Healthcare element evaluation support system

Abstract

To provide a healthcare element evaluation support system for extracting data items that affect determination from among a wide variety of data items, for specific individuals having various backgrounds.SOLUTION: In a healthcare element evaluation support system, a learning unit edits and manages a plurality of data groups composed of data having three or more data items belonging to three or more kinds of data item groups including a determination data item group and an inspection data item group, to construct a database, and learns data of the database through a neural network. An inference unit infers, using the trained neural network, the changes in an inference value of a specific data item serving as an index and set in an output layer, a quantified degree of influence indicating influence of a data item set in an input layer on the specific data item serving as an index, and an inference value of the specific data item serving as an index, set in the output layer by changing a data value of the data item set in the input layer.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a health management element evaluation support system for estimating the degree of influence of each data on judgment data and the change in judgment data caused by the change of each data in a data group including judgment data, inspection data, history data, lifestyle data, etc., which is linked to a specific individual for the purpose of maintaining or improving the health status of the specific individual.

Background Art

[0002] Conventionally, techniques for judging and evaluating an individual's health status by performing related data analysis using artificial intelligence have been studied. For example, in a data selection device, a learning device, and a program (see, for example, Patent Document 1) and an inspection system and an interview system (see, for example, Patent Document 2), techniques for accurately diagnosing a health status or a disease state are shown. These techniques are systems for obtaining correct judgment results, such as improving the reliability of judgment results by narrowing down data items used in the study.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, these systems are aimed at making accurate determinations and make determinations by selecting data items that have a high correlation with the determination. Therefore, data items with an unknown correlation are excluded from the objects to be considered for the influence on the determination, and there is a problem that even data items that actually affect the determination result are not considered as data items that affect the determination. For example, even if there are various pieces of information linked to an individual, regarding whether there is information that should be noted to improve the health status or disease state among the information whose influence on the health status and disease state is unknown, the known systems cannot examine this problem.

[0005] Therefore, the present invention is a technology for extracting data items that affect a determination from a wide variety of data items in a specific individual having various backgrounds. For example, it is a technology that can organize information that should be noted to improve the health status or disease state according to the individual from various pieces of information related to the health and life linked to the individual. In order to improve and optimize "specific data items serving as indicators" such as the stage of a disease which is determination data, when a group of data items such as the physical and health status, lifestyle habits, and disease history, which vary widely from individual to individual, are in a specific state linked to a specific individual, the magnitude of the influence exerted by other data items on the "specific data items serving as indicators" is estimated, and a health management element evaluation support system is provided for estimating the change received by the "specific data items serving as indicators" due to changes in data items other than the "specific data items serving as indicators", and the purpose is to utilize the information for a specific individual to maintain a good health status in the work of organizing by relevant parties who pay attention to the health status of the specific individual, such as doctors and the patient himself / herself.

Means for Solving the Problem

[0006] The health management element evaluation support system of the present invention includes a data capture unit (100) that captures a plurality of data groups (101, 102, 103) composed of data of three or more data items belonging to three or more types of data item groups including a determination data item group and an inspection data item group, a database construction unit (200) that edits and manages the plurality of data to construct a database, a learning unit (300) that sets a specific data item as an output layer and learns the data in the database using a neural network, and a prediction unit (400) that sets prediction conditions as an input layer using the learned neural network, and predicts a predicted value of a specific data item serving as an index, a numerical influence degree indicating the influence of the data item set in the input layer on the specific data item serving as the index, and a change in the predicted value of the specific data item serving as the index set in the output layer caused by a change in the data value of the data item set in the input layer.

Effects of the Invention

[0007] According to the present invention, in a system in which a data group composed of a plurality of data is in a specific state, it is possible to infer the influence of other data items on a specific data item serving as an index, and it is possible to narrow down data items to be noted in order to improve the specific data item serving as the index from among a plurality of data items.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Embodiments for Carrying Out the Invention

[0009] Hereinafter, embodiments for carrying out the present invention will be described with reference to figures and examples. The health management element evaluation support system of the present invention includes a data capture unit (100) that captures a plurality of data groups (101, 102, 103) composed of a plurality of data, a database construction unit (200) that edits and manages the plurality of data to construct a database (240), a learning unit (300) that learns the data in the database using a neural network, and a prediction unit (400) that uses the trained neural network to predict a predicted value of a specific data item serving as an index set in an output layer, a numerical influence degree indicating the influence of a data item set in an input layer on the specific data item serving as the index, and a change that occurs in the predicted value of the specific data item serving as the index set in the output layer when the data value of the data item set in the input layer changes. By using the health management element evaluation support system of the present invention, for example, in a data group consisting of a plurality of data in diabetic patients, at different specific states indicated by the data group of each patient, the stage of nephropathy, which is a determination data item set as a specific data item serving as an index, the numerical influence degree of each other data item on the stage of the aforementioned nephropathy, and the change in the stage of nephropathy caused by the change of each data item can be inferred (hereinafter, expressed as "infer the relationship between the stage of nephropathy in diabetic patients and influencing factors").

[0010] The data capture unit (100) in the health management element evaluation support system of the present invention is a unit that captures a data group (101, 102, 103) consisting of a plurality of data items into the health management element evaluation support system. For example, when "inferring the relationship between the stage of nephropathy in diabetic patients and influencing factors", in the data capture unit (100), the "stage of nephropathy", which is a data item determined by a doctor, and a data group of 285 diabetic patients consisting of a plurality of various data items related to the patient (a data group consisting of 31 data items described later) are captured into the health management element evaluation support system. The 31 data items in the data group of diabetic patients are as follows. Gender, age, height, weight, age at diabetes diagnosis, smoking habit, severity of periodontal disease, systolic blood pressure, "the first" systolic blood pressure since data registration, diastolic blood pressure, "the first" diastolic blood pressure since data registration, HDL cholesterol (HDL-C) level, "the first" HDL cholesterol (HDL-C) level since data registration, LDL cholesterol (LDL-C) level, "the first" LDL cholesterol (LDL-C) level since data registration, ALT level, "the first" ALT level since data registration, blood glucose level, "the first" blood glucose level since data registration, HbA1c level, "the first" HbA1c level since data registration, corrected value of urinary protein quantification, "the first" corrected value of urinary protein quantification since data registration, urinary albumin level, "the first" urinary albumin level since data registration, creatinine (Cr) level, "the first" creatinine (Cr) level since data registration, eGFR level, "the first" eGFR level since data registration, stage of nephropathy, "the first" stage of nephropathy since data registration.

[0011] The data used in the health management element evaluation support system of the present invention is not limited to the above. In order to obtain information for maintaining health from various angles, a group of judgment data items including one or more judgment data items such as the stage of a disease and health status determined and certified by a doctor, a group of unique data items including one or more unique data items such as gender, age, and genetic information, a group of test data items including one or more test data items obtained by analyzing body tissues such as blood and diseased tissues, a group of body data items including one or more body data items such as height, weight, waist circumference, and body fat percentage, a group of measurement data items including one or more measurement data items such as blood pressure and body temperature, a group of lifestyle data items including one or more lifestyle data items such as exercise habits and sleep habits, a group of history data items including one or more history data items such as medical history and treatment history, a group of oral intake habit data items including one or more oral intake habit data items such as foods, beverages, and luxury goods, and a group of regional data items including one or more regional data items such as place of origin, place of residence, and length of residence. A data group composed of data of three or more data items belonging to three or more types of data item groups including the judgment data item group and the test data item group is used. It is a preferred embodiment to use a data group composed of data of a plurality of data items belonging to four or more types of data item groups, and it is more preferred to use a data group composed of data of a plurality of data items belonging to five or more types of data item groups. In order to find out data items that strongly affect the judgment data among various data items, it is preferable that the data group consists of data of 10 or more data items, and it is more preferable that it consists of data of 20 or more data items. In order to consider safe and effective countermeasures for multiple different disease types and health statuses, it is preferable that the plurality of data items constituting the data group include two or more judgment data items belonging to the judgment data item group. Since the influence of the data item group on the judgment data item group can be inferred, it is a preferred embodiment to include a data item group with unclear relevance in the data group.From the above viewpoints, it is preferable to use, in the data group, data selected from the data items included in the unique data item group, the physical data item group, the measurement data item group, the lifestyle data item group, the history data item group, the oral intake habit data item group, and the regional data item group. It is more preferable to use the data of the data items belonging to one or both of the lifestyle data item group and the oral intake habit data item group.

[0012] The database construction unit (200) in the health management element evaluation support system of the present invention sets and edits the database construction conditions when handling data such as the correction methods for missing data and abnormal data, the classification methods for data items, and the learning correction methods, and edits and manages the data group captured by the data capture unit according to the database construction conditions when handling the data to construct a database (240). For example, when "inferring the relationship between the stage of nephropathy and influencing factors in diabetic patients", a database (240) consisting of 31 data items including "the stage of nephropathy" was constructed based on the diabetic patient data group of 285 people in total captured by the data capture unit (100). For example, as the database construction conditions when handling data, when there is a data item with missing data in the diabetic patient data group, the average value of the data item in the diabetic patient data group with the same value of "the stage of nephropathy" can be supplemented to construct the database.

[0013] In the database construction unit (200) of the present invention, it is preferable to correct missing data and abnormal data in the data groups (101, 102, 103) in order to suppress an adverse effect on the learning unit (300). In order to improve the reliability of the database (240) used in the learning unit, it is preferable that the database (240) is corrected with estimated values obtained by learning and correcting missing data and abnormal data. As a method of supplementing missing data and correcting abnormal data, missing data and abnormal data can be manually input and corrected, but it is preferable to set rules for handling the data to alert to check for missing data and abnormal data in the data groups, and it is more preferable to automatically detect missing data and abnormal data and automatically input and correct them. For example, when there is a certain difference or more between the "stage of nephropathy" determined by a doctor and the "stage of nephropathy" estimated by the health management element evaluation support system of the present invention, it is preferable to extract the corresponding data group and prompt attention.

[0014] In order to improve the reliability of the database (240) used in the learning unit, it is preferable that the database (240) is corrected with estimated values obtained by learning and correcting missing data and abnormal data. In learning correction, using the corrected correction database (24α) for insufficient data and abnormal data, the function (210) for setting and editing database construction conditions is used to set data items including insufficient data and abnormal data that require learning correction work as the output layer (223) in the database construction unit, and the learning of the neural network (222) that performs learning in the database construction unit is carried out. Then, using the neural network (222) that has been learned to perform learning in the database construction unit, a data group including insufficient data and abnormal data is used as the input layer (221) in the database construction unit, and the output layer (223) in the database construction unit is obtained, so that estimated values obtained by learning for insufficient data and abnormal data can be obtained. By replacing the insufficient data and abnormal data in the correction database (24α) with the estimated values obtained by learning, a learning correction database (24α + 1) can be obtained. Repeating the learning correction work is preferable for further improving the reliability of the database (240). Also, identifying and storing separately the databases before and after performing the supplementary work for insufficient data and the correction work for abnormal data is preferable for confirming and examining the validity of the content of the supplementary work and the correction work. In order to examine the validity of the construction state of the database, it is preferable to have a function (230) for outputting and displaying the construction state of the database.

[0015] In the health management element evaluation support system of the present invention, in order to consider the state from various viewpoints, a plurality of inference results obtained using a plurality of databases (240) can be compared. For example, it is possible to prepare the plurality of types of databases (240) by changing the data items constituting the database, the number, and the types of data item groups. By providing one or a plurality of data items serving as classification criteria from among data items such as gender and oral intake habits of preferred product categories, setting judgment criteria for grouping in the data items serving as the criteria, and performing grouping of the data group, and constructing databases associated with each group using the grouped data group, obtaining a plurality of databases is a preferable aspect in examining the influence of differences in data items serving as classification criteria on the inference result. In order to avoid divergence of considerations regarding the inference result, it is preferable that the plurality of databases used to obtain the plurality of different inference results be two or more and ten or less, and more preferably three or more and five or less.

[0016] The learning unit (300) in the health management element evaluation support system of the present invention is a process of setting a specific data item as an index as an output layer and learning the data items of the database (240) by a neural network. For example, when "inferring the relationship between the stage of nephropathy and influencing factors in diabetic patients", the above-mentioned 31 data items are set in the input layer (321) of the learning unit, and "the stage of nephropathy" is set as the output layer (323) in the learning unit, and learning of the neutral network (322) is performed using the database (240) based on the data group of 285 diabetic patients. In order to examine the validity of the learning content, it is preferable to have a function (330) for outputting and displaying the learning result.

[0017] In the inference unit (400) of the health management element evaluation support system of the present invention, the inference conditions are set as the input layer (421) in the inference unit, and the learned neural network (322) obtained by the above-mentioned learning unit (300) is used. As the sum of the weights of the synapses connecting the artificial neurons of the neural network from the data item set in the input layer (421) in the inference unit to the specific data item serving as the index set in the output layer (423) in the inference unit, the quantification influence degree indicating the influence of the data item set in the input layer (421) in the inference unit on the "specific data item serving as the index" set in the output layer (423) in the inference unit, and the change in the inferred value of the specific data item serving as the index set in the output layer (423) in the inference unit due to the change in the data of the data item set in the input layer (421) in the inference unit are inferred. As shown in FIG. 5, the quantification influence degree can be obtained by adding the weights of the synapses connecting the artificial neurons in the learned neural network. For example, in a learned neural network (500) where the input layer is X1, X2, X3, the output layer is Z1, Z2, Z3, and the hidden layer is Y1, Y2, the quantification influence degree of the input layer X1 on the output layer Z1 can be obtained by adding the weights (W) of the synapses connecting the output layer Z1 and the input layer X1 as "W1 11 + W1 12 + W2 11 + W2 21".

[0018] For example, when "inferring the relationship between the stage of nephropathy and influencing factors in diabetic patients", the data of the above-mentioned 31 data items in a specific diabetic patient is set as the input layer (421) in the inference unit, and from the above-mentioned learning model, the "stage of nephropathy" set in the output layer (423) in the inference unit and the quantified influence degrees of other data items on the "stage of nephropathy" are inferred. As shown in FIG. 6, in an example of the screen for inputting, operating, and displaying in the health management element evaluation support system according to the present invention, in the data group of a specific diabetic patient, the value of the original data group of the "stage of nephropathy" set in the output layer (423) in the inference unit and the value inferred from the learning model, the order of the quantified influence degrees of each data item on the "stage of nephropathy", and the result of inferring the change received by the "stage of nephropathy" when "body weight", "systolic blood pressure", and "eGFR" are changed can be confirmed. Thus, in the health management element evaluation support system of the present invention, in a specific data group, when an arbitrary data item is changed, the result of inferring the change received by a specific data item serving as an index can be confirmed. In the arbitrary data item and the specific data item serving as an index in the specific data group, it is a preferable mode for confirming the influence when the arbitrary data item is changed to display the state before the arbitrary data item is changed and the inferred result after the change in a comparative form.

[0019] Also, in the health management element evaluation support system of the present invention, in order to confirm the influence of each data item on a specific data item serving as an index in the learned neural network, the quantified influence degree of each data item can be displayed. As shown in FIG. 7, in order to compare the influence of each data item on a specific data item serving as an index, it is preferable to compare and display the quantified influence degrees of each data item. By comparing the quantified influence degrees of each data item in a plurality of learned neural networks obtained using a plurality of databases, the influence of the conditions provided in constructing the database on the inference result can be examined. By utilizing these functions, when a doctor or a patient deliberates alone on the improvement guidelines for life and treatment for improving health and preventing diseases, or when multiple people such as doctors or between a doctor and a patient have a discussion, it can be used as a deliberation tool or a communication tool. In order to examine the validity of the speculation content, it is preferable to have a function (430) for outputting and displaying the speculation result.

[0020] In an information processing apparatus that executes the health management element evaluation support system of the present invention, in order to construct a database in cooperation from various places, share and execute calculations among multiple people or places, and widely share and utilize the calculation results, it is preferable that the apparatus configuration includes communication modules (651, 652, 653) for information exchange between the outside, devices, and terminals. For example, as shown in FIG. 8, there is an information processing apparatus including a data capture module (601) for capturing a data group, a storage module (611) for storing a database, an input module (621) for setting and editing various conditions such as database construction conditions, learning conditions, and speculation conditions, a calculation processing module (631) for executing database management, learning, speculation, etc., an output / display module (641) for confirming the construction state of the database, learning conditions, speculation conditions, speculation results, etc., and a communication module (651) for information exchange with the outside. In data capture, database construction, neural network learning, speculation of specific data items serving as indicators, numerical influence degrees, etc., it becomes possible to execute operations and functions such as common inputs, calculation processing, output / display using common modules, thereby simplifying the apparatus configuration. Also, for example, as shown in FIG. 9, it may be an information processing apparatus in which a plurality of terminals are connected by a network. Each device or terminal connected to the network has at least a part of the functions constituting the present invention and their execution modules, and it is necessary that the overall system integrating each terminal satisfies the functions constituting the present invention.

Industrial Applicability

[0021] By using the health management element evaluation support system of the present invention, in a state system having a plurality of data including those with unclear relevance and its changes, it becomes possible to analyze the influence of other data on specific data and its changes. It can be used as an indicator for proceeding with judgment work in cases such as "the influence of individual data on the state system and between elements is unclear", "it is difficult to measure each data with high accuracy enough to be evaluated alone", and "the change in the influence exerted by each data due to differences in the state system itself is unclear". Therefore, the influence evaluation support system of the present invention can be used as a support system for considering lifestyle guidance and treatment policies, such as when inferring major factors affecting the progression of diseases and the target levels to be improved in the field of health and medicine. For example, when trying to reduce the burden on the body caused by examinations and it becomes necessary to suppress the amount of specimens such as blood and cells used in the examinations, and high examination accuracy that is sufficient to judge only by individual examinations cannot be expected, or when the influence of data such as inherent factors such as physique, age, gender, medical history, and lifestyle habits such as smoking, drinking, diet, exercise habits, sleep time, and lifestyle rhythm is suspected. The health management element evaluation support system of the present invention can be used when proceeding with judgment work or trying to share the content among multiple people. The health management element evaluation support system of the present invention can be used not only to maintain the health of humans, pets, and livestock, but also in various applications such as when considering product quality and equipment management in the manufacturing industry and when attempting to analyze influencing factors in the academic field.

Explanation of Signs

[0022] 100: Data acquisition unit 101: Data group a1 102: Data group a2 103: Data group ax 200: Database construction unit 210: Function for setting and editing database construction conditions 220: Database management function 221: Input layer in the database construction unit 222: Neural network for learning in the database construction unit 223: Output layer in the database construction unit 230: Function to output and display the construction state of the database 240: Database 24α: Corrected database with insufficient and abnormal data corrected 24α+1: Learning correction database with insufficient and abnormal data corrected by the learning correction function 300: Learning unit 310: Function to set and edit learning conditions 320: Learning function 321: Input layer in the learning unit 322: Neural network for learning in the learning unit 323: Output layer in the learning unit 330: Function to output and display learning results 400: Inference unit 410: Function to set and edit inference conditions 420: Inference function 421: Input layer in the inference unit 423: Output layer in the inference unit 430: Function to output and display inference results 500: Weights of synapses connecting artificial neurons of the neural network 601, 602: Data acquisition module 611: Memory module 621, 622, 623: Input module 631, 632, 633: Arithmetic processing module 641, 642, 643: Output / display module 651, 652, 653: Communication module

Claims

1. A data acquisition unit that acquires a plurality of data groups composed of data of three or more data items belonging to three or more types of data item groups including a determination data item group and an inspection data item group; a database construction unit that edits and manages the plurality of data groups to construct a database; a learning unit that sets a specific data item as an output layer and learns the data in the database using a neural network; and using the learned neural network, sets inference conditions as an input layer, and outputs an inference value of a specific data item serving as an index set in the output layer, a numerical influence degree indicating the influence of the data item set in the input layer on the specific data item serving as the index, and an inference unit that infers a change that occurs in the inference value of the specific data item serving as the index set in the output layer when the data value of the data item set in the input layer changes. A health management element evaluation support system.

2. The health management element evaluation support system according to claim 1, characterized in that, in an arbitrary data item of a specific data group and a specific data item serving as an index, the state before changing the arbitrary data item and the inferred result after changing are displayed in a comparative form.

3. The health management element evaluation support system according to claims 1 to 2, characterized in that the numerical influence degrees of each data item in a plurality of learned neural networks obtained using a plurality of databases are compared and displayed.

4. The health management element evaluation support system according to claims 1 to 3, characterized in that one or more data items serving as classification criteria are provided among the data items, a judgment criterion for grouping is set for the data items serving as the criteria, the data group is grouped, and a plurality of databases are obtained by constructing databases associated with each group using the grouped data groups.

5. The health management element evaluation support system according to claims 1 to 4, characterized in that two or more and ten or less types of estimation results obtained using two or more and ten or less types of databases are compared and displayed.

Images