Fall probability prediction device, fall probability prediction system, fall probability prediction method, and fall probability prediction program

Abstract

This fall probability prediction device has a computer function for outputting a numerical value calculated on the basis of an input value. The fall probability prediction device comprises a trained model that has been subjected to machine learning so as to output a fall probability curve, which represents a fall probability as a function of age, using the age of a plurality of individuals and the presence or absence of a fall for each of the plurality of individuals as explanatory variables. Upon input of the age of a subject as an input value, the fall probability prediction device outputs a fall probability curve for predicting the fall probability of the subject at a time later than the time of the input.

Description

Falling probability prediction device, falling probability prediction system, falling probability prediction method, and falling probability prediction program

[0008] ,

[0001] The present invention relates to a technique for predicting the future possibility of a subject falling.

[0002] Patent Document 1 describes a factor estimation system for estimating factors of a fall risk indicating the possibility of falling.

[0003] The system of Patent Document 1 measures the electromyogram value and walking state of a non-measurer. The system of Patent Document 1 predicts the presence or absence of a fall risk at the measurement time based on the measurement results.

[0004] Japanese Patent No. 7122696

[0005] However, the conventional technologies such as Patent Document 1 predict whether a person will fall or not at the measurement time, and cannot predict the future possibility of falling. Also, the conventional technologies such as Patent Document 1 cannot predict the future possibility of falling.

[0006] Therefore, an object of the present invention is to provide a technique for predicting the future possibility of a person to be predicted falling on a continuous time axis. <00​​​The probability of falling is strongly related to the progression of age. For example, among several individuals with similar health conditions, the progression of the probability of falling with respect to age shows a specific relationship. By generating a pre-trained model based on this relationship, it is possible to obtain a continuous curve showing the progression of the probability of falling with respect to age after the input age. Therefore, it becomes possible to evaluate the progression of the probability of falling with respect to future age for the target individual on a continuous time axis.

[0009] According to this invention, the likelihood of a subject falling in the future can be predicted on a continuous time axis.

[0010] Figure 1 is a functional block diagram of a fall probability prediction system according to the first embodiment of the present invention. Figure 2 is a functional block diagram of a learning model according to the first embodiment of the present invention. Figure 3 is a table showing an example of explanatory variables for the learning algorithm according to the first embodiment of the present invention. Figure 4 is a graph showing an example of a fall probability curve according to the first embodiment of the present invention. Figure 5 is a functional block diagram showing a derivative example of the fall probability prediction system according to the first embodiment of the present invention. Figure 6 is a flowchart showing an example of a method for generating a trained model for generating a fall probability curve in the fall probability prediction method according to the first embodiment of the present invention. Figure 7 is a functional block diagram of a fall probability prediction system according to the second embodiment of the present invention. Figure 8 is a table showing an example of explanatory variables for the learning algorithm according to the second embodiment of the present invention. Figure 9 is a functional block diagram of a fall probability prediction system according to the third embodiment of the present invention. Figure 10 is a graph showing an example of a fall probability curve according to the third embodiment of the present invention. Figure 11 is a functional block diagram of a fall probability prediction system according to the fourth embodiment of the present invention. Figure 12 is a functional block diagram showing an example of a muscle activity information acquisition unit according to the fourth embodiment of the present invention. Figure 13 is a graph showing an example of a fall probability curve according to the fourth embodiment of the present invention. Figure 14 is a functional block diagram of a fall probability prediction system according to a fifth embodiment of the present invention.

[0011] [First Embodiment] A fall probability prediction system and a fall probability prediction method according to the first embodiment of the present invention will be described with reference to the figures. Figure 1 is a functional block diagram of the fall probability prediction system according to the first embodiment of the present invention. Figure 2 is a functional block diagram of the learning model according to the first embodiment of the present invention. Figure 3 is a table showing an example of explanatory variables for the learning algorithm according to the first embodiment of the present invention. Figure 4 is a graph showing an example of a fall probability curve according to the first embodiment of the present invention.

[0012] As shown in Figure 1, the fall probability prediction system 10 includes an age acquisition unit 21 and a fall probability curve calculation unit 30. When the fall probability is predicted using a device with computer functions (a computing device: for example, a personal computer), the age acquisition unit 21 and the fall probability curve calculation unit 30 are provided in this device.

[0013] The age acquisition unit 21 acquires the age of the person to be predicted and outputs it to the fall probability curve calculation unit 30. When the age is input, the fall probability curve calculation unit 30 calculates a fall probability curve that predicts the probability of the person to be predicted falling at a time after the input.

[0014] Specifically, the fall probability curve calculation unit 30 includes a pre-trained model that has been trained to output a fall probability curve, which is a function of age and fall probability, using the ages of multiple people and whether or not each person has fallen as explanatory variables. The fall probability curve calculation unit 30 calculates a fall probability curve that predicts the fall probability of the person to be predicted by inputting the age of the person to be predicted into this pre-trained model.

[0015] The trained model is generated as follows:

[0016] In generating the trained model, as shown in Figure 2, the ages and fall information of multiple individuals are input to the learning algorithm 300 as explanatory variables.

[0017] The ages of multiple individuals are entered by associating each individual's identification information (ID). For each ID, if the individual fell before the time of input, the age and fall information are entered as 1; if the individual has not fallen before the time of input, the current age and fall information are entered as 0.

[0018] For example, in the example shown in Figure 3, for person ID 1, it is entered that they fell when they were 65 years old (fall event "1"). Similarly, for people ID 2 and ID 4, it is entered that person ID 2 fell when they were 66 years old, and person ID 4 fell when they were 73 years old (fall event "1"). On the other hand, for people ID 3 and ID 4, since they have not fallen at the time of input, it is entered that they have not fallen (fall event "0"), and their age at the time of input is also entered.

[0019] Whether or not a fall has occurred can be determined by setting a minimum age for assessment and applying the assessment to individuals above this minimum age, or it can be limited to individuals within specific conditions. For example, the assessment could be based on whether or not a fall occurred after the age of 65, or it could be limited to cases occurring within a care facility, specifically focusing on whether or not a fall occurred after admission to the facility.

[0020] These multiple individuals are not limited to those focused on a specific person. Therefore, it is possible to easily collect fall information of various ages necessary for generating a trained model in a short period of time.

[0021] Here, the learning algorithm 300 is generated using a survival time analysis method. In this process, the presence or absence of a fall corresponds to the presence or absence of an event according to the survival time analysis method. That is, a fall is set to an event value of "1", and no fall is set to an event value of "0".

[0022] Then, as the learning algorithm 300, for example, the Cox proportional hazards model is used. In this Cox proportional hazards model, the model is trained using the hazard ratio. That is, by learning the change in the likelihood of a fall by comparing it with the case where all the event values ​​of the explanatory variables are 0 (where there is no fall), a trained model is generated.

[0023] The generation of this trained model allows for the creation of a fall probability curve f1, which is a function of age and fall probability, as shown in Figure 4. This fall probability curve represents how much the probability of falling increases with age, assuming that the probability of falling at a baseline age (e.g., 20 years old), where falls due to muscle mass are considered to be almost nonexistent, is 0%.

[0024] Furthermore, since this fall probability curve f1 includes both people who have experienced falls and those who have not, it is possible to use a large number of explanatory variables for training. Consequently, the fall probability curve f1 has improved prediction accuracy of fall probability compared to a fall probability curve calculated using only people who have experienced falls.

[0025] Furthermore, the fall probability curve f1 does not represent the probability of falling at a single point in the future, but rather the probability of falling is represented by a time function with continuous values. Therefore, the fall probability curve f1 can also represent the trend of the probability of falling in response to changes (increases) in age.

[0026] While the aforementioned fall probability prediction system 10 only performed the processing up to the calculation of the fall probability curve, the fall probability prediction system 10X shown below can calculate the age (period) at which the probability of falling increases significantly.

[0027] Figure 5 is a functional block diagram showing a derivative example of the fall probability prediction system according to the first embodiment of the present invention. The fall probability prediction system 10X shown in Figure 5 differs from the fall probability prediction system 10 shown in Figure 1 above in that it additionally includes a high-probability timing calculation unit 40. The other components of the fall probability prediction system 10X are the same as those of the fall probability prediction system 10, and the explanation of the similar parts will be omitted.

[0028] The fall probability prediction system 10X includes an age acquisition unit 21, a fall probability curve calculation unit 30, and a high-probability period calculation unit 40.

[0029] The high-probability period calculation unit 40 receives numerical input of the relationship between age and the probability of falling, based on the fall probability curve obtained by the fall probability curve calculation unit 30.

[0030] The high-probability period calculation unit 40 calculates the magnitude of the change in the probability of falling in response to a change (increase) in age. In other words, the high-probability period calculation unit 40 calculates the derivative of the fall probability curve (the inflection point of the fall probability curve).

[0031] The high-probability period calculation unit 40 calculates the age at which the differential value is greater than or equal to a threshold from this calculation result. The high-probability period calculation unit 40 outputs the calculated number of years of operation. Alternatively, the high-probability period calculation unit 40 may subtract the current age (the input age of the person to be predicted) from the age at which the probability of falling becomes high and output the remaining number of years until the probability of falling becomes high. The high-decline period may be calculated as the age at which the probability of falling becomes highest in the future progression of the probability of falling, even if the differential value is not greater than or equal to a threshold.

[0032] By using this configuration and processing, users (the individuals being predicted to fall and their instructors for strength training, etc.) can easily understand the age (period) at which they are most likely to fall. This allows users, for example, to understand what level of muscle strength they can achieve by the age (period) at which they are most likely to fall.

[0033] (Method for generating a fall probability curve) Figure 6 is a flowchart showing an example of a method for generating a trained model for generating a fall probability curve in the fall probability prediction method according to the first embodiment of the present invention. The specific details of each step in the flowchart shown in Figure 6 are described in the above-mentioned explanation of the configuration, so they are omitted unless further explanation is necessary.

[0034] The fall probability prediction system 10 obtains the age (S11). The fall probability prediction system 10 obtains fall information (whether or not a fall occurred) (S12).

[0035] The fall probability prediction system 10 takes age and fall information as input (explanatory variables) and generates a trained model that outputs an age-fall probability curve (S13).

[0036] Furthermore, after receiving an age input from the user, the fall probability prediction system 10 generates and outputs a corresponding fall probability curve. The fall probability prediction system 10 then calculates and outputs the age (period) at which the probability of falling increases, based on this fall probability curve.

[0037] [Second Embodiment] A fall probability prediction system and a fall probability prediction method according to a second embodiment of the present invention will be described with reference to the figures. Figure 7 is a functional block diagram of the fall probability prediction system according to the second embodiment of the present invention. Figure 8 is a table showing an example of explanatory variables for the learning algorithm according to the second embodiment of the present invention.

[0038] As shown in Figure 7, the fall probability prediction system 10A according to the second embodiment differs from the fall probability prediction system 10 according to the first embodiment in that it includes a body information acquisition unit 22 and in the trained model in the fall probability curve calculation unit 30A. The other components of the fall probability prediction system 10A are the same as those of the fall probability prediction system 10, and the explanation of the similar parts will be omitted.

[0039] The fall probability prediction system 10A includes a body information acquisition unit 22. The body information acquisition unit 22 includes a weight acquisition unit 221 and a gender acquisition unit 222.

[0040] The weight acquisition unit 221 acquires the weight of the person to be predicted. The weight acquisition unit 221 outputs the weight to the fall probability curve calculation unit 30A. The gender acquisition unit 222 acquires the gender of the person to be predicted. The gender acquisition unit 222 outputs the gender to the fall probability curve calculation unit 30A. In this case, weight and gender correspond to physical information.

[0041] The fall probability curve calculation unit 30A generates a trained model that outputs a fall probability curve, using age, physical information (weight, sex), and fall information as explanatory variables (inputs). The fall probability curve calculation unit 30A calculates the fall probability curve using this trained model.

[0042] With this configuration, the fall probability prediction system 10A can learn using further physical information and calculate a fall probability curve with higher accuracy. Furthermore, this allows the fall probability prediction system 10A to calculate with higher accuracy the periods when the probability of falling is high.

[0043] Note that the physical information is not limited to weight and gender, and may be any information indicating the physical characteristics of the person to be predicted, such as height, presence or absence of work (normal physical activity level), eyesight, hearing ability, etc.

[0044] [Third Embodiment] A fall probability prediction system and a fall probability prediction method according to the third embodiment of the present invention will be described with reference to the drawings. FIG. 9 is a functional block diagram of the fall probability prediction system according to the third embodiment of the present invention. FIG. 10 is a graph showing an example of a fall probability curve according to the third embodiment of the present invention.

[0045] As shown in FIG. 9, the fall probability prediction system 10B according to the third embodiment is different from the fall probability prediction system 10A according to the second embodiment in that the learned model in the fall probability curve calculation unit 30B is different, and in that it includes a fall factor identification unit 50B. Other configurations of the fall probability prediction system 10B are the same as those of the fall probability prediction system 10A, and descriptions of the same parts will be omitted.

[0046] The fall probability prediction system 10B includes a fall probability curve calculation unit 30B and a fall factor identification unit 50B.

[0047] The fall probability curve calculation unit 30B classifies specific information in the physical information into a plurality of conditions. For example, the fall probability curve calculation unit 30B classifies the input (explanatory variable) according to whether the weight at a specific age (for example, the youngest age at which falls are likely to increase (as an example, 60 years old)) is above or below a threshold value (threshold value for physical information, for example, 40 kg). Here, the weight is not the normal weight but the weight based on muscle mass. The weight based on muscle mass is measured using, for example, a body composition analyzer or the like.

[0048] The fall probability curve calculation unit 30B performs learning using the set of inputs where the weight is greater than or equal to the threshold for body information, and outputs a fall probability curve f31 (see FIG. 10). The fall probability curve calculation unit 30B performs learning using the set of inputs where the weight is less than the threshold for body information, and outputs a fall probability curve f32 (see FIG. 10). The vertical axis in FIG. 10 indicates, for example, the proportion of people who fall for the first time at a certain age or older. Alternatively, it may be the proportion of people who have fallen and have completed treatment and then fall after discharge, as long as it is the probability of a fall occurring in a group corresponding to certain conditions.

[0049] The fall factor identification unit 50B detects the inflection point (the position where the derivative value is large) of the fall probability curve for each set. The fall factor identification unit 50B calculates, for each set, the age (time period) at which the fall probability increases significantly from the detected inflection point.

[0050] For example, in the above case, when the weight is greater than or equal to the threshold for body information, the fall factor identification unit 50B calculates the age (time period) at which the fall probability increases significantly from the fall probability curve f31. When the weight is less than the threshold for body information, the fall factor identification unit 50B calculates the age (time period) at which the fall probability increases significantly from the fall probability curve f32.

[0051] The fall factor identification unit 50B compares the ages (time periods) at which the fall probability increases significantly for each classified set, and identifies how the classified information affects the increase in the fall probability.

[0052] For example, if the age (time period) at which the fall probability increases significantly is different between classifications, the fall factor identification unit 50B identifies that the classified information is a major factor for the significant increase in the fall probability. In the above case, if the age (time period) at which the fall probability increases significantly occurs earlier when the weight is less than the threshold than when the weight is greater than or equal to the threshold, the fall factor identification unit 50B identifies that the weight (muscle mass) is a major factor affecting the fall probability.

[0053] Note that the fall factor identification unit 50B may identify factors based on the difference in the magnitude of the fall probability or the magnitude of the change amount, or may identify factors based on the temporal difference in the age (time period) at which the fall probability increases significantly.

[0054] This allows the fall probability prediction system 10B to notify the user of physical factors that increase the probability of falling (increase the risk of falling). Therefore, for example, the user can take these factors into consideration when making future actions (lifestyle habits, etc.) for themselves or the person they are supervising. For example, in this case, the user or their supervisor can take care to ensure that their weight (muscle mass) does not fall below a threshold when making future actions (lifestyle habits, etc.).

[0055] In the configuration and processing described above, the fall probability prediction system 10B classified one type of information contained in the physical information to identify factors causing a significant or rapid increase in the probability of falling. However, the fall probability prediction system 10B may classify multiple types of information to identify factors causing a significant or rapid increase in the probability of falling. Furthermore, the fall probability prediction system 10B can also compare the fall probability curves of multiple pieces of information to identify which pieces of information have a significant influence as factors causing a significant or rapid increase in the probability of falling.

[0056] [Fourth Embodiment] A fall probability prediction system and a fall probability prediction method according to the fourth embodiment of the present invention will be described with reference to the figures. Figure 11 is a functional block diagram of the fall probability prediction system according to the fourth embodiment of the present invention. Figure 12 is a functional block diagram showing an example of a muscle activity information acquisition unit according to the fourth embodiment of the present invention. Figure 13 is a graph showing an example of a fall probability curve according to the fourth embodiment of the present invention.

[0057] As shown in Figure 11, the fall probability prediction system 10C according to the fourth embodiment differs from the fall probability prediction system 10B according to the third embodiment in that it includes a muscle activity information acquisition unit 23, differs in the trained model in the fall probability curve calculation unit 30C, and differs in the processing of the fall factor identification unit 50B. The other components of the fall probability prediction system 10C are the same as those of the fall probability prediction system 10B, and the explanation of the similar parts will be omitted.

[0058] The fall probability prediction system 10C includes a muscle activity information acquisition unit 23, a fall probability curve calculation unit 30C, and a fall factor identification unit 50C.

[0059] As shown in Figure 12, the muscle activity information acquisition unit 23 includes a sensor 2311, a sensor 2312, a muscle activity amount calculation unit 232, and a muscle activity information generation unit 233.

[0060] Sensors 2311 and 2312 are, for example, sensors attached to the person being predicted during prediction, and sensors attached to multiple people that generate the biosignals that form the basis of the explanatory variables during the generation of a trained model. Sensors 2311 and 2312 measure biosignals (for example, tremor signals of muscles related to walking) that correlate with multiple muscle activity levels that form the basis of the explanatory variables of the person being measured (the person being predicted during prediction, and multiple people during the generation of a trained model). Sensors 2311 and 2312 output the measured biosignals to the muscle activity level calculation unit 232.

[0061] Sensors 2311 and 2312 are constructed using, for example, a piezoelectric film, but may also be composed of an acceleration sensor, an angular velocity sensor, a microphone, a vibration sensor, an electromyography sensor, or a myosound sensor. When a piezoelectric film is used, tremor signals can be measured accurately as biological signals.

[0062] Sensors 2311 and 2312 may be of the same type or different types. If they are different types, they may be placed in approximately the same location on the person being predicted, or in different locations. If they are of the same type, sensors 2311 and 2312 may be placed on multiple muscles related to walking, for example, on the person being predicted. There may be one sensor or three or more sensors.

[0063] The muscle activity calculation unit 232 calculates the amount of muscle activity of muscles related to walking based on biological signals from sensors 2311 and 2312. The muscle activity calculation unit 232 generates muscle activity information based on the amount of muscle activity over a predetermined period.

[0064] For example, muscle activity information may directly represent muscle activity, such as the average muscle activity over a predetermined period (e.g., one week) or muscle activity for each muscle group, or it may indirectly represent muscle activity, such as the time the sensor was worn (exercise awareness), exercise frequency, average walking time, commuting time, or average number of steps. Note that muscle activity information may include multiple of these types of information.

[0065] The fall probability curve calculation unit 30C classifies specific information in the muscle activity information into multiple conditions. For example, the fall probability curve calculation unit 30C classifies the input (explanatory variable) based on whether the average muscle activity is above or below a threshold (a threshold for muscle activity information, e.g., 7000).

[0066] The fall probability curve calculation unit 30C learns using input sets in which the average muscle activity is greater than or equal to the threshold for muscle activity information, and outputs a fall probability curve f31C (see Figure 13). The fall probability curve calculation unit 30C learns using input sets in which the average muscle activity is less than the threshold for muscle activity information, and outputs a fall probability curve f32C (see Figure 13).

[0067] The fall risk identification unit 50C detects inflection points (positions where the differential value is large) in the fall probability curve for each group. From the detected inflection points, the fall risk identification unit 50C calculates the age (period) in which the probability of falling increases significantly for each group.

[0068] For example, in the above case, the fall risk identification unit 50C calculates the age (time) at which the probability of falling increases significantly when the average muscle activity is above the threshold for muscle activity information, from the fall risk curve f31C. The fall risk identification unit 50C also calculates the age (time) at which the probability of falling increases significantly when the average muscle activity is below the threshold for muscle activity information, from the fall risk curve f32C.

[0069] The fall risk identification unit 50C compares the age (period) at which the probability of falling increases significantly for each classified group, and identifies how the classified information influences the increase in the probability of falling.

[0070] For example, the fall risk identification unit 50C identifies the classified information as a major factor in the significant increase in fall probability if the age (time) at which the probability of falling increases significantly differs between classifications. In the above case, the fall risk identification unit 50C identifies the average muscle activity level as a major factor influencing the probability of falling if, in the case where the average muscle activity level is above a threshold and the case where the average muscle activity level is below a threshold, the age (time) at which the probability of falling increases significantly occurs earlier in the case where the average muscle activity level is below a threshold.

[0071] The fall risk identification unit 50C may identify factors based on differences in the magnitude or magnitude of change in the probability of falling, or it may identify factors based on the time difference in the age (period) at which the probability of falling increases significantly.

[0072] This allows the fall probability prediction system 10C to notify the user of muscle activity factors that increase the probability of falling (increase the risk of falling). Therefore, for example, the user can take these factors into consideration when conducting future muscle training or exercise for themselves or the person they are training. For example, in this case, the user or their instructor can take care to ensure that the average muscle activity level does not fall below a threshold when conducting future muscle training or exercise.

[0073] In the configuration and processing described above, the fall probability prediction system 10C classified one type of information contained in muscle activity information to identify factors causing a significant or rapid increase in the probability of falling. However, the fall probability prediction system 10C may classify multiple types of information to identify factors causing a significant or rapid increase in the probability of falling. Furthermore, the fall probability prediction system 10C can also compare the fall probability curves of multiple pieces of information to identify which piece of information has a significant influence as a factor causing a significant or rapid increase in the probability of falling.

[0074] [Fifth Embodiment] A fall probability prediction system and a fall probability prediction method according to the fifth embodiment of the present invention will be described with reference to the figures. Figure 14 is a functional block diagram of the fall probability prediction system according to the fifth embodiment of the present invention.

[0075] As shown in Figure 14, the fall probability prediction system 10D according to the fifth embodiment differs from the fall probability prediction system 10C according to the fourth embodiment in that it includes a preventive measures unit 60D. The fall probability curve calculation unit 30D and fall factor identification unit 50D shown in Figure 14 have the same configuration as the fall probability curve calculation unit 30D and fall factor identification unit 50D shown in Figure 11 according to the third embodiment. The other configurations of the fall probability prediction system 10D are the same as those of the fall probability prediction system 10C, and the explanation of the similar parts will be omitted.

[0076] The fall probability prediction system 10D includes a preventive measures unit 60D. The preventive measures unit 60D receives muscle activity information that significantly affects the probability of falling from the fall factor identification unit 50D.

[0077] The preventive measure unit 60D is, for example, an EMS (Electrical Muscle Stimulation) device, which is attached to the person being targeted for prediction. Based on muscle activity information that greatly influences the probability of falling, the preventive measure unit 60D provides stimulation to activate the muscles that are the source of (obtain) this muscle activity information.

[0078] As a result, the fall probability prediction system 10D can effectively prevent falls.

[0079] In the above example, a pre-trained model was generated as a preliminary step using the ages and fall information of multiple individuals, as well as, if necessary, physical information and muscle activity information. However, for predicting the probability of falling, it is not necessary to continue using the generated pre-trained model. The pre-trained model may be further learned and updated based on the subsequent fall status of the individuals for whom the fall probability curve was calculated using this system.

[0080] Furthermore, the configurations of each of the embodiments described above can be combined as appropriate. For example, it is possible to combine physical information and muscle activity information to perform processing such as calculating the period when the probability of falling is high.

[0081] 10, 10A, 10B, 10C, 10D, 10X: Fall probability prediction system 21: Age acquisition unit 22: Physical information acquisition unit 23: Muscle activity information acquisition unit 30, 30A, 30B, 30C, 30D: Fall probability curve calculation unit 40: High probability period calculation unit 50B, 50C, 50D: Fall factor identification unit 60D: Preventive measures unit 221: Weight acquisition unit 222: Gender acquisition unit 232: Muscle activity amount calculation unit 233: Muscle activity information generation unit 300: Learning algorithm 2311, 2312: Sensors f1, f31, f31C, f32, f32C: Fall probability curve

Claims

1. A fall probability prediction device equipped with a computer function that outputs a numerical value calculated based on input values, and includes a trained model that has been trained to output a fall probability curve which is a function of age and fall probability, with the ages of multiple people and whether or not each of the multiple people has fallen as explanatory variables, and when the age of the person to be predicted is input as the input value, it outputs a fall probability curve that predicts the probability of the person to be predicted falling at a time after the input time.

2. The fall probability prediction device according to claim 1, which detects an inflection point in the fall probability curve and further outputs information about the inflection point that is located after the time the age was input, and the period after the time the fall probability increases significantly.

3. The pre-trained model is a pre-trained model that has been machine-trained by further including the physical information of the multiple persons as explanatory variables, and the fall probability prediction device further includes the physical information of the person to be predicted as an input value, as described in claim 1 or claim 2.

4. The fall probability prediction device according to claim 3, wherein the physical information is at least one of height, weight, sex, and muscle activity level.

5. The physical information is muscle activity, which is determined by calculating from tremor, as described in claim 4, for the fall probability prediction device.

6. The fall probability prediction device according to any one of claims 1 to 5, wherein the trained model is created by survival time analysis.

7. A fall probability prediction system comprising a user terminal and a calculation processing device that outputs based on values ​​entered into the user terminal, wherein the user terminal includes an age acquisition unit that acquires the age of a person to be predicted, and when the age is entered, transmits the information to the calculation processing device, and the calculation processing device includes a fall probability curve calculation unit that calculates a fall probability curve that predicts the probability of the person to be predicted falling at a time after the time of input, and the fall probability curve calculation unit includes a trained model that has been trained to use machine learning to output the fall probability curve, which is a function of the age and the probability of falling, with the ages of multiple people and whether or not the multiple people have fallen as explanatory variables.

8. The fall probability prediction system according to claim 7, wherein the calculation processing device includes a high-probability period calculation unit that detects an inflection point in the fall probability curve and calculates information on the inflection point that is located after the time the age was input, and a period after the time the fall probability is input when the fall probability increases significantly.

9. A fall probability prediction system according to claim 7 or claim 8, comprising a body information acquisition unit that acquires the body information of the person to be predicted, wherein the fall probability curve calculation unit includes the trained model that has been machine-trained to further include the body information of the multiple persons as explanatory variables.

10. The fall probability prediction system according to claim 9, wherein the physical information can be classified into multiple states using thresholds for physical information, and the fall probability curve calculation unit uses a trained model that outputs the fall probability curve for each state of the physical information.

11. The fall probability prediction system according to claim 10, wherein the calculation processing device includes a fall factor identification unit that detects inflection points of the fall probability curve for each state of the physical information and identifies physical information that affects the fall from the inflection points.

12. A fall probability prediction system according to any one of claims 7 to 11, comprising a muscle activity information acquisition unit that acquires muscle activity information of the person to be predicted, and the fall probability curve calculation unit includes the trained model that has been machine-trained to further include the muscle activity information of the multiple persons as explanatory variables.

13. The fall probability prediction system according to claim 12, wherein the muscle activity information acquisition unit comprises: a sensor that measures a biosignal correlated with the muscle activity level of the person to be predicted; a muscle activity level calculation unit that calculates the muscle activity level from the biosignal; and a muscle activity information generation unit that generates the muscle activity information based on the muscle activity level over a predetermined period.

14. The muscle activity information can be classified into multiple states using a threshold for muscle activity information, and the fall probability curve calculation unit uses a trained model that outputs the fall probability curve for each state of the muscle activity information, the fall probability prediction system according to claim 12 or claim 13.

15. The fall probability prediction system according to claim 14, wherein the calculation processing device includes a fall factor identification unit that detects inflection points of the fall probability curve for each state of the muscle activity information and identifies muscle activity information that affects falls from the inflection points.

16. The fall probability prediction system according to claim 15, further comprising a preventive measure unit that provides stimulation to activate the muscle activity that forms the basis of muscle activity information that affects falls.

17. The fall probability prediction system according to any one of claims 7 to 16, wherein the trained model is trained using a survival time analysis method.

18. A method for predicting the probability of falling, comprising: an age acquisition step of obtaining the age of a person to be predicted; and a fall probability curve calculation step of calculating a fall probability curve that predicts the probability of the person to be predicted falling at a time after the age has been input, wherein the fall probability curve calculation step includes a trained model that has been trained to output a fall probability curve that is a function of the age and the probability of falling, with the ages of multiple people and whether or not the multiple people have fallen as explanatory variables.

19. A fall probability prediction program for computational processing on a computer, which uses a computation that includes a trained model that has been trained to output a fall probability curve with the ages of multiple people and whether or not the multiple people have fallen as explanatory variables, to take the age of the person to be predicted as input, calculate and output a fall probability curve that shows the relationship between the probability of the person to be predicted falling and the age of the person to be predicted at a time after the input time.

Images