Comprehensive management system for diabetics

Abstract

The invention relates to an integrated management system for diabetic patients, which comprises the following steps of: constructing a high-dimensional metabolic data model, performing multi-modal fusion on genome information, metabolome data and life behavior data, and mining a potential mode of metabolic characteristics of the patients; carrying out patient typing by utilizing a clustering algorithm comprising a self-organizing mapping network, and identifying different metabolism types comprising an insulin resistance type and an insulin secretion insufficiency type; developing a hierarchical management strategy based on a typing result, wherein the hierarchical management strategy comprises diet, exercise and medicine intervention schemes for different types; collecting dynamic quantitative indexes of the metabolic load, and integrating the blood glucose fluctuation amplitude, insulin sensitivity and daily nutrition intake of the patient; constructing a metabolic load prediction model, and combining a time sequence deep learning algorithm with real-time data, including blood glucose, diet and exercise, of a patient to measure a future metabolic load; and establishing a causal relationship model between patient behaviors and metabolic reactions, and analyzing an interaction mode of diet, exercise and metabolic loads by adopting a causal inference technology.

Description

Technical Field

[0001] The invention relates to a comprehensive management system for diabetic patients. Background Art

[0002] At present, the integrated management system for diabetes patients with the patent number 202410550285X has introduced blood sugar monitoring, information archive management, cloud storage terminal and patient health diagnosis module in diabetes management, which has certain innovation and practicality, but there are still some deficiencies and disadvantages in actual application. Specifically, it is manifested in the following aspects:

[0003] First, the monitoring and management of diabetic patients by the system mainly focuses on the static analysis and periodic evaluation of blood glucose parameters. The core is to collect and calculate the fasting, postprandial and bedtime blood glucose data of patients to obtain the blood glucose stability evaluation coefficient. However, this evaluation method ignores the dynamic characteristics of blood glucose fluctuations in patients and fails to fully reflect the comprehensive impact of blood glucose fluctuations on metabolic load. Blood glucose fluctuations in diabetic patients, especially the alternating state of hyperglycemia and hypoglycemia, have a significant impact on the patient's metabolism and complication risk, and the calculation of blood glucose values ​​based solely on fixed time points cannot fully capture these dynamic changes. This deficiency may lead to weak pertinence of management programs in dealing with patients' complex metabolic states. Secondly, the system lacks fusion analysis of multidimensional data and only manages based on blood glucose parameters, without fully considering other key factors that affect metabolic load, such as insulin sensitivity, dietary structure, exercise intensity and rest mode. The management of diabetes is a highly complex process, and relying solely on single-dimensional blood glucose monitoring is not enough to fully reflect the patient's health status. The system does not integrate patients' behavioral data (such as diet, exercise and sleep) with metabolic data for correlation analysis, resulting in deficiencies in personalization and precision of its management program, making it difficult to meet the individual needs of patients.

[0004] Third, the feedback mechanism and regulation ability of the system are relatively simple. The system mainly provides evaluation reports for patients through the calculation results of blood glucose parameters, and does not dynamically adjust the patient's intervention strategy in combination with real-time monitoring data. For example, patients may experience strenuous exercise, high-carb diet or lack of sleep during the day, and these short-term behaviors will significantly affect blood glucose fluctuations and metabolic status. However, the existing system lacks a real-time feedback regulation mechanism and cannot dynamically adjust the patient's diet, exercise and medication regimen according to real-time data, which may result in patients not receiving timely intervention in high-risk situations, thereby increasing the risk of complications. Fourth, the system has limited ability to identify and handle abnormal events. The existing system records and evaluates the patient's blood glucose status through cloud storage terminals and blood glucose monitoring and analysis modules, but does not set up clear early warning mechanisms and emergency treatment plans for abnormal events (such as acute hyperglycemia, hypoglycemia or excessive blood glucose fluctuations). When the patient's blood glucose fluctuations exceed the safe range, the system cannot actively prompt the patient to take measures, nor can it provide clear decision support for doctors. This passive management mode may increase the patient's health risks.

[0005] In addition, the system lacks in-depth analysis and intervention support for the linkage between behavior and metabolism. The diet, exercise, medication and lifestyle of diabetic patients are interrelated and jointly affect their metabolic state, but the system only provides blood sugar assessment results and fails to analyze the impact path of patient behavior on metabolic load based on causal inference technology. Summary of the invention

[0006] The purpose of the present invention is to provide a comprehensive management system for diabetic patients, thereby solving some of the drawbacks and shortcomings pointed out in the background technology.

[0007] The present invention solves the above-mentioned technical problems by adopting the following technical solution, which includes the following steps:

[0008] S1. Accurate classification and stratified management of diabetes based on individual metabolic characteristics:

[0009] S1.1. Build a high-dimensional metabolic data model to integrate genomic information, metabolomics data, and life behavior data in a multimodal manner to explore the potential patterns of patients' metabolic characteristics;

[0010] S1.1. Use clustering algorithms including self-organizing map networks to classify patients and identify different metabolic types including insulin resistance and insulin secretion deficiency;

[0011] S1.3. Develop tiered management strategies based on the classification results, including diet, exercise and drug intervention plans for different classifications;

[0012] S2. Dynamic metabolic load monitoring and prediction model:

[0013] S2.1. Collect dynamic quantitative indicators of metabolic load, integrating the patient's blood sugar fluctuation range, insulin sensitivity and daily nutritional intake;

[0014] S2.2. Construct a metabolic load prediction model to measure future metabolic load by combining patients’ real-time data including blood sugar, diet, and exercise through a time series deep learning algorithm;

[0015] S3. Bidirectional feedback regulation mechanism between behavior and metabolism:

[0016] S3.1. Establish a causal relationship model between patient behavior and metabolic response, and use causal inference techniques to analyze the interaction pattern between diet, exercise and metabolic load;

[0017] S3.2. Develop a bidirectional feedback control algorithm between behavior and metabolism to adjust the behavior intervention plan according to the real-time metabolic data and update the metabolic model parameters according to the behavior data;

[0018] S3.3. Integrate personalized behavioral incentive modules to enhance patient compliance with intervention measures using health points system, goal setting and reward mechanism;

[0019] S4. Intelligent diagnosis and early warning system for metabolic abnormalities:

[0020] S4.1. Identify potential risk signals in real-time metabolic data, including hyperglycemia, hypoglycemia, or other metabolic imbalances, based on anomaly detection algorithms including deep autoencoders;

[0021] S4.2. Propose a metabolic abnormality scoring system to dynamically quantify metabolic risk levels;

[0022] S4.3. For high-risk events including acute hypoglycemia, an integrated early warning and intervention recommendation system is used to link the behavior control module to provide emergency plans.

[0023] Furthermore, the dynamic metabolic load monitoring and prediction model construction method includes: obtaining the blood sugar fluctuation amplitude as a time series function through a continuous blood sugar monitoring device; dynamically evaluating insulin sensitivity in combination with the patient's insulin usage and blood sugar response law; collecting nutrient intake in real time through a diet recording device to quantify the impact of daily diet on metabolic load; the expression formula is:

[0024]

[0025] Among them, MLI(t) represents the metabolic load index at time t; w1, w2, w3 are the weight factors of blood glucose fluctuation amplitude, insulin sensitivity and daily nutritional load, which represent the contribution ratio of each index to metabolic load; G′(t) is the rate of change of blood glucose over time, reflecting the dynamic characteristics of short-term blood glucose fluctuation; G(t) is the real-time blood glucose value, Gtarget is the target blood sugar level, and the difference between the two represents the degree to which the current blood sugar deviates from the healthy target; I(t) is the real-time insulin dosage, which measures the patient's responsiveness to insulin; N(t) is the daily nutritional load, which is a weighted combination of nutrients such as carbohydrates, proteins, and fats.

[0026] Furthermore, the dynamic metabolic load monitoring and prediction model construction method includes: using an improved time series deep learning model to integrate blood sugar fluctuations, exercise and dietary behavior data to generate time-dependent features; through an adaptive learning mechanism, dynamically adjusting the prediction model parameters to adapt to individual differences in patients; the expression formula is:

[0027]

[0028] Among them, P(t+Δt) is the metabolic load value predicted at the future time t+Δt; is a time series deep learning model, and the parameter θ is obtained through training; D(t) is the metabolic load feature input at time t, including blood glucose fluctuations, insulin sensitivity, and nutritional intake; B(t) is the patient's behavioral data input, including exercise intensity, eating time, and rest pattern.

[0029] Furthermore, the dynamic metabolic load monitoring and prediction model construction method includes: predicting the future trend of metabolic load changes by integrating blood sugar, diet and exercise data in real time; introducing an explanatory analysis module, using reverse tracing technology to identify the main influencing factors that cause predicted changes, and providing transparent health advice; the expression formula is:

[0030] F future (t+Δt)=P(t+Δt)+∫ t t+Δt α·ω(D(t))dt,

[0031] Among them, F future (t+Δt) is the predicted metabolic load value at the future time t+Δt; P(t+Δt) is the basic output of the predicted value; The dynamic adjustment value during the prediction period is calculated based on the metabolic load change rate ω(D(t)), and the weight coefficient α reflects the sensitivity of the patient's metabolic load to behavioral changes.

[0032] Furthermore, the behavior and metabolism bidirectional feedback regulation mechanism method includes: by collecting patients' long-term behavior data including diet, exercise and sleep and metabolic data, using causal inference technology including structured causal model to establish a causal relationship network between behavior and metabolic load; defining the quantitative relationship between key intervention variables including exercise intensity and carbohydrate intake and target metabolic indicators, and removing non-causal associated variables; and the causal relationship between behavior and metabolic load is quantified as:

[0033] ΔM=∫0 T (α1·B1(t)+α2·B2(t)β·C(t)) d t,

[0034] Among them: ΔM is the cumulative change of metabolic load, which indicates the change of metabolic state of the patient in time period T; B1(t) is the exercise intensity of the patient at time t; B2(t) is the dietary carbohydrate intake of the patient at time t; C(t) is the interference of the common dependent variable on the metabolic load at time t; α1 and α2 are the causal influence coefficients of exercise and diet on metabolic load; β is the inhibition coefficient of the common dependent variable, which indicates the intensity of the negative effect.

[0035] Furthermore, the behavior and metabolism bidirectional feedback regulation mechanism method includes:

[0036] Through multimodal data fusion technology, dietary intake, exercise parameters and metabolic changes are modeled synchronously, and the interactive relationship between behavior and metabolism is updated in real time. Dynamic causal graphs are used to identify the immediate impact and long-term trend of short-term behavior on metabolism in real time. The metabolic interaction pattern analysis model is:

[0037]

[0038] Where: R(t) is the real-time response rate of behavior to metabolic load at time t; M(t) is the metabolic load state at time t; B(t) is the behavioral data input at time t; γ is the immediate response coefficient, which represents the short-term impact intensity of behavior on metabolic load; η is the weight of the impact of long-term behavior on metabolic load; φ(B(t), t) is the nonlinear relationship function between behavior and metabolism, which is used to capture the interactive pattern including the delayed effect of exercise on blood glucose.

[0039] Furthermore, the behavior and metabolism bidirectional feedback regulation mechanism method includes:

[0040] Based on causal inference and real-time interaction pattern analysis, a two-way feedback control algorithm is developed; the patient's behavioral intervention plan is dynamically adjusted according to real-time metabolic data, and the metabolic model parameters are updated according to the behavioral data; and an abnormal event warning mechanism is introduced; and the behavior and metabolism two-way feedback control equation is:

[0041] B new (t) = B current (t)+k·(M target M(t))+δ·ψ(A(t)),

[0042] in:

[0043] B new (t) is the behavioral intervention adjustment value after feedback at time t; Bcurrent (t) is the current behavior data at time t; M target is the target metabolic load value; M(t) is the real-time metabolic load state at time t; k is the feedback gain coefficient, which controls the intervention intensity; δ is the abnormal response coefficient; ψ(A(t)) is the abnormal event adjustment function, which dynamically corrects the behavior based on the abnormal event A(t).

[0044] The comprehensive management system for diabetic patients of the present invention has the following beneficial effects:

[0045] Real-time collection of multi-dimensional data such as blood sugar, insulin sensitivity, diet, exercise, etc. of patients, comprehensive construction of dynamic metabolic load model, accurate assessment of patients' metabolic status, and providing scientific basis for personalized management. Through the two-way feedback control mechanism, the system can dynamically adjust patients' diet, exercise and drug regimens according to real-time metabolic data, and optimize metabolic model parameters in real time according to patients' behavioral data, ensuring dynamic adaptability and personalization of intervention strategies.

[0046] The integrated deep learning-based anomaly detection algorithm can promptly identify hyperglycemia, hypoglycemia or other metabolic abnormalities, and provide personalized emergency intervention plans through metabolic risk scoring and early warning mechanisms to effectively reduce the risk of acute complications. Using causal inference technology, the system can accurately identify the causal relationship between behaviors such as diet, exercise and rest and metabolic load, eliminate non-causal factors, focus on key intervention variables, and improve the scientificity and pertinence of intervention effects.

[0047] Through the adaptive learning mechanism, the model parameters are continuously optimized according to the patient's real-time data and long-term characteristics, and the management plan is tailored to the patient's physiological characteristics and living habits to avoid the limitations of universal management plans. Through modules such as the health points system, goal setting and reward mechanism, the system can enhance the patient's compliance and participation in the health management plan, gradually help patients form a healthy lifestyle, and improve long-term management effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a comprehensive management system for diabetic patients according to the present invention.

[0049] Figure 2 The present invention is a flow chart of the method for building a dynamic metabolic load monitoring and prediction model.

[0050] Figure 3 The figure is a flow chart of the behavior and metabolism bidirectional feedback regulation mechanism method of the present invention. DETAILED DESCRIPTION

[0051] The specific implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings.

[0052] Combined with the process Figure 1 As shown in the figure, a comprehensive management system for diabetic patients is centered on the precise classification and hierarchical management of diabetes based on individual metabolic characteristics. First, the system collects multidimensional data of patients, including genomic information (such as genetic risk factors), metabolome data (such as blood glucose metabolites, insulin sensitivity) and life behavior data (such as eating habits, exercise intensity), and constructs a high-dimensional metabolic data model. The model integrates these heterogeneous data in a multimodal manner to capture the potential patterns of patients' metabolic characteristics. In order to classify patients, the system introduces clustering algorithms such as self-organizing map networks (SOM), which can perform adaptive learning and unsupervised grouping of patients based on the intrinsic characteristics of high-dimensional data, thereby identifying different metabolic types, such as insulin resistance, insulin secretion deficiency or mixed metabolic abnormalities. Each metabolic type corresponds to a different pathological mechanism. For example, patients with insulin resistance mainly show a decrease in the efficiency of insulin function, while patients with insulin secretion deficiency show a significant decrease in insulin production capacity. Based on the classification results, the system further develops a tiered management strategy and implements personalized intervention measures for patients. For example, for patients with insulin resistance, a low-carb diet and high-intensity interval training are recommended to improve insulin sensitivity; for patients with insufficient insulin secretion, the type and dosage of insulin supplement drugs are adjusted, and a dietary structure with a risk of hypoglycemia is designed.

[0053] The patient's metabolic load dynamic quantitative indicators are collected. These indicators integrate the patient's multidimensional metabolic data, including blood sugar fluctuation range, insulin sensitivity and daily nutrient intake. The blood sugar fluctuation range is obtained through continuous glucose monitoring equipment (CGM), which can quantify the range and rate of blood sugar changes in patients in different time periods; insulin sensitivity dynamically evaluates the patient's insulin utilization efficiency by analyzing the time series data of insulin dosage and blood sugar response; daily nutrient intake is based on image recognition technology or barcode scanning equipment, and calculates the quantitative data of carbohydrates, proteins and fats consumed by patients in real time, thereby reflecting the direct impact of diet on metabolic load. On this basis, the system uses a time series deep learning algorithm to build a metabolic load prediction model, which can combine the patient's real-time data, including variables such as blood sugar, diet and exercise, to capture the dynamic interaction between these data. By extracting features from time series, the model can not only reflect the direct impact of short-term behavior on metabolic load (such as the acute change of blood sugar caused by a high-carb diet), but also predict the impact of long-term behavior patterns on metabolic trends (such as the improvement of insulin sensitivity caused by continuous exercise).

[0054] A causal relationship model between patient behavior and metabolic response is established. By collecting behavioral data such as diet and exercise, as well as metabolic data such as blood sugar fluctuations and insulin sensitivity, causal inference techniques (such as structured causal models or instrumental variable methods) are used to deeply analyze the direct effects of diet and exercise on metabolic load. By removing potential common variable interference, the model can clarify the causal relationship between behavioral variables (such as exercise intensity and carbohydrate intake) and metabolic indicators (such as blood sugar fluctuation amplitude and hypoglycemia risk), thereby identifying behavioral patterns with key intervention value. Based on this causal relationship model, a bidirectional feedback regulation algorithm for behavior and metabolism is systematically developed, which realizes bidirectional dynamic adjustment through real-time analysis of the patient's metabolic state and behavioral data. At the metabolic level, the algorithm dynamically adjusts the behavioral intervention plan based on real-time metabolic data (such as continuous blood sugar monitoring values ​​or insulin use efficiency), such as recommending patients to optimize their diet structure or adjust their exercise plan to avoid excessive metabolic load. At the behavioral level, the algorithm dynamically updates metabolic model parameters through behavioral data, such as adjusting the weight of insulin sensitivity or blood sugar response rate, thereby improving the adaptability of the model to individual metabolic states. In addition, the system integrates a personalized behavioral incentive module to enhance patient compliance with intervention measures through a health points system, goal setting, and reward mechanism. For example, the system rewards patients with health points based on their completion of goals (such as daily exercise steps and degree of dietary plan achievement), and further encourages patients to continue to maintain good behavioral habits through points redemption for health services or reward mechanisms.

[0055] By collecting patients' metabolic data in real time, including blood sugar, insulin sensitivity, dietary intake, and exercise parameters, and using anomaly detection algorithms such as deep autoencoders, an intelligent model that can identify potential metabolic risk signals is constructed. The deep autoencoder maps high-dimensional data to a low-dimensional latent space by learning the characteristics of the patient's normal metabolic pattern, and detects the difference between the input data and the reconstructed data in the reconstruction stage to identify abnormal signals that deviate from the normal metabolic state, such as hyperglycemia, hypoglycemia, or other metabolic imbalances. To further quantify metabolic risks, the system proposes a metabolic abnormality scoring system that dynamically generates a metabolic risk score by comprehensively evaluating the patient's real-time blood sugar fluctuation amplitude, the frequency and duration of metabolic load exceeding the normal range, and the probability of abnormal events (such as acute hypoglycemia). The scoring system can not only reflect the patient's current metabolic risk level, but also predict future risk trends, thereby providing a scientific basis for early intervention. For high-risk events detected, such as acute hypoglycemia, the system integrates early warning and intervention suggestion modules, and sends alerts to patients and doctors in a timely manner through multi-terminal linkage (such as mobile phone apps, smart watches, or medical monitoring platforms). The early warning system also links with the behavior control module to provide personalized emergency intervention plans based on the patient's real-time metabolic status, such as advising the patient to immediately consume fast-absorbing carbohydrates, adjust insulin dosage, or suspend high-intensity exercise. Through this closed-loop mechanism, the system can not only quickly identify and respond to metabolic abnormalities, but also effectively reduce the probability of acute complications, significantly improving the safety of diabetes management and the quality of life of patients.

[0056] Embodiment 1:

[0057] Combined with the process Figure 2 As shown in the figure, Mr. Zhang is a 45-year-old type 2 diabetic patient weighing 80 kg. He has poor blood sugar control, frequent blood sugar fluctuations and insulin resistance. He began to use a comprehensive management system for diabetic patients. The system tailors a metabolic management plan for Mr. Zhang based on a dynamic metabolic load monitoring and prediction model. The following is the management process of Mr. Zhang by the system, including the complete process of data collection, formula calculation and model application.

[0058] Blood sugar fluctuation range G′(t): Mr. Zhang wears a continuous glucose monitor (CGM) to record his blood sugar level in real time. In the past 24 hours, Mr. Zhang's blood sugar fluctuated between 4mmol / L and 12mmol / L, and the blood sugar change rate was calculated as G′(t) = 1.2mmol / L·h.

[0059] Insulin use I(t): Mr. Zhang injected insulin three times a day, each dose was 6 units. Through monitoring, it was found that the blood sugar dropped by 4mmol / L after insulin use, reflecting that his insulin sensitivity was at a moderate level.

[0060] Nutritional intake N(t): Mr. Zhang recorded his daily dietary intake, including 200 grams of carbohydrates, 80 grams of protein, and 60 grams of fat. Based on the metabolic weights of the nutrients, the system calculated that Mr. Zhang's daily nutritional load was N(t) = 2.5 metabolic units.

[0061] Weight factors w1, w2, w3: According to Mr. Zhang's metabolic characteristics and doctor's advice, the weight factors are set as follows:

[0062] w1 = 0.5: moderate effect of blood sugar fluctuations on metabolic load;

[0063] w2 = 0.3: low to moderate effect of insulin sensitivity on metabolic load;

[0064] w3 = 0.2: Low impact of daily nutrient intake on metabolic load.

[0065] Target blood sugar target : The doctor set the target blood sugar level for Mr. Zhang at 6mmol / L.

[0066] The system calculates Mr. Zhang's metabolic load index MLI(t) for one day based on the formula:

[0067]

[0068] Assuming that a day is 24 hours, the integral from t0 = 0 to t = 24 is calculated as follows:

[0069] 1. Contribution of blood sugar fluctuations:

[0070] ∫0 24 w1·G′(t)dt=0.5·1.2·24=14.4

[0071] 2. Contribution to insulin sensitivity:

[0072] Mr. Zhang's average real-time blood sugar is 8mmol / L, which is far from the target blood sugar level. target =6 is G(t)G target =2mmol / L, calculated as:

[0073]

[0074] 3. Contribution of nutritional intake:

[0075] ∫0 24 w3·N(t)dt=0.2·2.5·24=12.0

[0076] The final metabolic load index is:

[0077] MLI(t)=14.4+2.4+12.0=28.8

[0078] Based on MLI(t)=28.8, the system determines that Mr. Zhang's metabolic load is higher than the ideal range (recommended value MLI<20). The system proposes the following intervention measures:

[0079] Reduce carbohydrate intake to 150 grams and increase dietary fiber intake to stabilize blood sugar fluctuations. It is recommended to do 30 minutes of moderate-intensity exercise (such as brisk walking) every day to enhance insulin sensitivity and reduce MLI. According to the characteristics of blood sugar fluctuations, adjust the insulin dosage plan and increase the evening insulin dose to reduce nighttime blood sugar fluctuations.

[0080] Mr. Zhang re-evaluated his MLI(t) after one week of intervention. The blood sugar fluctuation range decreased to G′(t)=0.8mmol / L·h, the efficiency of insulin use improved, and the deviation from the target blood sugar level decreased to G(t)G tar get =1mmmol / L, the nutritional load is reduced to N(t)=2.0 metabolic units. The calculation results show that the new MLI(t)=18.0, and the metabolic load is significantly improved. The system will continue to dynamically track and optimize the plan to ensure that Mr. Zhang's metabolic state remains within a stable range.

[0081] After a week of intervention, Mr. Zhang's metabolic load index MLI(t) dropped to 18.0, but further optimization of management is still needed to stabilize the metabolic state. The system uses Mr. Zhang's real-time data to predict his future metabolic load trend through an improved time series deep learning model. The formula is:

[0082]

[0083] in:

[0084] P(t+Δt): predicted metabolic load value at time t+Δt. Transformer-based time series deep learning model, parameter θ is obtained through training. D(t): metabolic load characteristics at time t, including blood sugar fluctuation G′(t) = 0.8mmol / L·h, insulin sensitivity I(t) = 6 units, nutritional load N(t) = 2.0 metabolic units. B(t): behavioral data input, including exercise intensity S(t) = 0.5 (moderate intensity exercise), diet time distribution is evenly distributed among three meals, and rest mode R(t) = 7 hours.

[0085] The system collected Mr. Zhang's metabolic data D(t) and behavioral data B(t) for a week to form a time series input. The specific values ​​are as follows:

[0086] D(t)={G′(t),I(t),N(t)}={0.8,6,2.0}.

[0087] B(t)={S(t),R(t)}={0.5,7}.

[0088] Through the embedding layer of the time series deep learning model, metabolic features and behavioral data are integrated to extract time-dependent features.

[0089] The model is trained using data from the past week, with the initial value range of the parameter θ being [0.1, 1.0]. After iterative optimization, it finally converges to:

[0090] Blood sugar fluctuation weight: θ1 = 0.7

[0091] Insulin sensitivity weight: θ2 = 0.5

[0092] Nutritional load weight: θ3 = 0.4

[0093] Behavior weight: θ4 = 0.6

[0094] These parameters reflect the proportion of factors affecting Mr. Zhang's current metabolic state.

[0095] The model calculates the predicted metabolic load for the next 4 hours (Δt=4) based on the real-time inputs D(t) and B(t). The formula is:

[0096]

[0097] Substitute the data into:

[0098]

[0099] Among them, ∈ = 0.2 is the system noise term. The calculation result is:

[0100] P(t+4)=0.56+0.083+0.8+0.3+0.2=1.943.

[0101] The system predicts that Mr. Zhang's metabolic load index in the next 4 hours will be 1.943.

[0102] Based on the prediction results, the system determined that Mr. Zhang's metabolic load was at a critical value and recommended the following personalized interventions:

[0103] 1. Dietary adjustment: Reduce carbohydrate intake at dinner to 40 grams and increase protein intake to stabilize blood sugar fluctuations.

[0104] 2. Exercise adjustment: Do 20 minutes of light exercise after dinner, such as slow walking.

[0105] 3. Rest adjustment: It is recommended to extend sleep time to 8 hours to improve insulin sensitivity.

[0106] After Mr. Zhang followed the advice, the system re-collected data and found that P(t+8) in the next cycle dropped to 1.5, indicating that the intervention measures were effective.

[0107] This embodiment uses real-time fusion of blood sugar, diet and exercise data to predict the change trend of future metabolic load, and introduces an explanatory analysis module to identify the main factors affecting the predicted changes through reverse tracing technology. The following is an application example and calculation description based on Mr. Zhang's actual data.

[0108] The system continuously monitors Mr. Zhang's metabolic characteristic data D(t) and behavioral data B(t), and performs multimodal fusion. The real-time data for the past 4 hours includes:

[0109] Blood sugar data: Continuous blood sugar monitoring shows a blood sugar change rate of G′(t) = 0.6mmmol / L·h, current blood sugar G(t) = 7.2mmol / L, target blood sugar G target =6.0mmol / L. Dietary data: Mr. Zhang consumes 60g of carbohydrates, 30g of protein, and 20g of fat. The calculated nutritional load is N(t)=1.8.

[0110] Exercise data: Mr. Zhang completed a low-intensity walk with exercise intensity S(t)=0.4.

[0111] The system uses this data to predict in real time the changes in metabolic load in the future t+Δt=t+4 hours.

[0112] The future trend of metabolic load is expressed by the formula:

[0113]

[0114] in:

[0115] F future (t+Δt): Predicted metabolic load value for the next 4 hours. P(t+Δt): Basic prediction output, calculated by the time series deep learning model in the previous stage, with a value of P(t+Δt)=2.0. Dynamic adjustment value, based on metabolic load change rate ω(D(t)) and weight coefficient α.

[0116] Set the model parameters as follows:

[0117] The weight coefficient α = 0.6, indicating that Mr. Zhang is moderately sensitive to behavioral changes.

[0118] The metabolic load change rate ω(D(t)) = 0.5, combining the dynamic effects of diet and exercise on metabolism.

[0119] Substitute the data and calculate:

[0120] ∫t t+Δt α·ω(D(t))dt=α·ω(D(t))·Δt=0.6·0.5·4=1.2.

[0121] Final prediction of future metabolic load:

[0122] F future (t+Δt)=P(t+Δt)+1.2=2.0+1.2=3.2.

[0123] The forecast results show that Mr. Zhang's metabolic load will rise to 3.2 in the next 4 hours, approaching the warning range (above 3.5).

[0124] The system uses reverse tracing technology to identify the main factors that lead to increased metabolic load. The results show:

[0125] 1. Blood sugar deviates from target level (G(t)G target =1.2mmol / L) has the greatest impact on metabolic load, with a weighted contribution of 50%.

[0126] 2. Nutritional load N(t) = 1.8 times higher, contributing 30% to the metabolic load.

[0127] 3. Low-intensity exercise S(t)=0.4 has limited effect on improving metabolism, with a contribution rate of only 20%.

[0128] Based on the analysis results, the system makes the following recommendations:

[0129] Reduce carbohydrate intake to 40 grams at the next meal and increase dietary fiber to alleviate blood sugar rise. It is recommended to add a moderate intensity exercise (such as 15 minutes of brisk walking) to increase the exercise intensity to S(t) = 0.6. If necessary, adjust the evening insulin dose to ensure that blood sugar remains within the target range.

[0130] After Mr. Zhang implemented the intervention measures as recommended, the system re-predicted the metabolic load for the next cycle t+8 hours, and the calculation results showed that F future (t+8)=2.5, which is significantly lower than the current predicted value of 3.2, indicating that the intervention measures are effective.

[0131] Embodiment 2:

[0132] Combined with the process Figure 3 As shown in the figure, Mr. Zhang is a 50-year-old type 2 diabetes patient weighing 85 kg. He has problems with excessive dietary carbohydrate intake and lack of regular exercise in his daily life. In order to optimize metabolic load management, the system activates a bidirectional feedback regulation mechanism between behavior and metabolism. By collecting his behavioral and metabolic data over a long period of time, a causal relationship network is established, and the intervention strategy is dynamically adjusted.

[0133] The system collected Mr. Zhang’s behavior data for the past 30 days, including:

[0134] Dietary data: Average daily carbohydrate intake is 250 grams.

[0135] Exercise data: The exercise intensity is low, with an average daily step count of 3,000 steps (equivalent to light exercise).

[0136] Sleep data: The average sleep time per night is 6 hours, which means there is a certain degree of sleep deficiency.

[0137] Metabolic data include blood sugar fluctuations, insulin sensitivity, and daily metabolic load index (MLI). Through causal inference technology, a causal relationship network between Mr. Zhang's behavior and metabolic load was systematically constructed, and the following key intervention variables were identified: exercise intensity B1(t) has a significant positive effect on metabolic load, with a causal influence coefficient of α1=0.7; carbohydrate intake B2(t) has a large negative effect on metabolic load, with a causal influence coefficient of α2=-0.6; common dependent variables C(t) (such as lack of sleep, stress) have a moderate interference effect on metabolic load, with an inhibition coefficient of β=0.5.

[0138] The quantitative formula for the causal relationship between behavior and metabolic load is:

[0139] ΔM=∫0 T (α1·B1(t)+α2·B2(t)β·C(t)) d t.

[0140] The system is based on the data of the past 7 days (T=7 days) and substitutes Mr. Zhang’s average behavior and metabolic characteristics:

[0141] B1(t) = 0.4 (mild exercise intensity);

[0142] B2(t) = 250 g / day;

[0143] C(t) = 0.6 (sleep deprivation index, full score 1.0).

[0144] Substituting the parameters into the formula:

[0145] ΔM=∫0 7 (0.7 0.4 0.6 250 0.5 0.6) d t.

[0146] Calculate item by item:

[0147] 0.7 0.4 = 0.28;

[0148] -0.6 250 = -150.0;

[0149] -0.5·0.6=-0.3.

[0150] therefore:

[0151] ΔM=∫0 7 (-150.02) d t = -1050.14.

[0152] The negative result shows that Mr. Zhang's metabolic load status continues to deteriorate, mainly caused by excessive carbohydrate intake and lack of sleep. Based on the causal network and model calculation results, the system generates the following intervention strategies:

[0153] Reduce daily carbohydrate intake to 180 grams (a 28% reduction) and increase the proportion of dietary fiber to stabilize blood sugar.

[0154] Add one moderate-intensity exercise session per day (such as brisk walking for 30 minutes) to increase the exercise intensity B1(t) to 0.6.

[0155] By adjusting the work and rest schedule, we ensured that Mr. Zhang slept at least 7 hours a day and lowered C(t) to 0.4.

[0156] After one week, the system collects data again and updates behavioral and metabolic indicators:

[0157] B1(t)=0.6;

[0158] B2(t)=180;

[0159] C(t)=0.43002

[0160] Substitute into the formula to calculate:

[0161] ΔM=∫0 7 (0.7 0.6 0.6 1800.5 0.4) d t.

[0162] Calculate item by item:

[0163] 0.7 0.6 = 0.42;

[0164] -0.6 180 = -108.0;

[0165] -0.5·0.4=-0.2.

[0166] therefore:

[0167] ΔM=∫0 7 (-107.78) d t = -754.46.

[0168] The results showed that Mr. Zhang's metabolic load decreased by about 28%, a significant improvement. Feedback model analysis showed that dietary adjustments contributed about 60% of the improvement, increased exercise contributed 30%, and sleep optimization contributed 10%. Through this causal inference model and two-way feedback mechanism, the system successfully identified the key metabolic influencing factors in Mr. Zhang's behavior and significantly reduced the metabolic load through personalized intervention. This closed-loop management model is highly scientific and feasible, providing reliable health management support for diabetic patients while improving patients' intervention compliance and quality of life.

[0169] Mr. Zhang's metabolic load has improved in the past two weeks of intervention, but blood sugar fluctuations are still large after a high-carb diet. The system starts a behavior and metabolism interaction analysis model, and optimizes the behavioral intervention strategy for Mr. Zhang through real-time monitoring and calculation.

[0170] The system collected multimodal data of Mr. Zhang in the past 48 hours: Dietary intake data: Mr. Zhang's carbohydrate intake in the past 24 hours was 180 grams and 220 grams, and his protein intake was 70 grams and 80 grams.

[0171] Exercise parameters: The exercise in the past two days was 40 minutes of moderate-intensity brisk walking and 30 minutes of jogging.

[0172] Metabolic change data: Continuous blood glucose monitoring data showed that the blood glucose fluctuation range G′(t) was 1.0mmol / L·h and 1.2mmol / L·h respectively, and the metabolic load state M(t) fluctuated between 2.5 and 3.8 in two days. Through multimodal data fusion technology, the system synchronously models the relationship between diet, exercise and metabolism, generating behavioral data input B(t).

[0173] The system uses a dynamic causal graph to analyze Mr. Zhang's behavior and metabolic interactions. The metabolic interaction pattern is described by the following formula:

[0174]

[0175] in:

[0176] R(t) is the real-time response rate of behavior to metabolic load;

[0177] M(t) is the metabolic load state, currently M(t) = 3.2;

[0178] B(t) is the behavioral data input, including dietary intake and movement parameters;

[0179] γ = 0.7 (immediate response coefficient, ranging from 0.5 to 1.0), reflecting the intensity of the short-term impact of behavior on metabolic load;

[0180] η=0.6 (long-term behavior impact weight, value range is 0.3, 0.8);

[0181] φ(B(t), t) is a nonlinear relationship function between behavior and metabolism, which is used to describe the delayed effect of exercise on blood glucose.

[0182] Substitute Mr. Zhang’s data:

[0183] Carbohydrate intake B1(t) = 200 g (average for two days);

[0184] Exercise intensity B2(t)=0.6 (weighted intensity of two-day exercise).

[0185] Calculate the immediate response:

[0186]

[0187] in:

[0188]

[0189] Calculated:

[0190]

[0191] Calculate the long-term response:

[0192]

[0193] Where φ(B(t),t)=0.04·B1(t)0.03·B2(t), substitute:

[0194] φ(B(t),t)=0.04·2000.03·0.6=7.986.

[0195] Calculated:

[0196]

[0197] The final metabolic interaction model response rate was:

[0198] R(t)=2.8+9.583=12.383.

[0199] The system analysis results show that Mr. Zhang's high-carb diet is the main factor leading to increased metabolic load, contributing 70%; exercise has limited effect on improving metabolic load, contributing only 20%, and the remaining 10% is determined by other behavioral factors (such as rest patterns). Based on this analysis, the system provides the following health recommendations:

[0200] Reduce daily carbohydrate intake to 150 grams and increase the proportion of protein to enhance satiety and reduce blood sugar fluctuations. Extend the daily moderate-intensity exercise time to 60 minutes (such as a combination of brisk walking and light running) and spread it out over two periods of time to reduce acute changes in blood sugar. Ensure at least 7 hours of high-quality sleep every day to reduce metabolic load interference caused by lack of sleep.

[0201] After Mr. Zhang implemented the suggestions, the system re-monitored his metabolic load over the next two days and found that M(t) stabilized at 2.8, the contribution of reduced carbohydrate intake to metabolic load dropped to 50%, and the contribution of improved exercise increased to 40%. The prediction results show that further optimization of behavioral intervention will help maintain metabolic stability in the long term. Through real-time modeling of behavioral and metabolic interaction patterns, the system successfully identified the main metabolic influencing factors in Mr. Zhang's behavior and significantly improved the metabolic load status through personalized intervention.

[0202] In order to further optimize the intervention effect, the system of this embodiment starts a two-way feedback control algorithm based on causal inference and real-time interactive mode, realizes personalized metabolic management by dynamically adjusting the behavioral intervention plan and updating the metabolic model parameters, and introduces an abnormal event early warning mechanism to quickly respond to potential risks.

[0203] In the past 24 hours, Mr. Zhang's real-time data is as follows:

[0204] Current behavior data B current (t):

[0205] Diet: Carbohydrate intake is 200 grams;

[0206] Exercise: Walking intensity is low, with about 4000 steps per day (intensity S(t) = 0.4).

[0207] Current metabolic data M(t): metabolic load state is 3.5 (higher than the ideal range, target metabolic load M target =2.5). Abnormal event monitoring: The system detected a large fluctuation in blood sugar at night (fluctuation amplitude A(t) = 1.8mmol / L), which is a high-risk event.

[0208] Based on the above data, the system substitutes the following feedback control equation for intervention optimization:

[0209] B new (t) = B current (t)+k·(M target M(t))+δ·ψ(A(t)).

[0210] The system sets feedback control parameters based on Mr. Zhang’s individual characteristics:

[0211] k = 0.6 (feedback gain coefficient, range 0.5, 1.0), used to control the effect of metabolic load deviation on intervention intensity;

[0212] δ = 0.4 (abnormal response coefficient, range 0.3, 0.8), used to quickly respond to abnormal events;

[0213] ψ(A(t))=1.5·A(t), abnormal event adjustment function, reflects the intensity of the impact of hyperglycemia fluctuations on behavioral intervention.

[0214] Substituting the data into the formula:

[0215] B new (t) = B current (t)+k·(M target M(t))+δ·ψ(A(t)).

[0216] Calculate item by item:

[0217] Dietary intervention (carbohydrate intake adjustment):

[0218] Current intake B current (t) = 200 g;

[0219] k·(M tar get M(t)) = 0.6·(2.53.5) = -0.6·1.0 = -0.6 g;

[0220] δ·ψ(A(t))=0.4·(1.5·1.8)=0.4·2.7=1.08 grams.

[0221] Adjusted carbohydrate intake:

[0222] B new (t) = 2000.6 + 1.08 = 200.48 g.

[0223] Exercise intervention (walking intensity adjustment):

[0224] Current strength S(t) = 0.4;

[0225] k·(M target M(t))=0.6·(2.53.5)=-0.6;

[0226] δ·ψ(A(t))=0.4·(1.5·1.8)=1.08.

[0227] Adjusted exercise intensity:

[0228] S new (t) = 0.40.6 + 1.08 = 0.88.

[0229] Based on the above calculation results, the system generates the following specific suggestions for Mr. Zhang:

[0230] Control the carbohydrate intake of dinner to less than 200 grams, and increase the intake of dietary fiber to stabilize blood sugar fluctuations. Add 30 minutes of brisk walking after dinner and increase the walking intensity to moderate intensity (S(t)=0.88). Based on abnormal blood sugar fluctuations, it is recommended to test blood sugar at night and supplement low GI carbohydrates appropriately to avoid the risk of hypoglycemia.

[0231] After the intervention, the system re-collected Mr. Zhang's metabolic data and found that the metabolic load M(t) dropped to 2.8 and the blood sugar fluctuation range A(t) dropped to 1.2mmol / L. Substituting into the formula for verification:

[0232] B new (t) = B current (t)+k·(M target M(t))+δ·ψ(A(t)).

[0233] Then calculate:

[0234] After diet adjustment new (t) = 2000.6 + 0.48 = 199.88 g;

[0235] Sport-tuned S new (t)=0.40.42+0.72=0.7.

[0236] Data verification shows that the model results are consistent with actual feedback, indicating that the system intervention measures are accurate and effective.

[0237] Through this two-way feedback control mechanism, the system successfully dynamically adjusted Mr. Zhang's behavioral intervention plan, and used the abnormal event warning mechanism to quickly respond to the risk of blood sugar fluctuations. The results showed that the intervention plan not only reduced Mr. Zhang's metabolic load, but also effectively improved blood sugar fluctuations, further verifying the scientificity and practicality of the plan. This management method based on real-time feedback and causal inference can significantly improve the health management effect of diabetic patients, while improving patients' compliance and quality of life.

Claims

1. A comprehensive management system for diabetic patients, characterized in that The following steps are involved: S1. Accurate classification and stratified management of diabetes based on individual metabolic characteristics: S1.

1. Build a high-dimensional metabolic data model to integrate genomic information, metabolomics data, and life behavior data in a multimodal manner to explore the potential patterns of patients' metabolic characteristics; S1.

1. Use clustering algorithms including self-organizing map networks to classify patients and identify different metabolic types including insulin resistance and insulin secretion deficiency; S1.

3. Develop tiered management strategies based on the classification results, including diet, exercise and drug intervention plans for different classifications; S2. Dynamic metabolic load monitoring and prediction model: S2.

1. Collect dynamic quantitative indicators of metabolic load, integrating the patient's blood sugar fluctuation range, insulin sensitivity and daily nutritional intake; S2.

2. Construct a metabolic load prediction model to measure future metabolic load by combining patients’ real-time data including blood sugar, diet, and exercise through a time series deep learning algorithm; S3. Bidirectional feedback regulation mechanism between behavior and metabolism: S3.

1. Establish a causal relationship model between patient behavior and metabolic response, and use causal inference techniques to analyze the interaction pattern between diet, exercise and metabolic load; S3.

2. Develop a bidirectional feedback control algorithm between behavior and metabolism to adjust the behavior intervention plan according to the real-time metabolic data and update the metabolic model parameters according to the behavior data; S3.

3. Integrate personalized behavioral incentive modules to enhance patient compliance with intervention measures using health points system, goal setting and reward mechanism; S4. Intelligent diagnosis and early warning system for metabolic abnormalities: S4.

1. Identify potential risk signals in real-time metabolic data, including hyperglycemia, hypoglycemia, or other metabolic imbalances, based on anomaly detection algorithms including deep autoencoders; S4.

2. Propose a metabolic abnormality scoring system to dynamically quantify metabolic risk levels; S4.

3. For high-risk events including acute hypoglycemia, an integrated early warning and intervention recommendation system is used to link the behavior control module to provide emergency plans.

2. A comprehensive management system for diabetic patients according to claim 1, characterized in that The dynamic metabolic load monitoring and prediction model construction method includes: obtaining the blood sugar fluctuation amplitude as a time series function through a continuous blood sugar monitoring device; dynamically evaluating insulin sensitivity in combination with the patient's insulin usage and blood sugar response law; collecting nutrient intake in real time through a diet recording device to quantify the impact of daily diet on metabolic load; the expression formula is: Among them, MLI(t) represents the metabolic load index at time t; w1, w2, w3 are the weight factors of blood glucose fluctuation amplitude, insulin sensitivity and daily nutritional load, which represent the contribution ratio of each index to metabolic load; G′(t) is the rate of change of blood glucose over time, reflecting the dynamic characteristics of short-term blood glucose fluctuation; G(t) is the real-time blood glucose value, G target is the target blood sugar level, and the difference between the two represents the degree to which the current blood sugar deviates from the healthy target; I(t) is the real-time insulin dosage, which measures the patient's responsiveness to insulin; N(t) is the daily nutritional load, which is a weighted combination of nutrients such as carbohydrates, proteins, and fats.

3. A comprehensive management system for diabetic patients according to claim 2, characterized in that The method for constructing a dynamic metabolic load monitoring and prediction model includes: using an improved time series deep learning model to integrate blood sugar fluctuations, exercise and dietary behavior data to generate time-dependent features; and dynamically adjusting the prediction model parameters to adapt to individual differences in patients through an adaptive learning mechanism.

4. A comprehensive management system for diabetic patients according to claim 3, characterized in that The dynamic metabolic load monitoring and prediction model construction method includes: predicting the future metabolic load change trend by integrating blood sugar, diet and exercise data in real time; introducing an explanatory analysis module, using reverse tracing technology to identify the main influencing factors that cause the predicted changes, and providing transparent health advice; the expression formula is: F future (t+Δt)=P(t+Δt)+∫ t t+Δt α·ω(D(t))dt, Among them, F future (t+Δt) is the predicted metabolic load value at the future time t+Δt; P(t+Δt) is the basic output of the predicted value; The dynamic adjustment value during the prediction period is calculated based on the metabolic load change rate ω(D(t)), and the weight coefficient α reflects the sensitivity of the patient's metabolic load to behavioral changes.

5. A comprehensive management system for diabetic patients according to claim 1, characterized in that The behavior and metabolism bidirectional feedback regulation mechanism method includes: by collecting patients' long-term behavior data including diet, exercise and sleep and metabolic data, using causal inference technology including structured causal model to establish a causal relationship network between behavior and metabolic load; defining the quantitative relationship between key intervention variables including exercise intensity and carbohydrate intake and target metabolic indicators, and removing non-causal associated variables; and the causal relationship between behavior and metabolic load is quantified as: ΔM=∫0 T (α1·B1(t)+α2·B2(t)β·C(t))d t , Among them: ΔM is the cumulative change of metabolic load, which indicates the change of metabolic state of the patient in time period T; B1(t) is the exercise intensity of the patient at time t; B2(t) is the dietary carbohydrate intake of the patient at time t; C(t) is the interference of the common dependent variable on the metabolic load at time t; α1 and α2 are the causal influence coefficients of exercise and diet on metabolic load; β is the inhibition coefficient of the common dependent variable, which indicates the intensity of the negative effect.

6. A comprehensive management system for diabetic patients according to claim 5, characterized in that The behavior and metabolism bidirectional feedback regulation mechanism method includes: Through multimodal data fusion technology, dietary intake, exercise parameters and metabolic changes are modeled synchronously, and the interactive relationship between behavior and metabolism is updated in real time. Dynamic causal graphs are used to identify the immediate impact and long-term trend of short-term behavior on metabolism in real time. The metabolic interaction pattern analysis model is: Where: R(t) is the real-time response rate of behavior to metabolic load at time t; M(t) is the metabolic load state at time t; B(t) is the behavioral data input at time t; γ is the immediate response coefficient, which represents the short-term impact intensity of behavior on metabolic load; η is the weight of the impact of long-term behavior on metabolic load; φ(B(t), t) is the nonlinear relationship function between behavior and metabolism, which is used to capture the interactive pattern including the delayed effect of exercise on blood glucose.

7. A comprehensive management system for diabetic patients according to claim 6, characterized in that The behavior and metabolism bidirectional feedback regulation mechanism method includes: Based on causal inference and real-time interaction pattern analysis, a two-way feedback control algorithm is developed; the patient's behavioral intervention plan is dynamically adjusted according to real-time metabolic data, and the metabolic model parameters are updated according to the behavioral data; and an abnormal event warning mechanism is introduced; and the behavior and metabolism two-way feedback control equation is: B new (t)=B current (t)+k·(M target M(t))+δ·ψ(A(t)), in: B new (t) is the behavioral intervention adjustment value after feedback at time t; B current (t) is the current behavior data at time t; M target is the target metabolic load value; M(t) is the real-time metabolic load state at time t; k is the feedback gain coefficient, which controls the intervention intensity; δ is the abnormal response coefficient; ψ(A(t)) is the abnormal event adjustment function, which dynamically corrects the behavior based on the abnormal event A(t).

Images