KR102442873B1 - A disease prediction system for planning insurance - Google Patents

Abstract

Basic health-related information such as body information, past history, family history, and lifestyle are collected through the questionnaire, past health checkup information and demographic variables are collected, and missing data among these collected data is corrected, It relates to a disease prediction service system for insurance design that predicts a predicted disease and the probability of the occurrence of the expected disease, and a sample data collection unit that collects demographic information, health checkup data, and onset data of past patients as sample data ; a questionnaire input unit that receives the customer's demographic information as a questionnaire; a health checkup collection unit that collects customer health checkup data; a missing value correction unit that extracts a similar group similar to the customer from the sample data, and replaces the missing data among the customer's health checkup data with a representative value of the extracted similar group to correct; And, by providing a configuration including a disease prediction unit for predicting a customer's disease using the customer's demographic information and the corrected health checkup data, and correcting the customer's missing data using the sample data of the past patients, the customer Even if a part of the health checkup result is omitted, the probability of occurrence of the expected disease of the customer can be predicted.

Description

{ A disease prediction system for planning insurance }

The present invention collects customer's health-related basic information such as body information, past history, family history, and lifestyle through a questionnaire, collects past health checkup information and demographic variables, and corrects missing data among these collected data, It relates to a disease prediction service system for insurance design that predicts the expected disease of the customer and the probability of the occurrence of the expected disease.

In general, insurance is a system in which a certain amount of money is paid in advance in case of an accident or illness that may occur in the future, and a certain amount of money is paid when a contracted condition is met. A typical example is health insurance. As the standard of living increases and medical technology advances, the life expectancy of a person is also increasing, thereby increasing the burden of medical expenses. Therefore, in order to reduce the burden of medical expenses due to the increase in expected life expectancy, insurance companies such as cancer insurance and medical expenses insurance are purchased to prepare for diseases. Insurance companies pay more as the customer's illness occurs.

On the other hand, the type of disease and the probability of occurrence vary depending on various factors such as health status, constitution, and lifestyle for each individual. Nevertheless, since health insurance is simply purchased according to gender, age, and occupation, high premiums may be paid for diseases with a relatively low probability of contracting, or cases may occur where it is impossible to prepare for essential diseases.

Conversely, from an insurance company's point of view, if more diseases occur than expected at the time of subscription, damages such as mortality loss may occur. Accordingly, various attempts are being made in the insurance industry to maximize the customer's needs and at the same time reduce the possibility of occurrence of quasi-loss (or quasi-loss risk).

For this purpose, after receiving basic medical treatment at various medical facilities, a technique for enrolling in insurance for expected related diseases according to the medical results of these medical facilities and the current status of the insured has been proposed [Patent Documents 1 and 2]. The prior art is intended to provide accurate insurance premium calculation and accuracy by the initial treatment stage to both the insured and the insurer.

However, although the prior art assumes that the health information of the insured is completely collected, there is a problem in that the health data of the insured cannot be perfectly obtained in reality.

There is a problem in that insurance products are selected by considering only the current health condition of the insured, but medical expenses incurred according to the condition of the insured are not taken into account.

For example, the results of 4 blood lipid levels (total cholesterol, high-density cholesterol, low-density cholesterol, and triglycerides) are intermittently performed instead of being tested every year after a specific point in time. Therefore, even if the customer's health checkup data is brought, there are cases where there is no result value for some of the above blood lipid level items among customers.

In addition, some items are omitted from the examination center during actual health examinations, so complete examination results do not always remain. Also, even when the customer directly inputs the results of the health checkup, the customer may not be able to input some result values.

Therefore, even if some of the customer's health-related information is missing, a technique for correcting the missing value of the corresponding item is very necessary.

Korean Patent Publication No. 10-2003-0023667 (published on March 19, 2003) Korean Patent Publication No. 10-2015-0049993 (published on May 8, 2015)

An object of the present invention is to solve the above problems, and collect basic health-related information of customers such as body information, past history, family history, and lifestyle through a questionnaire, and collect past health checkup information and demographic variables and to provide a disease prediction service system for insurance design that corrects missing data among the collected data to predict the expected disease of the customer and the probability of occurrence of the expected disease.

In addition, an object of the present invention is to collect sample data based on demographic variables for health checkup information and onset data of patients, and apply the sample data based on demographic factors of the customer, so that the health of the customer It is to provide a disease prediction service system for insurance design that corrects missing data of examination data.

In order to achieve the above object, the present invention relates to a disease prediction service system for insurance design, comprising: a sample data collection unit that collects demographic information, health checkup data, and onset data of past patients as sample data; a questionnaire input unit that receives the customer's demographic information as a questionnaire; a health checkup collection unit that collects customer health checkup data; a missing value correction unit that extracts a similar group similar to the customer from the sample data, and replaces the missing data among the customer's health checkup data with a representative value of the extracted similar group to correct; and a disease prediction unit for predicting the customer's disease by using the customer's demographic information and the corrected health checkup data.

In addition, the present invention provides a disease prediction service system for insurance design, wherein the system further comprises a similar variable selector for selecting a similar group extraction variable, and the missing value correction unit is the same as the data value of the customer's similar group extraction variable. It is characterized in that data having a value is extracted as a similar group.

Further, in the present invention, in a disease prediction service system for insurance design, the similar variable selector configures a plurality of combinations by combining predetermined candidate variables, and with respect to a predetermined test set, the customer's examination for each combination Estimate the data, obtain the difference value between the estimated examination data and the actual data of the test set, set the statistical value of the difference value of all test sets for each combination as an error, and select combinations having an error within a predetermined error range. It is characterized in that the variables of the combination having the smallest number of variables among the extracted combinations are selected as the similar group extraction variables.

In addition, the present invention is characterized in that in the disease prediction service system for insurance design, the similar group extraction variable consists of gender, age, and BMI.

In addition, in the disease prediction service system for insurance design, the present invention is characterized in that the missing value correcting unit forms a similar group to have a predetermined number, but randomly extracts it from sample data having the same value.

In addition, the present invention provides a disease prediction service system for insurance design, wherein the missing value correcting unit obtains a patient group and a control group, respectively, to form a similar group, and configures the similar group by combining the same number of patient groups and control groups, and the control group is a sample It is characterized in that the data is extracted from the data of patients who have not been diagnosed with the disease, and the patient group is extracted from the data of the patients diagnosed with the disease from the sample data.

Further, in the present invention, in the disease prediction service system for insurance design, the representative value of the similar group is set as the 'mode' of the corresponding values in the case of categorical data, and the 'median' of the corresponding values in the case of continuous data. characterized by being

In addition, in the present invention, in the disease prediction service system for insurance design, the demographic information includes gender, age, residential area, insurance subscription type, income decile, disability presence, height, and weight, and the examination data is , waist circumference, systolic blood pressure, diastolic blood pressure, fasting blood sugar, total cholesterol, high-density cholesterol, low-density cholesterol, triglycide, hemoglobin, urine protein, serum creatinine, serum GioT, serum GPT, gamma GTP do it with

As described above, according to the disease prediction service system for insurance design according to the present invention, in addition to personal physical information, family history, and habits, the possibility of onset is predicted by considering demographic variables such as residential area, disability, income, and insurance subscription. , an effect that allows the user to more accurately predict a disease that is likely to develop in the future is obtained.

In addition, according to the disease prediction service system for insurance design according to the present invention, by correcting the customer's missing data using the sample data of past patients, even if a part of the customer's health checkup result is omitted, the expected disease of the customer is corrected. An effect that can predict the probability of an outbreak is obtained.

1 is a block diagram of an entire system for implementing the present invention.
2 is a block diagram of the configuration of a disease prediction service system for insurance design according to an embodiment of the present invention.
3 is a table showing input variables of a disease prediction model according to an embodiment of the present invention.

Hereinafter, specific contents for carrying out the present invention will be described with reference to the drawings.

In addition, in demonstrating this invention, the same part is attached|subjected by the same code|symbol, and the repetition description is abbreviate|omitted.

First, the configuration of the entire system for implementing the present invention will be described with reference to FIG. 1 .

As shown in Figure 1, the entire system for implementing the present invention is a smart mobile terminal 10 such as a smart phone of an insurance planner, an insurance design client 20 installed in the smart mobile terminal 10 to support insurance design, It consists of an insurance design server 30 that provides an insurance design service, and a prediction server 50 that predicts a probability of occurrence of a predicted disease. The smart mobile terminal 10 and the insurance design server 30 are connected to each other through a network 80 such as the Internet. In addition, the database 40 for storing health checkup results, statistical information of demographic-based health checkup results, disease prediction information, and the like may be additionally configured.

First, the smart mobile terminal 10 is a mobile terminal used by an insurance planner, and is a smart terminal equipped with a normal computing function, such as a smartphone, a phablet, and a tablet PC. In particular, the smart mobile terminal 10 is a terminal in which an application or a mobile application (or an app, an application) can be installed and executed.

Next, the insurance design client 20 is a mobile application (or app, application) that is installed and executed on the smart mobile terminal 10, and is a program system that provides an insurance design service by interworking with the insurance design server 30 .

That is, the insurance design client 20 provides services related to insurance design support, such as collecting the customer's health status, etc. through a questionnaire, or predicting and providing the customer's possible disease or invention probability.

Next, the insurance design server 30 is a normal application server, which collects the customer's past examination data, predicts the customer's disease, and provides an insurance design according to the prediction result.

In particular, the insurance design server 30 requests the prediction server 50 to predict the customer's disease, and receives prediction information from the prediction server 50 .

In addition, the prediction server 50 collects sample data such as demographic data and past examination/diagnosis and onset data of patients, and predicts the customer's disease by using this. In particular, upon receiving the customer's demographic information and health checkup information, the disease for the customer is predicted.

In addition, when the prediction server 50 determines that some of the health checkup information of the customer is missing, the prediction server 50 corrects the missing missing data by using the sample data.

Meanwhile, the insurance design server 30 and the prediction server 50 are functionally separated servers, and may be implemented as a single server. Hereinafter, for convenience of explanation, it will be described that the function of the prediction server 50 is implemented in the insurance design server 30 .

Meanwhile, the client 20 and the insurance design server 30 (and the prediction server 50 ) may be implemented according to a typical method of configuring a client and a server. That is, the functions of the entire system can be divided according to the performance of the client or the amount of communication between the server and the server. As an example, the client 20 simply collects data for a questionnaire or outputs the analysis result, and the insurance design server 30 collects and analyzes the collected sample data, or predicts the customer's disease. . As another example, the insurance design server 30 may perform a customer's past examination data collection operation, and the client 20 may directly perform all or a part of an analysis operation or a prediction operation. Hereinafter, the disease prediction service system will be described, but it may be implemented in various forms of sharing depending on the server-client configuration method.

Next, the database 40 includes a sample data DB 41 for storing sample data such as demographic information and health examination results of patients, and a customer health information DB 42 for storing demographic information and health examination results of customers. ), and a prediction information DB 43 that stores the customer's possible disease and its probability, risk, and the like. However, the configuration of the database 40 is only a preferred embodiment, and in developing a specific device, it may be configured in a different structure according to the database construction theory in consideration of the ease and efficiency of access and search.

Next, a disease prediction service system 300 for insurance design according to an embodiment of the present invention will be described with reference to FIG. 2 .

As described above, the disease prediction service system 300 according to the present invention is implemented in a server-client manner with the insurance design client 20 and the insurance design server 30 (or the prediction server 50 ).

As shown in FIG. 2 , the disease prediction service system 300 according to an embodiment of the present invention includes a sample data collection unit 31 that collects demographic information, health checkup information, onset data, etc. of past patients as sample data. ), a questionnaire input unit 32 that receives the customer's demographic information as a questionnaire, a health checkup collection unit 33 that collects the customer's health checkup results, and a missing value correction unit that corrects missing data among the customer's health checkup results (34), and a disease prediction unit 35 for predicting the customer's disease by using the customer's demographic information and the corrected health checkup result. In addition, it may be configured to further include a similar variable selector 36 for selecting a similar group extraction variable.

First, the sample data collection unit 31 collects demographic information, health checkup information, onset data, and the like of past patients as sample data.

Demographic information is data representing a patient's health status, and consists of age, gender, height, weight, presence of a disability, lifestyle, and the like.

In addition, the health checkup information is the patient's health checkup data, and is data measured (checked) when performing a health checkup, such as blood pressure, cholesterol level, hemoglobin level, and urine protein level.

In addition, the onset data is data on the disease onset of the patient, and indicates whether the patient has the disease.

On the other hand, sample data is collected by being classified or identified by demographic information. That is, personal information that identifies the patient, such as the patient's name, is excluded, and medical data is collected based on health status information such as age, gender, height, weight, disability, and lifestyle.

In particular, preferably, the sample data uses a sample cohort DB. The total data that is building the sample cohort DB refers to the data of 1 million people. The 1 million subjects were stratified based on the gender and age of the citizens and the distribution of residential areas, so it can be said that the results derived from this data are representative of the whole nation.

Next, the questionnaire input unit 32 receives the customer's demographic information as a questionnaire.

As described above, the customer's demographic information consists of age, gender, height, weight, presence or absence of a disability, lifestyle, income quintile, past medical history, family medical history, and the like. Customer's demographic information is obtained by questionnaire. Therefore, in the following, questionnaire data are mixed with demographic data.

As an example, if an interface for questioning demographic information is provided through the insurance planner's smart mobile terminal 10, the customer directly responds to each questionnaire and receives the customer's information.

Preferably, the questionnaire consists of 26 items. In other words, the questionnaire data includes gender, age, region of residence, insurance subscription type, income quintile, disability, type of examination institution, height, weight, person (stroke, heart disease, high blood pressure, diabetes, dyslipidemia, pulmonary tuberculosis, and other diseases including cancer) ), family history (stroke, heart disease, high blood pressure, diabetes, liver disease, cancer) history, smoking status, smoking period, smoking amount per day, drinking habits, alcohol consumption per day, exercise amount per week, etc.

Next, the health checkup collection unit 33 collects the customer's health checkup data.

The customer's health checkup data consists of data measured during the health checkup, such as systolic blood pressure, diastolic blood pressure, pre-meal blood sugar, total cholesterol, high-density cholesterol, low-density cholesterol, triglycerides, and urine protein.

At this time, the customer's health checkup data is directly input by the customer, or the most recent health checkup data is obtained from the Health Insurance Corporation, a medical data institution (Health Insight), etc. through the customer's accredited authentication process. By accessing the site from the smart mobile terminal 10, the customer's authentication information can be entered and directly collected.

In general, when the National Health Insurance Corporation general health checkup is conducted, the items collected by the questionnaire before the checkup are usually called questionnaire data. However, when importing the customer's health checkup data, the questionnaire data during the health checkup process cannot be imported. Accordingly, the corresponding data is received through the paper-weight input unit 32 in advance. In addition, the user's 'height' and 'weight' are included in the health checkup, but for the same reason as above, this service is included in the questionnaire.

Screening data that are usually measured during a health checkup include waist circumference, systolic blood pressure, diastolic blood pressure, fasting blood sugar, total cholesterol, HDL cholesterol, LDL cholesterol, triglycerides, hemoglobin, urine protein, serum creatinine, serum GioT, serum GPT, and Gamma GT.

On the other hand, among the health checkup results, the four screening data for blood lipid levels (total cholesterol, high-density cholesterol, low-density cholesterol, and triglycerides) are intermittently examined without being examined every year after a certain point in time. Therefore, when retrieving the recent health checkup data, the corresponding value may be missing.

In addition, some items may be omitted from the examination center during the actual health examination. Also, when the customer directly inputs, the customer may not be able to input some examination data, resulting in missing data.

Next, the similar variable selecting unit 36 selects an extraction variable (or a similar group extraction variable) for extracting a similar group from the sample data.

In order to obtain an accurate disease occurrence prediction value, not only the user's demographic information and various health check-up information are required, but also the user's past disease history, family disease history, and daily life habits are required. However, in reality, it is impossible for the user to memorize all of the above information, and there are many cases where it is difficult for the user to obtain his/her own health information, or there is a difficulty in using the disease prediction model due to problems with personal information protection.

Therefore, even if the accuracy is somewhat lower, the user should be able to use the disease prediction model more easily with simple questionnaires and questionnaires that can be memorized right away (age, weight, height, daily lifestyle, personal and family disease history, etc.). In other words, since it is difficult to use the future disease occurrence prediction model through all indicators related to the patient's health, it is difficult to derive results only from questionnaires and questionnaires that can be easily answered, or health where the rest of the responses cannot be answered only with the simple items. It would be very convenient to be able to estimate all relevant figures.

The greater the effect of the missing value to be estimated on the performance in the disease prediction model, the more the actual classification or the performance degradation of the predictive model becomes. Therefore, accurate estimation of the missing value is an important factor in the overall model performance and output.

In the present invention, groups similar to customer characteristics are extracted from sample data (sample cohort DB), and missing values of customers are replaced with representative values (mean values, median values, etc.) of the extracted similar groups for correction.

However, when extracting similar groups, the following factors should be considered in setting the extraction variables.

First, the data of the group to be extracted should be the group with minimal missing. That is, if there are many missing data in the extracted group, an accurate and objective similar group cannot be formed.

In addition, the extracted variable should be the variable with the least change (even over time) among the learning variables input to the development of the disease prediction model. For example, demographic and socioeconomic variables (region of residence, income level, insurance subscription type, etc.), health checkup results, and daily lifestyle information can change significantly depending on time and conditions. Therefore, these variables should be excluded as much as possible in estimating the replacement value.

In addition, the number of extraction variables should be minimized to reduce the load on estimation value calculation. Using all of the learning variables input in the actual disease prediction model development stage as a standard for estimating missing values has a practical limitation in excessively increasing the implementation time of the technology. In particular, since the load increases as the number of variables increases, it is efficient to minimize the number of variables.

The similar variable selector 36 receives predefined candidate variables. That is, the manager or the like selects a variable with little change or a variable that is not missing, and defines candidate variables in advance. Preferably, the candidate variables consist of gender, age, height, weight, BMI, history, and the like.

Next, the similar variable selection unit 36 constitutes a combination of candidate variables. As an example, a combination of each candidate variable such as {sex, age}, {sex, age, height}, {sex, age, weight}, and {sex, age, BMI} is configured.

Next, the similar variable selector 36 receives a preset test set. The test set is multiple audience data.

Next, the similar variable selector 36 estimates the customer's examination data of the test set for each combination, and calculates difference values between the estimated data and the actual data. That is, the test set includes the customer's demographic information, examination data, onset data, and the like. Therefore, it is assumed that a specific item (variable) of the examination data is missing data, and the data value of the corresponding variable is estimated. Then, the difference value is calculated by comparing the difference between the estimated value and the actual data of the test set.

Next, the similar variable selector 36 averages the difference values of the test sets for each combination, obtains an average of the difference values for each combination, and sets the average as an error for each combination.

Next, the similar variable selector 36 selects a combination having an error within a predetermined error range, and selects a combination having the smallest error from the actual value among the selected combinations.

Finally, the similar variable selection unit 36 selects variables belonging to the fewest combinations as the similar group extraction variables.

Preferably, the similar group extraction variables are selected as gender, age, and BMI variables. In other words, it was found that the importance of additional variables and statistics and performance of the calculated results were not superiorly improved in selecting factors other than gender, age, and BMI as the missing value estimation criteria. In addition, if more variables other than the above three factors are considered, there are too many variables by type (continuous and categorical), so it cannot contribute to the estimation of the estimated value because there is already a missing value in the relevant variable, or implement a technique to estimate the missing value time consumption will be very large.

Next, the missing value correction unit 34 corrects the missing data among the results of the customer's health checkup. In particular, the missing value correction unit 34 corrects missing data among the data of the input variables of the disease prediction model.

That is, the missing value correcting unit 34 extracts a similar group from the sample data, sets the missing data of the corresponding customer as a representative value of the similar group, and corrects it.

A similar group refers to a set of data partially extracted from standard data by a similar group extraction variable. That is, the similar group is a subset of the standard data. At this time, data having the same value (value in the same range) as the value of the customer's extraction variable (similar group extraction variable) is extracted.

Also, the number of similar groups may be set to a predetermined number. In this case, data having the same value are randomly extracted. That is, the number of similar groups may be limited to 10,000, and all data in the database may be utilized.

In addition, the similarity group extraction variable is composed of a plurality of predetermined variables. Preferably, the similar-group extraction variables consist of gender, age, and BMI values. The BMI value is an index for determining the degree of obesity, and is calculated using the ratio of height to weight. When the values of the similar group extraction variables are all the same (in the same range), the corresponding data is extracted.

That is, the same sample data as the customer's gender, age, and BMI are extracted as a similar group. At this time, whether the extracted variables are the same, the values of the variables are divided into ranges or categories, and if they are included in the corresponding range or category, they are determined to be the same. For example, the age may be divided into categories of 5 years old, from 19 years old to 85 years old or older, and the gender may be divided into two categories such as male and female. In addition, the BMI value can be divided into a total of 6 stages from underweight to 3rd stage obesity, and underweight is classified as less than 18.5 and 3rd stage obesity as 35 or higher. In this case, data from a 43-year-old woman with a BMI of 31 and data from a 41-year-old woman with a BMI of 33 are classified into the same group (similar group).

In addition, when extracting a similar group, a patient group and a control group are obtained, respectively, and a similar group is formed. Preferably, a similar group is formed by combining the same number of patient groups and control groups.

The patient group is a set of data of patients diagnosed with a disease in the sample data, and the control group is a set of data of patients who are not diagnosed with a disease in the sample data. In other words, the data set for the ‘patient group’ data set includes the medical questionnaire data for the year of the patient who was first diagnosed with the disease code of the disease to be predicted (ex. type 2 diabetes: E11) as the main diagnosis and the medical examination data 2 years ago. In addition, during the total follow-up period of the sample data (or sample cohort DB) for 12 years, as the main and minor diagnoses, none of the relevant diseases (ex. all except metabolic syndrome E00-99 including diabetes) The data of people who have not been diagnosed with the disease are set as the 'control' dataset. At this time, the number of controls is randomly extracted with the same number as the number of datasets in the patient group. The data of the control group is also composed of the questionnaire data for the current year and the health checkup data from 2 years ago.

In particular, preferably, the similar group sets the patient group and control group of the sample data to be 5:5. That is, as for the learning data for each disease, the patient group and control group are set to 5:5. However, if the missing value is patient group data, the replacement value is calculated only from the patient group data. Conversely, if the data is control data, the replacement value is calculated only from the control group data.

On the other hand, the representative value of the similar group is set as the 'mode' of the corresponding values in the case of categorical data and the 'median' of the corresponding values in the case of continuous data.

In other words, the basic principle of correcting missing values is to insert the values most representative of the people. If a missing value occurs in the data input by the user, the median value of the continuous variable (ex. height, weight, blood pressure, blood sugar, etc.) in the entire training dataset constructed for each disease is the median value, and the categorical variable (ex. residential area, For income level, drinking, smoking, exercise, etc.), the mode value is replaced with the corrected value.

For example, it is assumed that the similar group consists of all 10,000 data, and the customer's missing data are low-density cholesterol and triglycerides. At this time, since low-density cholesterol and triglycerides are continuous values, the customer's missing data is set and corrected as the median values of low-density cholesterol and triglycerides, respectively, of 10,000 data of similar groups.

In summary, the missing value correcting unit 34 extracts a similar group of customers from the sample data, obtains a representative value of the similar group, and sets and corrects the missing data with the obtained representative value.

Next, the disease prediction unit 35 predicts the customer's disease by using the customer's demographic information and the corrected health checkup data.

Preferably, the disease prediction unit 35 predicts the customer's disease using the disease prediction model. When the disease prediction model receives an input value of a predetermined input variable, the disease prediction model outputs an onset probability of each predetermined disease variable.

In particular, the disease prediction model is composed of a neural network, etc., and internal variables are learned by learning data. And when the disease prediction model is trained, when the values of the input variables are input, the disease prediction model outputs the probability of occurrence of each disease.

The disease prediction model is the result of an artificial intelligence neural network that machine-learned millions of health check-up results from thousands to tens of thousands of people selected to represent domestic patients for each disease, demographic factors, and lifestyles. The calculation result may vary freely due to the user's steady improvement in health behavior, etc.

3 shows the input variables of the disease prediction model. As shown in FIG. 3 , all of them consist of a total of 44 input variables. In particular, among these input variables, the proportion of missing values for variables 15-18 (related to blood lipid levels) is high.

Also, preferably, the output variable consists of the incidence probabilities for 12 diseases (or 12 major diseases). In particular, the diseases are breast cancer, five major cancers, cancer integration, cerebrovascular disease, osteoporosis, cataract, hypertension, obesity, diabetes, COPD (chronic obstructive pulmonary disease), joint disease, dyslipidemia, and the like.

In addition, when the output value of the corresponding disease is greater than or equal to a predetermined reference probability, the disease prediction unit 35 selects the corresponding disease as a possible disease.

As mentioned above, although the invention made by the present inventors has been described in detail according to the above embodiments, the present invention is not limited to the above embodiments, and various modifications can be made without departing from the gist of the present invention.

10: smart mobile terminal 20: insurance design client
30: insurance design server 31: sample data collection unit
32: questionnaire input unit 33: health checkup collection unit
34: missing value correction unit 35: disease prediction unit
36: similar variable selection unit
40: database 41: sample data DB
42: Customer health information DB 43: Prediction information DB
50: prediction server 80: network

Claims (8)

In the disease prediction service system for insurance design,
a sample data collection unit that collects demographic information, health checkup data, and outbreak data of past patients as sample data;
a questionnaire input unit that receives the customer's demographic information as a questionnaire;
a health checkup collection unit that collects customer health checkup data;
a missing value correction unit that extracts a similar group similar to the customer from the sample data, and replaces the missing data among the customer's health checkup data with a representative value of the extracted similar group to correct; and,
A disease prediction unit that predicts the customer's disease using the customer's demographic information and the corrected health checkup data;
The system further includes a similar variable selection unit for selecting a similar group extraction variable,
The missing value correction unit extracts data having the same value as the data value of the similar group extraction variable of the customer into the similar group,
The similar variable selection unit configures a plurality of combinations by combining predetermined candidate variables, estimates the customer's examination data for each combination with respect to a predetermined test set, and estimates the examination data and the actual data of the test set. obtain the difference value of , set the statistical value of the difference value of all test sets for each combination as the error, extract combinations having an error within a predetermined error range, and select the combination with the smallest number of variables among the extracted combinations. A disease prediction service system for insurance design, characterized in that the variables are selected as similar group extraction variables.
delete delete The method of claim 1,
The disease prediction service system for insurance design, characterized in that the similar group extraction variable consists of gender, age, and BMI.
The method of claim 1,
The missing value correcting unit forms a similarity group to have a predetermined number of data, but randomly extracts data having the same value as the data value of the customer's similarity group extraction variable from the sample data. Disease prediction service system.
6. The method of claim 5,
The missing value correction unit obtains a patient group and a control group, respectively, to form a similar group, and constitutes the similar group by combining the same number of patient groups and control groups, and the control group is extracted from the data of patients who have not been diagnosed with the disease from the sample data, and the patient group A disease prediction service system for insurance design, characterized in that the sample data is extracted from the data of patients diagnosed with the disease.
The method of claim 1,
The representative value of the similar group is set as the 'mode' of the corresponding values in the case of categorical data and the 'median' of the corresponding values in the case of continuous data.
The method of claim 1,
The demographic information includes gender, age, region of residence, insurance subscription type, income decile, disability, height, and weight,
The health checkup data includes waist circumference, systolic blood pressure, diastolic blood pressure, fasting blood sugar, total cholesterol, high-density cholesterol, low-density cholesterol, triglyceride, hemoglobin, urine protein, serum creatinine, serum G.T, serum GPT, and gamma GTP. Disease prediction service system for insurance design, characterized in that it comprises.

Images