Diabetic complication disease process intelligent deduction service method and system

Abstract

The invention relates to the technical field of medical prediction, in particular to an intelligent deduction service method and system for the morbidity process of diabetic complications, and the method comprises the steps: collecting continuous blood glucose monitoring, heart rate variability and skin electrical activity data of a user; based on the heart rate variability and the skin electrical activity data, an emotional pressure index for quantifying the autonomic nerve excitation degree is calculated; inputting the continuous blood glucose data and the emotional pressure index into a deduction rule knowledge base containing a medical production rule for reasoning; obtaining brain function risk and / or blood sugar control difficulty assessment; and generating visual report feedback containing personalized pressure-reducing non-drug intervention suggestions. According to the method, multiple data are fused, reasoning is conducted through a medical rule base, the problems that single-index early warning lags and machine learning black boxes lack interpretability are solved, brain function risks are prospectively deduced, the influence of pressure on blood glucose control is quantified, adverse circulation is broken, and intervention suggestions with physiological bases are provided.

Description

Technical Field

[0001] This invention relates to the field of medical prediction technology, and more specifically, to a method and system for intelligently predicting the pathogenesis of diabetic complications. Background Technology

[0002] Diabetes, a prevalent chronic metabolic disease worldwide, poses a core threat not so much from the disease itself, but from the various complications resulting from long-term or rapid loss of blood sugar control. Clinical data and pathological studies have clearly shown that both type 1 and type 2 diabetes can induce acute and chronic complications, with only slight differences in the specific scenarios or timing of their occurrence. However, "the type does not determine whether complications will occur, but only the manifestation of uncontrolled blood sugar." Monitoring any type can provide crucial information for predicting the progression of complications. The core trigger for acute complications (such as diabetic ketoacidosis (DKA) and hyperosmolar hyperglycemic state (HHS)) is a rapid increase in blood glucose and failure of insulin regulation in a short period of time. This process can occur in both type 1 and type 2 diabetes. Chronic complications (such as diabetic retinopathy, diabetic nephropathy, diabetic cardiovascular disease, diabetic neuropathy, diabetic foot, and stroke) are the result of "long-term cumulative damage from hyperglycemia" and are not absolutely related to diabetes type. The only difference is that type 1 diabetes is more common in adolescents (with a longer course) and type 2 diabetes is often accompanied by metabolic abnormalities (such as hypertension and hyperlipidemia), so there may be slight differences in the timing of complication onset. However, the essence of both needs to be predicted by monitoring the long-term blood glucose status of patients with different types. Whether monitoring blood glucose fluctuations (such as sudden spikes in real-time blood glucose or abnormal insulin dosage) in patients with type 1 or type 2 diabetes, the risk of acute and chronic complications can be identified at an early stage. However, current clinical risk prediction for diabetic complications mainly relies on static indicators such as glycated hemoglobin (HbA1c) and fasting blood glucose. However, HbA1c cannot reflect the sudden rise in blood glucose caused by fluctuations in insulin dosage in patients with type 1 diabetes, nor can it reflect the peak changes in postprandial blood glucose in patients with type 2 diabetes. This results in prediction lag and insufficient accuracy, making it difficult to support the early prediction of complication progression. While some systems have begun to use machine learning models for prediction, these "black box" models lack interpretability, making it difficult for doctors and patients to understand the specific physiological mechanisms behind the risks, thus affecting the targeting of interventions.

[0003] Furthermore, traditional methods rarely focus on the synergistic negative impact of autonomic nervous system status (as an objective quantitative indicator of emotional stress) and blood glucose fluctuations on brain function and health, resulting in insufficient early warning capabilities for diabetic central nervous system complications (such as mood disorders, cognitive decline, and epilepsy risk).

[0004] In view of this, a smart prediction service method and system for the pathogenesis of diabetic complications is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a smart prediction service method and system for the pathogenesis of diabetic complications, so as to solve the problems of delayed prediction of diabetic complication risks, insufficient early warning capabilities, and insufficient targeting of intervention measures.

[0006] To solve the above-mentioned technical problems, the present invention provides a method for intelligently predicting the pathogenesis of diabetic complications, comprising the following steps: S1. Collect the user's physiological data, including continuous blood glucose monitoring data, heart rate variability data, and skin conductance data; S2. Based on heart rate variability and skin conductance data, an emotional stress index is calculated to quantify the level of excitation of the user's autonomic nervous system. S3. Input the continuous blood glucose monitoring data and emotional stress index into the preset deduction rule knowledge base for logical reasoning; the deduction rule knowledge base contains several production rules based on medical knowledge, including a first rule subset for deducing brain function health risks and a second rule subset for deducing the difficulty of blood glucose control. S4. Based on the results of logical reasoning, deduce the user's specific brain function health risk status and / or blood glucose control difficulty assessment related to diabetes; S5. Generate a visualization report of the deduction results and provide feedback to the user; the visualization report also includes providing the user with personalized, non-pharmacological intervention suggestions for reducing the emotional stress index.

[0007] As a further improvement to this technical solution, in step S2, the emotional stress index is calculated using the following formula: ,in, The standard deviation of heart rate variability; The change in skin conductance level relative to baseline is expressed by the formula: , This serves as a baseline reference value for each user. and These are weighting coefficients. and The value of is determined through machine learning training and optimization on historical data samples, where The value range is from 0.4 to 0.7. The value range is from 0.3 to 0.6, and satisfies the following condition: .

[0008] As a further improvement to this technical solution, in step S3, the first rule subset includes inference rules based on amygdala functional activity; The inference rule is defined as follows: when continuous blood glucose monitoring data indicates... When the emotional stress index is ≥0.6, the inference rule is triggered, and it is presumed that the amygdala's functional activity is abnormal.

[0009] As a further improvement to this technical solution, abnormal amygdala functional activity is a risk of hyperfunction. The criteria for determining the risk of hyperfunction are: the average low-frequency to high-frequency power ratio in heart rate variability increases by more than 25% from the baseline value within one week after the trigger rule.

[0010] As a further improvement to this technical solution, in step S4, if the risk of amygdala hyperfunction is deduced, then the increased risk of epileptic seizures is deduced according to the following rules: When a state of amygdala hyperfunction persists for more than 4 weeks, and continuous blood glucose monitoring data shows ≥2 hypoglycemic events during this period, increased excitability of the limbic system is presumed, indicating an increased risk of seizures. The hypoglycemic events are defined as follows: .

[0011] As a further improvement to this technical solution, in step S3, the second rule subset is used to infer the negative impact of emotional stress on the difficulty of future blood sugar control. The derivation rule for the second rule subset is: if If the condition lasts for more than 4 weeks, the output projection conclusion is that the probability of achieving blood sugar target within the next 1 to 3 months will decrease by more than 30%, where achieving blood sugar target means that the user's glucose target range is exceeded for more than 70% of the time.

[0012] A smart prediction service system for the pathogenesis of diabetic complications, wherein the smart prediction service system for the pathogenesis of diabetic complications is used to implement the above-mentioned smart prediction service method for the pathogenesis of diabetic complications, including: The data acquisition module is used to collect users' physiological data from wearable devices and smart terminals; The data processing module is used to calculate the emotional stress index based on heart rate variability and skin conductance data; The intelligent deduction module contains a deduction rule knowledge base, which is used to perform logical reasoning based on production rules and output deduction results. Among them, the production rules based on medical knowledge, the first rule subset for deducing brain function health risks, and the second rule subset for deducing the difficulty of blood sugar control in the deduction rule knowledge base are constructed by structurally extracting medical literature and clinical expert experience, and retrospectively analyzing and verifying them using historical patient data. The user interaction module is used to generate visual reports, provide personalized intervention suggestions, and display feedback information to users.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This intelligent prediction service method and system for the pathogenesis of diabetic complications creatively integrates continuous blood glucose monitoring data with the Emotional Stress Index (EPSI) for quantifying autonomic nervous function. Based on medical mechanisms, it constructs a hierarchical and progressive production rule base for logical reasoning. This effectively overcomes the shortcomings of existing technologies, such as the lag in early warning caused by relying on a single biochemical indicator, and the lack of interpretability and inability to guide clinical intervention in machine learning "black box" models. It can not only prospectively predict brain function health risks (such as amygdala dysfunction and increased risk of epileptic seizures), but also quantify the negative impact of emotional stress on the difficulty of blood glucose control. Thus, it provides patients with personalized non-drug intervention suggestions with clear physiological basis, breaking the vicious cycle of "poor blood glucose control → high stress → worse blood glucose".

[0014] 2. In the intelligent inference service method and system for the pathogenesis of diabetic complications, by collecting heart rate variability (HRV) and electrical skin activity (EDA) data and using a machine learning-optimized formula to calculate the emotional stress index (EPSI), the system overcomes the recall bias and strong subjectivity problems of traditional subjective questionnaires, and provides reliable and objective quantitative input for subsequent inference.

[0015] 3. The intelligent simulation service method and system for the pathogenesis of diabetic complications adopts production rules rather than a "black box" model, so that each risk conclusion is supported by clear physiological rules and data thresholds that can be understood by doctors and patients, which enhances the reliability and credibility of the results and facilitates clinical adoption and application.

[0016] 4. In this intelligent prediction service method and system for the pathogenesis of diabetic complications, the rule base design simulates the thinking of clinical experts, which can reveal the deep connections between different physiological indicators. The rules based on multi-dimensional cross-validation improve the accuracy of early warning of brain function risks. It not only warns of "functional abnormalities", but also further quantifies "hyperfunction" through standards such as the low-frequency to high-frequency power ratio in heart rate variability. It combines short-term abnormalities with long-term persistent states to deduce the more serious clinical consequence of "increased risk of epileptic seizures", providing clinicians with clear and targeted diagnostic clues and intervention basis. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall method of the present invention; Figure 2 This is an overall system block diagram of the present invention. Detailed Implementation

[0018] 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.

[0019] Example 1 Currently, existing technologies for predicting the risk of diabetic complications largely rely on single biochemical indicators (such as HbA1c) or imaging examinations, which are subject to lag. Although some systems have begun to use machine learning models for prediction, these "black box" models lack interpretability, making it difficult for doctors and patients to understand the specific physiological mechanisms behind the risks, thus affecting the targeting of interventions. In addition, traditional methods rarely pay attention to the synergistic negative impact of autonomic nervous system status (as an objective quantitative indicator of emotional stress) and blood glucose fluctuations on brain function, resulting in insufficient early warning capabilities for diabetic central nervous system complications (such as mood disorders, cognitive decline, and epilepsy risk). In view of this, please refer to Figure 1 As shown, one of the objectives of this invention is to provide a method for intelligently predicting the pathogenesis of diabetic complications. This method includes the following steps: S1. Collect the user's physiological data, including continuous blood glucose monitoring data, heart rate variability data, and skin conductance data; Among them, the continuous glucose monitoring (CGM) device used is Abbott FreestyleLibre3, with a sampling frequency of 15 minutes / time, a blood glucose measurement range of 2.2 to 22.2 mmol / L, and an error control within ±10% (compliant with ISO15197:2013 standard), to ensure the dynamic nature and accuracy of blood glucose data; The heart rate variability (HRV) data acquisition device was an Apple Watch Series 9, with a sampling frequency of 5 minutes / time. Only "resting state" data was collected (judgment criteria: steps per minute < 50, excluding temporary interference such as exercise and emotional excitement). The focus was on collecting the "sinus RR interval standard deviation (SDNN)" (range 0-500ms) in HRV. The FitbitSense2 was selected as the electrical activity of the skin (EDA) acquisition device, with a sampling frequency of 1 minute / time. Abnormal values ​​were filtered out simultaneously (judgment criteria: skin conductance level SCL < 0.5 μS or > 100 μS, excluding environmental interference such as dry skin and excessive sweating). The smart terminal uses the mobile phone that users use every day, and receives data from the above three types of devices in real time through the device's native API (such as Apple HealthKit, Fitbit SDK); The mobile device stores data in a structured format of "User ID-Collection Time-Indicator Value" through a local MySQL database. A data detection cycle is set to 1 minute: every minute, each device is checked for new data. If new data is found, it is immediately stored in the local database and simultaneously uploaded to the cloud database (to avoid local data loss). If the device is offline, the local database will cache the data of the last 24 hours and automatically re-upload it after the network is connected to ensure data integrity. Through the above-mentioned technical means, the standardization and high-frequency collection of three types of core physiological data have been achieved, with high data integrity (excluding extreme situations such as equipment power failure), providing a high-quality data foundation for subsequent emotional stress quantification and risk projection.

[0020] Considering that traditional stress assessments lack quantitative formulas, and a single indicator cannot reflect the overall excitation state of the autonomic nervous system; and that weighting coefficients and baseline values ​​lack scientific basis; therefore: S2. Based on heart rate variability and skin conductance data, an emotional stress index is calculated to quantify the level of excitation in the user's autonomic nervous system; specifically: The ECG signal is preprocessed (denoising, R-wave detection) by uploading it to a cloud server via a mobile app, and then calculated. (Standard deviation of all normal sinus intervals); calculate the amplitude of change of skin conductance signals. Finally, the emotional stress index is calculated using the formula. ; Emotional Stress Index Calculated using the following formula: ,in, The standard deviation of heart rate variability; The change in skin conductance level relative to baseline is expressed by the formula: ,current The difference from the baseline; a positive value indicates that the sympathetic nerve excitation level is higher than the baseline. This serves as a baseline reference value for each user. and These are weighting coefficients. and The value of is determined through machine learning training and optimization on historical data samples, where The value range is from 0.4 to 0.7. The value range is from 0.3 to 0.6, and satisfies the following condition: ; The specific calculation and parameter optimization process is as follows: First is the user's personal baseline value ( , The determination of ) The average value of the user's "standard resting state" data for 7 consecutive days was selected as the baseline. The standard resting state is defined as maintaining an empty stomach, lying still, in an environment free from external noise interference (such as a bedroom) within 30 minutes after waking up each morning, and avoiding behaviors such as getting up, eating, and looking at mobile phones. Next, data filtering was performed to exclude outliers from the 7-day resting data (such as outliers on a particular day). If the value suddenly drops below 50% of the mean, it is considered a temporary disturbance; the mean is calculated using the remaining 6 days of data. Taking patient A as an example: the 7-day resting SDNN data were 75, 72, 78, 74, 76, 73, and 77 ms, with a mean of 75 ms. Therefore... =75ms; 7-day resting SCL data were 12.5, 13.1, 12.8, 13.3, 12.9, 13.0, and 12.7 μS, with a mean of 12.9 μS. Therefore... =12.9μS.

[0021] Next are the weighting coefficients. , The optimization process is as follows: First, a sample selection process was conducted, collecting 6 months of longitudinal data from 1000 patients with type 2 diabetes (disease duration 3–10 years), including daily data. , Data, and weekly GAD-7 scale scores (the gold standard for clinical stress assessment, 0-21 points, ≥10 points for moderate anxiety); Then, target optimization is performed to minimize... The mean squared error (MSE) of the GAD-7 score is used to maximize the Pearson correlation coefficient R (a measure of the strength of the correlation). The training method used was a random forest regression model (based on the Scikit-learn library), which iterated through... ∈[0.4,0.7]、 All combinations of [0.3, 0.6] (a total of 16 combinations); The final optimized result is when =0.55、 When =0.45, The correlation with the GAD-7 score was R=0.82 (p<0.001, statistically significant) and MSE=1.2, which was improved compared with the single SDNN index (R=0.51) or the single SCL index (R=0.48), proving that this coefficient can most accurately quantify emotional stress. Taking patient A's real-time data at a certain moment as an example (currently) =60ms, current =15.2μS): Step 1: Calculation Item: 1 - 60 / 75 = 0.2, multiplied by =0.55, therefore 0.11; Step 3: Calculation 15.2 - 12.9 = 2.3 μS =2.3 / 12.9≈0.178, multiplied by =0.45, therefore 0.0801; Step 3: Summation and range constraint (0-1, to avoid outliers): 0.11 + 0.0801 ≈ 0.1901, that is... =0.19 (no obvious pressure); If the current =50ms, Current =16.5μS, then =0.55×(1-50 / 75)+0.45×((16.5-12.9) / 12.9)=0.55×0.333+0.45×0.279≈0.183+0.126≈0.309 (mild pressure); Through the above steps, the recall bias and subjectivity of subjective questionnaires are overcome, achieving continuous, passive, and objective monitoring of emotional stress. Simultaneously, weights learned from real data through machine learning (…) , Furthermore, satisfying the constraint that the sum is 1 is superior to setting it empirically (e.g., 0.5 for each), making The correlation between the index and the actual emotional state has been improved, which resolves the ambiguity of traditional subjective assessment and provides a calculable stress indicator for subsequent risk projection.

[0022] Given that the extrapolation rules lack medical basis, and the thresholds (such as "blood glucose fluctuation range" and "EPSI threshold") are arbitrarily set without historical data verification, the extrapolation results are unreliable. Therefore: S3. Input continuous blood glucose monitoring data and emotional stress index into a preset deduction rule knowledge base for logical reasoning; the deduction rule knowledge base contains several production rules based on medical knowledge, including a first subset of rules for deducing brain function health risks and a second subset of rules for deducing the difficulty of blood glucose control; the construction process, threshold basis, and reasoning logic of each rule are as follows: The overall construction process of the inference rule knowledge base: First, rule extraction was performed. A literature search was conducted, retrieving 237 relevant articles published between 2018 and 2023 in authoritative international journals such as *DiabetesCare* and *Neurology*. Association conclusions regarding "glucose fluctuations and amygdala function," "emotional stress and epilepsy risk," and "stress and glucose target achievement rate" were extracted. Next, through expert screening, 50 clinical experts (30 from endocrinology and 20 from neurology, all with associate chief physician or higher titles) were invited to score the validity of the literature conclusions (1-5 points, ≥4 points indicating high consensus). Ultimately, 18 high consensus conclusions were selected (e.g., "The greater the glucose fluctuation, the higher the risk of amygdala abnormality"). Then, rule structuring was performed, transforming the 18 high consensus conclusions into production rules of "IF condition (including specific threshold) THEN conclusion" (e.g., "IF daily average glucose fluctuation ≥4.0 mmol / LANDEPSI ≥0.6 THEN amygdala abnormality").

[0023] The threshold verification and optimization process is as follows: A retrospective analysis was conducted on 5000 patients with type 2 diabetes (3-year follow-up data, including fMRI brain function tests and blood glucose control records): For each rule's threshold (e.g., 4.0 mmol / L, 0.6), multiple sets of values ​​were tested, and the accuracy of the rule under different thresholds ("number of correct deductions / total number of deductions") was calculated. Adjust the thresholds until the accuracy is ≥85%, and finally determine the final thresholds for each rule, specifically applied to the first and second rule subsets as follows: The first subset of rules contains inference rules based on amygdala functional activity; inference rule 1 is defined as: when continuous blood glucose monitoring data indicates... When the emotional stress index is ≥0.6, the inference rule is triggered, and it is presumed that the amygdala's functional activity is abnormal. The threshold is determined based on: A chi-square test was performed on the "daily average blood glucose fluctuation range" and "amygdala abnormalities (fMRI results)" of 5000 patients, and the results are as follows: When the average daily blood glucose fluctuation range (mmol / L) was <3.0, the incidence of amygdala abnormalities was 8.2%, and the difference (p value) between the control group (<3.0mmol / L) and the baseline group (no significant difference). When the average daily blood glucose fluctuation range (mmol / L) was 3.0-4.0, the incidence of amygdala abnormality was 19.5%, and the difference (p value) between the control group (<3.0mmol / L) and the control group (p value) was <0.05 (statistically significant). When the average daily blood glucose fluctuation range (mmol / L) was ≥4.0, the incidence of amygdala abnormality was 38.7%, and the difference (p value) between the control group (<3.0mmol / L) and the control group (p value) was <0.01 (highly significant). It can be seen that when the daily average blood glucose fluctuation is ≥4.0 mmol / L, the amygdala abnormality rate increases sharply (4.7 times that of the <3.0 mmol / L group). Therefore, 4.0 mmol / L is taken as the critical value for blood glucose fluctuation. At the same time, Association analysis with GAD-7 scores showed: When the threshold is ≥0.6, 85% of patients have a GAD-7 score ≥10 (moderate anxiety). At this threshold, autonomic nervous system excitation significantly exacerbates amygdala overactivation; therefore, 0.6 is used as the threshold value. Critical value; For example: If patient A's average daily blood glucose fluctuation in a certain week is 4.2 mmol / L, during the same period If the mean value is 0.63, then inference rule 1 is triggered, presuming that the amygdala function is abnormal; Abnormal amygdala function is a risk of hyperfunction; the risk of hyperfunction is defined by extrapolation rule 2, the standard being: the mean low-frequency to high-frequency power ratio (LF / HF) in heart rate variability increases by more than 25% from the baseline value within one week after triggering the rule; LF / HF is a frequency domain index of HRV. LF (0.04–0.15 Hz) reflects the level of sympathetic nerve excitation, and HF (0.15–0.4 Hz) reflects the level of parasympathetic nerve excitation. An elevated LF / HF indicates sympathetic dominance (autonomic imbalance). The baseline value of LF / HF is compared with... Baseline values ​​were calculated synchronously (mean LF / HF under 7-day resting state). The threshold is determined based on: For 1200 patients who triggered inference rule 2, the association between "LF / HF change amplitude" and "fMRI amygdala hyperactivity (oxygen level-dependent signal elevation)" was analyzed: When the LF / HF ratio increased by more than 25% from baseline, the positive rate of amygdala hyperplasia was 78.3%. When the LF / HF ratio increased by ≤25% from baseline, the positive rate of amygdala hyperplasia was only 19.1%. The difference between the two is highly significant (p < 0.001), so 25% is taken as the critical value for the change in LF / HF to ensure the accuracy of the rule; For example, after patient A triggers the inference rule for amygdala function activity, the mean LF / HF ratio within one week is 3.2 (the baseline LF / HF ratio is 2.5), and the change is (3.2-2.5) / 2.5×100%=28%>25%. Therefore, inference rule 2 is triggered, and it is presumed that there is a risk of amygdala hyperfunction. If a risk of amygdala hyperfunction is deduced, then according to deduction rule 3, an increased risk of seizures is deduced: when the risk of amygdala hyperfunction persists for more than 4 weeks, and continuous blood glucose monitoring data shows ≥2 hypoglycemic events during this period, then increased excitability of the limbic system is presumed, and an increased risk of seizures is present. The hypoglycemic events are defined as follows: The duration of hyperactivity risk is the number of consecutive weeks from the first time the rule of abnormal amygdala function activity as a risk of hyperactivity is triggered to the current time (if the rule of abnormal amygdala function activity as a risk of hyperactivity is not triggered in a certain week during the period, the duration is recalculated). The threshold is determined based on: When the duration was >4 weeks, 800 patients who triggered the rule of amygdala functional activity abnormality as a risk of hyperactivity were followed up. It was found that the seizure rate within 3 months was 2.1% in patients whose hyperactivity lasted ≤4 weeks and >4 weeks; the seizure rate increased to 9.7% in patients whose hyperactivity lasted >4 weeks (the difference was highly significant, p < 0.01). When there are ≥2 hypoglycemic events and blood glucose <3.9mmol / L, the brain glucose supply is insufficient, which will further aggravate the excitability of the limbic system (including the amygdala); stratified analysis of patients with hyperactivity lasting >4 weeks showed that the seizure rate of patients with ≥2 hypoglycemic events was 15.3%, which was 4.0 times that of patients with <2 events (3.8%) (p<0.001). In summary, "lasting for more than 4 weeks" and "hypoglycemia ≥ 2 times" were determined as the critical conditions; For example, if patient A's risk of amygdala hyperactivity persists for 5 weeks (>4 weeks), during which a total of 3 hypoglycemic events occur (blood glucose levels of 3.7, 3.5, and 3.8 mmol / L, all <3.9 mmol / L), then rule 3 is triggered, presuming that the patient has an increased risk of seizures.

[0024] The second rule subset is used to extrapolate the negative impact of emotional stress on the future difficulty of blood sugar control; the extrapolation rule for the second rule subset is defined as extrapolation rule 4, specifically: if If the condition persists for more than 4 weeks, the output projection conclusion is that the probability of achieving blood glucose targets within the next 1 to 3 months will decrease by more than 30%, where achieving blood glucose targets is defined as the user's time within the glucose target range (TIR) ​​> 70%. The threshold is determined based on: A 3-month follow-up study was conducted on 1000 patients to analyze… The association between "≥0.5 duration" and "TIR > 70% rate in the next 3 months": when When the duration of ≥0.5 (weeks) was ≤2, the TIR >70% rate after 3 months was 68.2%, which was the baseline group (no significant difference) compared with the control group (≤2 weeks). when When the duration of ≥0.5 was 2–4 weeks, the TIR > 70% rate after 3 months was 45.5%, which was 33.3% lower than the control group (≤2 weeks) (p < 0.05). when When the duration of ≥0.5 weeks was >4, the TIR >70% rate after 3 months was 32.7%, which was 52.0% lower than the target achievement rate of the control group (≤2 weeks) (p < 0.01). It can be seen that when When ≥0.5 persists for more than 4 weeks, the compliance rate decreases by more than 50% compared to the baseline group; therefore, a decrease of more than 30% is taken as the conclusion (leaving a margin to avoid interference from extreme data); at the same time, A score of ≥0.5 corresponds to a GAD-7 score of ≥8 (mild anxiety), at which point stress has begun to affect insulin sensitivity, making blood sugar control more difficult. For example: Patient A had 5 weeks of [something] in the past 6 weeks. If the value is ≥0.5 and the number of days is ≥5 (cumulative duration of 5 weeks > 4 weeks), then according to rule 4, the conclusion is that "the probability of patient A achieving target blood glucose (TIR > 70%) will decrease by more than 30% in the next 1-3 months."

[0025] Given that current simulations only output a binary conclusion of "risky / no risk" without risk level classification, they cannot guide clinical prioritization. Therefore: S4. Based on the results of logical reasoning, deduce the user's specific brain function health risk status and / or blood glucose control difficulty assessment related to diabetes; the specific content is as follows: Based on the reasoning results of step S3, "Risk Level Quantification (Level 1-5, Level 5 is the highest)" is added. The specific level classification criteria and examples are as follows: When the risk type is amygdala dysfunction, the condition for triggering Level 1 (low risk) is occasional triggering rule 1 (less than once a week); the condition for triggering Level 2 (mild risk) is continuous triggering rule 1 (more than once a week). When the risk type is amygdala hyperfunction risk, the condition for triggering level 3 (moderate risk) is occasional triggering rule 2 (less than once a week); the condition for triggering level 4 (high risk) is continuous triggering rule 2 (more than once a week). When the risk type is increased risk of epileptic seizures, the condition for triggering level 5 (very high risk) is triggering rule 3. When the risk type is difficulty in blood glucose control, the condition for triggering Level 1 (low risk) is that Rule 4 is not triggered; the condition for triggering Level 2 (mild risk) is that Rule 4 is triggered (lasting for 4-6 weeks); and the condition for triggering Level 3 (moderate risk) is that Rule 4 is triggered (lasting for >6 weeks). For example: Based on the reasoning results of step S3, the final deduction result for patient A is: "Abnormal amygdala function (level 2), risk of amygdala hyperfunction (level 4), increased risk of epileptic seizures (level 5), and a decrease of more than 30% in the probability of achieving blood glucose targets in the next 1-3 months (level 3)". Through the aforementioned technical means, risk level quantification enables doctors to quickly determine clinical priorities (e.g., level 5 risk requires immediate intervention, while level 2 risk can be followed up periodically), thus resolving the ambiguity of traditional binary conclusions.

[0026] Given the current technology, the report only presents data tables without intuitive correlation analysis; intervention recommendations are generalized (e.g., "more exercise"), lacking personalized adaptation. Therefore: S5. Generate a visualization report of the deduction results and provide feedback to the user; the visualization report also includes personalized, non-pharmacological intervention suggestions for reducing the emotional stress index; the specific content includes the following: Visual report design: The format uses a "timeline + relationship diagram", with the horizontal axis representing the past 30 days and the vertical axis divided into three layers: First layer (red curve): Daily average blood glucose fluctuation range (unit: mmol / L), marked with the 4.0 mmol / L threshold; Second layer (blue curve): Daily average The values ​​are marked with two critical lines: 0.5 and 0.6. Third layer (colored bars): Daily risk level (yellow = level 1, orange = level 2, red = level 3-5); At the same time, key annotations are made, and the risk type is marked with a red asterisk on the "date of triggering the rule" (e.g., "2025-08-15: triggering the amygdala abnormality rule"), so that users can intuitively understand the relationship between "blood sugar - stress - risk".

[0027] Personalized intervention suggestion generation logic: The three-step method of "cause analysis → habit matching → solution selection" is adopted, as follows: Step 1: Conduct trigger analysis (based on user physiological data), through analysis Peak hours (e.g., 22:00-24:00) High blood pressure → poor sleep quality; weekdays 9:00-18:00 High levels of stress (work pressure) and peak blood sugar fluctuations (e.g., high blood sugar 2 hours after a meal – dietary issues) are key factors to identify the core triggers for stress and blood sugar fluctuations. Step 2: Perform habit matching (based on user behavior data), collect user lifestyle habits (e.g., occupation: programmer → sedentary; lunch break: 30 minutes → short; exercise frequency: <1 time per week → less exercise), and eliminate intervention solutions that conflict with habits (e.g., do not recommend "long-distance outdoor running" for sedentary users); Step 3: Screening of intervention plans (based on the intervention library), which contains 200+ non-pharmacological intervention plans (verified by 50 clinical experts). Based on the triggers and habits, select 3-5 of the most suitable plans and specify the frequency, duration and specific operation of implementation. Example of intervention recommendation (Patient A): The cause analysis for patient A was "nighttime". High (22:00-24:00 average 0.72), weekdays The patient's condition was described as "high (mean 0.65 from 9:00 to 18:00)," and their habits were "programmer (sedentary), 30-minute lunch break, 0 times of exercise per week." The final intervention recommendation was: Sleep intervention: Go to sleep before 10:30 p.m. every night, turn off electronic devices 1 hour before bedtime, and play white noise (we recommend the "Rain Sound" mode of the "Tide" app) for 20 minutes each time; Stress reduction during work breaks: Perform the "4-7-8 breathing exercise" once every 1 hour of work (inhale for 4 seconds, hold your breath for 7 seconds, exhale for 8 seconds), repeating 5 sets each time; Lunch break intervention: Use 15 minutes of your lunch break for mindfulness meditation (we recommend the "Workplace Stress Reduction" course on the "Headspace" App, which lasts 10 minutes). Intervention for prolonged sitting: Get up and move around for 2 minutes every 30 minutes, and do simple stretches such as "chest expansion + waist twist" (10 times each). Dietary recommendations: Consume 100g of spinach at lunch and dinner daily (containing magnesium, which helps relieve nerve excitation), and avoid caffeine (such as coffee and strong tea) after 3 pm. Through the aforementioned technical means, visualized reports enhance patients' and doctors' understanding of the correlation between "blood sugar-stress-risk" and improve the implementation rate of personalized intervention recommendations.

[0028] The method described in Example 1 is applicable in principle to all types of diabetic patients. The core pathophysiological mechanism on which it is based, namely, autonomic dysfunction caused by blood glucose fluctuations and emotional stress, is a common feature of type 1 and type 2 diabetes.

[0029] Example 1 selected patients with type 2 diabetes as the main subjects for implementation based on the following considerations: First, the type 2 diabetes patient population is the largest and has significant public health needs; second, the comorbidity of blood glucose fluctuations and mood disorders is common and insidious in this population, making early warning clinical value particularly prominent; third, adult type 2 patients have better compliance with long-term wearable monitoring devices, which is conducive to continuous and stable data collection and ensures the stable operation of the simulation system.

[0030] Data collection, threshold determination, and rule construction were all based on a large cohort study of patients with type 2 diabetes. It is worth noting that the core deductive framework of this invention (such as the EPSI index formula and rule engine architecture) is also applicable to patients with type 1 diabetes; only some threshold parameters need to be adaptively calibrated according to the characteristics of this population.

[0031] Example 2 To implement the method of Example 1, therefore, please refer to Figure 2 As shown, the purpose of Embodiment 2 is to provide an intelligent prediction service system for the pathogenesis of diabetic complications. This intelligent prediction service system for the pathogenesis of diabetic complications includes: The data acquisition module is used to collect users' physiological data from wearable devices and smart terminals; among which, wearable devices include CGM, Apple Watch, and Fitbit; and smart terminals include mobile phones. The data processing module, connected to the data acquisition module, is used to calculate the emotional stress index based on heart rate variability and skin conductance activity data. The data processing module is deployed on the mobile device and includes local scheduling of the data acquisition module and real-time calculation of the data processing module (such as incremental calculation of EPSI), reducing the pressure on the cloud. The intelligent deduction module, connected to the data processing module, contains a deduction rule knowledge base. It performs logical reasoning based on production rules and outputs deduction results. The production rules based on medical knowledge, the first subset of rules for deducing brain function health risks, and the second subset of rules for deducing the difficulty of blood sugar control in the deduction rule knowledge base are constructed through structured extraction from medical literature and clinical expert experience, and retrospective analysis and verification using historical patient data. The intelligent deduction module is deployed on an Alibaba Cloud server, which includes the intelligent deduction module (embedded in the rule knowledge base) and a cloud database (storing user data and the rule base), responsible for rule reasoning and long-term data storage. The user interaction module, connected to the intelligent simulation module, is used to generate visual reports, provide personalized intervention suggestions, and display feedback information to users. The overall interaction logic is as follows: the data acquisition module collects data → the data processing module performs preprocessing and real-time calculations → the intelligent inference module performs reasoning → the user interaction module obtains the results from the intelligent inference module and displays them.

[0032] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligently predicting the pathogenesis of diabetic complications, characterized in that, Includes the following steps: S1. Collect the user's physiological data, including continuous blood glucose monitoring data, heart rate variability data, and skin conductance data; S2. Based on heart rate variability and skin conductance data, an emotional stress index is calculated to quantify the level of excitation of the user's autonomic nervous system. S3. Input the continuous blood glucose monitoring data and emotional stress index into the preset deduction rule knowledge base for logical reasoning; the deduction rule knowledge base contains several production rules based on medical knowledge, including a first rule subset for deducing brain function health risks and a second rule subset for deducing the difficulty of blood glucose control. S4. Based on the results of logical reasoning, deduce the user's specific brain function health risk status and / or blood glucose control difficulty assessment related to diabetes; S5. Generate a visualization report of the deduction results and provide feedback to the user; the visualization report also includes providing the user with personalized, non-pharmacological intervention suggestions for reducing the emotional stress index.

2. The intelligent prediction service method for the pathogenesis of diabetic complications according to claim 1, characterized in that, In step S2, the emotional stress index is calculated using the following formula: ,in, The standard deviation of heart rate variability; The change in skin conductance level relative to baseline is expressed by the formula: , This serves as a baseline reference value for each user. and These are weighting coefficients. and The value of is determined through machine learning training and optimization on historical data samples, where The value range is from 0.4 to 0.

7. The value range is from 0.3 to 0.6, and satisfies the following condition: .

3. The intelligent prediction service method for the pathogenesis of diabetic complications according to claim 1, characterized in that: In step S3, the first rule subset includes inference rules based on amygdala functional activity; The inference rule is defined as follows: when continuous blood glucose monitoring data indicates... When the emotional stress index is ≥0.6, the inference rule is triggered, and it is presumed that the amygdala's functional activity is abnormal.

4. The intelligent prediction service method for the pathogenesis of diabetic complications according to claim 3, characterized in that: Abnormal amygdala function is a risk factor for hyperfunction. The criteria for determining the risk of hyperfunction are: the average low-frequency to high-frequency power ratio in heart rate variability increases by more than 25% from the baseline value within one week after the trigger rule.

5. The intelligent prediction service method for the pathogenesis of diabetic complications according to claim 4, characterized in that, In step S4, if the deduction indicates a risk of amygdala hyperfunction, then the increased risk of epileptic seizures is deduced according to the following rules: When a state of amygdala hyperfunction persists for more than 4 weeks, and continuous blood glucose monitoring data shows ≥2 hypoglycemic events during this period, increased excitability of the limbic system is presumed, indicating an increased risk of seizures. The hypoglycemic events are defined as follows: .

6. The intelligent prediction service method for the pathogenesis of diabetic complications according to claim 1, characterized in that, In step S3, the second rule subset is used to infer the negative impact of emotional stress on the difficulty of future blood sugar control; The derivation rule for the second rule subset is: if If the condition lasts for more than 4 weeks, the output projection conclusion is that the probability of achieving blood sugar target within the next 1 to 3 months will decrease by more than 30%, where achieving blood sugar target means that the user's glucose target range is exceeded for more than 70% of the time.

7. A smart prediction service system for the pathogenesis of diabetic complications, wherein the smart prediction service system for the pathogenesis of diabetic complications is used to implement the smart prediction service method for the pathogenesis of diabetic complications as described in any one of claims 1 to 6, characterized in that, include: The data acquisition module is used to collect users' physiological data from wearable devices and smart terminals; The data processing module is used to calculate the emotional stress index based on heart rate variability and skin conductance data; The intelligent deduction module contains a deduction rule knowledge base, which is used to perform logical reasoning based on production rules and output deduction results. Among them, the production rules based on medical knowledge, the first rule subset for deducing brain function health risks, and the second rule subset for deducing the difficulty of blood sugar control in the deduction rule knowledge base are constructed by structurally extracting medical literature and clinical expert experience, and retrospectively analyzing and verifying them using historical patient data. The user interaction module is used to generate visual reports, provide personalized intervention suggestions, and display feedback information to users.

Images