JP5645761B2 - Medical data analysis method, medical data analysis device, and program - Google Patents

Description

  The present invention relates to an analysis method, a medical data analysis apparatus, and a program for analyzing medical data by a computer.

Regular medical examinations are generally performed.
In a health checkup, for example, physical data such as sex, age, height, weight, body mass index (BMI), smoking / drinking habits, food preferences and content, sleep status, quality and quantity of exercise, etc. Various data such as lifestyle data, blood pressure, medication status, past medical history, subjective symptoms, physiological data such as diagnosis of disease status by medical personnel, test result data obtained by blood test, urine test, etc. .
These data are provided to a doctor who makes a diagnosis, and are used for evaluation of the health condition of a health check-up examinee, maintenance of health, prevention / early detection of diseases, and the like.

Aside from health examinations conducted in the city, we try to obtain more medical-related information by conducting detailed inspections and analyzes such as component analysis and microbial analysis on the specimen obtained from the human body. Attempts are being made at laboratories and universities that conduct basic medical research.
In such basic medical laboratories, for example, metabolomic analysis of blood (all component analysis), presence of symbiotic microorganisms contained in oral mucus, feces, urine, saliva, nasal mucus, skin and vaginal fluid, etc. Based on this analysis, details of patient characteristics are specified.

  Basic medical data obtained by such basic medical examinations / analyses estimates the effects of minute components in blood and symbiotic microorganisms on the human body, and diagnoses disease conditions and prevents / predicts diseases It is used for an attempt.

By the way, various data obtained in a general medical examination (data obtained directly from a patient, hereinafter referred to as clinical data) and basic medical data are completely different data. Basic medical data is currently handled only by laboratories and research laboratories conducting basic medical research. For this reason, there are very few opportunities for doctors who handle clinical data to access basic medical data.
Therefore, attempts to extract the relationship between clinical data and basic medical data, and to obtain medically useful findings using both clinical data and basic medical data have been made until now. It was hardly done.

  The present invention has been made in view of such circumstances, and a medical data analysis method capable of obtaining diagnostic techniques and medically and scientifically useful knowledge using both clinical data and basic medical data. An object of the present invention is to provide a medical data analysis apparatus and program.

The medical data analysis method of the first invention includes at least one of physical data related to a plurality of patients, data related to a patient's lifestyle, data related to a patient's disease state, and data related to a test result obtained from a patient. containing either data, and clinical medical data is data obtained when the examination and / or treatment of the patient hands, the patient's feces, urine, saliva, nasal mucus, to at least one of skin and vaginal fluid On the other hand, basic medical data that is data on the results of basic medical examinations and / or analyzes on the specimens obtained from patients, including data on symbiotic microorganisms obtained by conducting basic medical examinations and / or analyses. Is a medical data analysis method for a medical data analysis device that performs data analysis based on data, and the medical data analysis device relates to the same patient. A first step of associating the clinical medical data with the basic medical data, and a medical data analysis device are associated with one item of the clinical medical data selected in advance in the first step. A second step of performing data analysis using a data mining method based on the basic medical data, and the second step includes a step in the basic medical data by the data mining method. The method further includes a third step of generating a classification model for extracting the similar population and for classifying the structural characteristics of the entire basic medical data.

The medical data analysis apparatus according to the second invention includes at least one of physical data related to a plurality of patients, data related to a patient's lifestyle, data related to a patient's disease state, and data related to a test result of a test object obtained from the patient. containing either data, and clinical medical data is data obtained when the examination and / or treatment of the patient hands, the patient's feces, urine, saliva, nasal mucus, to at least one of skin and vaginal fluid On the other hand, basic medical data that is data on the results of basic medical examinations and / or analyzes on the specimens obtained from patients, including data on symbiotic microorganisms obtained by conducting basic medical examinations and / or analyses. A medical data analysis device that performs data analysis based on data, and stores the clinical medical data and the basic medical data A memory unit, an input unit that receives an input operation, and a control unit, wherein the control unit associates the clinical medical data and the basic medical data related to the same patient, and inputs via the input unit Based on the basic medical data associated with one item of the clinical medical data preselected by the operation, a similar group in the basic medical data is extracted by the data mining technique, and the basic medical data is extracted. A classification model for classifying the structural characteristics of the entire target data is generated and data analysis is performed.

The program of the third invention is at least one of physical data related to a plurality of patients, data related to a patient's lifestyle, data related to a patient's disease state, and data related to a test result obtained from a patient. Including clinical data, including data, which is obtained when the patient is examined and / or treated in the field, and / or feces, urine, saliva, nasal mucus, skin and vaginal fluid . Basic medical data that is data relating to the results of basic medical examinations and / or analyzes on the specimen obtained from the patient, including data on symbiotic microorganisms obtained by conducting basic medical examinations and / or analyzes; Is a program executed by a computer included in a medical data analysis apparatus that performs data analysis, the clinical program relating to the same patient A first procedure for associating the medical data with the basic medical data, and the basic medical data associated in the first procedure with respect to one item of the clinical medical data selected in advance In the second procedure for performing data analysis using a data mining technique, and in the second procedure, a similar population in the basic medical data is extracted by the data mining technique, and the basic medical data as a whole is extracted. And causing the computer to execute a third procedure for generating a classification model for classifying the structural characteristics of the computer.

  According to the present invention, there is provided a medical data analysis method, a medical data analysis device, and a program capable of obtaining diagnostic techniques and medically and scientifically useful knowledge using both clinical data and basic medical data. Can be provided.

FIG. 1 is a diagram illustrating an example of a business model of a medical data analysis method. FIG. 2 is a diagram illustrating a configuration example of the medical data analysis apparatus 100. FIG. 3 is a flowchart showing an example of a medical data analysis method using a data mining technique. FIG. 4 is a diagram illustrating an example of data that is a target of the medical data analysis method. FIG. 5 is a diagram showing an example of a decision tree obtained as a result of analysis using HbA _1c as a characteristic by the C & RT method. FIG. 6 is a table showing an example of threshold values when categorizing patients for HbA _1c . FIG. 7 is a diagram illustrating an example of a decision tree obtained as a result of analysis using systolic blood pressure as a characteristic by the C & RT method. FIG. 8 is a table showing an example of threshold values for categorizing patients with respect to systolic blood pressure. FIG. 9 is a diagram illustrating an example of a decision tree obtained as a result of analyzing LDL-C and HDL-C as characteristics by the C & RT method. FIG. 10 is a table showing an example of threshold values when performing patient categorization for LDL-C and HDL-C.

Hereinafter, embodiments of the present invention will be described.
First, the types of data handled in this embodiment will be described.

(1) Clinical data In this embodiment, for example, various data obtained from patients in medical examinations and examinations at general medical facilities such as hospitals in the city are collectively referred to as clinical data.
Here, a general medical facility means a medical facility that does not include a research institute or university that conducts basic medical research described later.
The clinical data is data obtained by, for example, a doctor's inquiry, a questionnaire for patients, measurement of physical data, blood collection, blood tests after urine collection, urinalysis, and the like.
Clinical data includes, for example, physical data such as gender, age, height, weight, and obesity, smoking / drinking habits, dietary preferences and content, sleep status, quality and quantity of exercise, It includes blood pressure, medication status, past medical history, subjective symptoms, physiological data such as examination and diagnosis of disease status by medical personnel, test / analysis result data obtained by blood test, urinalysis, and the like.

(2) Basic medical data In this embodiment, for example, obtained by performing metabolomic analysis (examination / analysis of all components) on blood collected from a patient at a research institute or university conducting basic medical research. Data and data on the presence of symbiotic microorganisms obtained from patient feces, urine, saliva, nasal mucus, skin, vaginal fluid, etc. are referred to as basic medical data.

  In the present embodiment, a medical data analysis method for obtaining diagnostic techniques and medically and scientifically useful knowledge using both clinical data and basic medical data will be described.

Business Model FIG. 1 is a diagram illustrating an example of a business model of the medical data analysis method of the present embodiment.
FIG. 1 shows an example of a business model.
As shown in FIG. 1, the service provider 1, the patient 2, and the hospital 3 are included.

The business operator 1 collects clinical data and test objects from the patient 2 and the hospital 3, and based on the collected test objects, for example, a basic medicine generated by a research institute or university that conducts basic medical research (not shown). The target data. Then, using the collected clinical data and the acquired basic medical data, analysis according to a predetermined purpose is performed by a data mining technique, and an analysis result is obtained.
Or you may produce | generate basic medical data based on the to-be-inspected thing which the provider 1 itself collected from the patient.
Data mining is a technique for analyzing the accumulated data and searching for correlations and features between items hidden in the target characteristics to predict characteristics trends.
A specific method of analysis by the data mining method performed by the business operator 1 will be described in detail later.
Based on the analysis result, the business operator 1 can create specific advice information for health including a life improvement method and a control method for in vivo symbiotic microorganisms for each patient 2. Then, along with the analysis result, advice information can be provided to the patient 2 and the hospital 3.

The patients 2 are, for example, a plurality of people extracted from the citizens by the business operator 1, and are subject to analysis performed by the business operator 1. The number of patients 2 is not particularly limited in the present invention. What is necessary is just to set it as an appropriate number of persons according to the purpose of the analysis which the provider 1 performs. Moreover, the conditions for extracting the patient 2 may be set according to the purpose of analysis performed by the business operator 1.
The hospital 3 examines the patient 2 and obtains clinical data according to a request from the business operator 1, for example. The number of hospitals 3 is not necessarily one, and different hospitals 3 may be used for each patient.

  In the example shown in FIG. 1, the subject of analysis performed by the operator 1 is a human patient 2, but the subject of analysis in the present embodiment is not limited to a human, and may be, for example, livestock. . In this case, the hospital 3 is an animal hospital or a livestock industry.

The medical data analyzer company 1 has a medical data analyzer 100 and performs analysis processing using the clinical data and basic medical data described above.
Hereinafter, a configuration example of the medical data analysis apparatus 100 will be described.
FIG. 2 is a diagram illustrating a configuration example of the medical data analysis apparatus 100.
As illustrated in FIG. 2, the medical data analysis apparatus 100 is a computer having elements of a storage unit 101, a display unit 102, a control unit 103, and an input unit 104.

The storage unit 101 is a storage device such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or a flash memory. The storage unit 101 stores various data (including the above-described clinical data and basic medical data), a predetermined program, data necessary for executing the program, and the like.
The display unit 102 is a display device such as a liquid crystal display or a CRT (Cathode Ray Tube).
The control unit 103 is a main processing device such as a CPU (Central Processing Unit), for example, and executes a predetermined program stored in the storage unit 101 to perform a predetermined process.
The input unit 104 is a data input device such as a keyboard, a mouse, or a scanner. The input unit 104 is a data input terminal and accepts online data input.

The analysis method described below is executed by the medical data analysis apparatus 100 which is a computer having such a storage unit 101, display unit 102, control unit 103, and input unit 104. Specifically, a program for executing such an analysis method is stored in the storage unit 101, and the analysis is performed by causing the control unit 103 to execute the program in response to an operation via the input unit 104.
In addition to the configuration described above, the medical data analysis apparatus 100 may include an output unit that performs output such as printing.

Analysis Method Next, a specific method of analysis performed by the data mining method using both clinical data and basic medical data performed by the medical data analysis apparatus 100 of the business operator 1 will be described.
FIG. 3 is a flowchart illustrating an example of an analysis method using a data mining technique performed in the medical data analysis apparatus 100.

Step ST1:
Data of clinical data acquired from a plurality of patients 2 and hospital 3 is input.
Clinical data obtained from a plurality of patients 2 and hospitals 3 in advance is used. The method for collecting clinical data from the patient 2 is not limited in the present invention. The clinical data is generally generated when the patient 2 goes to the hospital 3.

Step ST2:
Basic medical data generated based on a test object acquired in advance from the patient 2 is input.
Here, the test object is blood, feces, urine, saliva, nostril mucus, skin, vaginal fluid, etc. collected from the patient 2. In the present embodiment, a case will be described in which analysis is particularly performed on intestinal resident bacteria, which are symbiotic microorganisms present in stool, using stool as a test object.
In addition, about the method of extract | collecting a test object from the patient 2, it is not limited in this invention. The inspected object may be collected by the patient 2 or may be collected in the hospital 3.
A method for generating basic medical data will be described later.
Step ST3:
Based on the clinical data input in step ST1 and the basic medical data input in step ST2, data arrangement is performed in order to suitably perform analysis by a data mining technique.
First, the acquired clinical data and basic medical data are arranged in the row (vertical) direction with the patient 2 (test subject) and the data items in the column (horizontal) direction, and the two-dimensional data (table). Generate.

FIG. 4A shows an example of clinical data as two-dimensional data.
FIG. 4B shows an example of basic medical data as two-dimensional data.
In FIG. 4A, patient names are shown in the column direction, and clinical data items for each patient 2 such as HbA _1c , systolic blood pressure, LDL-C, HDL-C, etc. are shown in the row direction. ing.
In FIG. 4 (b), the names of patients in the column direction and the item names of basic medical data for each patient such as B332, B494, B641, B657... And a name indicating a quantitative structure). Details of item names such as B332 and B494 will be described later.

In order to analyze clinical data and basic medical data by data mining techniques, it is necessary to represent these data sets in a single table. Therefore, in this step, a process for creating a single table is performed based on the clinical data illustrated in FIG. 4A and the basic medical data illustrated in FIG.
Specifically, based on the clinical data illustrated in FIG. 4A and the basic medical data illustrated in FIG. 4B, the clinical data and the basic medical data relating to the same patient are associated with each other.
FIG. 4 (c) shows an example of a single table created based on the clinical data shown in FIG. 4 (a) and the basic medical data shown in FIG. 4 (b).
The table shown in FIG. 4 (c) is a single table showing both clinical data shown in FIG. 4 (a) and basic medical data shown in FIG. 4 (b) for each patient name. . The patient name is shown in the column direction, and the clinical data item names such as HbA _1c , systolic blood pressure, LDL-C, HDL-C... Are shown in the row direction, and then B332, B494, B641, B657.・ Item names of basic medical data such as
Since clinical data and basic medical data related to the same patient are associated with each other, patient C exists in the table of FIG. 4A and does not have the display of FIG. 4B, and the table of FIG. Patient E, which does not exist in FIG. 4 (a), has been deleted from the single table shown in FIG. 4 (c).
By such processing, the correspondence between clinical data and basic medical data for each patient becomes clear.

Step ST4:
In this step, an input of “characteristic” that is an object of analysis is received via the input unit 104 of the medical data analysis apparatus 100.
Here, the “characteristic” means an item of data corresponding to the purpose of analysis determined in advance by the business operator 1, for example.
The “characteristic” is selected from one item of clinical data corresponding to a disease to be analyzed, for example.
For example, in the example of FIG. 4 described above, as shown in FIG. 4A, clinical data includes various items such as HbA _1c , systolic blood pressure, LDL-C, HDL-C. Yes.
Here, HbA _1c is an item deeply related to diabetes, systolic blood pressure is a high item related to hypertension, and LDL-C and HDL-C are related to dyslipidemia (hyperlipidemia). It is a high item.

That is, for example, the business operator 1 selects HbA _1c as the “characteristic” when the purpose is to analyze diabetes. Similarly, when the purpose is to analyze hypertension, the systolic blood pressure is “characteristic”, and when the purpose is to analyze dyslipidemia, LDL-C and What is necessary is just to select HDL-C.
In addition, when the purpose is to analyze a disease other than the above-described example, clinical data of an item highly related to the disease may be input in step 1 in advance and the item may be selected in this step.
As described above, the business operator 1 selects one item of clinical data in accordance with the purpose of analysis, and inputs the selected item via the input unit 104. In the subsequent steps, analysis is performed according to the value of the selected selected item.

  In the present embodiment, as described above, the clinical data includes data described in sentences instead of numerical values or category values, such as doctor's diagnosis contents and results of a questionnaire for the patient 2. In this embodiment, although not illustrated in FIG. 4A, description data other than numerical data can also be set as characteristics (details will be described later).

The above-mentioned “characteristic” is selected by an operation via the input unit 104 of the business operator 1.
For example, the business operator 1 may acquire clinical data of items highly related to the disease from the patient 2 and the hospital 3 according to the disease to be analyzed, or may randomly select the patient. 2 and the clinical data acquired from the hospital 3, the business operator 1 may arbitrarily determine a “characteristic” to be analyzed.

Step ST5:
Based on the two-dimensional data generated in step ST3, the “characteristic” selected in step ST4 is subjected to data analysis by a data mining technique.
Data mining is a technique for predicting the trend of characteristics by analyzing accumulated data and searching for correlations and features between items hidden in the target characteristics.

Data mining techniques include, for example, C & RT method, CHAID (Chi-square Automatic Interaction Detection) method, QUEST (Quick, Unbiased, Efficient, Statistical Tree) method, and C5.0 method decision tree (rule set) construction And a classification model, and a calculation method that provides a classification model without constructing a decision tree, such as a Bayesian method, a logistic regression method, a neural network algorithm, and an SVM (Support Vector Machine) method.
A decision tree represents a classification result algorithm corresponding to an input pattern using a tree structure.
A classification model is a data mining algorithm that focuses on and / or organizes the characteristics of data contained in a given numerical document, and the data is individually based on the status of belonging to a similar group. To classify and classify the structural characteristics of the entire document with good reproducibility, and to apply similar numerical results to the model to easily obtain similar classification results and prediction probabilities. Is an arithmetic expression group characterized by the following.
A classification model can be constructed for each calculation method of data mining. Further, even if the same numerical material and the same calculation method are used, if the target characteristics are different, different classification models can be generated and stored in an electronic file or the like.

Data mining techniques used by the data mining analysis method analysis, for example, C & RT method, CHAID method, QUEST method, C5.0 method, Bayes method, logistic regression, neural networks algorithm, SVM (Support Vector Machine), Of these methods, a method arbitrarily selected by the operator 1 in accordance with the purpose and nature of the analysis may be used.
Among the data mining calculation methods exemplified above, when the C & RT method, the CHAID method, the QUEST method, and the C5.0 method are used, a decision tree and a classification model are provided, and a Bayes method, a logistic regression method, a neural network are provided. When the algorithm, SVM is used, only the classification model is provided and the decision tree is not provided.

Hereinafter, a case where the C & RT method is used will be described as an example of an arithmetic method for providing a decision tree.
(1) C & RT method The C & RT method constructs a decision tree by dividing the target into two in order to create as homogeneous a data subset as possible for the objective variable.
Specifically, when the impurity (Gini coefficient) of data is defined and the original data (parent node data) is divided into two subsets (child node data), the impurity of the child node is The degree of improvement indicating how much improvement has been made with respect to the impurity is established as an evaluation standard for the classification work.
Then, the process of recursively searching for the separation point (the classification variable and its value) that maximizes the improvement is repeated until the stop rule is satisfied.

  The impurity g (t) and the improvement degree f (t) are calculated as follows.

  However, in Equation 1 and Equation 2,

である。

It is.

Here, π (j) is the prior probability of category j, N _j (t) is the number of cases of category j at node t (in this embodiment, the number of patients), and N _j is the category at the root node. j is the number of cases. Also, P _L is the percentage of cases that are sent to the first child node in node t, P _R is the percentage of cases that are sent to the second child node in the node t.
In the C & RT method, separation is performed so that the improvement degree f (t) is maximized. That is, fractionation is performed so that the impurity is most greatly reduced.
In this way, a decision tree is constructed and output.
In addition to using the Gini coefficient as described above, there are other methods for creating decision trees by the C & RT method, such as using Twining, least square deviation (LSD), etc. You may make it use properly by personality.

Next, a case where a logistic regression method is used will be described as an example of an arithmetic method that does not provide a decision tree.
(2) Logistic regression method The probability π _ij of the response category j in the subpopulation i is as follows.

ここで、Ｊは最後のカテゴリである。
ｘ’_ｉβ_ｊは、次のように表現される。

Here, J is the last category.
x ′ _i β _j is expressed as follows.

ただし、ｊ＝１、・・・Ｊである。

However, j = 1,... J.

  Equation 6 is the inverse function of the logit transform. When J = 2, this model is the same as the binomial logistic regression model. For this reason, the above model is considered to be an extension of the binary logistic regression model from the binary response to the multinomial nominal response.

  The log likelihood of this model is obtained by the following equation.

Here, the Newton-Raphson method is used to obtain the parameter B that maximizes the log likelihood. However, this method is the same as the Fisher scoring algorithm of this model because the expected value of the second derivative of 1 for the parameter B is the same as the expected value of the observation target.
Let ∂l / ∂B be the (J−1) p × 1 vector of the first derivative of 1 with respect to parameter B.
Further, let [∂ ² l / ∂B∂B] be the (J−1) p × (J−1) p matrix of the second derivative 1 with respect to the parameter B.

Here, [∂ ² l / ∂ B∂ B] is given as follows.

Δ_ｉは、次のような（Ｊ−１）×（Ｊ−１）行列である。

Δ _i is the following (J−1) × (J−1) matrix.

ここで、π_ｉ ^（−Ｊ）＝π_ｉ１，．．．，π_ｉＪ−１であり、Diag（π_ｉ ^（−Ｊ））はπ_ｉ ^（−Ｊ）のβ_ｊ＊対角行列である。

Here, π _i ^(−J) = π _i1,. . . , Π _iJ−1 and Diag (π _i ^(−J) ) is a β _{j *} diagonal matrix of π _i ^(−J) .

Assuming that the parameter estimation value at the iteration ν is B ^(ν) , the parameter estimation value B ^{(ν + 1) at} the iteration ν + 1 is given by the following equation.

ξ＞０はｌ（Ｂ^{（ν＋１）}）−ｌ（Ｂ^（ν））≧０となるようなステップ基準スカラーであり、Ｘ^＊は独立ベクトルの（Ｊ−１）ｐ×（Ｊ−１）行列となる。
ｌ（Ｂ^{（ν＋１）}）−ｌ（Ｂ^（ν））＜０の場合は、段階二分法を使用し、νがステップの最大数とすると、ξの値のセットは｛１／２^ν：ν＝０，．．．，ν−１｝となる。

ξ> 0 is a step reference scalar such that l (B ^{(ν + 1)} ) −l (B ^(ν) ) ≧ 0, and X ^* is a (J−1) p × (J−1) matrix of independent vectors. It becomes.
If l (B ^{(ν + 1)} ) −l (B ^(ν) ) <0, then using the step bisection method and ν being the maximum number of steps, the set of values of ξ is {1/2 ^ν : ν = 0,. . . , Ν−1}.

_Given two convergence criteria ε _k > 0 and ε _p > 0, an iteration is considered converged if any of the following criteria is met:
(1) | l (B ^{(ν + 1)} ) −l (B ^(ν) ) | <ε _k
(2) max _i | B _i ^{(ν + 1)} −B _i ^(ν) | <ε _p
(3) The maximum value of the above element in ∂l / ∂B ^{(ν + 1)} is less than min (ε _k , ε _p ) A classification model is generated by such a calculation method.

-Method for generating basic medical data Next, a method for generating basic medical data will be described.
In the present embodiment, a case will be described in which analysis values of intestinal resident bacteria contained in feces collected from a patient are used as basic medical data.

First, a method for obtaining an analysis value of intestinal resident bacteria will be described in detail.
For example, a terminal-RFLP (Terminal Restriction Fragment Length Polymorphism Analysis: T-RFLP) method is used as a method for obtaining an analysis value of intestinal resident bacteria.
The T-RFLP method extracts a 16S rDNA gene derived from a microorganism from an object to be examined, amplifies the template DNA by PCR (Polymerase Chain Reaction), and cleaves a restriction enzyme (specific base sequence site of DNA). In this method, fragment inspection / analysis is performed after digestion with an enzyme having a property of oxidization, and the difference in the cleavage site of the restriction enzyme is used to measure the difference in peak position and intensity.
In the T-RFLP method, each DNA fragment is classified and measured as an OTU (Operational Taxonomic Unit) derived from a symbiotic microorganism.

By such an analysis, for example, basic medical data as shown in FIG. 4B is obtained.
Here, item names such as “B332” and “B494” indicate the restriction enzyme used and the peak position obtained using the restriction enzyme.
That is, in the case of “B332”, for example, it means an OTU whose peak position is “332” obtained using a restriction enzyme BslI. That is, the item name is formed by combining the initial letter of the restriction enzyme and the peak position obtained using the restriction enzyme.
In the present embodiment, an example in which BslI is used as a restriction enzyme has been shown, but the present invention is not limited to this. Other restriction enzymes such as MspI, AluI, HaeIII and the like may be used.

There are a large number of microbial species belonging to each OTU, and most of the human symbiotic microorganisms are unclear about their species names and physiological functions. It can only be inferred by using restriction enzymes. However, if an innovative and reproducible DNA testing method is developed in the future, it will be possible to obtain materials including the names of individual microorganisms and species that are more accurate than the current OTU as basic medical data. Seem.
Due to such circumstances, in this embodiment, OTU indicating the restriction enzyme used and the peak position obtained using the restriction enzyme is adopted as the item name.

Specific example of data analysis by data mining method Hereinafter, a specific example of data analysis by the data mining method will be shown.
As an example, a data analysis result in the case where each item shown in FIG.
[1] When HbA _1c is used as a characteristic A case where HbA _1c (hemoglobin A _1c ) is selected as a characteristic will be described.
HbA _1c is hemoglobin (hemoglobin) associated with glucose and is an item having a great relevance to diabetes.
FIG. 5 is an example of a decision tree obtained as a result of analysis using HbA _1c as a characteristic by the C & RT method. Figure 5 is based on clinical data and basic medical data obtained from questionnaires, interviews, and various tests including blood, targeting 121 men and women who were concerned about lifestyle-related diseases in regular health examinations. is there.

In FIG. 5, the decision tree extends from the left side to the right side.
The leftmost node, node 0, is referred to as a root node, and all patients subjected to data analysis are included in this node.
All patients included in node 0 are classified in advance into four categories A to D.
This categorization is performed in advance by the operator 1 through, for example, an operation via the input unit 104 of the operator 1 according to the purpose and character of the analysis.
In the example shown in FIG. 5, categorization is performed according to the number of values of HbA _1c . The categorization may be performed according to, for example, whether the value of HbA _1c is equal to or less than a predetermined threshold value or greater than the predetermined threshold value.

FIG. 6 is a table showing an example of threshold values used for categorizing HbA _1c in FIG.
As shown in FIG. 6, category A is the category to which the patient with the lowest HbA _1c concentration belongs, and category D is the category to which the patient with the highest HbA _1c concentration belongs. A patient group belonging to category D having the highest HbA _1c concentration is a severely diabetic patient.

In the examples shown in FIGS. 5 and 6, the category to which patients with normal HbA _1c concentration belong (category A), the category to which patients in need of attention (category B and C), and patients with severe diabetes are shown. The purpose is to obtain knowledge about the relationship between the symbiotic microorganisms and the HbA _1c concentration, and thus the relationship between the symbiotic microorganisms and diabetes by performing category classification to the category D to which it belongs.

In FIG. 5, each node indicates the number of patients corresponding to the value of HbA _{1c in} each category A to D in the “n” column, and the ratio of the number of patients in each category to the total number of patients. This is shown in the “%” column. Therefore, in node 0, the total in the “n” column is 121 people in the total number of patients, and the total in the “%” column is “100.000”%.
In the example shown in FIG. 5, the number of patients corresponding to category A out of the total number of patients 121 is 42, and the ratio is 34.711%. The number of patients corresponding to category B is 46, and the ratio is 38.017%. The number of patients corresponding to category C is 14, and the ratio is 11.570%. The number of patients corresponding to category D is 19, and the ratio is 15.702%.

In FIG. 5, node 0 is classified into node 1 and node 2.
This classification condition is described between node 0 and nodes 1 and 2 in FIG. Node "B494_tr" described in the right side of 0 indicates the OTU contribute to fractionation, described on the left side of node 1 '<= 21.923 "and written to the left of the node 2"> 21.9 23 "Indicates the boundary value at the time of classification.
That is, in FIG. 5, when the node 0 is sorted into the nodes 1 and 2, if the value of the OTU “B494_tr” is equal to or less than the boundary value 21.923, the value is set to the node 1; Are separated.

Here, the OTU that contributes to the classification is selected to have the highest degree of improvement by calculating the degree of improvement by the C & RT method as described above. In the case of classification from the node 0 to the nodes 1 and 2 in FIG. 5, “B494_tr” is obtained.
“B494_tr” means an OTU having a peak position of 494 due to the restriction enzyme BslI starting with B, and is one of the items of basic medical data described above. Note that “_tr” of “B494_tr” indicates that a value normalized for each item is used in the above-described calculation by the C & RT method.

  In the example shown in FIG. 5, node 1 includes 41 people in category A and 14 people in category B, and 0 patients in categories C and D. On the other hand, the node 2 includes one category A, 32 categories B, 14 categories C, and 19 categories D.

  In the example shown in FIG. 5, the node 3 includes only category A patients, and the node 15 includes almost only category A patients. In addition, only the patient of category B gathers at node 9, and many patients of category B gather at node 7, node 16, node 18, and node 26. Nodes 11 and 25 have only category C patients. All patients of category D are gathered only at node 12.

From the above, at node 11 and node 12, all the members of category D having the highest HbA _1c concentration (which is a serious patient with diabetes) and the patients of group C having the next highest HbA _1c concentration gathered. I understand that. According to FIG. 5, since the OTUs to be sorted into nodes 11 and 12 are “B494” and “B332”, it is easily estimated from the above analysis that these OTUs have a great relationship with the occurrence of diabetes. can do.

[2] When systolic blood pressure is used as a characteristic A case where systolic blood pressure is selected as a characteristic will be described.
Systolic blood pressure is a characteristic that has great relevance to hypertension.
FIG. 7 is an example of a decision tree obtained as a result of analysis using systolic blood pressure as a characteristic by the C & RT method. Figure 7 is based on clinical data and basic medical data obtained from questionnaires, interviews, and various tests including blood, targeting 121 men and women who were concerned about lifestyle-related diseases in regular health examinations. is there.

  In the example illustrated in FIG. 7, as in the example illustrated in FIG. 5, the classification is determined from the node 0 so that the improvement degree by the C & RT method is increased, the decision tree is extended toward the right side, and each node is single. The classification process is performed until the category or a state close thereto.

In the example illustrated in FIG. 7, all patients are classified in advance into four categories J to M by the business operator 1 in advance. Each patient is classified into one of categories J to M according to the value of systolic blood pressure. FIG. 8 shows an example of threshold values used for categorization according to systolic blood pressure values.
In the categories J to M classified by the threshold shown in FIG. 8, the category M is the category to which the patient with the highest systolic blood pressure belongs.

According to FIG. 7, all of the patients belonging to category M having the highest systolic blood pressure are included in node 10, and many of the patients belonging to category L, which is the next category having the highest systolic blood pressure, belong to node 9. included.
Therefore, it can be easily estimated from the decision tree shown in FIG. 7 that OTUs such as “B469”, “B124”, and “B366” have a great relevance to the occurrence of hypertension.
[3] When LDL-C and HDL-C are used as characteristics A case where LDL-C and HDL-C (cholesterol value) are selected as characteristics will be described.
LDL-C and HDL-C are properties that have great relevance to dyslipidemia (hyperlipidemia).
FIG. 9 is an example of a decision tree obtained as a result of analyzing the characteristics of LDL-C and HDL-C by the C & RT method. Figure 9 is based on clinical data and basic medical data obtained from questionnaires, interviews, and various tests including blood targeting 121 men and women who were concerned about lifestyle-related diseases in regular health examinations. is there.

  In the example shown in FIG. 9, as in the examples shown in FIGS. 5 and 7, the classification is determined from the node 0 so that the improvement degree by the C & RT method is increased, the decision tree is extended toward the right side, and each node is The separation process is performed until a single category or a state close thereto.

In the example illustrated in FIG. 9, all patients are classified in advance into four categories P to S by the business operator 1 in advance. Each patient is classified into one of categories P to S according to the values of LDL-C and HDL-C. FIG. 10 shows an example of threshold values used for categorization according to the values of LDL-C and HDL-C.
In categories P to S categorized by threshold values as shown in FIG. 10, category S is a category to which the most severe dyslipidemic patient belongs, and category P is a category to which a normal patient belongs. Categories Q and R are categories to which patients who are in need of attention regarding dyslipidemia belong.

According to FIG. 9, node 1 includes all patients in category S to which patients with serious dyslipidemia belong, and node 4 includes those in categories Q and R to which patients in need of attention belong. All patients are included.
Node 5 includes many patients of category Q, and node 12 includes all members of category R. The OTU related to the classification from the node 4 to the node 5 and the node 12 is “B990”. That is, considering the separation from the node 12 to the node 13 (all are included in the category P to which normal patients belong) and the node 14 (including the patients in the categories Q and R to which the patients in need of attention are included) It can be inferred that the subtle concentration difference of “B990” greatly affects the onset of dyslipidemia.

As described above, according to the medical data analysis method of the present embodiment, clinical data obtained in general medical examinations and examinations, feces, urine, saliva, nasal mucus, skin, vaginal fluid, etc. Because data analysis by data mining techniques is performed based on basic medical data generated by testing and / or analysis in basic medical laboratories based on test objects such as blood, clinical data And a basic medical data can be organically associated with each other, and a decision tree can be constructed in which these relationships can be easily grasped.
In addition, among the multiple items included in the clinical data, an item corresponding to the purpose of data analysis is selected, and data analysis is performed based on the selected item (characteristic) and the numerical data of each item of basic medical data. be able to. Therefore, it is possible to construct a decision tree according to the purpose simply by changing the target item from the same data (clinical data and basic medical data).

  Furthermore, as a data mining technique used for data analysis, a target characteristic can be obtained by using a calculation method (for example, C & RT method, CHAID method, QUEST method, C5.0 method, etc.) capable of constructing a decision tree. It is possible to visually and easily grasp which node contains a patient corresponding to, and to easily understand the items of basic medical data corresponding to diseases etc. according to the target characteristics Can do.

In addition, by applying pre-built decision trees and classification models to new patient data other than the clinical and basic medical data patients used to build the decision trees and classification models, The situation can be predicted and sorted.
Specifically, for example, when only basic medical data of a new patient is acquired, if the decision tree (see FIG. 5) constructed for the characteristic “HbA _1c ” in the above-described embodiment is referred to, the constructed decision is made. In the tree, OTUs corresponding to classification and classification boundary values are made clear, so by applying the already generated classification model to new patient data, each new patient is assigned to which node. It is possible to easily infer whether it is included, thereby making it possible to predict and classify a new patient's disease state with respect to a target characteristic.

The present invention is not limited to the embodiment described above.
That is, those skilled in the art may make various modifications, combinations, subcombinations, and alternatives regarding the components of the above-described embodiments within the technical scope of the present invention or an equivalent scope thereof.

  In the above-described embodiment, in the data analysis for constructing the decision trees shown in FIGS. 5, 7, and 9, each of the categories is divided into four categories according to the selected characteristic values. This is not a limitation. By dividing into more categories, for example, 6 to 8 categories, it is possible to find a category that includes a more serious patient, and to more appropriately identify an OTU related to characteristics.

  In the above-described embodiment, the example in which the business operator 1 performs categorization according to the target characteristic value in advance in the data analysis for constructing the decision trees shown in FIGS. The invention is not limited to this. The clinical data is not necessarily numerical values or category data, and may be described by a list of sentences and phrases such as a response result of a questionnaire by the patient 2 and a diagnosis result of a doctor. For example, the business operator 1 may select items in such a list of sentences and phrases as characteristics and categorize them according to the contents of the description. For example, select the item “Subjective Symptom” in which “None”, “Pain in the abdomen”, “Pain in the leg”, etc. are described as data, and select “ You may divide and may categorize by "site with a subjective symptom". In this way, the category division can be freely set by the business operator 1 according to the purpose of analysis.

  In the above-described embodiment, the C & RT method and the logistic regression method have been described as examples of the data mining method. However, the data mining calculation method used in the present invention is not limited thereto. For example, CHAID method, QUEST method, C5.0 method, etc., which provide decision tree construction and classification model, Bayes method, logistic regression method, neural network algorithm, SVM, etc. Of the calculation methods that provide the classification model without construction, for example, the operator 1 may arbitrarily select and use a calculation method that matches the purpose of the analysis. In the case of a calculation method that does not construct a decision tree, it is impossible to visually grasp the classification and relevance of nodes. Data analysis and separation are possible.

In addition, when an arithmetic method that generates a classification model without constructing a decision tree is adopted, it is possible to further calculate the mathematical importance described below and estimate items closely related to the target characteristics. Is possible.
The mathematical importance is calculated by, for example, nearest neighbor analysis.
The mathematical importance is used as a measure of dissimilarity for identifying the data pattern of each component based on the similarity to the target characteristic and classifying the distance.
Specifically, if the importance of the component is FI _(p) and the error rate or the error sum of squares when the patterns are compared is e, the generated numerical model is represented by X ₍₁₎ , X ₍₂₎ ,. . . , X _(m) (1 ≦ m ≦ P ⁰ ₎ , the importance of the model OTU X _(p) is calculated by the following equation.

First, the component X _(p) is removed from the model, and the remaining components X ₍₁₎ , X ₍₂₎ ,. . . , X _(p−1) , X _{(p + 1),.} . . , X _(m) , the error rate or error sum of squares e _(p) is calculated and compared.
Then, FI _(p) = e _(p) + 1 / m is calculated, and finally the importance of the component X _(p) is obtained.
Here, X is a two-dimensional P × N matrix having an element X _pn , p = 1,..., P is a component, and n = 1,. Further, P is the number of dimensions of the target characteristic, which is the number of components in the continuous type, and the number of categories of the entire component in the category type characteristics.

  Moreover, although embodiment mentioned above demonstrated the case where the stool of patient 2 was used as a to-be-tested object as basic medical data, and the data about intestinal resident bacteria were used, this invention is not limited to this. For example, urine, saliva, nasal mucus, skin and vaginal fluid may be used as test objects, and symbiotic microorganisms contained therein may be analyzed. Further, for example, the result of metabolome analysis (all component analysis: comprehensive analysis of metabolites) using blood as the test object may be used as basic medical data.

  DESCRIPTION OF SYMBOLS 1 ... Business operator, 2 ... Patient, 3 ... Hospital, 100 ... Medical data analyzer, 101 ... Memory | storage part, 102 ... Display part, 103 ... Control part, 104 ... Input section

Claims (7)

Practical data on patients, including at least one of physical data on multiple patients, data on patient lifestyle, data on patient's disease status, and data on test results obtained from patients Clinical medical data, which is data obtained during examination and / or treatment, and / or basic medical examination and / or examination of stool, urine, saliva, nasal mucus, skin and vaginal fluid. contains data about probiotic microorganisms obtained by performing analysis, and basic medical data is data about the results of the basic medical examination and / or analysis of the object to be inspected obtained from a patient, a based on, performing data analysis A medical data analysis method for a medical data analyzer,
The medical data analysis device first associates the clinical medical data and the basic medical data for the same patient;
The medical data analysis device for preselected a component of the clinical medical data, based on the basic medical data associated in the first step, data analysis using data mining techniques A second step of performing
Have
The second step is a third step of extracting a similar group in the basic medical data by the data mining technique and generating a classification model for classifying structural characteristics of the entire basic medical data. A medical data analysis method further comprising: In the second step, at least one of the C & RT method, the CHAID method, the QUEST method, and the C5.0 method is used as the data mining method, and the medical data analysis device uses the method to determine the decision tree and the classification The medical data analysis method according to claim 1, wherein a model is constructed. In the second step, at least one of a Bayesian method, a logistic regression method, a neural network algorithm, and an SVM (Support Vector Machine) is used as the data mining method. The medical data analysis method according to claim 1, wherein a classification model is constructed. In the second step, the medical data analysis device sets one item of the pre-selected clinical medical data as an objective variable, and sets each item of the associated basic medical data as an explanatory variable. The medical data analysis method according to any one of claims 1 to 3, wherein the classification model is constructed by a data mining technique. The physical data regarding the plurality of patients includes at least one of sex, age, height, weight, and obesity,
The data on the lifestyle of the patient includes at least one of smoking habit, supper habit, meal preference, meal content, sleep situation, exercise quality, exercise quantity,
The data regarding the disease state of the patient includes at least one of blood pressure, medication status, past medical history, subjective symptoms, and diagnostic results by medical personnel,
The medical data analysis according to any one of claims 1 to 4, wherein the data related to the test result of the test object obtained from the patient includes at least one of a blood test result and a urine test result. Method. Practical data on patients, including at least one of physical data on multiple patients, data on patient lifestyle, data on patient's disease status, and data on test results obtained from patients Clinical medical data, which is data obtained during examination and / or treatment, and / or basic medical examination and / or examination of stool, urine, saliva, nasal mucus, skin and vaginal fluid. contains data about probiotic microorganisms obtained by performing analysis, and basic medical data is data about the results of the basic medical examination and / or analysis of the object to be inspected obtained from a patient, a based on, performing data analysis A medical data analysis device,
A storage unit for storing the clinical medical data and the basic medical data;
An input unit that accepts input operations;
A control unit;
Have
The control unit associates the clinical medical data and the basic medical data related to the same patient, and is associated with one item of the clinical medical data selected in advance by an input operation via the input unit. Based on the basic medical data, a data mining technique is used to extract a similar group in the basic medical data and generate a classification model to separate the structural characteristics of the entire basic medical data. Medical data analysis device. Practical data on patients, including at least one of physical data on multiple patients, data on patient lifestyle, data on patient's disease status, and data on test results obtained from patients Clinical medical data, which is data obtained during examination and / or treatment, and / or basic medical examination and / or examination of stool, urine, saliva, nasal mucus, skin and vaginal fluid. contains data about probiotic microorganisms obtained by performing analysis, and basic medical data is data about the results of the basic medical examination and / or analysis of the object to be inspected obtained from a patient, a based on, performing data analysis A program executed by a computer included in the medical data analysis device,
A first procedure for associating the clinical medical data and the basic medical data for the same patient;
A second procedure for performing data analysis using a data mining method based on the basic medical data associated in the first procedure with respect to one item of the clinical medical data selected in advance; ,
In the second procedure, a third procedure for extracting a similar group in the basic medical data by the data mining technique and generating a classification model for classifying the structural characteristics of the entire basic medical data. When,
A program for causing the computer to execute.

Images