System and method for predicting risk of abnormal glucose metabolism

Abstract

This invention discloses a system and method for predicting the risk of abnormal glucose metabolism, relating to the field of medical data processing technology. The system includes: a data acquisition module that acquires continuous blood glucose monitoring data, postpartum weight retention, and clinical characteristic parameters of gestational diabetes in target subjects with a history of gestational diabetes within a set postpartum time window; a feature extraction module that extracts six dynamic blood glucose features: mean glucose value, glucose coefficient of variation, continuous overlap net blood glucose effect, time percentage within the target range, hyperglycemia index, and diurnal mean absolute difference; a risk prediction module that combines the above features, postpartum weight retention, and clinical characteristic parameters into a feature vector and inputs it into a machine learning classification model to obtain a risk level of abnormal glucose metabolism; and an early warning module that outputs risk grading results and referral prompts when the risk level meets set conditions. This invention improves the screening efficiency of postpartum abnormal glucose metabolism without increasing the burden of venous blood collection.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to a system and method for predicting the risk of abnormal glucose metabolism. Background Technology

[0002] Women with a history of gestational diabetes mellitus (GDM) have a higher risk of developing glucose metabolism abnormalities in the postpartum period, and some of these individuals will gradually progress to impaired glucose tolerance (IGT) or type 2 diabetes mellitus (T2DM).

[0003] Currently, the assessment of postpartum glucose metabolism in this population mainly relies on the oral glucose tolerance test (OGTT). This test requires subjects to fast and undergo multiple venous blood draws, which is cumbersome and uncomfortable. As a result, actual participation in this test is often limited in primary care and community settings, and some high-risk individuals are not identified in a timely manner.

[0004] There are two other types of auxiliary assessment methods in the current technology. The first is based on building statistical or machine learning prediction models based on clinical history data. This type of method lacks characterization of individual real-time glycemic dynamics information. The second is based on extracting glycemic fluctuation indicators from continuous glucose monitoring (CGM) data and combining them with machine learning for classification. This type of method usually extracts a large number of fluctuation indicators at once, and there is information redundancy between the indicators. It has relatively high requirements for computing resources and does not take into account the weight change factors specific to the postpartum period. Summary of the Invention

[0005] How can we provide a system and method for predicting the risk of abnormal glucose metabolism to solve the problems existing in current technologies?

[0006] According to one aspect of the present invention, a system for predicting the risk of abnormal glucose metabolism is provided, comprising a data acquisition module, a feature extraction module, a risk prediction module, and an early warning module. The data acquisition module is configured to acquire continuous blood glucose monitoring data of a target subject within a set postpartum time window, as well as the postpartum weight retention and clinical characteristic parameters of gestational diabetes mellitus of the target subject. The target subject is a postpartum individual who has previously suffered from gestational diabetes mellitus, and the postpartum weight retention is the difference between the target subject's current weight and its pre-pregnancy weight. The feature extraction module is configured to extract dynamic blood glucose features from the continuous blood glucose monitoring data. The dynamic blood glucose features include the average glucose value representing the baseline blood glucose level, the glucose coefficient of variation representing the relative fluctuation of blood glucose within the day, the continuous overlapping net blood glucose effect representing the inertia of blood glucose changes, the proportion of time within the target range representing the level of exposure to meet the standard, the hyperglycemia index representing the degree of hyperglycemia risk, and the daily average absolute difference representing the stability of blood glucose within the day. The continuous overlapping net blood glucose effect represents the dispersion of the difference between the glucose value at each sampling time and the glucose value at a set time before that sampling time. The proportion of time within the target range represents the ratio of the time during which the glucose value falls within the set target range to the total monitoring time. The risk prediction module is configured to combine the dynamic blood glucose characteristics, the postpartum weight retention, and the clinical characteristic parameters of gestational diabetes into a feature vector, and input the feature vector into a pre-trained machine learning classification model to obtain the risk degree of abnormal glucose metabolism of the target object. The early warning module is configured to output risk classification results and referral prompts in response to the risk level meeting the set early warning conditions.

[0007] In conjunction with the first aspect, in an optional implementation, the feature extraction module is further configured to perform smoothing filtering and outlier removal on the continuous blood glucose monitoring data before extracting the dynamic blood glucose features.

[0008] In conjunction with the first aspect, in one optional embodiment, the set duration is one hour, the continuous overlapping net blood glucose function is the standard deviation of the difference between the glucose value at each sampling time and the glucose value one hour before that sampling time; the glucose coefficient of variation is the ratio of the standard deviation of the continuous blood glucose monitoring data to the average glucose value; and the diurnal average absolute difference is the average of the absolute values ​​of the difference between the glucose values ​​at the same time on two adjacent days.

[0009] In conjunction with the first aspect, in one optional embodiment, the target range is set to 3.9 mmol / L to 10.0 mmol / L; the hyperglycemia index is an index characterizing the degree of hyperglycemia risk obtained by summing up glucose values ​​above a set glucose threshold after nonlinear conversion according to a set conversion coefficient.

[0010] In conjunction with the first aspect, in one optional implementation, the continuous blood glucose monitoring data refers to glucose data collected continuously from the target subject for no less than fourteen days within the set postpartum time window; the set postpartum time window is four to twelve weeks after delivery.

[0011] In conjunction with the first aspect, in an optional implementation, the risk prediction module is further configured to binarize the postpartum weight retention amount into a postpartum weight retention categorical variable according to a set weight threshold, and combine the postpartum weight retention categorical variable into the feature vector.

[0012] In conjunction with the first aspect, in one alternative implementation, the set weight threshold is 5 kg.

[0013] In conjunction with the first aspect, in one optional implementation, the machine learning classification model includes at least one of an extreme gradient boosting model, a random forest model, and a logistic regression model, or an ensemble learning model composed of the aforementioned models.

[0014] In conjunction with the first aspect, in an optional implementation, the early warning module is configured to determine that the set early warning condition is met in response to the risk level being greater than a set risk threshold; the risk classification result includes low risk, medium risk, and high risk; the referral prompt information includes information suggesting that the target subject be referred for a glucose tolerance test.

[0015] Secondly, the present invention provides a data processing method for predicting the risk of abnormal glucose metabolism, comprising: Continuous blood glucose monitoring data of the target subject within a set postpartum time window, as well as the postpartum weight retention and gestational diabetes clinical characteristic parameters of the target subject, wherein the target subject is a postpartum individual who has previously suffered from gestational diabetes, and the postpartum weight retention is the difference between the target subject's current weight and its pre-pregnancy weight. Dynamic blood glucose features are extracted from the continuous blood glucose monitoring data. These dynamic blood glucose features include average glucose value, glucose coefficient of variation, continuous overlap net blood glucose effect, time percentage within the target range, hyperglycemia index, and diurnal average absolute difference. The continuous overlap net blood glucose effect characterizes the degree of dispersion of the difference between the glucose value at each sampling time and the glucose value at a set time before that sampling time. The dynamic blood glucose characteristics, postpartum weight retention, and clinical characteristic parameters of gestational diabetes are combined into a feature vector, and the feature vector is input into a pre-trained machine learning classification model to obtain the risk of abnormal glucose metabolism of the target object. In response to the risk level meeting the set early warning conditions, the risk classification result and referral prompt information are output.

[0016] In conjunction with the second aspect, in an optional implementation, before extracting the dynamic blood glucose features, the continuous blood glucose monitoring data is further subjected to smoothing filtering and outlier removal.

[0017] In conjunction with the second aspect, in an optional implementation, the set duration is one hour, the continuous overlapping net blood glucose function is the standard deviation of the difference between the glucose value at each sampling time and the glucose value one hour before that sampling time; the glucose coefficient of variation is the ratio of the standard deviation of the continuous blood glucose monitoring data to the average glucose value; the diurnal average absolute difference is the average of the absolute values ​​of the difference between the glucose values ​​at the same time on two adjacent days; and the target range for the time percentage within the target range is set to a target range of 3.9 mmol / L to 10.0 mmol / L.

[0018] In conjunction with the second aspect, in an optional implementation, the postpartum weight retention is binarized according to a set weight threshold and combined into the feature vector, wherein the set weight threshold is 5 kg.

[0019] In conjunction with the second aspect, in an optional implementation, the machine learning classification model includes at least one of an extreme gradient boosting model, a random forest model, and a logistic regression model, or an ensemble learning model composed of the aforementioned models.

[0020] According to another aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the processor executes a computer program on the memory to implement the above-described data processing method; and a computer-readable storage medium thereon, wherein a computer program stored thereon is executed by a processor to implement the above-described data processing method.

[0021] The embodiments of this invention can achieve the following beneficial effects: First, data collection can be completed at home and in the community environment by means of wearable continuous glucose monitoring and routine weight measurement without fasting and venous blood sampling, which is conducive to improving the participation rate of postpartum screening; Second, the six dynamic blood glucose features used respectively characterize the dimensions with low correlation such as baseline level, intraday fluctuation, change inertia, target exposure, hyperglycemia risk and intraday stability, with low information redundancy among them, and the consumption of computing resources is reduced at the same data scale compared with the scheme using dozens of derived indicators; Third, dynamic indicators such as continuous overlapping net blood glucose effect and intraday mean absolute difference may change before the average glucose value and the time proportion within the target range have changed significantly, which helps to provide early risk warning of postpartum glucose metabolism abnormalities; Fourth, in the verification of one embodiment, the model using the combination of six dynamic blood glucose features and postpartum weight retention has improved indicators such as the area under the receiver operating characteristic curve compared with the model using only clinical features and the ablation model that removes relevant features. Attached Figure Description

[0022] The accompanying drawings, which form part of this specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention. The invention will become more clearly understood from the following detailed description, referring to the accompanying drawings, in which: Figure 1 This is a schematic diagram of the structure of the glucose metabolism abnormality risk prediction system provided in an embodiment of the present invention; Figure 2 A flowchart illustrating the data processing method for predicting the risk of abnormal glucose metabolism provided in an embodiment of the present invention; Figure 3 This is a time-series diagram illustrating the joint screening and referral process between the community and the hospital, as provided in an embodiment of the present invention. Detailed Implementation

[0023] The present application will be further described below with reference to specific embodiments and accompanying drawings. It is understood that the illustrative embodiments of this disclosure include, but are not limited to, related methods, devices, and systems. The specific embodiments described herein are merely for explaining the present application and not for limiting it. Furthermore, for ease of description, the accompanying drawings show only the parts relevant to the present application, and not all of the structures or processes.

[0024] The following specific embodiments illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Although the description of this application will be presented in conjunction with preferred embodiments, this does not mean that the features of this invention are limited to this embodiment. On the contrary, the purpose of describing the invention in conjunction with embodiments is to cover other options or modifications that may be derived based on the claims of this application. To provide a thorough understanding of this application, many specific details will be included in the following description. This application may also be implemented without using these details. Furthermore, to avoid confusion or obscuring the focus of this application, some specific details will be omitted in the description. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0025] See Figure 1 The present invention provides a glucose metabolism abnormality risk prediction system 100, including a data acquisition module 10, a feature extraction module 20, a risk prediction module 30 and an early warning module 40.

[0026] The data acquisition module 10 is configured to acquire continuous blood glucose monitoring data of the target subject within a set postpartum time window, as well as the target subject's postpartum weight retention and clinical characteristic parameters of gestational diabetes; wherein, the postpartum weight retention is the difference between the target subject's current weight and its pre-pregnancy weight. The reason why the difference between the current weight and the pre-pregnancy weight is used for the postpartum weight retention, rather than the current weight itself, is because this difference directly characterizes the weight load that has not subsided after delivery and is more related to insulin resistance and glucose metabolism load in the postpartum stage. Optionally, continuous blood glucose monitoring data is collected by a continuous blood glucose monitor 11 at a set sampling interval, such as 5 to 15 minutes. The continuous blood glucose monitoring data consists of glucose data collected continuously for at least fourteen days within a set postpartum time window for the target subject. Collecting data for at least fourteen days helps to cover differences in diet and rest between weekdays and rest days, making subsequent characteristics more robust. The set postpartum time window can be four to twelve weeks after delivery, which coincides with the time window for postpartum glucose metabolism re-examination. In other optional embodiments, the set postpartum time window can also be set to other periods within one year postpartum. Postpartum weight retention can be obtained by subtracting the current weight measured by the weight measuring device 12 from the pre-recorded pre-pregnancy weight. Optionally, clinical characteristic parameters of gestational diabetes are entered by the medical history entry unit 13, including at least one of the following: gestational blood glucose control level, whether insulin therapy was received during pregnancy, and gestational week of diagnosis of gestational diabetes. Using these parameters, rather than simply whether or not the condition is present, can maintain discriminatory power within the target population who all have a history of gestational diabetes.

[0027] The feature extraction module 20 is configured to extract dynamic blood glucose features from continuous blood glucose monitoring data. Optionally, the feature extraction module 20 includes a preprocessing unit 21 and a feature calculation unit 22. Before extracting dynamic blood glucose features, the preprocessing unit 21 performs smoothing filtering and outlier removal on the continuous blood glucose monitoring data; smoothing filtering, for example, uses moving average to suppress sensor noise, and outlier removal, for example, is based on interquartile range (ICM) to determine and remove outliers, thereby reducing the interference of acquisition noise on subsequent feature extraction.

[0028] The dynamic blood glucose features extracted by feature calculation unit 22 include the following six items: average glucose value, which is the average glucose value within the monitoring period, representing the baseline blood glucose level; glucose coefficient of variation, which is the ratio of the standard deviation of continuous blood glucose monitoring data to the average glucose value, representing the relative fluctuation of blood glucose within the day. Since it is normalized to the mean, it can reflect the relative fluctuation amplitude after removing the influence of the baseline level; continuous overlap net blood glucose effect, which is the dispersion of the difference between the glucose value at each sampling time and the glucose value at a set time before that sampling time, representing the inertia of blood glucose change; time proportion within the target range, which is the ratio of the time when the glucose value falls within the set target range to the total monitoring time, representing the degree of exposure to the target; hyperglycemia index, which is an index representing the degree of hyperglycemia risk obtained by accumulating glucose values ​​that are higher than the set glucose threshold after nonlinear transformation according to the set conversion coefficient. It amplifies the risk weight of the hyperglycemia segment through nonlinear transformation and does not overlap with the hypoglycemia segment; and the average absolute difference between daytime glucose values, which is the average of the absolute values ​​of the difference between glucose values ​​at the same time on two adjacent days, representing the stability of blood glucose during the day and reflecting the metabolic buffering capacity across days.

[0029] In one embodiment, the above-mentioned dynamic blood glucose characteristics can be calculated as follows: the average glucose value is the arithmetic mean of all glucose sampling values ​​within the monitoring period; the glucose coefficient of variation is the ratio of the standard deviation of glucose values ​​within the period to the average glucose value, and can be expressed as a percentage; the continuous overlapping net blood glucose effect is the standard deviation of the difference sequence formed by the difference between the glucose value at each sampling time and the glucose value at a previously set time interval; the target range time percentage is the ratio of the number of sampling points whose glucose values ​​fall within the set target range to the total number of sampling points; the hyperglycemia index is obtained by summing the glucose values ​​above the set glucose threshold after logarithmic nonlinear transformation and multiplying by a set coefficient; the diurnal average absolute difference is the average of the absolute value of the difference between glucose values ​​at the same time on two adjacent days over all corresponding times.

[0030] For continuous overlapping net glucose monitoring, in one optional embodiment, the set duration is one hour, and the dispersion is the standard deviation, i.e., the standard deviation of the difference between the glucose value at each sampling time and the glucose value one hour prior. In other optional embodiments, the set duration is two or four hours, and the dispersion can also be variance or interquartile range. Continuous overlapping net glucose monitoring characterizes the rate of change of blood glucose on a short timescale, which can be correlated with the tendency of delayed early-phase insulin secretion, thus helping to reflect early changes in regulatory capacity before baseline levels have changed. Regarding the time percentage within the target range, in one optional embodiment, the target range is set to 3.9 mmol / L to 10.0 mmol / L, and in other optional embodiments, it can be set separately according to screening needs. The above six dynamic blood glucose characteristics characterize dimensions with low correlation and low information redundancy among them, thus maintaining the amount of information while keeping the number of features small, reducing the requirements for the amount of collected data and computing resources.

[0031] The risk prediction module 30 is configured to combine the aforementioned six dynamic blood glucose characteristics, postpartum weight retention, and gestational diabetes clinical characteristic parameters into a feature vector, and input the feature vector into a pre-trained machine learning classification model to obtain the risk level of abnormal glucose metabolism in the target object. Optionally, the risk prediction module 30 includes a vector construction unit 31 and a classification model 32. The vector construction unit 31 binarizes postpartum weight retention into a postpartum weight retention classification variable according to a set weight threshold, and then combines this classification variable with the six dynamic blood glucose characteristics and gestational diabetes clinical characteristic parameters into a feature vector; in an optional embodiment, the set weight threshold is 5 kg. The classification model 32 is a pre-trained machine learning classification model, which may optionally include at least one of extreme gradient boosting model, random forest model, and logistic regression model, or an ensemble learning model composed of the aforementioned models; during training, the glucose tolerance test results obtained from follow-up are used as labels, and the samples are divided into training set, validation set, and test set for cross-validation. The classification model 32 outputs a risk score for abnormal glucose metabolism, such as the probability value of the target subject developing impaired glucose tolerance or type 2 diabetes.

[0032] The early warning module 40 is configured to output a risk grading result and referral prompt information in response to a risk level meeting a set early warning condition. Optionally, the early warning module 40 includes a risk grading unit 41 and a referral prompt unit 42. The risk grading unit 41 determines that the set early warning condition is met in response to a risk level greater than a set risk threshold, and classifies the risk grading result into low risk, medium risk, and high risk. The referral prompt unit 42 generates referral prompt information when the set early warning condition is met, the referral prompt information including information suggesting that the target patient be referred to a higher-level medical institution for a glucose tolerance test.

[0033] It should be noted that the above-mentioned risk levels and risk classifications for abnormal glucose metabolism are intermediate results for reference by community or primary healthcare workers, used for screening and triage management, and do not constitute a diagnosis of the disease; whether the target individuals are indeed suffering from abnormal glucose metabolism needs to be confirmed by higher-level medical institutions through glucose tolerance tests, etc.

[0034] In one optional implementation, when constructing the feature vector, the risk prediction module, in addition to listing the dynamic blood glucose features, postpartum weight retention, and gestational diabetes clinical feature parameters in parallel, also constructs interactive features that reflect the synergistic effect of the two, such as the product of postpartum weight retention and the time percentage within the target range, and the product of postpartum weight retention and the glucose variation coefficient. Compared to simply inputting each feature in parallel, introducing interactive features helps to characterize the synergistic effect between weight load and blood glucose fluctuations.

[0035] In one optional implementation, the data acquisition module determines the validity of continuous blood glucose monitoring data. The determination criteria include an effective wearing time of no less than a set number of days, a daily effective sampling point ratio of no less than a set threshold, and a continuous missing period of no more than a set duration. When the conditions are not met, a prompt to re-wear the device or extend the collection period is generated, and short-term missing data is filled by interpolation to improve the reliability of home-collected data.

[0036] In one alternative implementation, the system employs a stratified dual-channel screening: the first channel performs basic risk stratification based solely on clinical characteristic parameters of gestational diabetes and postpartum weight retention, suitable for target subjects who have not yet worn continuous glucose monitoring devices; for target subjects whose basic risk reaches a set level, the second channel collects continuous glucose monitoring data and extracts the dynamic glucose characteristics for fine screening, and the results of the two channels are merged to output the final risk score, thereby achieving a balance between convenience and accuracy.

[0037] In one optional implementation, the risk prediction module outputs the risk level of abnormal glucose metabolism in the target subject within multiple set time windows such as one year, three years, and five years postpartum, based on the same set of inputs or inputs from different follow-up nodes. The risk level is dynamically updated as the postpartum period extends, and different levels of follow-up and intervention prompts are provided for different time windows.

[0038] In one optional implementation, the early warning module outputs the risk level and, based on the degree of contribution of each feature to the risk level, outputs the main contributing features, such as indicating high postpartum weight retention, low time percentage within the target range, or high glucose variation coefficient, to help medical staff understand the source of the risk; and generates graded health management recommendations based on the risk grading results: for low-risk individuals, it recommends routine follow-up; for medium-risk individuals, it recommends weight control, adjusting diet and exercise, and retesting continuous blood glucose monitoring after a set time; and for high-risk individuals, it recommends immediate referral to a higher-level medical institution for a glucose tolerance test.

[0039] See Figure 2 Based on the same technical concept as the above system, this invention also provides a data processing method for predicting the risk of abnormal glucose metabolism, comprising: step S100, acquiring continuous blood glucose monitoring data of the target subject within a set postpartum time window, as well as the postpartum weight retention and gestational diabetes clinical characteristic parameters of the target subject; step S200, performing smoothing filtering and outlier removal on the continuous blood glucose monitoring data; step S300, extracting six dynamic blood glucose features: mean glucose value, glucose coefficient of variation, continuous overlapping net blood glucose effect, time proportion within the target range, hyperglycemia index, and daily mean absolute difference; step S400, binarizing the postpartum weight retention and combining it with the above features and gestational diabetes clinical characteristic parameters to form a feature vector; step S500, inputting the feature vector into a pre-trained machine learning classification model and outputting a risk level of abnormal glucose metabolism; step S600, determining whether the risk level meets the set warning conditions; if it does, executing step S700 to generate risk grading results and referral prompts; if it does not meet the conditions, executing step S800 to output routine follow-up prompts. Optionally, in step S300, the continuous overlapping net blood glucose effect is calculated as the standard deviation of the difference between the glucose value at each sampling time and the glucose value one hour prior; the glucose coefficient of variation is the ratio of the standard deviation to the average glucose value; the diurnal mean absolute difference is the average of the absolute values ​​of the differences between the glucose values ​​at the same time on two adjacent days; and the target range for the time percentage within the target range is set to 3.9 mmol / L to 10.0 mmol / L. In step S400, postpartum weight retention is binarized according to a set weight threshold, which is 5 kg. The aforementioned machine learning classification model may optionally include at least one of extreme gradient boosting model, random forest model, and logistic regression model, or an ensemble learning model composed of the aforementioned models. The specific implementation of each step of this data processing method is consistent with the corresponding module of the aforementioned system.

[0040] See Figure 3In the application scenarios of closely integrated medical consortia or medical communities, community health service centers collect and upload continuous blood glucose monitoring and weight data. The cloud-based risk prediction system runs a prediction model to calculate the risk level and determine the risk grade. Those identified as high-risk are referred by the endocrinology department of a higher-level hospital, undergo glucose tolerance tests for diagnosis, and are then subject to tiered intervention and management. The management results are fed back to the community for follow-up management, thus forming a closed loop of screening and referral.

[0041] In one alternative implementation, feature extraction can employ alternative or complementary schemes. For example, sample entropy can be used to replace or supplement the average glucose value to characterize the complexity of the blood glucose time series, or the average glucose value during the nighttime period can be introduced to characterize the baseline blood glucose level. In another alternative implementation, risk prediction can employ a Long Short-Term Memory network or a temporal convolutional network, using continuous blood glucose monitoring data and postpartum weight retention as inputs for end-to-end representation learning and outputting a risk score. In terms of deployment, the above data processing can be distributed to communities via mobile terminal applications, or it can be embedded as a backend service in hospital information systems or public health chronic disease management platforms.

[0042] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the aforementioned data processing method for predicting the risk of abnormal glucose metabolism. The present invention also provides a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the aforementioned data processing method for predicting the risk of abnormal glucose metabolism; the computer program can be divided into program modules corresponding to the functions of the aforementioned modules.

[0043] In one example of validation, glucose tolerance test results obtained during follow-up were used as labels to ablate and control glucose metabolism abnormality prediction models constructed with different feature combinations. The area under the receiver operating characteristic curve (AUC, including 95% confidence interval), sensitivity, and specificity of each model are shown in the table below: As shown in the table above, when the complete combination of gestational diabetes history, postpartum weight retention, and six dynamic blood glucose characteristics is used, the area under the receiver operating characteristic curve reaches 0.809 and the sensitivity reaches 90.9%, which is an improvement of about 0.087 compared with the baseline model using only gestational diabetes history (0.722), and is higher than any single or two-characteristic combination. This indicates that the combination of gestational diabetes history, postpartum weight retention, and six dynamic blood glucose characteristics has a positive effect on the prediction of postpartum glucose metabolism abnormality risk, and the three complement each other.

[0044] The methods and systems of the present invention may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. Furthermore, in some embodiments, the invention may also be implemented as a program recorded on a recording medium, the program comprising machine-readable instructions for implementing the methods according to the invention. Therefore, the invention also covers recording media storing programs for performing the methods according to the invention.

[0045] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system for predicting the risk of abnormal glucose metabolism, characterized in that, It includes a data acquisition module, a feature extraction module, a risk prediction module, and an early warning module: The data acquisition module is configured to acquire continuous blood glucose monitoring data of the target object within a set postpartum time window, as well as the postpartum weight retention and gestational diabetes clinical characteristic parameters of the target object; the target object is a postpartum individual who has previously suffered from gestational diabetes, and the postpartum weight retention is the difference between the target object's current weight and its pre-pregnancy weight; The feature extraction module is configured to extract dynamic blood glucose features from the continuous blood glucose monitoring data. The dynamic blood glucose features include the average glucose value representing the baseline blood glucose level, the glucose coefficient of variation representing the relative fluctuation of blood glucose within the day, the continuous overlapping net blood glucose effect representing the inertia of blood glucose changes, the proportion of time within the target range representing the level of exposure to meet the standard, the hyperglycemia index representing the degree of hyperglycemia risk, and the daily average absolute difference representing the stability of blood glucose within the day. The continuous overlapping net blood glucose effect represents the dispersion of the difference between the glucose value at each sampling time and the glucose value at a set time before that sampling time. The proportion of time within the target range represents the ratio of the time during which the glucose value falls within the set target range to the total monitoring time. The risk prediction module is configured to combine the dynamic blood glucose characteristics, the postpartum weight retention, and the clinical characteristic parameters of gestational diabetes into a feature vector, and input the feature vector into a pre-trained machine learning classification model to obtain the risk degree of abnormal glucose metabolism of the target object. The early warning module is configured to output risk classification results and referral prompts in response to the risk level meeting the set early warning conditions.

2. The glucose metabolism abnormality risk prediction system according to claim 1, characterized in that, The feature extraction module is further configured to perform smoothing filtering and outlier removal on the continuous blood glucose monitoring data before extracting the dynamic blood glucose features.

3. The glucose metabolism abnormality risk prediction system according to claim 1, characterized in that, The set duration is one hour, and the continuous overlapping net blood glucose function is the standard deviation of the difference between the glucose value at each sampling time and the glucose value one hour before that sampling time; the glucose coefficient of variation is the ratio of the standard deviation of the continuous blood glucose monitoring data to the average glucose value; the diurnal average absolute difference is the average of the absolute values ​​of the difference between the glucose values ​​at the same time on two adjacent days.

4. The glucose metabolism abnormality risk prediction system according to claim 1, characterized in that, The target range is set to 3.9 mmol / L to 10.0 mmol / L; the hyperglycemia index is an index that characterizes the degree of hyperglycemia risk, obtained by summing up glucose values ​​that are higher than the set glucose threshold after nonlinear conversion according to a set conversion coefficient.

5. The glucose metabolism abnormality risk prediction system according to claim 1, characterized in that, The continuous blood glucose monitoring data refers to glucose data collected continuously for no less than fourteen days within the set postpartum time window for the target subject; the set postpartum time window is four to twelve weeks after delivery.

6. The glucose metabolism abnormality risk prediction system according to claim 1, characterized in that, The risk prediction module is further configured to binarize the postpartum weight retention amount into a postpartum weight retention classification variable according to a set weight threshold, and combine the postpartum weight retention classification variable into the feature vector.

7. The glucose metabolism abnormality risk prediction system according to claim 6, characterized in that, The set weight threshold is 5 kg.

8. The glucose metabolism abnormality risk prediction system according to claim 1, characterized in that, The machine learning classification model includes at least one of the extreme gradient boosting model, random forest model, and logistic regression model, or an ensemble learning model composed of the aforementioned models.

9. The glucose metabolism abnormality risk prediction system according to claim 1, characterized in that, The early warning module is configured to determine that the set early warning conditions are met in response to the risk level being greater than a set risk threshold; the risk classification results include low risk, medium risk, and high risk; the referral prompt information includes information suggesting that the target subject be referred for a glucose tolerance test.

10. A data processing method for predicting the risk of abnormal glucose metabolism, characterized in that, include: Continuous blood glucose monitoring data of the target subject within a set postpartum time window, as well as the postpartum weight retention and gestational diabetes clinical characteristic parameters of the target subject, wherein the target subject is a postpartum individual who has previously suffered from gestational diabetes, and the postpartum weight retention is the difference between the target subject's current weight and its pre-pregnancy weight. Dynamic blood glucose features are extracted from the continuous blood glucose monitoring data. These dynamic blood glucose features include average glucose value, glucose coefficient of variation, continuous overlap net blood glucose effect, time percentage within the target range, hyperglycemia index, and diurnal average absolute difference. The continuous overlap net blood glucose effect characterizes the degree of dispersion of the difference between the glucose value at each sampling time and the glucose value at a set time before that sampling time. The dynamic blood glucose characteristics, postpartum weight retention, and clinical characteristic parameters of gestational diabetes are combined into a feature vector, and the feature vector is input into a pre-trained machine learning classification model to obtain the risk of abnormal glucose metabolism of the target object. In response to the risk level meeting the set early warning conditions, the risk classification result and referral prompt information are output.

11. The data processing method for predicting the risk of abnormal glucose metabolism according to claim 10, characterized in that, Before extracting the dynamic blood glucose features, the continuous blood glucose monitoring data is also smoothed, filtered, and outlier removed.

12. The data processing method for predicting the risk of abnormal glucose metabolism according to claim 10, characterized in that, The set duration is one hour; the continuous overlapping net blood glucose effect is the standard deviation of the difference between the glucose value at each sampling time and the glucose value one hour before that sampling time; the glucose coefficient of variation is the ratio of the standard deviation of the continuous blood glucose monitoring data to the average glucose value; the inter-day average absolute difference is the average of the absolute values ​​of the difference between the glucose values ​​at the same time on two adjacent days; the target range of time percentage is set to 3.9 mmol / L to 10.0 mmol / L.

13. The data processing method for predicting the risk of abnormal glucose metabolism according to claim 10, characterized in that, The postpartum weight retention is binarized according to a set weight threshold and combined into the feature vector, where the set weight threshold is 5 kg.

14. The data processing method for predicting the risk of abnormal glucose metabolism according to claim 10, characterized in that, The machine learning classification model includes at least one of the extreme gradient boosting model, random forest model, and logistic regression model, or an ensemble learning model composed of the aforementioned models.

15. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the computer program, it implements the data processing method for predicting the risk of abnormal glucose metabolism as described in any one of claims 10 to 14.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the data processing method for predicting the risk of abnormal glucose metabolism as described in any one of claims 10 to 14.

Images