Method, support device, support program, and support system for supporting advance prediction of target time water removal amount by dialysis or occurrence of blood pressure reduction per unit time

Abstract

To predict, in advance and without relying on a physician's judgment, a water removal amount per unit time by dialysis and the presence or absence of occurrence of blood pressure reduction per unit time by dialysis.SOLUTION: A method by a support device inputs analysis datasets (i) to (iv) into a support model constituted by artificial intelligence, and outputs labels indicating a target time water removal amount on the dialysis implementation day and presence or absence of occurrence of blood pressure reduction per unit time. (i) The first dialysis dataset in at least one past session immediately prior to the dialysis implementation day acquired from a prediction subject. (ii) The profile dataset of the prediction subject. (iii) The second dialysis dataset actually acquired prior to the unit time being the prediction target of the target time water removal amount on the dialysis implementation day for the prediction subject, or prior to the unit time for predicting occurrence of blood pressure reduction by dialysis. (iv) The specimen test dataset immediately prior to the dialysis implementation day acquired from the prediction subject.SELECTED DRAWING: Figure 15

Description

[Technical Field]

[0001] This specification discloses a method, an assistance device, an assistance program, and an assistance system that use artificial intelligence to assist in the advance prediction of the target time amount of water removal by dialysis or the occurrence of blood pressure drops per unit time. [Background technology]

[0002] Intradialytic hypotension is an adverse event that occurs during fluid removal during dialysis, and is accompanied by symptoms such as leg cramps, malaise, and loss of consciousness, and has a negative impact on life prognosis. Hypotension occurs in 5-10% of patients.

[0003] Non-Patent Document 1 describes that the performance of a Dual-Channel Combiner Network (DCCN) was evaluated using a dialysis event dataset. Patent Document 1 describes a personalized DCCN. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2022-216618A1 [Non-patent literature]

[0005] [Non-Patent Document 1] https: / / nijingchao.github.io / paper / iclr22_scgm.pdf Summary of the Invention [Problem to be solved by the invention]

[0006] The amount of fluid removed per unit time during dialysis is usually determined by the doctor. Furthermore, if blood pressure drops during dialysis, it can worsen the patient's condition, so doctors and nurses must take action. However, while multiple patients typically undergo dialysis in a dialysis room, the number of staff members stationed there is limited. Therefore, if a patient's blood pressure suddenly drops during dialysis, many of the staff members stationed there will be busy dealing with the situation. Furthermore, if another patient experiences a drop in blood pressure during dialysis, there is a risk that response may be delayed.

[0007] Therefore, if predictions can be made before dialysis begins, the number of staff can be increased in advance, enabling safe and secure dialysis treatment. Furthermore, if the amount of fluid removal that does not cause a drop in blood pressure can be known in advance, safer dialysis treatment can be performed.

[0008] The present invention aims to predict in advance the amount of water removed per unit time by dialysis, and to predict whether or not a drop in blood pressure will occur per unit time by dialysis, without relying on the judgment of a doctor. [Means for solving the problem]

[0009] One embodiment relates to an assistance device 20, 70 that assists in the advance prediction of a target time water removal volume due to dialysis on a dialysis day or the occurrence of a blood pressure decrease due to dialysis per unit time for a prediction subject. The target time water removal volume is a target amount of water removal per unit time after the start of dialysis that is the subject of prediction on a dialysis day. The assistance device 20, 70 is directly connected to the dialysis machine. The assistance device 20, 70 includes a control unit 200, 700. The control unit 200, 700 inputs the following analysis datasets (i) and (ii), (i) to (iii), (i) and (ii) and (iv), or (i) to (iv) into an assistance model configured with artificial intelligence, and outputs a label indicating the target time water removal volume or the occurrence of a blood pressure decrease per unit time on the dialysis day: (i) a first dialysis data set obtained from the subject of the prediction at least once immediately before the dialysis date; (ii) a profile dataset including a profile of the subject and a profile of a dialysis device that the subject will use on the dialysis date; (iii) A second dialysis data set that is actually acquired for the prediction subject on the dialysis performance date before the unit time that is the target time water removal amount prediction target or before the unit time that predicts the occurrence of blood pressure decrease due to dialysis, the second dialysis data set being the second dialysis data set acquired from a dialysis machine connected to the assistance device; (iv) A specimen test dataset obtained from the subject of prediction immediately prior to the date of dialysis.

[0010] The first dialysis data set includes blood flow rate per unit time, fluid pressure of the dialysis fluid per unit time, amount of fluid removed per unit time, rate of change in circulating blood volume per specified time, fluid removal rate per unit time, systolic blood pressure and diastolic blood pressure per unit time, whether or not a drop in blood pressure occurred during dialysis per unit time, temperature of the dialysis fluid on the day of dialysis, target total amount of fluid removed determined by a doctor on the day of dialysis, dry weight on the day of dialysis, previous post-dialysis weight, weight before the start of dialysis on the day of dialysis, weight increase from before the start of dialysis on the day of dialysis based on the previous post-dialysis weight, increase from dry weight on the day of dialysis, post-dialysis weight on the day of dialysis, systolic blood pressure and diastolic blood pressure before the start of dialysis on the day of dialysis, standard dialysis time on the day of dialysis, and dialysis time on the day of dialysis. The profile of the predicted subject described in (ii) includes the dialysis start date, dialysis time period, gender, and age group of the predicted subject, and the profile of the dialysis machine includes the type of dialysis method. The second dialysis data set described in (iii) includes the dry weight obtained for the prediction subject on the day of dialysis, the previous value of the post-dialysis weight, the pre-dialysis weight on the day of dialysis obtained on the day of dialysis, the increase in weight up to the start of dialysis on the day of dialysis based on the previous value of the post-dialysis weight, the blood flow rate at each unit time, the fluid pressure of the dialysis fluid at each unit time, the amount of water removed at each unit time, the rate of change in circulating blood volume at each specified time, the water removal rate at each unit time, the systolic blood pressure and diastolic blood pressure at each unit time, and whether or not a drop in blood pressure has occurred at each unit time. The specimen test dataset described in (iv) above includes biochemical test data and blood test data.

[0011] Preferably, the target amount of water removal per unit time immediately after the start of dialysis or the occurrence of a drop in blood pressure per unit time is predicted by inputting the analysis datasets (i) and (ii) into the support model.

[0012] Preferably, the target time water removal amount or the occurrence of blood pressure reduction per unit time is predicted for multiple unit times.

[0013] Preferably, the assistance model is trained using the following training datasets (A) and (B), or (A), (B), and (C) obtained from a plurality of dialysis patients: (A) a dialysis dataset obtained from each of the plurality of dialysis patients; (B) a profile data set including a profile of each of the plurality of dialysis patients and a profile of a dialysis machine used by each of the patients; (C) A specimen test dataset obtained from each of the plurality of dialysis patients.

[0014] The dialysis dataset in (A) includes blood flow rate per unit time, fluid pressure of the dialysis fluid per unit time, volume of fluid removed per unit time, rate of change in circulating blood volume per specified time, volume of fluid removed per unit time, systolic blood pressure and diastolic blood pressure per unit time, whether or not a drop in blood pressure occurred during dialysis per unit time, temperature of the dialysis fluid on the day of dialysis, target total volume of fluid removed determined by the doctor on the day of dialysis, dry weight on the day of dialysis, previous post-dialysis weight, weight before the start of dialysis on the day of dialysis, weight gain from before the start of dialysis on the day of dialysis based on the previous post-dialysis weight, gain from dry weight on the day of dialysis, post-dialysis weight on the day of dialysis, systolic blood pressure and diastolic blood pressure before the start of dialysis on the day of dialysis, standard dialysis time on the day of dialysis, and dialysis time on the day of dialysis. The profile of each of the plurality of dialysis patients in (B) includes the start date of dialysis, the time period during which dialysis is performed, the gender and age group of each dialysis patient, and the profile of the dialysis machine includes the type of dialysis method. The specimen test dataset described in (C) above includes biochemical test data and blood test data.

[0015] Preferably, the artificial intelligence that constitutes the assistance model is an algorithm having a structure other than a neural network structure, or an algorithm having a neural network structure.

[0016] Preferably, the algorithm having a structure other than a neural network structure is a gradient boosting algorithm or a random forest algorithm; Preferably, the algorithm having the neural network structure is a Dual-Channel Combiner Network algorithm or a Long Short-Term Memory algorithm.

[0017] In one embodiment, the present invention relates to a method for assisting in the advance prediction of a target time water removal amount by dialysis on a dialysis execution date or the occurrence of blood pressure decrease due to dialysis per unit time for a prediction subject. The target time water removal amount is a target amount of water removal per unit time after the start of dialysis to be predicted on the dialysis execution date. The method is aided by a support model that is comprised of artificial intelligence. The assistance model is configured to input the following analysis datasets (i and (ii), (i) to (iii), (i, (ii) and (iv), or (i) to (iv) into the assistance model, and output a label indicating the target time water removal amount on the dialysis day or whether or not a decrease in blood pressure has occurred per unit time: (i) a first dialysis data set obtained from the subject of the prediction at least once immediately before the dialysis date; (ii) a profile dataset including a profile of the subject and a profile of a dialysis device that the subject will use on the dialysis date; (iii) a second dialysis data set actually acquired for the prediction subject on the dialysis execution date before the unit time for which the target time water removal amount is predicted or before the unit time for which the occurrence of blood pressure decrease due to dialysis is predicted, the second dialysis data set being acquired from a dialysis machine connected to the assistance device according to claim 1; (iv) A specimen test dataset obtained from the subject of prediction immediately prior to the date of dialysis.

[0018] One embodiment relates to an assistance program for assisting in the advance prediction of a target amount of water removal per unit time due to dialysis on a dialysis day or the occurrence of a decrease in blood pressure due to dialysis per unit time for a prediction subject, the program causing the computer to execute, when executed by a computer, the following analysis datasets (i) and (ii), (i) to (iii), (i), (ii), and (iv), or (i) to (iv) to be input into an assistance model configured with artificial intelligence, and outputting a label indicating the target amount of water removal per unit time on a dialysis day or the occurrence of a decrease in blood pressure per unit time due to dialysis. The target amount of water removal per unit time is a target amount of water removal per unit time after the start of dialysis to be predicted on the dialysis day: (i) a first dialysis data set obtained from the subject of the prediction at least once immediately before the dialysis date; (ii) a profile dataset including a profile of the subject and a profile of a dialysis device that the subject will use on the dialysis date; (iii) a second dialysis data set actually acquired for the prediction subject on the dialysis execution date before the unit time for which the target time water removal amount is predicted or before the unit time for which the occurrence of blood pressure decrease due to dialysis is predicted, the second dialysis data set being acquired from a dialysis machine connected to the assistance device according to claim 1; (iv) A specimen test dataset obtained from the subject of prediction immediately prior to the date of dialysis.

[0019] One embodiment relates to a support device 30, 80 that supports advance prediction of a target time water removal amount due to dialysis on a dialysis day or the occurrence of a blood pressure drop due to dialysis per unit time for a prediction subject. The target time water removal amount is a target amount of water removal per unit time after the start of dialysis that is the subject of prediction on a dialysis day. The support device 30, 80 is a server device. The support device 30, 80 includes a control unit 300, 800. The control unit 300, 800 inputs the following analysis data sets (i) and (ii), (i) to (iii), (i), (ii) and (iv), or (i) to (iv) into a support model configured by artificial intelligence, and outputs a label indicating the target time water removal amount on the dialysis day or whether or not a drop in blood pressure has occurred per unit time: (i) a first dialysis data set obtained from the subject of the prediction at least once immediately before the dialysis date; (ii) a profile dataset including a profile of the subject and a profile of a dialysis device that the subject will use on the dialysis date; (iii) A second dialysis data set actually acquired for the prediction subject on the dialysis execution date before the unit time for which the target time water removal amount is predicted or before the unit time for which the occurrence of blood pressure decrease due to dialysis is predicted, and the second dialysis data set is transmitted to the server device via a network; (iv) A specimen test dataset obtained from the subject of prediction immediately prior to the date of dialysis.

[0020] Preferably, the first dialysis data set includes blood flow rate per unit time, fluid pressure of the dialysis fluid per unit time, fluid removal volume per unit time, rate of change in circulating blood volume per specified time, fluid removal rate per unit time, systolic blood pressure and diastolic blood pressure per unit time, whether or not a drop in blood pressure occurred during dialysis per unit time, fluid temperature of the dialysis fluid on the day of dialysis, target total fluid removal volume determined by a doctor on the day of dialysis, dry weight on the day of dialysis, previous post-dialysis weight, weight before the start of dialysis on the day of dialysis, weight increase from the previous post-dialysis weight to the start of dialysis on the day of dialysis, increase from dry weight on the day of dialysis, post-dialysis weight on the day of dialysis, systolic blood pressure and diastolic blood pressure before the start of dialysis on the day of dialysis, standard dialysis time on the day of dialysis, and dialysis time on the day of dialysis. Preferably, the profile of the predicted subject described in (ii) includes the dialysis start date, dialysis time period, gender, and age group of the predicted subject, and the profile of the dialysis machine includes the type of dialysis method. Preferably, the second dialysis data set described in (iii) includes the dry weight obtained for the prediction subject on the day of dialysis, the previous value of the post-dialysis weight, the pre-dialysis weight on the day of dialysis obtained on the day of dialysis, the increase in weight up to the start of dialysis on the day of dialysis based on the previous value of the post-dialysis weight, the blood flow rate at each unit time, the fluid pressure of the dialysis fluid at each unit time, the amount of water removed at each unit time, the rate of change in circulating blood volume at each specified time, the water removal rate at each unit time, the systolic blood pressure and diastolic blood pressure at each unit time, and whether or not a drop in blood pressure has occurred at each unit time. Preferably, the specimen test dataset described in (iv) above includes biochemical test data and blood test data.

[0021] Preferably, the target time water removal volume per unit time immediately after the start of dialysis is predicted by inputting the analysis datasets (i) and (ii) into the support model.

[0022] Preferably, the target time water removal amount is predicted for a plurality of unit times.

[0023] Preferably, the assistance model is trained using the following training datasets (A) and (B), or (A), (B), and (C) obtained from a plurality of dialysis patients: (A) a dialysis dataset obtained from each of the plurality of dialysis patients; (B) a profile data set including a profile of each of the plurality of dialysis patients and a profile of a dialysis machine used by each of the patients; (C) A specimen test dataset obtained from each of the plurality of dialysis patients.

[0024] Preferably, the dialysis data set in (A) includes blood flow rate per unit time, fluid pressure of the dialysis fluid per unit time, amount of fluid removed per unit time, rate of change in circulating blood volume per specified time, rate of fluid removal per unit time, systolic blood pressure and diastolic blood pressure per unit time, whether or not a drop in blood pressure occurred during dialysis per unit time, temperature of the dialysis fluid on the day of dialysis, target total amount of fluid removed determined by a doctor on the day of dialysis, dry weight on the day of dialysis, previous value of post-dialysis weight, weight before the start of dialysis on the day of dialysis, weight increase from the previous value of post-dialysis weight to the start of dialysis on the day of dialysis, increase from dry weight on the day of dialysis, post-dialysis weight on the day of dialysis, systolic blood pressure and diastolic blood pressure before the start of dialysis on the day of dialysis, standard dialysis time on the day of dialysis, and dialysis time on the day of dialysis. Preferably, the profile of each of the multiple dialysis patients in (B) includes the start date of dialysis, the time period during which dialysis is performed, the gender and age group of each dialysis patient, and the profile of the dialysis machine includes the type of dialysis method. Preferably, the specimen test dataset described in (C) above includes biochemical test data and blood test data.

[0025] Preferably, the assistance model is an algorithm having a structure other than a neural network structure, or an algorithm having a neural network structure.

[0026] Preferably, the algorithm having a structure other than a neural network structure is a gradient boosting algorithm or a random forest algorithm. Preferably, the algorithm having the neural network structure is a Dual-Channel Combiner Network algorithm or a Long Short-Term Memory algorithm.

[0027] In one embodiment, the present invention relates to a method for assisting in the advance prediction of a target time water removal amount by dialysis on a dialysis execution date or the occurrence of blood pressure decrease due to dialysis per unit time for a prediction subject. The target time water removal amount is a target amount of water removal per unit time after the start of dialysis to be predicted on the dialysis execution date. The method is aided by a support model that is comprised of artificial intelligence. The assistance model is configured to input the following analysis datasets (i and (ii), (i) to (iii), (i, (ii) and (iv), or (i) to (iv) into the assistance model, and output a label indicating the target time water removal amount on the dialysis day or whether or not a decrease in blood pressure has occurred per unit time: (i) a first dialysis data set obtained from the subject of the prediction at least once immediately before the dialysis date; (ii) a profile dataset including a profile of the subject and a profile of a dialysis device that the subject will use on the dialysis date; (iii) A second dialysis data set actually acquired for the prediction subject on the dialysis execution date before the unit time for which the target time water removal amount is predicted or before the unit time for which the occurrence of blood pressure decrease due to dialysis is predicted, and the second dialysis data set is transmitted to the server device via a network; (iv) A specimen test dataset obtained from the subject of prediction immediately prior to the date of dialysis.

[0028] One embodiment relates to an assistance program for assisting in the advance prediction of a target amount of water removal per unit time due to dialysis on a dialysis day or the occurrence of a decrease in blood pressure due to dialysis per unit time for a prediction subject, the program causing the computer to execute, when executed by a computer, the following analysis datasets (i) and (ii), (i) to (iii), (i), (ii), and (iv), or (i) to (iv) to be input into an assistance model configured with artificial intelligence, and outputting a label indicating the target amount of water removal per unit time on a dialysis day or the occurrence of a decrease in blood pressure per unit time due to dialysis. The target amount of water removal per unit time is a target amount of water removal per unit time after the start of dialysis to be predicted on the dialysis day: (i) a first dialysis data set obtained from the subject of the prediction at least once immediately before the dialysis date; (ii) a profile dataset including a profile of the subject and a profile of a dialysis device that the subject will use on the dialysis date; (iii) A second dialysis data set actually acquired for the prediction subject on the dialysis execution date before the unit time for which the target time water removal amount is predicted or before the unit time for which the occurrence of blood pressure decrease due to dialysis is predicted, and the second dialysis data set is transmitted to the server device via a network; (iv) A specimen test dataset obtained from the subject of prediction immediately prior to the date of dialysis.

[0029] One embodiment relates to a support system 1000, 2000 that supports advance prediction of a target time water removal amount by dialysis on a dialysis day or the occurrence of blood pressure decrease due to dialysis per unit time for a prediction subject. The target time water removal amount is a target amount of water removal per unit time after the start of dialysis to be predicted on the dialysis day. The support systems 1000 and 2000 include support devices 30 and 80, which are server devices, and a user terminal 40 used by a user. The support devices 30 and 80 include control units 300 and 800. The control unit 300, 800 receives the following analysis datasets (i) and (ii), (i) to (iii), (i), (ii) and (iv), or (i) to (iv), inputs the analysis datasets into a support model configured by artificial intelligence, and outputs a label indicating the target time water removal amount on the dialysis day or whether or not a decrease in blood pressure occurred per unit time: (i) a first dialysis data set obtained from the subject of the prediction at least once immediately before the dialysis date; (ii) a profile dataset including a profile of the subject and a profile of a dialysis device that the subject will use on the dialysis date; (iii) A second dialysis data set that was actually acquired for the prediction subject on the dialysis execution date before the unit time for which the target time water removal amount is predicted, and that was transmitted from the user terminal 40 to the support device 30, 80 via the network; (iv) A specimen test dataset obtained from the subject of prediction immediately prior to the date of dialysis. [Effects of the Invention]

[0030] For the subject of the prediction, it is possible to predict in advance the target amount of fluid removal per unit time by dialysis on the day of dialysis, or the occurrence of a drop in blood pressure per unit time. [Brief explanation of the drawings]

[0031] [Figure 1] 1 shows the hardware configuration of a training device 10. [Figure 2] 10 shows the processing flow of the training program 1042. [Figure 3] 1 shows the hardware configuration of the support device 20. [Figure 4] 10 shows the processing flow of the support program 2042. [Figure 5] 1 shows an overview of a support system 1000. [Figure 6] 1 shows the hardware configuration of the support device 30. [Figure 7] 2 shows the hardware configuration of the user terminal 40. [Figure 8] 1 shows the processing flow of the support system 1000. [Figure 9] 1 shows the hardware configuration of a training device 60. [Figure 10] 6 shows the processing flow of the training program 6042. [Figure 11] 1 shows the hardware configuration of the support device 70. [Figure 12] 7 shows the processing flow of the support program 7042. [Figure 13] 2 shows an overview of the support system 2000. [Figure 14]1 shows the hardware configuration of the support device 80. [Figure 15] 2 shows the processing flow of the support system 2000. [Figure 16] The analytical data used in the analysis is shown below. [Figure 17] This shows the prediction accuracy of the assistance model when using the dialysis data set immediately before the unit time to be predicted. [Figure 18] This shows the prediction accuracy of the assistance model when using dialysis data sets other than those immediately before the unit time to be predicted. [Figure 19] This shows the prediction accuracy of the blood pressure decrease prediction support model when using dialysis data sets immediately before the unit time to be predicted. DETAILED DESCRIPTION OF THE INVENTION

[0032] I. Prediction of target time fluid removal volume by dialysis 1. Support model for predicting the target amount of fluid removal during dialysis An embodiment of the present invention relates to a support model (hereinafter also simply referred to as a "water removal amount prediction support model") that supports advance prediction of a target time water removal amount by dialysis on a dialysis implementation day for a prediction subject.

[0033] The target time water removal volume is the target volume of water removal predicted in advance for a patient undergoing dialysis. The target volume is predicted for each unit of time after the start of dialysis. In other words, the target time water removal volume is the target volume of water removal per unit of time after the start of dialysis that is the target for the day of dialysis.

[0034] Here, "unit time" refers to a unit obtained by dividing the time from the start of dialysis to the end of dialysis (generally 4 or 5 hours) into fixed intervals. The unit time can be 2 hours, 1 hour, 30 minutes, 10 minutes, 5 minutes, etc. Preferably, it is 1 hour.

[0035] First, a training method for training an artificial intelligence to function as a water removal amount prediction support model will be described.

[0036] 1-1. Overview (1) Creation of a water removal volume prediction support model The artificial intelligence that constitutes the water removal amount prediction support model may be an algorithm having a structure other than a neural network structure, or an algorithm having a neural network structure. Examples of algorithms having a structure other than a neural network structure include a gradient boosting algorithm and a random forest algorithm. Examples of algorithms having a neural network structure include a dual-channel combiner network algorithm and a long short-term memory algorithm.

[0037] When the artificial intelligence is an algorithm having a neural network structure, it is preferably composed of a first algorithm structure and a second algorithm structure, where the first algorithm structure and the second algorithm structure are different.

[0038] For example, since a static parameter data set is input to the first algorithm, the first algorithm is an algorithm used to analyze static parameters. An example of the first algorithm is a feedforward neural network. Preferably, it is a feedforward neural network. Among them, a deep neural network (DNN) is more preferable. More preferably, the first algorithm is a fully connected neural network.

[0039] For example, the second algorithm is an algorithm used to analyze dynamic parameters because a temporal parameter data set is input to the second algorithm. The second algorithm may be, for example, a recurrent neural network. Preferably, the second algorithm is a recurrent neural network. Among them, a long short-term memory (LSTM) is more preferred.

[0040] An algorithm having such a different structure is known as a Dual-Channel Combiner Network (DCCN), as described in Non-Patent Document 1 and Patent Document 1. Non-Patent Document 1 and Patent Document 1 are incorporated herein by reference.

[0041] As the activation function in DCCN, ReLU (Rectified Linear Unit) or identity function can be used.

[0042] The static parameter datasets may include a profile dataset and a specimen test dataset, preferably a profile dataset. The occasional parameter datasets may include a dialysis dataset. Preferably, the static parameter datasets do not include a specimen test dataset.

[0043] When the artificial intelligence is an algorithm having a structure other than a neural network structure, both the static parameter data set and the ad hoc parameter data set are treated as the static parameter data set, and preferably the static parameter data set does not include the specimen test data set.

[0044] The specimen test dataset includes, for example, data from biochemical tests and blood tests. These data can be obtained through specimen tests at ordinary hospitals or testing centers. The specimen test dataset may preferably include serum albumin concentration, red blood cell count, hematocrit value, mean corpuscular hemoglobin (MCH), and platelet count. Preferably, the specimen test dataset includes serum albumin concentration, serum BUN concentration, serum creatinine concentration, serum calcium concentration, corrected calcium concentration, serum inorganic phosphorus concentration, serum iron concentration, total iron binding capacity (TIBC), serum sodium concentration, serum potassium concentration, serum chloride concentration, white blood cell count, red blood cell count, hemoglobin concentration, hematocrit value, mean corpuscular hemoglobin (MCH), platelet count, and serum CRP concentration. Specimen tests are generally performed about twice a month.

[0045] The dialysis dataset used for training is data acquired during each dialysis session. The dialysis dataset includes, for example, blood flow rate per unit time, dialysate pressure per unit time, fluid removal volume per unit time, rate of change in circulating blood volume per predetermined time, fluid removal rate per unit time, systolic blood pressure and diastolic blood pressure per unit time, whether or not blood pressure drops during dialysis per unit time, dialysate temperature on the dialysis day, target total fluid removal volume determined by a physician on the dialysis day, dry weight on the dialysis day, previous post-dialysis weight, pre-dialysis weight on the dialysis day, weight gain from the pre-dialysis weight on the dialysis day based on the previous post-dialysis weight, gain from dry weight on the dialysis day, post-dialysis weight on the dialysis day, systolic blood pressure and diastolic blood pressure before the start of dialysis on the dialysis day, standard dialysis time on the dialysis day, and dialysis time on the dialysis day. Dialysis is generally performed approximately three times a week. Here, the amount of water removed per unit time, blood flow per unit time, fluid pressure of the dialysis fluid per unit time, systolic blood pressure and diastolic blood pressure per unit time, and whether or not blood pressure has decreased per unit time may also include data for the unit time on the day of dialysis, going back from just before the unit time for which the target time water removal amount is predicted to the start of dialysis.

[0046] In principle, dry weight can be calculated from the cardiothoracic ratio of each dialysis patient. The cardiothoracic ratio is calculated from the images taken of chest X-rays taken once or twice a month for each dialysis patient. The standard value for the cardiothoracic ratio is approximately 50%, so doctors determine the dry weight based on this standard value. However, for some dialysis patients, determining the amount of fluid removal during dialysis based on the calculated dry weight can cause a drop in blood pressure during dialysis. In such cases, doctors will make adjustments, such as increasing the dry weight for each patient.

[0047] The occurrence of a blood pressure drop can be defined as a drop in systolic blood pressure of 20 mmHg or more from the previous measurement and a drop of 110 mmHg or less.

[0048] The circulating blood volume change rate (ΔBV%) is calculated for each predetermined time. The predetermined time may be the same as the unit time, but may also be different. If the predetermined time is shorter than the unit time (for example, if the unit time is 1 hour and the predetermined time is 10 minutes), all circulating blood volume change rates obtained within the unit time are used. If the predetermined time is longer than the unit time (for example, if the unit time is 1 hour and the predetermined time is 90 minutes), the unit time in which the water removal volume used for the correct label was obtained, or the circulating blood volume change rate immediately before the unit time being predicted, is used.

[0049] The profile dataset includes (a) profiles of dialysis patients or prediction subjects from which training data is obtained, and (b) profiles of the types of dialysis methods used by the dialysis patients or prediction subjects from which training data is obtained. The profile data includes data that is basically unchanged or that changes over time, such as age groups.

[0050] The profile of the dialysis patient or prediction target from which training data is obtained includes the time of day when dialysis is performed (morning or afternoon), the date when dialysis started, gender, and age group. Age groups can be quantified, for example, into teens, twenties, thirties, forties, fifties, sixties, seventies, eighties, and nineties.

[0051] In principle, the time of day when dialysis is performed (morning or afternoon) is the same for each patient. Types of dialysis procedures are hemodialysis (HD) therapy or hemodiafiltration (HDF) therapy.

[0052] For the various types of data mentioned above, if the data is continuous quantitative data (numerical data), the numerical value is used, and if the data is qualitative data (non-continuous data) or age group, a label value according to the type of quality is used.

[0053] The artificial intelligence is trained using a training dataset obtained from multiple dialysis patients to function as a water removal volume prediction support model. The multiple patients may be 10 or more, 100 or more, 1,000 or more, 10,000 or more, or 100,000 or more. Because each patient undergoes dialysis multiple times, the training dataset may be 20 or more, 200 or more, 2,000 or more, 20,000 or more, or 200,000 or more.

[0054] For training, (A) a dialysis dataset obtained from each of a plurality of dialysis patients, and (B) a profile dataset including a profile of each of the plurality of dialysis patients and a profile of the dialysis machine used by each of the patients are used. In addition to (A) and (B), (C) a specimen test dataset obtained from each of the plurality of dialysis patients may also be used for training. The above datasets (A) and (B), or (A), (B), and (C), are linked to each patient. The above training datasets (A) and (B), or (A), (B), and (C) are referred to as the training dataset.

[0055] Since each dialysis patient undergoes multiple dialysis sessions, training uses all possible input data sets of (A) and (B) or (A), (B) and (C) obtained for each of multiple patients.

[0056] If the algorithm has a structure other than a neural network structure, the correct label is the amount of water removed per unit time for each past dialysis session for each patient, and the artificial intelligence is trained using a training dataset. In this case, for the amount of water removed per unit time, blood flow per unit time, dialysate pressure per unit time, systolic blood pressure and diastolic blood pressure per unit time, and whether or not blood pressure has decreased per unit time, if the unit time of water removal input to the artificial intelligence is the Xth unit time, at least one set, preferably two or more sets, and more preferably all sets of datasets going back from the X-1st unit time immediately preceding the Xth unit time to the first unit time at which dialysis started are input to the artificial intelligence.

[0057] When the algorithm has a neural network structure, training involves inputting a training dataset into the input layer of the artificial intelligence, and inputting the amount of water removed per unit time for each patient's past dialysis session into the output layer of the artificial intelligence. In this case, for the amount of water removed per unit time, blood flow per unit time, dialysate pressure per unit time, systolic blood pressure and diastolic blood pressure per unit time, and whether or not blood pressure has decreased per unit time, if the unit time of water removal input into the output layer of the artificial intelligence is designated as the Xth unit time, at least one set, preferably two or more sets, and more preferably all sets of data going back from the X-1st unit time immediately preceding the Xth unit time to the first unit time at which dialysis began are input into the input layer of the algorithm. The artificial intelligence trained with the above training dataset is called a water removal volume prediction support model.

[0058] It is preferable that the water removal volume prediction support model is constructed for each unit time in one dialysis treatment. Since there are multiple unit times to be predicted in one dialysis treatment, multiple water removal volume prediction support models can be constructed for each unit time. For example, if one dialysis session lasts five hours and one hour is used as the unit time, the following five support models are constructed for one dialysis session for one prediction subject. Each of the five support models corresponds to a different unit time. First water removal volume prediction support model: predicts the target time water removal volume from the start of dialysis to one hour after the start of dialysis (first unit time), Second water removal volume prediction support model: predicts the target time water removal volume from 1 hour to 2 hours after the start of dialysis (second unit time), Third water removal volume prediction support model: predicts the target time water removal volume from 2 to 3 hours after the start of dialysis (third unit time), 4th water removal volume prediction support model: predicts the target time water removal volume from 3 to 4 hours after the start of dialysis (4th unit time), 5th water removal volume prediction support model: Predicts the target time water removal volume from 4 to 5 hours after the start of dialysis (5th unit time).

[0059] For the first water removal volume prediction support model, because part of the dialysis data set for that day (water removal volume per unit time, blood flow rate per unit time, dialysate pressure per unit time, systolic and diastolic blood pressure per unit time, and whether or not blood pressure has decreased per unit time) is not available before the first unit time on the day of prediction in (A) above, this data is not used to train the artificial intelligence. From the second water removal volume prediction support model onwards, dialysis data sets acquired for each unit time on that day are also used. For example, the second water removal volume prediction support model is a support model that predicts the target time water removal volume for the second unit time. The second water removal volume prediction support model also uses the water removal volume per unit time, blood flow rate per unit time, dialysate pressure per unit time, systolic and diastolic blood pressure per unit time, and whether or not blood pressure has decreased per unit time acquired at the first unit time. For example, the fifth water removal volume prediction support model is a support model that predicts the target time water removal volume for the fifth unit time. The fifth water removal volume prediction support model is trained using the water removal volume per unit time, blood flow per unit time, dialysis fluid pressure per unit time, systolic blood pressure and diastolic blood pressure per unit time, and whether or not there is a decrease in blood pressure per unit time, all of which are obtained within at least one unit time from the first unit time to the fourth unit time.

[0060] Furthermore, since dialysis patients undergo dialysis approximately three times a week, their dialysis datasets and specimen test datasets are accumulated each time dialysis is performed. Therefore, when new datasets are acquired, they can be added to the training dataset and used to retrain the trained AI. This is expected to improve the prediction accuracy of the assistance model.

[0061] (2) Personalization of trained AI Optionally, the water removal volume assistance model may be personalized for the person being predicted. In this training, to further personalize the artificial intelligence trained by the above method to fit a specific person being predicted, (1) a dataset of multiple past dialysis sessions obtained from the specific person being predicted, and (2) a profile dataset containing a profile of the specific person being predicted and a profile of the dialysis machine used by the person being predicted are used. In addition to (1) and (2), (3) a specimen test dataset obtained from the person being predicted may also be used. The training datasets (1) and (2), or (1), (2), and (3) above, are referred to as individual training datasets.

[0062] In the training for individualization, all inputtable datasets (1) and (2) or datasets (1), (2), and (3) obtained for a specific prediction subject are used. In the training for individualization, the individual training dataset is input to an artificial intelligence trained with datasets (A) and (B) or (A), (B), and (C) described in 1-1.(1) above, and the water removal volume per unit time for each past dialysis session of the specific prediction subject is input as a correct answer label. As in 1-1.(1) above, if the unit time of water removal volume to be input as a correct answer label is Xth, at least one set, preferably two or more sets, and more preferably all sets of datasets going back from the unit time X-1 immediately before unit time X to the first dialysis session at which dialysis started are input to the trained artificial intelligence. The artificial intelligence trained in this manner is called an individualized water removal volume prediction support model.

[0063] The individualized water removal volume prediction support model constructed in this manner can also be used as a support model for supporting advance prediction of the target time water removal volume by dialysis on the dialysis date for the specific prediction subject. By using the individualized water removal volume prediction support model, higher prediction accuracy can be achieved. Furthermore, the individualized water removal volume prediction support model can accurately predict the target time water removal volume for the prediction subject even if the facility that obtained the initial training data is different from the facility that obtained the individualized training data. Furthermore, since the prediction subject receives dialysis therapy continuously, individual training datasets are accumulated each time. The individualized water removal volume prediction support model can be retrained with the added individual training dataset each time an individual training dataset for the prediction subject is added, thereby becoming a support model that is more suited to each prediction subject.

[0064] 1-2. Support model training device 10 Figure 1 shows the hardware configuration of a training device 10 (hereinafter simply referred to as "training device 10") for a water removal volume prediction support model that supports advance prediction of the target time water removal volume by dialysis on the day of dialysis for a prediction subject. The training device 10 may be connected to an input device 111 and an output device 112 .

[0065] In the training device 10, a processing unit (CPU) 101, a memory 102, a ROM (read only memory) 103, a storage device 104, and an interface 106 are connected to each other via a bus 109 so as to be able to communicate data with each other. The processing unit 101, the memory 102, and the ROM 103 function as a control unit 100 of the training device 10.

[0066] The processing unit 101 is the CPU of the training device 10. The processing unit 101 may cooperate with a GPU or MPU. The processing unit 101 cooperates with an operating system (OS) 1041 stored in the storage device 104 or the ROM 103 to execute an artificial intelligence training program 1042 (hereinafter also simply referred to as the "training program 1042") that supports advance prediction of the target time water removal amount by dialysis on the dialysis execution date for a prediction target person described below, thereby causing the computer to function as the training device 10.

[0067] The ROM 103 stores a training program 1042 executed by the processing unit 101 and data used therefor. The ROM 103 also stores a boot program executed by the processing unit 101 when the training device 10 is started up, as well as programs and settings related to the operation of the hardware of the training device 10.

[0068] The storage device 104 non-volatilely stores an operating system (OS) 1041, a training program 1042 (hereinafter simply referred to as the "training program 1042") for training an artificial intelligence that constitutes a water removal amount prediction support model described below, and a model database 1043. The model database 1043 stores the artificial intelligence before training, a water removal amount prediction support model constituted by the trained artificial intelligence, and / or an individualized water removal amount prediction support model. The model database 1043 can also store a training dataset and an individual training dataset.

[0069] The input device 111 is composed of a touch panel, a keyboard, a mouse, a pen tablet, a microphone, etc., and is used to input text or voice to the training device 10. The input device 111 may be connected to the training device 10 from outside, or may be integrated with the training device 10. The output device 112 is configured by, for example, a display, a printer, and the like.

[0070] The processing unit 101 may acquire the application software and various settings required for controlling the training apparatus 10 via a network instead of reading them from the ROM 103 or the storage device 104. The application program may be stored in a storage device of a server computer on the network, and the training apparatus 10 may access this server computer to download a training program 1042 and store it in the ROM 103 or the storage device 104.

[0071] An operating system that provides a graphical user interface environment, such as Windows (registered trademark) manufactured and sold by Microsoft Corporation or the open-source Linux (registered trademark), is installed in ROM 103 or storage device 104. The training program runs on the operating system. In other words, training device 10 may be a personal computer or the like.

[0072] 1-3. Processing by training program 1042 2 shows the flow of processing performed by the training program 1042. The control unit 100 executes the training program 1042, causing the computer to function as the training device 10. The control unit 100 receives a processing start request input by the user from the input device 111, for example, and executes the training program 1042 to start the training processing.

[0073] In step S11, the control unit 100 reads the pre-training artificial intelligence and the training data set from the model database 1043, inputs the training data set into the artificial intelligence according to the method described in 1-1.(1) above, and trains the artificial intelligence to construct a trained artificial intelligence. The control unit 100 stores the trained artificial intelligence in the storage device 104.

[0074] In step S12, the control unit 100 receives a user command to start the verification process and verifies the trained AI. The verification can be performed using the error between the amount of water removal per unit time for each prediction subject and the predicted target time amount of water removal for each prediction subject predicted by the trained AI as an index. For example, the index can be the accuracy rate, mean absolute percentage error (MAPE), or mean absolute error (MAE). MAE is calculated using the following formula:

number

[0075] The accuracy rate is the accuracy rate of the predicted water removal volume for each prediction subject predicted by the trained AI, assuming that the difference between the predicted water removal volume for each prediction subject predicted by the trained AI and the water removal volume difference in the same unit time is within 12% (maximum 300 mL). A higher accuracy rate indicates a smaller error. A smaller MAE value indicates a smaller error. Therefore, if the accuracy rate is higher than the reference value, or if the MAPE or MAE is smaller than a predetermined reference value, the trained AI can be evaluated as being suitable for prediction. If the accuracy rate is higher than the reference value, or if the MAPE or MAE is larger than the predetermined reference value, the user can review the training dataset and adjust the training parameters for each algorithm, and then return to step S11 and train again. Acceptable accuracy rates are, for example, 80% or higher, 85% or higher, 88% or higher, or 90% or higher. An acceptable MAPE is, for example, 15% or less, preferably 10% or less, and more preferably 8% or less. In step S12, the trained artificial intelligence that has been verified is used as a water removal amount prediction support model.

[0076] 2. Prediction of water removal volume within a target time using a water removal volume prediction support model 2-1. Overview The method for supporting the advance prediction of the target time water removal volume by dialysis on the day of dialysis for a prediction subject (hereinafter, sometimes simply referred to as the "water removal volume prediction support method") can obtain the target time water removal volume (e.g., mL) as an output value by inputting the analysis dataset of each prediction subject into the water removal volume prediction support model constructed in 1-1(1) and 1-3 above. The following (i) and (ii), (i) to (iii), (i) and (ii) and (iv), or (i) to (iv) are used as the analysis dataset: (i) a first dialysis data set obtained from the subject of the prediction at least once immediately before the dialysis date; (ii) a profile dataset including a profile of the subject to be predicted and a profile of a dialysis machine to be used by the subject to be predicted; (iii) a second dialysis data set actually acquired for the subject on the dialysis day before the unit time for which the target time water removal volume is to be predicted; (iv) A specimen test dataset obtained from the subject of prediction immediately prior to the date of dialysis administration.

[0077] The first dialysis data set includes blood flow rate per unit time, fluid pressure of the dialysis fluid per unit time, amount of fluid removed per unit time, rate of change in circulating blood volume per specified time, fluid removal rate per unit time, systolic blood pressure and diastolic blood pressure per unit time, whether or not a drop in blood pressure occurred during dialysis per unit time, temperature of the dialysis fluid on the day of dialysis, target total amount of fluid removed determined by the doctor on the day of dialysis, dry weight on the day of dialysis, previous post-dialysis weight, weight before the start of dialysis on the day of dialysis, weight gain from before the start of dialysis on the day of dialysis based on the previous post-dialysis weight, gain from dry weight on the day of dialysis, post-dialysis weight on the day of dialysis, systolic blood pressure and diastolic blood pressure before the start of dialysis on the day of dialysis, standard dialysis time on the day of dialysis, and dialysis time on the day of dialysis.

[0078] The profile of the subject in (ii) above includes the dialysis start date, dialysis time period, gender, and age group of the subject, and the profile of the dialysis machine includes the type of dialysis method.

[0079] The second dialysis data set in (iii) above includes the dry weight acquired on the dialysis day for the prediction subject, the previous post-dialysis weight, the pre-dialysis weight on the dialysis day acquired on the dialysis day, the weight increase from the previous post-dialysis weight to the start of dialysis on the dialysis day, the blood flow rate at each unit time, the fluid pressure of the dialysate at each unit time, the amount of water removed at each unit time, the rate of change in circulating blood volume at each predetermined time, the water removal rate at each unit time, the systolic blood pressure and diastolic blood pressure at each unit time, and whether or not blood pressure has decreased at each unit time. If the support device 20 is directly connected to the dialysis device, the second dialysis data is acquired from the dialysis device connected to the support device 20. If the support device 20 is a server device, the second dialysis data is acquired from the dialysis device connected to the support device 20 via a network.

[0080] The second dialysis data set is not limited to the one actually acquired for each subject on the dialysis day before the unit time for which the target time water removal volume is predicted. For example, if the unit time for predicting the target time water removal volume is X, at least one set, preferably two or more sets, and more preferably all sets of data from the unit time (X-1) immediately preceding the unit time X to the first time when dialysis started are used for analysis. However, since some of the dialysis data sets for that day (water removal volume per unit time, blood flow per unit time, dialysate pressure per unit time, systolic blood pressure and diastolic blood pressure per unit time, and whether or not blood pressure decreased per unit time) are not available before the first unit time on the prediction day, this data is not used in the analysis.

[0081] The specimen test dataset described in (iv) above includes the most recent biochemical test and blood test data.

[0082] For other data sets and explanations of the data, the explanation in 1-1.(1) above is incorporated herein by reference.

[0083] Here, the number of sets of dialysis data sets is at least one set immediately prior to the date of dialysis, and preferably five sets immediately prior to the date of dialysis. A dialysis data set is acquired for each dialysis session, but specimen test data is acquired approximately twice a month, so one set of specimen test data is acquired for every five dialysis data sets. The profile data set does not change in principle except for the age group.

[0084] 2-2.Water removal volume prediction support device that supports advance prediction of target time water removal volume by dialysis 2-2-1. Configuration of the support device 20 FIG. 3 shows the hardware configuration of a water removal amount prediction support device 20 (hereinafter simply referred to as "support device 20") that supports advance prediction of the target time water removal amount by dialysis on the day of dialysis for a prediction subject.

[0085] The assistance device 20 may be connected to an input device 211 and an output device 212 .

[0086] In the support device 20, a CPU 201, a memory 202, a ROM 203, a storage device 204, and an interface 206 are connected to each other via a bus 209 so as to be able to communicate data with each other. The processing unit 201, the memory 202, and the ROM 203 function as a control unit 200 of the support device 20.

[0087] Each component of the support device 20 is similar to the corresponding component of the training device 10 except for the configuration of the storage device 204 .

[0088] The support device 20 may be directly connected to the dialysis machine 50. "Directly connected" refers to a form in which the support device 20 and the dialysis machine 50 are communicatively connected by wire or wirelessly so as to enable real-time communication, or are integrated.

[0089] The storage device 204 non-volatilely stores an operating system (OS) 2041, a water removal amount prediction support program 2042 (hereinafter simply referred to as the "support program 2042") that supports advance prediction of the target time water removal amount by dialysis on the dialysis execution date for a prediction subject described below, a model database 2043, and an analysis database 2044. The model database 2043 stores a water removal amount prediction support model. Furthermore, the analysis database 2044 can store an analysis dataset.

[0090] 2-2-2. Processing under the Support Program 2024 4 shows the flow of processing performed by the assistance program 2042. The control unit 200 executes the assistance program 2042, causing the computer to function as the assistance device 20.

[0091] The control unit 200 receives a processing start request input by the user from the input device 211, for example, and executes the assistance program 2042 to start the assistance processing.

[0092] In step S21, the control unit 200 reads out the water removal amount prediction support model and the analysis data set of the prediction target from the model database 2043, and inputs the analysis data set of the prediction target into the water removal amount prediction support model.

[0093] In step S22, the control unit 200 outputs the predicted value output from the water removal amount prediction support model to the output device 212 as the target time water removal amount for the person to be predicted.

[0094] 2-3. Support system for predicting the target time for fluid removal during dialysis 2-3-1. Configuration of the Support System 1000 5 shows an overview of a water removal volume prediction support system 1000 (hereinafter simply referred to as "support system 1000") that supports advance prediction of the target time water removal volume by dialysis on the day of dialysis for a prediction subject. In the support system 1000, a water removal volume prediction support device 30 (hereinafter simply referred to as "support device 30") that supports advance prediction of the target time water removal volume by dialysis on the day of dialysis for a prediction subject and a user terminal 40 are communicably connected via a network 99.

[0095] (1) Configuration of the support device 30 6 shows the hardware configuration of the support device 30. The support device 30 is a server device.

[0096] The assistance device 30 may be connected to an input device 311 and an output device 312 .

[0097] In the support device 30, a CPU 301, a memory 302, a ROM 303, a storage device 304, and an interface 306 are connected to each other via a bus 309 so as to be able to communicate data with each other. The processing unit 301, the memory 302, and the ROM 303 function as a control unit 300 of the support device 30.

[0098] Each component of the support device 30 is similar to the corresponding component of the training device 10 except for the fact that it is connected to a network 99 and the configuration of the storage device 304 .

[0099] The storage device 304 non-volatilely stores an operating system (OS) 3041, a water removal amount prediction support program 3042 (hereinafter simply referred to as the "support program 3042") that supports advance prediction of the target time water removal amount by dialysis on the dialysis execution date for a prediction subject described below, a model database 3043, and an analysis database 3044. The processing by the support program 3042 is similar to that of the support program 2042. The model database 3043 stores a water removal amount prediction support model. Furthermore, the analysis database 3044 can store an analysis dataset.

[0100] Here, the prediction of the target time water removal amount in the support system 1000 is performed by the support device 30. For this reason, the analysis database 4044 stores any of the analysis data sets (i) and (ii), (i) to (iii), (i), (ii) and (iv), (iii), (iv), (iii) and (iv), or (i) to (iv) in a non-volatile or volatile manner.

[0101] (2) Configuration of user terminal 40 FIG. 7 shows the configuration of the user terminal 40. The user terminal 40 may be connected to an input device 411 and an output device 412. The input device 411 and the output device 412 may be integrated into the user terminal 40. In other words, the user terminal may be a desktop computer, a laptop computer, a tablet, etc.

[0102] In the user terminal 40, the CPU 401, memory 402, ROM 403, storage device 404, and interface 406 are connected to each other via a bus 409 so as to be able to communicate data with each other. The processing unit 401, memory 402, and ROM 403 function as a control unit 400 of the user terminal 40.

[0103] The storage device 404 stores, in a non-volatile or volatile manner, an operating system (OS) 4041, user terminal software 4042, and an analysis database 4044. The analysis database 4044 can store analysis datasets.

[0104] 2-3-2. Processing flow of the support system 1000 FIG. 8 shows the processing flow of the support system 1000.

[0105] In step S41, the control unit 400 of the user terminal 40 accepts a request to transmit the analysis dataset of the predicted subject input by the user from the input device 411, for example, and transmits the analysis dataset of the predicted subject to the support device 30. The transmission request is input via the user terminal software.

[0106] In step S51, the control unit 300 of the assistance device 30 receives the analysis data set transmitted by the user terminal 40. This reception serves as a trigger, or upon receiving a processing start command input by the operator from the input device 311, the control unit 300 executes the assistance program 3042 and starts the assistance processing.

[0107] In step S52, the control unit 300 of the support device 30 executes the support program 3042 and starts the support process. In step S52, the control unit 300 of the support device 30 reads out the water removal amount prediction support model from the model database 3043 and inputs the analysis data set of the prediction target person into the water removal amount prediction support model. The control unit 300 of the support device 30 outputs a predicted value from the water removal amount prediction support model.

[0108] In step S53, the control unit 300 of the support device 30 outputs the predicted value output from the water removal amount prediction support model to the user terminal 40 as the target time water removal amount of the person to be predicted.

[0109] In step S42, the control unit 400 of the user terminal 40 receives the target time water removal amount output from the support device 30. In step S43, the control unit 400 of the user terminal 40 outputs the target time water removal amount received from the assistance device 30 to the output device 412 via the user terminal software 4042.

[0110] In the above, steps S41, S42, and S43 are processes performed by the user terminal software 4042, and steps S51, S52, and S53 are processes performed by the assistance program 3042.

[0111] The analysis datasets transmitted in step S41 may be analysis datasets (i) to (iii), (i), (ii), and (iv), or all of (i) to (iv). Alternatively, for example, if analysis datasets (i) and (ii) are pre-stored in the analysis database 3044 of the support device 30, only analysis dataset (iii), only (iv), or only (iii) and (iv) may be transmitted. Alternatively, if analysis datasets (i), (ii), and (iv) are pre-stored in the analysis database 3044, only analysis dataset (iii) is transmitted. In other words, at least analysis dataset (iii) is transmitted via the network in step S41. The analysis datasets (iii) and (iv) may be transmitted, for example, from a terminal of an electronic medical record system serving as the user terminal 40. The analysis dataset (iii) may also be a dataset output from a dialysis machine to the user terminal 40. The analysis datasets (iii) and (iv) may also be manually input by the user.

[0112] II. Prediction of blood pressure drop due to dialysis per unit time 1. A support model that helps predict the occurrence of blood pressure drops due to dialysis per unit time One embodiment of the present invention relates to a support model (hereinafter simply referred to as a "blood pressure decrease occurrence prediction support model") that supports advance prediction of the occurrence of blood pressure decrease due to dialysis for each unit time on a dialysis day for a prediction subject.

[0113] The explanation of "unit time" is given in I.1 above.

[0114] First, a training method for training an artificial intelligence to function as a blood pressure reduction occurrence support model will be described.

[0115] 1-1. Overview (1) Creation of a blood pressure drop prediction support model The explanation of the artificial intelligence that constitutes the water removal volume prediction support model, the explanation of each data set, the training method, and the individualization method are incorporated herein by reference in the explanation of I.1-1 above. However, "predicting water removal volume" should be read as "whether or not blood pressure drop will occur," and the term "water removal volume prediction support model" should be read as "blood pressure drop prediction support model." Furthermore, in training the water removal volume prediction support model, the correct answer label is "water removal volume per unit time in each past dialysis session for each patient," but in this embodiment, a label indicating "whether or not blood pressure drop will occur per unit time in each past dialysis session for each patient" is used. The label may be text such as "yes" or "no," or may be a number such as "1" or "0."

[0116] 1-2. Support model training device 60 Figure 9 shows the hardware configuration of a training device 60 (hereinafter simply referred to as "training device 60") for a blood pressure decrease occurrence prediction support model that supports advance prediction of the occurrence of blood pressure decrease due to dialysis for each unit time on the dialysis day for a prediction subject. The training device 60 may be connected to an input device 611 and an output device 612 .

[0117] In the training device 60, a processing unit (CPU) 601, a memory 602, a ROM (read only memory) 603, a storage device 604, and an interface 606 are connected to each other via a bus 609 so as to be able to communicate data with each other. The processing unit 601, the memory 602, and the ROM 603 function as a control unit 600 of the training device 60.

[0118] The storage device 604 non-volatilely stores an operating system (OS) 6041, a training program 6042 (hereinafter simply referred to as the "training program 6042") for training an artificial intelligence that constitutes a blood pressure decrease occurrence prediction support model described below, and a model database 6043. The model database 6043 stores the artificial intelligence before training, the blood pressure decrease occurrence prediction support model constituted by the trained artificial intelligence, and / or the individualized blood pressure decrease occurrence prediction support model. The model database 6043 can also store a training dataset and an individual training dataset.

[0119] The hardware configuration of training device 60 is similar to the corresponding configuration of training device 10 , except for storage device 604 .

[0120] 1-3. Processing by training program 6042 10 shows the flow of processing performed by the training program 6042. The control unit 600 executes the training program 6042, causing the computer to function as the training device 60. The control unit 600 receives a processing start request input by the user from the input device 611, for example, and executes the training program 6042 to start the training processing.

[0121] Step S61 in Figure 10 is the same as step S11 except for the correct label. A trained AI is constructed by inputting the training data set into the AI according to the method described in II.1-1.(1) above and training the AI. The control unit 600 stores the trained AI in the storage device 604.

[0122] The control unit 600 receives a user command to start the verification process and verifies the trained AI. The verification can be performed using an index as to whether or not the occurrence of a blood pressure drop per unit time for each prediction subject matches the occurrence of a blood pressure drop for each prediction subject predicted by the trained AI. For example, the index can be an F score or an AUC. The F score is calculated using the following formula:

number

[0123] Here, recall indicates the percentage of actual blood pressure drops that the trained AI was able to predict, and precision indicates the percentage of blood pressure drops that actually occurred among the blood pressure drops predicted by the AI. AUC (area under the curve) refers to the area under the ROC curve, ranging from 0.5 to 1.0, with 1.0 indicating perfect prediction. ROC curves are abbreviations for receiver operating characteristic, and are plotted by plotting recall on the vertical axis and false positive rate on the horizontal axis, or recall on the vertical axis and precision on the horizontal axis. The evaluation of the trained AI's accuracy using F-scores and AUC is similar to the case of accuracy in I.1-3 above. The trained AI that has been verified in step S62 is used as a blood pressure drop prediction support model.

[0124] 2.Predicting the occurrence of hypotension using a hypotension prediction support model 2-1. Overview The method for supporting the advance prediction of blood pressure decrease due to dialysis for each unit time on the dialysis day for a prediction subject (hereinafter, sometimes simply referred to as the "blood pressure decrease prediction support method") inputs the analysis dataset for each prediction subject into the blood pressure decrease prediction support model constructed in II.1-1(1) and II.1-3 above, and can obtain a label indicating whether or not blood pressure decrease will occur for each unit time as an output value. The analysis dataset is the same as in I.2-1 above.

[0125] 2-2. Blood pressure drop prediction support device that supports advance prediction of blood pressure drop due to dialysis per unit time 2-2-1.Configuration of the blood pressure drop prediction support device Figure 11 shows the hardware configuration of a blood pressure drop occurrence prediction support device 70 (hereinafter simply referred to as the "support device 70") that supports advance prediction of the occurrence of blood pressure drop due to dialysis for each unit time on the dialysis day for a prediction subject.

[0126] The assistive device 70 may be connected to an input device 711 and an output device 712 .

[0127] In the support device 70, a CPU 701, a memory 702, a ROM 703, a storage device 704, and an interface 706 are connected to each other via a bus 709 so as to be able to communicate data with each other. The processing unit 701, the memory 702, and the ROM 703 function as a control unit 700 of the support device 70.

[0128] Each component of the support device 70 is the same as the corresponding component of the support device 20, except for the configuration of the storage device 704.

[0129] The support device 70 may be directly connected to the dialysis machine 50. "Directly connected" refers to a form in which the support device 70 and the dialysis machine 50 are communicatively connected by wire or wirelessly so as to enable real-time communication, or are integrated.

[0130] The storage device 704 non-volatilely stores an operating system (OS) 7041, a blood pressure decrease occurrence prediction support program 7042 (hereinafter simply referred to as the "support program 7042") that supports advance prediction of the occurrence of blood pressure decrease due to dialysis for each unit time on a dialysis performance day for a prediction subject described below, a model database 7043, and an analysis database 7044. The model database 7043 stores a blood pressure decrease occurrence prediction support model. Furthermore, the analysis database 7044 can store an analysis dataset.

[0131] 2-2-2. Processing by the support program 7024 12 shows the flow of processing performed by the assistance program 7042. The control unit 700 executes the assistance program 7042, causing the computer to function as the assistance device .

[0132] The control unit 700 receives a processing start request input by the user from the input device 711, for example, and executes the assistance program 7042 to start the assistance processing.

[0133] In step S71, the control unit 700 reads out the blood pressure decrease occurrence prediction support model and the analysis data set of the prediction target from the model database 7043, and inputs the analysis data set of the prediction target to the blood pressure decrease occurrence prediction support model.

[0134] In step S72, the control unit 700 outputs the label output from the blood pressure decrease occurrence prediction support model to the output device 712 as a label indicating whether or not the prediction subject will experience a decrease in blood pressure.

[0135] 2-3. Blood pressure drop prediction support system that supports advance prediction of blood pressure drop due to dialysis per unit time 2-3-1. Configuration of Support System 2000 13 shows an overview of a support system 2000 (hereinafter simply referred to as "support system 2000") that supports advance prediction of the occurrence of a drop in blood pressure due to dialysis for each unit time on a dialysis day for a prediction subject. In the support system 2000, a support device 80 (hereinafter simply referred to as "support device 80") that supports advance prediction of the occurrence of a drop in blood pressure due to dialysis for each unit time on a dialysis day for a prediction subject and a user terminal 40 are communicably connected via a network 99.

[0136] (1) Configuration of the support device 80 FIG. 14 shows the hardware configuration of the support device 80.

[0137] The assistive device 80 may be connected to an input device 811 and an output device 812 .

[0138] In the support device 80, a CPU 801, a memory 802, a ROM 803, a storage device 804, and an interface 806 are connected to one another via a bus 809 so as to be able to communicate data with one another.

[0139] The hardware configuration of the support device 80 is the same as that of the support device 30, except for the storage device 804. The support device 80 is a server device.

[0140] The storage device 804 non-volatilely stores an operating system (OS) 8041, an assistance program 8042 (hereinafter simply referred to as the "assistance program 8042") that supports advance prediction of the occurrence of blood pressure decrease due to dialysis for each unit time on a dialysis day for a prediction subject described below, a model database 8043, and an analysis database 8044. The processing by the assistance program 8042 is similar to that of the assistance program 7042. The model database 8043 stores a blood pressure decrease occurrence prediction assistance model. Furthermore, the analysis database 8044 can store an analysis dataset.

[0141] Here, the prediction of the occurrence of a drop in blood pressure in the support system 2000 is performed by the support device 80. For this reason, the analysis database 8044 stores any of the analysis data sets (i) and (ii), (i) to (iii), (i), (ii) and (iv), (iii), (iv), (iii) and (iv), or (i) to (iv) in a non-volatile or volatile manner.

[0142] (2) Configuration of user terminal 40 The user terminal 40 is as described above in I.2-3-1.

[0143] 2-3-2. Processing flow of the support system 2000 FIG. 15 shows the processing flow of the support system 2000.

[0144] In step S411, the control unit 400 of the user terminal 40 accepts a request to send the analysis dataset of the predicted subject input by the user from the input device 411, for example, and sends the analysis dataset of the predicted subject to the support device 80. The transmission request is input via the user terminal software.

[0145] In step S81, the control unit 800 of the assistance device 80 receives the analysis data set transmitted by the user terminal 40. This reception serves as a trigger, or upon receiving a processing start command input by the operator from the input device 811, the control unit 800 executes the assistance program 8042 and starts the assistance processing.

[0146] In step S82, the control unit 800 of the support device 80 executes the support program 8042 and starts the support processing. In step S82, the control unit 800 of the support device 80 reads out the blood pressure decrease occurrence prediction support model from the model database 8043, and inputs the analysis data set of the prediction subject into the blood pressure decrease occurrence prediction support model. The control unit 800 of the support device 80 outputs a prediction label indicating whether or not blood pressure decrease due to dialysis will occur from the blood pressure decrease occurrence prediction support model. The label may be text such as "yes" or "no," or may be a number such as "1" or "0."

[0147] In step S83, the control unit 800 of the support device 80 outputs the predicted value output from the blood pressure decrease occurrence prediction support model to the user terminal 40 as the target time water removal amount of the person to be predicted.

[0148] In step S412, the control unit 400 of the user terminal 40 receives the predicted label output from the assistance device 80. In step S413, the control unit 400 of the user terminal 40 converts the predicted label received from the assistance device 30 or the predicted label into text and outputs it to the output device 412 via the user terminal software 4042.

[0149] In the above, steps S411, S412, and S413 are processes performed by the user terminal software 4042, and steps S81, S82, and S83 are processes performed by the assistance program 8042.

[0150] The analysis datasets transmitted in step S411 may be analysis datasets (i) to (iii), (i), (ii), and (iv), or all of (i) to (iv). Alternatively, for example, if analysis datasets (i) and (ii) are stored in advance in the analysis database 8044 of the support device 80, only analysis dataset (iii), only (iv), or only (iii) and (iv) may be transmitted. Alternatively, if analysis datasets (i), (ii), and (iv) are stored in advance in the analysis database 8044, only analysis dataset (iii) is transmitted. In other words, at least analysis dataset (iii) is transmitted via the network in step S411. The analysis datasets (iii) and (iv) may be transmitted, for example, from a terminal of an electronic medical record system serving as the user terminal 40. The analysis dataset (iii) may also be a dataset output from a dialysis machine to the user terminal 40. The analysis datasets (iii) and (iv) may also be manually input by the user.

[0151] III. Storage medium storing the program One embodiment of the present invention relates to a program product, such as a media drive, that stores training programs 1042, 6042, assistance programs 2042, 3042, 7042, 8042, user terminal software 4042, and / or retraining programs. Specifically, the training programs 1042, 6042, assistance programs 2042, 3042, 7042, 8042, user terminal software 4042, and / or retraining programs may be stored on a media drive, such as a hard disk, a semiconductor memory device such as a flash memory, or an optical disk. The media drive may also be a computer, such as a server device. The format of the program recording on the media drive is not limited as long as each device can read the program. It is preferable that the recording on the media drive be non-volatile.

[0152] IV. Variations In the above, the unit time is shown as a time interval from the start of dialysis, but it may also be shown as a time of day.

[0153] If the number of dialysis sessions for the person being predicted increases, the individualized water removal volume prediction support model or the individualized blood pressure decrease occurrence prediction support model may be retrained by adding a new individual training dataset containing the data set for the increased number of dialysis sessions, or by updating the individual training dataset.

[0154] V. Verification of effectiveness 1. Dataset A dialysis dataset, a specimen test dataset, and a profile dataset were obtained from each patient undergoing dialysis.

[0155] Next, data from patients with few dialysis sessions and patients with missing data were excluded, and ultimately data from 138,000 dialysis sessions was used for analysis. The data was then divided into a learning (training) dataset, a testing dataset, a calibration dataset, and an evaluation dataset, and used to train the artificial intelligence and evaluate its accuracy.

[0156] The dialysis dataset used for the analysis is shown in Figure 16. The analysis dataset (i) used was the five dialysis datasets most recent to the day of dialysis for the prediction subject. Specifically, the datasets included the dialysate temperature, pre-dialysis blood pressure, dry weight, previous post-dialysis weight, pre-dialysis weight on the day of dialysis, weight gain, post-dialysis weight on the day of dialysis, fluid removal volume per unit time, systolic and diastolic blood pressure per unit time, and whether or not a decrease in blood pressure occurred during dialysis per unit time. The analysis dataset (ii) used patient attributes, day of the week, and number of days since the first dialysis. The analysis dataset (iii) used the dialysate temperature, pre-dialysis blood pressure, dry weight, previous post-dialysis weight, pre-dialysis weight, weight gain, blood flow per unit time, dialysate pressure per unit time, fluid removal volume per unit time, systolic and diastolic blood pressure at each unit time, and whether or not a decrease in blood pressure occurred at each unit time on the day of dialysis. A decrease in blood pressure during dialysis was defined as a decrease in systolic blood pressure of 20 mmHg or more from the previous measurement and 110 mmHg or less at the time of decrease.

[0157] 2.Prediction accuracy Figure 17 shows the accuracy of predictions of the target time water removal volume using dialysis data set stretching back from just before the unit time being predicted to the start of dialysis on the day of dialysis. Good prediction accuracy was observed from the second hour onwards. Next, a dialysis dataset for a unit time before the unit time to be predicted but not immediately before the unit time to be predicted was used as the second dialysis dataset for prediction. Specifically, the dialysis data acquired in the first unit time (0 to 1 hour) was used to predict the amount of water removed after 3 hours, 4 hours, and 5 hours. The dialysis data acquired in the second unit time (1 to 2 hours) was used to predict the amount of water removed after 4 hours and 5 hours. The dialysis data acquired in the third unit time (2 to 3 hours) was used to predict the amount of water removed after 5 hours. The results are shown in Figure 18. Even when using a dialysis dataset for a unit time not immediately before the unit time to be predicted, the amount of water removed could be predicted with high accuracy. It was also shown that the more dialysis datasets acquired on the day of dialysis, the higher the accuracy. Figure 19 shows the prediction accuracy when predicting the occurrence of a drop in blood pressure in a unit time period using dialysis data set stretching back from just before the unit time to be predicted to the start of dialysis on the day of dialysis. Prediction accuracy was evaluated using AUC. As time passes, the amount of data available for prediction increases, and so prediction accuracy improves. [Explanation of symbols]

[0158] 20,70 Support equipment 200,700 Control section 30,80 Support equipment 300,800 Control section 40 user terminals 1000,2000 Support System

Claims

1. An assistance device that assists in the advance prediction of a target time water removal amount by dialysis on a dialysis implementation day or the occurrence of blood pressure decrease due to dialysis per unit time for a prediction subject, wherein the target time water removal amount is a target amount of water removal per unit time after the start of dialysis to be predicted on the dialysis implementation day, and the assistance device is directly connected to a dialysis machine; The assistance device includes a control unit, The control unit inputs the following analysis data sets (i) and (ii), (i) to (iii), (i), (ii), and (iv), or (i) to (iv) into an assistance model configured by artificial intelligence, and outputs a label indicating the target time water removal amount on the dialysis performance day or whether or not a decrease in blood pressure has occurred per unit time. (i) a first dialysis data set of at least one previous dialysis session immediately preceding the dialysis date obtained from the subject; (ii) a profile dataset including a profile of the predicted subject and a profile of a dialysis machine that the predicted subject will use on a dialysis performance date; (iii) A second dialysis data set that is actually acquired for the prediction subject on the dialysis execution date before the unit time that is the target time water removal amount prediction target, or before the unit time that predicts the occurrence of blood pressure decrease due to dialysis, wherein the second dialysis data set is the second dialysis data set acquired from a dialysis machine connected to the assistance device; (iv) A specimen test dataset obtained from the subject of prediction most recently prior to the date of dialysis administration.

2. The support model is The assistance device according to claim 1, which is trained using the following training datasets (A) and (B), or (A), (B), and (C) obtained from a plurality of dialysis patients: (A) a dialysis dataset obtained from each of the plurality of dialysis patients; (B) a profile data set including a profile of each of the plurality of dialysis patients and a profile of a dialysis machine used by each of the patients; (C) A specimen test dataset obtained from each of the plurality of dialysis patients.

3. the first dialysis data set includes blood flow rate per unit time, fluid pressure of the dialysis fluid per unit time, fluid removal volume per unit time, rate of change in circulating blood volume per predetermined time, fluid removal rate per unit time, systolic blood pressure and diastolic blood pressure per unit time, whether or not a decrease in blood pressure occurred during dialysis per unit time, temperature of the dialysis fluid on the day of dialysis, target total fluid removal volume determined by a doctor on the day of dialysis, dry weight on the day of dialysis, previous value of post-dialysis weight, weight before the start of dialysis on the day of dialysis, weight increase from the previous value of post-dialysis weight to the start of dialysis on the day of dialysis, increase from dry weight on the day of dialysis, post-dialysis weight on the day of dialysis, systolic blood pressure and diastolic blood pressure before the start of dialysis on the day of dialysis, standard dialysis time on the day of dialysis, and dialysis time on the day of dialysis; The profile of the subject described in (ii) includes a dialysis start date, a dialysis time slot, a gender, and an age group of the subject, and the profile of the dialysis device includes a type of dialysis method; The second dialysis data set described in (iii) includes the dry weight obtained for the prediction subject on the dialysis day, the previous value of the post-dialysis weight, the pre-dialysis weight on the dialysis day obtained on the dialysis day, the weight increase amount up to the start of dialysis on the dialysis day based on the previous value of the post-dialysis weight, the blood flow rate at each unit time, the fluid pressure of the dialysis fluid at each unit time, the amount of water removed at each unit time, the rate of change in circulating blood volume at each predetermined time, the water removal speed at each unit time, the systolic blood pressure and diastolic blood pressure at each unit time, and whether or not a decrease in blood pressure has occurred at each unit time, The specimen test dataset described in (iv) includes biochemical test data and blood test data, The support device according to claim 1 .

4. The dialysis dataset in (A) includes blood flow rate per unit time, fluid pressure of the dialysis fluid per unit time, fluid removal volume per unit time, rate of change in circulating blood volume per predetermined time, fluid removal rate per unit time, systolic blood pressure and diastolic blood pressure per unit time, presence or absence of blood pressure decrease during dialysis per unit time, temperature of the dialysis fluid on the day of dialysis, target total fluid removal volume determined by a doctor on the day of dialysis, dry weight on the day of dialysis, previous value of post-dialysis weight, weight before the start of dialysis on the day of dialysis, weight increase from the previous value of post-dialysis weight to the start of dialysis on the day of dialysis, increase from dry weight on the day of dialysis, post-dialysis weight on the day of dialysis, systolic blood pressure and diastolic blood pressure before the start of dialysis on the day of dialysis, standard dialysis time on the day of dialysis, and dialysis time on the day of dialysis. The profile of each of the plurality of dialysis patients in (B) includes a dialysis start date, a dialysis time period, a gender, and an age group of each dialysis patient, and the profile of the dialysis machine includes a type of dialysis method; The support device according to claim 2 , wherein the specimen test data set described in (C) includes biochemical test data and blood test data.

5. 2. The assistance device according to claim 1, wherein the target amount of water removal per unit time immediately after the start of dialysis or the occurrence of a drop in blood pressure per unit time is predicted by inputting the analysis data sets of (i) and (ii) into the assistance model.

6. The assistance device according to claim 1 , wherein the target time water removal amount or the occurrence of a blood pressure decrease per unit time is predicted for a plurality of unit times.

7. 2. The assistance device according to claim 1, wherein the artificial intelligence constituting the assistance model is an algorithm having a structure other than a neural network structure, or an algorithm having a neural network structure.

8. The algorithm having a structure other than a neural network structure is a gradient boosting algorithm or a random forest algorithm; 8. The assistance device according to claim 7, wherein the algorithm having a neural network structure is a Dual-Channel Combiner Network algorithm or a Long Short-Term Memory algorithm.

9. A method for assisting in the advance prediction of a target time water removal amount by dialysis on a dialysis implementation day or the occurrence of blood pressure reduction due to dialysis per unit time for a prediction subject, wherein the target time water removal amount is a target amount of water removal per unit time after the start of dialysis to be predicted on the dialysis implementation day; The method is assisted by an assistance model comprised of artificial intelligence; The method, wherein the assistance model is configured to input the following analysis datasets (i) and (ii), (i) to (iii), (i), (ii), and (iv), or (i) to (iv) into the assistance model, and output a label indicating the target time water removal amount on the dialysis performance day or whether or not a decrease in blood pressure has occurred per unit time: (i) a first dialysis data set of at least one previous dialysis session immediately preceding the dialysis date obtained from the subject; (ii) a profile dataset including a profile of the predicted subject and a profile of a dialysis machine that the predicted subject will use on a dialysis performance date; (iii) A second dialysis data set actually acquired for the prediction subject on the dialysis performance date before the unit time for which the target time water removal amount is predicted or before the unit time for which the occurrence of blood pressure decrease due to dialysis is predicted, wherein the second dialysis data set is acquired from a dialysis machine connected to the assistance device according to claim 1; (iv) A specimen test dataset obtained from the subject of prediction most recently prior to the date of dialysis administration.

10. When you run it on a computer, 1. A support program for supporting advance prediction of a target amount of water removal per unit time by dialysis on a dialysis day or the occurrence of a decrease in blood pressure per unit time by dialysis for a subject, the support program executing a step of inputting the following analysis datasets (i) and (ii), (i) to (iii), (i), (ii), and (iv), or (i) to (iv) into a support model configured with artificial intelligence, and outputting a label indicating the target amount of water removal per unit time on a dialysis day or the occurrence of a decrease in blood pressure per unit time, wherein the target amount of water removal per unit time is a target amount of water removal per unit time after the start of dialysis that is to be predicted on the dialysis day: (i) a first dialysis data set of at least one previous dialysis session immediately preceding the dialysis date obtained from the subject; (ii) a profile dataset including a profile of the predicted subject and a profile of a dialysis machine that the predicted subject will use on a dialysis performance date; (iii) A second dialysis data set actually acquired for the prediction subject on the dialysis performance date before the unit time for which the target time water removal amount is predicted or before the unit time for which the occurrence of blood pressure decrease due to dialysis is predicted, wherein the second dialysis data set is acquired from a dialysis machine connected to the assistance device according to claim 1; (iv) A specimen test dataset obtained from the subject of prediction most recently prior to the date of dialysis administration.

11. An assistance device that assists in the advance prediction of a target time water removal amount by dialysis on a dialysis implementation day or the occurrence of blood pressure decrease due to dialysis per unit time for a prediction subject, wherein the target time water removal amount is a target amount of water removal per unit time after the start of dialysis to be predicted on the dialysis implementation day, and the assistance device is a server device; The assistance device includes a control unit, The control unit inputs the following analysis data sets (i) and (ii), (i) to (iii), (i), (ii), and (iv), or (i) to (iv) into an assistance model configured by artificial intelligence, and outputs a label indicating the target time water removal amount on the dialysis performance day or whether or not a decrease in blood pressure has occurred per unit time. (i) a first dialysis data set of at least one previous dialysis session immediately preceding the dialysis date obtained from the subject; (ii) a profile dataset including a profile of the predicted subject and a profile of a dialysis machine that the predicted subject will use on a dialysis performance date; (iii) A second dialysis data set actually acquired for the prediction subject on the dialysis execution date before the unit time for which the target time water removal amount is predicted or before the unit time for which the occurrence of blood pressure decrease due to dialysis is predicted, and the second dialysis data set is transmitted to the server device via a network; (iv) A specimen test dataset obtained from the subject of prediction most recently prior to the date of dialysis administration.

12. The support model is The assistance device according to claim 11, which is trained using the following training datasets (A) and (B), or (A), (B), and (C) obtained from a plurality of dialysis patients: (A) a dialysis dataset obtained from each of the plurality of dialysis patients; (B) a profile data set including a profile of each of the plurality of dialysis patients and a profile of a dialysis machine used by each of the patients; (C) A specimen test dataset obtained from each of the plurality of dialysis patients.

13. the first dialysis data set includes blood flow rate per unit time, fluid pressure of the dialysis fluid per unit time, fluid removal volume per unit time, rate of change in circulating blood volume per predetermined time, fluid removal rate per unit time, systolic blood pressure and diastolic blood pressure per unit time, whether or not a decrease in blood pressure occurred during dialysis per unit time, temperature of the dialysis fluid on the day of dialysis, target total fluid removal volume determined by a doctor on the day of dialysis, dry weight on the day of dialysis, previous value of post-dialysis weight, weight before the start of dialysis on the day of dialysis, weight increase from the previous value of post-dialysis weight to the start of dialysis on the day of dialysis, increase from dry weight on the day of dialysis, post-dialysis weight on the day of dialysis, systolic blood pressure and diastolic blood pressure before the start of dialysis on the day of dialysis, standard dialysis time on the day of dialysis, and dialysis time on the day of dialysis; The profile of the subject described in (ii) includes a dialysis start date, a dialysis time slot, a gender, and an age group of the subject, and the profile of the dialysis device includes a type of dialysis method; The second dialysis data set described in (iii) includes the dry weight obtained for the prediction subject on the dialysis day, the previous value of the post-dialysis weight, the pre-dialysis weight on the dialysis day obtained on the dialysis day, the weight increase amount up to the start of dialysis on the dialysis day based on the previous value of the post-dialysis weight, the blood flow rate at each unit time, the fluid pressure of the dialysis fluid at each unit time, the amount of water removed at each unit time, the rate of change in circulating blood volume at each predetermined time, the water removal speed at each unit time, the systolic blood pressure and diastolic blood pressure at each unit time, and whether or not a decrease in blood pressure has occurred at each unit time, The specimen test dataset described in (iv) includes biochemical test data and blood test data, The assistance device according to claim 11.

14. The dialysis dataset in (A) includes blood flow rate per unit time, fluid pressure of the dialysis fluid per unit time, fluid removal volume per unit time, rate of change in circulating blood volume per predetermined time, fluid removal rate per unit time, systolic blood pressure and diastolic blood pressure per unit time, presence or absence of blood pressure decrease during dialysis per unit time, temperature of the dialysis fluid on the day of dialysis, target total fluid removal volume determined by a doctor on the day of dialysis, dry weight on the day of dialysis, previous value of post-dialysis weight, weight before the start of dialysis on the day of dialysis, weight increase from the previous value of post-dialysis weight to the start of dialysis on the day of dialysis, increase from dry weight on the day of dialysis, post-dialysis weight on the day of dialysis, systolic blood pressure and diastolic blood pressure before the start of dialysis on the day of dialysis, standard dialysis time on the day of dialysis, and dialysis time on the day of dialysis. The profile of each of the plurality of dialysis patients in (B) includes a dialysis start date, a dialysis time period, a gender, and an age group of each dialysis patient, and the profile of the dialysis machine includes a type of dialysis method; The support device according to claim 12 , wherein the specimen test data set described in (C) includes biochemical test data and blood test data.

15. The assistance device according to claim 11, wherein the target time water removal volume in a unit time immediately after the start of dialysis is predicted by inputting the analysis data sets of (i) and (ii) into the assistance model.

16. The assistance device according to claim 11 , wherein the target time water removal amount is predicted for a plurality of unit times.

17. The assistance device according to claim 11 , wherein the assistance model is an algorithm having a structure other than a neural network structure, or an algorithm having a neural network structure.

18. The algorithm having a structure other than a neural network structure is a gradient boosting algorithm or a random forest algorithm; 18. The assistance device according to claim 17, wherein the algorithm having a neural network structure is a Dual-Channel Combiner Network algorithm or a Long Short-Term Memory algorithm.

19. A method for assisting in the advance prediction of a target time water removal amount by dialysis on a dialysis implementation day or the occurrence of blood pressure reduction due to dialysis per unit time for a prediction subject, wherein the target time water removal amount is a target amount of water removal per unit time after the start of dialysis to be predicted on the dialysis implementation day; The method is assisted by an assistance model comprised of artificial intelligence; The method, wherein the assistance model is configured to input the following analysis datasets (i) and (ii), (i) to (iii), (i), (ii), and (iv), or (i) to (iv) into the assistance model, and output a label indicating the target time water removal amount on the dialysis performance day or whether or not a decrease in blood pressure has occurred per unit time: (i) a first dialysis data set of at least one previous dialysis session immediately preceding the dialysis date obtained from the subject; (ii) a profile dataset including a profile of the predicted subject and a profile of a dialysis machine that the predicted subject will use on a dialysis performance date; (iii) A second dialysis data set actually acquired for the prediction subject on the dialysis execution date before the unit time for which the target time water removal amount is predicted or before the unit time for which the occurrence of blood pressure decrease due to dialysis is predicted, and the second dialysis data set is transmitted to the server device via a network; (iv) A specimen test dataset obtained from the subject of prediction most recently prior to the date of dialysis administration.

20. When you run it on a computer, 1. A support program for supporting advance prediction of a target amount of water removal per unit time by dialysis on a dialysis day or the occurrence of a decrease in blood pressure per unit time by dialysis for a subject, the support program executing a step of inputting the following analysis datasets (i) and (ii), (i) to (iii), (i), (ii), and (iv), or (i) to (iv) into a support model configured with artificial intelligence, and outputting a label indicating the target amount of water removal per unit time on a dialysis day or the occurrence of a decrease in blood pressure per unit time, wherein the target amount of water removal per unit time is a target amount of water removal per unit time after the start of dialysis that is to be predicted on the dialysis day: (i) a first dialysis data set of at least one previous dialysis session immediately preceding the dialysis date obtained from the subject; (ii) a profile dataset including a profile of the predicted subject and a profile of a dialysis machine that the predicted subject will use on a dialysis performance date; (iii) A second dialysis data set actually acquired for the prediction subject on the dialysis execution date before the unit time for which the target time water removal amount is predicted or before the unit time for which the occurrence of blood pressure decrease due to dialysis is predicted, and the second dialysis data set is transmitted to the server device via a network; (iv) A specimen test dataset obtained from the subject of prediction most recently prior to the date of dialysis administration.

21. A support system for supporting advance prediction of a target time water removal amount by dialysis on a dialysis implementation day or the occurrence of blood pressure decrease due to dialysis per unit time for a prediction subject, wherein the target time water removal amount is a target amount of water removal per unit time after the start of dialysis to be predicted on the dialysis implementation day, The assistance system includes an assistance device that is a server device and a user terminal used by a user, The assistance device includes a control unit, the control unit receives analysis datasets of the following (i) and (ii), (i) to (iii), (i), (ii), and (iv), or (i) to (iv), inputs the analysis datasets into an assistance model configured by artificial intelligence, and outputs a label indicating the target time water removal amount on the dialysis performance day or whether or not a decrease in blood pressure has occurred per unit time, (i) a first dialysis data set of at least one previous dialysis session immediately preceding the dialysis date obtained from the subject; (ii) a profile dataset including a profile of the predicted subject and a profile of a dialysis machine that the predicted subject will use on a dialysis performance date; (iii) A second dialysis data set actually acquired for the prediction subject on the dialysis implementation date before the unit time for which the target time water removal amount is predicted, and transmitted from a user terminal to the support device via a network; (iv) A specimen test dataset obtained from the subject of prediction most recently prior to the date of dialysis administration.

Images