JP2013206337A - Health care guidance target person selection support device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a health care guidance target person selection support device which solves a problem that it is difficult to select a guidance target person from the viewpoint of whether or not a person is likely to attend a health care guidance.SOLUTION: A memory stores: first medical examination data which is, as to an individual who is selected as a medical care guidance target person in a past term, medical examination data in the term; a first label value representative of whether or not a person attended a health care guidance in the term; second medical examination data which is medical examination data of an individual who is a health care guidance target person candidate of a current term; and a selecting condition. A processor is programmed so that it learns a first discrimination model for calculating a probability at which an individual attends a health care guidance from medical examination data of the individual by using first medical examination data and the first label value; inputs second medical examination data to the learned first discrimination model to calculate, for each individual who is a candidate of a health care guidance target person candidate of a current term, a first probability at which the individual will attend the health care guidance; and selects, on the basis of the calculated first probability, a health care guidance target person satisfying a selecting condition out of individuals who are health care guidance target person candidates.

Description

  The present invention relates to an apparatus for supporting an insurer's health guidance business, and more particularly to an apparatus for supporting selection of a health guidance target person.

  In recent years, an increase in medical expenses that an insurer bears on an insured person has put pressure on the insurer's finances, and there is an urgent need to create measures for reducing medical expenses. As one of the measures, the insurer is carrying out health guidance for health promotion of the insured. Specifically, a certain standard is set for the health check result of the insured person, and health guidance is given to an individual (applicable person) that satisfies the standard. For example, health guidance is provided for individuals who need to improve their lifestyle habits or who are at high risk of future sickness.

  However, since the insurer has limited costs for health guidance, it is difficult to provide guidance to all individuals who need guidance. For this reason, it is important how to select a health guidance target person who implements health guidance among individuals who need guidance.

  An example of a technique for selecting a health guidance subject is described in Patent Document 1. In the technology described in Patent Document 1 (hereinafter referred to as related technology of the present invention), an individual provides health guidance based on past medical examination data and health guidance results (improvement or non-improvement). Learn the discriminant model that calculates the probability of participation and improvement. Then, the health diagnosis data of candidates for health guidance for this term is input to the learned discrimination model, the probability of improvement for each candidate is calculated, and the health guidance target is selected using this calculated probability as a priority. . Thus, in order to prioritize individuals who have a high possibility of improvement by health guidance, individuals with high health guidance effects can be selected and health guidance can be implemented.

JP 2009-151404 A

  By the way, the insurer requests the individual selected as the health guidance target to participate in the health guidance, but it cannot be said that all the individuals who have made the request participate. Therefore, for example, when there are multiple individuals who have almost the same probability of improvement by health guidance, it is important to select those who are more likely to participate in health guidance, but support such selection. Currently, there is no technology that can help.

  The object of the present invention is to provide a health guidance target person selection support that solves the problem that it is difficult to select a subject person from the viewpoint of whether or not it is expected to participate in health guidance. To provide an apparatus.

The health guidance target person selection support device according to one aspect of the present invention is:
The first health check data that is the health check data of the individual selected as the health guidance target in the past period and the first label indicating whether or not the individual has participated in the health guidance of the period A memory for storing values, second health check data that is personal health check data of individuals who are candidates for health guidance for this term, and selection conditions;
A processor connected to the memory,
The processor
Using the first health check data and the first label value, learning a first discrimination model for calculating a probability that the individual participates in health guidance from the individual health check data;
The second health check data is input to the learned first discrimination model, and a first probability of participating in health guidance is calculated for each individual who is the candidate for health guidance in the current term.
Based on the calculated first probability, it is programmed to select a health guidance target person who satisfies the selection conditions from individuals who are candidates for the health guidance target of the current term.

  Since the present invention has the above-described configuration, it is possible to select a person to be instructed from the viewpoint of whether or not there is a possibility of participating in health instruction.

It is a block diagram of a 1st embodiment of the present invention. It is a flowchart which shows an example of the process of the 1st Embodiment of this invention. It is a figure which shows the formula used for description of the selection phase in the 1st Embodiment of this invention. It is a figure which shows the formula used for description of the selection phase in the 1st Embodiment of this invention. It is a figure which shows the formula used for description of the selection phase in the 1st Embodiment of this invention. It is a block diagram of the 2nd Embodiment of this invention. It is a flowchart which shows an example of the process of the 2nd Embodiment of this invention.

  Next, embodiments of the present invention will be described in detail with reference to the drawings.

[First embodiment]
Referring to FIG. 1, a health guidance target person selection support apparatus 1 according to the first embodiment of the present invention includes a plurality of individuals who are candidates for health guidance in health examinations performed every certain period. Therefore, it has the function to select the health guidance target person. In the present embodiment, the three types of probabilities calculated for each individual, that is, the probability of participating in health guidance, the probability of participating in health guidance and not being a candidate for health guidance in the next period, and participating in health guidance Without selecting the candidate for health guidance based on the probability that the candidate will not be the candidate for health guidance in the next period. Here, the period is arbitrary, but in the following description, one year is one period.

  The health guidance target person selection support apparatus 1 includes a communication interface unit (communication I / F unit) 11, an operation input unit 12, a screen display unit 13, a storage unit 14, and a processor 15 as hardware.

  The communication I / F unit 11 includes a dedicated data communication circuit, and has a function of performing data communication with various devices (not shown) connected via a communication line (not shown). The operation input unit 12 includes an operation input device such as a keyboard and a mouse, and has a function of detecting an operator operation and outputting the operation to the processor 15. The screen display unit 13 includes a screen display device such as an LCD or a PDP, and has a function of displaying various information such as an operation menu and a selection result on the screen in response to an instruction from the processor 15.

  The storage unit 14 includes a storage device such as a hard disk or a semiconductor memory, and has a function of storing processing information and programs 14P necessary for various processes in the processor 15. The program 14P is a program that realizes various processing units by being read and executed by the processor 15, and is read by an external device (not shown) or a computer via a data input / output function such as the communication I / F unit 11. It is read in advance from a possible storage medium (not shown) and stored in the storage unit 14. As main processing information stored in the storage unit 14, the past health diagnosis data 14A, three types of flags 14B to 14D, the current period health diagnosis data 14E, three types of discrimination models 14F to 14H, and three types Are 14I to 14K, a selection condition 14L, and a selection result 14M.

  The past health checkup data 14A is composed of the health checkup data of the individual who falls under a predetermined standard (for example, the diagnosis standard of metabolic syndrome) in the health check in the past year and is selected as the health guidance target person in the current year. . The past medical examination data 14A is classified by year and individual. The health check-up data for one individual for one year consists of a personal ID that uniquely identifies the individual, personal information such as the year of visit, age of visit, sex, etc., height, weight, waist circumference, minimum blood pressure, maximum blood pressure, blood glucose level, It has various test values such as neutral fat.

  The participation flag 14B is a label value for each individual indicating whether or not an individual who meets a predetermined standard in the health examination in the past year and is selected as a health guidance target person participated in the health guidance in the current year. In general, health guidance is conducted continuously multiple times. The participation flag 14B is set to a value indicating participation when participating in the health guidance even once, and set to a value indicating non-participation when not participating in the health guidance once.

  The participation improvement flag 14C corresponds to a predetermined standard in the health examination in the past year and the individual selected as the health guidance target person participates in the health guidance of the current year and becomes a candidate for health guidance target in the following year. It is a label value for each individual indicating whether or not.

  The non-participation improvement flag 14D is an individual label indicating whether or not the health examination target in the past year has fallen out of the candidate for health guidance in the next fiscal year without participating in the health guidance in the current fiscal year. Value.

  The health check data 14E for this term is composed of the health check data for the current year of individuals who meet the predetermined criteria in the health check for this year, that is, individuals who are candidates for health guidance. The items of the health examination data of one individual for this year are the same as the health examination data of the past year.

  The participation determination model 14F is a model that shows the relationship between the individual's health diagnosis data and whether or not the individual participates in health guidance. In addition, the participation improvement determination model 14G is a model that shows a relationship between individual health checkup data and whether or not an individual participates in health guidance and is excluded from candidates for health guidance in the next year. Further, the non-participation improvement determination model 14H is a model that shows a relationship between an individual's health diagnosis data and whether or not an individual is not participating in health guidance and is not a candidate for health guidance in the next fiscal year. The discriminant models 14F to 14H may be a linear regression model, a logistic regression model, or the like. The discriminant models 14F to 14H are generally polynomials composed of a plurality of explanatory variables and their coefficients (parameters).

  Participation probability 14I is the participation probability for individual health guidance calculated by inputting health checkup data 14E of this term into participation determination model 14F. The participation improvement probability 14J is an individual probability calculated by inputting the health examination data 14E of this term into the participation improvement determination model 14G. The participation improvement probability 14J will be excluded from candidates for health guidance in the next period after participating in health guidance. Shows the probability. Further, the non-participation improvement probability 14K is an individual probability calculated by inputting the health examination data 14E of this term into the non-participation improvement discrimination model 14H, and does not participate in the health guidance and is excluded from candidates for health guidance in the next period. Indicates the probability of

  The selection condition 14 </ b> L indicates a constraint condition set by the insurer when selecting a health guidance target person. The selection condition 14L is, for example, the maximum number of health guidance target persons or the maximum cost that an insurer can spend on health guidance.

  The selection result 14M is a list of individuals selected as health guidance subjects.

  The processor 15 has a microprocessor such as a CPU and its peripheral circuits, and reads and executes the program 14P from the storage unit 14, thereby causing the hardware and the program 14P to cooperate to implement various processing units. have. As main processing units realized by the processor 15, there are an input unit 15A, a discriminant model learning unit 15B, a probability calculating unit 15C, and an instruction target person selecting unit 15D.

  The input unit 15A inputs the past medical examination data 14A, the three types of flags 14B to 14D, the current medical examination data 14E, and the selection condition 14L from the communication I / F unit 11 or the operation input unit 12, and stores them. The function of storing in the unit 14 is provided.

  The discrimination model learning unit 15B reads the past health examination data 14A and the participation flag 14B from the storage unit 14, learns the participation discrimination model 14F using these data, and stores the participation discrimination model 14F after learning as the storage unit 14 Has the function of saving. Further, the discrimination model learning unit 15B reads the past health examination data 14A and the participation improvement flag 14C from the storage unit 14, learns the participation improvement discrimination model 14G using these data, and participates in the improvement improvement discrimination model 14G after learning. Is stored in the storage unit 14. Further, the discrimination model learning unit 15B reads the past health examination data 14A and the non-participation improvement flag 14D from the storage unit 14, learns the non-participation improvement discrimination model 14H using these data, and the non-participation improvement discrimination model 14H after learning. Is stored in the storage unit 14.

  The probability calculation unit 15C reads the participation determination model 14F and the current period health check data 14E from the storage unit 14, inputs the current period health check data 14E to the participation determination model 14F, and is a candidate for health guidance target for the current period. It has a function of calculating a probability 14I of participating in health guidance for each individual and storing it in the storage unit 14. The probability calculation unit 15C reads the participation improvement determination model 14G and the current period health check data 14E from the storage unit 14, inputs the current period health check data 14E to the participation improvement determination model 14G, and receives the current period health guidance target person. For each individual who is a candidate, he / she has the function of calculating the probability 14J of participating in the health guidance and not being a health guidance candidate candidate in the next year and storing it in the storage unit 14. Further, the probability calculation unit 15C reads the non-participation improvement determination model 14H and the current period health checkup data 14E from the storage unit 14, inputs the current period health checkup data 14E into the nonparticipation improvement determination model 14H, For each individual who is a candidate, it has a function of calculating the probability 14K of not being a candidate for health guidance in the next year without participating in health guidance and storing it in the storage unit 14.

  The instruction target person selecting unit 15D reads the three types of probabilities 14I to 14K and the selection condition 14L from the storage unit 14, and based on these three types of probabilities 14I to 14K and the selection condition 14L, the subject of health instruction for this term A health guidance target person is selected from candidate individuals so as to satisfy the selection condition 14L, and is stored in the storage unit 14 as a selection result 14M. The instruction target person selecting unit 15 </ b> D has a function of reading the selection result 14 </ b> M from the storage unit 14 and outputting the selection result 14 </ b> M to the screen display unit 13 or outputting it to the outside through the communication I / F unit 11.

  Next, the operation of the health guidance target person selection support apparatus 1 according to the present embodiment will be described with reference to FIG.

  The operation of the health guidance target person selection support apparatus 1 according to the present embodiment is divided into a learning phase, a score calculation phase, and a selection phase.

  In the learning phase, the following operations are performed. First, the input unit 15A inputs the past health diagnosis data 14A and the three types of flags 14B to 14D from the communication I / F unit 11 or the operation input unit 12, and stores them in the storage unit 14 (step S11).

  Next, the discrimination model learning unit 15B reads the past examination data 14A and the three types of flags 14B to 14D from the storage unit 14, and learns the three types of discrimination models 14F to 14H (steps S12 to S14). The order of learning may be arbitrary. In the example of FIG. 2, learning is performed in the order of the participation determination model 14F, the participation improvement determination model 14G, and the non-participation improvement determination model 14H. In the learning of the participation determination model 14F, the past health examination data 14A and the participation flag 14B are used. In the learning of the participation improvement discrimination model 14G, the past health diagnosis data 14A and the participation improvement flag 14C are used. In the learning of the non-participation improvement discrimination model 14H, the past medical examination data 14A and the non-participation improvement flag 14D are used.

  In the score calculation phase, the following operations are performed. First, the input unit 15A inputs the current period health check data 14E from the communication I / F unit 11 or the operation input unit 12, and stores it in the storage unit 14 (step S15).

  Next, the probability calculation unit 15C reads the current period health diagnosis data 14E and the learned three types of discrimination models 14F to 14H from the storage unit 14, and calculates three types of probabilities 14I to 14K (steps S16 to S18). . The order of calculation may be arbitrary. In the example of FIG. 2, the participation probability 14I, the participation improvement probability 14J, and the non-participation improvement probability 14K are calculated in this order. The calculation of the participation probability 14I of a certain individual is performed by inputting the health check data 14E of this individual in the participation determination model 14F. The calculation of the participation improvement probability 14J of a certain individual is performed by inputting the health diagnosis data 14E of this individual in the participation improvement determination model 14G. The calculation of the non-participation improvement probability 14K of a certain individual is performed by inputting the individual's current health check data 14E to the non-participation improvement discrimination model 14H.

  In the selection phase, the following operations are performed. First, the input unit 15A inputs the selection condition 14L from the communication I / F unit 11 or the operation input unit 12, and stores it in the storage unit 14 (step S19).

  Next, the guidance target person selecting unit 15D reads the three types of probabilities 14I to 14K and the selection condition 14L from the storage unit 14, and satisfies the selection condition 14L from the individual who is the candidate for the current health guidance target. A person is selected, and the selection result 14M is stored in the storage unit 14 (step S20). More specifically, the guidance target person selecting unit 15D preferentially selects individuals who have a higher probability of participating in health guidance and have a higher probability of improving through health guidance, while having a probability of participating in health guidance. The health guidance target person is selected so that individuals who are lower and have a higher improvement probability outside the health guidance are not preferentially selected. Then, the instruction target person selecting unit 15D outputs the above selection result 14M to the screen display unit 13 or outputs it to the outside through the communication I / F unit 11 (step S21).

Subsequently, operations of the learning phase, the score calculation phase, and the selection phase will be described in more detail. In the following, subscripts are expressed with an underline. For example, A _B is expressed as A_B. The superscript is expressed with a hat. For example, A ^B is written as A ^ B.

(1) Details of learning phase (1-1) Step S11
Past medical examination data 14A and three types of flags 14B to 14D are input. Let X_n (n = 1, 2, .., N) be the entered health checkup data. X_n is health check data of health guidance target person n. X_nj (j = 1,..., M) represents the measurement result of the health diagnosis item j of the health guidance target person n. M is the number of health examination items.

  Furthermore, let the inputted participation flag 14B, participation improvement flag 14C, and non-participation improvement flag 14D be Y_n, Z_n, S_n (n = 1,..., N). That is, Y_n is a flag indicating whether the health guidance target person n participated in the health guidance of the current year (Y_n = 1) or not (Y_n = 0). Z_n is whether or not the health guidance target person n participated in the health guidance for the current year and was excluded from the health guidance in the next year's health guidance (improvement) (Z_n = 1) or not (deteriorated) (Z_n = 0). The health guidance target person n of Y_n = 0 who did not participate in the health guidance in the current year is assumed to have Zn = 0. S_n was not subject to health guidance for the current year, and was not subject to health guidance for the next year (improvement) (S_n = 1) (deteriorated) (S_n = 0). The health guidance target n of Y_n = 1 who participated in the health guidance of the current year is assumed to have S_n = 0.

  As preprocessing, binarization of health check data is performed. A threshold for binarization is set for each health check item, and X_nj (j = 1,..., M) is binarized to 0 or 1 based on the threshold. The threshold value of each inspection item may use the health guidance criteria set by the Ministry of Health, Labor and Welfare. Note that this process is not necessary when the input health checkup data is already binarized.

(1-2) Steps S12 to S14
The values obtained from the discriminant model are the following P, Q, and R. A model for calculating this value is determined and its parameters are learned.

  P (Y_n = 1) represents the probability that the health guidance target person n participates in health guidance. The probability (non-participation probability) that the health guidance target person n does not participate in the health guidance is obtained by 1-P (Y_n = 1). This is P (Y_n = 0).

  Q is the probability of participating in health guidance and not (improves) health guidance in the next year's health guidance = participation improvement rate. Q (Z_n = 1) represents the probability that the health guidance target person n participates in the health guidance and is excluded (improves) in the health guidance in the next year. Q (Z_n = 0) represents a probability that the health guidance target person n participates in the health guidance and does not fall out of the scope of health guidance in the next year's health guidance (does not improve or deteriorates).

  R is the rate of improvement that does not participate in the health guidance for the current year and is excluded (improves) in the health guidance for the next year. R (S_n = 1) represents the probability that the health guidance target person n does not participate in the health guidance and is excluded (improves) in the health guidance in the next year. R (S_n = 0) represents a probability that the health guidance target person n does not participate in the health guidance and is not excluded from the health guidance target in the next year's health guidance (does not improve or deteriorates). A feature of the present embodiment is that P representing the probability that the health guidance target person n participates in health guidance is introduced.

  P, Q, and R are learned based on different models. Here, the learning of the P model will be described as an example.

  Here, X_n with Y_n = 1 is called a positive example, and X_n with Y_n = 0 is called a negative example. For example, a logistic regression model that can output a probability that a certain X_n is a positive example or a negative example may be used. The logistic regression model is a model that is often applied to a binary discrimination problem that discriminates whether X_n to Y_n = 1 or Y_n = 0. Hereinafter, the mathematical structure of logistic regression will be described.

Let X be an M-dimensional explanatory variable corresponding to the medical examination data, and Y be a random variable indicating whether or not he / she participated in health guidance in the current year (Y = 1 represents participation, Y = 0 represents non-participation) . When W is an M-dimensional weight vector, the logistic regression model is expressed by the following equation.
P (Y = 1 | X; W) = 1 / (1 + exp (W ^ {T} X))… (1)
P (Y = 0 | X; W) = 1-P (Y = 1 | X; W) (2)
However, P (● | ○; ★) represents the conditional probability of ● when ★ is given as a parameter and ○ is given. A superscript T represents transposition of a vector.

Logistic regression calculates the value of parameter W by optimizing the following objective function when given positive examples and negative examples {X_n, Y_n} (n = 1, ..., N) as learning data To do. However, X_n and Y_n are realized values of X and Y, respectively.
L (W) = \ sum ^ {N} _ {n = 1} logP (Y_n | X_n, W) −λ * || W || (3)
Here, λ is a parameter for adjusting the balance between the likelihood (first term on the right side) and the penalty term (second term on the right side), and | W | is the norm of W. Usually, norm 2 or norm 1 is used. \ Sum ^ {N} _ {n = 1} represents the total from n = 1 to N.

  Note that L (W) is a convex function related to W, and can be maximized by a method according to the gradient method. The parameter value that maximizes L (W) is W *. Here, the parameter W * of the model of P is W * _P.

  In the learning of the QW model, the health check data X_n of the health guidance subject n with Z_n = 1 is a positive example, the X_n of the health guidance subject n with Z_n = 0 is a negative example, and X_n to Z_n = 1 Or a logistic model that discriminates whether Z_n = 0. The parameter W * of the Q model is W * _Q.

  In the learning of the RW model, the health check data X_n of the health guidance subject n with S_n = 1 is a positive example, and the X_n of the health guidance subject n with S_n = 0 is a negative example. Or a logistic model that discriminates whether S_n = 0. The parameter W * of the R model is W * _R.

  The participation discrimination model 14F, the participation improvement discrimination model 14G, and the non-participation improvement discrimination model 14H, which are processing results of the learning phase, are expressed (defined) by the learned model parameters W * _P, W * _Q, and W * _R.

(2) Details of score calculation phase (2-1) Step S15
The health check data 14E of the selection subject for this term is input. Let the input data be X_n (n = 1, 2,..., N). X_nj (j = 1,..., M) represents the measurement result of the health check item j of the selection target person n. M is the number of health examination items.

  As preprocessing, binarization of health check data is performed. A threshold for binarization is set for each health check item, and X_nj (j = 1,..., M) is binarized to 0 or 1 based on the threshold. The threshold value of each inspection item may use the health guidance criteria set by the Ministry of Health, Labor and Welfare. If binarization has already been performed at the time of input, the binarization process is omitted.

(2-2) Steps S16 to S18
First, the probability P (Y_n = 1) that X_n participates in health guidance is calculated. Specifically, using W * _P, calculation is performed as P (Y_n = 1 | X_n; W * _P) = P (Y_n) for X_n.

  Also, the probability Q (Z_n = 1) that X_n participates in the health guidance and is excluded (improves) in the health guidance in the next fiscal year is calculated. Specifically, using W * _Q, calculation is performed as Q (Z_n = 1 | X_n; W * _Q) = Q (Z_n) for X_n.

  Further, the probability R (S_n = 1) that X_n does not participate in the health guidance for the current year and is excluded from (improves) the health guidance in the next year's health guidance is calculated. Specifically, using W * _R, calculation is performed as R (S_n = 1 | X_n; W * _R) = R (S_n) for X_n.

  The participation probability 14I, participation improvement probability 14J, and non-participation improvement probability 14K, which are the processing results of the score calculation phase, are calculated as P (Y_n), Q (Z_n), R (S_n) (n = 1,. N).

(3) Details of selection phase (3-1) Step S19
The constraint condition ST is input. Here, for example, the maximum number of health guidance participants may be used as the constraint condition ST. This constraint condition ST and P (Y_n), Q (Z_n), and R (S_n) (n = 1,..., N) calculated in the score calculation phase are input to the selection phase.

(3-2) Step S20
Explain the selection target as C name. K_c is a symbol indicating that the selection target person c is selected (K_c = 1) and not selected (K_c = 0). For example, if the selection target person 1 is not selected, 2 is selected,..., C is selected, K_1 = 0, K_2 = 1,.

Here, when the selection target is determined as K_1 = 0, K_2 = 1, ..., K_C = 1, the selection target of K_c = 0 does not participate in health guidance, and the selection target of K_c = 1 does not participate in health guidance The probability of participation is expressed as follows.
P '(K_1 = 0, K_2 = 1,…, K_C = 1) = (1-P (Y_1)) P (Y_2)… P (Y_C)… (4)

In addition, even if the selection target of K_c = 0 does not participate in the health guidance of the corresponding year, the probability of improvement by the health check of the next year, and the selection target of K_c = 1 participates in the health guidance of the relevant year and the health of the next year The probability of improvement by diagnosis is expressed as follows.
P ”(K_1 = 0, K_2 = 1,…, K_C = 1) = R (S_1) Q (Z_2)… Q (Z_C)… (5)

Taking the logarithm of each value, the above equations (4) and (5) become the following equations.
ln (P '(K_1 = 0, K_2 = 1,…, K_C = 1)) = ln (1-P (Y_1)) + ln (P (Y_2)) +… + ln (P (Y_C))… ( 6)
ln (P ”(K_1 = 0, K_2 = 1,…, K_C = 1)) = ln (R (S_1)) + ln (Q (Z_2)) +… + ln (Q (Z_C))… (7)

And if K_1 = 0, K_2 = 1, ..., K_C = 1, the selection target person is selected, K_c = 0 selection target person does not participate in health guidance, K_c = 1 selection target person participates in health guidance However, even if the selection target of K_c = 0 does not participate in the health guidance of the corresponding year, the health check of the next year will improve, and the selection target of K_c = 1 will participate in the health guidance of the relevant year, The probability of an event that improves with is given by:
L (K_1 = 0, K_2 = 1,…, K_C = 1) = P '(K_1 = 0, K_1 = 1,…, K_C = 1) P ”(K_1 = 0, K_2 = 1,…, K_C = 1 )
… (8)

Taking the logarithm of the above equation (8), the following equation is obtained.
ln (L (K_1 = 0, K_2 = 1,…, K_C = 1))
= ln (1-P (Y_1)) + ln (P (Y_2)) +… + ln (P (Y_C)) + ln (R (S_1)) + ln (Q (Z_2)) +… + ln (Q (Z_C))
… (9)
Here, ln (L (K_1 = 0, K_2 = 1,..., K_C = 1)) expressed by the above equation (9) is called an objective function.

  Maximizing the objective function is to select a selection target who has a high participation probability and a high probability of improvement through health guidance, has a high non-participation probability (1-participation probability), and has a high health improvement probability outside the health guidance. It is to avoid selecting the target person.

  Consider writing the objective function in another form.

First, as shown in FIG.
ln (P '(K_1, K_2,…, K_C)) = Const. + 1 ^ {T} (2P-1) K… (10)
It becomes.
Also, as shown in FIG.
ln (P ”(K_1, K_2,…, K_C)) = Const. + 1 ^ {T} (QR) K… (11)
It becomes.
Therefore,
ln (L (K_0 = 1, K_2 = 1,…, K_C = 1)) = (F ^ {T} + G ^ {T}) K + Const.… (12)
It becomes. Here, F = 1 ^ {T} (2P-1), G = 1 ^ {T} (QR).
Therefore, objective function = 1 ^ {T} R1 + 1 ^ {T} (QR) K (13)
It becomes.

  Now, if the restriction H of the number of health guidance participants is the constraint ST, calculate the value of ln (P (Y_n)) + ln (Q (Z_n)), and sort all users in descending order by this value, If you select H people from the top (K_n = 1 (n \ in (H people from the top))), you can find each selection target that maximizes the objective function.

  In addition, when the total cost J of the insurer's health guidance (budget) is the constraint ST, assuming that the health guidance cost per person is u, from H '= J / u, ln (P (Y_n)) Sort all users in descending order by + ln (Q (Z_n)) and select H 'people from the top (K_n = 1 (n \ in (H' people from the top))). Thus, it is possible to obtain each selection target person who maximizes the objective function.

  By the way, according to the selection criteria for health guidance subjects set by the Ministry of Health, Labor and Welfare, health guidance subjects are divided into two types: active supporters and motivation supporters. The health guidance menu established by the Ministry of Health, Labor and Welfare is different for active supporters and motivation supporters. Therefore, assuming that the total cost J of the insurer's health guidance (budget) is the constraint ST, the health guidance cost per active supporter is u1, and the health guidance cost per motivation supporter is u2. In order to maximize the objective function under the constraint ST = J, it is necessary to formulate an optimization problem and solve it as a linear programming problem as follows.

min
:-(F ^ {T} + G ^ {T}) K
sub.to:
1 ^ {T} AK <= J
K_n
> = 0 (n = 1, .., N)… (14)

Here, K is an N-dimensional vector having K_n \ in {0, 1} as elements. A is A_nn = u1 (if
K_n = active supporter) and A_nn = u2 (if X_n = motivation supporter) is a diagonal matrix having nn components.

  For example, when K_1 and K_C are active supporters and K_2 is a motivation supporter, A is a value shown in FIG.

  The selection result 14M, which is the processing result of the selection phase, is data of K_n (n = 1,..., N), that is, a list of instruction target / non-target persons.

  As described above, according to the present embodiment, the probability of participating in health guidance from among a plurality of individuals who are candidates for health guidance in the health examination of the current fiscal year, participating in health guidance and subject to health guidance in the next fiscal year. Select the health guidance target person to meet the selection conditions specified by the insurer based on the probability of being out of the candidate and the probability of not being in the health guidance candidate in the next fiscal year without participating in the health guidance. Can do.

[Second Embodiment]
Referring to FIG. 6, the health guidance target person selection support apparatus 2 according to the second embodiment of the present invention includes a plurality of individuals who are candidates for health guidance in health examinations performed every certain period. Therefore, it has the function of selecting individuals who are more likely to participate in health guidance as health guidance subjects. Here, the period is arbitrary, but in the following description, one year is one period.

  The health guidance target person selection support apparatus 2 includes a communication I / F unit 21, an operation input unit 22, a screen display unit 23, a storage unit 24, and a processor 25 as hardware.

  Communication I / F unit 21, operation input unit 22, and screen display unit 23 are the same as communication I / F unit 11, operation input unit 12, and screen display unit 13 in the first embodiment shown in FIG. It has a function.

  The storage unit 24 includes a storage device such as a hard disk or a semiconductor memory, and has a function of storing processing information and programs 24P necessary for various processes in the processor 25. The program 24P is a program that realizes various processing units by being read and executed by the processor 25, and can be read by an external device (not shown) or a computer via a data input / output function such as the communication I / F unit 21. It is read in advance from a possible storage medium (not shown) and stored in the storage unit 24. As main processing information stored in the storage unit 24, past medical examination data 24A, participation flag 24B, current period medical examination data 24C, participation determination model 24D, participation probability 24E, selection condition 24F, There is a selection result 24G.

  The past health examination data 24A, participation flag 24B, current term health examination data 24C, participation discrimination model 24D, participation probability 24E, selection condition 24F, and selection result 24G are the same as those in the first embodiment shown in FIG. This is the same as the health check data 14A, participation flag 14B, current term health check data 14C, participation determination model 14F, participation probability 14I, selection conditions 14L, and selection results 14M.

  The processor 25 includes a microprocessor such as a CPU and its peripheral circuits, and reads and executes the program 24P from the storage unit 24, thereby realizing the various processing units in cooperation with the hardware and the program 24P. have. As main processing units realized by the processor 25, there are an input unit 25A, a participation determination model learning unit 25B, a participation probability calculation unit 25C, and an instruction target person selection unit 25D.

  The input unit 25A has a function of inputting past health examination data 24A, participation flag 24B, and current period health examination data 24C from the communication I / F unit 11 or the operation input unit 12, and storing them in the storage unit 14. .

  The participation determination model learning unit 25B has a function of reading past health examination data 24A and participation flag 24B from the storage unit 24, learning the participation determination model 24D using these data, and storing them in the storage unit 24. Specifically, the parameters of each explanatory variable of the polynomial constituting the participation determination model 24D are learned and stored.

  The participation probability calculation unit 25C reads the participation determination model 24D and the current period health check data 24C from the storage unit 24, inputs the current period health check data 24C to the participation determination model 24D, A probability 24E of participating in health guidance for each individual is calculated and stored in the storage unit 24.

  The guidance target person selecting unit 25D reads the participation probability 24E and the selection condition 24F from the storage unit 24, and based on the participation probability 24E and the selection condition 24F, the selection condition 24F is selected from the individuals who are candidates for the current health guidance target. The health guidance target person is selected so as to satisfy, and the storage unit 24 has a function of saving it as the selection result 24G. The instruction target person selecting unit 25 </ b> D has a function of reading the selection result 24 </ b> G from the storage unit 24 and outputting the selection result 24 </ b> G to the screen display unit 23 or to the outside through the communication I / F unit 21.

  Next, the operation of the health guidance target person selection support apparatus 2 according to the present embodiment will be described with reference to FIG.

  First, the input unit 25A inputs the past medical examination data 24A, the participation flag 24B, the current medical examination data 24C, and the selection condition 24F from the communication I / F unit 21 or the operation input unit 22, and stores them in the storage unit 24. Store (step S31).

  Next, the participation determination model learning unit 25B learns a participation determination model 24D that calculates the probability that the individual participates in health guidance from the individual health diagnosis data using the past health diagnosis data 24A and the participation flag 24B. (Step S32). This learning is performed by the same method as the participation determination model 14F in the first embodiment.

  Next, the participation probability calculation unit 25C calculates the participation probability 24E for health guidance for each candidate for health guidance target for this term by inputting the health diagnosis data 24C for this term to the participation discrimination model 24D after learning. (Step S33).

  Finally, the guidance target person selecting unit 25D selects a health guidance target person so as to satisfy the selection condition 24F based on the calculated individual participation probability 24E, and stores the selection result 24G in the storage unit 24. (Step S34).

  For example, when the selection condition 24F is the maximum number of health guidance target persons, the guidance target person selection unit 25D sorts the individual participation probabilities 24E in descending order of probability, and the maximum number of parts from the top with the highest probability. Select individuals for health guidance. In addition, when the selection condition 24F is the total cost (budget) for the insurer to receive the health guidance, the total cost is divided by the health guidance cost per person to obtain the maximum number of health guidance target persons. The other participation probabilities 24E are sorted in descending order of probability, and a maximum number of individuals from the top with the highest probability are selected as health guidance subjects.

  Then, the instruction target person selecting unit 25D outputs the selection result 24G to the screen display unit 23, or outputs it to the outside through the communication I / F unit 21 (step S35).

  As described above, the present embodiment is based on the health examination data for the past year selected as the health guidance target person and the information on whether or not the health guidance for that year has been participated. Learn the model that estimates the participation probability, and then input the health checkup data of the candidates for the current year's health guidance into the learned model to calculate the participation probability in the health guidance, and based on the calculated participation probability The guidance target is selected. For this reason, the insurer can select the person to be instructed from the viewpoint of whether or not there is a possibility of participating in the health instruction.

  Although the present invention has been described with reference to some embodiments, the present invention is not limited to the above embodiments, and various other additions and modifications can be made. For example, in each of the above embodiments, learning of a discrimination model such as a participation discrimination model, calculation of a probability such as a participation probability, and selection of a person to be instructed are performed by the same processor. And may be executed in a distributed manner.

DESCRIPTION OF SYMBOLS 1 ... Health guidance object selection support apparatus 11 ... Communication I / F part 12 ... Operation input part 13 ... Screen display part 14 ... Memory | storage part 14A ... Past medical examination data 14B ... Participation flag 14C ... Participation improvement flag 14D ... Non-participation improvement Flag 14E ... Health checkup data for this term 14F ... Participation discrimination model 14G ... Participation improvement discrimination model 14H ... Nonparticipation improvement discrimination model 14I ... Participation probability 14J ... Participation improvement probability 14K ... Nonparticipation improvement probability 14L ... Selection condition 14P ... Selection result 14P ... Program DESCRIPTION OF SYMBOLS 15 ... Processor 15A ... Input part 15B ... Discrimination model learning part 15C ... Probability calculation part 15D ... Target person selection part

Claims (7)

The first health check data that is the health check data of the individual selected as the health guidance target in the past period and the first label indicating whether or not the individual has participated in the health guidance of the period A memory for storing values, second health check data that is personal health check data of individuals who are candidates for health guidance for this term, and selection conditions;
A processor connected to the memory,
The processor is
Using the first health check data and the first label value, learning a first discrimination model for calculating a probability that the individual participates in health guidance from the individual health check data;
The second health check data is input to the learned first discrimination model, and a first probability of participating in health guidance is calculated for each individual who is a candidate for health guidance in this term,
A health guidance target person selection support apparatus programmed to select a health guidance target person who satisfies the selection condition from individuals who are candidates for health guidance target persons in the current term based on the calculated first probability. The memory further includes a second label value indicating whether or not the individual has participated in the health guidance for the period and is not a candidate for health guidance in the next period; Storing the third label value indicating whether or not the candidate for health guidance was excluded from the next period without participating,
The processor is
In the learning, using the first health check data and the second label value, the individual participates in health guidance from the individual's health check data and is excluded from candidates for health guidance in the next period. And learning the second discriminant model for calculating the probability of becoming, and using the first health examination data and the third label value, the individual participates in health guidance from the individual health examination data. Without learning the third discriminant model that calculates the probability of not being a candidate for health guidance in the next period,
In the calculation, the second health check data is input to the learned second discriminating model, and the individual who is the candidate for the health guidance for this term participates in health guidance for the next period. For each individual who is a candidate for health guidance for the current term, the second probability of being outside the training target candidate is calculated and the second health diagnosis data is input to the learned third discrimination model. Calculate the third probability of not being a candidate for health guidance in the next period without participating in health guidance,
In the selection, based on the calculated first probability, second probability, and third probability, a health guidance target person satisfying the selection condition is selected from individuals who are candidates for the health guidance target of the current term. The health guidance target person selection support apparatus according to claim 1, which is programmed in the apparatus. In the selection, the processor preferentially selects individuals who have a higher probability of participating in the health guidance and a higher probability of improvement by the health guidance, and have a lower probability of participating in the health guidance and outside the health guidance. The health guidance target person selection support apparatus according to claim 2, programmed so that individuals with higher improvement probabilities are not preferentially selected. 2. The health guidance target person selection support apparatus according to claim 1, wherein the processor is programmed to preferentially select an individual who has a higher probability of participating in health guidance in the selection. 5. The health guidance target person selection support apparatus according to claim 1, wherein the selection condition is one of a maximum number of health guidance target persons and a maximum value of a budget that an insurer can spend on health guidance. First health check data that is health check data of the individual selected for the health guidance target in the past period and the first label indicating whether or not the individual has participated in the health guidance for the period A health guidance target comprising: a memory for storing a value, second health examination data that is personal health examination data of a candidate for health guidance for the current term, selection conditions; and a processor connected to the memory A health guidance target person selection support method executed by a person selection support device,
The processor is
Using the first health check data and the first label value, learning a first discrimination model for calculating a probability that the individual participates in health guidance from the individual health check data;
The second health check data is input to the learned first discrimination model, and a first probability of participating in health guidance is calculated for each individual who is a candidate for health guidance in this term,
A health guidance target person selection support method for selecting a health guidance target person who satisfies the selection condition from an individual who is a candidate for a health guidance target for the current term based on the calculated first probability. First health check data that is health check data of the individual selected for the health guidance target in the past period and the first label indicating whether or not the individual has participated in the health guidance for the period A processor connected to a memory for storing the value, the second health check data that is the health check data of the individual who is a candidate for health guidance for this term, and the selection condition;
Learning a first discriminant model for calculating a probability that the individual participates in health guidance from the individual's health diagnosis data using the first health diagnosis data and the first label value;
Inputting the second health check data into the learned first discriminating model, and calculating a first probability of participating in health guidance for each individual who is a candidate for health guidance in this term;
The program for performing the step which selects the health guidance object person who satisfy | fills the said selection conditions from the individual who is the candidate for health guidance object of this term based on the calculated 1st probability.

Images