JP6929124B2 - Forecasting systems, forecasting methods, and forecasting programs - Google Patents

Description

One aspect of the invention relates to prediction systems, prediction methods, and prediction programs.

Various computer systems for managing medical or human health are known. For example, Patent Document 1 below describes a medical data analysis device that predicts changes in the health condition of a subject. This analyzer deeply records medical data in a series of ages of a series of subjects given in the form of a bug-of-word in which different types of data are mixed as frequency factors, as a collection of individual data for each subject and age. Apply learning. Then, the analysis device outputs the representation data of the individual data in each intermediate layer as a compressed representation of a bug of word in which different types of data are mixed as frequency elements.

JP-A-2017-27307

Methods for predicting human health are known. However, it is convenient if the state can be predicted more accurately, and further, various further information can be obtained based on the prediction result. Therefore, a mechanism for more accurately predicting the future state of the subject is desired.

The prediction system according to one aspect of the present invention is an acquisition unit that acquires the past consultation record of the subject from a predetermined database, and is based on the acquisition unit and the consultation record in which the consultation record indicates at least medical treatment. A conversion unit that generates a plurality of input vectors and normalizes the element group of each input vector within a predetermined numerical range, wherein the element group includes at least an element indicating a medical treatment. It is a prediction unit that obtains an output vector that indicates the prediction of the state of the target person at a predetermined time in the future by inputting each of the converted input vectors to the learning algorithm, and the element group of the output vector is the element group of the input vector. It is provided with the prediction unit corresponding to the above.

The prediction method according to one aspect of the present invention is a prediction method executed by a prediction system including at least one processor, and is an acquisition step of acquiring a past consultation record of a subject from a predetermined database. An acquisition step in which the consultation record indicates at least a medical procedure, and a conversion step in which a plurality of input vectors are generated based on the consultation record and the element group of each input vector is normalized within a predetermined numerical range. By inputting the transformation step and each normalized input vector into the learning algorithm, the group includes at least an element indicating the medical treatment, so that an output vector indicating a prediction of the subject's state at a predetermined time in the future is obtained. It is a prediction step to be obtained, and includes the prediction step in which the element group of the output vector corresponds to the element group of the input vector.

The prediction program according to one aspect of the present invention is an acquisition step of acquiring a past consultation record of a subject from a predetermined database, and is based on the acquisition step and the consultation record in which the consultation record indicates at least a medical treatment. A conversion step that generates a plurality of input vectors and normalizes the element group of each input vector within a predetermined numerical range, wherein the element group includes at least an element indicating a medical treatment. It is a prediction step to obtain an output vector indicating the prediction of the state of the target person at a predetermined time in the future by inputting each of the converted input vectors to the learning algorithm, and the element group of the output vector is the element group of the input vector. The computer system is made to perform the prediction step corresponding to the above.

In such an aspect, by normalizing the input vector expressing the consultation record of the subject and then inputting the input vector into the learning algorithm, an output vector showing the prediction of the future state of the subject can be obtained. The input vector contains at least an element indicating a medical procedure, and the output vector corresponds to the input vector. By outputting an output vector that includes at least an element indicating a medical procedure, the condition of the subject can be predicted more accurately.

According to one aspect of the present invention, the future state of the subject can be predicted more accurately.

It is a figure which shows the hardware configuration of the computer used in the prediction system which concerns on embodiment. It is a figure which shows the functional structure of the prediction system which concerns on embodiment. It is a flowchart which shows the operation of the prediction system which concerns on embodiment. It is a figure which shows the structure of the prediction program which concerns on embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same or equivalent elements are designated by the same reference numerals, and duplicate description will be omitted.

The prediction system 10 according to the embodiment is a computer system that predicts the future state of the subject. For example, the prediction system 10 predicts when and what kind of medical treatment the subject will receive, and predicts what the physical condition of the subject will be in the future. A "medical procedure" is a procedure performed by a medical institution on a patient and includes, for example, at least one of medical examination, treatment, and prescription of a drug. More specifically, medical procedures are performed by medically qualified persons (eg, doctors, nurses, pharmacists, etc.). When a person (patient) receives medical treatment, it is called "visiting". The "subject" is a person (patient) who is the target of the prediction process.

FIG. 1 shows a general hardware configuration of the computer 100 constituting the prediction system 10. For example, the computer 100 includes a processor 101, a main storage unit 102, an auxiliary storage unit 103, a communication control unit 104, an input device 105, and an output device 106. Processor 101 executes operating systems and application programs. The main storage unit 102 is composed of, for example, a ROM and a RAM. The auxiliary storage unit 103 is composed of, for example, a hard disk or a flash memory, and generally stores a larger amount of data than the main storage unit 102. The communication control unit 104 is composed of, for example, a network card or a wireless communication module. The input device 105 is composed of, for example, a keyboard, a mouse, a touch panel, and the like. The output device 106 is composed of, for example, a monitor and a speaker.

Each functional element of the prediction system 10 is realized by loading a predetermined software (for example, a prediction program P1 described later) on the processor 101 or the main storage unit 102 and executing the software. The processor 101 operates the communication control unit 104, the input device 105, or the output device 106 according to the software, and reads and writes data in the main storage unit 102 or the auxiliary storage unit 103. The data or database required for processing is stored in the main storage unit 102 or the auxiliary storage unit 103.

The prediction system 10 may be composed of one computer or a plurality of computers. When a plurality of computers are used, one prediction system 10 is logically constructed by connecting these computers via a communication network such as the Internet or an intranet.

The prediction system 10 can access the database group 20 via a communication network such as the Internet or an intranet. The database group 20 is a set of databases for storing people's consultation records. A database is a functional element or device that stores a set of data to accommodate any data manipulation (eg, extract, add, delete, overwrite, etc.) from a processor or external computer. The method of implementing the database is not limited, but a database management system (DBMS) may be used, for example. A "visit record" is data that at least indicates the medical procedure that a person has received in the past.

As shown in FIG. 2, in the present embodiment, the database group 20 includes a receipt database 21, a medical examination database 22, and a questionnaire database 23. The location where the receipt database 21, the medical examination database 22, and the questionnaire database 23 are provided is not limited. For example, at least a part of these databases may be managed by a computer system different from the prediction system 10, or may be a part of the prediction system 10. In addition, the management entity of each database is not limited. For example, the operator of the prediction system 10 may be provided with data from another institution and manage a set of these data as a database group 20.

The medical receipt database 21 is a database that stores medical receipt data indicating medical treatments that a person has received in the past. For example, the receipt data is generated by a medical institution and stored in the receipt database 21. The receipt database 21 may be a set of receipt databases of each medical institution, or may be one database in which the receipt data of each medical institution is aggregated.

Each record of the receipt data includes the subject ID, the subject's attributes, the medical institution ID, the date of consultation, and the details of the medical procedure. The subject ID is an identifier that uniquely identifies a person who has undergone medical treatment. The subject's attributes are data that indicate the subject's nature or characteristics, such as gender and age. The medical institution ID is an identifier that uniquely identifies the medical institution that has performed the medical procedure. The date of consultation is the date on which the medical procedure was performed. The details of the medical procedure are data showing one or more medical examination items, one or more treatment items, and one or more prescription drugs. Data showing prescription drugs include drug type, dose, and days of administration. Since there are so many types of examinations, treatments, and prescription drugs, the details of a medical procedure can contain so many (eg, thousands, tens of thousands) of data items.

The medical examination database 22 is a database that stores medical examination data showing the results of medical examinations that a person has received in the past. For example, medical examination data is generated by a medical institution and stored in the medical examination database 22. Similar to the medical receipt database 21, the medical examination database 22 may be a set of medical examination databases of each medical institution, or may be one database in which medical examination data of each medical institution is aggregated.

Each record of the medical examination data includes the subject ID, the subject's attributes, the medical institution ID, the medical examination date, and the test result. The subject ID is an identifier that uniquely identifies the person who has undergone the medical examination. The attributes of the target person are the same as those of the receipt data. The medical institution ID is an identifier that uniquely identifies the medical institution that performed the medical examination. The medical examination date is the date when the medical examination was carried out. The test result is the result of a medical examination, and more specifically, it is a set of a plurality of test values. Since the health examination includes various test items (eg, at least one of the legal and non-statutory items), the test result can include a large number (eg, tens or hundreds) of data items.

The questionnaire database 23 is a database that stores questionnaire data showing the results of questionnaires related to health that people have answered in the past. Examples of health-related questionnaires include, but are not limited to, stress checks conducted by companies and lifestyle-related questionnaires conducted during health examinations. The questionnaire data is generated by a company or a medical institution and stored in the questionnaire database 23. Similar to the receipt database 21, the questionnaire database 23 may be a set of questionnaire databases of each company or each medical institution, or may be one database in which the questionnaire data of each company or each medical institution are aggregated. ..

Each record of the questionnaire data includes the subject ID, the executing agency ID, the response date, and the questionnaire result. The target person ID is an identifier that uniquely identifies the person who responded to the questionnaire. The executing agency ID is an identifier that uniquely identifies the institution (for example, a company or medical institution) that conducted the questionnaire. The answer date is the date when the answer to the questionnaire is entered. Questionnaire results are data showing multiple answers, including various answers such as stress checks, lifestyle habits, and awareness surveys.

FIG. 2 shows the functional configuration of the prediction system 10. As shown in FIG. 2, the prediction system 10 includes an acquisition unit 11, a conversion unit 12, and a prediction unit 13 as functional components.

The acquisition unit 11 is a functional element that acquires the past consultation record of the subject. When the target person ID is automatically or by the operator of the prediction system 10, the acquisition unit 11 reads out the consultation record corresponding to the target person ID from the database group 20. Specifically, the acquisition unit 11 reads the receipt data, the medical examination data, and the questionnaire data corresponding to the subject ID from the receipt database 21, the medical examination database 22, and the questionnaire database 23, respectively. Then, the acquisition unit 11 outputs the past consultation record composed of these data to the conversion unit 12.

The conversion unit 12 is a functional element that generates a plurality of input vectors based on past consultation records. The input vector is data represented by a matrix of N rows and 1 column (N> 1), and is input to a learning algorithm described later. One input vector corresponds to the medical examination record at one time point in the past, and this input vector indicates the receipt data, the medical examination data, and the questionnaire data at that time point. Receipt data including details of medical treatment (examination, treatment, and prescription drug) consists of P data items, and medical examination data including test results (test values) consists of Q data items, and questionnaire results (answers) ) Is included in the questionnaire data, which consists of R data items. In this case, the input vector v _t at a certain temporary point t can be expressed by the following equation (1). Elements a _{1 to} a _P indicate each data item of the receipt data, elements b _{1 to} b _Q indicate each data item of the medical examination data, and elements c _{1 to} c _R indicate each data item of the questionnaire data. The number of elements (number of rows) of each input vector can be enormous depending on the number of data items of each data.

In general, a person receives multiple medical procedures, undergoes regular health examinations, and regularly answers health questionnaires. For example, a person receives medical treatment monthly for a specific period of time, receives a health checkup once a year, and answers a questionnaire once a year (for example, during a health checkup). Therefore, the acquisition unit 11 acquires the receipt data, the medical examination data, or the questionnaire data at a plurality of past time points for one subject ID, and obtains a plurality of input vectors indicating the consultation records at the plurality of time points. Generate from. Assuming that the number of input vectors obtained is M, the past consultation record of the subject can be represented by a set V of input vectors represented by the following equation (2). The input vectors in the set V are arranged in chronological order. Therefore, this set V is time series data.
V = {v ₁ , v ₂ , ..., v _M } ... (2)

The element group of the input vector is represented by a numerical value. On the other hand, there may be data items that are not represented numerically (for example, data items expressed in text) in the receipt data, the medical examination data, and the questionnaire data. In this case, the conversion unit 12 converts the data item into a numerical value based on a predetermined rule. The rules for converting non-numeric values into numerical values are described in advance in the program code or setting file that realizes the conversion unit 12, and the conversion unit 12 executes the conversion according to the description.

The method of converting a non-numeric value into a numerical value is not limited, and the conversion method may differ for each data item. For example, the conversion unit 12 may convert non-numeric data into binary values (0 or 1), or within a predetermined range (for example, 0 to 1, 0 to 10, 1 to 100, -5 to +5, etc.). It may be converted into a numerical value, or may be converted using the value contained in the original data. As an example, it is assumed that item A with receipt data indicates an injection containing a specific component B. The conversion unit 12 may substitute a binary value (0: no injection, 1: with injection) indicating the presence or absence of injection into the element corresponding to this item A, or substitute the amount of the applied component B into the element. May be good.

The dates indicated by the receipt data, medical examination data, or questionnaire data (examination date, medical examination date, and response date) can be different from each other. Also, the number of these data can be different from each other. For example, when the consultation day interval is smaller than the medical examination day interval and the response date interval, the elements corresponding to the medical examination data and the questionnaire data are empty values in at least one input vector. When an empty value is generated in this way, the acquisition unit 11 replaces the empty value with a predetermined significant numerical value (for example, 0). This is called padding, and this process sets significant numerical values for all elements of all input vectors.

The conversion unit 12 normalizes the element group of each generated input vector within a predetermined numerical range. The numerical range is not limited, but for example, the conversion unit 12 may normalize each element in the range of 0 or more and 1 or less. The normalization rules are also described in advance in the program code or the setting file that realizes the conversion unit 12, and the conversion unit 12 executes the normalization according to the description.

The specific rules of normalization are not limited, and the normalization method may differ from element to element. For example, the conversion unit 12 may convert an element into a binary value (0 or 1), or may convert it into an arbitrary value between 0 and 1. Depending on the data item, the value may not change even if it is normalized. Equation (3) below shows a specific example of a set of normalized input vectors.

The conversion unit 12 may increase the number of 0 elements in the input vector by performing sparse regularization on any element of the normalized input vector. This sparse regularization can be expected to improve learning accuracy while avoiding overfitting. The rules for sparse regularization are also described in advance in the program code or the setting file that realizes the conversion unit 12, and the conversion unit 12 executes the sparse regularization according to the description.

Which element is subject to sparse regularization may be arbitrarily determined, and the conversion unit 12 may execute sparse regularization for all elements of the input vector, or execute sparse regularization for only some elements. You may. For example, the conversion unit 12 may execute sparse regularization only for the element indicating the medical procedure in the receipt data. Alternatively, the conversion unit 12 may execute sparse regularization only for the element corresponding to the medical procedure and the element corresponding to the questionnaire data in the receipt data. The conversion unit 12 does not execute sparse regularization for elements that require accuracy (for example, elements corresponding to the test results of medical examination data), but performs sparse regularization for some or all of the other elements. You may do it.

The transforming unit 12 may compress the normalized input vector or the input vector to which both normalization and sparse normalization are applied. Compression is the process of reducing the number of dimensions of the input vector. The conversion unit 12 reduces the number of dimensions of each input vector by removing elements (rows) that are 0 in all the input vectors in the set V from these input vectors.

In this way, the conversion unit 12 generates a plurality of input vectors (executes padding if necessary) and normalizes each input vector. If necessary, the transforming unit 12 further performs at least one of sparse normalization and compression on the input vector. The conversion unit 12 outputs the set V of the input vectors obtained by this series of processing to the prediction unit 13.

The prediction unit 13 is a functional element that obtains an output vector indicating a prediction of the state of the subject (for example, a consultation prediction) at a predetermined time in the future by inputting at least each normalized input vector to the learning algorithm. be. Examples of the learning algorithm (learning model) include LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), and Attention. However, the learning algorithm used by the prediction unit 13 is not limited to these, and any algorithm may be used.

The prediction unit 13 divides a plurality of input vectors arranged in time series into the first half and the second half. Subsequently, the prediction unit 13 first causes the learning algorithm to process the input vector in the first half to learn the learning algorithm. Subsequently, the prediction unit 13 verifies the learning algorithm by letting the learning algorithm process the input vector in the latter half, and the learning algorithm estimated to have the highest calculation system (hereinafter, this is also referred to as the "best learning algorithm"). To get. That is, the set of input vectors in the first half is training data, and the set of input vectors in the second half is test data.

Subsequently, the prediction unit 13 uses the obtained best learning algorithm to obtain an output vector indicating the prediction of the state of the target person at a predetermined time in the future (hereinafter, also referred to as “target time”). The future time to be predicted is input to the prediction system 10 in advance, and is input together with, for example, the target person ID. The prediction unit 13 divides the period from the present to the target time into a plurality of unit periods. Then, the prediction unit 13 sequentially obtains the output vector at the time when the unit period elapses from the present to the target time.

As an example, the set _{V = {v 1, v 2} , ..., v M} of input vectors corresponding to the past time with, and to calculate the output vector _{v y} in the target time _{t y.} First, half _{V F = {v 1, ...} , v i} of the prediction unit 13 is set V to learn a learning algorithm using a second half _{V L = {v i + 1} , ..., v M} of set V with By verifying the learning algorithm, the best learning algorithm is obtained.

Then, the estimating unit 13 divides the current time t _c a period until the target time t _y to N unit period k. Subsequently, the prediction unit 13 inputs the set V = {v ₁ , ..., V _M } of the input vectors to the best learning algorithm, so that the output vector v indicating the prediction of the state _{in the future time (t c + 1k).} ´ ₁ is obtained.

Then, the estimating unit 13 except for the oldest input vector v ₁ in the set V, is added instead its output vector v _'1 to set V in. As a result, the set V of the input vectors changes to _{{v 2} , ..., v _M , _{v ′ 1}.} Prediction unit 13 the set _{V = {v 2, ...,} v M, v'1} by inputting the best learning algorithm, the output vector indicating the predicted state at a future time _(t c + 2k) v' _{Get 2} .

Then, the estimating unit 13 except for the oldest input vector v ₂ in the set V, by adding the output vectors v _'2 to set V instead, the set of input vectors _{V = {v 3, ...,} v _{M, v '1,} obtain v' _2}. Prediction unit 13 the set _{V = {v 3, ...,} v M, v'1, v'2} indicates a by entering the best learning algorithms, prediction of the state in the future time _(t c + 3k) obtain an output vector v _'3.

Thus, prediction unit 13 continue to seek an output vector while shifting towards a period shown by a set of input vectors in the future, finally, the target timing _{t y (= t c + Nk} ) output vectors in v _'N Ask for. The finally obtained output vector element group corresponds to the input vector element group. Therefore, the output vector includes at least one or more elements indicating the medical procedure (prediction regarding the item of the receipt data) and other elements regarding the physical condition of the person (prediction regarding the item of the medical examination data data or the questionnaire data, for example, various types). Test value of.) Can be further included.

From the finally obtained output vector, it is possible to know what kind of medical treatment the subject will receive at a predetermined time in the future, and to know the future state of the subject's body. However, since each element of this output vector is normalized, it is difficult for humans to understand even if the normalized value is output as it is. Therefore, the prediction unit 13 may process this output vector. For example, the prediction unit 13 may return the value of the element of the output vector to the unnormalized value by canceling the normalization of the output vector (non-normalization). Alternatively, the prediction unit 13 may convert at least a part of the elements of the output vector or the denormalized output vector in the normalized state into text.

Alternatively, the prediction unit 13 may generate additional data that can be derived from the elements of the output vector. For example, the prediction unit 13 determines the medical expenses at a predetermined time in the future based on the value of each element related to the medical procedure and the unit price (medical fee) of each medical procedure (each medical examination, each treatment, and each prescription drug). Estimates may be obtained.

The prediction unit 13 outputs at least a part of the output vector itself, the processed output vector, and the data derived from the output vector as the prediction result. The output method is not limited. For example, the prediction unit 13 may display the prediction result on a monitor or print it with a printer, may store the prediction result in a predetermined database, or transmit the prediction result to another computer system. You may.

As described above, the format and output method of the prediction result are not limited. In any case, the user of the prediction system 10 can estimate when and what kind of medical treatment the subject will receive in the future by referring to the prediction result. The user can also know when the subject will stop treatment and how much medical expenses will be incurred in the future. Furthermore, the user can also know how the test value indicating the physical condition of the subject will change in the future.

Next, the operation of the prediction system 10 will be described with reference to FIG. 3, and the prediction method according to the present embodiment will be described. FIG. 3 is a flowchart showing the operation of the prediction system 10.

First, the acquisition unit 11 acquires the past consultation records (receipt data, medical examination data, and questionnaire data) of the subject from the database group 20 (step S11, acquisition step).

Subsequently, the conversion unit 12 generates a plurality of input vectors based on the consultation record (step S12, conversion step). First, the conversion unit 12 generates a set of input vectors based on the consultation record (step S121). If necessary, in this process, the conversion unit 12 converts non-numeric data into numerical values and executes padding. Subsequently, the conversion unit 12 normalizes the element group of each input vector (step S122), and executes sparse regularization or compression on the normalized input vector as needed (step S123).

Subsequently, the prediction unit 13 inputs at least each normalized input vector into the learning algorithm to show the prediction of the state of the subject (for example, consultation prediction) at a predetermined time (target time) in the future. Obtain the output vector (step S13). The prediction unit 13 learns a learning algorithm using a set of input vectors, and further verifies the learning result to obtain the best learning algorithm (step S131). Then, the prediction unit 13 obtains the output vector for each predetermined unit period from the present to the target time. Let t _{c be} the current time, k be the unit period, and be t _c + ik (i = 1, ..., N) for the future time. The prediction unit 13 first obtains an output vector indicating a state prediction (for example, consultation prediction) in a _{future time (t c} + 1k), and calculates the output vector while incrementing the variable i until the _{target time (t c + Nk).} Repeat (steps S132 to S135). Finally, the prediction unit 13 outputs a prediction result based on the output vector at the target time (step S14).

Next, the prediction program P1 for making at least one computer function as the prediction system 10 will be described with reference to FIG. FIG. 4 is a diagram showing the configuration of the prediction program P1.

The prediction program P1 includes a main module P10, an acquisition module P11, a conversion module P12, and a prediction module P13. The main module P10 is a part that comprehensively manages the consultation prediction. By executing the acquisition module P11, the conversion module P12, and the prediction module P13, the acquisition unit 11, the conversion unit 12, and the prediction unit 13 are realized.

The prediction program P1 may be provided after being fixedly recorded on a tangible recording medium such as a CD-ROM, a DVD-ROM, or a semiconductor memory. Alternatively, the prediction program P1 may be provided via a communication network as a data signal superimposed on a carrier wave.

As described above, the prediction system according to one aspect of the present invention is an acquisition unit that acquires past consultation records of a subject from a predetermined database, and the acquisition unit indicates at least medical treatment. A conversion unit that generates a plurality of input vectors based on a medical examination record and normalizes the element group of each input vector within a predetermined numerical range, and the element group includes at least an element indicating a medical treatment. It is a prediction unit that obtains an output vector that indicates the prediction of the state of the target person at a predetermined time in the future by inputting the conversion unit and each normalized input vector to the learning algorithm, and is an element group of the output vector. Is provided with the prediction unit corresponding to the element group of the input vector.

The prediction method according to one aspect of the present invention is a prediction method executed by a prediction system including at least one processor, and is an acquisition step of acquiring a past consultation record of a subject from a predetermined database. An acquisition step in which the consultation record indicates at least a medical procedure, and a conversion step in which a plurality of input vectors are generated based on the consultation record and the element group of each input vector is normalized within a predetermined numerical range. By inputting the transformation step and each normalized input vector into the learning algorithm, the group includes at least an element indicating the medical treatment, so that an output vector indicating a prediction of the subject's state at a predetermined time in the future is obtained. It is a prediction step to be obtained, and includes the prediction step in which the element group of the output vector corresponds to the element group of the input vector.

The prediction program according to one aspect of the present invention is an acquisition step of acquiring a past consultation record of a subject from a predetermined database, and is based on the acquisition step and the consultation record in which the consultation record indicates at least a medical treatment. A conversion step that generates a plurality of input vectors and normalizes the element group of each input vector within a predetermined numerical range, wherein the element group includes at least an element indicating a medical treatment. It is a prediction step to obtain an output vector indicating the prediction of the state of the target person at a predetermined time in the future by inputting each of the converted input vectors to the learning algorithm, and the element group of the output vector is the element group of the input vector. The computer system is made to perform the prediction step corresponding to the above.

In such an aspect, by normalizing the input vector expressing the consultation record of the subject and then inputting the input vector into the learning algorithm, an output vector showing the prediction of the future state of the subject can be obtained. The input vector contains at least an element indicating a medical procedure, and the output vector corresponds to the input vector. By outputting an output vector containing at least an element indicating a medical procedure, a person's condition can be predicted more accurately. For example, the user can refer to or use the output vector to know what kind of medical treatment the subject will receive at a predetermined time in the future (that is, it is possible to predict the visit of the subject. ).

Values indicating medical procedures are expressed in various formats such as texts and numerical values, so simply digitizing these values and inputting them into the learning algorithm may not improve the accuracy of the calculation or take a long time to calculate. there is a possibility. However, by normalizing the input vector containing the elements indicating the medical procedure, it becomes possible to accurately and quickly predict the future state of the subject, which needs to be obtained using a wide variety of data. ..

In the prediction system according to the other aspect, the transforming unit may perform sparse regularization on the elements indicating the medical procedure in the elements of the normalized input vector. Since there are so many types of medical procedures, the number of elements that indicate medical procedures is also very large. By making this element sparse regularized, the number of elements having a value of 0 increases, so that it is possible to improve the learning accuracy and shorten the calculation time while avoiding overfitting.

In the prediction system according to the other aspect, the prediction unit may obtain the medical expenses at a predetermined time in the future based on the output vector. In this case, future medical costs can be predicted.

The present invention has been described in detail above based on the embodiment. However, the present invention is not limited to the above embodiment. The present invention can be modified in various ways without departing from the gist thereof.

In the above embodiment, the consultation record is composed of the receipt data, the medical examination data, and the questionnaire data, but at least one of the medical examination data and the questionnaire data may be omitted. Alternatively, the consultation record may be composed of at least the receipt data and the exercise data showing the result of the exercise ability measurement that the person has received in the past. Therefore, the medical examination record may be composed of medical receipt data, medical examination data, questionnaire data, and exercise data.

The processing procedure of the prediction method executed by at least one processor is not limited to the example in the above embodiment. For example, some of the steps (processes) described above may be omitted, or each step may be executed in a different order. Further, any two or more steps among the above-mentioned steps may be combined, or a part of the steps may be modified or deleted. Alternatively, other steps may be performed in addition to each of the above steps.

When comparing the magnitude relations of two numbers in a forecasting system, either of the two criteria "greater than or equal to" and "greater than" may be used, and the two criteria "less than or equal to" and "less than" Either of them may be used. The selection of such criteria does not change the technical significance of the process of comparing the magnitude relations of two numbers.

10 ... Prediction system, 11 ... Acquisition unit, 12 ... Conversion unit, 13 ... Prediction unit, 20 ... Database group, 21 ... Receipt database, 22 ... Medical examination database, 23 ... Questionnaire database, P1 ... Prediction program, P10 ... Main module , P11 ... Acquisition module, P12 ... Conversion module, P13 ... Prediction module.

Claims (5)

An acquisition unit that acquires the past consultation record of the subject from a predetermined database, and the acquisition unit that the consultation record indicates at least a medical procedure.
A conversion unit that generates a plurality of input vectors based on the consultation record and normalizes the element group of each input vector within a predetermined numerical range, and the element group includes at least an element indicating the medical procedure. , The conversion unit and
Wherein by inputting the normalized set a learning algorithm for each input vector has been, a prediction unit for obtaining the final output vector indicating a prediction of the subject's status at a future predetermined period, the final The element group of the output vector corresponds to the element group of the input vector, and the prediction unit is provided .
The prediction unit
The period from the present to the predetermined time is divided into a plurality of unit periods.
An intermediate output vector showing a prediction of the state of the subject at the time when the unit period elapses while changing the set for each of the plurality of unit periods in order from the present to the predetermined time. Is executed to obtain the final output vector, and the final output vector is obtained.
In the iterative process, the set is input to the learning algorithm to acquire the intermediate output vector at the elapse of the i-th unit period, and then the oldest input vector in the set is removed and the set is removed. By inputting the set modified by adding the acquired intermediate output vector to the learning algorithm, the process of acquiring the intermediate output vector at the elapse of the (i + 1) th unit period is included.
Prediction system. The transforming unit performs sparse regularization on the element indicating the medical procedure in the element group of the normalized input vector.
The prediction system according to claim 1. The prediction unit obtains medical expenses at a predetermined time in the future based on the final output vector.
The prediction system according to claim 1 or 2. A prediction method performed by a prediction system with at least one processor.
An acquisition step of acquiring a subject's past consultation record from a predetermined database, wherein the consultation record indicates at least a medical procedure.
A conversion step in which a plurality of input vectors are generated based on the consultation record and the element groups of each input vector are normalized within a predetermined numerical range, wherein the element group includes at least an element indicating the medical procedure. , The conversion step and
By inputting the set of each normalized input vector into the learning algorithm, it is a prediction step of obtaining a final output vector indicating the prediction of the state of the subject at a predetermined time in the future , which is the final step. element group of minimal output vector corresponding to the element group of the input vector, see contains the said prediction step,
In the prediction step
The period from the present to the predetermined time is divided into a plurality of unit periods.
An intermediate output vector showing a prediction of the state of the subject at the time when the unit period elapses while changing the set for each of the plurality of unit periods in order from the present to the predetermined time. Is executed to obtain the final output vector, and the final output vector is obtained.
In the iterative process, the set is input to the learning algorithm to acquire the intermediate output vector at the elapse of the i-th unit period, and then the oldest input vector in the set is removed and the set is removed. By inputting the set modified by adding the acquired intermediate output vector to the learning algorithm, the process of acquiring the intermediate output vector at the elapse of the (i + 1) th unit period is included.
Prediction method. An acquisition step of acquiring a subject's past consultation record from a predetermined database, wherein the consultation record indicates at least a medical procedure.
A conversion step in which a plurality of input vectors are generated based on the consultation record and the element groups of each input vector are normalized within a predetermined numerical range, wherein the element group includes at least an element indicating the medical procedure. , The conversion step and
By inputting the set of each normalized input vector into the learning algorithm, it is a prediction step of obtaining a final output vector indicating the prediction of the state of the subject at a predetermined time in the future , which is the final step. The computer system is made to perform the prediction step in which the elements of the output vector correspond to the elements of the input vector.
In the prediction step
The period from the present to the predetermined time is divided into a plurality of unit periods.
An intermediate output vector showing a prediction of the state of the subject at the time when the unit period elapses while changing the set for each of the plurality of unit periods in order from the present to the predetermined time. Is executed to obtain the final output vector, and the final output vector is obtained.
In the iterative process, the set is input to the learning algorithm to acquire the intermediate output vector at the elapse of the i-th unit period, and then the oldest input vector in the set is removed and the set is removed. By inputting the set modified by adding the acquired intermediate output vector to the learning algorithm, the process of acquiring the intermediate output vector at the elapse of the (i + 1) th unit period is included.
Prediction program.

Images