JPH11296598A - System and method for predicting blood-sugar level and record medium where same method is recorded - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the system and method for blood-sugar level prediction which are enabled to predict the level of sugar in the blood daily on the basis of measurement data on the level of sugar in the blood so as to a doctor can obtain support information for determining a proper insulin dosage without any time lag, and to provide the recording medium where the method is recorded. SOLUTION: Sugar-blood level measurement data on a diabetic are stored as sugar-blood level time-series data to a sugar-blood level time-series file 2 and a dynamics estimation part 3 stores an embedding dimension (n) and a delay time τ capable of representing phase properties most suitably that the sugar-blood level time-series data have to a parameter file 4; and a sugar-blood level prediction part 5 predicts the level of sugar in the blood of the near future by a local fuzzy reconstituting method on the basis of the sugar-blood level time-series data and parameters and stores it to a predicted sugar-blood level file 6, an inslin dosage time-series file 8 stores inslin doages as time-series data, and a display part 9 displays data of the respective files as information for determining an inslin dosage.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and system for predicting a blood sugar level of a diabetic patient by computer processing.

[0002]

2. Description of the Related Art In the treatment of a diabetic patient, the dosage of insulin is adjusted based on the blood sugar level of the patient. Blood glucose management is performed in an open circle in which the patient or the physician only measures the blood glucose level, or insulin is administered once a month or two weeks with the feeling of the physician based on the blood glucose measurement data. A feedback method of adjusting the amount is employed. Also, the insulin dose may be adjusted on a daily basis by negotiating an insulin scale.

[0003]

The treatment of a physician for insulin therapy for a diabetic patient is one of the following.

(1) Based on the blood sugar level measurement data, an insulin dose is determined about twice a month based on the experience and feeling of a doctor.

[0005] (2) An insulin dose for a blood glucose level is determined, and the test is performed once to three times a day based on this scale.

[0006] In these treatment methods, the control of the blood sugar level involves feedback with a large time lag, so that the blood sugar level change may become unstable. For example, increasing the insulin dose to lower the average blood glucose level may result in hypoglycemia. Conversely, reducing the insulin dose can cause blood sugar levels to be too high.

[0007] Under such circumstances, in order to determine an insulin dose for obtaining an appropriate blood sugar level control effect based on blood sugar level measurement data, blood glucose level control without a time lag requires daily measurement of blood glucose level. It is demanded to keep the change within a proper range in the long term while minimizing the change in the temperature.

[0008] An object of the present invention is to provide a blood glucose level that can predict a daily blood glucose level based on blood glucose level measurement data so that a doctor can obtain support information for determining an appropriate insulin dose without a time lag. It is an object of the present invention to provide a value prediction system, a blood glucose level prediction method, and a recording medium on which the method is recorded.

[0009]

DISCLOSURE OF THE INVENTION The present invention analyzes the instability of blood glucose control, elucidates the existence of chaos in the behavior of blood glucose over time based on this analysis, and uses the local fuzzy reconstruction method to analyze the current blood glucose. The blood sugar level in the near future (after tomorrow) can be predicted from the value. These matters will be described below. The near future prediction by the local fuzzy reconstruction method has already been proposed by the present inventors (Japanese Patent Laid-Open No. 7-239838).

(Time-dependent behavior of blood sugar level and chaos phenomenon)
Diabetes patients whose blood glucose data were analyzed were insulin-dependent (IDDM) 5 cases and non-dependent (NIDDM)
Five cases. The time series measurement data of these patients for a minimum of one and a half years to a maximum of ten years was measured at daily intervals. FIG.
Shows part of the time-series data of one NIDDM case and two IDDM cases showing good glycemic control.

The percentage of HbA _1C that is used clinically as a control index is the clinically approximated average of glycemic control over the past 1-2 months. Case 1 which becomes the data of FIG. 2 is HbA
_1C was 5-6%, Case 2 was 5-6% HbA _1C , Case 3 was 9-10% HbA _1C , and the control state was almost constant during the course.

Case 1 is considered to be non-insulin-dependent diabetes mellitus, and the blood glucose control function via the endogenous insulin secretion is insufficient but remaining. Cases 2 and 3 are considered to have insulin-dependent diabetes with insulin secretion capacity close to 0 and blood sugar regulation function by endogenous insulin close to 0.

[0013] The data of these three cases was converted to FFT (Fa
When spectrum analysis was performed using (stFourierTransform), frequency components appeared in a wide band. In addition, when the autocorrelation function is taken, almost 0
Converged. The maximum Lyapunov exponent is positive and 3
One case was chaotic.

Next, the attractors projected onto the three-dimensional space for these three cases are shown in FIG. The attractor shows a columnar shape in Case 1, a triangular pyramid shape in Case 2, and a spherical shape in Case 3. Each fractal dimension is 2.27 in case 1, 2.73 in case 2, and 3.54 in case 3, indicating that the fractal dimension increases as the shape of the attractor becomes more complicated.

In these three cases, the cylinder of case 1 and the triangular pyramid of case 2 were good and the same degree in the evaluation of the control level by HbA _1C , and the attractors differed in shape between IDDM and NIDD.
This is probably due to the difference in M's ability to regulate blood sugar.

Both the pyramidal pyramid of Case 2 and the sphere of Case 3 are IDDM and are controlled by the same continuous insulin subcutaneous infusion therapy (CSII), and the control level is good (good control) or insufficient (poor). control).

The same examination was performed for all other DM cases, but all cases showed chaos, and the attractor shape was any one of these three types or a mixed shape.

These three types of attractors also change the number of data, and when observed from various directions, are actually based on a spiral shape as shown in FIG. 4, and this spiral is probably based on three or four small parameters and noise. It is thought to change into a triangular pyramid, cylinder, sphere, etc.

It has been inferred from many other cases that the parameters are present in the residual or control level of the ability to control blood glucose through endogenous insulin.

As described above, the time-dependent behavior of the blood glucose level of a diabetic patient appears at first glance to be an irregular phenomenon, that is, a nondeterministic phenomenon governed by chance, but its behavior can be determined deterministically. I was able to elucidate the phenomenon, a deterministic chaotic phenomenon.

(Prediction of Blood Sugar Level by Local Reconstruction Method) In the deterministic chaos phenomenon, if nonlinear deterministic regularity can be estimated, the "sensitive sensitivity to the initial value" of chaos is obtained from observation data at a certain point in time. Thus, it is possible to predict data in the near future until losing deterministic causality.

The near future prediction for such a deterministic chaotic phenomenon is based on Takens' theory that "the state space and attractors of the original dynamical system are reconstructed from one observation time series data". . The outline of this theory is as follows.

From the observed time series data y (t), vectors (y (t), y (t−τ), y (t−2)
τ), y (t− (n−1) τ) (where τ is the delay time). This vector would indicate a point n-dimensional reconstructed state space R ^n.

Therefore, by changing t, a trajectory can be drawn in this n-dimensional reconstructed state space. If,
A target system deterministic dynamical system, the observed time series data assuming that obtained through the observation system corresponding to C ¹ continuous function from the state space of the dynamical system into a one-dimensional Euclidean space R For this reconstruction trajectory, if n is sufficiently large, the embedding of the original deterministic system (embeddin
g).

That is, if any attractor appears in the dynamical system, an attractor that preserves the phase structure of the attractor is reproduced in the reconstructed state space. n is usually referred to as “embedded dimension”. In order for the reconstruction operation to be “embedded”, when the dimension of the state space of the original dynamical system is m, the following equation is satisfied. Has proven to be sufficient.

[0026]

## EQU1 ## However, this is a sufficient condition, and depending on data, 2 m
Embedding may be performed even if the value is less than +1. Further, n>
If 2d (where d is the box count dimension of the attractor of the original dynamical system), it is also shown that the reconstruction operation is a one-to-one mapping.

As described above, since the change in the blood sugar level is a deterministic chaotic phenomenon, the time series data of the blood sugar level is reconstructed on the basis of the Taken's embedding theorem and the reconstruction state space and the attractor. Further, it is possible to predict the blood sugar level in the near future based on this attractor.

More specifically, as shown in FIG.
Reconstruction is performed by embedding the time-series data y (t) of the blood glucose level observed at equal sampling intervals into an n-dimensional state space with an embedding dimension n and a delay time τ using Takens' embedding theorem. The vector is obtained.

[0029]

X (t) = (y (t), y (t−τ),..., Y
(T− (n−1) τ) where t = ((n−1) τ + 1) to YY: the number of time-series data y (t) This operation is repeated for a large number of y (t) data. And a smooth manifold consisting of a finite number of data vectors in the n-dimensional reconstructed state space. FIG. 5B shows the trajectory of the attractor when embedded in the three-dimensional reconstruction state space.

With respect to the trajectory of this attractor, the near future trajectory of the current data vector is estimated by using the data vector including the time series data of the blood glucose level measured most recently and the trajectory of the data vector in the vicinity thereof. The data vector at the step destination can be obtained. That is, from the current blood glucose level data vector and its neighboring data vector, a predicted value of the blood glucose level in the near future (after tomorrow) can be obtained from the current blood glucose level data. This is a local reconstruction.

That is, as shown in FIG. 6, a data vector z (T) obtained by the latest data is plotted in an n-dimensional reconstructed state space, and a data vector in the vicinity thereof is x (i). Since the data x (i) is past data, the state x (i + s) s steps ahead is known. Using this, the current data vector z
A predicted value z (T + s) s steps ahead of (T) can be predicted. Then, a predicted value y (t + s) s steps ahead of the original time-series data can be obtained from the predicted value z (T + s).

(Prediction of blood glucose level by local fuzzy reconstruction method) In the prediction by the local reconstruction method, the state x
The change to the state x (i + s) after s steps of (i) is
It is thought to be based on deterministic dynamics. The dynamics are x (i) and x (i +
Using s), it can be expressed in a linguistic expression as follows. Where i∈N (z (T)) and N (z (T)) are z
A set of indices i of neighborhood x (i) of (T).

[0033]

(Equation 3) x (T): a set representing data vectors near z (T) in the n-dimensional reconstructed state space x (T + s): a set representing data vectors after s steps of x (T) x (i) is z ( Since the data vector is in the vicinity of T), if the step s is before the loss of deterministic causality due to the "sensitive dependence on the initial value" of chaos, the state z (T)
From, it can be assumed that the dynamics of state z (T + s) are approximately equivalent to the dynamics of states x (i) through x (i + s).

When the attractor embedded in the n-dimensional reconstructed state space is a smooth manifold, z (T) to z
The vector distance to (T + s) is affected by the vector distance from z (T) to x (i). In other words, it can be considered that the closer to the trajectory of x (i) from z (T), the larger the effect on the trajectory from z (T) to z (T + s), and the farther the trajectory is, the smaller the effect.

By the way,

[0036]

X (i) = (y (i), y (i−τ),..., Y (i− (n−1) τ)) x (i + s) = (y (i + s), y (i + s) −τ),..., Y (i + s− (n−1τ)) (1) Therefore, focusing on the j-axis in the n-dimensional reconstructed state space, the expression (1) becomes

[0037]

## EQU00005 ## IF aj (T) is yj (i) THEN aj (T + s) is y (i + s) (j = 1 to n) (2) where aj (T): a neighborhood value of z (T) The j-axis component of x (i) in the n-dimensional reconstruction state space aj (T + s): the j-axis component of x (i + s) in the n-dimensional reconstruction state space n: embedding dimension number

The trajectory from z (T) to z (T + s) is
Although affected by the vector distance from z (T) to x (i), this effect is expressed in a non-linear manner since the attractor that is the trajectory of this vector is a smooth manifold.
Therefore, in order to make the effect nonlinear, if Expression (2) is expressed by a fuzzy function,

[0039]

## EQU00006 ## IF aj (T) is y'j (i) THEn aj (T + s) is y'j (i + s) where (j = 1 to n) (3) Normally, the function y (i) is In the case of fuzzy conversion, the symbol "~" is used. Here, the symbol "'" is used.

Also,

[0041]

(7) z (T) = (y (T), y (T−τ),..., Y (T−
(n-1) τ)), so that j (T) in the n-dimensional reconstructed state space
The axis component is yj (T). Therefore, the data vector z (T)
If the predicted value of the data vector z (T + s) after s steps is z ″ (T + s), the j-axis component is aj in equation (3).
By substituting yj (T) for (T) and performing fuzzy inference, it can be obtained as aj (T + s). This method is called "Local Fuzzy Reconstruction"
I will call it the law.

As a specific example, the embedding dimension n
= 3, the delay time τ = 4, and the number N of data vectors included in the vicinity N = 3.

Each data vector is represented by

[0044]

(8) z (T) = (y1 (T), y2 (T-4), y3 (T-8)) x (a) = (y1 (a), y2 (a-4), y3 (a) -8)) x (b) = (y1 (b), y2 (b-4), y3 (b-8)) x (c) = (y1 (c), y2 (c-4), y3 (c −8)) z ″ (T + s) = (y1 (T + s), y2 (T + s−4), y3 (T + s−8)) x (a + s) = (y1 (a + s), y2 (a + s−4), y3 ( a + s-8)) x (b + s) = (y1 (b + s), y2 (b + s-4), y3 (b + s-8)) x (c + s) = (y1 (c + s), y2 (c + s-4), y3 ( c + s-8)), the fuzzy rule expressed by the equation (3) is obtained by the equation (4)
(5) and (6).

For the first axis of the reconstructed state space,

[0046]

IF a1 (T) is y′1 (a) THEN a1 (T + s) is y′1 (a + s) IF a1 (T) is y′1 (b) THEN a1 (T + s) is y′1 ( b + s) IF a1 (T) is y′1 (c) THEN a1 (T + s) is y′1 (c + s) (4) For the second axis of the reconstructed state space,

[0047]

## EQU10 ## IF a2 (T) is y'2 (a-4) THEN a2 (T + s) is y'2 (a + s-4) IF a2 (T) is y'2 (b-4) THEN a2 (T + s ) is y'2 (b + s-4) IF a2 (T) is y'2 (c-4) THEN a2 (T + s) is y'2 (c + s-4) ... (5) The third axis of the reconstructed state space about,

[0048]

## EQU11 ## IF a3 (T) is y'3 (a-8) THEN a3 (T + s) is y'3 (a + s-8) IF a3 (T) is y'3 (b-8) THEN a3 (T + s ) is y'3 (b + s-8) IF a3 (T) is y'3 (c-8) THEN a3 (T + s) is y'3 (c + s-8) ... (6) Also, the membership function is x ( Since a), x (b), and x (c) are data vectors near z (T), each of the reconstructed state spaces in the antecedent of the fuzzy rules (4), (5), and (6) The membership function of the axis is as shown in FIG.

Since the membership function of the consequent part cannot limit the table set to a finite range, it is expressed in a crisp expression.

For the dynamics expressed by the fuzzy rules and the membership functions, a1 (T) = y
When fuzzy inference is performed using 1 (T), a2 (T) = y2 (T), and a3 (T) = y2 (T) as input data,

[0051]

Y′1 (T + s) = a1 (T + s) y ″ 2 (T + s−4) = a2 (T + s) y ″ 3 (T + s−8) = a3 (T + s) (7) The predicted value y ″ 1 (T + s) s steps ahead of the series data y1 (T) is obtained as a1 (T + s).

As described above, the prediction value z (T + s) is obtained by using the interpolation capability and the local approximation capability of the fuzzy inference.
, And a time-series predicted value y (t + s) s steps ahead can be obtained from z (T + s).

In order to apply the prediction by the local fuzzy reconstruction method to the prediction of the blood glucose level, the current blood glucose level data vector z () is obtained from an attractor constructed by embedding the blood glucose level time-series data in a multidimensional state space. T) and s from the past neighboring data vector x (i) and the data vector x (i) in which the nearest one is selected using the Euclidean distance as a measure.
The data vector x (i + s) at the step destination is obtained, the predicted value z (T + s) at the s step destination of z (T) is obtained from these data vectors, and the time-series predicted blood glucose value y is obtained.
It is obtained as (t + s).

(Prediction experiment by local fuzzy reconstruction method)
The inventors of the present application have confirmed by experiments that it is possible to predict the time-dependent behavior of the blood glucose level from the present time using the chaos theory from the blood glucose level measurement data.

In this experiment, the blood glucose of each case was predicted one day ahead by computer software using the local fuzzy reconstruction method, and the results shown in FIG. 8 were obtained as a result of comparison with the measured values. The figure shows the prediction results for Case 1 with an average of 2
Prediction was possible with an error of 0 mg / dl or less, and sufficient clinically usable accuracy was obtained. Good prediction results were obtained for other cases as well.

From this prediction result, when the predicted value is clinically above or below a certain level, by making appropriate programming to change the amount of insulin effective at that point in a small amount, the best blood glucose control system without time lag can be obtained. There is a possibility that it can be built.

As described above, the present invention is characterized by the following blood glucose level prediction system, blood glucose level prediction method, and recording medium recording the method.

(Blood Sugar Level Prediction System) A time series measurement data storage means for storing blood sugar level measurement data as time series data in a blood glucose level time series file, and a time series data stored in the blood glucose level time series file. A dynamics estimating unit for estimating dynamics that can best represent topological properties, parameter embedding means for embedding the estimated dynamics in a multidimensional state space, and a parameter storing means for storing a delay time τ as parameters, Based on the blood glucose level stored in the blood glucose level time series file and the corresponding parameters, a blood glucose level prediction / storing to predict the near future blood glucose level by the local fuzzy reconstruction method and store the predicted blood glucose level in the predicted blood glucose level file Means, and display means for displaying data of each file.

(Prediction method of blood glucose level) The latest and past blood glucose level measurement data y (t) are prepared as time-series data, and the time-series data is embedded in a multidimensional state space by the Taketens embedding theorem, whereby the attractor can be used. And selecting a data vector z (T) on the attractor containing the latest blood glucose measurement data y (T), and selecting a plurality of data vectors on another trajectory passing through a space near the data vector z (T). Selecting a nearby data vector x (i) that is close to the Euclidean distance as a measure, selecting a data vector x (i + s) s steps ahead of the data vector x (i) to be predicted from the attractor, Predicted value s steps ahead of data vector z (T) by local fuzzy reconstruction using data vector z (T), x (i), x (i + s) (T + s) infers, the predicted value z
Predicted blood sugar value y (T + s) s steps ahead from (T + s)
Is obtained.

(Recording medium on which blood glucose level prediction method is recorded)
A procedure for collecting and recording the latest and past blood glucose level measurement data y (t) as time-series data, a procedure for embedding the time-series data in a multidimensional state space by embedding the Takentens theorem, Selecting a data vector z (T) on the attractor including the blood glucose level measurement data y (T);
A procedure of selecting a plurality of neighboring data vectors x (i) on another trajectory passing through the neighboring space of (T) as a measure using a Euclidean distance as a measure, and a step of selecting the data vector x (i) from the attractor. A procedure for selecting a data vector x (i + s) s steps ahead to be predicted; and a procedure for selecting the data vectors z (T), x (i), x (i +
s) using a local fuzzy reconstruction method to infer a predicted value z (T + s) s steps ahead of the data vector z (T), and a predicted blood glucose value y s steps ahead from the predicted value z (T + s) The procedure for obtaining (T + s) is recorded on a computer-readable recording medium as a program for causing a computer to execute the procedure.

[0061]

FIG. 1 is a system configuration diagram showing an embodiment of the present invention. The self-measured blood sugar level input unit 1 transmits a blood sugar level self-measured daily by a diabetic patient to a medical center or the like using communication means such as the Internet, PHS, personal computer communication, pager, and FAX.

The blood glucose level time series file 2 is provided as an external storage device of a computer system such as a medical center, and stores self-measured blood glucose level data transmitted from the blood glucose level input unit 1 as time series data for each patient. Keep it.

The dynamics estimating unit 3 estimates dynamics that can best represent the topological properties of the patient-specific time-series data stored in the file 2.

This dynamics estimation is performed by predicting one step ahead as a parameter for embedding in the multidimensional state space, that is, an initial value for embedding the first half of the patient-specific file, and then assuming that the first half + 1 data is known. Is predicted one step ahead. When the correlation coefficient between the predicted value and the actually measured blood glucose level when this process is repeated until there is no more data is the highest, “embedded dimension n” and “delay time τ” are obtained.

The dynamics estimation is performed when a certain amount of self-measurement value is collected and when a change in the dynamics (for example, a change in the blood glucose level of the patient is changed from poor control to fair control or good control).
This is executed when the prediction error becomes larger than a certain value due to (shift to 1).

The optimum embedding parameter file 4 stores “embedding dimension n” and “delay time τ” obtained by the dynamics estimating unit 3 as parameters for each patient.

The blood glucose level predicting section 5 stores the blood glucose level measurement data for each patient stored in the blood glucose level time-series file 2 and the optimum embedding parameters corresponding to the data.
And predicts the blood glucose level 1 to n steps ahead by the local fuzzy reconstruction method.

In this blood glucose level prediction, an attractor is constructed by embedding time-series data in a multidimensional state space by the Takentens embedding theorem, and the latest blood glucose level measurement data y
A data vector z (T) on the attractor including (T) is selected, and a plurality of neighboring data vectors x on another trajectory passing through the neighboring space of the data vector z (T).
(I) is selected using the Euclidean distance as a measure, and a data vector x (i + s) s steps ahead of the data vector x (i) to be predicted from the attractor.
And the data vectors z (T), x (i), x (i
+ S) using the local fuzzy reconstruction method to infer a predicted value z (T + s) s steps ahead of the data vector z (T), and predict a blood glucose value y (T + s) s steps ahead from this predicted value z (T + s). ).

The predicted blood sugar level file 6 is stored in the blood sugar level predicting section 5.
Save the blood glucose level data predicted in the above for each patient.

The insulin dose input section 7 is used to input the amount of insulin actually administered by the diabetic patient via the Internet, PH, or the like.
The data is transmitted to a medical center or the like using communication means such as S, personal computer communication, pager, and facsimile.

The insulin dose time-series file 8 is provided as an external storage device of a computer system such as a medical center, and stores the insulin dose data transmitted from the insulin dose input unit 7 as patient-specific time-series data. Keep it.

The display unit 9 displays each data for each patient retrieved from the blood sugar level time-series file 2, the predicted blood sugar level file 6, and the insulin dose time-series file 8, and provides the doctor with necessary support for diabetes care. Give as information. This display shows the current blood glucose level of the patient, the predicted blood glucose level in the near future, the history information of the insulin dose up to the present, and the information necessary for medical support such as the prediction confidence level and error range as necessary. To be.

With the above system configuration, instead of the conventional insulin administration treatment based on the experience and feeling of the doctor, the doctor can determine the appropriate insulin dose from the predicted blood glucose level based on the dynamics of the blood glucose level change for each individual patient. The determination can be made, and the blood sugar level control without a time lag makes it possible to keep the blood sugar level within an appropriate range in the long term while reducing the daily change.

In addition, the patient can actively use the self-measurement data, and the doctor can provide the patient with a daily instruction based on the predicted blood glucose level, thereby increasing the motivation of the patient for the self-blood glucose level measurement. Improvement can be expected.

[0075]

As described above, according to the present invention, attention is paid to the fact that the time-dependent behavior of the blood glucose level is a chaotic phenomenon, and the local fuzzy reconstruction method is used to calculate the current blood glucose level from the blood glucose level measurement time series data. Since the blood sugar level in the near future (after tomorrow) is predicted, there is an effect that support information for a doctor to determine an appropriate insulin dose can be obtained without a time lag.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a blood sugar level prediction system showing an embodiment of the present invention.

FIG. 2 is a part of time series data of a diabetic patient.

FIG. 3 is an example of an attractor projected onto a three-dimensional space.

FIG. 4 is a detailed view of an attractor shape in a three-dimensional space.

FIG. 5 is an explanatory diagram of embedding time-series data in an n-dimensional reconstruction space.

FIG. 6 shows x (T + s) from x (T) by the local reconstruction method.
FIG.

FIG. 7 is an example of an antecedent membership function in the local fuzzy reconstruction method.

FIG. 8 shows prediction results of Case 1.

[Explanation of symbols]

 Reference Signs List 1 self-monitoring blood sugar level input unit 2 blood sugar level time-series file 3 dynamics estimation unit 4 optimal embedding parameter file 5 blood sugar level prediction unit 6 predicted blood sugar level file 7 insulin dose input unit 8 insulin dose Time series file 9 ... Display

────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] July 31, 1998

[Procedure amendment 1]

[Document name to be amended] Statement

[Correction target item name] Full text

[Correction method] Change

[Correction contents]

[Document Name] Statement

Patent application title: Blood glucose predicting system and predicting method, and recording medium recording this method

[Claims]

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and system for predicting a blood sugar level of a diabetic patient by computer processing.

[0002]

2. Description of the Related Art In the treatment of a diabetic patient, the dosage of insulin is adjusted based on the blood sugar level of the patient. Blood glucose management is performed in an open circle in which the patient or the physician only measures the blood glucose level, or insulin is administered once a month or two weeks with the feeling of the physician based on the blood glucose measurement data. A feedback method of adjusting the amount is employed. Also, the insulin dose may be adjusted on a daily basis by negotiating an insulin scale.

[0003]

The treatment of a physician for insulin therapy for a diabetic patient is one of the following.

(1) Based on the blood sugar level measurement data, an insulin dose is determined about twice a month based on the experience and feeling of a doctor.

[0005] (2) An insulin dose for a blood glucose level is determined, and the test is performed once to three times a day based on this scale.

[0006] In these treatment methods, the control of the blood sugar level involves feedback with a large time lag, so that the blood sugar level change may become unstable. For example, increasing the insulin dose to lower the average blood glucose level may result in hypoglycemia. Conversely, reducing the insulin dose can cause blood sugar levels to be too high.

[0007] Under such circumstances, in order to determine an insulin dose for obtaining an appropriate blood sugar level control effect based on blood sugar level measurement data, blood glucose level control without a time lag requires daily measurement of blood glucose level. It is demanded to keep the change within a proper range in the long term while minimizing the change in the temperature.

[0008] An object of the present invention is to provide a blood glucose level that can predict a daily blood glucose level based on blood glucose level measurement data so that a doctor can obtain support information for determining an appropriate insulin dose without a time lag. It is an object of the present invention to provide a value prediction system, a blood glucose level prediction method, and a recording medium on which the method is recorded.

[0009]

DISCLOSURE OF THE INVENTION The present invention analyzes the instability of blood glucose control, elucidates the existence of chaos in the behavior of blood glucose over time based on this analysis, and uses the local fuzzy reconstruction method to analyze the current blood glucose. The blood sugar level in the near future (after tomorrow) can be predicted from the value. These matters will be described below. The near future prediction by the local fuzzy reconstruction method has already been proposed by the present inventors (Japanese Patent Laid-Open No. 7-239838).

(Time-dependent behavior of blood sugar level and chaos phenomenon)
Diabetes patients whose blood glucose data were analyzed were insulin-dependent (IDDM) 5 cases and non-dependent (NIDDM)
Five cases. The time series measurement data of these patients for a minimum of one and a half years to a maximum of ten years was measured at daily intervals. FIG.
Shows part of the time-series data of one NIDDM case and two IDDM cases showing good glycemic control.

[0011] Clinically used as an indicator of control is the% of _HbAlc , which is clinically considered to roughly represent the average of glycemic control over the past one to two months. Case 1 which becomes the data of FIG. 2 is Hb
A _lc 5 to 6% Case 2 _{HbA lc} 5 to 6% Case 3 in _{HbA lc} is 9-10%, during the course control state has been nearly constant.

Case 1 is considered to be non-insulin-dependent diabetes mellitus, and the blood glucose control function via the endogenous insulin secretion is insufficient but remaining. Cases 2 and 3 are considered to have insulin-dependent diabetes with insulin secretion capacity close to 0 and blood sugar regulation function by endogenous insulin close to 0.

[0013] The data of these three cases was converted to FFT ( Fa
When the spectrum analysis was performed using the “ st Fourier Transform” , frequency components appeared in a wide band. In addition, when the autocorrelation function was taken, it converged to almost 0 with an increase in time. The maximum Lyapunov exponent is positive, three cases showed the potential to be chaotic.

Next, the attractors projected onto the three-dimensional space for these three cases are shown in FIG. The attractor shows a columnar shape in Case 1, a triangular pyramid shape in Case 2, and a spherical shape in Case 3. Each fractal dimension is 2.27 in Case 1, 2.73 in Case 2, and Case 3
Is 3.54, which indicates that the fractal dimension increases as the shape of the attractor increases.

[0015] These three cases, triangular pyramid _{HbA lc} cylinder and case 2 in Case 1 in the evaluation of the control level by the the same extent in good (good Control), the difference in shape of the attractor IDDM and NIDD
This is probably due to the difference in M's ability to regulate blood sugar.

Both the pyramidal pyramid of Case 2 and the sphere of Case 3 are IDDM and are controlled by the same continuous insulin subcutaneous infusion therapy (CSII), and the control level is good (good control) or insufficient (poor). control).

The same examination was performed for all other DM cases, but all cases showed chaos, and the attractor shape was any of these three types or a mixed shape.

These three types of attractors also change the number of data, and when observed from various directions, they are actually based on a spiral shape as shown in FIG. 4, and this spiral is probably based on three or four small parameters and noise. It is thought to change into a triangular pyramid, cylinder, sphere, etc.

It has been inferred from many other cases that the parameters are present in the residual or control level of the ability to control blood glucose through endogenous insulin.

As described above, the time-dependent behavior of the blood glucose level of a diabetic patient appears to be an irregular phenomenon, that is, a nondeterministic phenomenon governed by chance, but its behavior can be determined deterministically. I was able to elucidate the phenomenon, a deterministic chaotic phenomenon.

(Prediction of Blood Sugar Level by Local Reconstruction Method) In the deterministic chaos phenomenon, if nonlinear deterministic regularity can be estimated, the "sensitive sensitivity to the initial value" of chaos is obtained from observation data at a certain point in time. Thus, it is possible to predict data in the near future until losing deterministic causality.

The near future prediction for such a deterministic chaotic phenomenon is based on Takens' theory that "attractors are reconstructed from one observed time series data into the state space of the original dynamical system". . The outline of this theory is as follows.

From the observed time series data y (t), vectors (y (t), y (t−τ), y (t−2)
τ), y (t− (n−1) τ)) (τ is a delay time). This vector indicates one point of the n-dimensional reconstructed state space R ⁿ .

Therefore, by changing t, a trajectory can be drawn in this n-dimensional reconstructed state space. If,
A target system deterministic dynamical system, the observed time series data assuming that obtained through the observation system corresponding to C ¹ continuous function from the state space of the dynamical system into a one-dimensional Euclidean space R For this reconstruction trajectory, if n is sufficiently large, the embedding of the original deterministic system (embeddin
g).

That is, if any attractor appears in the dynamical system, an attractor that preserves the phase structure of the attractor is reproduced in the reconstructed state space. n is usually referred to as “embedded dimension”. In order for the reconstruction operation to be “embedded”, when the dimension of the state space of the original dynamical system is m, the following equation is satisfied. Has proven to be sufficient.

[0026]

## EQU1 ## However, this is a sufficient condition, and depending on data, 2 m
Embedding may be performed even if the value is less than +1. Further, n>
If 2d (where d is the box count dimension of the attractor of the original dynamical system), it is also shown that the reconstruction operation is a one-to-one mapping.

As described above, since the change in the blood glucose level is a deterministic chaotic phenomenon, the time series data of the blood glucose level is stored in the reconstruction state space based on Taken's embedding theorem. Reconstruction is performed, and a blood glucose level in the near future can be predicted based on the attractor.

More specifically, as shown in FIG.
Reconstruction is performed by embedding the time-series data y (t) of the blood glucose level observed at equal sampling intervals into an n-dimensional state space with an embedding dimension n) delay time τ using the Taken's embedding theorem. The vector is obtained.

[0029]

X (t) = (y (t), y (t−τ),..., Y
(T− (n−1) τ) where t = 1 to L L: the number of data of the time-series data y (t) When this operation is repeatedly performed on a large number of y (t) data, an n-dimensional reconstruction state is obtained. A smooth manifold composed of a finite number of data vectors can be constructed in space. FIG. 5B shows the trajectory of the attractor when embedded in the three-dimensional reconstruction state space.

With respect to the trajectory of this attractor, the near future trajectory of the current data vector is estimated by using the data vector including the time series data of the blood glucose level measured most recently and the trajectory of the data vector in the vicinity thereof. The data vector at the step destination can be obtained. That is, from the current blood glucose level data vector and its neighboring data vector, a predicted value of the blood glucose level in the near future (after tomorrow) can be obtained from the current blood glucose level data .

That is, as shown in FIG. 6, a data vector z (T) obtained by the latest data is plotted in an n-dimensional reconstructed state space, and a data vector in the vicinity thereof is x (i). Since the data x (i) is past data, the state x (i + s) s steps ahead is known. Using this, the current data vector z
A predicted value z (T + s) s steps ahead of (T) can be predicted. Then, a predicted value y (t + s) s steps ahead of the original time-series data can be obtained from the predicted value z (T + s).

(Prediction of blood glucose level by local fuzzy reconstruction method) In the prediction by the local reconstruction method, the state x
The change to the state x (i + s) after s steps of (i) is
It is thought to be based on deterministic dynamics. The dynamics are x (i) and x (i +
Using s), it can be expressed in a linguistic expression as follows. Where i∈N (z (T)) and N (z (T)) are z
A set of indices i of neighborhood x (i) of (T).

[0033]

## EQU00003 ## IF x (T) is x (i) THEN x (T + s) is x (i + s) (1) x (T): A data vector near z (T) in the n-dimensional reconstructed state space. Set x (T + s): set x (i) representing a data vector after s steps of x (T) Since x (i) is a data vector in the vicinity of z (T), step s is “sensitive to chaos” Dependency ", before loss of deterministic causality, approximates the dynamics from state z (T) to state z (T + s) with the dynamics from state x (i) to state x (i + s) It can be assumed that

When the attractor embedded in the n-dimensional reconstruction state space is a smooth manifold, the trajectory from z (T) to z (T + s) is the vector from z (T) to x (i). Affected by distance. That is, it can be considered that the closer to the trajectory of x (i) from z (T), the larger the effect on the trajectory from z (T) to z (T + s), and the farther the trajectory is, the smaller the effect.

By the way,

[0036]

X (i) = (y (i), y (i−τ),..., Y (i− (n−1) τ)) x (i + s) = (y (i + s), y (i + s) −τ),..., Y (i + s− (n− 1) τ))... ( 2 ) Therefore, focusing on the j-axis in the n-dimensional reconstructed state space, the equation (1) becomes

[0037]

## EQU00005 ## IF aj (T) is yj (i) THEN aj (T + s) isy (i + s) (j = 1 to n) ( 3 ) where aj (T): neighborhood of z (T) J-axis component in the n-dimensional reconstruction state space of the value x (i) aj (T + s): j-axis component in the n-dimensional reconstruction state space of x (i + s) n: embedding dimension number

The trajectory from z (T) to z (T + s) is affected by the vector distance from z (T) to x (i). The trajectory of this vector is a smooth manifold. Therefore, this effect is expressed in a nonlinear form. Therefore, in order to make the influence nonlinear, if Expression ( 3 ) is expressed by a fuzzy function,

[0039]

## EQU00006 ## IF aj (T) is y'j (i) THEN aj (T + s) is y'j (i + s) (j = 1 to n) ( 4 ) Normally, the function y (i) is In the case of fuzzy conversion, the symbol "~" is used. Here, the symbol "'" is used.

Also,

[0041]

(7) z (T) = (y (T), y (T−τ),...
y (T− (n−1) τ)), the j-axis component of z (T) in the n-dimensional reconstructed state space is yj (T). Therefore, the data vector z (T + s) after s steps of the data vector z (T)
Is assumed to be z ″ (T + s), its j-axis component is
By substituting yj (T) for aj (T) in equation ( 4 ) and performing fuzzy inference, it can be obtained as aj (T + s). This method is referred to as “local fuzzy reconstruction (Loca
l FuzzyReconstruction method)
Call .

As a specific example, the embedding dimension n
= 3, the delay time τ = 4, and the number N of data vectors included in the vicinity N = 3.

Each data vector is represented by

[0044]

(8) z (T) = (y1 (T), y2 (T-4), y3 (T-8)) x (a) = (y1 (a), y2 (a-4), y3 (a) -8)) x (b) = (y1 (b), y2 (b-4), y3 (b-8)) x (c) = (y1 (c), y2 (c-4), y3 (c) −8)) z ″ (T + s) = (y1 (T + s), y2 (T + s−4), y3 (T + s−8)) x (a + s) = (y1 (a + s), y2 (a + s−4), y3 (A + s-8)) x (b + s) = (y1 (b + s), y2 (b + s-4), y3 (b + s-8)) x (c + s) = (y1 (c + s), y2 (c + s-4), y3 (C + s-8)), the fuzzy rule expressed by the equation ( 4 ) is expressed by the equation
(5), (6), and (7) .

For the first axis of the reconstructed state space,

[0046]

## EQU9 ## IFa1 (T) is y'1 (a) THEN a1 (T + s) is y'1 (a + s) IFa1 (T) is y'1 (b) THEn a1 (T + s) is y'1 (b + s) IFa1 (T) is y′1 (c) THEN a1 (T + s) is y′1 (c + s) ( 5 ) For the second axis of the reconstructed state space,

[0047]

## EQU10 ## IFa2 (T) is y'2 (a-4) THEN a2 (T + s) is y'2 (a + s-4) IFa2 (T) is y'2 (b-4) THEN a2 (T + s) is y′2 (b + s−4) IFa2 (T) is y′2 (c−4) THEN a2 (T + s) is y′2 (c + s−4) ( 6 ) For the third axis of the reconstructed state space,

[0048]

IFa3 (T) is y'3 (a-8) THEN a3 (T + s) isy'3 (a + s-8) IFa3 (T) is y'3 (b-8) THEN a3 (T + s) is y′3 (b + s−8) IFa3 (T) is y′3 (c−8) THEN a3 (T + s) is y′3 (c + s−8) ( 7 ) Also, the membership function is x (a), x (b), x
Since (c) is a data vector near z (T), the membership function of each axis of the reconstructed state space in the antecedent of the fuzzy rules (5), (6) and (7) is shown in FIG. Become like

Since the membership function of the consequent part cannot limit the table set to a finite range, it is expressed in a crisp expression.

For the dynamics expressed by the above fuzzy rules and membership functions, a1 (T) =
y1 (T), a2 (T) = y2 (T), a3 (T) = y
When fuzzy inference is performed using 3 (T) as input data,

[0051]

Y′1 (T + s) = a1 (T + s) y ″ 2 (T + s−4) = a2 (T + s) y ″ 3 (T + s−8) = a3 (T + s) ( 8 ) The predicted value y ″ 1 (T + s) s steps ahead of the series data y1 (T) is obtained as a1 (T + s).

As described above, the prediction value z (T + s) is obtained by using the interpolation capability and the local approximation capability of the fuzzy inference.
, And a time-series predicted value y (t + s) s steps ahead can be obtained from z (T + s).

In order to apply the prediction by the local fuzzy reconstruction method to the prediction of the blood glucose level, the current blood glucose level data vector z () is obtained from an attractor constructed by embedding the blood glucose level time-series data in a multidimensional state space. T), a past neighboring data vector x (i) and a data vector x (i) in which a plurality of close Euclidean distances are selected as measures.
To obtain a data vector x (i + s) s steps ahead, and a predicted value z (T + s) s steps ahead of z (T) from these data vectors, which is used as a time-series predicted blood glucose value y (t + s). Ask.

(Prediction experiment by local fuzzy reconstruction method)
The inventors of the present application have confirmed by experiments that it is possible to predict the time-dependent behavior of the blood glucose level from the present time using the chaos theory from the blood glucose level measurement data.

In this experiment, the blood glucose of each case was predicted one day ahead by computer software using the local fuzzy reconstruction method, and the results shown in FIG. 8 were obtained as a result of comparison with the measured values. The figure shows the prediction results for Case 1 with an average of 2
Prediction was possible with an error of 0 mg / dl or less, and sufficient clinically usable accuracy was obtained. Good prediction results were obtained for other cases as well.

From this prediction result, when the predicted value is clinically higher or lower than a certain level, by making appropriate programming to change the amount of insulin effective at that point in a small amount, the best blood glucose control system without time lag can be obtained. There is a possibility that it can be built.

As described above, the present invention is characterized by the following blood glucose level prediction system, blood glucose level prediction method, and recording medium recording the method.

(Blood Sugar Level Prediction System) A time series measurement data storage means for storing blood sugar level measurement data as time series data in a blood glucose level time series file, and a time series data stored in the blood glucose level time series file. A dynamics estimating unit for estimating dynamics that can best represent topological properties, parameter embedding means for embedding the estimated dynamics in a multidimensional state space, and a parameter storing means for storing a delay time τ as parameters, Based on the blood glucose level stored in the blood glucose level time series file and the corresponding parameters, a blood glucose level prediction / storing to predict the near future blood glucose level by the local fuzzy reconstruction method and store the predicted blood glucose level in the predicted blood glucose level file Means, and display means for displaying data of each file.

(Prediction method of blood sugar level) The latest and past blood sugar level measurement data y (t) are prepared as time-series data, and the time-series data is embedded in a multidimensional state space by the Taken 's embedding theorem, whereby the attractor can be used. And selecting a data vector z (T) on said attractor containing the latest blood glucose measurement data y (T),
A plurality of neighboring data vectors x (i) on another trajectory passing through the neighboring space of (T) are selected using Euclidean distance as a measure, and the data vector x (i) will be predicted from the attractor. The data vector x (i + s) at the s-step ahead is selected, and s of the data vector z (T) is determined by the local fuzzy reconstruction method using the data vectors z (T), x (i), and x (i + s). Infer the predicted value z (T + s) at the step destination, and calculate the predicted value z
Predicted blood sugar value y (T + s) s steps ahead from (T + s)
Is obtained.

(Recording medium on which blood glucose level prediction method is recorded)
Data and procedures for collecting and recording the latest and past blood glucose measurement data y (t) is a time-series data, the time-series data
A procedure for configuring an attractor by embedding in a multidimensional state space according to the Kens 's embedding theorem, a procedure for selecting a data vector z (T) on the attractor including the latest blood glucose level measurement data y (T), Vector z
A procedure of selecting a plurality of neighboring data vectors x (i) on another trajectory passing through the neighboring space of (T) as a measure using a Euclidean distance as a measure, and a step of selecting the data vector x (i) from the attractor. A procedure for selecting a data vector x (i + s) s steps ahead to be predicted; and a procedure for selecting the data vectors z (T), x (i), x (i +
s) using a local fuzzy reconstruction method to infer a predicted value z (T + s) s steps ahead of the data vector z (T), and a predicted blood glucose value y s steps ahead from the predicted value z (T + s) The procedure for obtaining (T + s) is recorded on a computer-readable recording medium as a program for causing a computer to execute the procedure.

[0061]

FIG. 1 is a system configuration diagram showing an embodiment of the present invention. The self-measured blood sugar level input unit 1 transmits a blood sugar level self-measured daily by a diabetic patient to a medical center or the like using communication means such as the Internet, PHS, personal computer communication, pager, and FAX.

The blood glucose level time series file 2 is provided as an external storage device of a computer system such as a medical center, and stores self-measured blood glucose level data transmitted from the blood glucose level input unit 1 as time series data for each patient. Keep it.

The dynamics estimating unit 3 estimates dynamics that can best represent the topological properties of the patient-specific time-series data stored in the file 2.

This dynamics estimation is performed by predicting one step ahead as a parameter for embedding in the multidimensional state space, that is, an initial value for embedding the first half of the patient-specific file, and then assuming that the first half + 1 data is known. Is predicted one step ahead. This processing is obtained as the “embedded dimension n” and the “delay time τ” when the prediction performance when the data is repeated until there is no more data is the best .

The dynamics estimation is performed when a certain amount of self-measurement value is collected and when a change in the dynamics (for example, a change in the blood glucose level of the patient is changed from poor control to fair control or good control).
This is executed when the predicted performance is reduced due to (shift to 1).

The optimum embedding parameter file 4 stores “embedding dimension n” and “delay time τ” obtained by the dynamics estimating unit 3 as parameters for each patient.

The blood glucose level predicting section 5 stores the blood glucose level measurement data for each patient stored in the blood glucose level time-series file 2 and the optimum embedding parameters corresponding to the data.
And predicts the blood glucose level 1 to n steps ahead by the local fuzzy reconstruction method.

This blood sugar level prediction is based on time series data .
The attractor is constructed by embedding in the multidimensional state space by the data embedding theorem, and the latest blood glucose measurement data y
A data vector z (T) on the attractor including (T) is selected, and a plurality of neighboring data vectors x on another trajectory passing through the neighboring space of the data vector z (T).
(I) is selected using the Euclidean distance as a measure, and a data vector x (i + s) s steps ahead of the data vector x (i) to be predicted from the attractor.
And the data vectors z (T), x (i), x (i
+ S) using the local fuzzy reconstruction method to infer a predicted value z (T + s) s steps ahead of the data vector z (T), and predict a blood glucose value y (T + s) s steps ahead from this predicted value z (T + s). ).

The predicted blood sugar level file 6 is stored in the blood sugar level predicting section 5.
Save the blood glucose level data predicted in the above for each patient.

The insulin dose input section 7 is used to input the amount of insulin actually administered by the diabetic patient via the Internet, PH, or the like.
The data is transmitted to a medical center or the like using communication means such as S, personal computer communication, pager, and facsimile.

The insulin dose time-series file 8 is provided as an external storage device of a computer system such as a medical center, and stores the insulin dose data transmitted from the insulin dose input unit 7 as patient-specific time-series data. Keep it.

The display unit 9 displays each data for each patient retrieved from the blood sugar level time-series file 2, the predicted blood sugar level file 6, and the insulin dose time-series file 8, and provides the doctor with necessary support for diabetes care. Give as information. This display shows the current blood glucose level of the patient, the predicted blood glucose level in the near future, the history information of the insulin dose up to the present, and the information necessary for medical support such as the prediction confidence level and error range as necessary. To be.

With the above system configuration, instead of the conventional insulin administration treatment based on the experience and feeling of the doctor, the doctor can determine the appropriate insulin dose from the predicted blood glucose level based on the dynamics of the blood glucose level change for each individual patient. The determination can be made, and the blood sugar level control without a time lag makes it possible to keep the blood sugar level within an appropriate range in the long term while reducing the daily change.

In addition, the patient can actively use the self-measurement data, and the doctor can provide the patient with a daily instruction based on the predicted blood glucose level, thereby increasing the motivation of the patient for the self-blood glucose level measurement. Improvement can be expected.

[0075]

As described above, according to the present invention, attention is paid to the fact that the time-dependent behavior of the blood glucose level is a chaotic phenomenon, and the local fuzzy reconstruction method is used to calculate the current blood glucose level from the blood glucose level measurement time series data. Since the blood sugar level in the near future (after tomorrow) is predicted, there is an effect that support information for a doctor to determine an appropriate insulin dose can be obtained without a time lag.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a blood sugar level prediction system showing an embodiment of the present invention.

FIG. 2 is a part of time series data of a diabetic patient.

FIG. 3 is an example of an attractor projected onto a three-dimensional space.

FIG. 4 is a detailed view of an attractor shape in a three-dimensional space.

FIG. 5 is an explanatory diagram of embedding time-series data in an n-dimensional reconstruction space.

FIG. 6 shows x (T + s) from x (T) by the local reconstruction method.
FIG.

FIG. 7 is an example of an antecedent membership function in the local fuzzy reconstruction method.

FIG. 8 shows prediction results of Case 1.

[Description of Signs] 1 ... Self-measured blood sugar level input unit 2 ... Blood sugar level time series file 3 ... Dynamics estimation unit 4 ... Optimal embedding parameter file 5 ... Blood sugar level prediction unit 6 ... Predicted blood sugar level file 7 ... Insulin dose input unit 8… Insulin dose time series file 9… Display

 ──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Seizaburo Arita 6-2-2804 Takenodai, Nishi-ku, Kobe City, Hyogo Prefecture (72) Inventor Masaya Yoneda 2-2034-16-1 Tsutakadai, Okayama City, Okayama Prefecture (72) Invention The person, Mr. Tadashi Hikami 2-1-1-17 Osaki, Shinagawa-ku, Tokyo Inside Meidensha Co., Ltd.

Claims (3)

[Claims] 1. A time series measurement data storage means for storing blood glucose level measurement data as time series data in a blood glucose level time series file; and a topological property of the time series data stored in the blood glucose level time series file. A dynamics estimating unit for estimating dynamics that can be well represented; a parameter storage unit for storing an embedding dimension n and a delay time τ as parameters for embedding the estimated dynamics in a multidimensional state space; and the blood glucose level time series file. A blood glucose level prediction and storage unit that predicts a near future blood glucose level by a local fuzzy reconstruction method and stores the predicted blood glucose level in a predicted blood glucose level file, based on the blood glucose level stored therein and the corresponding parameter. A blood glucose level prediction system, comprising: display means for displaying file data. 2. The latest and past blood glucose measurement data y
(T) is prepared as time-series data, and the time-series data is embedded in a multidimensional state space by the Takentens embedding theorem to form an attractor, and on the attractor including the latest blood glucose level measurement data y (T). Selecting a data vector z (T), selecting a plurality of neighboring data vectors x (i) on another trajectory passing through the neighboring space of the data vector z (T) and using the Euclidean distance as a measure, An s-step ahead data vector x (i +) for which the data vector x (i) is to be predicted from the attractor
s) and the data vectors z (T), x (i), x (i + s)
And the data vector z by the local fuzzy reconstruction method using
A method for predicting a blood glucose level, comprising inferring a predicted value z (T + s) s steps ahead of (T) and obtaining a predicted blood glucose level y (T + s) s steps ahead from the predicted value z (T + s). 3. The latest and past blood glucose measurement data y
A procedure for collecting and recording (t) as time-series data; a procedure for configuring the attractor by embedding the time-series data in a multidimensional state space by the Takentens embedding theorem; And selecting a data vector z (T) on the attractor including: a plurality of neighboring data vectors x (i) on another trajectory passing through the neighboring space of the data vector z (T). A procedure for selecting a measure close to the measure, and a data vector x (i +) s steps ahead of the data vector x (i) to be predicted from the attractor.
s), and the data vector z (T), x (i), x (i + s)
And the data vector z by the local fuzzy reconstruction method using
A program for causing a computer to execute a procedure of inferring a predicted value z (T + s) s steps ahead of (T) and a procedure of calculating a predicted blood glucose level y (T + s) s steps ahead from the predicted value z (T + s). Recording method for predicting a blood sugar level, wherein the method is recorded on a computer-readable recording medium.