Diabetic complication risk assessment method and system

Abstract

The invention relates to the technical field of diabetes prediction, in particular to a risk assessment method and system for diabetic complications, and the method comprises the following steps: obtaining CGM data, and carrying out the preprocessing of the CGM data; dividing the preprocessed CGM data into the CGM data in different time periods; performing feature extraction on the CGM data of each time period; taking the extracted features as input of a machine learning algorithm, and adopting the machine learning algorithm for training; analyzing the influence of the extracted features on diabetic complications to obtain important features; a logistic regression model is adopted, and the corresponding relation between changes of the important features in different time periods and changes of the diabetic complication risk is calculated; according to the corresponding relation, the blood glucose value in the time period corresponding to the features is precisely regulated and controlled. According to the method, the CGM data in one day are divided into different time periods, the influence of the change of the CGM data in each time period on the diabetic complications is analyzed, and support is provided for precise management of the diabetic complications.

Description

Technical Field

[0001] This application relates to the technical field of diabetes prediction, and particularly to a method and system for assessing the risk of diabetes complications. Background Art

[0002] With the improvement of people's living standards and the acceleration of the aging process of the population in China, the prevalence of diabetes has been increasing year by year. At present, the number of diabetes patients in China has exceeded 120 million, ranking first in the world. Without scientific and effective blood glucose management, disorders of glucose metabolism will lead to a series of serious complications, such as diabetic nephropathy, retinopathy, peripheral neuropathy, etc.

[0003] Existing research shows that there is a certain correlation between blood glucose changes and diabetes complications. Therefore, providing precise management for diabetes complications based on continuous glucose monitoring (CGM) data is a current research hotspot. CGM is a technology that continuously monitors the glucose concentration in interstitial fluid of subcutaneous tissue through a glucose sensor. Usually, CGM can continuously obtain the blood glucose values of patients for 24 hours. Currently, researchers mainly calculate the standard deviation of blood glucose (SDBG), coefficient of variation (CV), maximum amplitude of glycemic excursions (LAGE), mean absolute glucose (MAG), time in range (TIR), etc. of CGM data, and analyze whether there is a correlation between these indicators and complications. However, it is not clear which time periods of blood glucose changes within a day have the greatest impact on diabetes complications. Summary of the Invention

[0004] An embodiment of this application provides a method and system for assessing the risk of diabetes complications. By dividing the CGM data within a day into different time periods and analyzing the impact of the changes in CGM data of each time period on diabetes complications, it provides important support for the precise management of diabetes complications based on CGM data.

[0005] To solve the above technical problems, in a first aspect, an embodiment of this application provides a method for assessing the risk of diabetes complications, including the following steps: First, obtain CGM data and preprocess the CGM data; then, divide the preprocessed CGM data into CGM data of different time periods; next, extract features from the CGM data of each time period; then, use the extracted features as the input of a machine learning algorithm, and the output is the risk of diabetes complications, and train using the machine learning algorithm; after training is completed, analyze the impact of the extracted features on diabetes complications to obtain important features for assessing diabetes complications; next, use a logistic regression model to calculate the corresponding relationship between the changes in important features of different time periods and the changes in the risk of complications of diabetes patients, and refine the impact of important features on diabetes complications; finally, according to the corresponding relationship, precisely regulate the blood glucose value of the time period corresponding to the feature to achieve precise management of diabetes complications.

[0006] In some exemplary embodiments, the preprocessing includes abnormal data processing and missing data processing; the abnormal data processing includes: if a certain CGM value shows an abnormal phenomenon of being significantly larger or smaller compared to the adjacent CGM values, then the average value of the two adjacent CGM values is used to replace the abnormal value; the missing data processing includes: if a certain CGM value is missing due to a temporary failure of the CGM device, the K-nearest neighbor value-taking method is adopted, and based on the K nearest samples of the missing data sample, the K values are weighted and averaged to estimate the missing data of the sample, so as to complete the data supplementation; the CGM data after preprocessing can be expressed as follows:

[0007] database = {G1, G2, G3, G4, G5...., G N}

[0008] Among them, G1 represents the first value of the CGM, G2 represents the second value of the CGM, and so on.

[0009] In some exemplary embodiments, the preprocessed CGM data is divided into CGM data for different time periods, including: according to the diet and sleep time of diabetic patients, the preprocessed CGM data is divided into: CGM data for the fasting-breakfast blood glucose time period, CGM data for the lunch blood glucose time period, CGM data for the dinner blood glucose time period, CGM data for the pre-sleep blood glucose time period, and CGM data for the night blood glucose time period.

[0010] In some exemplary embodiments, the fasting-breakfast blood glucose time period is 06:00 - 11:00; the lunch blood glucose time period is 11:00 - 16:00; the dinner blood glucose time period is 16:00 - 20:00; the pre-sleep blood glucose time period is 20:00 - 01:00; the night blood glucose time period is 01:00 - 06:00.

[0011] In some exemplary embodiments, feature extraction is performed on the CGM data for each time period, and the extracted features include: average value, standard deviation, variance, coefficient of variation, blood glucose fluctuation range value, blood glucose fluctuation change value, average absolute difference of blood glucose, average gradient change value of blood glucose, J-index, hypoglycemia index, hyperglycemia index, and blood glucose target range index.

[0012] In some exemplary embodiments, the coefficient of variation is calculated by dividing the standard deviation by the mean; the blood glucose fluctuation range value is calculated by subtracting the minimum value from the maximum value of the CGM data; the blood glucose fluctuation change value is calculated by dividing the blood glucose fluctuation range value by the mean; the mean absolute difference of blood glucose is obtained by first calculating the sum of the absolute values of all adjacent CGM differences and then calculating the mean; the mean gradient change value of blood glucose is obtained by first calculating the sum of all adjacent CGM gradient values and then calculating the mean; the J-index is obtained by calculating the square of the sum of the mean and the standard deviation; the hypoglycemia index is obtained by first performing a statistical transformation on each CGM data, calculating the hypoglycemia risk based on the transformation result, and then calculating the mean of all hypoglycemia risks within that time period; the hyperglycemia index is obtained by first performing a statistical transformation on each CGM data, then calculating the hyperglycemia risk based on the transformation result, and then calculating the mean of all hyperglycemia risks within that time period; the blood glucose target range index is obtained by calculating the ratio of the number of blood glucose values within the target range of 3.9 mmol / L to 10 mmol / L to the total number of blood glucose values in the CGM data for each time period.

[0013] In some exemplary embodiments, multiple machine learning algorithms are used for training. The machine learning algorithms include random forest, decision tree, and support vector machine. The extracted features are used as the input of the machine learning algorithms, and the output is the risk of diabetic complications for training. After training, based on the SHAP analysis method, the marginal contribution degrees of all features are visualized, and they are sorted according to the contribution degrees of the features. The top 10% of the features are selected as the important features for evaluating diabetic complications, and the time periods corresponding to the top 10% of the features are determined.

[0014] In some exemplary embodiments, a logistic regression model is used to calculate the correspondence between the changes in important features in different time periods and the changes in the risk of diabetic complications in patients, including: using the logistic regression model to calculate the OR value and the 95% confidence interval of diabetic complications in patients for the top 10% of the features respectively, so as to determine the proportion of increase or decrease in the risk of diabetic complications in patients when each feature increases by 10%.

[0015] In some exemplary embodiments, according to the correspondence, the blood glucose value in the time period corresponding to the feature is accurately regulated, including: accurately regulating the blood glucose value in that time period according to the change in the OR value of diabetic complications in patients caused by the change in the CGM feature in different time periods, so as to provide important support for reducing the occurrence of diabetic complications and realizing the accurate management of diabetic complications.

[0016] Second aspect, the embodiments of the present application further provide a diabetes complication risk assessment system, which performs risk assessment based on the diabetes complication risk assessment method described in the above embodiments, including: a data acquisition and preprocessing module, a data partitioning module, a feature extraction module, a training module, an analysis module, and a regulation module that are connected in sequence; the data acquisition and preprocessing module is used to acquire CGM data and preprocess the CGM data; the data partitioning module is used to partition the preprocessed CGM data into CGM data of different time periods; the feature extraction module is used to extract features from the CGM data of each time period; the training module is used to use the extracted features as the input of a machine learning algorithm, and the output is the risk of diabetes complications, and train using a machine learning algorithm; after training is completed, analyze the influence of the extracted features on diabetes complications to obtain important features for evaluating diabetes complications; the analysis module is used to use a logistic regression model to calculate the correspondence between the changes in important features in different time periods and the changes in the complication risk of diabetes patients, and refine the influence of important features on diabetes complications; the regulation module is used to accurately regulate the blood glucose value of the time period corresponding to the feature according to the correspondence to achieve precise management of diabetes complications.

[0017] The technical solutions provided by the embodiments of the present application have at least the following advantages:

[0018] The embodiments of the present application provide a diabetes complication risk assessment method and system. The method includes the following steps: First, acquire CGM data and preprocess the CGM data; then, partition the preprocessed CGM data into CGM data of different time periods; next, extract features from the CGM data of each time period; then, use the extracted features as the input of a machine learning algorithm, and the output is the risk of diabetes complications, and train using a machine learning algorithm; after training is completed, analyze the influence of the extracted features on diabetes complications to obtain important features for evaluating diabetes complications; next, use a logistic regression model to calculate the correspondence between the changes in important features in different time periods and the changes in the complication risk of diabetes patients, and refine the influence of important features on diabetes complications; finally, accurately regulate the blood glucose value of the time period corresponding to the feature according to the correspondence to achieve precise management of diabetes complications. The present application proposes a new method for processing CGM data. By partitioning the CGM data within a day into different time periods and analyzing the influence of the changes in the CGM data of each time period on diabetes complications, it provides support for achieving precise management of diabetes complications. Description of the Drawings

[0019] One or more embodiments are exemplarily illustrated by the pictures in the corresponding drawings. These exemplary illustrations do not constitute limitations on the embodiments. Unless otherwise stated, the figures in the drawings do not constitute a proportional limitation.

[0020] Figure 1 The flowchart of a method for assessing the risk of diabetic complications provided by an embodiment of the present application.

[0021] Figure 2 The module structure diagram of a system for assessing the risk of diabetic complications provided by an embodiment of the present application. Detailed implementation manners

[0022] As can be seen from the background art, the existing indicators for calculating CGM data are mainly calculated from the CGM data within one day, and it is not clear which time periods within one day have the greatest impact on diabetic complications in terms of blood glucose changes.

[0023] To solve the above technical problems, the present application provides a method and system for assessing the risk of diabetic complications. The method includes the following steps: First, obtain CGM data and preprocess the CGM data; then, divide the preprocessed CGM data into CGM data for different time periods; next, extract features from the CGM data for each time period; then, use the extracted features as the input of a machine learning algorithm, and the output is the risk of diabetic complications, and use the machine learning algorithm for training; after the training is completed, analyze the impact of the extracted features on diabetic complications to obtain the important features for assessing diabetic complications; next, use a logistic regression model to calculate the corresponding relationship between the changes in the important features for different time periods and the changes in the complication risk of diabetic patients, and refine the impact of the important features on diabetic complications; finally, according to the corresponding relationship, accurately regulate the blood glucose value of the time period corresponding to the feature to achieve precise management of diabetic complications. The present application provides a new method for processing CGM data. By dividing the CGM data within one day into different time periods and analyzing the impact of the changes in the CGM data for each time period on diabetic complications, it provides support for achieving precise management of diabetic complications.

[0024] The following will elaborate on each embodiment of the present application in conjunction with the accompanying drawings. However, those of ordinary skill in the art can understand that in each embodiment of the present application, many technical details are provided for the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in the present application can still be implemented.

[0025] Refer to Figure 1 , an embodiment of the present application provides a method for assessing the risk of diabetic complications, including the following steps:

[0026] Step S1: Obtain CGM data and preprocess the CGM data.

[0027] Step S2: Divide the preprocessed CGM data into CGM data for different time periods.

[0028] Step S3: Extract features from the CGM data for each time period.

[0029] Step S4: Use the extracted features as the input of a machine learning algorithm, and the output is the risk of diabetic complications, and train using the machine learning algorithm.

[0030] Step S5: After training, analyze the impact of the extracted features on diabetic complications to obtain important features for evaluating diabetic complications.

[0031] Step S6: Use a logistic regression model to calculate the corresponding relationship between the changes in important features in different time periods and the changes in the risk of diabetic complications in diabetic patients, and refine the impact of important features on diabetic complications.

[0032] Step S7: According to the corresponding relationship, accurately regulate the blood glucose value in the time period corresponding to the feature to achieve precise management of diabetic complications.

[0033] In some embodiments, the preprocessing in Step S1 includes abnormal data processing and missing data processing; among them, abnormal data processing includes: if a certain CGM value shows an abnormal phenomenon of being significantly larger or smaller compared with the adjacent CGM value, then replace the abnormal value with the average value of the two adjacent CGM values; missing data processing includes: if a certain CGM value is missing due to a temporary failure of the CGM device, then use the K-nearest neighbor value-taking method, and according to the K nearest samples of the missing data sample, weight and average the K values to estimate the missing data of the sample to complete data supplementation; the CGM data after preprocessing can be expressed as follows:

[0034] database={G1,G2,G3,G4,G5....,G N}

[0035] Among them, G1 represents the first value of the CGM, G2 represents the second value of the CGM, and so on.

[0036] It should be noted that when the blood glucose monitoring frequency of the CGM device is 5 minutes, 12 blood glucose values can be obtained within 1 hour, and 288 blood glucose values can be obtained within 1 day (24 hours).

[0037] In some embodiments, in step S2, the preprocessed CGM data is divided into CGM data for different time periods, including: according to the diet and sleep time of the diabetic patient, the preprocessed CGM data is divided into: CGM data for the fasting - breakfast blood glucose time period, CGM data for the lunch blood glucose time period, CGM data for the dinner blood glucose time period, CGM data for the bedtime blood glucose time period, and CGM data for the night - time blood glucose time period.

[0038] Among them, according to the diet and sleep time of the diabetic patient, the CGM data of one day is divided into multiple time periods, such as 06:00 - 11:00, 11:00 - 16:00, 16:00 - 20:00, 20:00 - 01:00, 01:00 - 06:00, which respectively correspond to the fasting - breakfast blood glucose time period (T1), the lunch blood glucose time period (T2), the dinner blood glucose time period (T3), the bedtime blood glucose time period (T4), and the night - time blood glucose time period (T5). The above - mentioned each time period can be corrected according to the actual situation.

[0039] In some embodiments, in step S3, feature extraction is performed on the CGM data for each time period, and the extracted features include: mean value, standard deviation, variance, coefficient of variation, blood glucose fluctuation range value, blood glucose fluctuation change value, mean absolute difference of blood glucose, mean gradient change value of blood glucose, J - index, hypoglycemia index, hyperglycemia index, and blood glucose target range index.

[0040] In some embodiments, the coefficient of variation is calculated by dividing the standard deviation by the mean value; the blood glucose fluctuation range value is calculated by subtracting the minimum value from the maximum value of the CGM data; the blood glucose fluctuation change value is calculated by dividing the blood glucose fluctuation range value by the mean value; the mean absolute difference of blood glucose is obtained by first calculating the sum of the absolute values of all adjacent CGM differences and then calculating the mean value; the mean gradient change value of blood glucose is obtained by first calculating the sum of all adjacent CGM gradient values and then calculating the mean value; the J - index is obtained by calculating the square of the sum of the mean value and the standard deviation; the hypoglycemia index is obtained by first performing statistical transformation on each CGM data, calculating the hypoglycemia risk according to the transformation result, and then calculating the mean value of all hypoglycemia risks within this time period; the hyperglycemia index is obtained by first performing statistical transformation on each CGM data, then calculating the hyperglycemia risk according to the transformation result, and then calculating the mean value of all hyperglycemia risks within this time period; the blood glucose target range index is obtained by calculating the ratio of the number of blood glucose values within the target range of 3.9 mmol / L to 10 mmol / L to the total number of blood glucose values in the CGM data for each time period.

[0041] In some embodiments, in step S4, multiple machine learning algorithms are used for training. The machine learning algorithms include random forest, decision tree, and support vector machine. The extracted features are used as the input of the machine learning algorithms, and the output is the risk of diabetic complications for training. After the training is completed, based on the SHAP (SHapley Additive exPlanations) analysis method, the marginal contribution degrees of all features are visualized and sorted according to the contribution degrees of the features. The top 10% of the features are selected as the important features for evaluating diabetic complications, and the time periods corresponding to the top 10% of the features are determined.

[0042] It should be noted that after obtaining the top 10% of the features, by analyzing the time periods corresponding to the top 10% of the features, it is possible to clarify which time periods of blood glucose changes have the greatest impact on diabetic complications, and thus derive similar conclusions such as that the average blood glucose value in the nighttime blood glucose time period (T5) has a great impact on diabetic complications, and the average absolute difference in blood glucose in the pre-bedtime blood glucose time period (T4) has a great impact on diabetic complications.

[0043] In some embodiments, in step S5, a logistic regression model is used to calculate the corresponding relationship between the changes in important features in different time periods and the changes in the risk of diabetic complications in patients, including: using the logistic regression model to calculate the OR value (odds ratio) and 95% confidence interval of diabetic complications in patients for the top 10% of the features respectively, so as to determine the proportion of increase or decrease in the risk of diabetic complications in patients when each feature increases by 10%.

[0044] In some embodiments, in step S6, according to the corresponding relationship, the blood glucose values in the time periods corresponding to the features are accurately regulated, including: accurately regulating the blood glucose values in the time periods according to the changes in the OR values of diabetic complications in patients caused by the changes in CGM features in different time periods, so as to provide important support for reducing the occurrence of diabetic complications and realizing the precise management of diabetic complications.

[0045] See Figure 2, an embodiment of the present application also provides a diabetes complication risk assessment system, which performs risk assessment based on the diabetes complication risk assessment method described in the above embodiment, including: a data acquisition and preprocessing module 101, a data division module 102, a feature extraction module 103, a training module 104, an analysis module 105, and a regulation module 106 that are connected in sequence; the data acquisition and preprocessing module 101 is used to acquire CGM data and preprocess the CGM data; the data division module 102 is used to divide the preprocessed CGM data into CGM data for different time periods; the feature extraction module 103 is used to extract features from the CGM data for each time period; the training module 104 is used to use the extracted features as the input of a machine learning algorithm, and the output is the risk of diabetes complications, and training is performed using a machine learning algorithm; after training is completed, analyze the impact of the extracted features on diabetes complications to obtain important features for evaluating diabetes complications; the analysis module 105 is used to use a logistic regression model to calculate the corresponding relationship between the changes in important features for different time periods and the changes in the complication risk of diabetes patients, and refine the impact of important features on diabetes complications; the regulation module 106 is used to accurately regulate the blood glucose value for the time period corresponding to the feature according to the corresponding relationship to achieve precise management of diabetes complications.

[0046] With the above technical solutions, an embodiment of the present application provides a diabetes complication risk assessment method and system. The method includes the following steps: First, acquire CGM data and preprocess the CGM data; then, divide the preprocessed CGM data into CGM data for different time periods; next, extract features from the CGM data for each time period; then, use the extracted features as the input of a machine learning algorithm, and the output is the risk of diabetes complications, and training is performed using a machine learning algorithm; after training is completed, analyze the impact of the extracted features on diabetes complications to obtain important features for evaluating diabetes complications; next, use a logistic regression model to calculate the corresponding relationship between the changes in important features for different time periods and the changes in the complication risk of diabetes patients, and refine the impact of important features on diabetes complications; finally, according to the corresponding relationship, accurately regulate the blood glucose value for the time period corresponding to the feature to achieve precise management of diabetes complications. The present application proposes a new method for processing CGM data. By dividing the CGM data within a day into different time periods and analyzing the impact of the changes in the CGM data for each time period on diabetes complications, it provides support for achieving precise management of diabetes complications.

[0047] Those of ordinary skill in the art can understand that the above embodiments are specific examples for implementing the present application. In actual applications, various changes can be made in form and details without departing from the spirit and scope of the present application. Any person skilled in the art can make their respective changes and modifications without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application should be subject to the scope defined by the claims.

Claims

1. A method for assessing the risk of diabetic complications, characterized in that: The following steps are involved: Acquire CGM data and pre-process the CGM data; Dividing the preprocessed CGM data into CGM data of different time periods; Extract features from CGM data in each time period; The extracted features are used as input to the machine learning algorithm, and the output is the risk of diabetes complications, which is trained using the machine learning algorithm; After the training is completed, the impact of the extracted features on diabetic complications is analyzed to obtain important features for evaluating diabetic complications; A logistic regression model was used to calculate the corresponding relationship between the changes in important characteristics in different time periods and the changes in the risk of complications in diabetic patients, and to refine the impact of important characteristics on diabetic complications; According to the corresponding relationship, the blood sugar level in the time period corresponding to the characteristics is accurately regulated to achieve accurate management of diabetic complications.

2. The method for assessing the risk of diabetes complications according to claim 1, characterized in that: The preprocessing includes abnormal data processing and missing data processing; The abnormal data processing includes: if a CGM value is significantly larger or smaller than an adjacent CGM value, the abnormal value is replaced by the average of the two adjacent CGM values; The missing data processing includes: if a CGM value is missing due to a temporary failure of the CGM device, the K-nearest neighbor method is used to estimate the missing data of the sample by weighted averaging the K values ​​according to the K samples closest to the missing data sample, so as to supplement the data; the CGM data after preprocessing can be expressed as follows: database={G1,G2,G3,G4,G5....,G N } Among them, G1 represents the first value of CGM, G2 represents the second value of CGM, and so on.

3. The method for assessing the risk of diabetes complications according to claim 1, characterized in that: The preprocessed CGM data is divided into CGM data of different time periods, including: According to the diet and sleep time of diabetic patients, the preprocessed CGM data is divided into: CGM data of fasting-breakfast blood sugar time period, CGM data of lunch blood sugar time period, CGM data of dinner blood sugar time period, CGM data of bedtime blood sugar time period, and CGM data of nighttime blood sugar time period.

4. The method for assessing the risk of diabetes complications according to claim 3, characterized in that: The fasting-breakfast blood sugar time period is 06:00-11:00; the lunch blood sugar time period is 11:00-16:00; the dinner blood sugar time period is 16:00-20:00; the bedtime blood sugar time period is 20:00-01:00; and the night blood sugar time period is 01:00-06:

00.

5. The method for assessing the risk of diabetes complications according to claim 1, characterized in that: Feature extraction is performed on the CGM data of each time period, and the extracted features include: mean, standard deviation, variance, coefficient of variation, blood glucose fluctuation range value, blood glucose fluctuation change value, average blood glucose absolute difference value, average blood glucose gradient change value, J index, hypoglycemia index, hyperglycemia index and blood glucose target range index.

6. The method for assessing the risk of diabetes complications according to claim 5, characterized in that: The coefficient of variation is calculated by dividing the standard deviation by the mean; The blood sugar fluctuation range is calculated by subtracting the minimum value from the maximum value of the CGM data; The blood sugar fluctuation change value is calculated by dividing the blood sugar fluctuation range value by the average value; The blood glucose average absolute difference is obtained by first calculating the sum of the absolute values ​​of all two adjacent CGM differences and then calculating the average value; The average blood glucose gradient change value is obtained by first calculating the sum of all two adjacent CGM gradient values ​​and then calculating the average value; The J index is obtained by calculating the square of the sum of the mean and the standard deviation; The hypoglycemia index is obtained by first performing statistical transformation on each CGM data, calculating the hypoglycemia risk according to the transformation result, and then calculating the average value of all hypoglycemia risks in the time period; The hyperglycemia index is obtained by first performing statistical transformation on each CGM data, then calculating the hyperglycemia risk according to the transformation result, and then calculating the average value of all hyperglycemia risks in the time period; The blood glucose target range index is obtained by calculating the ratio of the number of blood glucose values ​​within the target range of 3.9 mmol / L to 10 mmol / L to the total blood glucose value number in the CGM data of each time period.

7. The method for assessing the risk of diabetes complications according to claim 1, characterized in that: A plurality of machine learning algorithms are used for training, wherein the machine learning algorithms include random forest, decision tree and support vector machine; the extracted features are used as input of the machine learning algorithm, and the output is the risk of diabetic complications for training; After training, the marginal contributions of all features were visualized based on the SHAP analysis method, and the features were sorted according to their contribution. The top 10% of features were selected as important features for evaluating diabetic complications, and the time periods corresponding to the top 10% of features were determined.

8. The method for assessing the risk of diabetes complications according to claim 1, characterized in that: Logistic regression models were used to calculate the relationship between changes in important characteristics at different time periods and changes in the risk of complications in patients with diabetes, including: A logistic regression model was used to calculate the OR values ​​and 95% confidence intervals of diabetic complications for the top 10% of the characteristics, so as to determine the proportion of increased or decreased risk of diabetic complications in patients when each characteristic increased by 10%.

9. The method for assessing the risk of diabetes complications according to claim 1, characterized in that: According to the corresponding relationship, the blood glucose level in the time period corresponding to the characteristic is accurately regulated, including: According to the changes in the OR value of complications of diabetic patients caused by the changes in CGM characteristics in different time periods, the blood sugar level in this time period is accurately regulated, thereby providing important support for reducing the occurrence of diabetic complications and achieving accurate management of diabetic complications.

10. A diabetes complication risk assessment system, which performs risk assessment based on the diabetes complication risk assessment method according to any one of claims 1 to 9, characterized in that: include: A data acquisition and preprocessing module, a data partitioning module, a feature extraction module, a training module, an analysis module, and a control module connected in sequence; The data acquisition and preprocessing module is used to acquire CGM data and preprocess the CGM data; The data division module is used to divide the preprocessed CGM data into CGM data of different time periods; The feature extraction module is used to extract features from the CGM data of each time period; The training module is used to use the extracted features as input to the machine learning algorithm, and the output is the risk of diabetic complications, and the machine learning algorithm is used for training; After the training is completed, the impact of the extracted features on diabetic complications is analyzed to obtain important features for evaluating diabetic complications; The analysis module is used to use a logistic regression model to calculate the corresponding relationship between the changes in important features in different time periods and the changes in the risk of complications of diabetic patients, and to refine the impact of important features on diabetic complications; The control module is used to accurately control the blood glucose value in the time period corresponding to the feature according to the corresponding relationship, so as to achieve accurate management of diabetic complications.

Images