figure 17, on February 21st, the day after the incident on February 20th when the subject choked on mochi, the grading score indicated by a was 5, and remained at 5 until the day of hospitalization. Meanwhile, the caregiver recorded that between February 21st and February 23rd, subject 50 had noticed things like choking and appearing lethargic, but at times judged as no observation (no problem). Furthermore, on February 23rd, despite subject 50's condition not having returned to normal, the nurse made a negative judgment regarding the grading score of 5, judging that subject 50 was fine. In other words, the nurse missed the early signs of a health problem in subject 50 on February 23rd.

[0152] In contrast, the calculation results of the graded score show that the graded score was 5 from February 21st, as indicated by a, and remained at 5 thereafter. In other words, it can be seen that the early signs of a health problem in the subject, that is, small changes in the subject's physical condition that could lead to a health problem, were detected from three days before February 24th, when the graded score was 5 and subject 50 was hospitalized.

[0153] (Possibility of other embodiments) The above describes the information management server 10 and other components related to the embodiments and examples, i.e., the health condition detection method and health condition detection device related to the embodiments and examples. However, this disclosure is not limited to these embodiments and examples.

[0154] For example, each processing unit included in the information management server 10 according to the above embodiments and examples is typically implemented as an LSI (Large-Scale Integrated Circuit). These may be individually integrated into a single chip, or some or all of them may be integrated into a single chip.

[0155] Furthermore, integrated circuit implementation is not limited to LSIs; it can also be achieved with dedicated circuits or general-purpose processors. Field Programmable Gate Arrays (FPGAs), which can be programmed after LSI manufacturing, or reconfigurable processors, which allow for the reconfiguration of the connections and settings of circuit cells within the LSI, may also be used.

[0156] Furthermore, this disclosure may be implemented as a health condition detection method performed by an information management server 10 or the like, i.e., a health condition detection device.

[0157] Furthermore, in each of the above embodiments, each component may be implemented by being composed of dedicated hardware or by executing a software program suitable for each component. Each component may also be implemented 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.

[0158] Furthermore, the division of functional blocks in the block diagram is just one example; multiple functional blocks can be implemented as a single functional block, a single functional block can be divided into multiple parts, or some functions can be moved to other functional blocks. In addition, the functions of multiple functional blocks with similar functions can be processed in parallel or time-sharing by a single piece of hardware or software.

[0159] Furthermore, the order in which each step in the flowchart is performed is illustrative for the purpose of specifically illustrating this disclosure, and may be in a different order. Also, some of the above steps may be performed simultaneously (in parallel) with other steps.

[0160] The above describes a health condition detection device according to one or more embodiments, based on embodiments and examples, but this disclosure is not limited to these embodiments and examples. Without departing from the spirit of this disclosure, various modifications that a person skilled in the art can conceive of may be applied to these embodiments, examples, and modifications, and forms constructed by combining components from different embodiments, examples, and modifications may also be included within the scope of one or more embodiments. [Industrial applicability]

[0161] This disclosure can be used in a health condition detection method, health condition detection device, and program, and can be used in a health condition detection method, health condition detection device, and program that can detect small changes in a subject's health that could lead to a health abnormality, for example, as a precursor to a health abnormality in the subject. [Explanation of Symbols]

[0162] 10. Information Management Server 11 Transmitter / Receiver 12 Information Recording Section 13 Feature Calculation Unit 14. Model Creation Section 15 Model Update Section 16. Health Detection Unit 20 Sensing Unit 25 Recorded Data 30 Display terminal section 40 Communication Networks 50 Target Persons 60, 61 users 100 Health Detection System 161 Anomaly Score Calculation Unit 162 Calculation Result Recording Unit 163. Level-based score calculation unit 164 Factor Analysis Department 301 Level-based score display screen 302 Vital Signs Fluctuation Graph Display Screen 303 Risk Group Display Screen

Claims

1. A computer-based method for detecting physical condition, Activity data, including respiratory rate and heart rate, of the subject during a predetermined period is acquired. Based on the acquired activity data, several features are calculated, By inputting the calculated multiple features into a model trained to partition the distribution of activity data consisting of multiple features, an anomaly score indicating the degree of physical abnormality per predetermined period is obtained as an outlier from the distribution. Based on the acquired abnormality score, a graded score is calculated to indicate the degree of the subject's physical abnormality in stages. The calculated graded score will be output. Methods for detecting physical condition.

2. When calculating the aforementioned graded score, If the aforementioned grading score is above a predetermined value, a factor analysis is performed to analyze whether or not each element included in the activity data is a contributing factor. When outputting the aforementioned graded score, The system outputs the elements identified as factors in the factor analysis and the graded score. The method for detecting physical condition according to claim 1.

3. The activity data includes, among the subject's food intake, respiratory rate, heart rate, and bed absence rate during the predetermined period, at least the respiratory rate and heart rate. A method for detecting physical condition according to claim 1 or 2.

4. When calculating the aforementioned multiple feature quantities, The following are calculated as the plurality of features: the respiratory rate, the difference data of the respiratory rate, the heart rate, and the mean, maximum, standard deviation, skewness, kurtosis, and impulse factor obtained by subtracting the mean from the maximum value, of which at least the mean and maximum values ​​for the respiratory rate and the heart rate are used. The method for detecting physical condition according to claim 3.

5. The aforementioned model learns whether the activity data set is normal or abnormal through unsupervised learning using the activity data set. A method for detecting physical condition according to claim 1 or 2.

6. The aforementioned model is a model that separates outliers based on a decision tree. The method for detecting physical condition according to claim 5.

7. The aforementioned model is an Isolation Forest model. The method for detecting physical condition according to claim 6.

8. The aforementioned model is periodically updated using the acquired activity data. The method for detecting physical condition according to claim 5.

9. When outputting the aforementioned graded score, The calculated graded score is output to the monitor's terminal for the subject. The user interface of the terminal is configured to display information that allows the monitor to take action regarding any abnormalities in the subject's physical condition. The method for detecting physical condition according to claim 5.

10. A transmitting and receiving unit that acquires activity data including the respiratory rate and heart rate of the subject over a predetermined period, A feature calculation unit calculates multiple features based on the acquired activity data, An abnormality score calculation unit obtains an abnormality score indicating the degree of physical abnormality per predetermined period as an outlier from the distribution by inputting the calculated multiple features into a model trained by the model creation unit to divide the distribution of activity data set consisting of multiple features, The system includes a graded score calculation unit that calculates graded scores to indicate the degree of physical abnormality of the subject in stages based on the acquired abnormal score, The transmitting and receiving unit outputs the calculated graded score. A health condition detection device.

11. Activity data, including respiratory rate and heart rate, of the subject during a predetermined period is acquired. Based on the acquired activity data, several features are calculated, By inputting the calculated multiple features into a model (model creation unit) that has been trained to partition the distribution of activity data consisting of multiple features, an anomaly score indicating the degree of physical abnormality per predetermined period is obtained as an outlier from the distribution. Based on the acquired abnormality score, a graded score is calculated to indicate the degree of the subject's physical abnormality in stages. The computer is instructed to output the calculated graded score. program.

Images