Diabetes early warning and risk assessment method and system

Abstract

The invention relates to the technical field of disease prevention, and discloses a diabetes early warning and risk assessment method and system, and the method comprises the steps: collecting clinical detection data of a patient, generating complication baseline data according to the clinical detection data, and obtaining the real-time blood sugar data of the patient through portable equipment; a cross-clock early warning model is constructed based on a Bayesian inference network, complication baseline data is used as a slow clock, real-time blood glucose data is used as a fast clock, a three-layer gating mechanism is fused to adjust an observation weight, and a risk level is updated and dynamically output through logarithm Bayesian accumulation; and when the risk level exceeds a preset threshold value or the uncertainty of the cross-clock early warning model exceeds an upper limit, triggering review scheduling, and outputting the risk level and a review date. The method has high clinical implementability and expansibility, is especially suitable for individual risk management of diabetes, and can significantly improve the complication early screening efficiency and the blood sugar control quality.

Description

Technical Field

[0001] The present invention relates to the technical field of disease prevention, and particularly to a method and system for diabetes early warning and risk assessment. Background Art

[0002] Diabetes is a type of metabolic disease characterized by chronic hyperglycemia. As the disease progresses, it often causes a variety of serious complications, including diabetic retinopathy, diabetic nephropathy, peripheral neuropathy, and cardiovascular and cerebrovascular diseases. These complications often lack obvious symptoms in the early stage. Once obvious clinical manifestations appear, it usually has entered the middle and late stages, with great treatment difficulty and high intervention costs, which are the main causes of disability and death in diabetes.

[0003] Although continuous glucose monitoring (CGM) technology has been widely used in daily blood glucose control in recent years, there are still significant limitations in the early identification and prediction of complications. Currently, interventions mainly rely on periodic hospital tests, which are difficult to dynamically reflect the complex relationship between patients' blood glucose fluctuations and the evolution of various complications, and cannot reasonably arrange follow-up plans according to the individual risk status of patients, resulting in coexistence of resource waste and disease delay.

[0004] In view of the above problems, the present invention proposes a method for diabetes complication early warning and follow-up scheduling based on cross-clock reasoning, which realizes the fusion modeling of clinical test data and real-time blood glucose data, can dynamically evaluate the development risks of multiple complications, and intelligently recommends the reexamination time, improving the early screening rate and resource utilization efficiency. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is that there are still significant limitations in the early identification and prediction of diabetes complications. To solve the above technical problem, the present invention provides the following technical solution: A method for diabetes early warning and risk assessment, including: collecting clinical test data of patients, generating complication baseline data according to the clinical test data, and obtaining real-time blood glucose data of patients through a portable device; Constructing a cross-clock early warning model based on a Bayesian inference network, using the complication baseline data as the slow clock and the real-time blood glucose data as the fast clock, adjusting the observation weight through a three-layer gating mechanism, and dynamically outputting the risk level through logarithmic Bayesian cumulative update; When the risk level exceeds a preset threshold, or the uncertainty of the cross-clock early warning model exceeds the upper limit, trigger a reexamination schedule and output the risk level and reexamination date.

[0007] As a preferred embodiment of the diabetes warning and risk assessment method described in the present invention, wherein: the clinical test data includes data on the risk of retinopathy, nephropathy, neuropathy, and cardiovascular disease.

[0008] As a preferred embodiment of the diabetes warning and risk assessment method described in the present invention, wherein: the real-time blood glucose data includes collecting continuous blood glucose values within a fixed time window of a patient through a portable device and marking timestamps, and calculating the blood glucose fluctuation characteristics within the time window; Obtaining the behavior information record of the patient, aligning the behavior information record with the blood glucose fluctuation characteristics, and identifying the event-related fluctuation interval; Setting the blood glucose threshold for a fixed time window based on the event-related fluctuation interval, and comparing the blood glucose fluctuation characteristics with the blood glucose threshold to achieve blood glucose detection and warning.

[0009] As a preferred embodiment of the diabetes warning and risk assessment method described in the present invention, wherein: the cross-clock warning model includes a cross-clock warning model that adopts a dual-time-axis modeling framework, and each type of complication corresponds to a slow-clock hidden state trajectory, and each hidden state trajectory is updated by two types of inputs: The blood glucose fluctuation characteristics from the fast clock; The complication baseline data from the slow clock; A three-layer gating mechanism is established between the fast clock and the slow clock to generate gating weights for each fixed time window and regulate the response intensity of different complications to the fast clock data; Using logarithmic cumulative Bayesian updates to perform state updates based on the blood glucose fluctuation characteristics and the complication baseline data; Based on the mutual influence of complications, establishing risk transduction jump nodes between complications and calculating the risk conduction relationship between complications.

[0010] As a preferred embodiment of the diabetes warning and risk assessment method described in the present invention, wherein: the three-layer gating mechanism includes three layers: an individual sensitivity coefficient, a daily rhythm coefficient, and a real-time event coefficient; The individual sensitivity coefficient is a fixed value set according to the clinical test data, and the individual sensitivity coefficient is positively correlated with the degree of complication development and is updated each time the clinical test data is collected; The daily rhythm coefficient includes retrieving the real-time blood glucose data within the past month, obtaining the quantile median smoothing curve on a 24-hour scale as the daily baseline, and the daily rhythm coefficient is positively correlated with the degree to which the current window blood glucose data deviates from the daily baseline; The real-time event coefficient is determined according to the complication preset template library, and the calculation of the gating weight is expressed as: ; Wherein, Represents the gating weights within the time window ; Represents the individual sensitivity coefficient Represents the circadian rhythm coefficient Represents the real-time event coefficient

[0011] As a preferred embodiment of the diabetes warning and risk assessment method described in the present invention, wherein: the cumulative update by logarithmic Bayesian and dynamic output of the risk level includes initializing the corresponding state prior distribution of the complication baseline data and calculating the gating weights within the time window Based on the blood glucose fluctuation characteristics within the time window, fitting the observation likelihood under the complication baseline data, and performing Bayesian update through logarithmic accumulation, expressed as ; Wherein Represents at time point , after experiencing fast clock time windows, the posterior probability of the updated risk state of complication ; Represents the risk level of the risk state Represents complication at time point initial posterior probability Represents the th fast clock time window, the gating weight of complication ; Represents the th fast clock time window, the blood glucose fluctuation characteristics Represents the logarithmic calculation Obtain the posterior probability distribution through Softmax normalization ; Calculate the risk conduction relationship between complications through the risk transduction jump node, and calculate the risk jump factor for each pair of related complications ; Wherein Represents the source complication to the target complication at time point jump factor Represents the calculation function of the jump factor Represents the gating weight Represents the source complication blood glucose fluctuation characteristics Represents the target complication blood glucose fluctuation characteristics If Greater than a preset threshold , take as the gain term to adjust the posterior state distribution of the target complication : ; wherein, represents the posterior probability distribution of the complication after the jump factor adjustment; represents the normalization function; represents the posterior probability distribution of the unadjusted complication ; represents the jump gain coefficient; Map the posterior probability distribution of the complication to the risk level. When the risk level exceeds the preset threshold, trigger a review schedule.

[0012] As a preferred solution of a diabetes warning and risk assessment method according to the present invention, wherein: the review schedule includes, taking the current date as the base point, forming a set of candidate review dates within the range of 2 - 26 weeks in the future with a 7-day step. For each candidate review date, calculate the information gain and resource cost, generate a benefit index, and output the date corresponding to the highest benefit index as the review date.

[0013] A diabetes warning and risk assessment system adopting any of the methods of the present invention, wherein: a collection module, which collects the clinical test data of the patient, generates complication baseline data according to the clinical test data, and obtains the real-time blood glucose data of the patient through a portable device; A prediction module, which constructs a cross-clock warning model based on a Bayesian inference network, takes the complication baseline data as the slow clock, the real-time blood glucose data as the fast clock, fuses a three-layer gating mechanism to adjust the observation weight, and dynamically outputs the risk level through logarithmic Bayesian accumulation update; A scheduling module, when the risk level exceeds the preset threshold, or the uncertainty of the cross-clock warning model exceeds the upper limit, triggers a review schedule and outputs the risk level and the review date.

[0014] Advantages of the present invention: By constructing a cross-clock Bayesian inference network, the present invention realizes the deep fusion modeling of the slow clock and the fast clock, combines a multi-layer gating mechanism to dynamically adjust the observation weight, and continuously corrects the risk states of various complications driven by minute-level data. The introduced risk transduction jump mechanism can effectively model the risk conduction between comorbid symptoms and improve the linkage recognition ability. At the same time, based on risk benefits and review costs, it intelligently recommends review times to avoid the coexistence of resource waste and missed detections. The present invention has high clinical feasibility and scalability, is particularly suitable for individualized diabetes risk management, and can significantly improve the early screening efficiency of complications and the quality of blood glucose control. Description of the Drawings

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings; Figure 1 It is the overall flowchart of a diabetes warning and risk assessment method provided by an embodiment of the present invention. Specific embodiments

[0016] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0017] Embodiment 1, referring to Figure 1 , which is an embodiment of the present invention, provides a diabetes warning and risk assessment method, including: S1: Collect the clinical test data of the patient, generate complication baseline data according to the clinical test data, and obtain the real-time blood glucose data of the patient through a portable device.

[0018] In the management of diabetic patients, long-term abnormal blood glucose not only causes acute metabolic disorders, but more importantly, gradually induces a variety of chronic complications without obvious clinical symptoms, including retinopathy, renal function damage, neuropathy, and cardiovascular system lesions. In response to the early warning needs of these serious complications, the present invention realizes individualized high-risk identification and warning prompts through a two-dimensional monitoring mechanism that combines professional detection baselines and real-time blood glucose trend linkages.

[0019] Specifically, based on the currently clinically recognized four types of high-risk complications, the following four types of structured medical data are collected as individual complication risk baselines: Fundus photography images and AI automatic scoring results for assessing the risk of diabetic retinopathy (DR). Fundus photography equipment and AI automatic scoring have been widely used in clinical practice and have the ability to identify high-resolution lesions. A large number of studies have confirmed that the occurrence and progression of retinopathy are closely related to long-term high blood glucose levels and blood glucose variability.

[0020] The urinary albumin-to-creatinine ratio and estimated glomerular filtration rate are used to assess the risk of diabetic kidney disease (DKD). The relevant detection methods have been standardized globally and are the core indicators for the clinical grading of diabetic kidney disease. Persistent hyperglycemia has been proven to cause glomerular filtration dysfunction and is an important driving factor in the progression of DKD.

[0021] Nerve conduction velocity or quantitative sensory score is used to assess the risk of diabetic peripheral neuropathy (DPN). Such indicators are the internationally common diagnostic basis for neuropathy, and long-term hyperglycemia and blood glucose fluctuations are positively correlated with nerve conduction dysfunction.

[0022] Coronary artery calcium score, level or echocardiographic indices of cardiac function are used to assess the risk of diabetic cardiovascular complications. The above-mentioned biomarkers and imaging parameters are widely used in the risk prediction of coronary heart disease and heart failure. The promoting effect of blood glucose fluctuations on vascular endothelial function, inflammatory response and atherosclerosis has been fully confirmed clinically.

[0023] It should be noted that the present invention focuses on complications that are difficult to detect by portable devices. Detection that can be achieved by portable devices is not within the scope of consideration of the present invention. For example, for diabetic foot or plantar ulcers, there are already mature wearable products such as smart insoles and smart socks that can achieve continuous home monitoring. In addition, for autonomic dysfunction, sleep apnea syndrome and diabetes-related cognitive dysfunction, there is currently a lack of standardized early detection programs or prediction models, and they also do not belong to the modeling objectives focused on by the present invention.

[0024] Furthermore, the system first obtains the patient's basic information for initializing the individual feature vector, and then obtains four types of clinical detection data, namely, retinopathy risk data, nephropathy risk data, neuropathy risk data and cardiovascular risk data, through hospitals, etc. The specific forms of the clinical detection data and the acquisition methods are as described above, and there are already mature technologies, so the present invention will not describe them in detail and directly obtains the detection results.

[0025] Map the clinical detection data to complication baseline data. The clinical detection data are the specific data collected clinically. According to clinical consensus, the specific data are mapped into the risk levels corresponding to the complications, such as being mapped into four levels: low, medium, high and extremely high. The larger the value, the higher the incidence risk of the corresponding complication. This risk level is the complication baseline data.

[0026] It should be noted that since the clinical detection data of these complications can only be obtained through professional hospital testing, the complication baseline data constructed by the present invention is data for long-term use. Based on the complication baseline data, the probability of complications in a certain period of time in the future is estimated according to the degree and time of the impact of blood sugar fluctuations on complications, so as to realize the detection and early warning of these complications, and issue an alarm when the risk is high and prompt the patient to go to the hospital for testing.

[0027] Furthermore, the patient's blood sugar is collected in real time through a wearable CGM (continuous glucose monitoring) device, with a sampling interval of no more than 5 minutes. The obtained blood sugar data is uploaded to the patient's mobile device. Each record includes the blood sugar value, collection timestamp, abnormal signal mark and device calibration status.

[0028] Secondly, patients use the APP to record behavioral events that are closely related to blood sugar fluctuations, including meal time and type, exercise time and type, medication use time, sleep start and end time, etc. The system can also be connected to smart wearable devices (such as watches, sports rings) to realize automatic recognition and synchronous labeling of some behaviors.

[0029] After being uploaded, the real-time blood glucose data undergoes data cleaning and denoising. The system calculates key dynamic characteristic indicators based on a sliding window of 5 to 15 minutes, including blood glucose variability, high blood glucose exposure time, low blood glucose exposure time, average blood glucose, blood glucose standard deviation and maximum change rate, etc., to construct blood glucose fluctuation characteristics.

[0030] The system further aligns the behavioral events with the blood sugar trend sequence, and automatically marks the event-related blood sugar fluctuation interval through the time window sliding mechanism for subsequent inference model input. The behavioral annotation data and the blood sugar feature matrix are input into the cross-clock Bayesian inference network together, and work together with the complication risk baseline under the slow clock to achieve dynamic update of the complication progression probability.

[0031] When the system detects that the dynamic pattern of blood sugar presents a high-risk structure (such as blood sugar continues to rise / fall above the threshold and the fluctuation amplitude is abnormal), or deviates significantly from the historical baseline, it will automatically issue risk warning signals of different levels and link the intervention module to generate personalized treatment suggestions.

[0032] S2: A cross-clock early warning model is constructed based on the Bayesian inference network. The baseline data of complications is used as the slow clock and the real-time blood glucose data is used as the fast clock. The three-layer gating mechanism is integrated to adjust the observation weights. The risk level is updated and dynamically output through logarithmic Bayesian accumulation.

[0033] Furthermore, to achieve real-time risk monitoring and personalized dynamic prediction of typical diabetic complications, the present invention constructs a cross-clock warning model. The cross-clock warning model adopts a modeling framework with two time axes. Each type of complication corresponds to a slow-clock hidden state trajectory, and each hidden state trajectory is updated by two types of inputs: The blood glucose fluctuation characteristics of the fast clock; The complication baseline data of the slow clock.

[0034] It should be noted that some diabetic patients have little blood glucose fluctuation, but maintain a relatively high level for many years, which is likely to induce chronic complications such as retinopathy and glomerulosclerosis.

[0035] To detect the complication pressure brought by this relatively high blood glucose level, in the slow-clock input features, in addition to the complication baseline data, an indicator of hyperglycemic exposure time calculated based on the continuous glucose monitoring system is further introduced, such as the cumulative time (TAR-90d) when the blood glucose value is higher than 10 mmol / L in the past 90 days. This indicator is used to measure the load degree of an individual in a hyperglycemic state for a long time, serves as an important input factor for the accumulation of complication risks, and is mapped to the initial prior distribution of the complication state inference model together with the baseline vector.

[0036] This indicator affects the complication state transition probability during the modeling process. When TAR-90d is higher than the upper quartile of the normal population distribution, the migration rate from the early state to the mid-late state is enhanced and adjusted. By introducing this long-term indicator, the present invention can effectively identify the risk of late-onset complications with "relatively stable blood glucose value but long-term high level".

[0037] Furthermore, the influence degrees of the fluctuations of continuous glucose monitoring (fast clock) data on various complications are not the same, and they change dynamically with individual constitutions and scenarios. If all blood glucose fluctuations are used to update the complication hidden state without selection, it will not only amplify the noise but also weaken the value of the semi-annual clinical baseline (slow clock data). Therefore, the present invention proposes a three-layer interpretable gating framework to dynamically adjust the importance weights of blood glucose windows for different complications in the form of hierarchical coefficients. The higher the gating value, the greater the correction effect of the current blood glucose window on the risk assessment of a certain complication.

[0038] Specifically, the three-layer gating mechanism includes three layers: individual sensitivity coefficient, diurnal rhythm coefficient, and real-time event coefficient. The value of each coefficient is obtained by mapping the numerical range to a fixed value, which can not only adjust the influence of the fast clock on the slow clock but also calculate the value quickly without occupying computing power.

[0039] Among them, the individual sensitivity coefficient is a fixed value set according to clinical test data. The individual sensitivity coefficient is positively correlated with the degree of complication development and is updated each time clinical test data is collected. For example, in combination with the complication baseline data, a fixed individual sensitivity coefficient is set according to the risk level. A low risk corresponds to 0.1, a medium risk corresponds to 0.3, etc.

[0040] The within-day rhythm coefficient includes retrieving real-time blood glucose data within the past month, obtaining the quantile median smoothing curve at 24-hour intervals as the within-day baseline. The within-day rhythm coefficient is positively correlated with the degree to which the current window blood glucose data deviates from the within-day baseline. For example, a deviation safety range is set. If the current window blood glucose falls within the baseline ±10% interval, the within-day rhythm coefficient takes 0.5. If it exceeds the baseline ±10% interval, the within-day rhythm coefficient takes 1.

[0041] The real-time event coefficient is determined according to the complication preset template library. For example, the DR high-risk template is that the blood glucose continuously > 10 mmol / L for ≥ 2 hours at night. The high-risk templates and thresholds for each complication are set based on professional knowledge and can also be adjusted according to the patient's situation. When the high-risk template is met, the real-time event coefficient takes 1.

[0042] The system calculates the gating weight in real time within each time window, expressed as: ; Among them, represents the gating weight within the time window ; represents the individual sensitivity coefficient; represents the within-day rhythm coefficient; represents the real-time event coefficient. Taking as the scaling coefficient of the fast clock observation likelihood, it is input into the cross-clock early warning model to complete the posterior update of the hidden state.

[0043] Furthermore, to achieve dynamic coupling between minute-level blood glucose fluctuations (fast clock data) and semi-annual clinical indicators (slow clock data), a Bayesian update structure is adopted to maintain continuous probability inference of the hidden state. This structure enables the system to continuously correct the estimated probability of the progression stage of various complications under the continuous input of non-invasive data streams, thereby achieving continuous risk quantification and hierarchical classification.

[0044] To avoid misjudgment or over-sensitivity of the model caused by high-frequency observations, the present invention designs a three-layer gating function to dynamically assign a weight factor related to complications to the observation data of each fast clock window. Considering that the gating coefficient may be significantly less than 1, to prevent numerical precision loss in the probability product, the present invention adopts a Bayesian posterior cumulative formula in logarithmic domain form, expressed as: ; Among them, Indicates at the time point , after fast clock time windows, the posterior probability of the updated risk state of the complication ; Indicates the risk level of the risk state; Indicates the complication at the time point initial posterior probability; Indicates the th fast clock time window, the gating weight of the complication ; Indicates the th fast clock time window, the blood glucose fluctuation characteristics of the complication Indicates logarithmic calculation. Finally, the posterior probability distribution is obtained through Softmax normalization.

[0045] It should be noted that the gating coefficients calculated by the three-layer gating mechanism are generally between 0.3 and 1. The decimal multiplication factor will cause precision loss in the multi-window cumulative multiplication, and the logarithmic accumulation can maintain stability. Compared with the hard judgment state jump or fixed grading, using the continuous probability distribution can retain the trend and ambiguity of the disease condition change, which is beneficial to dynamic warning and follow-up scheduling.

[0046] Furthermore, a large number of studies have shown that diabetic complications such as DKD, CV, and DPN have obvious intersections in the pathological mechanism. And for the multi-task Bayesian network, under the condition of associated targets, introducing intermediate associated variables can improve the accuracy of the main task. Therefore, the present invention proposes a risk transduction jump node structure, which explicitly introduces a cross-information transduction path between the complication risks to achieve risk sharing and early warning between high-risk channels.

[0047] Specifically, a risk transduction jump node is added between each pair of potentially related complication hidden state nodes. The node does not represent an independent disease stage, but represents the influence intensity and direction of the source risk on the target risk. Calculate the risk jump factor for each pair of related complications: ; Wherein, represents the jump factor of the source complication on the target complication at the time point ; represents the calculation function of the jump factor; represents the gating weight; represents the blood glucose fluctuation characteristics of the source complication ; represents the blood glucose fluctuation characteristics of the target complication .

[0048] It should be noted that It is not a fixed formula set uniformly. Instead, a discriminative calculation model should be designed according to the medical relevance and data characteristics between different complications. Taking DKD→CV as an example, there is a large amount of pathological basis for such a relationship in medical research. The jump factor can be constructed based on the following two dimensions: First, whether the fast clock gating weight is synchronously enhanced.

[0049] If and rise synchronously, it indicates that the source symptom has a weighted upregulation, which may have a recessive induction on the target symptom. The similarity score can be obtained by calculating the Pearson correlation or DTW dynamic matching degree of the gating coefficient within the sliding window.

[0050] Second, whether the fast clock observation characteristics match the known cross-high-risk event template.

[0051] For example: persistent hyperglycemia at night + large morning fluctuations → is related to the simultaneous deterioration of DKD + CV. Combining with the preset template library of real-time event coefficients, if a cross-high-risk is matched, directly take a relatively high value.

[0052] If is greater than the preset threshold , take as the gain term to adjust the posterior state distribution of the target complication : ; Among them, represents the posterior probability distribution of the complication after being adjusted by the jump factor; represents the normalization function; represents the posterior probability distribution of the unadjusted complication ; represents the jump gain coefficient.

[0053] Map the posterior probability distribution of the complication to the risk level. When the risk level exceeds the preset threshold, trigger the reexamination scheduling.

[0054] S3: When the risk level exceeds the preset threshold, or the uncertainty of the cross-clock warning model exceeds the upper limit, trigger the reexamination scheduling, and output the risk level and the reexamination date.

[0055] Furthermore, the review scheduling is only initiated when the system determines that the user is currently in a high-risk state or there is significant uncertainty in the inference model. Specifically, if the high-risk level probability of any complication in the inference result of the cross-clock warning model exceeds the set threshold, or the system detects that the entropy value remains high in consecutive sliding windows, the system deems that the current state requires supplementary slow-clock information for calibration. In this case, the review scheduling is automatically activated to evaluate the best review time within the next few weeks and make a decision by comprehensively considering the risk-benefit and the cost of medical visits.

[0056] The review scheduling dynamically determines the optimal date of the next in-hospital review by integrating the information entropy decline potential of the comprehensive risk curve and the consumption of medical resources.

[0057] First, cache the current entropy values of each complication and generate a set of candidate review dates on a weekly basis within the range of 2 - 26 weeks. Subsequently, based on historical follow-up data, the system predicts the information gain that can be brought by conducting an examination at the candidate dates, and introduces parameters such as the hospital outpatient load and the convenience of patient medical visits to calculate the comprehensive benefit index. Select the date with the maximum benefit index as the recommended review date, and at the same time output alternative dates within adjacent ±1 week for doctors to adjust according to the actual outpatient schedule.

[0058] Through the above mechanism, the system can keep the uncertainty of the complication risk at a controllable level while maintaining an acceptable patient burden, effectively reducing the probability of missed detection of complications caused by insufficient follow-up, and avoiding unnecessary high-frequency examinations.

[0059] Embodiment 2: In an exemplary embodiment, a diabetes warning and risk assessment system is further provided, including an acquisition module that acquires the clinical test data of patients, generates complication baseline data based on the clinical test data, and obtains the real-time blood glucose data of patients through a portable device.

[0060] A prediction module that constructs a cross-clock warning model based on a Bayesian inference network, uses the complication baseline data as the slow clock and the real-time blood glucose data as the fast clock, fuses a three-layer gating mechanism to adjust the observation weight, and dynamically outputs the risk level through logarithmic Bayesian accumulation and update.

[0061] A scheduling module that triggers the review scheduling and outputs the risk level and the review date when the risk level exceeds the preset threshold or the uncertainty of the cross-clock warning model exceeds the upper limit.

[0062] If the above functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0063] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0064] More specific examples (non-exhaustive list) of computer-readable media include the following: electrical connection parts with one or more wirings (electronic devices), portable computer disk cartridges (magnetic devices), random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memories), optical fiber devices, and portable compact disc read-only memories (CDROM). Additionally, the computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or otherwise processing it as necessary, and then storing it in a computer memory.

[0065] It should be understood that each part of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits with logic gate circuits for implementing logic functions on data signals, application specific integrated circuits with suitable combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0066] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for diabetes warning and risk assessment, characterized in that, Including: Collect the clinical test data of the patient, generate complication baseline data according to the clinical test data, and obtain the real-time blood glucose data of the patient through a portable device; Construct a cross-clock early warning model based on the Bayesian inference network, use the complication baseline data as the slow clock and the real-time blood glucose data as the fast clock, fuse a three-layer gating mechanism to adjust the observation weight, and dynamically output the risk level through logarithmic Bayesian accumulation update; When the risk level exceeds the preset threshold or the uncertainty of the cross-clock early warning model exceeds the upper limit, trigger a reexamination schedule and output the risk level and reexamination date.

2. The diabetes warning and risk assessment method according to claim 1, characterized in that: The clinical test data includes diabetic retinopathy risk data, nephropathy risk data, neuropathy risk data, and cardiovascular risk data.

3. The diabetes warning and risk assessment method according to claim 2, wherein: The real-time blood glucose data includes collecting continuous blood glucose values within a fixed time window of the patient through a portable device and marking timestamps, and calculating the blood glucose fluctuation characteristics within the time window; Obtain the behavior information record of the patient, align the behavior information record with the blood glucose fluctuation characteristics, and identify the event-related fluctuation interval; Set the blood glucose threshold for the fixed time window based on the event-related fluctuation interval, and compare the blood glucose fluctuation characteristics with the blood glucose threshold to achieve blood glucose detection and early warning.

4. The diabetes warning and risk assessment method according to claim 3, characterized in that: The cross-clock early warning model includes a modeling framework with a dual time axis. Each type of complication corresponds to a slow clock hidden state trajectory, and each hidden state trajectory is updated by two types of inputs: Blood glucose fluctuation characteristics from the fast clock; Complication baseline data from the slow clock; A three-layer gating mechanism is established between the fast clock and the slow clock to generate gating weights for each fixed time window and regulate the response intensity of different complications to the fast clock data; Adopt logarithmic cumulative Bayesian update to perform state update according to the blood glucose fluctuation characteristics and complication baseline data; Based on the mutual influence of complications, establish risk transduction jump nodes between complications and calculate the risk conduction relationship between complications.

5. The diabetes warning and risk assessment method according to claim 4, wherein: The three-layer gating mechanism includes three layers: individual sensitivity coefficient, daily rhythm coefficient, and real-time event coefficient; The individual sensitivity coefficient is a fixed value set according to the clinical test data. The individual sensitivity coefficient is positively correlated with the degree of complication development and is updated each time the clinical test data is collected; The daily rhythm coefficient includes retrieving the real-time blood glucose data within the past month, obtaining the quantile median smoothing curve according to the 24-hour scale as the daily baseline, and the daily rhythm coefficient is positively correlated with the degree of deviation of the current window blood glucose data from the daily baseline; The real-time event coefficient is determined according to the preset template library of complications. The calculation of the gating weight is expressed as: ； Among them, represents the gating weight within the time window; represents the individual sensitivity coefficient; represents the circadian rhythm coefficient; represents the real-time event coefficient.

6. The method for diabetes warning and risk assessment according to claim 5, characterized in that: The dynamic output of the risk level through logarithmic Bayesian accumulation update includes initializing the corresponding state prior distribution of the complication baseline data and calculating the gating weight within the time window; Based on the blood glucose fluctuation characteristics within the time window, fit the observation likelihood under the complication baseline data and perform Bayesian update through logarithmic accumulation, expressed as: ； Among them, represents the posterior probability of the updated risk state of the complication after fast clock time windows at the time point; ; represents the risk level of the risk state; represents the initial posterior probability of the complication at the time point ; represents the gating weight of the complication in the th fast clock time window; represents the blood glucose fluctuation characteristics in the th fast clock time window; represents logarithmic calculation; Obtain the posterior probability distribution through Softmax normalization ; Calculate the risk conduction relationship between complications through the risk transduction jump node, and calculate the risk jump factor for each pair of related complications: ； Among them, represents the source complication for the target complication at the time point jump factor; represents the calculation function of the jump factor; represents the gating weight; represents the source complication blood glucose fluctuation characteristics; represents the target complication blood glucose fluctuation characteristics; If is greater than a preset threshold , then is used as a gain term to adjust the posterior state distribution of the target complication : ； Among them, represents the posterior probability distribution of complications after jump factor adjustment; the posterior probability distribution; represents the normalization function; represents the posterior probability distribution of unadjusted complications the posterior probability distribution; represents the jump gain coefficient; Map the posterior probability distribution of the complication to the risk level. When the risk level exceeds the preset threshold, trigger a reexamination schedule.

7. The diabetes warning and risk assessment method according to claim 6, wherein: The reexamination scheduling includes, taking the current date as the base point, forming a set of candidate reexamination dates within the range of 2 - 26 weeks in the future with a 7-day step. For each candidate reexamination date, calculate the information gain and resource cost, generate a benefit index, and output the date corresponding to the highest benefit index as the reexamination date.

8. A diabetes warning and risk assessment system, applied to the diabetes warning and risk assessment method according to any one of claims 1 to 7, characterized in that, including, a collection module that collects the clinical test data of the patient, generates complication baseline data based on the clinical test data, and obtains the real-time blood glucose data of the patient through a portable device; a prediction module that constructs a cross-clock early warning model based on a Bayesian inference network, uses the complication baseline data as the slow clock and the real-time blood glucose data as the fast clock, fuses a three-layer gating mechanism to adjust the observation weight, and dynamically outputs the risk level through logarithmic Bayesian accumulation update; a scheduling module that triggers the reexamination scheduling and outputs the risk level and the reexamination date when the risk level exceeds the preset threshold or the uncertainty of the cross-clock early warning model exceeds the upper limit.

Images