KR20200084807A - Apparatus and method for predicting chronic kidney disease - Google Patents

Abstract

The present invention relates to a device for predicting chronic kidney disease. The device for predicting chronic kidney disease comprises: a genome marker selection unit that selects genome markers related to a plurality of diseases; an integrated genome index calculation unit that calculates an integrated genome index using the genome markers related to the plurality of diseases selected by the genome marker selection unit; an integrated risk factor building unit that derives factors related to the plurality of diseases and builds an integrated risk factor model; a disease occurrence prediction model generation unit that constructs a disease occurrence prediction model by inputting the integrated genome index and the integrated risk factor model; and a disease prediction unit that predicts the occurrence of chronic kidney disease of a subject by inputting new disease prediction data into the disease occurrence prediction model. Accordingly, high-risk groups in clinical trials can be identified and managed.

Description

Apparatus and method for predicting chronic kidney disease occurrence{APPARATUS AND METHOD FOR PREDICTING CHRONIC KIDNEY DISEASE}

The present invention relates to an apparatus and method for predicting the development of chronic kidney disease.

Among the diseases in which the health risk prediction tool is implemented and the intervention for the high-risk group is actively performed, breast cancer is representative, and according to the risk assessment model implemented in the West, it can be divided into three types.

One is a model that predicts the likelihood of absolute occurrence in the general population as a combination of baseline risk and risk, and the other is a method of predicting the likelihood of occurrence according to the relative risk of risk factors. The third is a model used specifically for predicting hereditary breast cancer occurrence, and may predict the likelihood of breast cancer development based on the likelihood of having a BRCA gene mutation or the possession of a BRCA gene mutation based on family history.

Currently, in Korea, the Korean Family Medicine Association has developed a Korean health risk prediction tool, and by applying this, a personalized health management program service is provided on the website of the National Health Insurance Corporation, <Health iN>.

However, the existing disease incidence risk model has limitations in that it is limited in application to the general population because it predicts a new disease incidence risk based on healthy subjects without disease. In addition, it presents a risk probability value for one result, mainly due to the occurrence or death of a disease, and has a limitation of predicting only one disease compared to entering many variables. And above all, these existing technologies have limitations in that they are inadequate to manage the health status of individuals who change periodically by identifying and correcting health risk factors.

Accordingly, demographic indicators such as age and gender, sociological factors, and disease family history factors, based on the individual's genetic characteristics, taking into account the timing of exposure and the temporal continuity of the condition that can be changed afterwards and the natural history of the disease. Factors related to environmental behavior that can be changed and exposed afterwards, and obesity-related metrics and blood markers that recognize blood abnormalities and blood disease markers that recognize blood abnormalities according to these behavior factors. Health condition management and personalized preventive measures through the provision of an integrated disease outbreak prediction model that identifies the risk of hypertension, diabetes, and their accompanying diseases and chronic kidney disease in the future, taking into account the disease's biological continuity Is needed.

The background technology of the present application is disclosed in Korean Patent Registration No. 10-0931300.

The present application is to solve the problems of the prior art described above, starting with the genetic characteristics that an individual has from birth, considering changes in environmental factors and changes in blood markers according to them, according to time sequence A device for predicting the occurrence of chronic kidney disease, taking into account risk factors and reflecting them in the form of those who change from time to time, and constructing an integrated model to predict the risks of high blood pressure, diabetes and their accompanying diseases and chronic kidney disease. It aims to provide a method.

However, the technical problems to be achieved by the embodiments of the present application are not limited to the technical problems as described above, and other technical problems may exist.

As a technical means for achieving the above technical problem, the apparatus for predicting the development of chronic kidney disease according to an embodiment of the present application is selected from a genomic marker selection unit and the genomic marker selection unit to select genomic markers associated with a plurality of diseases. An integrated genomic index calculation unit that calculates an integrated genomic index using genomic markers associated with a plurality of diseases, an integrated risk factor construction unit that derives factors related to the plurality of diseases and builds an integrated risk factor model, the integrated genomic index, and Using the integrated risk factor model as an input, a disease outbreak prediction model generator for constructing a disease outbreak prediction model and a disease predictor for predicting the occurrence of chronic kidney disease in a subject by inputting new disease prediction data into the disease outbreak prediction model It can contain.

In addition, the genome marker selection unit selects a first genome marker associated with the first disease, verifies the selected first genome marker in each of the urban cohort and rural cohort, and determines the first genome marker, and Related second genome markers can be selected, and the selected second genome markers can be verified in each of the urban cohort and rural cohort to determine the second genome marker.

In addition, the genome marker selection unit selects a single-base polymorphism (SNP) marker associated with a plurality of diseases based on a regression analysis algorithm, the single-base polymorphism (SNP) marker, the first dielectric marker, and the second dielectric marker Core gene information can be derived by considering at least one of the above.

In addition, the integrated risk factor building unit includes at least one of basic demographic factors, sociological factors, disease and family history factors, environmental behavior-related factors, obesity-related indicator factors, blood abnormality indicator factors, and underlying disease morbidity factors. Factors related to the plurality of diseases may be derived by considering any one, and an integrated risk factor model may be constructed.

In addition, the apparatus for predicting the occurrence of chronic kidney disease may further include an explanatory variable derivation unit that derives data of a subject having at least one of a plurality of diseases as an explanatory variable.

In addition, the disease incidence model generation unit, by inputting a plurality of state variables, including the state variables and health status variables of patients with chronic kidney disease, the key genetic information and disease risk of chronic kidney disease as input, the plurality of A disease outbreak prediction model for learning the degree of relationship between at least one of state variables and key genetic information and the disease risk of the chronic kidney disease may be generated.

In addition, the disease occurrence prediction model, when the first state variable and the previous time hidden layer of the plurality of state variables as the input layer, and the second state variable or the current time state variable of the plurality of state variables as the hidden layer, the When performing the first learning to learn the degree of the relationship between the input layer and the hidden layer, and using the hidden layer and the genetic information as the input layer and the disease risk as the output layer, learning the degree of the relationship between the hidden layer and the output layer By performing the second learning, the degree of relationship between at least one of the plurality of state variables and genetic information and the disease risk of the chronic kidney disease is learned, and the first learning is based on [Equation 1]. , Learning the degree of the relationship between the input layer and the hidden layer,

[Equation 1]

는 t 시점에서의 은닉층이고, 상기

는 입력층과 은닉층 사이의 제1유형의 관계의 정도를 나타내는 제1가중치이고, 상기

은 이전 시점 은닉층이고,

는 입력층과 은닉층 사이의 제2유형의 관계의 정도를 나타내는 제2가중치이고,

는 t시점에서의 제1상태 변수일 수 있다.At this time, above

Is a hidden layer at time t, and

Is a first weight value indicating the degree of relationship of the first type between the input layer and the hidden layer, wherein

Is the hidden layer from the previous point,

Is a second weight value indicating the degree of relationship of the second type between the input layer and the hidden layer,

May be the first state variable at time t.

In addition, the second learning is to learn the degree of the relationship between the hidden layer and the output layer based on [Equation 1] and [Equation 2],

[Equation 2]

는 은닉층과 출력층 사이의 관계의 정도를 나타내는 제3가중치이고,

는 t 시점에서의 은닉층이고, 상기

는 입력층 중 유전자 정보와 출력층 사이의 관계의 정도를 나타내는 제4가중치이고, z는 입력층 중 유전자 정보일 수 있다.At this time, y is an output layer, and

Is a third weight value indicating the degree of relationship between the hidden layer and the output layer,

Is a hidden layer at time t, and

Is a fourth weight value indicating the degree of relationship between the gene information and the output layer in the input layer, and z may be the gene information in the input layer.

In addition, the disease occurrence prediction model generation unit, based on [Equation 3], a machine learning model for learning the degree of relationship between at least one of the plurality of state variables and genetic information and the disease risk of the chronic kidney disease. To update the weight to the error that occurs during generation,

[Equation 3]

는 오차에 따른 과적합(overfitting)을 방지하기 위한 L2 정규식일 수 있다. The E is the detection value of the error of the disease occurrence prediction model generation unit, the t is whether the chronic kidney disease occurs, the y is the disease risk predicted through the machine learning model,

May be an L2 regular expression to prevent overfitting due to errors.

In addition, the disease prediction unit may provide disease prevention management information associated with the prediction result of chronic kidney disease in the subject.

According to one embodiment of the present application, the method for predicting the development of chronic kidney disease comprises: selecting a genomic marker associated with a plurality of diseases, calculating an integrated genomic index using genomic markers associated with a plurality of selected diseases, and the plurality Constructing an integrated risk factor model by deriving factors related to diseases of the disease, constructing a disease outbreak prediction model by inputting the integrated genomic indicator and the integrated risk factor model, and predicting a new disease in the disease outbreak prediction model It may include a step of predicting the occurrence of chronic kidney disease in the subject by using the data as input.

The above-described problem solving means are merely exemplary and should not be construed as limiting the present application. In addition to the exemplary embodiments described above, additional embodiments may exist in the drawings and detailed description of the invention.

According to the above-described problem solving means of the present application, starting from the genetic characteristics of the person, based on the change in environmental factors and the change in clinical values, hypertension, diabetes, and chronic forms of their multiple diseases Identify subjects with a high probability of developing kidney disease, and provide preventive measures through early diagnosis on subjects with high probability, and treat them in advance even before they become worse (cardiovascular disease and death) even in patients with illness. By performing the secondary prevention, it is possible to reduce the risk of complications such as cardiovascular disease, chronic kidney disease, metabolic disorders, and neurological diseases, thereby improving the quality of life.

According to the above-mentioned problem solving means of the present application, by presenting information on a disease risk pattern according to a risk factor of hypertension, diabetes, and chronic kidney disease, these risk factors are controlled through lifestyle correction, and accordingly By identifying the form of reduction, an individual's active self can lead to health care.

According to the above-described problem solving means of the present application, by using a model built on a machine learning basis, it is possible to identify and manage a real community general population group or a high-risk group in a clinical trial through high predictive power.

According to the above-described problem solving means of the present application, by predicting and preventing metabolic diseases such as hypertension, diabetes, and chronic kidney disease, it is possible to help lower the risk of multiple diseases of the disease to increase the risk of death.

According to the above-described problem solving means of the present application, the algorithm may be used in a product using a web (WEB) and an app (APP) aimed at a disease risk prediction model or a personal health care service.

However, the effects obtainable herein are not limited to the effects described above, and other effects may exist.

1 is a schematic configuration diagram of a device for predicting the development of chronic kidney disease according to an embodiment of the present application.
Figure 2 is a schematic block diagram of a chronic kidney disease device according to an embodiment of the present application.
FIG. 3 is a diagram showing a list of information that can be used as reference dielectric information for extending genetic variation information of a chronic kidney disease device according to an embodiment of the present application.
4 is a view for explaining a significant genomic marker associated with hypertension according to an embodiment of the present application.
5 is a view for explaining a significant genomic marker associated with diabetes according to an embodiment of the present application.
6 is a view for explaining the results of the hypertension integrated genomic score construction according to an embodiment of the present application.
7 is a view for explaining the results of the diabetes integrated genomic score construction according to an embodiment of the present application.
FIG. 8 is a diagram showing variables sequentially included in a predictive model in consideration of the timing of factor exposure according to an embodiment of the present application and the temporal continuity of a state that can be changed thereafter and the natural history of the disease.
9 is a diagram for explaining a deep learning model structure that integrates time series data and genetic data according to an embodiment of the present application.
10 is a schematic diagram of a disease outbreak prediction model according to an embodiment of the present application.
11 is a schematic diagram of a random forest prediction model according to an embodiment of the present application.
12 is a diagram schematically showing a process of training models in various cases according to an embodiment of the present application by an ensemble technique.
13 is a view showing a comparison of the predicted risk of developing hypertension according to a statistical and machine learning model according to an embodiment of the present application.
14 is a diagram showing a comparison of predicted risk of developing diabetes according to a statistical and machine learning model according to an embodiment of the present application.
15 is a view showing a comparison of predicted risk of developing chronic kidney disease according to a statistical and machine learning model according to an embodiment of the present application.
16 is a flowchart illustrating a method for predicting the development of chronic kidney disease according to an embodiment of the present application.

Hereinafter, embodiments of the present application will be described in detail with reference to the accompanying drawings so that those skilled in the art to which the present application pertains may easily practice. However, the present application may be implemented in various different forms and is not limited to the embodiments described herein. In addition, in order to clearly describe the present application in the drawings, parts irrelevant to the description are omitted, and like reference numerals are assigned to similar parts throughout the specification.

Throughout this specification, when a part is "connected" to another part, it is not only "directly connected", but also "electrically connected" or "indirectly connected" with another element in between. "It includes the case where it is.

Throughout this specification, when one member is positioned on another member “on”, “on top”, “top”, “bottom”, “bottom”, “bottom”, it means that one member is on another member This includes cases where there is another member between the two members as well as when in contact.

Throughout the present specification, when a part “includes” a certain component, it means that the component may further include other components, not to exclude other components, unless otherwise stated.

Based on the genetic characteristics of an individual, demographic indicators such as age and gender, sociological factors, disease family history factors, and later Factors related to environmental behavior that can be exposed to change, obesity-related metrics that change according to these behavior factors, blood markers for recognizing blood abnormalities, and disease and pre-disease conditions caused by changes in the state of all these processes. It relates to a method of predicting an integrated disease outbreak model that identifies the risk of three diseases, including hypertension, diabetes, and their comorbidities and chronic kidney disease, in consideration of the biological continuity of the disease.

1 is a schematic configuration diagram of a device for predicting the development of chronic kidney disease according to an embodiment of the present application.

Referring to FIG. 1, the chronic kidney disease occurrence prediction apparatus 10 and the disease prediction server 20 may be linked through a network. Illustratively, the disease prediction server 20 may include a genomic data source of the Ansan-Anseong cohort, which is part of the Korean genomic epidemiology research project of the Center for Disease Control and traced tracking data from the 1st to the 7th. The disease prediction server 20 may provide information on the genomic and trace data sources of the Ansan-Anseong cohort, which is part of the Korean Genomic Epidemiology Project of the Korea Centers for Disease Control and Prevention, through the network as a device for predicting the occurrence of chronic kidney disease.

Chronic kidney disease occurrence prediction apparatus 10 includes all kinds of servers, terminals, or devices that transmit and receive data, contents, and various communication signals with the disease prediction server 20 through a network and have data storage and processing functions. can do.

The chronic kidney disease occurrence prediction apparatus 10 is a device that is linked to the disease prediction server 20 through a network, for example, a smart phone, a smart pad, a tablet PC, a wearable device, and the like. Personal Communication System (GSM), Global System for Mobile communication (GSM), Personal Digital Cellular (PDC), Personal Handyphone System (PHS), Personal Digital Assistant (PDA), International Mobile Telecommunication (IMT)-2000, Code Division Multiple Access (CDMA) )-2000, W-CDMA (W-Code Division Multiple Access), all types of wireless communication devices such as Wibro (Wireless Broadband Internet) terminals, and fixed terminals such as desktop computers and smart TVs.

Examples of networks for sharing information between the chronic kidney disease occurrence prediction apparatus 10 and the disease prediction server 20 include a 3rd Generation Partnership Project (3GPP) network, a Long Term Evolution (LTE) network, a 5G network, and World Interoperability (WIMAX). for Microwave Access (Wi-Fi) network, wired and wireless Internet (LAN), Local Area Network (LAN), Wireless Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), Bluetooth network, Wifi network , NFC (Near Field Communication) network, satellite broadcasting network, analog broadcasting network, DMB (Digital Multimedia Broadcasting) network, etc. may be included, but is not limited thereto.

The method for predicting the occurrence of chronic kidney disease described below may be performed in the apparatus for predicting the occurrence of chronic kidney disease 10. As another example, each step of the method for predicting the occurrence of chronic kidney disease may be performed in the disease prediction server 20. As another example, some steps of each step of the method for predicting the occurrence of chronic kidney disease may be performed by the apparatus for predicting the occurrence of chronic kidney disease, and the remaining steps may be performed by the disease prediction server 20. For example, the apparatus for predicting the occurrence of chronic kidney disease (10 is a step of a method for predicting a disease of chronic kidney disease, receives user input, transmits the received user input to the server, and transmits from the server in response to the user input) Only the function of displaying the displayed information on the screen may be performed, and the remaining steps of the method for predicting the occurrence of chronic kidney disease may be performed by the disease prediction server 20. Hereinafter, for convenience of description, prediction of the occurrence of chronic kidney disease A description will be given of an example in which the method for predicting the occurrence of chronic kidney disease in the device 10 is performed.

According to an embodiment of the present application, the apparatus for predicting the occurrence of chronic kidney disease 10 targets Koreans, including genomic information and population/social factors, environmental ecological factors, and obesity, which consist of a total of six repetitive measurements from 1 to 7 stages. It is possible to predict the occurrence of chronic kidney disease by receiving the cohort data of the disease management headquarters, which has investigated information such as body measurements, key blood marker information, and current disease morbidity for related factors, from the disease prediction server 20.

In addition, the apparatus for predicting the occurrence of chronic kidney disease uses cohort data of the Korea Centers for Disease Control and Prevention, which surveys information on genomic information, lifestyle, obesity-related body measurements, major blood marker information, and current disease status for Koreans. Therefore, it can predict the occurrence of chronic kidney disease. Chronic kidney disease outbreak prediction device (10) builds a model for the Ansan Anseong cohort of the Center for Disease Control and Prevention through three cohorts of urban-rural and rural-based cohorts, and urban-rural-based integrated cohorts. Verification can be performed based on the built model.

In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 may integrate hypertension, diabetes, chronic kidney disease morbidity and new incidence in the entire population for analysis of genomic markers, and classify the subject of morbidity/occurrence into the patient group. In addition, the apparatus for predicting the development of chronic kidney disease 10 may classify a control group of subjects who are not all affected by morbidity/occurrence and select genomic markers related to each disease of hypertension and diabetes. The apparatus for predicting the occurrence of chronic kidney disease 10 may construct a “integrated genomic score” by integrating a patient group and a control group. In particular, the genome information here was calculated using other microarray-based genotyping (genotyping) to calculate hypertension and diabetes markers, but the chronic kidney disease outbreak prediction device (10) is based on the latest version of 1K genome 3.0. The information was extended and applied to the analysis of genome markers.

In addition, the apparatus for predicting the development of chronic kidney disease 10 includes hypertension, diabetes, and diabetes-hypertension (high hypertension and diabetes are concurrently high due to biological pathological mechanisms) and chronic kidney disease (hypertension and diabetes are 4 of chronic kidney disease). About the factors related to each disease morbidity and each disease outbreak, and integrating them, for the two major diseases that occupy the largest scale among the major causative diseases), ' Integrated risk factors.

In addition, the apparatus for predicting the occurrence of chronic kidney disease cannot be seen as a risk factor of the disease, but in view of the composition of the general population, patients with hypertension, pre-hypertensive disease, diabetes, pre-diabetes Exist, and in these cases, the risk of developing other diseases is higher than that of normal conditions (e.g., hypertensive pre-diseases have a higher risk of developing diabetes and chronic kidney disease than normal conditions, and hypertensive patients have normal or pre-existing risks). The risk is higher than the disease status), and to apply the status of this one-group population directly to the risk prediction model, the'baseline's disease-to-disease status indicators' (hypertension, diabetes, chronic kidney disease morbidity and all diseases) Evaluation indicators for state morbidity) can be added to explanatory variables.

In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 was defined as the occurrence of hypertension and diabetes and the accompanying condition and chronic kidney disease as hypertension, diabetes mellitus, and their accompanying diseases as a result of the occurrence risk prediction model, and a description of the risk prediction model As for the variables, the'integrated risk factor panel' can be constructed by integrating both the'integrated genomic score of the baseline survey', the'integrated risk factor of repeated measures', and the'indicator of disease-to-disease status of the baseline survey'.

In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 can construct a risk prediction model by applying a statistically-based time-variable proportional probability risk model and a machine learning method such as neural network and random forest. In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 may select a model using a random forest having the highest predictive power as the final “integrated risk prediction model” through repeated training and comparison of the predictive power of the model.

In addition, the apparatus for predicting the occurrence of chronic kidney disease, in the machine learning method model, includes the'temporal continuity of the time and the state of the disease that can be changed after and when the factors are exposed' in the machine learning method model. It can be considered, and after all the explanatory variables are included, the result variables can also be generated to be included according to the natural history of disease and biological continuity of disease.

In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 is a factor that can be exposed from birth (dielectric) -> demographic indicators (age and gender) -> sociological factors (education level, income level, marital status) -> medical history Family history factor -> Behavioral factors that may change and be exposed afterwards -> Obesity-related measurement indicators that may change due to behavioral factors -> Blood markers that recognize hematologic abnormalities that may be altered by them -> All states It is possible to predict whether or not chronic kidney disease occurs through the same flow as the disease caused by the change and whether or not the entire disease state -> hypertension or diabetes alone disease -> hypertension-diabetic comorbid disease -> chronic kidney disease.

Figure 2 is a schematic block diagram of a chronic kidney disease device according to an embodiment of the present application.

Referring to FIG. 2, the chronic kidney disease device 10 includes a genomic marker selection unit 11, an integrated genomic index calculation unit 12, an integrated risk factor construction unit 13, an explanatory variable derivation unit 14, and disease outbreak prediction A model generation unit 15 and a disease prediction unit 16 may be included. However, the configuration of the chronic kidney disease device 10 is not limited thereto.

According to one embodiment of the present application, the genomic marker selection unit 11 may select genomic markers associated with a plurality of diseases. For example, a plurality of diseases may include hypertension, diabetes, diabetes-hypertension and chronic kidney disease. Illustratively, the genome marker selector 11 has a genome among all subjects in the Ansan Antarctic Cohort in order to select a genome marker that can be exposed from birth, whether or not blood pressure and blood pressure medications are taken, and whether fasting blood glucose and glycated hemoglobin diabetes are taken. With the information, it is possible to select targets capable of defining hypertension and diabetes as targets for selection of genomic markers related to multiple diseases.

In addition, the genome marker selection unit 11 may select the first dielectric marker associated with the first disease, and verify the selected first dielectric marker in each of the urban cohort and the rural cohort to determine the first dielectric marker. As an example, the first disease may be hypertension. For example, the genomic marker selection unit 11 may be provided with the Ansan Anseong cohort information from the disease prediction server 20. The genome marker selection unit 11 selects a first genome marker (eg, a hypertension-related genome marker) related to the first disease from the Ansan Anseong cohort, and verifies this in a city cohort and a rural cohort to determine the final hypertension-related genome marker Can be selected. Illustratively, the genome marker selection unit 11 targets 8,840 subjects who have genomic data and can define hypertension (with information on whether to take blood pressure or blood pressure medication) among 10,030 subjects in the Ansan Anseong Cohort. The first genetic marker associated with the disease can be selected. In addition, the genome marker selection unit 11 may integrate base data and tracking data. The genome marker selection unit 11 may be defined as a hypertensive patient group in the case of a subject having a systolic blood pressure equal to or greater than the first blood pressure, a diastolic blood pressure equal to or greater than the second blood pressure, or a person taking blood pressure medication. In addition, subjects who are never included in morbidity and development can be set as hypertension controls. In other words, the genome marker selection unit 11 integrates the basal data and the tracking data, and the hypertensive blood pressure (SBP) is greater than 130 mm/Hg, the diastolic blood pressure (DBP) is greater than 90 mm/Hg, or the patient is taking hypertension. It can be defined as, and subjects who have never been included in morbidity and development can be set as controls. As an example, the genome marker selection unit 11 can derive 7,444 people from the Ansan Anseong cohort, 858 people from the urban cohort, and 332 people from the rural cohort for the selection of genomic markers. However, this; It is not limited, and various embodiments may exist.

In addition, the genome marker selection unit 11 may select the second dielectric marker associated with the second disease, and verify the selected second dielectric marker in each of the urban cohort and rural cohort to determine the second dielectric marker. For example, the genomic marker selection unit 11 may be provided with the Ansan Anseong cohort information from the disease prediction server 20. The genome marker selection unit 11 selects a second genome marker (eg, diabetes-related genome marker) related to the second disease from the Ansan Anseong cohort, and verifies this in a city cohort and a rural cohort to determine the final hypertension-related genome marker Can be selected. The genome marker selection unit 11 is a second genome marker related to the second disease for 8,831 subjects who have genomic data and who can define diabetes (fasting blood glucose, HBA1C, whether to take diabetes drugs) among 10,030 subjects in the Ansan Anseong Cohort Can be selected. In addition, the genome marker selection unit 11 may be defined as diabetes when fasting blood glucose is greater than or equal to the first blood glucose level or glycated hemoglobin level is greater than or equal to the first blood glucose level or is taken from 1 to 7 periods. In other words, the genome marker selection unit 11 may define a case in which the fasting blood glucose is 126 or more, the glycated hemoglobin is 6.5% or more, or the diabetes drug is taken from 1st to 7th. The genomic marker selection unit 11 may be defined as a group of diabetic patients by integrating basis data and tracking data, having fasting blood glucose of 126 or more, glycated hemoglobin of 6.5% or more, or taking diabetes drugs. In addition, the dielectric marker selection unit 11 may set a subject that has never been included in morbidity and development as a control. As an example, the genome marker selection unit 11 finally, 7,444 people in the Ansan Anseong cohort, 858 people in the urban cohort, and 332 subjects in the rural cohort can be derived for the selection of genomic markers. However, this; It is not limited, and various embodiments may exist.

According to one embodiment of the present application, the genome marker selection unit 11 may expand gene information through genotype imputation of genome genotyping data. For example, genomic information in the Ansan Anseong population was analyzed by genome analysis using Affimetrix 6.0 array, and genome analysis was performed based on a genome chip composed of up to 600,000 SNPs at the time. However, since this chip is not able to provide information on all SNPs, it is a method to check SNPs at regular intervals, so it is difficult to obtain information on SNPs when the frequency is rare or heterogeneous in the Korean population. Of the 600,000 SNPs, only about 300,000 to 400,000 SNPs could be identified and genome markers could be found. The dielectric marker selection unit 11 can expand the genetic variation information by applying genotype imputation (hereinafter referred to as'genetic variation information extension'), which is a method of expanding the genetic variation information to compensate for the limitations of the dielectric chip. have. It can bring over tens of millions of genetic mutations based on information of more than several thousand people, and it can be used to expand genetic variation information and check the frequency of genetic variation. As a result, the dielectric marker selector 11 may extend the genetic variation information to hundreds of thousands of genetic variation information on the dielectric chip to about 80 million or more, which is the WGS level. The method of extending the genetic variation information is an analysis method that compares the reference dielectric information and the genetic variation information of the dielectric chip, and does not exist in the dielectric chip, but can statistically estimate and secure the genetic variation in the reference dielectric information.

FIG. 3 is a diagram showing a list of information that can be used as reference dielectric information for extending genetic variation information of a chronic kidney disease device according to an embodiment of the present application.

Referring to FIG. 3 as an example, FIG. 3 is a list of information that can be used as reference genome information of the genetic variation information extension, and the genome marker selection unit 11 is the reference genome information calculated from the top 1K Genome project. Eastern Asian results can be applied to obtain reference genome information for genetic variation information expansion.

As an example, the genome marker selection unit 11 may expand the genetic variation information based on a subject having genome information among the Ansan Anseong cohorts. Marker selector 11 uses QP (MAF: 0.01, missing rate per sample: 0.05, missing rate per SNP: 0.02, Hardy-weinberg: <10) using the'Plink' program before performing Imputation. -6) can be performed. The marker selector 11 can then change the annotation to'HG19' and remove the SNP that cannot be annotated. In addition, the marker selection unit 11 can use'Shapeit2, 1000 Genome Phase 3 East Asian population' for pre-phasing, and direct imputation can be performed using'impute2'. In addition, the marker selector 11 may use'Qctool v2' in the Plink 2.0 version for file converting. In addition, the marker selector 11-filtering may be performed based on probability 0.9, completion rate 0.98, and info (r2) 0.7 or higher among imputation results. The marker selection unit 11 can expand the original genotyping data to 31,563,540 SNPs after the final impution. In other existing studies, hypertension and diabetes markers were calculated using genomic analysis information (Affimetrix 6.0 array genotyping), but the marker selection unit 11 expands the information based on the latest version of 1K genome 3.0 (Imputation) and uses that information as a genomic marker It was used for analysis.

According to one embodiment of the present application, the genomic marker selection unit 11 selects a single-base polymorphism (SNP) marker associated with a plurality of diseases based on a regression analysis algorithm, a single-base polymorphism (SNP) marker, and a first genome Core gene information may be derived by considering at least one of the marker and the second dielectric marker. As an example, the genome marker selection unit 11 may calculate a risk (crude odds ratio [crude OR]) of the SNP marker itself to a disease using logistic regression. In addition, since the genome marker selection unit 11 has the largest effect on both the disease and other risk factors in the case of the gender and age variables, the results of the disease risk (Age-sex adjusted OR) of the SNP markers corrected for the two variables are shown. Can be calculated.

For example, the dielectric marker selection unit 11 may perform detailed quality control in SNPs from the input data. The dielectric marker selector 11 can remove 11,313,891 SNPs by Genotyping missing <5%. In addition, the dielectric marker selector 11 can remove 183 SNPs by HWE p <1E-06. In addition, the dielectric marker selection unit 11 may remove 17,950,690 SNPs by MAF (Minor allele frequency) <1%. The genome marker selection unit 11 may derive SNPs with a final 2,298,777 SNPs, and missing objects may not be further processed since QC has already been performed in advance in the questionnaire.

In addition, the genome marker selection unit 11 may perform correction based on a population group when heterogeneity between groups to be corrected exists through QQ plot and lambda. The genome marker selector 11 may select an SNP below a threshold based on the statistical significance of SNP according to Phenotype <1 x 10 ^-6 . The genome marker selector 11 may generate a Manhattan plot in which the P-value of each SNP is increased in the order of chromosome and physical distance for visualization of the analysis result. The dielectric marker selection unit 11 performed a Cochran-Armitage test (1df) under the assumption of the additive effect of each SNP index to calculate the raw p-value and can confirm the result through the Manhattan plot. In other words, the dielectric marker selection unit 11 provides a Manhattan plot in which the P-value of each SNP is increased in the order of chromosome and physical distance, so that the user can be provided with visualized analysis results. In addition, the genome marker selection unit 11 calculated the risk of the SNP marker itself on the disease (crude odds ratio [crude OR]) using logistic regression, and also gender and age variables, as well as disease Since the effect on the other risk factors at the same time is the largest variable, it is possible to calculate the disease risk (Age-sex adjusted OR) result of the SNP markers corrected for both variables.

In other words, the genome marker selection unit 11 may select a genome indicator that is a marker of a disease and can be used to predict disease risk. The genome marker selection unit 11 may select SNP markers commonly associated with a corresponding disease in the Crude and Age-sex adjusted results for each of a plurality of diseases. The genome marker selector 11 may select a marker (SNP that satisfies odds ratio (OR) ≥ 1) in the direction of increasing the risk of disease among the selection markers as a genetic marker. Diabetes is a common comorbid disease in hypertensive patients, and hypertension is a common comorbid disease in diabetic patients. Hypertension and diabetes affect the same target organs through the vascular network, and hypertension and diabetes are chronic kidney disease and endovascular diseases. , Renal failure, heart failure, ocular diseases, etc., and these diseases rapidly increase their incidence and mortality.

Therefore, the genome marker selection unit 11 integrates the genomic SNP markers of the first disease (hypertension) and the second disease (diabetes) and uses them for hypertension, diabetes, hypertension-diabetes, and chronic kidney disease to develop chronic kidney disease. Predictable. The genome marker selection unit 11 may construct a final genome marker panel by integrating SNP markers selected under such biological feasibility.

4 is a view for explaining a significant genomic marker associated with hypertension according to an embodiment of the present application. 4(a) is a crude model in which nothing is corrected using logistic regression. In addition, Figure 4 (b) is an age-sex adjusted model that additionally corrects gender and age. The genome marker selector 11 displays qq-plot and genetic index results related to hypertension in each of the Crude model, which does not correct anything using logistic regression, and the Age-sex adjusted model, which additionally corrects gender and age. You can check the results through the Manhattan plot.

In addition, the dielectric marker selection unit 11 may finally select an SNP having an Odds ratio (OR) of 1 or more among common SNPs in both Crude results and results of correcting age and gender. The final four selected SNPs are shown in Table 1. Table 1 is the final selected hypertension SNP.

SNP Gene CCHR PPOS MAF Alleles OR L95 U95 P rs35639474 6 0.928329 -/T 1.35641 1.19034 1.54564 4.77E-06 rs57808874 STOML3 13 0.756094 -/G 1.21593 1.11934 1.32085 3.66E-06 rs7040217 RAD23B 9 0.822863 A/C 1.19135 1.10399 1.28563 6.61E-06 rs7986278 13 0.960859 A/G 1.56804 1.32225 1.85952 2.33E-07

5 is a view for explaining a significant genomic marker associated with diabetes according to an embodiment of the present application. 5(a) is a crude model in which nothing is corrected using logistic regression. In addition, Figure 4 (b) is an age-sex adjusted model that additionally corrects gender and age. The genomic marker selection unit 11 may identify genetic markers related to diabetes in each of the Crude model, which has not corrected anything by using logistic regression, and the Age-sex adjusted model, which additionally corrects sex and age. In addition, the genome marker selection unit 11 may finally select an SNP having an Odds ratio (OR) of 1 or more and p-value<10e-6 among common SNPs in both Crude results and age and gender correction results. . Table 28 shows the 28 selected SNPs. Table 2 is the final selected diabetes SNP.

SNP Gene CHR POS MAF Alleles OR L95 U95 P rs10536170 CDKAL1 6 2.1E+07 0.535241 -/TATAT 1.2966 1.19898 1.40217 7.84E-11 rs138420022 CDKAL1 6 2.1E+07 0.532877 -/T 1.29703 1.19891 1.40318 9.18E-11 rs9356744 CDKAL1 6 2.1E+07 0.522241 C/T 1.29027 1.19407 1.39424 1.15E-10 rs34499031 CDKAL1 6 2.1E+07 0.525247 -/AA 1.28926 1.19316 1.3931 1.29E-10 rs9356748 CDKAL1 6 2.1E+07 0.521774 A/T 1.26924 1.17589 1.37001 9.55E-10 rs6906327 CDKAL1 6 2.1E+07 0.528224 A/G 1.2584 1.16749 1.3564 1.88E-09 rs9358356 CDKAL1 6 2.1E+07 0.525268 C/T 1.259 1.16703 1.35822 2.67E-09 rs9295474 CDKAL1 6 2.1E+07 0.52291 C/G 1.2532 1.16206 1.35148 4.66E-09 rs9348440 CDKAL1 6 2.1E+07 0.532331 C/T 1.24481 1.15518 1.34141 9.28E-09 rs9356743 CDKAL1 6 2.1E+07 0.509704 C/T 1.24565 1.15059 1.34856 5.85E-08 rs56099357 CDKAL1 6 2.1E+07 0.558728 -/T 1.23076 1.14152 1.32698 6.43E-08 rs4515379 CDKAL1 6 2.1E+07 0.563085 C/G 1.22286 1.13425 1.3184 1.59E-07 rs80151164 8 6.1E+07 0.934038 A/C 1.51167 1.27822 1.78775 1.38E-06 rs77039794 6 2.1E+07 0.556653 A/G 1.21547 1.12272 1.31589 1.45E-06 rs75897742 8 6.1E+07 0.940407 A/G 1.54327 1.29299 1.842 1.54E-06 rs2328529 CDKAL1 6 2.1E+07 0.586855 A/C 1.19771 1.11127 1.29087 2.35E-06 rs17193507 LOC107987429 6 3.1E+07 0.960873 G/T 1.58192 1.30641 1.91553 2.63E-06 rs75925737 8 6.1E+07 0.938786 A/G 1.52515 1.27831 1.81966 2.79E-06 rs116381594 8 6.1E+07 0.939492 C/G 1.52499 1.27806 1.81963 2.84E-06 rs114462576 8 6.1E+07 0.94065 G/T 1.52641 1.27809 1.82297 3.03E-06 rs115952070 8 6.1E+07 0.940802 C/T 1.52735 1.27851 1.82462 3.04E-06 rs116468467 8 6.1E+07 0.940623 C/T 1.52611 1.27792 1.82252 3.04E-06 rs114961690 8 6.1E+07 0.941057 A/G 1.52852 1.27901 1.82671 3.07E-06 rs2932331 5 3863434 0.015545 A/G 2.55321 1.71879 3.79272 3.44E-06 rs11187007 IDE 10 9.4E+07 0.326168 A/G 1.20482 1.11315 1.30405 3.94E-06 rs191296992 LPP 3 1.9E+08 0.98986 A/G 2.98624 1.87495 4.75618 4.09E-06 rs116198656 5 3867721 0.985373 A/G 2.58996 1.72272 3.8938 4.77E-06 rs115880178 8 6.1E+07 0.938456 C/T 1.50468 1.26266 1.79311 4.96E-06

According to one embodiment of the present application, the integrated genomic index calculation unit 12 may calculate the integrated genomic index using genomic markers associated with a plurality of diseases selected by the genomic marker selection unit 11. For example, in the case of the integrated dielectric marker panel constructed by the integrated dielectric index calculator 12, dozens of SNP markers are included. When these are used at the same time, the gene actually affects the disease. In many cases, it is considered to be 5-7% or less. When many SNPs enter the model at the same time, the genetic impact of the disease prediction is greater than the environmental impact. Thus violating biological feasibility. In order to solve this problem, the integrated genome index calculation unit 12 may calculate it in the form of a genomic score to calculate it as one index. Also, the integrated genome index calculation unit 12 may calculate a genome score. . The polygenic risk score can be calculated by multiplying the beta value of each SNP and the probability value of each individual SNP disease and then summing them and dividing by the total number of SNPs.

Here, X _j is a genomic score (polygenic risk score), OR _i is the beta value of each SNP, SNP _ij is the probability value for each individual SNP disease, m is the total number of SNPs.

[Equation 4]

According to one embodiment of the present application, the integrated risk factor building unit 13 may build factors associated with a plurality of diseases to construct an integrated risk factor model. A plurality of diseases may include hypertension, diabetes, diabetes-hypertension and chronic kidney disease. The integrated risk factor building unit 13 may derive factors related to each disease outbreak and combine the derived factors with environmental factors to build an integrated risk factor model for their condition.

6 is a view for explaining the results of the hypertensive integrated genomic score construction according to an embodiment of the present application.

Referring to FIG. 6 as an example, as a result of calculating the genomic score with the SNP marker finally selected by the integrated risk factor building unit 13, the distribution of PRS of hypertension is shown in FIG. 6.

Table 3 shows the default values for the hypertensive polygenic risk score in the integrated factor building unit (13).

Minimum Median medium Maximum 0.00 0.05240 0.06465 0.34822

7 is a view for explaining the results of the diabetes integrated genomic score construction according to an embodiment of the present application. Referring to FIG. 6 for example, the genomic score as the SNP marker finally selected by the integrated risk factor building unit 13 As a result, the distribution of PRS in diabetes is shown in FIG. 7.

Table 4 shows the default values for the hypertensive polygenic risk score in the integrated factor building unit (13).

Minimum Median medium Maximum 0.04813 0.1903 0.1996 0.57534

According to one embodiment of the present application, the integrated risk factor building unit 13 includes basic demographic factors, sociological factors, disease and family history factors, environmental behavior factors, obesity indicator factors, and blood abnormality indicator factors, Factors related to the plurality of diseases may be derived and an integrated risk factor model may be constructed by considering at least one of the underlying disease morbidity factors. For example, the integrated risk factor construction unit 13 may construct an integrated risk factor model by including variables corresponding to gender and age as basic demographic factors. In addition, the integrated risk factor building unit 13 may construct an integrated risk factor model by including variables corresponding to education level and income level as sociological factors. In addition, the integrated risk factor building unit 13 is a family history of cardiovascular disease as a disease and family history factor (a family history of hypertension family history, diabetes family history, heart disease family history, 1 if each has a family history, An integrated risk factor model can be constructed by including variables that correspond to zero (0), absence of all family history of the disease, and one or more cases (1 point).

In addition, the integrated risk factor building unit 13 may construct an integrated risk factor model by including variables corresponding to alcohol drinking state, regular exercise state, and cigarette smoking state as factors related to environmental behavior. In addition, the integrated risk factor building unit 13 may construct an integrated risk factor model by including variables corresponding to the obesity index, body mass index, and abdominal obesity index, as obesity-related indicator factors. In addition, the integrated risk factor building unit 13 includes a variable corresponding to lipid abnormality indicators TG, HDL-cholesterol, total cholesterol and liver abnormality indicator ALBUMIN as indicators for recognizing blood abnormalities. Can build.

In addition, the integrated risk factor building unit 13 is a cardiovascular disease disease history score (whether diagnosing congestive heart failure, diagnosing coronary artery disease, stroke, paralysis, etc.) as the underlying disease morbidity (blood abnormality, blood pressure abnormality, and medical history) Diagnosis of cerebrovascular disease: 3 cases of disease history are calculated by classifying 1 as a case of history, 0 as a case of absence, 0 as a case of no history, and 1 point as having one or more diseases) , Pre-hypertension, pre-diabetes, pre-diabetes, pre-diabetes, chronic kidney disease prevalence variables can be included to build an integrated risk factor model.

On the other hand, the integrated risk factor building unit 13 can build an integrated risk factor model including an integrated risk factor and an integrated genomic index (dielectric score). The integrated genomic index (dielectric score) may be a result calculated by the integrated genomic index calculation unit 12. The integrated genomic index calculation unit 12 may calculate a genomic score to include all genomic indices calculated from hypertension and diabetes.

In addition, the integrated risk factor building unit 13 may build an integrated risk factor model corresponding to each of a plurality of diseases. The integrated risk factor construction unit 13 may generate an integrated risk factor model predicting the occurrence of hypertension by allowing the genomic score to be calculated by considering only the genomic indicators associated with hypertension. In addition, the integrated risk factor building unit 13 may generate an integrated risk factor model for predicting the occurrence of diabetes by allowing the genomic score to be calculated in consideration of only genomic indicators associated with diabetes. In addition, the integrated risk factor building unit 13 may generate an integrated risk factor model by calculating the genomic scores based on all genomic indicators in the diabetes-hypertensive co-occurrence and chronic kidney disease integrated risk factor model.

According to one embodiment of the present application, the explanatory variable deriving unit 14 may derive data of a subject having at least one of a plurality of diseases as an explanatory variable. In other words, the explanatory variable deriving unit 14 may derive an explanatory variable to be included in the disease outbreak prediction model. The explanatory variables may include risk factors associated with hypertension and diabetes prevalence of the entire population and risk factors associated with hypertension and diabetes. In the case of hypertension-diabetes, the risk factors of the two diseases were set to be included, and no risk factors were derived.

The explanatory variable derivation unit 14 is a disease caused by hypertension and diabetes in the case of chronic kidney disease, but it is judged that it is difficult to cover both risk factors with only the risk factors of both diseases, so that the risk factors associated with each disease can be selected. have. The reason is another cause of chronic kidney disease, and kidney disease such as polycystic kidney disease and glomerulonephosis.

In addition, the explanatory variable derivation unit 14 did not analyze due to lack of power in the base group because the morbidity was less than 5% of the total group, and instead, the occurrence during the follow-up period was about 20% of the entire group, so the risk for occurrence Factors can be analyzed and included. The explanatory variable derivation unit 14 integrates all risk factors of high blood pressure, diabetes, and diabetes, and risk factors for chronic kidney disease, as well as integrated risks for factors related to environmental and other changes in the human body, other than genetic factors. Factor models can be built. Exemplarily, the genome marker selection unit 11 selects SNPs markers related to hypertension and diabetes by expanding the genome genotyping source information through 1k genome information-based imputation in relation to genetic factors, and the integrated genome index calculator 12 ) Can be converted into a polygenic risk score (PRS) that combines multiple markers to form a single genomic indicator. The explanatory variable derivation unit 14 can use this as an index to construct an integrated risk factor model. The variables included in the final model were thus composed of genomic indicators (dielectric scores) and integrated risk factors, and the final ones can be called the integrated risk factor model.

According to one embodiment of the present application, the explanatory variable derivation unit 14 cannot be seen as a risk factor for disease, but when viewing the compositional state of the general population, a hypertensive patient, a hypertensive all disease state, a diabetic patient, a diabetic all disease state Because patients are present and in these cases, the risk of developing other diseases is higher than that of normal conditions (e.g., hypertensive all disease status has a higher risk of developing diabetes and chronic kidney disease than normal conditions, and hypertensive patients are normal) However, the risk is higher than that of all diseases). To apply the condition of these general populations as an explanatory variable directly to the risk prediction model, the'baseline disease-to-disease status indicator' (hypertension, diabetes, chronic kidney disease) The evaluation index for morbidity and morbidity morbidity) can be derived.

According to one embodiment of the present application, the integrated risk factor building unit 13 is a result variable of a disease outbreak prediction model in relation to hypertension, diabetes, and the accompanying state of diabetes and chronic kidney disease due to hypertension, diabetes, and their accompanying diseases. The variable can be applied. In addition, the integrated risk factor building unit 13 integrates the results of calculating the genomic indicators of the integrated genomic index calculation unit 12, the integrated risk factors, and the disease-to-disease status index of the underlying investigation derived from the explanatory variable derivation unit 14 To build an integrated risk factor model.

In addition, the integrated risk factor building unit 13 may select indicators that reflect environmental exposure after birth and changes in them and biological changes in the human body. The integrated risk factor building unit 13 may exclude variables that are not directly related to disease prediction, such as personal identifiers and survey dates, and variables that are directly related to disease results or have a missing ratio of more than 20% for all variables. . Thereafter, the integrated risk factor building unit 13 selects three variable selection methods (forward, backward, stepwise) in a multinomial logistic regression model in case of hypertension and diabetes, and Cox proportional probability risk in case of hypertension, diabetes, chronic kidney disease. Based on the three variable selection methods (forward, backward, and stepwise) in the model, variables from at least two processes can be selected.

The integrated risk factor building unit (13) then uses the basic demographic factors (gender, age) and sociological factors (education to build an integrated development model for hypertension, diabetes, and their accompanying disease and chronic kidney disease based on clinical judgment criteria). Level, income level, marital status, family history factor (cardiovascular disease family history score: hypertension family history, diabetes family history, heart disease family history) 1 for each family history of 3 family history, 0 for none Too cows, family history of all diseases are 0, and one or more cases are classified as 1 point), factors related to environmental behavior (drinking, regular exercise, smoking status) and obesity index (body mass index, waist Circumference), indicators for recognizing hematologic abnormalities (lipid abnormalities TG, HDL-cholesterol, total cholesterol and anemia indicators hemoglobin, liver abnormality indicators ALT), underlying disease morbidity (cardiovascular disease history score (congestion) Whether to diagnose sexual heart failure, whether to diagnose coronary artery disease, whether to diagnose cerebrovascular diseases such as stroke and stroke, each case has a history of 1, 0 if none, and 0 if none. , Calculated by classifying one or more cases as one point) 19 variables can be finally selected including hypertension, pre-hypertension, diabetes, pre-diabetes, and chronic kidney disease). The variables included in the final model were therefore composed of genomic indicators (dielectric scores) and integrated risk factors, and the final ones can be called the “integrated risk panel”.

According to an embodiment of the present application, the disease occurrence prediction model generator 15 may build a disease occurrence prediction model by using an integrated genomic indicator and an integrated risk factor model as inputs. Illustratively, the disease outbreak prediction model uses the statistical method (proportional probability risk model) and the machine learning method (support vector machine, neural network, random forest) to develop hypertension, diabetes, hypertension-diabetic comorbidities and chronic kidney disease. You can build an integrated risk prediction model for. The disease occurrence prediction model generator 15 may build a disease occurrence prediction model by applying a first statistical technique and a second machine learning method. For example, the first statistical technique could be a Cox proportional hazards model. In addition, the second machine learning method may include any one of a support vector machine (SVM), a recurrent neural network (RNN), and a random forest (Random Forest). In addition, the disease occurrence prediction model generation unit 15 compares the disease occurrence prediction model generated based on various algorithms and selects a model using the most predictive random forest as the final integrated risk prediction model (disease occurrence prediction model). Can.

Illustratively, the disease occurrence prediction model generation unit 15 uses the Cox regression model to identify disease risk according to a combination of risk factors of known diseases using hypertension, diabetes genomic markers, and repeatedly measured factors. Disease risk by factor can be confirmed. The disease occurrence prediction model generation unit 15 includes a statistical model method using a Cox proportional risk model, a support vector machine (SVM) based on artificial neural networks as a machine learning method, and a recurrent neural network (RNN) as a method of deep learning. And a random forest (Random Forest, RF) can be used to build a disease outbreak prediction model. The disease occurrence prediction model generator 15 may set a model having the most predictive result among prediction models constructed from a plurality of neural networks as a final disease prediction model.

The disease incidence prediction model generator 15 inputs genomic marker information of a plurality of diseases, a risk factor variable of a plurality of diseases, and a disease risk of chronic kidney disease, thereby inputting genomic marker information and a plurality of diseases of a plurality of diseases A disease risk machine learning model that learns information of the relationship between disease risk factors of disease and disease risk to learn the degree of relationship between disease risk of chronic kidney disease can be generated. For example, the machine learning model may generate a machine learning model using the Cox regression model.

The disease incidence prediction model generator 15 uses the time-variant Cox regression model to determine the disease risk according to changes in factors measured repeatedly by using hypertension, diabetes genomic markers, and repetitively measured factors. Can do For the final disease outbreak prediction model, a statistical model method using a time-variant cox regression model and a support vector machine (SVM) based on an artificial neural network as a machine learning method, and a recurrent neural network (RNN) and a random method of deep learning You can build a disease outbreak prediction model using Forest (RF). The disease occurrence prediction model generator 15 may set a model having the most predictable result as a final disease prediction model.

According to one embodiment of the present application, the disease occurrence prediction model generation unit 15 divides the variables derived from the genomic marker selection unit 11, the integrated genomic index calculation unit 12, and the integrated risk factor construction unit 13 Can be separated by. For example, in the case of continuous variables such as the disease occurrence prediction model generation unit 15 body measurements and blood markers, they were divided into normal ranges and out-of-normal risk levels based on metabolic syndrome diagnosis criteria, and based on body mass index. In the case of categorical variables such as obesity as defined, age distribution defined by 4 sections (under 50, 50-60, 60-70, over 70), variables related to environmental behavior, and PRS defined as quartile In this case, dummy variables for each level can be created and classified to have binary values. By dividing the variables into dichotomous types in the disease occurrence prediction model generator 15, it is possible to evaluate the effect of each variable on disease occurrence by state.

In addition, the disease occurrence prediction model generator 15 may construct a disease occurrence prediction model by dividing it into a training data set and a test data set to verify the disease occurrence prediction model. As an example, the disease occurrence prediction model generation unit 15 needs to build a model and require a verification step, so Ansan Anseong cohort is trainng set, city cohort is test set1, rural cohort is test set2, and urban and rural integrated cohort Can be defined as test set3 to perform external verification.

The disease occurrence prediction model generator 15 can confirm the degree of generalization of the constructed model by checking the prediction degree by applying it to the training data set and the test data set. The disease outbreak prediction model generator 15 divides the entire subjects into a ratio of 7 to 3, and 70% of the subjects with normal subjects and each disease outbreak (hypertension, diabetes, hypertension-diabetes, chronic kidney disease) are developed for model development. 30% was used as test data for model verification. In the training data set, we tried to predict the risk of each outcome variable by changing only the outcome variable using the variables in the integrated risk factor panel.For this purpose, the statistical method is the time-variant Cox proportional probability risk model, and the machine learning method. As an alternative, we can construct a learning model by applying two different algorithms to each of two artificial neural network based methods (Support vector machine and deep learning recurrent neural network (RNN)) and random forest method.

The support vector machine is a supervised learning model, and by applying classification and regression analysis, when a set of data belonging to one of the two categories is given, it is non-stochastic to determine which category the new data belongs to based on the given set of data. This is a method of creating a bilinear linear classification model to identify and separate the closest learning data and the farthest hyperplane. The disease occurrence prediction model generation unit 15 applied the spline and variance analysis RBF kernel method among various application methods.

The random forest method is a kind of ensemble learning method used for classification, regression analysis, etc., and outputs an average predicted value of classification or regression analysis from a plurality of decision trees constructed in a training process and can operate in the next process. First, the tree structure and parameters are automatically learned from the learning data, so that the independent training step and learning can be performed to reach the end node, whether or not a disease occurs. In addition, in the typical random forest method, it is a principle to randomly select variables from data and make another model by randomly extracting other variables through reconstruction extraction. A situation may arise in which a variable with less relevance comes in, or those with intermediate results that should have occurred in the latter than the former come before variables corresponding to the cause. In order to solve this problem to some extent, an ensemble learning method is maintained, but an algorithm for grouping variables into groups and adjusting each group to include variables in a step-by-step manner can be developed and used in order to include variables in a tree under a biological temporal algorithm. .

Disease occurrence prediction model generation unit 15 Verification data for internal validity verification to verify whether the model generated in the training data set is reproduced in the test data set to test the predictive power of each model The set (test set) was subjected to 1,000 permutations using the boot-straping technique, and then the probability calculation method of each calculated model was applied as it was to verify that the prediction value of the training set and the prediction value of the test set matched. ROC-curve and AUC values in set can be suggested. In the case of a random forest, an AUC value can be calculated using ntree=300. The predictive power of each model was tested for the difference with a 95% confidence interval of AUC, and the model with the highest AUC can be selected as the final predictive model.

8 is a diagram showing variables sequentially included in a predictive model in consideration of the timing of factor exposure and temporal continuity of a state that can be changed afterwards and natural history of disease according to an embodiment of the present application.

8 is a demographic, sociological factors, environmental behavior-related factors, obesity-related measurement indicators, current health status according to blood markers, based on genomic scores using genomic markers that are significant for diseases based on the genetic characteristics of the individual. It is a diagram showing the flow from hypertension and diabetes, their multi-disease hypertension-diabetes, and further chronic kidney disease, as time-series data on disease morbidity at baseline is added.

9 is a diagram for explaining a deep learning model structure that integrates time series data and genetic data according to an embodiment of the present application. 10 is a schematic diagram of a disease outbreak prediction model according to an embodiment of the present application. 11 is a schematic diagram of a random forest prediction model according to an embodiment of the present application. 12 is a diagram schematically showing a process of training models in various cases according to an embodiment of the present application by an ensemble technique.

According to one embodiment of the present application, the disease incidence prediction model generation unit 15 measures a plurality of state variables including life status variables and health state variables of patients with chronic kidney disease, core genetic information, and disease risk of chronic kidney disease. As an input, a disease outbreak prediction model for learning the degree of relationship between at least one of a plurality of state variables and core genetic information and disease risk of chronic kidney disease may be generated.

The disease occurrence prediction model generation unit 15 may generate a machine learning model that learns information on a relationship between at least one of a plurality of state variables and genetic information and a disease risk of chronic kidney disease. For example, the machine learning model may generate a machine learning model using a recurrent neural network (RNN) and a multi-layer perceptron neural network (MLP).

According to one embodiment of the present application, the disease occurrence prediction model generator 15 may input a gene related to each disease of a chronic kidney disease by connecting a multi-layer perceptron neural network to a circulatory neural network. In addition, the disease occurrence prediction model generation unit 15 sequentially inputs and analyzes the correlation between variables as well as time-dependent correlation of each epidemiological variable through a plurality of state variables that are repeatedly measured and analyzed. can do.

In addition, the disease occurrence prediction model generation unit 15 may repeatedly measure information of a subject's subject state variable and subject's gene and input the repeatedly measured information. The disease occurrence prediction model generation unit 15 may check whether there is a change in the lifestyle for repeated measured values such as lifestyle, body measurement, and clinical values based on the subject's subject state variables and the information of the subject's genes. The disease outbreak prediction model generation unit 15 may classify groups showing similar patterns among repeated measured values, generate clusters for each, and distinguish groups showing similar lifestyle changes by gender and disease. The disease occurrence prediction model generation unit 15 may select a significant gene related to a change in lifestyle for each disease of chronic kidney disease based on the subject's target gene information. A significant gene may be a gene associated with each disease of chronic kidney disease.

For example, referring to FIG. 9, the disease occurrence prediction model generation unit 15 sequentially inputs the subject's genetic data of the repetitively measured subjects into the circulatory neural network among artificial neural networks, and changes the lifestyle of each disease of chronic kidney disease. A significant gene associated with can be connected to the circulatory neural network through a multi-layer perceptron.

For example, referring to FIG. 10, the disease occurrence prediction model generator 15 applies a circulatory neural network among artificial neural networks capable of inputting time series data such as a plurality of state variables including living state variables and health state variables. You can create machine learning models. The disease occurrence prediction model generator 15 may additionally connect a multi-layer perceptron neural network to the last layer of the existing circulatory neural network in order to integrally input genetic information collected at a single time point. The disease occurrence prediction model generation unit 15 may set the presence/absence of chronic kidney disease occurrence in the last output layer.

For example, the artificial neural network may be divided into three layers: an input layer, a hidden layer, and an output layer. Each layer is composed of nodes, and the input layer can receive input data from outside the system and transmit input data to the system. The hidden layer is located inside the system and can take input data, process the input data, and calculate the result. The output layer can calculate the system output value based on the input value and the current system state. The input layer may input values of a predictive variable (input variable) for deriving a predicted value (output variable). If there are n input values in the input layer, the input layer will have n nodes, and the value input to the input layer in the present application may be a plurality of state variables and genetic information including a living state variable and a health state. The hidden layer can receive input values from a plurality of input nodes, calculate weighted sums, and apply the values to the transfer function to transfer them to the output layer. For example, the input layer of the machine learning model may be a plurality of status information, genetic information, a hidden layer at a previous time, the hidden layer may be a plurality of status information, or a grouping of a plurality of status information, and the output layer may be a disease risk. It may indicate that.

According to one embodiment of the present application, when a first state variable among a plurality of state variables is used as an input layer and a second state variable among a plurality of state variables is a hidden layer, the disease occurrence prediction model is information on a relationship between the input layer and the hidden layer A first learning for learning may be performed. In addition, the machine learning model is the first learning that learns information of the relationship between the input layer and the hidden layer when the current viewpoint state variable of the plurality of state variables is the hidden layer and the current viewpoint state variable of the plurality of state variables is the hidden layer. You can do

The machine learning model can learn the degree of relationship between the input layer and the hidden layer based on Equation (1). The degree of relationship may mean a value obtained by calculating a weighted sum of information received at the input layer, but is not limited thereto.

는 t 시점에서의 은닉층이고,

은 t시점의 이전 시점 은닉층이고,

는 제 1 상태 변수이고,

는 입력층과 은닉층 사이의 제 1 유형의 관계의 정도를 나타내는 제 1 가중치이고,

는 입력층과 은닉층 사이의 제 2 유형의 관계의 정도를 나타내는 제 2 가중치이다. 예시적으로, [수학식 1]에서

는 t시점의 복수의 상태 변수 중 제 1 상태 변수이고,

는 t시점의 은닉층을 나타내고

는 복수의 상태 변수(입력 변수)와 은닉층간의 가중치이고,

는 은닉층들간의 가중치일 수 있으나, 이에 한정되는 것은 아니다. 일예로, 제 1 유형의 관계의 정도는 시간에 따른 복수의 상태 변수들관의 상관관계(가중치)일 수 있고, 제 2 유형의 관계의 정도는 복수의 상태 변수간의 상관관계(가중치)일 수 있으나, 이에 한정되진 않는다. At this time,

Is the hidden layer at time t,

Is the hidden layer before t

Is the first state variable,

Is a first weight indicating the degree of the first type of relationship between the input layer and the hidden layer,

Is a second weight representing the degree of the second type of relationship between the input layer and the hidden layer. For example, in [Equation 1]

Is a first state variable among a plurality of state variables at time t,

Denotes the hidden layer at time t

Is a weight between a plurality of state variables (input variables) and a hidden layer,

May be a weight between hidden layers, but is not limited thereto. For example, the degree of relationship of the first type may be a correlation (weight) of a plurality of state variables over time, and the degree of relationship of a second type may be a correlation (weight) of a plurality of state variables. However, it is not limited thereto.

The disease outbreak prediction model inputs a plurality of state variables (for example, individual lifestyle and health state variables) measured repeatedly in the circulatory neural network expressed in Equation 1, as well as lifestyle changes as well as correlation with time. You can even analyze the correlation between health status variables.

According to one embodiment of the present application, the disease occurrence prediction model may perform a second learning to learn information of a relationship between the hidden layer and the output layer when the hidden layer and the genetic information are used as the input layer and the disease risk is the output layer. In addition, the machine learning model may perform a second learning to learn information on the relationship between the hidden layer and the output layer when the hidden layer and the genetic information are used as the input layer and the disease risk is the output layer.

The machine learning model can learn the degree of the relationship between the hidden layer and the output layer based on [Equation 2]. The second learning can learn the degree of relationship between the hidden layer and the output layer based on [Equation 1] and [Equation 2]. The machine learning model can learn information on the relationship between the input layer, the hidden layer, and the output layer based on [Equation 1] and [Equation 2], and the predicted result of disease risk as a result of the output layer.

는 은닉층과 출력층 사이의 관계의 정도를 나타내는 제 3 가중치이고,

는 은닉층이고,

는 입력층 중 유전자 정보와 출력층 사이의 관계의 정도를 나타내는 제4 가중치이고, z는 입력층 중 유전자 정보일 수 있다. 일예로, 제 3 가중치는 질병 위험을 예측하기 위해 복수의 상태 변수와 출력층 사이의 관계를 나타낸 관계의 정도이고, 제 4가중치는 특정 유전자에 가중치를 부여하기 위한 유전자 정보와 출력층 사이의 관계의 정도일 수 있다. At this time, y is the output layer,

Is a third weight indicating the degree of relationship between the hidden layer and the output layer,

Is a hidden layer,

Is a fourth weight representing the degree of relationship between the gene information and the output layer in the input layer, and z may be the gene information in the input layer. In one example, the third weight is a degree of a relationship representing a relationship between a plurality of state variables and an output layer to predict a disease risk, and the fourth weight is a degree of a relationship between a gene information and an output layer for weighting a specific gene Can.

According to one embodiment of the present application, since the genetic information is collected at a single time point, a multi-layer perceptron neural network may be connected to and input to the last layer of the circulatory neural network as shown in [Equation 2] in order to integrate it into the circulatory neural network. Exemplarily, the genetic information was collected in the form of a single nucleotide polymorphism, and it is possible to convert and input previously known genetic information for each chronic kidney disease into a risk fator according to an allele. The machine learning model may learn the degree of the relationship between the hidden layer and the output layer, that is, the weight between the hidden layer and the output layer through the second learning.

According to one embodiment of the present application, the disease occurrence prediction model generator 15 determines the degree of relationship between at least one of a plurality of state variables and genetic information based on [Equation 3] and disease risk of chronic kidney disease. Weights can be updated for errors that occur when creating a predicted disease outbreak model.

는 오차에 따른 과적합(overfitting)을 방지하기 위한 L2 정규식이다. E is a disease outbreak prediction model (mechanical learning model is the detection value of an error, t is whether chronic kidney disease occurs, y is a disease risk predicted through a machine learning model,

Is an L2 regular expression to prevent overfitting due to errors.

[Equation 3] is an error formula of the disease risk machine learning model generator 140, and the calculated error can be used to learn the weight of the artificial neural network through a backpropagation algorithm. In order to prevent overfitting due to noise generated during the learning process, the L2 purification formula has been added, and t may indicate the presence or absence of each actual chronic kidney disease, but is not limited thereto.

According to an embodiment of the present application, the Cox proportional risk model applied in the disease incidence prediction model generator 15 was measured once for genomic indicators, demographic, social, and family history indicators, and the underlying conditions for hypertension, diabetes, and chronic kidney disease were investigated. Since it contains only the morbidity of, it can also be measured once. The disease occurrence prediction model generation unit 15 can predict the risk of disease occurrence based on measures measured at the time of the base investigation, such as behavioral factors, obesity-related indicators, and blood abnormality indicators. The disease occurrence model generation unit 15 may use the Cox proportional risk model to predict the risk of disease occurrence. The disease incidence prediction model generator 15 analyzes using the Cox proportional probability risk model, the effect (beta) value and the risk of disease (HR) hazard on the occurrence of hypertension, diabetes, and chronic kidney disease for each variable. can confirm.

Referring to FIG. 11 as an example, the model is a model that allows group-specific variables to be entered step by step in order to take into account the biological temporal and temporal relationships. As shown in FIG. 11, several models are ensembled through bagging and boosting. By training, you can create an optimal model that can be generalized. 11 is a simplified diagram of model selection and ensemble training and optimization model setup in various cases.

According to one embodiment of the present application, the disease prediction unit 16 may predict the occurrence of chronic kidney disease in a subject by inputting new disease prediction data into the disease outbreak prediction model.

The disease prediction unit 16 may predict the occurrence of chronic kidney disease in the subject by applying the genetic information of the subject, a plurality of integrated risk factor indicator blood markers, and the like to the disease outbreak prediction model. For example, the disease prediction unit 16 may predict a subject's disease risk by applying the subject's subject state variable and subject's gene information to the disease risk machine learning model. In addition, the disease predicting unit 16 may predict the subject's subject's disease risk by applying the subject's subject state variables and subject's genetic information to the disease risk machine learning model and genetic information statistical probability model.

According to one embodiment of the present application, the disease prediction unit 16 may predict the subject's disease risk by applying the subject's subject state variable and subject's genetic information to the machine learning model and the statistical probability model. In addition, the disease predicting unit 16 may visualize a result of predicting a disease risk of a subject based on a predetermined classification item. For example, the disease prediction unit 16 may construct a deep learning-based visualization algorithm to provide visualized results for each subject based on the disease occurrence prediction model of the disease occurrence prediction model generator 15. The disease prediction unit 16 may predict and visualize a change in an individual's disease risk path based on a change pattern of a negative factor. In addition, the disease prediction unit 16 may visualize and provide a safety path through which an individual's probability of disease risk may be reduced based on a change pattern of a positive factor. In addition, the disease predicting unit 16 considers the changes in the negative and positive factors collectively, and based on the change in lifestyle of each subject, chronic kidney disease and cardiovascular disease, which is the final health condition, chronic heart disease, and Personalized preventive care service model can be provided through guidance on risk aversion to death

In addition, the disease prediction unit 16 may provide disease prevention management information associated with the predicted outcome of a subject's chronic kidney disease. The disease prediction unit 16 may provide disease prevention management information associated with a prediction result of a chronic kidney disease occurrence of a target user terminal (not shown). As an example, the disease prevention management information may include information related to hypertension, diabetes, hypertension-diabetes accompanying condition, diet associated with chronic kidney disease subjects, exercise methods, and the like.

According to one embodiment of the present application, the apparatus for predicting the occurrence of chronic kidney disease 10 may select factors associated with each disease morbidity in the basal population. First, among the 501 variables of the chronic kidney disease occurrence prediction apparatus 10, variables may be excluded based on the following criteria. At this time, variables that are not related to the prediction of diseases (hypertension, diabetes, chronic kidney disease), such as personal identification and the date of investigation, can be excluded. In addition, variables directly related to the final state of the disease (hypertension, diabetes, chronic kidney disease), such as whether the disease was diagnosed or not, can be excluded. Also, variables in which the ratio of missing values exceeds 20% of the total data can be excluded. Chronic kidney disease incidence prediction device 10 may derive a total of 120 variables for consideration through this process.

In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 may select variables by creating a full polynomial logistic regression model for 120 variables with the disease morbidity defined as a response variable through a pre-processing step. Chronic kidney disease occurrence prediction device (10) After creating a multinomial logistic regression model, select variables by performing variable selection method (forward / backward / stepwise selection), and then select each of the three variables Among them, at least two variables can be selected. In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 may further select risk factors having a significant correlation with the occurrence of hypertension/diabetes/chronic kidney disease based on clinical and epidemiological significance through the following method. . Specifically, after performing the analysis based on each of the three variable selection methods through backward, forward, and stepwise in the Cox proportional risk model, it is possible to select variables from the results of applying at least two variable selection methods. The variables used in the model for predicting cardiovascular and metabolic diseases established in previous studies can be selected as additional explanatory variables. Table 5 shows the model for predicting cardiovascular disease established in previous studies.

Demographic
And family history Lifestyle Disease history Blood pressure Blood test
(Lipid) Other blood tests gender smoking diabetes SBP Total cholesterol Blood sugar age Body mass index CVD DBP HDL-cholesterol C reative
protein race Physical activity AF High blood pressure LDL-cholesterol Albumin Family history of cardiovascular disease Dietary Glucose tolerance
intolerance) TG Creatinine Psychological factors angina pectoris Socioeconomic factors Etc Drinking

Finally, the variables included in this model include factors such as smoking, drinking, obesity, and regular exercise as factors that should be improved based on health promotion disease prevention policy in Korea. Selected. Based on the above process, the chronic kidney disease incidence prediction apparatus 10 has sex, age, education level, income level, cardiovascular disease history, and family history of cardiovascular disease in order to build an integrated model of hypertension, diabetes, and chronic kidney disease. Scores, environmental behavior-related factors (drinking, regular exercise, smoking status), obesity indicators, body mass index, waist circumference, blood test (total cholesterol, HDL-cholesterol, TG), liver function blood markers such as albumin, basal status Finally, 19 variables can be selected, including disease morbidity and disease genomic markers. According to one embodiment of the present application, the integrated risk factor building unit 13 may evaluate the effects of variables selected in the integrated risk factor panel on the occurrence of hypertension and diabetes. Illustratively, the integrated risk factor building unit 13 may evaluate the effect on the occurrence of hypertension. Table 6 is an assessment of the impact of the integrated risk factor panel variable on the development of hypertension.

When the beta value in each factor is greater than 0, the risk of disease occurrence increases according to the factor, whereas when it is less than 0, the risk of hypertension decreases.

Variable beta Standard error HR (95% CI) ¹ P-value ¹ Age, years <50 ref 50-60 0.582 0.040 1.79 (1.65-1.94) <.0001 60-70 0.927 0.039 2.53 (2.34-2.73) <.0001 gender man ref Woman -0.119 0.032 0.89 (0.83-0.95) 0.001 Education level Elementary School ref high school -0.501 0.035 0.61 (0.57-0.65) <.0001 University or higher -0.634 0.055 0.53 (0.48-0.59) <.0001 Income, 10,000 won/month <1 million won/month ref 1 million to 2 million won/month -0.418 0.040 0.66 (0.61-0.71) <.0001 2-4 million won/month -0.609 0.042 0.54 (0.50-0.59) <.0001 4 million won/month or more -0.588 0.067 0.56 (0.49-0.63) <.0001 BMI, kg/m ² <23 ref 23-27.5 0.382 0.040 1.47 (1.36-1.59) <.0001 27.5+ 0.778 0.048 2.18 (1.98-2.39) <.0001 Waist circumference, cm Men <90cm, Women <85cm ref 90+ male, 85+ female 0.645 0.033 1.91 (1.79-2.04) <.0001 Smoking Never ref Past 0.148 0.044 1.16 (1.06-1.27) <.0001 Current 0.039 0.039 1.04 (0.96-1.12) 0.314 Drinking Never ref Past 0.056 0.068 1.06 (0.93-1.21) 0.411 Current 0.018 0.033 1.02 (0.95-1.09) 0.584 regular exercise No ref Yes -0.031 0.032 0.97 (0.91-1.03) 0.331 High blood pressure (Blood pressure) ¹ normal ref Prehypertension 1.629 0.044 5.10 (4.68-5.56) <.0001 High blood pressure 2.741 0.057 15.50 (13.86-17.333) <.0001 Diabetes Status (Fasting Glucose and HBA1C) ² normal ref Pre-diabetes 0.415 0.070 1.51 (1.32-1.74) <.0001 diabetes 0.528 0.047 1.70 (1.55-1.86) <.0001 Chronic kidney disease (creatinine) ³ normal ref CKD 0.694 0.087 2.00 (1.69-2.38) <.0001 Total cholesterol, mg/dL <200 ref 200-240 0.115 0.036 1.12 (1.05-1.20) 0.002 240+ 0.324 0.053 1.38 (1.25-1.54) <.0001 HDL-C, mg/dL 40+ men, 50+ women ref Men <40 Women <50 0.151 0.033 1.16 (1.09-1.24) <.0001 TG, mg/dL <150 ref 150+ 0.486 0.033 1.63 (1.53-1.73) <.0001 ALBUMIN, g/dL 3.4-5.4 ref <3.4 0.839 0.353 2.32 (1.16-4.63) 0.018 Cardiovascular disease history ⁴ 0 ref 1+ 0.650 0.098 1.92 (1.58-2.32) <.0001 Cardiovascular disease family history score ⁵ 0 ref 1+ 0.060 0.044 1.06 (0.97-1.16) 0.171 Hypertensive genome score Continuous scale Categorical scale Quantile1 ref Q2 0.070 0.054 1.07 (0.96-1.19) 0.201 Q3 0.095 0.054 1.10 (0.99-1.22) 0.078 Q4 0.223 0.052 1.25 (1.13-1.39) <.0001 Q5 0.299 0.051 1.35 (1.22-1.49) <.0001

In hypertension, SBP is 120mmHg or less, DBP is 80mmHg or less, SBP is 130mmHg or more, DBP is 80mmHg or more, or SBP is less than 140mmHg and DBP is less than 90mmHg. The abnormal case can be defined as hypertension. Diabetes status is normal when fasting blood glucose is less than 100mg/dL and HBA1C is less than 6.5%, and if 100≤ fasting blood sugar <126mg/dL and HBA1C is less than 6.5%, pre-diabetes, when fasting blood sugar is 126mmHg or more or HBA1C 6.5 or more Can be defined as diabetes.

Chronic kidney disease state can be defined as chronic nephropathy disease with an estimated glomerular filtration rate (eGRFR) value of less than 60 based on serum creatine concentration.

The history of cardiovascular disease scores is 1 for cases with a history of disease, 1 for cases with no history, and 0 for cases with a history of congestive heart failure, diagnosis of coronary artery disease, and diagnosis of cerebrovascular diseases such as stroke and stroke. It can be calculated by classifying a case where all of these are absent as 0 and a case when there is more than one case.

The family history of cardiovascular disease scores is 3 for family history of hypertension, family history of diabetes, family history of heart disease, 1 for family history, 0 for none, and 0 for all family history of disease. It can be calculated by classifying one case of more than one.

It can be seen that the risk of hypertension in men is higher as the age increases, compared to women under the age of 50, compared to women. At the education level, the rate of hypertension decreased as the period of education was longer, based on'below elementary school'. In addition, it can be seen that in general, the higher the income level, the lower the risk of occurrence. This can be expected to be the effect from the degree of attention to health status, the type of work involved, and the difference in knowledge required to maintain health, depending on the socio-economic level. It can be seen that the risk of hypertension increases with increasing body mass index, which is an index related to obesity, and waist circumference, which is an index of obesity. It can be confirmed that the risk of hypertension is not significant for drinking and regular exercise. However, in the case of smoking, it can be seen that the risk of hypertension in the past smokers increases significantly. In the case of blood markers indicating blood abnormalities, lipid abnormality indicators TG, HDL-cholesterol, total cholesterol and liver abnormality indicator Albumin were categorized based on clinical reference values. It can be seen that this statistically significant increase was observed. It can be seen that the cardiovascular disease history score indicating the current disease morbidity and the risk of hypertension increased as the morbidity of hypertension, diabetes, and chronic kidney disease increased. If you have a history of cardiovascular disease, you can see that the risk of hypertension increases but is not significant. It can be seen that the higher the genomic score made of the genome associated with hypertension, the higher the risk of hypertension.

According to one embodiment of the present application, the integrated risk factor building unit 13 may evaluate the effects of variables selected in the integrated risk factor panel on the occurrence of hypertension and diabetes. Illustratively, the integrated risk factor building unit 13 may evaluate the effect on diabetes incidence. Table 7 assesses the impact of the integrated risk factor panel variable on diabetes incidence.

When the beta value of each factor is greater than 0, the risk of disease occurrence increases according to the factor, whereas when it is less than 0, the risk of diabetes decreases.

Variable beta Standard error HR (95% CI) ¹ P-value ¹ Age, years <50 ref 50-60 0.539 0.065 1.71 (1.51-1.95) <.0001 60-70 0.639 0.065 1.89 (1.67-2.15) <.0001 gender man ref Woman -0.160 0.054 0.85 (0.77-0.95) 0.003 Education level Elementary School ref high school -0.341 0.058 0.71 (0.64-0.80) <.0001 University or higher -0.323 0.087 0.72 (0.61-0.86) <.0001 Income, 10,000 won/month <1 million won/month ref 1 million to 2 million won/month -0.235 0.066 0.79 (0.69-0.90) <.0001 2-4 million won/month -0.353 0.068 0.70 (0.62-0.80) <.0001 4 million won/month or more -0.246 0.105 0.78 (0.64-0.96) 0.019 BMI, kg/m ² <23 ref 23-27.5 0.505 0.072 1.66 (1.44-1.91) <.0001 27.5+ 1.151 0.079 3.16 (2.71-3.69) <.0001 Waist circumference, cm Men <90cm, Women <85cm ref 90+ male, 85+ female 0.880 0.054 2.41 (2.17-2.68) <.0001 Smoking Never ref Past 0.140 0.075 1.15 (0.99-1.33) 0.061 Current 0.244 0.062 1.28 (1.13-1.44) <.0001 Drinking Never ref Past 0.317 0.102 1.37 (1.13-1.68) 0.002 Current -0.023 0.056 0.98 (0.88-1.09) 0.675 regular exercise No ref Yes -0.077 0.053 0.93 (0.83-1.03) 0.148 High blood pressure (Blood pressure) ¹ normal ref Prehypertension 0.600 0.060 1.82 (1.62-2.05) <.0001 High blood pressure 0.883 0.078 2.42 (2.08-2.82) <.0001 Diabetes Status (Fasting Glucose and HBA1C) ² normal ref Pre-diabetes 1.927 0.093 6.87 (5.73-8.24) <.0001 Chronic kidney disease (creatinine) ³ normal ref CKD 0.203 0.157 1.23 (0.90-1.66) 0.195 Total cholesterol, mg/dL <200 ref 200-240 0.328 0.057 1.39 (1.24-1.55) <.0001 240+ 0.694 0.077 2.00 (1.72-2.33) <.0001 HDL-C, mg/dL 40+ men, 50+ women ref Men <40 Women <50 0.363 0.053 1.44 (1.30-1.60) <.0001 TG, mg/dL <150 ref 150+ 0.994 0.054 2.70 (2.43-3.00) <.0001 ALBUMIN, g/dL 3.4-5.4 ref <3.4 1.473 0.448 4.36 (1.81-10.50) 0.001 Cardiovascular disease history ⁴ 0 ref 1+ 0.415 0.162 1.51 (1.10-2.08) 0.011 Cardiovascular disease family history score ⁵ 0 ref 1+ -0.022 0.072 0.98 (0.85-1.23) 0.755 Diabetes genome score Continuous scale Categorical scale Quantile1 ref Q2 0.095 0.088 1.10 (0.93-1.31) 0.283 Q3 0.137 0.087 1.15 (0.97-1.36) 0.116 Q4 0.315 0.084 1.37 (1.16-1.62) <.0001 Q5 0.491 0.082 1.63 (1.39-1.92) <.0001

According to an embodiment of the present application, it can be confirmed that the risk of diabetes increases as the body mass index, which is an index related to obesity, and the waist circumference, which is an index of abdominal obesity, increase. In addition, it can be seen that the risk of diabetes increases significantly if you are a current smoker or have been drinking in the past. However, it can be confirmed that the risk of developing diabetes is not significant in the case of regular exercise. In the case of blood markers indicating blood abnormalities, lipid abnormality indicators TG, HDL-cholesterol, total cholesterol and liver abnormality indicators Alubmin were categorized based on clinical reference values. It can be seen that the occurrence has a statistically significant increase. In addition, it can be seen that the risk of diabetes increases as the prevalence of hypertension, pre-diabetes, and chronic kidney disease, which indicates the current disease prevalence. In addition, it was confirmed that the risk of developing diabetes increased significantly if there was a history of cardiovascular disease. However, it can be confirmed that the risk of developing diabetes was not significant in the case of family history of cardiovascular diseases. The higher the genomic score of diabetes-related genome, the higher the risk of diabetes. According to an embodiment of the present disclosure, the integrated risk factor building unit 13 chronic kidney disease of variables selected in the integrated risk factor panel. You can evaluate the impact on. Illustratively, the integrated risk factor building unit 13 may evaluate the effect on the occurrence of chronic kidney disease. Table 8 is an assessment of the impact of the integrated risk factor panel variable on the occurrence of chronic kidney disease.

When the beta value in each factor is greater than 0, the risk of disease occurrence increases according to the factor, whereas when it is less than 0, the risk of developing chronic kidney disease decreases.

Variable beta Standard error HR (95% CI) ¹ P-value ¹ Age, years <50 ref 50-60 0.692 0.072 2.00 (1.74-2.30) <.0001 60-70 1.545 0.064 4.69 (4.14-5.31) <.0001 gender man ref Woman 0.494 0.054 1.64 (1.47-1.82) <.0001 Education level Elementary School ref high school -0.637 0.055 0.53 (0.48-0.59) <.0001 University or higher -0.694 0.089 0.50 (0.42-0.59) <.0001 Income, 10,000 won/month <1 million won/month ref 1 million to 2 million won/month -0.519 0.064 0.60 (0.53-0.68) <.0001 2-4 million won/month -0.580 0.066 0.56 (0.49-0.64) <.0001 4 million won/month or more -0.753 0.116 0.47 (0.38-0.59) <.0001 BMI, kg/m ² <23 ref 23-27.5 0.171 0.062 1.19 (1.05-1.34) 0.006 27.5+ 0.489 0.076 1.63 (1.41-1.89) <.0001 Waist circumference, cm Men <90cm, Women <85cm ref 90+ male, 85+ female 0.587 0.052 1.80 (1.62-1.99) <.0001 Smoking Never ref Past -0.286 0.077 0.75 (0.65-0.87) <.0001 Current -0.416 0.068 0.66 (0.58-0.75) <.0001 Drinking Never ref Past -0.124 0.105 0.88 (0.72-1.09) 0.239 Current -0.522 0.055 0.59 (0.53-0.66) <.0001 regular exercise No ref Yes -0.094 0.052 0.91 (0.82-1.01) 0.068 High blood pressure (Blood pressure) ¹ normal ref Prehypertension 0.424 0.059 1.53 (1.36-1.71) <.0001 High blood pressure 0.800 0.077 2.23 (1.92-2.59) <.0001 Diabetes Status (Fasting Glucose and HBA1C) ² normal ref Pre-diabetes -0.060 0.136 0.94 (0.72-1.23) 0.660 diabetes 0.807 0.067 2.24 (1.97-2.56) <.0001 Total cholesterol, mg/dL <200 ref 200-240 0.314 0.056 1.37 (1.23-1.53) <.0001 240+ 0.478 0.083 1.61 (1.37-1.90) <.0001 HDL-C, mg/dL 40+ men, 50+ women ref Men <40 Women <50 0.430 0.054 1.54 (1.38-1.71) <.0001 TG, mg/dL <150 ref 150+ 0.447 0.052 1.56 (1.41-1.73) <.0001 ALBUMIN, g/dL 3.4-5.4 ref <3.4 2.457 0.501 11.67 (4.37-31.16) <.0001 Cardiovascular disease history ⁴ 0 ref 1+ 0.747 0.150 2.11 (1.57-2.83) <.0001 Cardiovascular disease family history score ⁵ 0 ref 1+ 0.055 0.070 1.06 (0.92-1.21) 0.439 Hypertensive genome score Continuous scale Categorical scale Quantile1 ref Q2 -0.015 0.086 0.99 (0.83-1.17) 0.865 Q3 0.102 0.084 1.11 (0.94-1.31) 0.223 Q4 0.118 0.082 1.13 (0.96-1.32) 0.154 Q5 0.107 0.082 1.12 (0.95-1.31) 0.192 Diabetes genome score Continuous scale Categorical scale Quantile1 ref Q2 -0.074 0.081 0.93 (0.79-1.09) 0.359 Q3 -0.011 0.079 0.99 (0.85-1.16) 0.890 Q4 -0.141 0.082 0.87 (0.74-1.02) 0.086 Q5 -0.082 0.081 0.92 (0.79-1.08) 0.316

It can be seen that the risk of developing chronic kidney disease increases with increasing body mass index, which is an index related to obesity, and waist circumference, which is an indicator of obesity. In addition, it can be confirmed that in the case of regular exercise, it is not significant for the occurrence of chronic kidney disease. In addition, in the case of blood markers indicating blood abnormalities, categories were classified based on lipid abnormality indicators TG, HDL-cholesterol, total cholesterol, and clinical reference values. Significantly increasing form. It can be seen that the risk of chronic kidney disease increases as the hypertension and diabetes mellitus indicate the current disease-related condition. As the family history of cardiovascular disease increases, the risk of developing chronic kidney disease increases. According to one embodiment of the present application, the disease occurrence prediction model generator 15 may compare the predictive power of the model when the disease occurrence prediction model is constructed using a statistical method and a plurality of machine learning methods. The disease incidence prediction model generator 15 compares the disease incidence prediction model using various AI learning methods, and finally determines the prediction model to which the method of predicting the highest risk of developing chronic kidney disease, hypertension, diabetes and their accompanying diseases is applied. It can be selected as a disease outbreak prediction model.

For example, ROC-curve predicts the model's predictive power when each disease prediction model is constructed using a statistical method (Cox model), three machine learning methods (artificial neural network, deep learning-based recurrent neural network (RNN), and random forest). The results were drawn and compared, and the calculated AUC values are shown in Tables 9 to 11.

Table 9 shows the results of predicting the risk of hypertension.

High blood pressure Ansan Anseong Urban infrastructure Rural base City + rural integration Statistical technique 0.855 0.605 0.654 0.616 Machine learning method Surf Vector Machine (SVM) 0.982 0.547 0.73 0.562 RNN 0.853 0.534 0.552 0.508 Random forest 0.988 0.683 0.623 0.664

Table 10 is a result of predicting the risk of diabetes.

diabetes Ansan Anseong Urban infrastructure Rural base City + rural integration Statistical technique 0.881 0.791 0.804 0.816 Machine learning method Surf Vector Machine (SVM) 0.993 0.692 0.704 0.711 RNN 0.853 0.624 0.640 0.605 Random forest 0.998 0.880 0.812 0.869

Table 11 shows the results of predicting the risk of developing chronic kidney disease.

Chronic kidney disease Ansan Anseong Urban infrastructure Rural base City + rural integration Statistical technique 0.748 0.742 0.860 0.804 Machine learning method Surf Vector Machine (SVM) 0.996 0.512 0.668 0.589 RNN 0.735 0.641 0.666 0.594 Random forest 0.998 0.803 0.851 0.824

13 is a diagram showing a comparison of the predicted risk of developing hypertension according to a statistical and machine learning model according to an embodiment of the present application, and FIG. 14 is a predicted risk of developing diabetes according to a statistical and machine learning model according to an embodiment of the present application. 15 is a diagram showing a comparison of the predicted risk of developing chronic kidney disease according to a statistical and machine learning model according to an embodiment of the present disclosure. As a result of comparison, a random forest method with the highest risk of developing chronic kidney disease, which is hypertension, diabetes and their accompanying diseases, can be selected as the final model.

According to one embodiment of the present application, the apparatus for predicting the development of chronic kidney disease 10 includes genomic markers for hypertension and diabetes, which are Korean phobias, and chronic diseases caused by these diseases and their accompanying morbidity and these diseases In predicting the risk of developing kidney disease, precise prediction is possible unlike the existing model. In particular, this genome marker is a marker selected using extended genetic information based on the latest version of 1K genome 3.0, unlike the previous studies.

In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 currently has a disease state (normal, total disease or disease morbidity) in the general population because the model includes a morbidity and non-morbidity that may have a higher risk of disease in the general population. Since the abnormal condition is considered, a disease prevention management plan can be proposed and applied to existing patients or non-abnormal patients.

In addition, in other existing models, a disease prediction model (model) was constructed by targeting the occurrence or morbidity of one disease, while the disease occurrence prediction model (model) constructed in the apparatus for predicting the occurrence of chronic kidney disease (model) is a biological disease. Unlike other models, it has specificity in that it has built an integrated model that can simultaneously identify four disease states in consideration of pathological mechanisms and natural history of disease.

In addition, unlike the existing model, the model for predicting disease outbreak built in the apparatus for predicting the occurrence of chronic kidney disease 10 was finally selected through training using several models, and in the current final model (depending on the disease Although different), the predictive power is at least 80% (-up to 91%), so it has a very high predictive power.

When assigning the parameters of the final selected model, the model was sequentially formed under biological algorithms taking into account both the time of factor exposure and the temporal continuity of the state that can be changed later and the continuity of the disease and the natural history of the disease. It can be seen as a predictable model of the state of biological change over time.

According to another embodiment of the present application, the apparatus for predicting the occurrence of chronic kidney disease 10 inputs genetic information of a patient with a chronic kidney disease and a disease risk of a chronic kidney disease, and thus, between the genetic information and the disease risk of a chronic kidney disease Genetic machine learning models that learn the degree of relationships can be created. In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 may select key genetic information from genetic information using a genetic information machine learning model. In addition, at least one of the plurality of state variables and the core genetic information, by inputting a plurality of state variables, the core genetic information, and the disease risk of the chronic kidney disease, including the living state variable and the health state variable of the chronic kidney disease patient A disease learning machine learning model can be created to learn the degree of the relationship between and the disease risk of chronic kidney disease. In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 may receive a subject's subject state variable and subject's gene information. In addition, the apparatus for predicting the occurrence of chronic kidney disease may predict the subject's disease risk by applying the subject's subject state variable and subject's gene information to the disease risk machine learning model.

In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 inputs the genetic information of the patient of the chronic kidney disease and the disease risk of the chronic kidney disease, and the probability of the disease of the chronic kidney disease according to the presence or absence of each of the genetic information Genetic statistical model of gene information can be generated. The apparatus for predicting chronic kidney disease occurrence may select key genetic information from genetic information using a statistical probability model of genetic information and a machine learning model of genetic information.

In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 inputs a plurality of status variables, genetic information, and disease risk of chronic kidney disease patients as inputs, and the presence or absence of at least one of the plurality of status variables and genetic information. Alternatively, a statistical probability model that probably represents the disease risk of chronic kidney disease may be generated according to the value. In addition, the subject's disease risk can be predicted by applying the subject's subject state variables and subject's gene information to the disease risk machine learning model and genetic information statistical probability model.

In addition, the apparatus for predicting the occurrence of chronic kidney disease 10 inputs a plurality of status variables, genetic information, and disease risk of chronic kidney disease in a patient with a chronic kidney disease, and at least one of the plurality of status variables associated with the chronic kidney disease Selecting the above state variable and generating a basic statistical probability model that probabilistically indicates the disease risk of chronic kidney disease for the presence or value of at least one state variable. In addition, a statistical probability model can be generated from the basic statistical probability model by applying a weight to the disease risk of the chronic kidney disease according to the presence or absence of genetic information associated with the chronic kidney disease.

In addition, the genetic information machine learning model of the apparatus for predicting the development of chronic kidney disease 10, when the first state variable among the plurality of state variables as the input layer and the second state variable among the plurality of state variables as the hidden layer, the input layer and By performing the first learning to learn the degree of the relationship between the hidden layer, and when the hidden layer and the genetic information as the input layer and the disease risk as the output layer, by performing the second learning to learn the degree of the relationship between the hidden layer and the output layer, plural It can learn the degree of the relationship between at least one or more of the state variables and genetic information and the disease risk of chronic kidney disease.

In addition, the genetic information machine learning model of the apparatus for predicting the occurrence of chronic kidney disease when the state variables of the previous time point of the plurality of state variables are input layers and the current time point state variables of the plurality of state variables are hidden layers, When performing the first learning to learn the degree of the relationship between the input layer and the hidden layer, and using the hidden layer and the genetic information as the input layer and the disease risk as the output layer, learning the degree of the relationship between the hidden layer and the output layer By performing the second learning, it is possible to learn the degree of relationship between at least one of the plurality of state variables and genetic information and the disease risk of the chronic kidney disease.

Hereinafter, based on the details described above, the operation flow of the present application will be briefly described.

16 is a flowchart illustrating a method for predicting the development of chronic kidney disease according to an embodiment of the present application.

The method for predicting the occurrence of chronic kidney disease may be performed by the apparatus for predicting the occurrence of chronic kidney disease 10 described above. Therefore, even if omitted, the description of the apparatus for predicting the development of chronic kidney disease 10 may be applied to the description of the method for predicting the development of chronic kidney disease.

In step S161, the chronic kidney disease occurrence prediction apparatus 10 may select a genomic marker associated with a plurality of diseases.

In step S162, the chronic kidney disease occurrence prediction apparatus 10 may calculate an integrated genomic index using genomic markers associated with a plurality of selected diseases.

In step S163, the chronic kidney disease occurrence prediction apparatus 10 may derive factors related to a plurality of diseases to construct an integrated risk factor model.

In step S164, the chronic kidney disease occurrence prediction apparatus 10 may build a disease outbreak prediction model by inputting an integrated genomic indicator and an integrated risk factor model.

In step S165, the chronic kidney disease occurrence prediction apparatus 10 may predict the occurrence of chronic kidney disease in a subject by inputting new disease prediction data into the disease occurrence prediction model.

In the above description, steps S161 to S165 may be further divided into additional steps or combined into fewer steps, according to an embodiment of the present application. In addition, some steps may be omitted if necessary, and the order between the steps may be changed.

The method for predicting the development of chronic kidney disease according to an embodiment of the present application may be implemented in a form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention or may be known and usable by those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs, DVDs, and magnetic media such as floptical disks. -Hardware devices specifically configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language code that can be executed by a computer using an interpreter, etc., as well as machine language codes made by a compiler. The hardware device described above may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

In addition, the above-described method for predicting the occurrence of chronic kidney disease may also be implemented in the form of a computer program or application executed by a computer stored in a recording medium.

The above description of the present application is for illustrative purposes, and a person having ordinary knowledge in the technical field to which the present application belongs will understand that it is possible to easily change to other specific forms without changing the technical spirit or essential characteristics of the present application. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope of the present application is indicated by the claims, which will be described later, rather than the detailed description, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts should be interpreted to be included in the scope of the present application.

10: Chronic kidney disease outbreak prediction device
11: genome marker selection section
12: Integrated dielectric index calculation unit
13: Integrated risk factor building department
14: Derivation of explanatory variables
15: disease outbreak prediction model generator
16: disease prediction department
20: disease prediction server

Claims (11)

In the apparatus for predicting the development of chronic kidney disease,
A genomic marker selection unit for selecting genomic markers associated with a plurality of diseases;
An integrated genomic index calculation unit for calculating an integrated genomic index using genomic markers related to a plurality of diseases selected by the genomic marker selection unit;
An integrated risk factor construction unit for deriving factors related to the plurality of diseases and constructing an integrated risk factor model;
A disease occurrence prediction model generator configured to build a disease occurrence prediction model using the integrated genomic indicator and the integrated risk factor model as inputs; And
A disease prediction unit that predicts the occurrence of chronic kidney disease in a subject by inputting new disease prediction data into the disease occurrence prediction model
Chronic kidney disease outbreak prediction apparatus comprising a. According to claim 1,
The dielectric marker selection unit,
The first genome marker associated with the first disease is selected, and the selected first genome marker is verified in each of the urban cohort and the rural cohort to determine the first genome marker, and the second genome marker associated with the second disease is selected. , Chronic kidney disease incidence prediction device to determine the selected second dielectric marker in the urban cohort and rural cohort, respectively, to determine the second dielectric marker. According to claim 2,
The dielectric marker selection unit,
Based on a regression analysis algorithm, a single base polymorphism (SNP) marker associated with a plurality of diseases is selected, and the core is considered by considering at least one of the single base polymorphism (SNP) marker, the first genome marker, and the second genome marker Genetic information, chronic kidney disease outbreak prediction device. According to claim 1,
The integrated risk factor building unit,
The plurality of diseases in consideration of at least one of basic demographic factors, sociological factors, disease and family history factors, environmental behavior-related factors, obesity-related indicator factors, hematologic abnormality indicator factors, and underlying disease morbidity factors A device for predicting the development of chronic kidney disease that derives factors related to and builds an integrated risk factor model. According to claim 1,
And a descriptive variable deriving unit for deriving data of a subject having at least one of a plurality of diseases as an explanatory variable. According to claim 3,
The disease occurrence prediction model generation unit,
At least one of the plurality of state variables and the core gene information, by inputting a plurality of state variables including the living state variable and the health state variable of the patient with chronic kidney disease, the core gene information and the disease risk of the chronic kidney disease And a disease outbreak prediction model for learning the degree of relationship between the disease risk of the chronic kidney disease. The method of claim 6,
The disease outbreak prediction model,
A degree of relationship between the input layer and the hidden layer when a first state variable and a previous time hidden layer among the plurality of state variables are used as an input layer and a second state variable or a current time state variable among the plurality of state variables is hidden. Do the first learning to learn,
When the hidden layer and the genetic information are used as input layers and the disease risk is set as the output layer, by performing a second learning to learn the degree of the relationship between the hidden layer and the output layer, at least one or more of the plurality of state variables and genetic information And the degree of relationship between the disease risk of the chronic kidney disease,
The first learning is to learn the degree of the relationship between the input layer and the hidden layer based on [Equation 1],
[Equation 1]

At this time, above

Is a hidden layer at time t, and

Is a first weight value indicating the degree of relationship of the first type between the input layer and the hidden layer, wherein

Is the hidden layer from the previous point,

Is a second weight value indicating the degree of relationship of the second type between the input layer and the hidden layer,

Is a first state variable at time t, a device for predicting the development of chronic kidney disease. The method of claim 7,
The second learning is to learn the degree of the relationship between the hidden layer and the output layer based on [Equation 1] and [Equation 2],
[Equation 2]

At this time, y is an output layer, and

Is a third weight value indicating the degree of relationship between the hidden layer and the output layer,

Is a hidden layer at time t, and

Is a fourth weighting value indicating the degree of relationship between the gene information and the output layer in the input layer, z is the genetic information in the input layer, chronic kidney disease outbreak prediction device. The method of claim 8,
The disease occurrence prediction model generation unit,
Based on [Equation 3], the weight is updated to an error that occurs when generating a machine learning model that learns the degree of relationship between at least one of the plurality of state variables and genetic information and the disease risk of the chronic kidney disease. Will be
[Equation 3]

The E is the detection value of the error of the disease occurrence prediction model generation unit, the t is whether the chronic kidney disease occurs, the y is the disease risk predicted through the machine learning model,

Is an L2 regular expression for preventing overfitting due to errors, and a device for predicting the development of chronic kidney disease. According to claim 1,
The disease prediction unit,
The apparatus for predicting chronic kidney disease outbreak is to provide disease prevention management information associated with the prediction result of chronic kidney disease outbreak in the subject. In the method of predicting the development of chronic kidney disease,
Selecting genomic markers associated with a plurality of diseases;
Calculating an integrated genomic index using genomic markers associated with the selected plurality of diseases;
Constructing an integrated risk factor model by deriving factors related to the plurality of diseases;
Constructing a disease outbreak prediction model using the integrated genomic index and the integrated risk factor model as inputs; And
Predicting the occurrence of chronic kidney disease in a subject by inputting new disease prediction data into the disease occurrence prediction model;
Method for predicting the occurrence of chronic kidney disease, including.

Images