Diabetes prediction method, diabetes prediction system, and diabetes prediction program

Abstract

To provide a technique for predicting the risk of a non-diabetes person developing diabetes or prediabetes in the future.SOLUTION: A diabetes prediction method is provided, comprising an information acquisition step of acquiring subject information of a population consisting of subjects, and a risk score computation step of performing data envelopment analysis to derive risk scores used for assessing the risk of the subjects developing diabetes in the future from the subject information acquired in the information acquisition step.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a diabetes prediction method, a diabetes prediction system, and a diabetes prediction program. [Background technology]

[0002] Current diabetes countermeasures are centered on secondary prevention, which involves early detection of people who have developed diabetes through the results of health checkups and other tests, and early treatment to prevent the disease from worsening.

[0003] Evaluation methods have been developed for determining whether a person's health condition poses a high risk of contracting a particular disease. For example, Patent Literature 1 discloses a method for determining the risk of contracting a particular disease using a machine learning model. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2023-113955 Summary of the Invention [Problem to be solved by the invention]

[0005] If we could understand the risk of developing diabetes based on health status, we could implement primary prevention to prevent the onset of diabetes before it actually occurs. Furthermore, if we could understand the risk of developing prediabetes, a condition in which blood glucose levels are higher than normal but not high enough to be classified as diabetes, we could lower the risk of developing diabetes through primary prevention. However, there are no known methods for predicting prediabetes.

[0006] An object of the present invention is to provide a technology that can predict the risk of a person who does not have diabetes developing diabetes or prediabetes in the future. [Means for solving the problem]

[0007] A diabetes prediction method according to one aspect of the present invention includes an information acquisition step of acquiring subject information of a population composed of subjects, and a risk score calculation step of calculating a risk score from the subject information acquired in the information acquisition step using data envelope analysis to evaluate the risk of the subject developing diabetes in the future.

[0008] The above-described diabetes prediction method calculates a risk score for each individual in a population of subjects, and makes it possible to relatively assess the risk of developing diabetes within the population based on the calculated risk score, thereby making it possible to predict the risk of developing diabetes or prediabetes in people who do not currently have diabetes.

[0009] The subject information includes BMI, waist circumference, blood pressure, and HbA1 C The risk score may be higher as the subject's risk of developing diabetes in the future increases. C If the risk is known, the risk score can be calculated without undergoing a blood test.

[0010] The method may further include an information conversion step of converting a part of the subject information acquired in the information acquisition step so that the minimum value and the maximum value in the population are interchanged, and the risk score calculation step may calculate the risk score from the converted information converted in the information conversion step using data envelopment analysis. By converting a part of the subject information, a DMU that can maximize output relative to input in data envelopment analysis satisfies the definition of efficient. This can be done.

[0011] Furthermore, the subject information may be information obtained from sources other than the blood test results of the subject. C If the risk scores are known, it is possible to calculate the risk score from BMI, waist circumference, and blood pressure even for subjects who have not undergone blood tests.

[0012] The subject information may be information obtained from the results of a blood test of the subject. The subject's risk score can be calculated based on the results of a medical examination or the like that includes a blood test.

[0013] Furthermore, the risk score calculation step may calculate the risk score from the converted information and unconverted information that has not been converted in the information conversion step. Unconverted information may be included if it satisfies the definition of data envelopment analysis.

[0014] The conversion information may also include the subject's BMI, waist circumference, blood pressure, GGT, TG, and HbA1 C The unconverted information may be the subject's HDL-C. Such converted and unconverted information can satisfy the definition that a DMU that can maximize output relative to input is efficient in data envelopment analysis.

[0015] The risk score calculation step may calculate the risk score, which becomes lower as the subject's risk of developing diabetes in the future increases, by taking the reciprocal of the non-risk score calculated using output-oriented data envelopment analysis from the converted information and unconverted information not converted in the information conversion step. Since the crude result obtained by output-oriented data envelopment analysis is an inefficiency score, it can be converted into an efficiency score by taking the reciprocal and used as the risk score.

[0016] Furthermore, the population may be a group that is actually targeted by an entity that implements the diabetes prediction method. Since factors that have a large impact on developing diabetes vary depending on the gender, age, environment, etc. of the subject, it is preferable that the risk score calculated by the diabetes prediction method be calculated for each population that is actually targeted by an entity that implements the diabetes prediction method according to the first embodiment, such as a region, occupation, workplace, or school.

[0017] The method may also include an extraction step of extracting a discriminant value that can determine whether a subject has a high probability of developing diabetes in the future from past information of the population, and if the risk score calculated in the risk score calculation step is lower than the high risk score extracted in the extraction step, the subject may be predicted to have a low risk of developing diabetes in the future. By extracting factors that contribute to a high risk of developing diabetes or pre-diabetes, a discriminant value that determines whether a subject has a high risk of developing diabetes or pre-diabetes can be specifically calculated.

[0018] The present invention can also be viewed as a diabetes prediction system or a diabetes prediction program. For example, a diabetes prediction system according to one aspect of the present invention may include: an information acquisition means for acquiring subject information of a population composed of subjects; and a risk score calculation means for calculating, using data envelopment analysis, a risk score used to assess the risk of the subject developing diabetes in the future from the subject information acquired by the information acquisition means. Based on the risk score calculated using the diabetes prediction system, the subject's future risk of developing diabetes or prediabetes can be predicted. [Effects of the Invention]

[0019] According to the present invention, it is possible to predict the risk of a person who does not have diabetes developing diabetes or prediabetes in the future. [Brief explanation of the drawings]

[0020] [Figure 1] FIG. 1 is a flowchart of the diabetes prediction method according to the first embodiment. [Figure 2] FIG. 2 is a flowchart of the diabetes prediction method according to the second embodiment. [Figure 3] Figure 3 shows the created decision tree (part 1). [Figure 4] Figure 4 shows the created decision tree (part 2). [Figure 5]FIG. 5 is a configuration diagram of a diabetes prediction device constituting a diabetes prediction system according to the third embodiment. [Figure 6] FIG. 6 is a functional block diagram of a diabetes prediction device used in the diabetes prediction method according to the third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0021] The following describes an embodiment of the present invention. The embodiment described below is one aspect of the present invention and does not limit the technical scope of the present invention.

[0022] The present invention includes the prediction of the possibility that a subject will develop diabetes or prediabetes, but does not include the final act of making a diagnosis. A physician may refer to the prediction results of the present invention to predict the possibility that a subject will develop diabetes or prediabetes and decide on a treatment plan, such as primary prevention. Therefore, the present invention can be rephrased as a technology that provides information for predicting and diagnosing the possibility that a subject will develop diabetes or prediabetes. The information for making a decision can also be rephrased as information that serves as a basis for making a decision.

[0023] First Embodiment A diabetes prediction method according to a first embodiment of the present invention will be described. A subject whose future onset of diabetes or prediabetes is predicted by the diabetes prediction method according to the first embodiment is a person who has been diagnosed as not having diabetes in a medical checkup or other examination.

[0024] The key to diabetes prevention is primary prevention, which aims to prevent the onset of diabetes in healthy people who have not yet developed diabetes, before it develops, and secondary prevention, which aims to prevent the disease from worsening through early detection and treatment once diabetes has developed.Currently, the main focus of diabetes prevention is secondary prevention.

[0025] In the diagnosis of diabetes, an OGTT (Oral Glucose Tolerance Test) is sometimes used. In an OGTT, a person suspected of having diabetes drinks a drink containing a certain amount of glucose, and after a certain period of time, blood is drawn to evaluate whether the person has diabetes based on the change in blood glucose over time. The OGTT is time-consuming and laborious to perform, and places a strain on the examinee's body, making it difficult to perform in tests such as health checkups. Therefore, tests such as health checkups usually use FPG (Fasting Plasma Glucose) or HbA1. C The test results are used to assess whether a patient has impaired glucose tolerance, which is the ability to process glucose to maintain normal blood sugar levels.

[0026] OGTT, FPG and HbA1 C Since the targets to be detected are different, FPG and HbA1 C Even if the results are within the normal range, the OGTT may indicate that the patient meets the criteria for diabetes or prediabetes. C Even among people who do not meet the criteria for diabetes or prediabetes, 11.5 out of 1,000 people develop diabetes in a year. Prediabetes is a condition in which blood glucose levels are higher than normal, but not high enough to be classified as diabetes.

[0027] The diabetes prediction method in the first embodiment is C Or both The risk of developing diabetes or prediabetes in a group of normal subjects is assessed based on a risk score calculated using Data Envelopment Analysis (DEA).

[0028] Data envelope analysis is known as a method used in the field of industrial engineering. Data envelope analysis uses linear programming to calculate a value called an efficiency score from any revenue and cost indicators for each business entity. Multiple revenue and cost indicators can be taken into account, and a single value called an efficiency score is output, making it possible to clearly compare business entities. This allows for relative comparison within a group based on the evaluation of each individual within the group, rather than finding a single optimal solution for the group as in regression analysis.

[0029] In the first embodiment, data envelopment analysis is used to calculate a risk score for each individual in a population of subjects, and the risk of developing diabetes or prediabetes within the population can be relatively assessed based on the calculated risk score. Because the risk score is calculated as a specific numerical value, subjects who are targets for primary prevention can be notified of their future risk. Furthermore, because data envelopment analysis reduces the risk of developing diabetes or prediabetes to a single value, subjects can more easily understand their own future risk of developing diabetes or prediabetes than if they were advised to take measures against diabetes based on an expert's judgment that combines the results of multiple physiological tests and blood tests. Therefore, subjects who are targets for primary prevention can recognize that they are targets for primary prevention, which can lead to motivation to improve their lifestyle habits and behavioral change toward diabetes prevention.

[0030] Furthermore, by using data envelopment analysis, it is possible to prioritize primary prevention interventions, such as conducting OGTT, for subjects predicted to be at high risk of developing diabetes or prediabetes in the future.

[0031] In addition, OGTT, FPG or HbA1 C Data envelopment analysis can be used to assess the future risk of developing diabetes or pre-diabetes for people who have been determined to have a low risk of developing diabetes and are being followed up approximately every three years. In this case, if it is possible to predict who is at high risk of developing diabetes or pre-diabetes within three years, it will be possible to prioritize interventions such as primary prevention before pre-diabetes develops.

[0032] Next, the diabetes prediction method according to the first embodiment will be described in detail with reference to Fig. 1. Fig. 1 is a flowchart of the diabetes prediction method according to the first embodiment. First, in an information acquisition step, subject information of a population made up of subjects is acquired (step S101).

[0033] The subject information includes diabetes risk factor information, which is a risk factor for diabetes, and HbA1, which can determine whether the subject has diabetes. C The diabetes risk factor information is the subject's BMI (Body Mass Index), waist circumference, blood pressure, GGT (Gamma-glutamyl Transferase), TG (Triglyceride), and HDL-C (High Density Lipoprotein Cholesterol). This information can be obtained during tests such as health checkups, so it is possible to predict whether a subject is at risk of developing diabetes or pre-diabetes in the future based on the test results of health checkups, etc. Note that diabetes risk factor information can also be limited to BMI, waist circumference, and blood pressure without blood test results. In this case, since it does not include the subject's blood information, HbA1 C If the blood glucose level is known, the test results of the subject's health checkup or the like may be used without conducting another blood test. In this case, the risk of diabetes can be predicted using information obtained from sources other than blood test results, so that the risk of the subject developing diabetes or prediabetes in the future can be predicted more simply and conveniently based on minimally invasive test results. The diabetes risk factor information is not limited to the above, and may be any information that can be a risk factor for diabetes. In the first embodiment, the information that can determine whether the subject has diabetes is HbA1, which is widely used in tests such as health checkups. C However, FPG or OGTT, which are used to determine whether a subject has diabetes, may also be used.

[0034] In the information conversion step S102 following step S101, the maximum and minimum values of part of the subject information acquired in the information acquisition step (step S101) are converted so that they are swapped (step S102). This step S102 is performed to satisfy the definition that a DMU that can maximize output relative to input is efficient in the data envelopment analysis method. The converted information with the maximum and minimum values converted is BMI, waist circumference, blood pressure, GGT, TG, and HbA1. C The unconverted information that was not converted in step S102 is HDL-C. The converted information is converted by subtracting each piece of information from the sum of the maximum and minimum values in the population. The information conversion is not limited to the above and may be performed by a method such as taking a reciprocal as long as the definition of data envelopment analysis is met. Furthermore, if the definition of data envelopment analysis is met with the unconverted information alone, the subject information does not need to be converted.

[0035] In the risk score calculation step S103 following step S102, a data envelopment analysis method is used to calculate a risk score to be used to assess the subject's risk of developing diabetes in the future from the converted information converted in the information conversion step (step S102) and the unconverted information not converted in the information conversion step (step S102).

[0036] The risk score calculation step (step S103) may use an output-oriented data envelope analysis method that focuses on improving output values. The crude result obtained by the output-oriented data envelope analysis method is an inefficiency score, so it can be converted into an efficiency score by taking its reciprocal, which can be used as the risk score. Theoretically, the risk score is a value ranging from 0.0 to 1.0, and the higher this value, the more resistant the subject is to diabetes and the lower the risk of developing diabetes. Therefore, the higher the risk of a subject developing diabetes or pre-diabetes in the future, the lower the risk score will be. Note that the inefficiency score and efficiency score usually indicate the results when using data envelope analysis in the field of industrial engineering.

[0037] The inefficiency score obtained by the output-oriented data envelopment analysis method may be treated as a non-risk score, and the higher the non-risk score, the more likely the subject is to develop diabetes or pre-diabetes in the future. C The higher the risk score obtained by the output-oriented data envelopment analysis, the more likely the subject is to develop diabetes or prediabetes in the future. In this case, the subject information does not need to include blood information. For example, the unconverted information may be BMI, waist circumference, blood pressure, and HbA1. C The higher the risk score obtained by the output-oriented data envelopment analysis, the higher the possibility that the subject will develop diabetes or prediabetes in the future. Furthermore, the calculation of the risk score in the risk score calculation step (step S103) may be performed using an input-oriented data envelopment analysis that focuses on improving input values.

[0038] Furthermore, the calculation of the risk score by the diabetes prediction method according to the first embodiment is performed for each population that is actually the target of the subject who implements the diabetes prediction method according to the first embodiment, such as region, occupation, workplace, or school. The factors that have a large impact on developing diabetes vary depending on the gender, age, environment, etc. of the subject, and since this method is a method for calculating the relative risk within an analysis population, it is preferable to perform the method as a collective population that is actually the target of the subject who implements the diabetes prediction method according to the first embodiment.

[0039] <Verification> The risk score calculated by data envelopment analysis and the subjects' risk of developing diabetes within three years were compared. We verified the correlation between the observed results.

[0040] A retrospective cohort study was conducted using data from specific health checkups of National Health Insurance subscribers living in Kaminoyama City, Yamagata Prefecture, from 2008 to 2013. The enrollment date for the study was the date of the health checkup, and the endpoints were the onset of diabetes, the onset of prediabetes, or the end of the observation period. The end of the observation period was the date of the last health checkup up to three years after the date of the health checkup for data from 2008 to 2010, the date of the last health checkup up to two years after the date of the health checkup for data from 2011, and the date of the last health checkup up to one year after the date of the health checkup for data from 2012. Because National Health Insurance covers individuals up to age 74, individuals aged 74 or older in each year were excluded from the analysis because their follow-up in the following year was not possible.

[0041] In this study, prediabetes was defined as HbA1 C Diabetes is defined as a HbA1 value of 5.7% or higher. C The HbA1 value was defined as 6.5% or higher. C Those with an HbA1 value of 5.6% were excluded to avoid bias, as the guidelines recommend that OGTT be performed. C The HbA1 value is less than 5.6%. C The values measured by the Japan Diabetes Society (JDS) were converted to NGSP (National Glycohemoglobin Standardization Program) values by adding 0.4%.

[0042] The data envelopment analysis method applied the output-oriented continuous returns to scale (CCR) model, with DMUs (Decision Making Units) as individuals. In this model, DMUs are responsible for converting input information into output information; in other words, the efficiency of the output information obtained based on the input information is evaluated. This model is also characterized by its non-parametric analysis, which does not require prior assumptions such as data distribution or hypotheses or conditions regarding the relationship between input and output when evaluating efficiency. Therefore, by collecting subject information, it is possible to evaluate risk for each group analyzed, and it is possible to reduce variation in the evaluation system due to hypotheses and conditions. In an output-oriented model, the DMU that produces the greatest output information for a specific input information is evaluated.

[0043] The input information was diabetes risk factor information, which is a factor in developing diabetes. The diabetes risk factor information was the subject's BMI, waist circumference, blood pressure, GGT, TG, and HDL-C. The output information was HbA1, which is information that can be used to determine whether the subject has diabetes. C It was decided.

[0044] In addition, BMI, waist circumference, blood pressure, GGT, TG and HbA1 C In this way, the results of the data envelope analysis obtained by converting the minimum and maximum values of some of the subject information showed that the BMI, waist circumference, blood pressure, GGT, and TG values were high, and even if the HDL-C value was low, the HbA1 C A DMU that achieves a low HbA1 value is considered to be efficient, with good output even with unfavorable input, meaning that the DMU has a low risk of developing diabetes or prediabetes. On the other hand, a DMU with lower BMI, waist circumference, blood pressure, GGT, and TG values and higher HDL-C values than such a DMU has a lower HbA1 value. CThis is an evaluation index that can be interpreted as being relatively inefficient unless the value is even lower, meaning that the DMU in question has a relatively high risk of developing diabetes or pre-diabetes in the future. Because the crude result obtained by output-oriented data envelopment analysis is an inefficiency score, we converted it into an efficiency score by taking its reciprocal and treated this as a risk score. Theoretically, the risk score ranges from 0.0 to 1.0, and the higher this value, the more resistant the individual is to diabetes and the lower the risk of developing diabetes or pre-diabetes.

[0045] In addition, we also analyzed the risk scores calculated by data envelopment analysis using the input information excluding blood test results from the above input information. Specifically, we used only BMI, waist circumference, and blood pressure as input information.

[0046] Statistical analysis was performed using R (version 4.1.0) for HbA1 C A regression analysis was conducted to examine the relationship between the risk score calculated using input information with blood test results and the risk score calculated using input information without blood test results and the onset of prediabetes or diabetes, which are turning points.

[0047] The results of the logistic regression analysis performed as a regression analysis showed that the crude odds ratio for developing diabetes associated with a 0.1 higher risk score calculated using input information including blood test results was 0.78 (95% confidence interval 0.62 to 0.97), and the crude odds ratio for developing diabetes associated with a 0.1 higher risk score calculated using input information excluding blood test results was 0.66 (95% confidence interval 0.52 to 0.84).

[0048] The subject of this analysis is HbA1 C This group is made up of so-called healthy people with an average blood sugar level of less than 5.6%, and whose physical measurements and blood test results are within the normal range. The risk score, which showed an association with the onset of diabetes and prediabetes within three years in this group, suggested its effectiveness as an indicator for identifying priority targets for primary prevention intervention.

[0049] As described above, the diabetes prediction method according to the first embodiment of the present invention uses data envelopment analysis to calculate a risk score for each individual in a population of subjects and to relatively assess the risk of developing diabetes or prediabetes within the population based on the calculated risk score. This makes it possible to predict the risk of a person who does not currently have diabetes developing diabetes or prediabetes in the future. This allows the subject to recognize that they are a target for primary prevention, which can motivate them to improve their lifestyle habits and lead to behavioral changes for diabetes prevention.

[0050] Furthermore, a physician may refer to the results of predicting the subject's risk of developing diabetes or prediabetes based on the risk score to ultimately predict the subject's risk of developing diabetes or prediabetes or determine a prevention policy, such as primary prevention. The method of the present invention for predicting the risk of developing diabetes or prediabetes can be an aid in the primary prevention of diabetes or prediabetes.

[0051] <Second embodiment> Next, a diabetes prediction method according to a second embodiment of the present invention will be described. After calculating the subject's risk score using data envelopment analysis, the subject's risk score for developing diabetes or prediabetes may be specifically predicted using the results of a decision tree analysis created based on past data of the population to which the subject belongs.

[0052] The diabetes prediction method according to the first embodiment allows a subject with a low risk score to know that he or she is at risk of developing diabetes or prediabetes in the future, but does not allow the specific value of the risk score that indicates a high risk of developing diabetes or prediabetes in the future to be known. On the other hand, the diabetes prediction method according to the second embodiment extracts factors that contribute to a high risk of developing diabetes or prediabetes through decision tree analysis and specifically calculates a risk score cutoff value (an example of the "discrimination value" referred to herein) that can determine whether a subject is at high risk of developing diabetes or prediabetes in the future. This makes it easy to cut off people who are at high risk of developing diabetes or prediabetes in the future, and allows people at high risk of developing diabetes or prediabetes to be given priority in taking measures against diabetes and implementing more intensive preventive measures.

[0053] Furthermore, since the decision tree can specifically grasp the cutoff value of the risk score indicating a high risk of developing diabetes or prediabetes, the insurer or the subject can easily grasp whether the subject is at high risk of developing diabetes or prediabetes in the future by using the risk score and the decision tree. Previously, this was only a portion of the population determined to have normal glucose tolerance, making it practically difficult to implement more intensive diabetes prevention interventions that devote sufficient human, material, and economic resources to such a large population. By identifying those at high risk of developing diabetes or prediabetes in the future, it becomes possible to provide individualized interventions such as health guidance and health classes. For example, if a subject has a high BMI, weight loss can reduce the risk of developing diabetes or prediabetes. For subjects, even if they do not currently have impaired glucose tolerance, being aware that they are at risk of developing diabetes or prediabetes in the future can motivate them to maintain a healthy state and lead to behavioral changes aimed at diabetes prevention.

[0054] The decision tree used to determine whether a subject is at risk of developing diabetes or prediabetes is created based on past data of the population to which the subject belongs. The decision tree can be appropriately updated and optimized by adding information on subjects whose future risk of developing diabetes or prediabetes is predicted by the diabetes prediction method according to the second embodiment.

[0055] Next, a diabetes prediction method using a decision tree will be described. FIG. 2 is a flowchart of the diabetes prediction method according to the second embodiment. First, a risk score for a subject is calculated, as in the first embodiment. Then, in an extraction step, a group likely to develop diabetes or prediabetes is identified from past information on the population using decision tree analysis, and a decision tree is created based on the results of extracting each factor required for the identification and the cutoff value for the risk score (step S104). If the risk score calculated in the risk score calculation step (step S103) does not satisfy the conditions for a high-risk group extracted by the result of decision tree analysis performed based on past data on the population to which the subject belongs, the subject is predicted to have a low risk of developing diabetes or prediabetes in the future.

[0056] The decision tree extracts factors that increase the likelihood that a subject will develop diabetes from past information about the population to which the input subject belongs. The information input into the decision tree includes the subject's age, gender, history of hypertension, history of dyslipidemia, smoking history, alcohol intake, and blood test results for LDL, AST, ALT, etc., at the time of past examinations of the population to which the subject belongs, as well as the risk score calculated in the risk score calculation step (step S103). This information can be obtained from test results such as health checkups, including interviews, and therefore factors that affect the onset of diabetes or prediabetes can be easily extracted. Therefore, the decision tree can be easily updated and optimized. Furthermore, by comparing the decision tree obtained by extracting factors that affect the onset of diabetes or prediabetes with the subject's risk score and the extracted factors, it is possible to easily identify subjects at high risk of developing diabetes or prediabetes. When notifying test results such as health checkups, not only the insurer but also the subject can be notified of their future risk of developing diabetes or prediabetes. Therefore, even if subjects with normal glucose tolerance are not interested in glucose tolerance, they can learn about their future risk of developing diabetes or prediabetes after undergoing a medical checkup or other examination, along with the results of the medical checkup, etc. Furthermore, it is relatively easy to provide an opportunity for all subjects whose test results show normal glucose tolerance to become interested in primary prevention of diabetes.

[0057] Furthermore, the past information of the population does not have to include blood test results. For example, it may only include the subject's age, sex, history of hypertension, history of dyslipidemia, smoking history, and alcohol intake at the time of the test. In this case, the subject can easily understand the subject's risk of developing diabetes or prediabetes in the future without undergoing a blood test. Note that the past information of the population to which the subject belongs that is used is not limited to the above, as long as it is a factor that increases the subject's likelihood of developing diabetes or prediabetes.

[0058] <Verification> A decision tree was created using the data for which risk scores were calculated in the verification of the first embodiment. Ta.

[0059] In the second embodiment, a decision tree is created using the decision tree analysis (CART). The analysis was performed using the rpart package using the CART algorithm. The minimum node size was set to 20 cases, and cost-complexity pruning (CP) was performed. CP values were calculated using the train function (caret package) under 10 iterations of 5-fold cross-validation to maximize the area under the receiver operating characteristic (ROC) curve (AUC). Actual data was applied to the resulting decision tree, and the sensitivity, accuracy, positive predictive value, negative predictive value, positive likelihood ratio, negative likelihood ratio, accuracy, and AUC were calculated. Note that the algorithm used to create the decision tree is not limited to the CART algorithm. For example, feature importance determined by random forests may also be used.

[0060] In creating the decision tree, the variable to be discriminated was the onset of diabetes within three years, and the input variables were the subject information and the results of data envelopment analysis. By creating a decision tree using these conditions, it is possible to extract factors that influence a subject's likelihood of developing diabetes within three years. Figures 3 and 4 show the created decision trees. Figure 3 is a decision tree when the population's past information includes blood test results. Figure 4 is a decision tree when the population's past information does not include blood test results. The populations used to create the decision trees are the same in both Figures 3 and 4.

[0061] In Figures 3 and 4, if a subject developed diabetes within the past three years, the column reads "affected," and the percentage of subjects who developed diabetes within the past three years is listed under "affected." If a subject did not develop diabetes within the past three years, the column reads "no abnormalities," and the percentage of subjects who did not develop diabetes within the past three years is listed under "no abnormalities." The number of subjects who met the conditions is indicated by "n."

[0062] In Figure 3, HbA1 CAmong subjects with a risk score of less than 0.78, 68.3% of the 31 subjects with an LCL-C level of less than 120 mg / dl developed diabetes within three years.

[0063] Figure 4 shows the HbA1 C Among the subjects with a risk score of less than 0.74, 65.3% of the 26 subjects aged 70 years or older, who had a blood glucose level of 5.5% or higher, were 61 years or older, and had a blood glucose level of less than 0.74, developed diabetes within three years. C It can be seen that of the 24 subjects under 70 years of age with hypertension, 58.3% of whom had a risk score of 5.5% or more, were 61 years of age or older, and had a risk score of less than 0.74, developed diabetes within three years.

[0064] In this way, decision tree analysis can identify factors that influence a subject's risk of developing diabetes or prediabetes and the cutoff value required for discrimination based on past population information. Therefore, it is possible to identify groups at high risk of developing diabetes or prediabetes and determine reference risk score cutoff values. In Figure 3, it is easy to see that a risk score of less than 0.78 indicates a high risk of developing diabetes or prediabetes, and in Figure 4, a high risk score of less than 0.74 indicates a high risk of developing diabetes or prediabetes. Therefore, subjects with a risk score indicating a high likelihood of developing diabetes or prediabetes extracted from the results of decision tree analysis can be determined to be at high risk of developing diabetes or prediabetes in the future, and preventive intervention can be implemented before the risk of developing diabetes increases. Furthermore, it can be seen that groups determined to be at high risk of developing diabetes or prediabetes using all factors are more likely to benefit from preventive intervention and therefore should be prioritized for implementation.

[0065] Previously, all people without diabetes who were the subjects of this study were given so-called healthy The campaign recommended lifestyle changes, such as reducing calorie intake and increasing moderate to high intensity physical activity, which are considered effective in preventing diabetes. However, the campaign was general in nature and targeted healthy people, so it must be said that the appeal to the target audience was low. In contrast, as shown in Figure 3, the HbA1 C By narrowing down the target population to those who are at high risk of developing diabetes or prediabetes in the future based on their age, risk score, and LCL-C, it becomes possible to notify the individuals of their high risk of developing the disease in the future and then provide more intensive intervention. This may contribute to primary prevention by increasing the appeal in terms of motivation and incentives for the individuals.

[0066] The diabetes prediction method according to the second embodiment performs a relative risk assessment based on given data, and therefore does not require any assumptions. Therefore, by collecting subject information from the population to which the subject belongs, it is possible to assess the risk for each population, and it is possible to suppress variations in the assessment system due to assumptions.

[0067] In this verification, the method for predicting future diabetes risk combining data envelopment analysis and decision tree analysis yielded relatively high specificity and negative predictive value values both when blood test results were included and when blood test results were not included. This indicates that the diabetes prediction method according to the second embodiment can accurately identify individuals with a low risk of developing the disease, even if they are excluded from priority intervention targets. Specificity, also known as the true negative rate, refers to the proportion of individuals who are correctly determined to be negative for the disease being detected by the test among those who do not have the disease being detected by the test. In the second embodiment, specificity refers to the proportion of individuals who are actually determined to be at low risk of developing diabetes among those who ultimately did not develop the disease. Negative predictive value refers to the proportion of individuals who actually did not have the disease being detected among those who tested negative. In the second embodiment, it refers to the proportion of individuals who were determined to be at low risk of developing the disease among those who tested negative.

[0068] As described above, the diabetes prediction method according to the second embodiment of the present invention can specifically calculate a group at high risk of developing diabetes or prediabetes and the cutoff value required to distinguish between these groups through decision tree analysis that extracts factors that contribute to a high risk of developing diabetes or prediabetes. This makes it possible to easily identify people at high risk of developing diabetes or prediabetes by excluding people at low risk of developing diabetes in the future, and thus allows people at high risk of developing diabetes or prediabetes to be prioritized for diabetes prevention measures and more intensive preventive interventions. Furthermore, because the decision tree allows the subject to specifically grasp the cutoff value, the subject can grasp their own risk of developing diabetes or prediabetes in the future, which can lead to behavioral changes aimed at diabetes prevention.

[0069] Furthermore, a physician can refer to the results of easily distinguishing people who are at high risk of developing diabetes or prediabetes in the future based on the decision tree analysis, and ultimately predict the risk of a subject developing diabetes or prediabetes or decide on a prevention policy such as primary prevention. The method of the present invention for predicting the risk of developing diabetes or prediabetes can be an aid in the primary prevention of diabetes or prediabetes.

[0070] <Third embodiment> Next, a third embodiment of the present invention will be described. In the third embodiment, a diabetes prediction system that predicts the possibility that a subject will develop diabetes or prediabetes in the future using the diabetes prediction methods according to the first and second embodiments and a diabetes prediction device including a diabetes prediction program may predict the risk of a subject developing diabetes or prediabetes in the future.

[0071] 5 is a configuration diagram of a diabetes prediction device 1 constituting a diabetes prediction system according to the third embodiment. The diabetes prediction system according to the third embodiment includes the diabetes prediction device 1. The diabetes prediction device 1 is a device that inputs subject information of a population made up of subjects who have undergone tests such as health checkups and been diagnosed as not having diabetes, and predicts whether the subjects are at risk of developing diabetes or pre-diabetes in the future based on the input subject information. Note that the diabetes prediction system may be composed of a single diabetes prediction device 1 as in the third embodiment, or may be composed of multiple devices selected from devices such as personal computers, tablet terminals, and smartphones.

[0072] The diabetes prediction device 1 is an information processing device that includes an input unit 2 for inputting subject information, an output unit 3 for outputting information calculated based on the information input by the input unit 2, for example, on a display, a non-volatile storage unit 4 such as an HDD (Hard Disk Drive) or SSD (Solid State Drive), a communication unit 5 for communicating with external devices and transmitting and receiving information, and a control unit 6 that has a CPU 6C and volatile memory 6M and controls the diabetes prediction device 1. The diabetes prediction device 1 runs a computer program that predicts whether a subject is at risk of developing diabetes in the future based on the subject information input to the input unit 2 or the subject information received by the communication unit 5.

[0073] Fig. 6 is a diagram showing functional blocks realized in the diabetes prediction device 1. The diabetes prediction device 1 has, as its main functions, an information acquisition unit 7 that acquires subject information, an information conversion unit 8 that converts information so that the minimum and maximum values of part of the subject information acquired by the information acquisition unit 7 are swapped, and a risk score calculation unit 9 that calculates a risk score. As shown in Fig. 6, in the diabetes prediction device 1, a program stored in the storage unit 4 of the diabetes prediction device 1 is expanded in the memory 6M and executed by the CPU 6C, thereby realizing acquisition of subject information by the information acquisition unit 7, conversion of the acquired information by the information conversion unit 8, and calculation of the risk score by the risk score calculation unit 9.

[0074] The information acquisition unit 7 acquires subject information. The acquired subject information may be subject information input to the input unit 2 of the diabetes prediction device 1, or subject information acquired by an external device and received by the communication unit 5 via a network.

[0075] The information conversion unit 8 converts part of the subject information acquired by the information acquisition unit 7 so that the minimum value and the maximum value are interchanged, and calculates conversion information.

[0076] The risk score calculation unit 9 calculates a risk score. The risk score is an index value used to evaluate the risk of a subject developing diabetes in the future. This allows the risk score to be calculated from the results of tests such as health checkups, allowing the subject to understand their own risk of developing diabetes or prediabetes in the future. Therefore, the subject can implement primary prevention by improving their lifestyle habits before the risk of developing diabetes increases even if they have no abnormalities in glucose tolerance.

[0077] The diabetes prediction device 1 may include a diabetes factor extraction unit (not shown). The diabetes factor extraction unit creates a decision tree by extracting diabetes factors that have a risk of developing diabetes from the past information of the population acquired by the information acquisition unit 7.

[0078] In this case, the past information of the population may be information previously acquired by the information acquisition unit 7, or may be information newly acquired by the information acquisition unit 7 for creating the decision tree.

[0079] Furthermore, physicians refer to the prediction results obtained using the diabetes prediction system and diabetes prediction program to ultimately predict the subject's risk of developing diabetes or prediabetes and determine a prevention policy, such as primary prevention. The system and program for predicting the risk of developing diabetes or prediabetes of the present invention can aid in the primary prevention of diabetes or prediabetes.

[0080] <Other embodiments> The features included in each of the embodiments disclosed above can be combined with each other. [Explanation of symbols]

[0081] 1. Diabetes prediction device 2. Input section 3. Output section 4...Storage section 5. Communications Department 6. Control section 6C··CPU 6M memory 7...Information acquisition section 8. Information conversion unit 9. Risk score calculation section

Claims

1. an information acquisition step of acquiring subject information of a population composed of subjects; and a risk score calculation step of calculating a risk score used to evaluate the risk of the subject developing diabetes in the future from the subject information acquired in the information acquisition step using data envelopment analysis. Diabetes prediction methods.

2. The subject information includes BMI, abdominal circumference, blood pressure, and HbA1 Ｃ and The higher the risk score, the higher the subject's risk of developing diabetes in the future. The method for predicting diabetes according to claim 1 .

3. an information conversion step of converting a part of the subject information acquired in the information acquisition step so that a minimum value and a maximum value in the population are interchanged; The risk score calculation step calculates the risk score from the converted information converted in the information conversion step using data envelopment analysis. The method for predicting diabetes according to claim 1 .

4. The subject information is information obtained from sources other than the subject's blood test results. The method for predicting diabetes according to claim 3 .

5. The subject information is information obtained from the blood test results of the subject. The method for predicting diabetes according to claim 3 .

6. the risk score calculation step calculates the risk score from the converted information and unconverted information that has not been converted in the information conversion step; The method for predicting diabetes according to claim 5 .

7. The conversion information includes the subject's BMI, waist circumference, blood pressure, GGT, TG, and HbA1 Ｃ and The unconverted information is the subject's HDL-C. The method for predicting diabetes according to claim 6.

8. The risk score calculation step calculates the risk score, which becomes lower as the subject's future risk of developing diabetes increases, by taking the reciprocal of the non-risk score calculated using output-oriented data envelopment analysis from the converted information and the unconverted information not converted in the information conversion step. The method for predicting diabetes according to any one of claims 3 to 7.

9. The population is a group of subjects who actually implement the diabetes prediction method. The method for predicting diabetes according to any one of claims 3 to 7.

10. an extraction step of extracting a discriminant value that can determine whether a subject has a high possibility of developing diabetes in the future from past information of the population, If the risk score calculated in the risk score calculation step is lower than the high risk score extracted in the extraction step, the subject is predicted to have a low risk of developing diabetes in the future. The method for predicting diabetes according to any one of claims 3 to 7.

11. an information acquisition means for acquiring subject information of a population composed of subjects; and a risk score calculation means for calculating a risk score to be used for evaluating the risk of the subject developing diabetes in the future from the subject information acquired by the information acquisition means using data envelope analysis. Diabetes prediction system.

12. On the computer, Obtaining subject information of a population composed of subjects; calculating a risk score to be used for assessing the risk of the subject developing diabetes in the future from the acquired subject information using data envelopment analysis; Diabetes Prediction Program.

Images