JP3226097B2 - Treatment data estimation method and treatment data estimation system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a method and system for automatically processing data related to medical care or welfare. More specifically, the present invention relates to a method and a system used for determining the degree of nursing care / amount of nursing care and the degree of nursing care / amount of nursing care.

[0002]

2. Description of the Related Art Observation on medical treatment and living conditions,
A tree regression model is known as a data processing system for automatically estimating the degree of care or the required amount of care of a care recipient based on data obtained from the survey. Neural networks and linear autoregressive techniques can also be used for such systems. In addition, such a system is not limited to the degree of care or the required amount of care of the treatment target person, but can also be used for calculating the degree of care or the required amount of care.

[0003]

However, the above-mentioned conventional system has the following problems. In the following, for convenience of explanation, the outline and problems of the conventional technology will be described by taking an example of a system for estimating the degree of care or the required amount of care, but the same applies to the degree of care or the required amount of care.

[0004] Regarding the estimation of the degree of care or the amount of care required by a tree regression model, for example, "Yoichi Seki, Takako Tsutsui,
Estimation of multi-objective variable tree regression model based on MDL criteria,
Proceedings of the 20th Symposium of the Japan Society of Applied Statistics, 69-
74 (October 1998), and "Tsuko Tsutsui, Research on Elderly Care Needs and Method of Estimating Care Needs, Medical Economics Research, Vol. 3, pp. 117-129 (December 1996
Mon) has a detailed explanation.

[0005] This method has an advantage that a regression model can be constructed even when the type of the scale of the input variable is arbitrary and the effect on the output variable is not anticipated. On the other hand, since the output result is determined by the bisection method based on the value of each input data, there is a problem that the estimation error is large when there is an input variable having a continuous effect. Also,
Since a model is created by sequentially adopting input variables from those considered to have high explanatory power, there is a drawback that it is difficult to optimize the order of adoption.

[0006] The linear autoregressive model expresses the degree of care or the required amount of care as an output result as a linear function of observation and survey results on medical treatment conditions and living conditions as input data. In such a system, the number of parameters specifying the linear regression model is one in the dimension of the input data.
Is added. This model is constructed under the assumption that the relationship between input data and output results is a linear relationship. Therefore, when the relationship between the input data and the output result is not a simple linear relationship, there is a problem that the estimation error increases. In reality, the relationship between input data and output results is not always a linear relationship.

[0007] On the other hand, when the relationship between input data and output result is not linear,
This has the advantage that the regression surface can be configured appropriately. However, the number of parameters for specifying a neural network is much larger than that of linear autoregression, and there is a problem that optimizing parameters requires enormous data and calculation costs.

The present invention has been made in view of the above situation, and has as its object to provide a method and system capable of realizing care data estimation with high accuracy and low calculation cost.

[0009]

Another object of the present invention is to provide a method and a system capable of realizing nursing data estimation with high accuracy and low calculation cost.

[0011]

In order to solve the above problems, in a first method of the present invention, an input vector having each survey value or each measured value as a component and an output vector having a care data as a component are provided. The input vector / output vector pair relating to the subject who has been configured and has already been recognized is stored as a reference database. Then, the distance between the input vector of the subject whose care data is to be newly estimated and the input vector in the reference database is calculated at an arbitrarily set phase, and the number of reference input vectors having a short distance can be arbitrarily set. Just to search. Then, the nursing care data for the subject is calculated by multiplying the output vector corresponding to the searched reference input vector by a weight according to the distance and taking an average. The nursing care data includes, for example, the degree of nursing care required and the amount of nursing care.

[0012]

[0013] In the second method of the present invention, an input vector having each survey value or each measured value as a component and an output vector having nursing data as a component are formed, and an input relating to a subject who has already been certified is input. The vector / output vector pair is stored as a reference database. Then, the distance between the input vector of the subject whose nursing data is to be newly estimated and the input vector in the reference database is calculated at an arbitrarily set phase, and the number of reference input vectors having a short distance can be arbitrarily set. Just to search. Then, nursing data for the subject is calculated by multiplying the output vector corresponding to the searched reference input vector by a weight according to the distance and taking an average. The nursing data includes, for example, the degree of nursing required and the amount of nursing.

In the method of the present invention described above, since treatment data is estimated by directly referring to a database established in the past, there is an advantage that the reliability of the estimation result is high. Further, as the results of the treatment data estimation accumulate, the contents of the database are enhanced, and the estimation accuracy is automatically improved accordingly.

In each of the above-described methods, preferably, a predetermined weight coefficient is used when weighting the reference output vector. Then, the weight coefficient is sequentially changed so that the error between the calculated treatment data and the referred treatment data is minimized. In other words, to realize a learning function as the whole system. As a result, the accuracy of the estimation processing is automatically improved.

[0016]

The first system of the present invention comprises an input vector having each survey value or each measured value as a component and an output vector having nursing data as a component. Data storage means for storing a vector pair as a reference database; calculating a distance between an input vector relating to a subject whose care data is to be newly estimated and an input vector in the reference database at an arbitrarily set phase; By arbitrarily setting the number of reference input vectors to be searched, multiplying the output vector corresponding to the searched reference input vector by a weight according to the distance and taking an average, the care data for the subject is obtained. Calculation means for calculating. The nursing care data includes, for example, the degree of nursing care required and the amount of nursing care.

Further, the second system of the present invention forms an input vector having each survey value or each measured value as a component and an output vector having nursing data as a component, and has an input vector relating to a subject who has been certified. Data storage means for storing the output vector pair as a reference database; calculating the distance between the input vector of the subject whose nursing data is to be newly estimated and the input vector in the reference database at an arbitrarily set phase; Numerical nursing for the subject is performed by searching for a reference number of reference input vectors arbitrarily settable, and averaging the output vectors corresponding to the searched reference input vector by weighting according to the distance. Computing means for calculating data. The nursing data includes, for example, the degree of nursing required and the amount of nursing.

In each of the above-described systems, preferably, a predetermined weight coefficient is used when weighting the reference output vector. Then, so that the error between the calculated treatment data and the referred treatment data is minimized,
The apparatus further includes weight coefficient optimization means for sequentially changing the weight coefficient.

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention realizes a system for estimating a degree of need for long-term care, which has a learning function in which the determination of model parameters is easier and the estimation error is smaller than in the prior art, and the estimation accuracy is improved with the expansion of the database. . The basic concept is described below. In the following, for convenience of explanation, an example of a system for estimating the degree of nursing or the required amount of nursing will be described, but it goes without saying that the present invention can be applied to a system for estimating the degree of nursing or the required amount of nursing.

FIG. 1 shows the basic concept of the present invention (a series of processes).
Is shown. In the present invention, for example, “G.
ara and R.A. M. May "Nonlinear forecas
tingas a way of distinguishing chaos from m
easurement error in time series ”, Natur
e, Vol. 344, p. p. 734-174 (199
0)) is improved so that it can be used for pattern matching. In pattern matching, nominal variables obtained by listening surveys are used as input information. Here, D regarding medical treatment status and living status
Think of input data as a set of individual survey items as a vector. It is assumed that a measurement result regarding the required amount of care or the degree of care is present for this input vector. Assume that such data has been obtained for N subjects.

The input vector is displayed as in the following equation, and the required amount of care or the degree of care as a corresponding output value is displayed as "y _i ". In the following equation, “d” is an index (d = 1,..., D) representing the survey item, and “i”
Is an index (i = 1,..., N) for identifying the survey subject. The already obtained input vector / output value pair is used as a reference database. When a new process is executed,
An input vector / output value pair as a processing result is added to this database.

[0023]

(Equation 1)

Now, it is assumed that an item corresponding to the input vector is newly investigated, and there is a care recipient who wants to calculate the degree of care or the required amount of care based on the result of the survey. This target person is particularly called “target person P”. The level of care or care required amount of the subject P is referred to as "y _p". Also, the input vector is represented as “X _p ”.

The degree of care or the required amount of care "y"_pTo the following
Estimate based on procedure. Input vector is D-dimensional
Think in Euclidean space. Note that the phase of the space is
It is not limited to the grid distance. Reference database
Input vector "X_iEnd point corresponds to the observation point
In this case, the distribution of the end points indicates the fulfillment of the database. "X
_iIs considered as a D + 1-dimensional polyhedron having a vertex as a vertex. Like that
Among the various polyhedrons, "X _pMinimum facet surrounding the end point of "
Explore the body. This is "X_i"And" X_p”
It can be performed by calculating and comparing. In this way
"X_pThe vertex of the smallest polyhedron surrounding the end point of
D + 1 "X" with points_{i (d)}(D = 1, ...,
D + 1) can be found. As a result,
Or nursing need "y_pIs estimated by the following equation.

[0026]

(Equation 2)

[0027]

(Equation 3)

In the above equations (1) and (2), “α” represents an appropriate coefficient, and “Q _d ” represents a distance between “X _{i (d )} ” and “X _p ”. Note that a Gaussian function exp (-αx ² ) can be used instead of exp (-αx) representing a weight coefficient. As the value of “α”, α = 1 may be set. In this way, the required amount of care or the degree of care of the new care recipient P can be automatically estimated at high speed by directly referring to the database obtained in the past. Since the present invention uses nominal variables of different categories as input variables, unlike time-series prediction, this method is regarded as pattern matching between reference data and new data.

To optimize the coefficient “α”, the following operation
Should be executed. From a reference database on N people
Input vector and output of M people at random using uniform random numbers
Pick out value pairs. These sets are called LIBRARY,
X_m, Y_m(M = 1,..., M). Remaining
Set consisting of input vector / output value pairs for NM persons
Is called PREDICT and X_q, Y_q(Q = 1, ..., N-
M). Next, use LIBRARY as reference data.
Based on the equations (1) and (2) under α = 1,
"X_qIs estimated. This is called "z_q"When
indicate. α = 1 is the initial value of “α”, but other values
It can also be used as an initial value. The predicted value “z _q"When
Actual measurement value "y_qIs evaluated by the following equation (3).
You.

[0030]

(Equation 4)

By changing “α” so that “E” shown in the above equation (3) is minimized, “α” can be optimized. It is considered that the steepest descent method or the conjugate gradient method can be applied as a general method for executing this. However, if “E” given by equation (3) is regarded as a function of “α”, “E” may have many minimum points for “α”. Therefore, there is a possibility that the optimum “α” cannot be searched for by the steepest descent method or the conjugate gradient method. To solve this problem, a method devised for learning neural networks or generalized radial basis function networks (B.
Caprile and F. Girosi, `` A nondeterministic minim
ization algorithm '', AI Memo (Artificial Intel
ligence Laboratory, Massachusetts Institute of Te
chnology), No. 1254 (September 1990). The procedure is shown in the following equation (4).

[0032]

“E” calculated for the initial value “α _old ” of “α” is stored as “E _old ”. next,
An independent uniform random number is generated within a range of “−ω or more and ω or less”, added to “α _old ”, and stored as “α _new ”.

Next, "α_newThe same process for
Is used to calculate the estimation error._new".
"E_newIs "E"_oldIs smaller than ""
_{ne w}”Is regarded as the initial value and the range is“ −2ω or more and 2ω ”.
Generates an independent uniform random number “ξ” based on Equation (4).
"Α_newIs determined. Conversely, "E_newIs "E"_o
ldIs greater than "α_old"Is regarded as the initial value.
None, independent uniform disturbance in the range of -0.5 Ω or more and 0.5 Ω
A number “数” is generated, and “α” is calculated based on Expression (4)._{ne w}"
decide. Repeat this procedure. And
"E_newIs the desired value "E _min"
Then, the iterative procedure is terminated, and the optimization process of “α”
To end. In case it does not fall below the desired value,
Set the maximum number of repetitions C and the number of repetitions exceeds C
Sometimes we end the optimization process. like this
FIG. 2 shows the flow of the optimization process.

[0035]

FIG. 3 shows an example of a configuration in which the present invention is implemented as a system for estimating a state requiring nursing care. This estimating system includes a data input / output unit 10, a reference database 12, a care degree estimating unit (arithmetic unit) 14, and an α optimizing unit 16. A reference data memory 18 and a prediction data memory 20 are mounted inside the α optimization unit 16. In the data input / output unit 10, new input data obtained by a survey or measurement is input, and a care-requiring state, which is an output value estimated by the care-degree estimation unit 14 for the input data, is displayed.

The input information supplied to the data input / output unit 10 is, for example, Takako Tsutsui, "[Introduction] Nursing Care Service Management (Nihon Keizai Shimbun)", Chapter 4, 152-1.
Page 53 (December 1998) "is selected as appropriate. In this case, the dimension of the input vector is 80 dimensions.

For example, it is assumed that "all assistance in changing operation" is selected as input information. The input value is 1 if all assistance is necessary, and -1 if not. Variables synthesized based on these input data can be used as components of the input vector. The database 12 stores the results of the 80 items of the same category for 10,000 care recipients and the results of the final determination of the state of need for nursing care corresponding to those results as input / output value pairs.

The final judgment result of the care-requiring state is expressed as one of the numbers 0, 1, 2, 3, 4, 5, and 6 as the care level. However, it is also possible to apply the total care time as the output data and express it as a real number of 0 or more in seconds. As an observation method of the total care time, for example,
A one-minute time study survey described in Takako Tsutsui, "[Introduction] Care Service Management" (Nihon Keizai Shimbun), Chapter 4, pages 116 to 118 (December 1998) "can be used.

In the care degree estimating section 14, the data input / output section 1
The state shown in FIG. 1 is estimated based on the reference data called from the reference database 12 for the survey result input from 0, thereby estimating the care-requiring state. Then, the estimated care-requiring state is transmitted to the data input / output unit 10 as the estimated care level or the estimated total care time. This estimation result is displayed on the data input / output unit 10. When the final determination result is determined, it is added to the reference database 12 and stored as a new input / output value pair.

The optimization of “α” is executed by the α optimization unit 16. The α optimizing unit 16 performs the processing shown in FIG. 2 while exchanging data with the reference database 12 and the nursing degree estimating unit 14 and performs optimization of “α”.
Then, the result is supplied to the care degree estimating unit 14. The care degree estimating unit 14 stores the optimized “α” in the memory.

When the present invention is implemented as a nursing degree estimation system, a system having the same configuration as that of FIG. 4 is used. That is, the nursing degree estimating system includes a data input / output unit 10, a reference database 12, a nursing degree estimating unit (arithmetic unit) 14,
α optimization unit 16. The reference data memory 18 and the prediction data memory 2
0 is mounted. In the data input / output unit 10, new input data obtained by a survey or measurement is input, and a nursing-required state, which is an output value estimated by the nursing degree estimating unit 14 for the input data, is displayed.

The input information supplied to the data input / output unit 10 includes, for example, Akihiro Igata, “Study on Nursing Work Volume Measurement by Comprehensive Payment Method”, pp. 151-162.
(1998 Welfare Science Research Policy Science Promotion Research Project,
About 300 items can be appropriately selected from the nursing service classification codes described in (March 1999).

In this case, the dimension of the input vector is 300. For each task item, the input value is 1 if the corresponding nursing is necessary, and -1 if not. Variables synthesized based on these input data can be used as components of the input vector. The reference database 12 stores 300 items of survey results of the same category for 10,000 nursing subjects and nursing degrees of 12 categories of Ministry of Health and Welfare corresponding to the survey results as input / output value pairs. Nursing levels are 1, 2, 3, 4, 5,
6, 7, 8, 9, 10, 11, and 12 are represented. However, it is also possible to use the total nursing time as output data and express it as a real number of 0 or more in seconds. For example, Takako Tsutsui, “[Introductory] Nursing Service Management”
(Nikkei Inc.), Chapter 4, pp. 116-118 (19
(December 1998) "can be used.

In the nursing degree estimating unit 14, the data input / output unit 1
The nursing level is estimated by performing the processing shown in FIG. 1 based on the reference data called from the reference database 12 for the survey result input from 0. The estimation result is supplied to the data input / output unit 10 as the degree of nursing or the estimated total nursing time, and is displayed. When the final determination result is determined, it is added to the reference database 12 and stored as a new input / output value pair. The optimization of “α” is executed by the α optimization unit 16. The α optimization unit 16 communicates with the reference database 12 and the nursing degree estimation unit 14 while
The process shown in FIG. 2 is performed to optimize “α”. Then, the result is transmitted to the nursing degree estimating unit 14.
The nursing degree estimating unit 14 stores the optimized “α” in the memory.

As is clear from this embodiment, since the degree of care or nursing is estimated by directly referring to the database established in the past, the reliability of the estimation result is high.
There is only one parameter that specifies the care degree or nursing degree estimating unit. Since all the estimation processing is executed by the electronic device, the processing speed is high. Further, since the database is enhanced as the results of the care degree determination are accumulated, the estimation accuracy is automatically improved accordingly. Further, optimization of “α” can always be performed, and a learning function is realized as a whole system.

The embodiments of the present invention have been described above. However, the present invention is not limited to these embodiments, and can be modified within the scope of the technical idea described in the claims. is there. In the above embodiment, the degree of nursing / necessary nursing care and the degree of nursing / necessary nursing care have been described. However, the present invention can be applied to estimation and calculation methods for various treatment data. For example, it is needless to say that the present invention can be applied to estimation of a customer's claim grade, management of a quality grade of an industrial product, and the like.

[0047]

As described above, according to the present invention, the model parameters can be easily determined, the estimation error is small, and the database is expanded and the estimation accuracy is improved. In other words, there is an effect that estimation of appropriate treatment data can be realized with high accuracy and low calculation cost.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a processing procedure of a method for estimating a required care amount or a care degree according to an embodiment of the present invention.

FIG. 2 is a flowchart illustrating a first half of an optimization processing procedure of “α” in a method of estimating a required care amount or a care degree according to an embodiment of the present invention.

FIG. 3 is a flowchart showing the latter half of the procedure for optimizing “α” in the method for estimating the required amount of care or the degree of care according to the embodiment of the present invention.

FIG. 4 is a block diagram illustrating a schematic configuration of a system according to an embodiment of the present invention.

[Explanation of symbols]

 12 Reference Database 14 Care Level Estimation Unit (Calculation Unit) 16 α Optimization Unit

──────────────────────────────────────────────────続 き Continuing from the front page (73) Patent holder 599102435 Chihaya T3 Co., Ltd. 13 Tansu-cho, Shinjuku-ku, Tokyo (72) Inventor Naoya Miyano 1-1, Oji-gakuen-cho, Hirosaki-shi, Aomori Pref. -4-56 (72) Inventor Takako Tsutsui 1-14-12-701 Chihaya, Toshima-ku, Tokyo (72) Inventor Yoichi Seki 151 3-Hitsuinaba 303, Motosoja-cho, Maebashi City, Gunma Prefecture (72) Inventor Hitoshi Taniguchi 13 Tanasu-cho, Shinjuku-ku, Tokyo Chihaya T3 Co., Ltd. (56) References JP-A-6-301664 (JP, A) Special Table 11-11-504739 (JP, A) (58) Field (Int.Cl. ⁷ , DB name) G06F 17/60 126 G06F 17/30 170 G06F 17/30 350 A61B 5/00

Claims (8)

(57) [Claims] 1. A treatment data estimating method for inputting a quantified survey or measurement result of nursing contents and medical contents relating to a target person whose degree of need of nursing care is to be certified and obtaining quantified nursing data of the target person. Constructing an input vector having each survey value or each measured value as a component and an output vector having nursing data as a component, and storing, as a reference database, an input vector / output vector pair relating to the subject who has been certified; Calculates the distance between the input vector of the subject whose care data is to be newly estimated and the input vector in the reference database at an arbitrarily set phase, and searches for the arbitrarily configurable number of reference input vectors with a short distance. And calculating an average by multiplying the output vector corresponding to the searched reference input vector by a weight according to the distance. Te,
Calculating the care data for the subject. 2. The method according to claim 1, wherein a predetermined weighting factor is used when weighting the reference output vector, and the error between the calculated estimated care degree and the referenced certified care degree is minimized. 2. The method for estimating treatment data according to claim 1, wherein the weighting factors are sequentially changed. 3. A treatment data estimating method for inputting a quantified survey or measurement result of nursing contents and medical contents relating to a subject whose nursing necessity is to be recognized and obtaining quantified nursing data of the subject. Constructing an input vector having each survey value or each measurement value as a component and an output vector having nursing data as a component, and storing, as a reference database, an input vector / output vector pair relating to the subject who has been certified; Calculates the distance between the input vector of the subject whose nursing data is to be newly estimated and the input vector in the reference database at an arbitrarily set phase, and searches for the arbitrarily configurable number of reference input vectors with a short distance. Averaging by multiplying the output vector corresponding to the searched reference input vector by a weight according to the distance. I,
Calculating nursing data for the subject. 4. A method for weighting the reference output vector, wherein a predetermined weighting coefficient is used, and the error between the calculated estimated nursing degree and the referred authorized nursing degree is minimized. 4. The treatment data estimation method according to claim 3, wherein the weight coefficient is sequentially changed. 5. A treatment data estimating system for inputting a digitized survey or measurement result of nursing contents and medical contents relating to a subject for whom the degree of nursing care is to be certified and obtaining digitized nursing data of the subject. A data storage means for constructing an input vector having each survey value or each measurement value as a component and an output vector having nursing data as a component, and storing, as a reference database, an input vector / output vector pair relating to the subject who has already been certified; The distance between the input vector of the subject whose care data is to be newly estimated and the input vector in the reference database is calculated at an arbitrarily set phase, and the number of reference input vectors having a short distance can be arbitrarily set. And the output vector corresponding to the searched reference input vector is weighted according to the distance to obtain an average. By treatment data estimation system, characterized in that a calculating means for calculating a care data with respect to the subject. 6. The arithmetic means uses a predetermined weighting factor when weighting the reference output vector, and minimizes an error between the calculated estimated care degree and the referenced certified care degree. The treatment data estimation system according to claim 5, further comprising a weight coefficient optimizing means for sequentially changing the weight coefficient. 7. A treatment data estimating system for inputting a digitized survey or measurement result of nursing contents and medical contents relating to a subject whose nursing degree is to be certified and obtaining quantified nursing data of the subject. A data storage means for constructing an input vector having each survey value or each measured value as a component and an output vector having nursing data as a component, and storing, as a reference database, an input vector / output vector pair relating to the subject who has been certified; The distance between the input vector of the subject whose nursing data is to be newly estimated and the input vector in the reference database is calculated at an arbitrarily set phase, and the number of reference input vectors having a short distance can be arbitrarily set. And the output vector corresponding to the searched reference input vector is weighted according to the distance to obtain an average. By treatment data estimation system, characterized in that a calculating means for calculating a nursing data for that subject. 8. The arithmetic means uses a predetermined weighting factor when weighting the reference output vector, and minimizes an error between the calculated estimated nursing degree and the referred authorized nursing degree. The treatment data estimating system according to claim 7, further comprising a weight coefficient optimizing means for sequentially changing the weight coefficient.