Early warning system and method for postpartum abnormal glucose metabolism of gestational diabetes mellitus
By using multidimensional data collection and machine learning algorithms, an early warning system for postpartum glucose metabolism abnormalities in gestational diabetes was constructed, which solved the problem of the lag in traditional blood glucose testing and enabled early risk assessment and personalized intervention.
Patent Information
- Application Number
- CN202511453400.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-13
- Publication Date
- 2025-11-14
AI Technical Summary
Current monitoring of postpartum glucose metabolism disorders mainly relies on routine blood glucose testing, which is lagging and makes it difficult to detect patients with early glucose metabolism disorders in a timely manner. Furthermore, a single blood glucose indicator is insufficient to comprehensively and accurately assess the patient's condition.
It employs a multi-dimensional data acquisition module, including a continuous glucose monitor, body fat scale, blood pressure monitor, blood biochemistry analyzer, diet record system, and smart bracelet, to collect physiological data, dietary habit data, and lifestyle data. It then uses machine learning algorithms to build predictive models for risk assessment and tiered early warning.
It enables early warning of postpartum glucose metabolism abnormalities, reduces the workload of medical staff, improves diagnostic and treatment efficiency, enhances patients' health awareness and self-management ability, and provides personalized treatment plans.
Smart Images

Figure CN120954755A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of diabetes monitoring technology, and more specifically to an early warning system and method for postpartum glucose metabolism abnormalities in gestational diabetes. Background Technology
[0002] Gestational diabetes mellitus (GDM) is a common pregnancy complication that poses a potential threat to the health of both mother and baby. Although most patients' blood sugar levels return to normal after delivery, a significant number experience persistent glucose metabolism abnormalities postpartum, and may even develop type 2 diabetes.
[0003] Currently, clinical monitoring of postpartum glucose metabolism disorders mainly focuses on routine blood glucose testing. This method is often lagging and cannot promptly detect patients in the early stages of glucose metabolism disorders. Moreover, relying solely on a single blood glucose indicator is insufficient to comprehensively and accurately assess a patient's glucose metabolism status, as glucose metabolism is influenced by multiple factors, including physiological data, dietary habits, and lifestyle (such as sleep and exercise).
[0004] Therefore, it is necessary to propose an early warning system and method for postpartum glucose metabolism abnormalities in gestational diabetes mellitus to address the above problems. Summary of the Invention
[0005] (a) Technical problems to be solved
[0006] The purpose of this invention is to address the problem that clinical monitoring of postpartum glucose metabolism abnormalities mainly focuses on routine blood glucose testing, which is often lagging and cannot promptly detect problems in patients in the early stages of glucose metabolism abnormalities. This invention provides an early warning system and method for postpartum glucose metabolism abnormalities in gestational diabetes mellitus.
[0007] (II) Technical Solution
[0008] To achieve the above objectives, the present invention specifically adopts the following technical solution:
[0009] An early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus includes a data acquisition module, a data processing and analysis module, and an early warning module. The data acquisition module collects data from patients with both abnormal and normal postpartum glucose metabolism, including physiological data, dietary data, and lifestyle data. The data processing and analysis module processes and analyzes the data collected by the data acquisition module, and constructs a predictive model using machine learning algorithms. The model is trained using the data from patients with both abnormal and normal postpartum glucose metabolism collected by the data acquisition module. The early warning module performs risk assessment on the physiological data of new patients to obtain risk probabilities, classifies the risks based on the risk probabilities, and issues early warning information based on the risk classification.
[0010] Furthermore, the physiological data includes waist circumference, blood glucose, weight, body fat percentage, muscle mass, blood pressure, blood lipids, and insulin levels; the dietary habit data includes food types, intake, meal times, and meal frequency; and the lifestyle data includes sleep duration, sleep quality, exercise type, exercise intensity, exercise time, and exercise frequency.
[0011] Furthermore, the data acquisition module includes a continuous glucose monitor, a body fat scale, a blood pressure monitor, a blood biochemistry analyzer, a diet recording system, and a smart bracelet; the continuous glucose monitor collects the patient's blood glucose data and uploads it to the system database; the body fat scale collects the patient's weight, body fat percentage, muscle mass, and body composition data; the blood pressure monitor measures blood pressure; and the blood biochemistry analyzer detects blood lipids and insulin levels.
[0012] The diet record system is used to record the patient's daily meal times, number of meals, types of food, and intake. It has a built-in nutritional database of food and can automatically calculate the total daily intake of calories, carbohydrates, protein, and fat nutrients based on the types of food consumed and the amount of fat consumed each day.
[0013] The smart bracelet monitors the patient's total sleep duration and sleep quality; it also records the patient's exercise type, exercise time, and weekly exercise frequency.
[0014] Furthermore, when processing data, the data processing and analysis module cleans the collected raw data to remove invalid data and outliers, corrects or removes blood glucose values that exceed the normal range and unreasonable dietary records; and normalizes different types of data to make each data comparable on the same scale.
[0015] When performing data analysis, the data processing and analysis module performs feature extraction, extracting features from physiological data, dietary habit data, and lifestyle data.
[0016] Furthermore, the machine learning algorithm is the random forest algorithm; the model is trained using historical case data, with the extracted features as input variables and whether the patient has postpartum glucose metabolism abnormalities as output variables; the model is evaluated using cross-validation, with the dataset divided into training, validation, and test sets, and the experiment repeated multiple times;
[0017] The random forest algorithm constructs the prediction model as follows:
[0018] Single decision tree construction (CART algorithm) utilizes the node splitting criterion of minimizing the Gini exponent, mathematically represented as:
[0019] ;in, Represents the sample set of the current node. This represents the total number of categories (in this scenario, K=2, meaning normal / abnormal glucose metabolism). Represents a node The Middle The proportion of class samples;
[0020] Split gain calculation: = ;in, Indicated by feature The sample set of child nodes after splitting. Indicates the number of candidate features (randomly selected) (One characteristic).
[0021] Furthermore, the early warning module performs risk assessment on the physiological data of new patients in the following manner:
[0022] Input data: physiological data, dietary habits data, and lifestyle data of new patients; preprocessed feature vectors Input the pre-trained risk prediction model to calculate the baseline risk probability of postpartum glucose metabolism abnormalities. The original model output is calibrated, and the basic risk probability calculation model output is as follows: = Where T represents the number of decision trees in the random forest. No. Each decision tree predicts the category (0 or 1) for a new sample.
[0023] The risk classification function of the early warning module for risk probability classification is as follows:
[0024] RiskLevel= ;
[0025] The warning module issues warning information when the RiskLevel is "medium risk" or "high risk".
[0026] The methods for generating warning content in warning information include:
[0027] Risk factor tracing: Ranking features by importance and quantifying the contribution of each feature to the risk probability using SHAP values: = ;in, Representation of features The SHAP value (positive indicates increased risk, negative indicates decreased risk); This represents the prediction of the base learner t on the feature subset S;
[0028] Top-N risk factors: Select the top 3 key characteristics with the largest absolute values as risk drivers;
[0029] Rule base matching: Based on risk level and key features, suggestions are matched from a predefined rule base.
[0030] Furthermore, the warning information is sent to patients and medical staff via SMS notifications, APP push messages, or pop-ups in the hospital information system.
[0031] A method for early warning of postpartum glucose metabolism abnormalities in gestational diabetes mellitus, comprising the following steps:
[0032] S1: Collect patients' physiological data, dietary habits data, and lifestyle data through the data acquisition module;
[0033] S2: Input the collected data into the data processing and analysis module to perform data processing, data analysis and predictive model construction, and calculate the risk probability of postpartum glucose metabolism abnormalities in patients.
[0034] S3: Based on the magnitude of the risk probability, the early warning module issues early warning information corresponding to the risk level.
[0035] (III) Beneficial Effects
[0036] The beneficial effects of this invention are as follows:
[0037] 1. This invention, through multidimensional data collection and analysis, can accurately assess the risk of postpartum glucose metabolism abnormalities in patients before they exhibit obvious abnormalities, enabling early warning and providing a valuable window for timely intervention. It fully considers the individual differences of each patient, focusing not only on blood glucose levels but also incorporating physiological data, dietary habits, and lifestyle data, enabling a more comprehensive and accurate assessment of the patient's glucose metabolism status and providing a basis for developing personalized treatment plans.
[0038] 2. This invention automatically collects and analyzes data, reducing the workload of medical staff and improving the efficiency of diagnosis and treatment. Simultaneously, the timely delivery of early warning information facilitates patient self-management and prompt medical attention. By providing patients with detailed risk assessment results and intervention suggestions, it enhances patients' health awareness, prompting them to proactively adjust their lifestyles and prevent the occurrence and development of abnormal glucose metabolism. Attached Figure Description
[0039] Figure 1 This is a schematic diagram of the early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus according to the present invention.
[0040] Figure 2 This is a schematic diagram of the early warning method for postpartum glucose metabolism abnormalities in gestational diabetes mellitus according to the present invention. Detailed Implementation
[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0042] Example 1
[0043] Please see Figure 1 An early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus includes a data acquisition module, a data processing and analysis module, and an early warning module. The data acquisition module collects data on postpartum patients with abnormal and normal glucose metabolism, including physiological data, dietary habit data, and lifestyle data. The data processing and analysis module processes and analyzes the data collected by the data acquisition module and uses machine learning algorithms to build a predictive model, training the model using the data from postpartum patients with abnormal and normal glucose metabolism collected by the data acquisition module. The early warning module performs risk assessment on the physiological data of new patients to obtain risk probabilities, classifies the risks based on the risk probabilities, and issues early warning information based on the risk classification. The early warning information is sent to patients and medical staff via SMS notifications, APP push messages, or pop-ups in the hospital information system.
[0044] Physiological data include waist circumference, blood sugar, weight, body fat percentage, muscle mass, blood pressure, blood lipids, and insulin levels; dietary data includes food types, intake, meal times, and meal frequency; and lifestyle data includes sleep duration, sleep quality, exercise type, exercise intensity, exercise time, and exercise frequency.
[0045] The data acquisition module includes a continuous glucose monitor (CGM), a body fat scale, a blood pressure monitor, a blood biochemistry analyzer, a diet recording system, and a smart bracelet. The CGM automatically collects the patient's blood glucose data at regular intervals (e.g., every 5 minutes) and uploads it to the system database. This device reflects the patient's blood glucose fluctuations throughout the day, including fasting blood glucose, postprandial blood glucose, and nighttime blood glucose. The body fat scale collects the patient's weight, body fat percentage, muscle mass, and body composition data. The blood pressure monitor measures blood pressure. The blood biochemistry analyzer detects blood lipids (total cholesterol, triglycerides, high-density lipoprotein cholesterol, and low-density lipoprotein cholesterol) and insulin levels. These physiological parameters are closely related to glucose metabolism and can serve as auxiliary diagnostic criteria.
[0046] The diet record system is used to record the patient's daily meal times, number of meals, types of food, and intake. It has a built-in nutritional database of food and can automatically calculate the total calories, carbohydrates, protein, and fat intake of the patient each day based on the types of food consumed and the amount of fat. Patients can identify food by taking a picture or manually enter the types of food they eat and the approximate amount they consume.
[0047] The smart bracelet monitors the patient's total sleep duration and sleep quality (deep sleep duration, light sleep duration, and number of awakenings). Insufficient sleep or poor sleep quality can affect hormone secretion, thereby interfering with glucose metabolism. The smart bracelet also records the patient's exercise type (such as walking, running, yoga), exercise intensity (steps, calories burned), exercise time, and weekly exercise frequency. Appropriate exercise helps improve insulin sensitivity and improve glucose metabolism; while lack of exercise will worsen glucose metabolism.
[0048] The data processing and analysis module performs data cleaning on the collected raw data, removing invalid data and outliers. For example, it corrects or removes blood glucose levels outside the normal range and unreasonable dietary records. Different types of data are normalized to ensure comparability on the same scale, facilitating subsequent data analysis and modeling. The data cleaning methods are as follows:
[0049] Define the rule base: Establish the medical range for each indicator (e.g., normal fasting blood glucose level is 3.9). (6.1 mmol / L), values outside this range are marked as potential abnormalities;
[0050] Statistical methods are used to assist in identifying extreme values using box plots (IQR); if a data point meets the following criteria... If so, it is judged as abnormal; (Third quartile minus First quartile);
[0051] Correction strategy: Replace values that exceed the reasonable range with nearby boundary values (e.g., blood glucose > upper limit → set to upper limit); if there is a spatiotemporal correlation (e.g., continuous blood glucose curves of the same patient), use time series models to predict reasonable values to replace outliers;
[0052] Logical consistency check: Verify the logical relationship between different parameters; for example, if the total daily calorie intake is greater than the basal metabolic rate × activity coefficient × 2, it is considered overeating; check whether the sequence of events conforms to common sense (e.g., blood sugar should decrease rather than increase after exercise); prioritize conflicting data (e.g., laboratory test results take precedence over self-reported dietary records).
[0053] Multiple imputation mechanisms: imputing missing items based on feature values of similar samples; inferring approximate values of missing elements using the low-rank property of the matrix (suitable for high-dimensional sparse data); adding binary flag bits to missing values that cannot be imputed, allowing the model to learn their impact autonomously;
[0054] The normalization process is as follows:
[0055] ZScore was used for physiological data (such as blood sugar and blood pressure) because of their natural large fluctuations; MinMax was used for dietary habit data (such as grams of carbohydrates) because of their clear dimensions and controllable range; Principal component analysis (PCA) was used to determine the contribution of each feature and to give greater weight to features with high variance; and logarithm was taken first and then normalized for lifestyle data (such as sleep duration).
[0056] The normalization implementation process is as follows:
[0057] Data is processed independently for each channel and normalized separately according to its source (CGM, questionnaire, sensor) to avoid cross-contamination;
[0058] The original baseline is preserved, and the mean and standard deviation of each column of data are stored for subsequent inverse transformation to restore the true value.
[0059] Batch real-time updates: When new data is added, statistical parameters are dynamically refreshed instead of fixed historical averages.
[0060] When performing data analysis, the data processing and analysis module performs feature extraction, extracting features from physiological data, dietary habit data, and lifestyle data.
[0061] Daily average blood glucose, blood glucose standard deviation, peak blood glucose time, MAGE index (mean glucose increment), visceral fat index, and insulin resistance index were extracted from physiological data; the visceral fat index was: Insulin resistance index is: ;
[0062] Nutrient ratio features were extracted from dietary habit data, and dietary pattern clustering was performed. Nutrient ratio features were obtained by calculating the carbohydrate energy ratio, using the mathematical formula: CHO = ;in, This indicates the total energy provided by carbohydrates on that day (kcal). This represents the total daily energy intake (kcal); the dietary pattern clustering method was obtained through principal component analysis (PCA) dimensionality reduction, and the mathematical formula is: X; where, =I,X represents the original food intake matrix (n×p). This represents the eigenvector matrix (p×k, where k is the principal component).
[0063] Sleep structure features and exercise performance features are extracted from lifestyle data; sleep structure features are obtained by calculating the percentage of deep sleep, using the following mathematical formula: ;in, The duration of deep sleep (in minutes). Total sleep duration (minutes); Exercise performance characteristics are obtained by calculating cumulative metabolic equivalents, using the following mathematical formula: ;in, This represents the metabolic equivalent value of the i-th exercise. This indicates the duration (in minutes) of the movement.
[0064] This embodiment constructs a multi-dimensional, real-time monitoring early warning system. By integrating physiological data collected from devices such as continuous glucose monitoring, body fat scales, and blood pressure monitors, and combining dietary records and lifestyle data tracked by smart bracelets, it achieves a comprehensive assessment of the risk of abnormal glucose metabolism. The system uses time series models to correct outliers and improves data quality through principal component analysis (PCA) and normalization, laying the foundation for subsequent accurate modeling. Its core advantage lies in covering key factors such as 24-hour blood glucose fluctuations, nutritional intake, and sleep-exercise patterns, significantly improving the timeliness and comprehensiveness of risk assessment.
[0065] Example 2
[0066] This embodiment is a further optimization based on Embodiment 1. Specifically, the machine learning algorithm is the random forest algorithm; historical case data (data of patients with known postpartum glucose metabolism abnormalities and normalities) are used to train the model, with the extracted features as input variables and whether the patient has postpartum glucose metabolism abnormalities as output variables. By continuously adjusting the model parameters, the model performance is optimized to achieve high accuracy and recall; cross-validation is used to evaluate the model, dividing the dataset into training, validation, and test sets, and repeating the experiment multiple times to ensure the stability and reliability of the model.
[0067] The random forest algorithm constructs the prediction model as follows:
[0068] Single decision tree construction (CART algorithm) utilizes the node splitting criterion of minimizing the Gini exponent, mathematically represented as:
[0069] ;in, Represents the sample set of the current node. This represents the total number of categories (in this scenario, K=2, meaning normal / abnormal glucose metabolism). Represents a node The Middle The proportion of class samples;
[0070] Split gain calculation: = ;in, Indicated by feature The sample set of child nodes after splitting. Indicates the number of candidate features (randomly selected) (features)
[0071] The predicted values for the leaf nodes are as follows:
[0072] Classification task: Take the majority category within the leaf node as the prediction result;
[0073] Regression task (e.g., predicting risk probability): take the mean of samples within leaf nodes.
[0074] The random forest ensemble learning method is as follows:
[0075] Autopilot sampling: Each decision tree uses a subset of samples with replacement drawn from the original dataset. The samples not selected constitute the out-of-bag samples for internal validation.
[0076] Random feature selection: When splitting at each node, from Randomly selected from the features The goal is to perform optimal segmentation based on the features; the objective is to break the strong correlation between features and improve generalization ability.
[0077] Integrated prediction rules:
[0078] For classification tasks, a majority vote method is used: ;
[0079] Regression task (risk probability output), taking the average of all tree predictions: = ;
[0080] Where T represents the number of decision trees, For the first The prediction results for the trees, the first The prediction output of each decision tree for a sample (class labels for classification tasks, continuous values for regression tasks). This represents an indicative function, a binary function that determines whether a condition is true (returning 1 if the condition is true, and 0 otherwise). Let c be the category label, where c ∈ C, and C is the set of categories in the classification task (e.g., in this scenario, C = {0, 1}, where 0 = normal and 1 = abnormal). For the first The prediction results for the trees, the first The probability that a sample output by each decision tree belongs to the positive class (abnormality);
[0081] The log-likelihood loss with L2 regularization is calculated, and its mathematical representation is as follows:
[0082] + ;
[0083] in, Indicates the true label (0=normal, 1=abnormal). The model predicts the probability of risk. This represents the L2 regularization coefficient (which controls complexity). This represents model parameters (such as decision tree node thresholds).
[0084] The implementation steps are as follows:
[0085] Feature extraction: Extract features from new patient data that are consistent with the training set (such as the mean blood glucose value during the same statistical window).
[0086] Missing value imputation: Use the same strategy as in the training phase (such as multiple imputation);
[0087] Model inference: Call the predict_proba method of the trained random forest model to obtain the raw probabilities;
[0088] Probability calibration: Applying PlateScaling to correct probabilities improves clinical reliability.
[0089] This embodiment employs the random forest algorithm to optimize the prediction model. It breaks down data redundancy through bootstrapping and random feature selection, and combines L2 regularization loss function to control model complexity, effectively improving generalization ability. Cross-validation and probability calibration ensure high reliability of high-risk / normal classification, enabling the model to achieve both high accuracy and low false positive rate in clinical scenarios. This optimization significantly enhances adaptability to individual differences, providing solid algorithmic support for subsequent early warning grading.
[0090] Example 3
[0091] This embodiment is a further optimization based on embodiment 2. Specifically, the early warning module performs risk assessment on the physiological data of new patients in the following way:
[0092] Input data: physiological data (e.g., blood glucose, blood pressure), dietary habits (e.g., carbohydrate intake), and lifestyle data (e.g., sleep duration, exercise level) of new patients; the preprocessed feature vectors... Input the pre-trained risk prediction model to calculate the baseline risk probability of postpartum glucose metabolism abnormalities. The original model output is calibrated to improve probability reliability. Basic risk probability calculation model output: Assuming a random forest model is used, its essence is the ensemble voting result of multiple decision trees. For a binary classification problem (normal / abnormal), the final risk probability is: = Where T represents the number of decision trees in the random forest. No. Each decision tree predicts the category (0 or 1) for a new sample.
[0093] The risk classification function of the early warning module for risk probability classification is as follows:
[0094] ;
[0095] Dynamic threshold adjustment: The threshold can be updated periodically based on clinical feedback (e.g., by using the optimal sensitivity / specificity balance point through ROC curves).
[0096] The trigger conditions for the early warning module to issue an early warning message are:
[0097] Warning activation: Triggered when RiskLevel is "Medium Risk" or "High Risk";
[0098] Suppression mechanism: Avoid repeated warnings (e.g., only one notification is sent for the same patient within 24 hours);
[0099] The methods for generating warning content in warning information include:
[0100] Risk factor tracing: Ranking features by importance and quantifying the contribution of each feature to the risk probability using SHAP values: = ;in, Representation of features The SHAP value (positive indicates increased risk, negative indicates decreased risk); This represents the prediction of the base learner t on the feature subset S;
[0101] Top-N risk factors: Select the top 3 key characteristics with the largest absolute values as risk drivers;
[0102] Rule base matching: Based on risk level and key features, suggestions are matched from a predefined rule base, for example:
[0103] High-risk, high-carbohydrate diet: Reduce refined carbohydrate intake and increase dietary fiber;
[0104] Medium risk + sleep deprivation: Ensure 78 hours of sleep per day;
[0105] High risk + lack of exercise: at least 150 minutes of moderate-intensity exercise per week.
[0106] This embodiment achieves personalized customization of early warning information through dynamic threshold adjustment and SHAP value tracing technology. The system not only triggers early warnings based on risk probability grading (e.g., "medium / high risk"), but also generates targeted suggestions (e.g., high-carbohydrate diet intervention) through rule base matching, and employs a 24-hour deduplication mechanism to avoid information redundancy. This design enhances the clinical usability of the early warning system, prompting patients to proactively adjust their behavior (e.g., increase exercise, improve sleep), forming a closed loop of doctor-patient collaborative management, and significantly improving intervention efficiency and patient compliance.
[0107] This invention, through multidimensional data collection and analysis, can accurately assess the risk of postpartum glucose metabolism abnormalities in patients before they exhibit obvious abnormalities, enabling early warning and providing a valuable window for timely intervention. It fully considers the individual differences of each patient, focusing not only on blood glucose levels but also incorporating physiological data, dietary habits, and lifestyle data, enabling a more comprehensive and accurate assessment of the patient's glucose metabolism status and providing a basis for developing personalized treatment plans.
[0108] The system automatically collects and analyzes data, reducing the workload of medical staff and improving the efficiency of diagnosis and treatment. Meanwhile, timely notifications of early warning information facilitate patient self-management and prompt medical attention. By providing patients with detailed risk assessment results and intervention recommendations, the system enhances patients' health awareness, encouraging them to proactively adjust their lifestyles and prevent the occurrence and development of abnormal glucose metabolism.
[0109] This invention also discloses an early warning method for a system for early warning of postpartum glucose metabolism abnormalities in gestational diabetes mellitus, comprising the following steps:
[0110] S1: Collect patients' physiological data, dietary habits data, and lifestyle data through the data acquisition module;
[0111] S2: Input the collected data into the data processing and analysis module to perform data processing, data analysis and predictive model construction, and calculate the risk probability of postpartum glucose metabolism abnormalities in patients.
[0112] S3: Based on the magnitude of the risk probability, the early warning module issues early warning information of the corresponding risk level;
[0113] Specifically:
[0114] Patient admission: Patients with gestational diabetes are admitted to the hospital before delivery. Medical staff will put a continuous glucose monitor on them and collect their first physiological data (weight, body fat percentage, blood pressure, blood lipids, insulin, etc.). At the same time, they will instruct the patients on how to use the diet record system and smart bracelet.
[0115] Inpatient monitoring: During hospitalization, patients' blood glucose, sleep and exercise are continuously monitored, and dietary records are collected daily. This data is then transmitted in real time to the data processing and analysis module for preliminary analysis.
[0116] Discharge preparation: Before the patient is discharged, a comprehensive examination and assessment are conducted, including re-measuring physiological parameters, analyzing dietary and exercise data during hospitalization, and using a trained model to calculate the probability of postpartum glucose metabolism abnormalities and record it in the early warning system.
[0117] Post-discharge follow-up: After discharge, patients continue to wear continuous glucose monitoring devices and other wearable devices, and go to the hospital for regular check-ups as prescribed by the doctor. At each check-up, the patient's physiological data, dietary habits and lifestyle data are updated, and the risk assessment is re-conducted. If the patient's risk level changes, the system will automatically issue an early warning.
[0118] This invention overcomes the limitations of single indicators by utilizing multi-dimensional real-time monitoring; it achieves automatic blood sampling every 5 minutes through a continuous glucose monitor (CGM), covering blood glucose fluctuations throughout the day (fasting / postprandial / nighttime), breaking through the limitation of traditional finger-prick blood tests that only reflect instantaneous values. It simultaneously collects metabolic-related indicators such as waist circumference, body fat percentage, blood pressure, blood lipids, and insulin levels to construct a comprehensive metabolic picture, which can better reflect the complexity of glucose metabolism than simple blood glucose testing.
[0119] This invention employs a random forest algorithm to process high-dimensional data (including daily average blood glucose standard deviation, MAGE index, HOMAIR, and other features), and identifies subclinical abnormalities through spatiotemporal correlation analysis. Based on SHAP values to quantify the contribution of each feature, a three-tiered early warning system of "low / medium / high" is established, offering improved sensitivity compared to traditional threshold-based judgments. Combined with rule-based matching technology, the system can output matching suggestions for high-risk + high-carbohydrate dietary combinations.
[0120] In summary, this invention, through multi-dimensional data perception and analysis, breaks through the limitations of traditional methods that rely solely on blood glucose monitoring, constructing a comprehensive, accurate, and forward-looking early warning system for postpartum glucose metabolism abnormalities. By integrating information from physiological, dietary, and lifestyle factors, and employing data analysis techniques and machine learning algorithms, it uncovers patterns hidden behind massive amounts of data, thereby achieving accurate prediction of individual risks. This not only helps improve the quality of medical services but also provides patients with more personalized and proactive health management solutions, yielding significant social and economic benefits.
[0121] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. The scope of patent protection of the present invention shall be determined by the claims. Similarly, any equivalent structural changes made based on the description and drawings of the present invention shall also be included within the scope of protection of the present invention.
Claims
1. An early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus, comprising a data acquisition module, a data processing and analysis module, and an early warning module, characterized in that: The data acquisition module is used to collect data on postpartum patients with abnormal and normal glucose metabolism, including physiological data, dietary habits data, and lifestyle data. The data processing and analysis module is used to process and analyze the data collected by the data acquisition module, and to build a predictive model using machine learning algorithms. The model is trained using the postpartum patient data with abnormal and normal glucose metabolism collected by the data acquisition module. The early warning module is used to perform risk assessment on the physiological data of new patients to obtain risk probabilities, classify risks according to risk probabilities, and issue early warning information based on risk classification.
2. The early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus according to claim 1, characterized in that: The physiological data includes waist circumference, blood glucose, weight, body fat percentage, muscle mass, blood pressure, blood lipids, and insulin levels; the dietary habit data includes food types, intake, meal times, and meal frequency; and the lifestyle data includes sleep duration, sleep quality, exercise type, exercise intensity, exercise time, and exercise frequency.
3. The early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus according to claim 1, characterized in that: The data acquisition module includes a continuous glucose monitor, a body fat scale, a blood pressure monitor, a blood biochemistry analyzer, a diet recording system, and a smart bracelet; the continuous glucose monitor collects the patient's blood glucose data and uploads it to the system database; the body fat scale collects the patient's weight, body fat percentage, muscle mass, and body composition data; A sphygmomanometer measures blood pressure; Blood biochemistry analyzers detect blood lipid and insulin levels; The diet record system is used to record the patient's daily meal times, number of meals, types of food, and intake. It has a built-in nutritional database of food and can automatically calculate the total daily intake of calories, carbohydrates, protein, and fat nutrients based on the types of food consumed and the amount of fat consumed each day. The smart bracelet monitors the patient's total sleep duration and sleep quality; it also records the patient's exercise type, exercise time, and weekly exercise frequency.
4. The early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus according to claim 1, characterized in that: When processing data, the data processing and analysis module cleans the collected raw data to remove invalid data and outliers, corrects or removes blood glucose values that are outside the normal range and unreasonable dietary records, and normalizes different types of data to make them comparable on the same scale. When performing data analysis, the data processing and analysis module performs feature extraction, extracting features from physiological data, dietary habit data, and lifestyle data.
5. The early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus according to claim 1, characterized in that: The machine learning algorithm is the random forest algorithm; the model is trained using historical case data, with the extracted features as input variables and whether the patient has postpartum glucose metabolism abnormalities as output variables; the model is evaluated using cross-validation, with the dataset divided into training, validation, and test sets, and the experiment repeated multiple times; The random forest algorithm constructs the prediction model as follows: Single decision tree construction (CART algorithm) utilizes the node splitting criterion of minimizing the Gini exponent, mathematically represented as: ;in, Represents the sample set of the current node. This represents the total number of categories (in this scenario, K=2, meaning normal / abnormal glucose metabolism). Represents a node The Middle The proportion of class samples; Split gain calculation: = ;in, Indicated by feature The sample set of child nodes after splitting. Indicates the number of candidate features (randomly selected) (One characteristic).
6. The early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus according to claim 1, characterized in that: The early warning module assesses the risk of new patients' physiological data in the following way: Input data: physiological data, dietary habits data, and lifestyle data of new patients; preprocessed feature vectors... Input the pre-trained risk prediction model to calculate the baseline risk probability of postpartum glucose metabolism abnormalities. The original model output is calibrated, and the basic risk probability calculation model output is as follows: = Where T represents the number of decision trees in the random forest. No. Each decision tree predicts the category (0 or 1) for a new sample. The risk classification function of the early warning module for risk probability classification is as follows: RiskLevel= ; The warning module issues warning information when the RiskLevel is "medium risk" or "high risk". The methods for generating warning content in warning information include: Risk factor tracing: Ranking features by importance and quantifying the contribution of each feature to the risk probability using SHAP values: = ;in, Representation of features The SHAP value (positive indicates increased risk, negative indicates decreased risk); This represents the prediction of the base learner t on the feature subset S; Top-N risk factors: Select the top 3 key characteristics with the largest absolute values as risk drivers; Rule base matching: Based on risk level and key features, suggestions are matched from a predefined rule base.
7. The early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus according to claim 1 or 6, characterized in that: The warning information is sent to patients and medical staff via SMS notifications, APP push messages, or pop-ups in the hospital information system.
8. The early warning method of the early warning system for postpartum glucose metabolism abnormalities in gestational diabetes mellitus according to any one of claims 1-7, characterized in that, Includes the following steps: S1: Collect patients' physiological data, dietary habits data, and lifestyle data through the data acquisition module; S2: Input the collected data into the data processing and analysis module to perform data processing, data analysis and predictive model construction, and calculate the risk probability of postpartum glucose metabolism abnormalities in patients. S3: Based on the magnitude of the risk probability, the early warning module issues early warning information corresponding to the risk level.
